Analysis and Correction of Wrong Technical Actions in Juvenile Sports Training Based on Deep Learning

Abstract

Scientific analysis of students’ incorrect actions in class, as well as timely and effective correction, is frequently an important link in the PE process. At the same time, it is an important symbol for assessing a teacher’s teaching level and quality. In this paper, the analysis and correction model of sports wrong technical movements is built using DCNN to address shortcomings in the process of detecting wrong movements in PE and training. This article is based on CNN and has been enhanced by DL. The model learns both manual and DL features; the manual features use an improved dense trajectory, the DL features use CNN based on motion information, and the generalization ability of the kernel support vector machine is used to fuse the two. The simulation results show that the accuracy of the wrong action judgment of this method can reach 92.16 percent, which is 4.6 percent higher than the method of combining NN with region prediction and 5.7 percent higher than the method of detecting image matching score. This method can accurately describe the characteristics of human motion and identify incorrect movements, improve the ability of judging and correcting incorrect movements in sports training, and help athletes improve their sports level.