双频通道差异增强的图像分类网络OA

Aiming at the problem that image feature differentiation is low in image classification networks,which in turn reduces feature expression ability,a dual-frequency channel difference enhancement for image classification(DCDENet)is proposed.The network is based on the ResNet-34 residual network.Firstly,a custom difference enhancement convolu-tion(CDEC)module is proposed to enhance features with high response value,suppress features with low response value,enhance the difference between high frequency features and low frequency features,and improve the expression ability of features.Secondly,the channel separation and reconstruction(CSR)module is proposed,which integrates the custom difference-enhanced convolution and channel reconstruction convolution,and uses the separation-transformation-difference-augmented-fusion strategy to improve the network's ability to extract key features and reduce the impact of redundant features on net-work information transmission.Finally,the CSR module is embedded into the residual block to improve the stability and convergence speed of training,enhance the nonlinear representation ability of features,and improve the classification ability of the network.The proposed method achieves 96.66%,80.08%and 97.55%classification accuracy on CIFAR-10,CIFAR-100 and SVHN data sets,respectively,with an average increase of 3.56,4.77 and 4.24 percentage points compared with the current advanced methods.Compared with the existing mainstream models,this network can effectively improve the expression ability of features and the ability to extract key features,reduce the redundant features in the channel,enhance the nonlinear representation ability of features,and effectively improve the classification ability of the model.