To address the issues of insufficient background noise suppression and difficulty in effectively focusing on key features in existing attention mechanisms due to imbalanced channel feature responses and lack of differentiated modulation of spatial features, this paper proposes a Visual Inhibition Attention Network for Image Classification (VIANet) inspired by the “central excitation-peripheral inhibition” mechanism in the biological visual system. The key innovation of this network lies in the construction of a visual inhibition attention module. In the channel dimension, this module applies moderate inhibition to high-response channels, enabling the network to extract complementary information from more channels and promoting a balanced distribution of channel responses, thereby enhancing the discriminability and robustness of channel representations. In the spatial dimension, by constructing a dynamic elliptical perception field model, inhibition is applied to peripheral areas based on spatial distance, effectively suppressing background noise while maintaining high resolution of central area features, and enhancing the modeling ability for features in target regions of interest. This module achieves the enhancement of important features and the suppression of redundant information through joint inhibition modulation of channels and space. Experimental results on multiple datasets, including CIFAR-10, CIFAR-100, SVHN, Imagenette, and Imagewoof, show that the proposed method outperforms existing mainstream methods in terms of classification accuracy.