Improved road vehicle classification network based on RepVGG-A0
Aiming at the problem that it is difficult to balance the detection accuracy and real-time in the current vehicle type recognition process,an improved road vehicle type recognition network based on RepVGG-A0 is pro-posed,which uses the idea of structural re-parameterization to fuse the multi-branch network to improve the network reasoning speed.The mixed void convolution is used to replace the traditional convolution,which strengthens the rec-ognition ability of the model for large targets.Integrating the residual structure coordinate attention(RES-CA)module into the network backbone improves the network's ability to extract effective feature information,and avoids the impact of gradient disappearance and gradient degradation.In addition,the label smoothing regularization method is used to improve the loss function,reduce the impact of model overfitting on the detection results,and improve the generaliza-tion of the model.After verification,the recognition accuracy of the method in this paper on the road vehicle data set BIT-Vehicle has reached 97.17％,which is 2.67％higher than the original model,and is superior to the existing ResNet-18,VGG and other network models,while ensuring the detection speed of the model.
model identificationstructural re-parameterizationresidual structuremixed void convolutionla-bel smoothing regularization
万方数据
维普
