Gait recognition is a promising biometric technology with unique advantages. It recognizes individuals at a distance by analyzing the gait characteristics of humans while walking. Existing model-based approaches mainly focus on the temporal and spatial features of joints. However, when the motion amplitude of the joints is small, especially in frontal (0°) and dorsal (180°) views, they are often overlooked. In these views, joint motions typically become less pronounced and more difficult to distinguish, making it difficult to accurately extract key joint features and correlations. This oversight leads to the loss of basic joint features and their correlations with neighboring joints, which significantly impacts recognition accuracy. To address this issue, we propose a channel topology graph convolution method for gait recognition to build different topologies that capture node correlations. we utilize a global adjacency matrix as a shared topology and construct a channel-specific topology by computing correlations unique to each channel. In addition, we introduce a Spatial Channel Attention mechanism to enhance sensitivity to subtle joint movements and emphasize critical node-channel information. Through extensive experiments on two common datasets, CASIA-B and OUMVLP-Pose, the proposed model is demonstrated to have higher recognition accuracy and significant robustness, thus highlighting its effectiveness in capturing subtle joint motions and node correlations.