Video action classification system and method based on hierarchical power parsing and coding

The invention relates to a video action classification system and method based on hierarchical power parsing and coding. The video action classification system comprises an apparent feature extractionmodule, a power parsing module, a hierarchical power coding module and a classification module, wherein the apparent feature extraction module extracts apparent features from each frame or adjacent several frames of the video and expresses the video into an apparent feature sequence or a low-dimensional state sequence; the power parsing module performs time sequence clustering on the apparent feature sequence or the low-dimensional state sequence, and divides the sequence into a plurality of stages in an unsupervised manner to obtain a stage parsing path; the hierarchical power coding moduleestablishes a two-layer power coding structure and performs coding on apparent features of image frames in each stage by using average pooling or sorting pooling in the first layer so as to obtain a first coding sequence, and perform coding on the coding sequence again by using sorting pooling in the second layer so as to obtain an overall video expression with fixed dimensions; and the classification module performs classification on the overall video expression by using a support vector machine (SVM).