Combining Gated SENet with ResNet Backbone for Image Classification

Image classification is a highly popular direction in the field of artificial intelligence, and its achievements hold significant reference value across numerous domains. Models and architectures such as ResNet, U-Net, Transformer, among others, are widely employed in various fields. Therefore, an efficient and accurate model plays a crucial role in image classification. In this paper, we propose a novel model named Gate Convolution SENet ResNet (GCSR-Net). It enhances the SENet module by introducing a convolutional layer and a gating unit, integrating these improvements into the original ResNet. The proposed GCSR-Net incorporates convolutional fusion to enhance its global contextual awareness and employs a gating mechanism to dynamically adjust the strength of feature information. It improves the accuracy of image classification and accelerates the convergence speed. Experimental results demonstrate its impressive performance on the CIFAR-10 and CIFAR-100 datasets, achieving a top-1 accuracy of 95.90% on CIFAR-10 and 80.32% on CIFAR-100, respectively. Furthermore, GCRS-Net converges after only 100 epochs without pretraining, accelerating by 50% in convergence speed. Therefore, the model proposed in this paper possesses reliability and robustness.