Neural network behavior recognition method adopting multi-dimensional correlation attention model

The invention discloses a neural network behavior recognition method adopting a multi-dimensional correlation attention model, and belongs to the technical field of computer vision, deep learning and behavior recognition. The method comprises the following steps: firstly, carrying out automatic feature extraction on an RGB image sequence or an RGB image and an optical flow sequence extracted from an action video by utilizing a three-dimensional convolutional neural network; and then automatically extracting multi-dimensional attention on the basis of the feature map by using a multi-dimensional correlation attention model, and performing weighted fusion on the generated feature map by using the extracted salient regions on time, space and feature channels to obtain a predicted value, and generating a final behavior recognition result. The multi-dimensional correlation among the three dimensions of time, space and feature channels is further explored, and the significance on the time, space and feature channel