Target detection model and method for pet standard behavior detection

The invention provides a target detection model and method for pet standard behavior detection. According to the scheme, a normalized attention module is optimized and improved, an attention mechanism is improved by using contribution factors of weights, scale factors of batch normalization are used, and importance of the weights is represented by using standard deviation; a global sparse attention mechanism module is optimized and improved, so that the global sparse attention mechanism module only calculates self-attention for a subset of the input Token and supports complete space interaction, the action range of the Token can be reduced, and meanwhile, the underlying information flow of modeling global and local contexts is also reserved; the self-attention overhead is further reduced through a sparse attention module, and better accuracy and delay balance are achieved; a cross attention module is optimized and improved, local information is captured by alternately applying attention inside the image Patch