Improved YOLOv7-tiny method for hazardous chemical substance vehicle detection

The invention discloses an improved YOLOv7-tiny method for hazardous chemical substance vehicle detection, which comprises the following steps: adding global response normalization into a traditional convolution module, placing the traditional convolution module behind a SiLU activation function to enable a model to pay attention to hazardous chemical substance vehicles at any position in a picture or a video, designing a fast spatial pyramid pooling module with enhanced channel features, and carrying out fast spatial pyramid pooling on the channel features. After three key pooling operations in a spatial pyramid pooling module are changed from a parallel structure to a serial node, the size of a pooling kernel is reduced to reduce the time complexity, and a compression excitation module is connected in a residual mode, so that relatively useless features are inhibited, important features are more concerned, a large-target path aggregation network is designed, and the path aggregation efficiency is improved.