Video anomaly detection algorithm based on multi-scale probability generation

A video anomaly detection algorithm based on multi-scale probability generation is provided, a model takes continuous eight frames of images as input, multi-scale image feature maps are extracted through an encoder, the feature maps of different scales are subjected to parallel fusion of time sequence features through a sequence module, then position coding is carried out, then the feature vectors are expanded into feature vectors, and the feature vectors are subjected to feature extraction; the feature vectors pass through a decoder and input disturbance to obtain fitting feature vectors, the fitting feature vectors are combined into a feature graph, the feature graph is input into a detection head, an abnormal probability graph is obtained, multi-scale feature graph design is adopted, the recognition capability of abnormal samples is improved, the model can fully learn global and local features of normal samples, and the detection accuracy is improved. And the perception capability of the model on the spati