Crowd counting method based on spatial pyramid attention network

The invention relates to a crowd counting method based on a spatial pyramid attention network, and the method comprises the following steps: obtaining a crowd image data set, carrying out the preprocessing of a crowd image through employing a Gaussian kernel, and generating a ground real density map; a crowded scene recognition algorithm CSRNet is selected as a basic structure, and a spatial pyramid attention network is constructed; inputting the acquired crowd images with different resolutions in the crowd counting data set and the corresponding crowd density map into a spatial pyramid attention network for training verification, outputting a corresponding estimated density map, and calculating the loss between the estimated density map and a ground real crowd density map input into the network to obtain an optimized spatial pyramid attention network; and storing the trained network weight, testing by using the test sample set, and judging the network effect. According to the method, the problems of severe s