The extraction of rivers in cold and arid regions is of great significance to the rational utilization of water resources, water conservancy planning and early warning of water disasters. In order to solve the problem of refined river extraction from remote sensing images, a segmentation network (AFR-LinkNet network) was proposed based on the LinkNet model. AFR-LinkNet introduced residual channel attention structure, asymmetric convolution module and dense skip connection structure on the basis of LinkNet, and replaced the original ReLU activation function with visual activation function FReLU. The residual channel attention structure can strengthen the features that were effective for segmentation tasks to improve the classification ability of the model and obtain more detailed information. The asymmetric convolution module was used to compress and accelerate the model. The FReLU activation function boosting network was used to extract fine spatial layout of rivers in remote sensing images. The experimental results on the river dataset in cold and arid regions showed that compared with FCN, UNet, ResNet50, LinkNet, DeepLabv3+ network, the intersection ratio of AFR-LinkNet network was improved by 26.4 percentage points, 22.7 percentage points, 17.6 percentage points, 12.0 percentage points and 9.7 percentage points respectively, the pixel accuracy was increased by 25.9 percentage points, 22.5 percentage points, 13.2 percentage points, 10.5 percentage points and 7.3 percentage points, respectively. After the introduction of the asymmetric convolution module, the intersection ratio was increased by 5.1 percentage points, and the pixel accuracy rate was increased by 2.9 percentage points. On this basis, after introducing the residual channel attention structure, the intersection ratio was improved by 2.2 percentage points, the pixel accuracy rate was improved by 2.3 percentage points, and its performance was better, and the extracted river coherence and details were better preserved. Therefore, AFR-LinkNet algorithm was of great and far-reaching significance for analyzing river distribution, water disaster warning, rational utilization of water resources and agricultural irrigation development in cold and arid regions of China, laying a foundation for the realization of sustainable development in China.