Personnel abnormal behavior identification system and method based on TSAFormer model

The invention designs a personnel abnormal behavior identification system and method based on a TSAFormer model. The system is composed of a data processing module, L identical TSAFormer blocks and an abnormal behavior classification module. Firstly, a data processing module extracts image frames from a monitoring video stream and converts the image frames into image block embedded representation, a behavior sequence modeling module performs time sequence displacement on the image blocks along a channel dimension, the operation of exchanging partial content with a previous frame timestamp is realized, and spatio-temporal information is obtained; then, the space-time information is sent to a space-time feature extraction module, and space-time features are captured by freezing a pre-trained image encoder vision transfer and combining a lightweight adapter; secondly, nonlinear transformation is carried out on the features, feature fusion is carried out by applying an attention mechanism, and interaction feature