The rapid development of the Internet has led to an exponential growth of multilingual information contained in the network, and traditional translation is difficult to meet the needs of users, and intelligent language translation has important research value and application prospects. This study adopts convolutional neural network to extract the visual features of translated images, and analyzes the correlation between image and text features by using the mechanism of region-selective attention to align the features of text and image in the translated information. Then the two information features are fused and processed, and input into the sequence model to realize the intelligent language translation, so as to obtain the intelligent language translation algorithm based on computer vision. The research results show that the intelligent language translation algorithm in this paper has comprehensive advantages in several key evaluation indexes, highlighting its performance in improving the quality of translation language generation. The application in translation real-world scenarios is able to maintain a low leakage (1.30%) and mistranslation rate (2.64%), and the translation response time is also able to be maintained at around 67.28ms. The proposed intelligent language translation algorithm has high advantages in performance test and application in real scenarios, has good generalization and applicability in multilingual translation, and is expected to be more widely used in the future.

The study analyzes the stylistic evolution of contemporary Chinese literary works using the MONK project. Text mining tools in the project are used to analyze the thematic classification, emotional tendency and stylistic type changes of the works. Among them, LDA model and GBDT algorithm are used to identify the thematic classification of Chinese modern and contemporary literary works, SO-PMI algorithm is used to identify the emotional tendency in the works, and the vector space model can classify the style of the works. Based on the above methods, the theme and emotional changes of modern and contemporary Chinese literary works can be categorized into 3 stages: the awakening of Enlightenmentism at the beginning of the 20th century, the diversified presentation during the revolutionary period, and the diversified development after the reform and opening up. The styles of modern and contemporary Chinese literary works can be divided into epic style, lyrical style, rural theme style and intellectual theme style.

The internal defects and concrete strength detection of concealed mass concrete structures (dams, fan foundations, tunnel arches, etc.) has been a difficult problem in the industry, and there is a lack of effective nondestructive testing technology, conventional single-sided nondestructive testing technology (ground-penetrating radar, ultrasonic array, impact echo method, etc.) in reinforced concrete structures can not be more than 3m in depth, and the practical application is limited. For this reason, we have developed a new face wave CT inspection technique based on elastic wave face wave, combining the excellent wavelength method and multiple filtering method to solve the problem of difficult extraction of frequency dispersion curves of the face wave in concrete, and through finite element simulation and example verification, it is confirmed that the method can detect the defects and strength of the concrete structure on a single side, and the effective detection depth is more than 4m, which has a strong practical application value.

With the development and popularization of information technology, more and more data are gathered in the hardware equipment and systems of major universities. However, how to obtain the required information from the massive data and make use of it has become the main problem faced by the information departments of colleges and universities. Although there are many studies on university management data, most of them are shallow-level analysis based on simple data, models and tools, which are not very helpful for improving the efficiency of university management data analysis. In recent years, with the rapid development of artificial intelligence and multimedia technology, artificial intelligence and multimedia technology have been applied to more and more fields to improve efficiency. This paper analyzed and studied the data analysis algorithm of higher education management under the background of the rapid development of artificial intelligence and multimedia technology, aiming to optimize the traditional higher education management data analysis algorithm and improve the efficiency of university management data analysis. This paper also compared the traditional higher education management data analysis algorithm and the optimized analysis algorithm. The comparison results showed that the optimized higher education management data analysis algorithm based on artificial intelligence and multimedia technology improved the efficiency by 11.4%.

As an important part of economic activities, logistics industry ushers in new development opportunities and challenges in the wave of digital transformation. The study explores the path of integration and development of digital economy industry and logistics industry, designs the path of building intelligent logistics ecosystem, and constructs the logistics distribution path optimization model based on time window. When analyzing and solving the logistics distribution path optimization problem, the ant colony algorithm (ACO) is improved by introducing the hierarchical idea of the artificial bee colony algorithm (ABC) and limiting the pheromone concentration on each path, controlling it within a known range, to make up for the shortcomings of the ant colony algorithm of precocious maturity and search stagnation. Using MATLAB software to simulate the logistics and distribution of M fresh food e-commerce enterprises, the comprehensive cost solved based on ABC-ACO algorithm is 75.64 yuan and 33.45 yuan less than the results of ACO and GA solving, respectively, and the optimal route traveling mileage is 21.35 km and 6.03 km shorter than the mileage solved by ACO and GA solving, respectively. It shows that the performance of the improved ant colony algorithm is better than that of the basic ant colony algorithm and the genetic algorithm, and it points out the direction for the future logistics and distribution of the distribution center. The empirical analysis found that the digital economy industry and logistics industry show a synergistic trend, and there is a large space for integration and development.

New media advertising is a standard mechanism used to increase the operating income of the platform, and intelligent optimization and recommendation of advertising content is a very beneficial mechanism for new media platforms. In this paper, a multi-task deep learning neural network model is used to realize the intelligent optimization of new media advertisement content, improve the model’s effect on advertisement content optimization through the attention mechanism in the model, and further improve the performance of the model by using the loss function. The model is then combined with blockchain technology to establish an intelligent optimization and recommendation system for new media advertisement content, which achieves personalized and accurate new media advertisement content recommendation. It is verified that the multi-task deep learning neural network model proposed in this paper achieves good results in the intelligent optimization of new media advertising content, and the system performance meets the functional and non-functional load requirements. In addition, under the application of intelligent content optimization and recommendation system, most of the new media users’ advertisement browsing duration is higher than 50 s. Compared with the traditional advertisement recommendation system, the advertisement content intelligent optimization system proposed in this study has the advantages of strong purpose, strong targeting, fast effect, low cost, etc., and it has obvious advantages in enhancing the user’s interest in the advertisement content.

Entity-relationship extraction task is one of the very important research directions in the field of natural language processing, aiming at identifying and determining the existence of specific relationships between entity pairs from unstructured text. The study firstly introduces the related theories of graph neural networks in terms of graph representation learning and graph neural networks, and then makes full use of the information of dependent syntactic trees to propose a relationship extraction model based on dependency graph convolution (DGGCN). The validity of the model and the entity extraction effect are verified through relevant experiments.The DGGCN model is fully experimented on the public datasets NYT and WebNLG, and the F1 value is effectively improved.According to the results of the ablation experiments, it is shown that the DGGCN model improves the entity and ternary extraction results by 0.5% and 4.3%, respectively. In the long and short distance entity extraction results, the DGGCN model outperforms the benchmark model in both long and short distance entity relations, but the extraction performance gap between short and long distance entity relations is still large and needs further improvement.

The concept of “Internet+Sports” has promoted the application of artificial intelligence and other emerging technologies in the field of sports. This paper mainly focuses on the special physical training, and explores the application and realization path of artificial intelligence technology in physical training test. In this paper, PSO-BP model is constructed based on BP neural network optimized by PSO intelligent algorithm and applied in physical training test. In addition, for the classification of physical training, this paper follows the basic principles of physical training system construction, establishes the physical training measurement index system through the results of expert solicitation, and determines the weights of each index by using the hierarchical analysis method. Through the empirical analysis of the PSO-BP model in this paper, it can be seen that the fitting results of the training samples of male and female students show that the corresponding correlation coefficients of male and female students are 0.99908 and 0.99898, respectively.The errors of the evaluation output values of the physical training measurements and the expected values are within ±3.5, and the prediction error of the BP neural network model optimized by the PSO algorithm is significantly reduced, and the relative errors of the evaluation of male and female students are reduced by 0.988% and 0.833%, respectively. The results show that the results of physical training measurement and evaluation using PSO-BP neural network model are more accurate, which proves that the performance of PSO-BP neural network in this paper has been effectively improved and optimized, and at the same time, it can meet the application requirements of physical training measurement and evaluation.

Urban safety development is one of the guarantees for the overall development of the city, and the study uses Delphi method, entropy weight method and TOPSIS method in the assessment of urban safety development. An improved Delphi-entropy weight-TOPSIS combination assessment model is constructed to evaluate the urban safety development. The evaluation index system of urban safety development is constructed, and the evaluation indexes of urban safety development are calculated by Delphi method and entropy weight method respectively, and the subjective and objective weights of the evaluation indexes of urban safety development are derived, and finally, the comprehensive weights are calculated by the method of combined weight assignment. The comprehensive weights of the guideline layer of the urban safety development evaluation index system are 0.1874, 0.2080, 0.2005, 0.2187, and 0.1854, respectively.The evaluation index system is used for empirical research, and City A is taken as the object of the research to assess its urban safety development status during the 10-year period from 2014 to 2023. From the evaluation results, it is known that the overall urban safety development of City A during the 10-year period shows an upward trend, with slight fluctuations in the process, but the overall development is good, and the evaluation score of urban safety development improves from 0.4657 points in 2014 to 0.6479 points in 2023.

In order to meet the needs of high-quality development of the civil engineering industry, it is necessary to carry out corresponding teaching reforms in the level drawing course as a core basic course. The purpose of this paper is to explore whether the case teaching of leveling drawing can effectively improve students’ ability of leveling drawing. By analyzing the level drawing course, the case teaching method of level drawing is designed. Students of a higher vocational school were selected as the experimental objects, and the questionnaire survey was used to understand the current learning status of the students’ drafting and to carry out the teaching practice, and the statistical analysis method was used to explore the teaching effectiveness. After the teaching practice, the students’ learning attitude, skill mastery and teaching satisfaction increased by 46.24% as a whole, which was significantly different from the learning status quo before the practice (p < 0.01). Meanwhile, there was an improvement of 8.52% and 5.57% in learning achievement over the pre-practice and comparison students, respectively. The results indicate that case teaching of leveled drafting can effectively improve students' learning attitudes, develop students' skill mastery, enhance teaching satisfaction, and it has a significant role in promoting students' learning outcomes in leveled drafting. This study confirms the value of case teaching of plain drawing in professional practice and has positive significance for improving the quality of education.

This paper analyzes the current teaching development direction of intelligent simulation context, and argues that the development and updating of semi-open human-computer dialogue systems based on scenes and topics need to rely on the division of labor between machines and human beings. In this regard, the development strategy of intelligent teaching of English dialogue interaction is proposed by combining the speech corpus annotation system designed based on artificial intelligence technology and the dialogue interaction teaching strategy derived from the interactive teaching model. The corpus annotation model of multi-layer perceptual machine is designed, which consists of real-time interaction module, core technology and algorithm module and data storage module. Draw a mind map of the framework for analyzing the effectiveness of dialog teaching, and develop a dialog interactive teaching strategy for pre-class dialog, in-class dialog and after-class dialog by invoking the interactive teaching model. Analyze the annotation results of the intelligent annotation model of speech corpus in LDC corpus and UN corpus. Observe and organize the teaching implementation effect of the dialogue interactive teaching strategy, and illustrate the pedagogical feasibility of the English dialogue interactive intelligent teaching development strategy proposed in this paper in combination with the learning achievement and questionnaire results. In the classroom English conversation, the zero feedback of dialog teaching is only 1.60%. The linguistic feedback of dialogue is reflected in the proportion of 33.87% of combing and summarizing, 33.55% of judging the correctness and error, and 30.99% of extending and pursuing questions. The effectiveness of English conversation classroom is improved.

In this paper, seven dimensional variables of marketing mode development and product quality improvement are selected under the 5G network security theory. In order to clarify the relationship between the two, the questionnaire was used to obtain the quantitative values of the seven dimensional variables, and then the Pearson correlation analysis was used to test whether the dimensional variables could be used in the regression analysis. After passing the test, multiple linear regression is used to interpret the relationship between the development of agricultural e-commerce marketing mode and product quality improvement. The regression equation of strengthening brand awareness on product quality is B1=0.511+0.439*B4+0.653*B5+ 0.327*B6+0.297*B7, which demonstrates the relationship between product quality enhancement and strengthening brand awareness in the development of agricultural products e-commerce marketing model in a more intuitive way. This study makes people have a more new cognition of the role relationship between the two, which is of great significance for practicing the goal of synergistic development of the two.

In this paper, we design a cloud-integrated financial robot based on artificial intelligence technology to provide advanced financial analysis and decision support for financial institutions. The robot platform is embedded with a large amount of financial domain knowledge and data, which can provide uninterrupted financial services to customers using a dialog engine. At the same time, it is equipped with the attention mechanism – long and short-term memory neural network model, in investment transactions and credit risk prediction, which can bring a new digital intelligence experience for financial institutions. The standard and similar sentence recognition accuracy of the article robot dialog engine can be stabilized at more than 90%, and the average access time of the user’s access request is about 0.15s. The importance distribution of financial credit risk indicator features is 5~24, and when the number of features takes the value of 10, the risk prediction accuracy of the robot in this article is the highest, 97.98%. When the prediction model is trained to 50~70 epochs, the Loss value of the financial robot converges to 0.15~0.17. The accuracy of the model chosen in this paper for risk prediction as well as stock prediction is 95.35 and 96.2% respectively. And the absolute difference between the predicted stock price and the true value of the model in this paper is 0~0.28 yuan. Combined with DMI strategy for stock trading, the return is 30.7%. The financial robot improves the user experience and increases the value of risk control as well as stock prediction for financial institutions.

This paper constructs the SAM agile iterative model according to the direction of online art course design, firstly by collecting art teaching related information and initiating the cognitive system of art teaching, and then entering into the iterative design phase to accomplish the development goal of the online art course incrementally through continuous iteration. Finally, after the double iteration phase, the software process enters the delivery phase to complete the design of the online art teaching course. The effect of the online art teaching course and its impact on delivery are analyzed in conjunction with the dynamic key-value memory network model based on the forgetting curve. The results of the memorization ability of art knowledge experiments in the pre-test have a mean value of 35.259, and the post-test has a mean value of 53.1254, while the Sig value of the paired test is 0.000, 0.000<0.05, which indicates that the effect of using the online course for art learning based on the Ebbinghaus forgetting curve is more significant on the learning of art knowledge than other applications. The regression results of the full sample model showed that overall instructors' use of big data aids for online art instruction significantly affects instructional delivery, t=1.245, P=0<0.05, which is significantly positive at the 1% level, indicating that the more adequate the use of these instructional methods, the higher the probability that students will rate their satisfaction with the instructional delivery.

In order to improve the accuracy of automatic detection of malicious code, this paper focuses on the “texture” features of malicious code and the characteristics of different types of malicious code, which are also different, and uses them for the automatic detection of unknown malicious code by using the four machine learning algorithms of KNN, RF, NB and SVM to perform single-feature detection and multi-feature (GLCM, LBP and ngram feature merging) detection respectively. Four machine learning algorithms, namely KNN, RF, NB, and SVM, are used to perform single-feature detection and multi-feature (GLCM, LBP, and n-gram feature merging) detection respectively, and analyze the accuracy of the spatial relationship feature-oriented malicious code detection scheme. A multi-version oriented data protection model is proposed for the data storage space, data version, quantity management and recovery requirements involved in service emergency response. The relative performance errors between its data protection scheme and the plaintext scheme and the simple add noise scheme are analyzed. In all four machine learning algorithms, the detection rate of fused features is higher than that of single features, and the maximum difference can reach more than 60%. When takes the value of 9 or 3, the data privacy protection algorithm, the plaintext algorithm, and the noise-only addition algorithm in this paper have similar accuracy rates. With proper noise selection, this paper’s scheme has good performance in real simulation.

Computer multimedia technology has brought unprecedented innovation to the film and television production industry. Multimedia technology in film and television post-production mainly focuses on two aspects of image processing and audio processing, this paper selects the skin color enhancement and voice enhancement for further research. Adaptive skin color enhancement method is proposed, IMCRA-OMLSA audio enhancement method is selected, and relevant experiments are designed to compare this paper’s method with other classical skin color enhancement and voice enhancement methods respectively, and the effectiveness of this paper’s method in skin color enhancement and audio enhancement is examined through the results of subjective and objective evaluation. The accuracy and F1 value of this paper’s adaptive skin color detection method are 0.961 and 0.945, respectively, and the performance of skin color detection is good. The adaptive skin color detection method in this paper has the best performance with a comprehensive evaluation score of 6.81. In the objective evaluation of speech enhancement, the PESQ, STOI, WSS, and RMSE values of IMCRA-OMLSA method in this paper are 2.03, 72.36, and 38.06, respectively, which are all optimal results. On subjective evaluation, the MOS value of IMCRA-OMLSA method is 1.88 which is the highest value.IMCRA-OMLSA method has the best performance for speech enhancement.

As an important part of the excellent traditional culture of the Chinese nation, Chinese Wushu condenses the wisdom of the Chinese nation, contains the genes of Chinese culture, and has important communication value. Based on the big data Hadoop technology, the article proposes a content recommendation design scheme for all-media communication of wushu cultural communication, and introduces the LFM model and MBGD algorithm to construct an intelligent recommendation model of wushu cultural communication content under the framework. Then, based on Lasswell’s 5W model, the fsQCA method was utilized to explore the relevant factors affecting the effect of martial arts cultural communication. When the number of hidden factors of LFM-MBGD intelligent recommendation model is 55, its RMSE is 0.92, and the HR@K value of the model can reach 62.12%. The consistency level of the existence and non-existence states of each conditional variable of the communication effect of wushu culture is less than 0.8, and the overall coverage rate and the coverage rate of each path are higher than 0.85.The wushu culture communication system driven by intelligent technology can start from building an online resource base of wushu culture, broadening the communication paths of wushu culture, sounding the laws and regulations of wushu culture communication, and building the brand of wushu to improve the communication effect of wushu culture.

In order to improve the intelligent processing capability of the server of the electric power information platform, the intelligent control platform of electric power informatization based on intelligent data analysis is designed. Taking the regional electric power headquarters as the base point, deploying the electric power informatization intelligent management and control workbench, connecting the necessary systems for electric power operation through the telecommunication management network (TMN), and completing the platform hardware structure design. Divide the platform monitoring function into four parts: query instruction issuance, feedback data reception, data parsing, and data storage, and monitor power data in real time. Deploy data collection algorithms on the data collection server to collect power data such as power harmonics, effective voltage and current, active and reactive power, and harmonic distortion. And Deep Belief Network (DBN) is used to train the anomaly detection model, which realizes the detection of abnormal behavior of the system. Determine the experimental methods and steps, and test the results: the server of the intelligent control platform for electric power informatization designed in this project passed the pressure test of the number of clicks per second and throughput of 100 and 500 simulated users, and has superior traffic processing capability. Application test of the platform, through the test to achieve the design requirements of the system’s various functional modules, in the distribution network line and equipment operation status monitoring, fault precision judgment, fault time statistical analysis and daily repair and other work has achieved certain results.

In today’s increasingly complex and dynamic network structure, cloud computing brings great convenience to computer users and meets people’s requirements for rapid computer data processing. The article firstly analyzes the cloud computing architecture and the security threats existing in the cloud environment, and explains the importance of cloud computing access control mechanism to ensure data security, starting from the traditional access control. Then it introduces the multi-authority attribute access control scheme based on blockchain and elliptic curve improvement, on the basis of which it proposes a blockchain-based cloud data security sharing model and a blockchain transaction privacy protection scheme, which both meets the user data privacy protection needs and realizes privacy computing. Finally, the security of the two schemes is analyzed, and compared with other schemes with the same mechanism. The results show that the blockchain-based cloud data security sharing scheme has better performance and scalability, which shows a stable linear growth of 1x, and the time load introduced by this scheme while enhancing the security of the encrypted data sharing system is acceptable compared to the other schemes to satisfy the application scenarios with large-scale access requests. At the same time, the blockchain transaction privacy protection scheme ensures data privacy while the average time consumed meets the user’s requirements for fast response.

STEM education emphasizes the in-depth integration between the knowledge of different disciplines, which is based on real problem solving, aims to establish an organic link between education and life, and takes the cultivation of composite talents with a sense of innovation and hands-on ability as the fundamental purpose. Aiming at the current problems of STEM education, the development path of STEM+ education based on digital visual virtual reality is proposed. Then, combining the DEMATEL method and Interpretive Structural Modeling (ISM), the dynamic factors affecting the development of STEM+ education are explored. Finally, the fuzzy set qualitative comparative analysis (fs/QCA) method was used to analyze the group path of STEM+ education high-quality development. The results of the analysis of motivational factors show that the governmental promotion among the extrinsic motivational factors has a high centrality and is a deep factor that drives the development of STEM+ education. Synergistic motivational factors play the largest role among the three dimensions and are the key to ensure the development of STEM+ education. Endogenous motivational factors are the direct motivational factors for the development of STEM+ education and need to be focused on control. The analysis of the grouping paths in region C, for example, shows that there are two high-level grouping paths and three non-high-level grouping paths, multiple grouping paths with different paths, and high-level grouping and non-high-level grouping are in an asymmetric state. There are some differences in the grouping paths in the east, center and west, and the three regions’ high-quality development of STEM+ education cannot be separated from the support of state factors and response factors. This paper provides a path reference for realizing high-level STEM+ education high-quality development.

In this study, the ecological environment landscape pattern index was selected to construct the ecological environment landscape data representation model under the self-organized feature mapping network (SOM) technique. The input data to the model were monitored under unsupervised conditions and made more sensitive to specific characterization information in the neuronal structure (Hebb), which resulted in different groups of regular data. The moving window method shows that the landscape index is unstable under the 1000-3000m window and the magnitude of change begins to decrease at the 4000m scale. The data tends to stabilize at 5000m scale, and the stability of the data decreases at 6000-7000m, and the anomalous data increases at this time. In terms of landscape level, the aggregation and connectivity of the overall landscape of the study area increased and landscape fragmentation, complexity and diversity decreased under the 4000m window. The land use change model based on SOM network can well reflect the law of land use change in the sample area, which greatly expands the spatial analysis research method of land use change.

Computer image-assisted design, as a product born in the era, provides more inspiration and creativity for art design. Based on the study of the basic theory of color design and the theory of color harmonization, an intelligent color matching model integrating visual aesthetics based on conditional generation adversarial network is proposed. Then a candidate graphic layout generation method based on visual saliency is proposed, which not only considers the visual saliency of each element in the image, but also considers how to generate candidate text regions under the constraints of aesthetic rules. In the visual analysis analysis experiment, under different color transformations, the F-value of the subject’s gaze time was 2.548, with significance P=0.051, which is not significant. The F-value of average gaze point is 6.398, significance P=0.002, significant difference is obvious. From this, it can be concluded that the artistic innovation design method proposed in this paper can make the subject’s point of interest change with a large difference, and the color that highlights the target object can significantly attract people’s attention, which is a feasible artistic innovation design scheme.

Aiming at the problem of large prediction error caused by the complex background of macroeconomic prediction, this paper proposes a macroeconomic prediction model based on time series clustering. The model adopts sparse self-encoder to deeply mine the features of the input vectors, constructs a bidirectional threshold cyclic unit network, and predicts the preliminary trend of the macroeconomy, and proposes a time series deep clustering algorithm that integrates the multi-scale feature extraction and clustering objectives of time series data into the same network. A sample generation strategy based on data augmentation and a multiclassification assistance module are used to extract the invariant patterns contained in the time series data to obtain a better representation for targeting time series clustering. Comparing this paper’s model with different forecasting models, the RMSE metrics are 0.0038 and 0.003 for the two time horizons, which are better than the other two models. The prediction range of this paper’s model for future GDP is 5.8%-5.9%, which is smaller than the GDP prediction range of the ARIMA model, indicating that this paper’s model is suitable for the realistic application of macroeconomic forecasting.

Accurate short-term load forecasting of distribution networks can ensure the normal life and production of the society, effectively reduce the cost of power generation, and improve the economic and social benefits. Aiming at the multivariate information that affects the power load, this paper utilizes factor analysis to reduce the dimensionality of the original influencing factors, and obtains the main influencing factors with the highest contribution rate, so as to guarantee the accuracy of the neural network prediction. On this basis, the neural network structure is improved by combining AlexNet and GRU, and the short-term load prediction model of distribution network is finally constructed. The relevant charge data of N village in 2023-2024 is used as a research sample to analyze the main influencing factors of its short-term load change, and three main influencing factors affecting the load change in the short term are identified as temperature, air pressure, and humidity factor. Based on the real data of N-village distribution network to carry out prediction simulation experiments, the load short-term prediction curve of this paper’s model has a better fitting degree and good stability, and the values of the prediction result evaluation indexes MRE, RMSE and MAE are smaller than those of the other comparative models, which are basically able to maintain a prediction accuracy of more than 90%.

Aiming at the allocation of teaching resources for school affairs scheduling, a decision-making model for school affairs scheduling is designed based on a multi-objective optimization model. The “conflict detection and repair” module is added after the “initial population generation” operation in the traditional genetic algorithm, which decouples the scheduling model and meets the needs of scheduling decision-making. The designed method is compared with the standard genetic algorithm and stochastic two-point crossover genetic algorithm on the data set, and then the efficiency of resource allocation for school scheduling is improved by solving an example problem. The average faculty satisfaction with scheduling is 2.8, which is about 17% higher than the second place NPGA. Applying the algorithms to a college scheduling project, the feasible solutions of the algorithms in this paper satisfy all the various constraints, and the results of the three-stage style algorithm in the self-selected course scheduling mode yield better solutions than the baseline algorithm based on the course set in any of the arithmetic cases. This paper provides an informative solution path for the allocation of school scheduling resources, which can satisfy the course allocation needs of the three parties: teachers, students and schools.

As the most intuitive visual phenomenon of animated films, color has emotional characteristics that are closely related to the viewers’ emotional experience. From the perspective of chromaticity and psychology, we explain the method of color emotion quantification, calculate the fuzzy affiliation degree and grey correlation degree for the uncertainty and fuzziness between color and emotion mapping, put forward the method of fuzzy grey correlation for emotion mapping in animated movies, and carry out the experiment of color emotion mapping in animated movies. Through the experiment, it is found that the character color schemes of warm, cold and neutral colors are suitable for the design of character color emotion experience in animated movies. Taking the animated film “Ne Zha: The Descent of the Magic Boy” as the research object, the correlation between color emotion mapping and character matching is further explored. Most of the H-value color blocks in Ne Zha are distributed between 0-60, which indicates warm and neutral tones, and the distribution of S-value and V-value color blocks shows a clear trend of decreasing color saturation, while the overall luminance remains basically stable. The whole film takes the proportion of red, blue, color purity changes and other aspects of color design to achieve the position of the characters, the character of the transformation of the transformation of the matching and implied.

Speech-text multimodal large model as a key tool in the operation of the power industry, its fault prediction performance directly affects the operational safety of mechanical equipment, this paper designs a detailed scheme for the optimization of its performance. Firstly, the structural design of the unimodal model is discussed, and the audio classifier based on Wav2Vec2 and the text classifier based on BERT are used to pre-train the model. Based on the above foundation, a multimodal model is introduced, with the cross-attention mechanism as the fusion strategy, so that the different modal information in the deep neural network is fused with each other, thus improving the accuracy and robustness of the recognition task. After completing the fault feature extraction task, on the premise of introducing the relevant theory of BNN, the structure of BBN is optimized, and after fusing the HC algorithm, BIC and annealing idea, the fault diagnosis method based on the improved BBN network is constructed by combining the fault feature extraction method in the electric power industry and the optimized BBN method. The effectiveness of the method is verified through simulation experiments. The prediction accuracy of this paper’s method for nine categories of fault data is above 90% at a high level, and the prediction accuracy of faults in some categories can reach 100%. The multimodal model fusion strategy proposed in this paper significantly improves the performance of fault feature recognition, in addition, the fault diagnosis method based on the improved BBN reduces the computational volume of the model and improves the fault prediction ability of the model.

Through the investigation of Chinese reading comprehension ability, the evaluation index system of Chinese reading comprehension ability is constructed, combining the hierarchical analysis method (AHP) and the data characteristic method (CRITIC) to combine the indexes to assign weights, and then using the fuzzy comprehensive evaluation model to calculate the indexes to quantify Chinese reading comprehension ability. After that, the indicators affecting Chinese reading comprehension ability in language education were screened and sorted out using a binary logistic regression model, and the Chinese reading comprehension ability education was optimized based on machine learning. This paper constructs a systematic evaluation model of Chinese reading comprehension in colleges and universities with 5 first-level indicators and 22 second-level indicators, and obtains the final score of the system of 87.73 points, the fuzzy comprehensive score of the five first-level indicators of “reading ability, general comprehension ability, deep comprehension ability, evaluation appreciation ability, and comprehensive application ability” is between 86.63 points and 88.68 points, and the fuzzy comprehensive score of 22 second-level indicators such as vocabulary, language comprehension ability and logical reasoning ability is between 80.68 points and 90.38 points. The final score of each indicator was 88.67, and the model was evaluated extremely well. In addition, the empirical analysis showed that all the indicators had a significant effect on Chinese reading comprehension (P < 0.05), and the language education should be optimized in terms of vocabulary mastery and the cultivation of critical thinking.

Reasonable and scientific supplier selection and resource allocation is a prerequisite for enterprises to optimize the quality of supply chain and avoid business risks. In this paper, we select multiple supplier evaluation indexes, use decision tree algorithm to train and calculate the hierarchy of suppliers to determine the supplier options that can be selected. Then the main body of procurement resource planning decision-making is divided into three types: purchaser, database vendor, and customer, to establish a multi-objective model for optimal allocation of procurement resources, and the model is optimized by genetic algorithm to solve the optimal allocation scheme of procurement resources. The supplier selection method based on decision tree can realize the optimal selection of suppliers by constructing a decision tree and transforming it into If-then classification rules. The procurement solutions based on genetic algorithm are 10.44%, 4.31%, and 5.14% higher than B, C, and D solutions, respectively, for better allocation of procurement resources.

Studying the influencing factors of logical reasoning ability can not only help teachers to find out the effective way to cultivate students’ logical reasoning ability, but also provide methodological and theoretical references for the relevant research in the area of artificial intelligence-driven program design education, which is of certain research value. The article firstly introduces the theory of structural equation modeling and the principle of algorithm used in model analysis. Then, taking the students of School S and School T as an example, we designed and distributed relevant collection questionnaires, and analyzed the data using SPSS to understand the overall status of students’ logical reasoning ability and the level of each dimension. Then we make reasonable assumptions about the factors affecting students’ logical reasoning ability, establish a structural equation model of the factors affecting logical reasoning ability, and analyze the effects and paths between the factors and on the logical reasoning ability. Finally, according to the experimental results, we propose targeted teaching reform methods. The results of the study show that: teacher’s activities, learning interest, learning attitude, classroom environment have a positive effect on students’ logical reasoning ability, in which the effect of classroom environment on logical reasoning ability is 0.48. Enhancing the teacher’s power and promoting the diversified development of students is an effective way to improve logical reasoning ability.

Personalized service is a targeted initiative for digital resource libraries to improve the quality of service and better play the function of culture and education. This paper proposes a digital book personalized recommendation algorithm based on artificial intelligence technology. After acquiring the borrowing data and pre-processing, the reader’s portrait is visualized with factor analysis and cluster analysis methods respectively. The traditional Slopeone algorithm is weighted and the collaborative filtering algorithm is improved. Combine the user profile with collaborative filtering to realize the personalized recommendation of digital books. User similarity calculates four types of readers such as pragmatic, youthful, recreational and curious. This paper’s algorithm outperforms CFRA and RABC algorithms under each parameter, with the highest recommendation accuracy and novelty, and realizes personalized library services.

There is an increasing demand for assisted training techniques in the sport of sparring. In this paper, a sparring multiple recognition and analysis system is designed and fabricated for the movements of sparring sports and used to recognize and analyze the players’ technical movements using the collected data and the model built using deep neural networks. The CNN-LSTM network is applied to extract the feature classification of the preprocessed sparring inertia data, and then the DTW algorithm is combined with the spatial distance classification method to realize the matching and recognition of sparring behaviors by stretching and compressing transformations of the time axis, effectively eliminating the distortion error in the time domain and obtaining the similar path with the shortest cumulative distance of the effective matches between different sequences. Experiments on the application of this paper’s system were conducted in two groups of sparring players, and after 12 weeks of training intervention, the average confrontation striking speed of the experimental group progressed from 0.36 seconds before the experiment to 0.32 seconds after the experiment, and the average performance of the control group progressed from 0.38 seconds before the experiment to 0.36 seconds after the experiment, which indicates that although the traditional resistance training also has a positive impact on the training effect of sparring training, the training effect of this paper’s system is more obvious The systematic training effect of this paper is more obvious. This paper makes an innovative exploration for the combination of sports programs such as sparring and cutting-edge information technology.

The continuous development of neural network makes the automated style migration technology also rise to a new height. This paper selects digital media art as the research field, constructs Cycle GAN, a cyclic consistent generative adversarial network structure applied to digital media art, on the basic framework of GAN, and optimizes it by adding bilinear interpolation and attention mechanism, so as to build up a style migration model for digital media art. In the style migration simulation experiment, the IS test values of this paper’s model on the photo2vangogh and photo2monet datasets are 5.32 and 6.03, and the FID test values are 97.52 and 75.55, which are better than the other comparative models. Similarly, the optimized performance of FID, SSIM and PSNR values on the dataset is also better than other comparative models, and the style migration performance of the model is verified. Using the model of this paper to design a digital topography with Chinese traditional ink painting as the content, we explore the correlation between the design attributes of the style migration design works in digital media art and the audience’s cognitive evaluation and overall perception. Among the design attributes, “plot relevance” (4.375) and “atmosphere rendering” (3.38) have the highest T-value, which is the most important influence on audience perception.

With the promulgation of relevant policies, virtual power plant market transactions are facing major adjustments, in order to promote the smooth entry of virtual power plants into market-oriented transactions and improve the economic benefits of virtual power plants, this paper proposes a virtual power plant market transaction model. The traditional virtual power plant resources are mathematically modeled, blockchain technology is introduced to build a decentralized trading framework, and fuzzy neural networks are combined to predict the power load of the virtual power plant. Then the decision-making model of virtual power plant participation in spot market trading is constructed by using two-stage stochastic planning theory with the goal of maximizing expected return. The results show that the prediction effect of the fuzzy logic-based virtual power plant market trading model is 2.925% higher than that of the traditional BP algorithm model, and its accuracy and stability are significantly improved. In addition, the distributed energy storage aggregated by the virtual power plant as well as the dynamic demand response rate is fast, the regulation is flexible, the short-time power throughput capability is strong, and it can accurately track the FM instructions. The cumulative FM capacity and FM mileage provided by the virtual power plant account for 84% and 99% of the total FM capacity demand in the system, respectively, making it highly competitive in the FM market. And under the premise of balancing riskiness and profitability, the bidding scheme of virtual power plant derived in this paper is more effective.

Some athletes’ lack of basic knowledge of exercise mechanism, mode, method, process and intensity has led to frequent occurrence of athletic risk events such as injury, disease and even sudden death, which seriously affects the physical and mental health of athletes and even threatens their lives. In this study, the data of athletes’ injury and disease risk characteristics were collected, and the feature selection method of Least Absolute Value Convergence and Selection Operator (LASSO) combined with Boruta’s algorithm was used to preprocess the data in order to eliminate redundant features. In terms of model construction, the prediction results of support vector machine, logistic regression, random forest algorithm and deep forest algorithm were integrated by using Stacking algorithm to construct the prediction model of athletes’ injury risk. After the predictive performance of the model is examined, it is used as an intervention for injury rehabilitation to carry out comparative experiments. The results show that the fusion model can effectively extract the feature importance of injury risk factors and predict the risk probability, and the prediction effect is better than that of a single model. Meanwhile, the intervention results show that the model has excellent effects on injury rehabilitation. This study can accurately predict injuries and illnesses, prevent the occurrence of injury and illness risk events in athletes, ensure the successful realization of sports goals, and play a role in assisting injury and illness rehabilitation.

The power optimization of wind farms and the optimal control of wind turbines require high-precision power ultra-short-term prediction for each wind turbine. In order to improve the performance of ultra-short-term prediction of wind power, this paper couples the LSTM model with the Logistic model and combines it with Graph Convolutional Neural Network (GCN) to construct the ultra-short-term prediction model of wind power based on Logistic-LSTM-GCN, and test and analyze the prediction performance of the model. Comparing the LASSO, XGboost, LSTM, GRU and TCN-LSTM models, the MAE and RMSE of this paper’s model are the lowest among all the models, which are 3.34% and 5.89%, respectively, and the R² is the highest, which is 79.76%. And the MAE and RMSE predicted by the model with inputs of four-dimensional spatio-temporal feature matrix are smaller than the model with inputs of one and two dimensions, and the R² value is larger than that of one and two-dimensional model. It indicates that the Logistic-LSTM-GCN model based on spatio-temporal information can extract the spatio-temporal information of wind farms more effectively, which improves the accuracy of wind cluster power prediction. In addition, with the increasing time step, the error indicators MAE, MAPE and RMSE are gradually increasing. Taking a time step of 4s for prediction, the prediction error of the model is minimized when considering multivariate variables such as wind speed, wind speed decomposition component, yaw error, wind direction, and rotor speed. This indicates that the multivariate LSTM, logistic and GCN coupled model can significantly improve the performance of ultra-short-term prediction of wind power.

The implementation of tax incentives is a powerful measure to reduce the burden of enterprises, build a new development pattern, and expand reform and opening-up. Some enterprises in nine provinces from 2010 to 2023 are sampled to verify the role of tax incentives in reducing the tax burden by using the double difference model. The weight coefficients are introduced as learning factors for the population center of mass, and the SWC-PSO algorithm is proposed to improve the shortcomings of PSO, which has low convergence accuracy and is prone to fall into local extremes, and to realize the mathematical planning for minimizing the tax burden of enterprises. After controlling the variables of tax policy and enterprise nature, the regression coefficient reflecting the enterprise tax burden is significantly negative at 1% level, and the tax burden of enterprises receiving tax incentives is significantly reduced, which proves the role of tax incentives in reducing the enterprise tax burden. After using SWC-PSO for planning, the sample units have a total of 1,779,919,000 yuan of tax relief, and the business tax rate of a construction project decreases from 3.35% to 0.42%, which indicates that the improved algorithm in this paper can plan the strategy of minimizing the tax burden of enterprises more efficiently.

Nowadays, all regions are actively carrying out rural ecological protection and restoration projects, and the overall trend of rural ecological protection projects is continuing to make positive progress, but there is still a lack of in-depth research on the evolution of the rural ecological environment and its mechanisms. In this study, the rural ecosystem distribution characteristics of Ganzhou District were investigated from various perspectives, including the number, spatial type distribution and spatial density distribution of different types of rural ecosystems in Ganzhou District. Geodetectors were used to study the spatial heterogeneity of rural ecosystems and the related driving factors behind them. Finally, the response of the service index of the integrated ecosystem to land use changes was studied, revealing the connection between different influencing factors and rural ecosystems’ heterogeneity, and proposing strategies and suggestions for the construction of a legal guarantee system for rural ecosystems in Ganzhou District. The most significant factor influencing the service index of rural ecosystem is the proportion of ecological land area, whose contribution rate is as high as 50%. The specific average coefficient of grassland is 7.99, which has the greatest influence on the CES index and is positively correlated with it. The X average coefficient of construction land is the only negative value, which is negatively correlated with the CES index. The results of this paper provide more specific scientific decisions for rural ecological environment protection, and in this way, further protection of rural ecosystems is realized from the improvement of ecological environment protection legislation, ecological compensation system and legal aid.

The extreme high temperature and erosive environment service environment in bridge construction puts forward higher requirements for high performance concrete and other aspects of performance. In this paper, compound mineral admixture is selected as a research breakthrough, and X-ray diffraction analysis (XRD) and Raman spectroscopy are used to explore the micromechanical behavior of compound mineral admixture in high-performance concrete. In the Raman spectral analysis, the stress distribution of the fitted curve of the compound mineral admixture is more flat and uniform, and the offset of the G’ peak position is higher than that of the reference concrete and the single-mineral-admixture concrete, and the stress can reach 2.5 MPa under 1% strain, showing good interfacial bond, stress transfer efficiency, etc. The physical phase data of the XRD also shows the frost resistance of compound mineral admixture, with the ability to mitigate carbon dioxide, and the ability to reduce the carbon footprint of the concrete, with the ability to reduce the carbon dioxide. The XRD data also show the frost resistance of the compound mineral admixture, which has the performance of slowing down carbonization. The NSGA-II algorithm is introduced and improved to propose a concrete proportion optimization model. The final evaluation function converges from 35 generations and the final value is 0.4558, which achieves the adaptive optimization of compound mineral admixture.

The undifferentiated recommended content of the existing library management system has been unable to meet the diversified and personalized needs of users, and for the large amount of user data accumulated in the library management system over the years, the value of the data is also yet to be tapped. This paper combines the requirements of personalized information recommendation in self-service libraries with the use of K-means clustering for the design of the label system and weight setting of user profiles. Then, based on traditional reinforcement learning, a reinforcement learning recommendation algorithm with Actor-Critic model is proposed, and the library information recommendation task is further modeled as a Markov decision-making process and utilizes reinforcement learning in order to automatically learn the optimal recommendation strategy, which rewards the users by maximizing the expected long-term accumulation. Meanwhile, the paper employs the DDPG algorithm to implement the parameter training of the Actor-Critic framework recommendation model to achieve better personalized recommendation performance. Comparing the recommendation model with the baseline model such as DeepFM on datasets such as Jester, this paper’s model scores 0.7708*, 0.1918, 0.7155, 0.3936 on ML (100k), Yahoo! Music, ML (1M), Jester, which is better than the traditional recommendation model as well as DeepFM based on Deep Learning and AFM works better because of its ability to model dynamically and to make decisions on good use of long-term rewards. The study makes an innovative exploration for accurate recommendation and improvement of user experience in libraries.

The field of artificial intelligence provides a new practical path for the inheritance and protection of non-heritage art. This paper proposes an innovative morphological design method for rattan weaving art based on fractal theory, and the grasshopper plug-in is selected to establish a parametric design model. The fractal graphics generated by the iterative function system are used as the input graphics, and the GrabCut algorithm and VGG16 neural network are combined to propose a graphic rendering method based on style migration containing elements of the cultural symbols of the Maonan Flower Bamboo Hat, and to realize the inheritance of the cultural symbols of the Maonan Flower Bamboo Hat. In the high preference survey, the A1 and A4 features of the sun hat in the questionnaire results are consistent with the preference results derived from the fractal design, and the questionnaire results of the handbag and handkerchief are also consistent with the preference results derived from the fractal design. It shows that the product form design method of Maonan flower bamboo hat cultural symbols based on fractal theory and style migration can play a certain role in promoting cultural inheritance.

In order to solve the shortcomings of the sound source separation method, this paper proposes a melody extraction method based on saliency and improved joint neural network, constructs the pitch saliency feature function according to the idea of harmonic energy superposition, pre-processes the audio, and then builds the joint neural network based on Res-CBAM according to the idea of joint neural network of music detection and pitch estimation classification to realize the melody pitch contour tracking. In addition, the calculation of the significance function is introduced to highlight the pitch significance features, so that the graphs input to the neural network have clearer melodic features. The results show that before and after the suppression of the accompaniment, the difference in the time-domain waveforms is not significant in the treble range, but there is a significant difference in the low-frequency range. In addition, the OA accuracy of the Res-CBAM algorithm proposed in this paper is up to 41.14% higher than other algorithms (P < 0.05), and the accuracy of the model is good. Applying this recognition model to teaching found that teaching with this model can significantly improve the subjects' perception of music (t=.197, p=0.002<0.05). It can be seen that the application of the Res-CBAM algorithm to actual music teaching is of great practical importance.

Semantics in public English texts are more challenging to understand accurately because they are influenced by specific contextual contexts. Traditional English text semantic understanding methods do not design their semantic understanding methods based on the conceptual semantic features of the text, and they have the problem of poor accuracy in understanding the deep semantics of English texts. For this reason, the article takes the public English text semantic algorithm as the research perspective, firstly conducts relevant theoretical research on English text semantic feature representation, then explores the text semantic extraction method based on the Dependency Tree-CRF, and deepens the understanding of English text semantics through the conceptualization and attention embedding methods. In the experiment of comparing the semantic coherence model with manual scoring, the experiment shows that by applying the semantic analysis model designed in this paper to the task of correcting the English writing of domestic college students and comparing it with the experimental results of manual scoring, it is found that the average absolute error between the scoring of the English compositions by this paper’s model and the scores of the compositions corrected by the teachers is 3.2051, i.e., the difference between the results of the manual correcting and the results of the correction by this paper’s model is It is not big, from which we can get that the model of this paper has good practical value.

Multidimensional vector space is the basis of lexical semantic correlation computation, which is able to assess the similarity between lexical semantics. In this paper, we implement a Japanese lexical named entity recognition and semantic relation calculation method based on this method. Dependency relations are fitted using N-Gram and knowledge expansion, contextual relations are corrected using collocation frequency, and semantic interactions are determined by semantic linking methods. The accuracy and recall of the identification of this method are higher than that of the spatial semantic role method by 0.78% and 4.93%, respectively, and the quantized values of the calculated correlations accurately reflect the strong and weak lexical semantic relationships. The results of the disambiguation experiments show that the maximum correlations computed using the method of this paper are consistent with the corresponding semantic items. Therefore, the method designed in this paper for recognizing named entities and calculating semantic relations of Japanese words has a relatively accurate recognition rate of semantic relations and has the ability of disambiguation.

Visual communication design, as an indispensable part of product design, plays an important role in enhancing the cultural connotation and aesthetic value of products. Based on fractal theory and supported by Iterative Function System (IFS), this paper studies the visual communication style design of patterns. Taking the flower pattern as an example, a method of automatic generation of flower pattern based on fractal geometry is proposed, and the effective value ranges of each parameter are derived through experiments and analyses to realize the digital visual communication design of the traditional handmade pattern. Then the generated fractal graphic is used as the content graphic, the style graphic is determined, the style migration technology is introduced, and the convolutional neural network model is constructed to build the style migration model of the product graphic, and experimental analyses are carried out to further improve the visual communication design of the product graphic. The average scores of this paper’s product graphic style migration method on aesthetics and style similarity are 3.95 and 3.81, respectively, and the p-values of the Mann-Whitney U-test are all less than 0.0001, which are significantly better than the baseline method. The average overall style similarity of this method on the real dataset is 86.27%, and the accuracy and mean square error on local style features are better than the VividGraph method, which has higher efficacy in performing product pattern style migration to realize visual communication style design.

Aiming at some configuration and scheduling problems of automated guided vehicles (AGV), shore bridges and yard bridges in the loading and unloading operation process of container terminals in the port logistics system, the flow characteristics of containers between ships and yards are analyzed in detail in the light of operational characteristics. Considering the intersection of AGVs with shore bridges at the quay front and the intersection of AGVs with yard bridges in the yard area, a container truck scheduling optimization model based on the objective of minimizing the operation cost is designed. And adaptive particle swarm algorithm (APSO-C) is used to solve the three-dimensional scheduling model of container in port logistics system. The results show that the fastest arrival scheduling rule is basically better than the shortest distance scheduling rule, and with the increase of the container task volume, the gap between the two scheduling rule optimization objectives in the same situation is getting bigger and bigger. Compared with the shortest distance, the fastest arrival has a shorter total completion time, which is more in line with the actual terminal operation scheduling. In addition, as the number of shore bridges increases, the operation time gap between single-load AGV mode and multi-load AGV mode is proportional to the number of shore bridges. Obviously the APSO-C algorithm has better performance in the container scheduling optimization process, which is more in line with the actual operation requirements of the terminal.

Based on the full active suspension and road input model, this paper introduces the fuzzy control theory and genetic algorithm design theory, adopts the fuzzy control method to control the actuator’s actuation force, creates the fuzzy control system of the automobile active suspension system, and optimizes the fuzzy control rules by using the improved genetic algorithm to ultimately realize the vibration damping effect enhancement in the process of driving the automobile vehicle. Simulation experiments and sample vehicle road experiments are used to verify the performance and utility of the fuzzy controller based on the improved genetic algorithm proposed in this paper. In the simulation experiments carried out with the help of Matlab/Simulink software, the control active suspension body controlled by the fuzzy controller based on the improved genetic algorithm reduces the root mean square value of angular acceleration of pendulum vibration, pitching rotation and lateral tilting rotation by 58.93%, 52.31% and 57.74%, respectively, compared with that of the conventional controller, the root mean square value of the dynamic deflection of the suspension is reduced, and the vehicle driving performance shows good stability and stability. The vehicle traveling shows good smoothness and stability. In the prototype road test, the root mean square value of the corresponding acceleration of the fuzzy-controlled active suspension optimized based on the improved genetic algorithm in this paper is reduced by 42.67%, 39.45% and 37.23%, respectively, compared with that of the passive suspension. Overall, the optimized design of fuzzy controller based on genetic algorithm proposed in this paper greatly improves the vibration damping effect of the active suspension system.

Semantic accuracy plays an important role in improving the quality of English translation teaching. This paper proposes a semantic translation model based on convolutional neural network. It is based on the semantic correlation expression and the statistical machine translation model of hierarchical phrases, and combines the convolutional neural network to propose a translation model optimization method that integrates sentence and document information. The method evaluates the semantic match between source language phrases and candidate target phrases by utilizing the sentence context of the source language phrases and the topic information of the documents in which they are located. The optimization method for evaluating the accuracy of English semantic translation is also given. In the simulated translation experiments, the accuracy of the translation correctness evaluation of this method is maintained at 92.5% and above, with high semantic accuracy. The research constructs a high and stable English semantic translation model, which provides informative aids for English translation teaching.

As an important part of Chinese traditional culture and art, how to efficiently realize the recognition, retrieval and style appreciation of calligraphy is of great significance. Aiming at the shortcomings of the traditional geometric feature recognition model with low recognition efficiency, this paper applies morphological neural network to the geometric feature recognition of calligraphy to design a geometric feature recognition model for calligraphy. Image enhancement is performed on the calligraphic graphics, the expansion pooling subnet is designed to replace the maximum pooling layer, and the calligraphic geometric feature recognition network is constructed by combining the residual block structure. The average recognition accuracy of this model in the geometric feature refinement recognition task is as high as 97.23%, which is higher than that of the comparative models such as CNN, LeNet-5, and the recognition accuracies are not less than 96% for the Euclidean, Liu, Zhao, and Yan styles. Using the model of this paper to explore the influence of calligraphic line fluidity and structural changes on the geometric features, it is analyzed that the “line” has a more significant influence on the geometric features of calligraphy than the “structure”. In the six types of traditional calligraphy, such as large seal, small seal, official script, regular script, line script, and cursive script, cursive script is only similar to the geometric characteristics of line script, and the geometric characteristics are very unique.