In this paper, the simpler wavelet transform method is used to remove the Gaussian noise contained in the image, reconstruct the approximation signal and the final processed detail signal, and obtain the estimated value of the noise signal to complete the image preprocessing. In the process of human static feature extraction, due to the structural limitations of the human body shoulder joint activity angle is low. Therefore, the local spherical coordinates are used to represent the joint angle characteristics of the human body in the static posture and, through the completion of the calculation of the distribution function of the action features of the target, to be tested to realize the recognition of the martial arts action. After completing the judgment matrix and consistency test, we finally establish the teaching effect index system for college wushu courses and analyze the teaching of ‘Internet Education’ wushu courses. The results show that the accuracy of the results of the four algorithms test dataset is stably maintained between 0.8 and 0.9, DNN algorithm, CNN algorithm, RNN algorithm, this paper’s algorithm loss average value of 1.34, 1.58, 1.75, 1.23, respectively, the combined accuracy and loss value, this paper’s algorithm compared to DNN algorithm, CNN algorithm, RNN algorithm performs better. Namely, the algorithm in this paper can accomplish action recognition with high accuracy. In the analysis of the teaching effect index evaluation of wushu course, the experts think that the matter dimension wushu course design index weight (0.800) > wushu teaching system (0.200), and wushu lecturers should design the course key points before the class. In this study, through the digital construction of teaching wushu courses in colleges and universities, we can achieve long-term development.