Training method of lightweight pyramid cavity volume accumulation network for crowd counting

The invention provides a training method of a lightweight pyramid cavity volume aggregation network for crowd counting, which comprises the following steps of: inputting a picture, firstly extracting feature information through a backbone network, and then inputting the extracted feature map into a plurality of multi-scale context aggregation modules in a cascade form; the module firstly uses convolution kernels with different voidage to extract multi-scale information, and then adaptively selects channel context feature information through a channel attention mechanism and aggregates the channel context feature information. The feature map is converted into a feature map with higher resolution through up-sampling every time through a multi-scale context aggregation module, finally, an estimated density map is output through a 1 * 1 convolution kernel, and the predicted number of people is obtained through integral summation. According to the method provided by the invention, multi-scale information is effect