Research On Medical Action Video Classification Technology Based On Deep Learning

In recent years,with the development of computer technology,artificial intelligence technology is changing with each passing day.Combining artificial intelligence technology with modern medicine and diagnosing diseases by corresponding technical means has become the development trend in the field of computer science.Developmental coordination disorder in children is a kind of disease that occurs in the growth stage of children.It is mainly manifested in children’s uncoordinated movements.Clinicians mainly make diagnosis by observing some movements of children,for screening work efficiency is lower.The main purpose of this paper is to extract the human body feature information from video images and classify the coordination of movements through deep learning technology.In terms of spatial information extraction and spatial and temporal feature information fusion of human body,two different deep learning network structures are used respectively,the spatial features are extracted by CNN,and the spatial and temporal features are fusion by GCN.Through in-depth mining of hidden features in the video data of medical actions,classifying the specific actions and obtaining the positive and abnormal conditions of the actions,it is beneficial to meet the needs of doctors for large-scale screening to a certain extent.The main work is as follows:1.When doctors watch the movements of children,they often pay attention to the key movements.In this paper,a new key frame extraction algorithm,multi-frame dimensionality-reduction difference method,is used to extract the key frames of each movement.2.Improve the fusion of VGG network structure and pose machine network,detect and extract human bone points in the video data,so as to obtain the spatial feature information of the video data.3.The temporal and spatial information of the medical action video data were fused based on the spatiotemporal graph convolutional network,and the action classification results were obtained through the classifier,which were compared with the conventional video classification methods to determine the selection of the video classification method.In this paper,an improved spatiotemporal graph neural network(CRM-ST-GCN)is proposed to improve the accuracy of the model for medical action video classification.4.In terms of visualization system design,Node.js is used as the development language to realize the visualization of medical action video classification based on Web,and the results of action video classification are displayed on the browser side.