Skeleton action recognition method based on multi-center and multi-mode graph convolutional network

The invention discloses a skeleton action recognition method based on a multi-center and multi-mode graph convolutional network. The method comprises the following steps: step 1, obtaining skeleton data and carrying out data preprocessing and data enhancement; 2, taking the joint flow state of the skeleton data processed in the step 1 as first-order skeleton data; 3, processing the joint flow state to obtain second-order skeleton data, wherein the second-order skeleton data comprises a skeleton flow state; 4, applying a GPT-3 model, and taking the human body action recognition data as input to generate text description data of offline actions; step 5, constructing and training a multi-center multi-modal graph convolutional network model; and step 6, inputting and outputting the joint flow state data, the skeleton flow state data and the text description data into a final prediction result. According to the method, the network can conveniently identify and detect the object under extreme scale change, and atte