Special vehicle detection method based on improved YOLOv7-tiny

The invention relates to the field of computer vision, and discloses a special vehicle detection method based on improved YOLOv7-tiny. According to the method, different numbers of ECA-Net attention modules are introduced into a backbone network and a neck network of a YOLOv7-tiny model respectively, so that redundant information is restrained, the feature extraction capacity of the model is enhanced, an ELAN module in the neck network of the original model is replaced by using ConvNext Block, and the feature extraction capacity of the model is improved. In this way, the receptive field is expanded, the utilization rate of the model for target core features is improved, a Focal EIoU Loss positioning loss function with more powerful functionality is adopted to accelerate prediction frame convergence, and the regression precision of the prediction frame is improved. Experiments on a self-made special vehicle data set show that the accuracy of the method is very considerable, the conditions of leak detection and