In order to improve the segmentation effect and performance of noisy images, a noisy image segmentation method based on multi-scale channel attention is proposed. A gray model of the noise image is constructed by taking a pixel point of the image as a unit. The bias field theory is introduced to construct a local segmentation model for noisy images. The global energy function of the noise image is constructed based on the Heaviside step function, and the gray offset field model of the noise image is obtained. The residual connection is added to the multi-scale channel attention module, and the three channels of the noise image features are pooled and fused to extract the features of different scales of the noise image and realize the noise image segmentation. Experimental results show that the method can preserve the edge features of noisy images and improve the accuracy and signal-to-noise ratio of image segmentation.