Teaching and correcting athletes’ techniques by analyzing and referring to the performance of professional tournament players can improve the teaching level and quality of wushu movements. In this paper, the performance of college students in UFC tournaments is taken as the research data, and the multilayer perceptron algorithm is used to process the images and carry out the global modeling of wushu fighting action images. The network coding design is used to improve the data transmission rate of the algorithm, and the activation function is used as the nonlinear expression method of the algorithm. The Tanh_Softsign activation function is improved to counteract the noise interference of the dataset images, in order to construct the multilayer perceptual machine algorithm and develop the learning of martial arts fighting action scores. After optimizing the learning of UFC martial arts action scores by this algorithm, this algorithm shows a high correlation between the performance scores of students and the professional teachers’ scores of an elective class of martial arts in a university with P>0.05, which indicates that the algorithm in this paper can accurately assess the students’ action performance.