Implementation and Rehabilitation Application of Sports Medical Deep Learning Model Driven by Big Data

A large number of unlabeled and limited style data greatly reduces the reuse possibility of existing motion sequences. Effective classification and fragment splicing have become an important way of data reuse. Aiming at these two problems, this paper focuses on the great success of deep learning in the field of graphics and iconography. Based on the theory of Restricted Boltzmann Machine (RBM), a spatio-temporal feature extraction model for human skeleton medical motion sequences is established. The research results are mainly manifested in three aspects. (1) In this paper, stack factor decomposition spatiotemporal feature model and discriminate RBM are used to construct semi-supervised combination model. (2) The underlying model firstly uses the idea of weight decomposition to construct the three channel generative RBM model; and then it extracts the abstract temporal and spatial characteristics of the original motion sequence. Furthermore, it identifies the behavior style of the current input segment at the top using the discriminate RBM model. Finally conduct the stylistic statistics of the whole motion sequence in the voting space. (3) An unsupervised similar frame detection model is constructed by using 3D convolution RBM's perception of human adjacent joints' linkage. In this way, Candidate frames for constructing graph model nodes are obtained. Trajectory and style switching control is realized based on attitude similarity screening criteria. The simulation experiment verifies the superiority and reliability of the algorithm, and it is effectively applied in rehabilitation training.