Deep Convolutional Network Design Based on YOLO Framework with Efficiency Enhancement Method in Target Detection Tasks

Compared with traditional target detection algorithms, deep learning-based target detection algorithms trained on rich sample data do not need to design features artificially, are better adapted to environmental changes, and the accuracy and efficiency of detection are dramatically improved. This paper relies on the deep convolutional neural network structure to construct the YOLOv5 target detection model. On the input side of the model, three data enhancement techniques, namely mosaic data enhancement, adaptive anchor frame and adaptive image scaling, are adopted respectively to improve the accuracy, generalization ability and detection speed of the model in the target detection task. Attention mechanism is introduced and YOLOv5 framework is improved to construct a new network model. For the efficiency of the target detection task, a loss function is added and a global average pooling operation is used for feature mapping to realize a fully convolutional network. Two widely used evaluation metrics are chosen to evaluate the target detection efficiency of the model. The experiments show that the MAP value of the improved YOLOv5n network model is 2.9979 percentage points higher than that of the original YOLOv5 model, and at the same time, the FPS is substantially improved by 31%. The time taken to complete 100 rounds of training is 20 min, which is 10 min shorter than the pre-improvement algorithm.