To overcome the shortcomings of existing methods in the segmentation of solar cell images, an improved U-Net structure for defect segmentation of solar cell images is proposed. First, the dense connection structure is introduced to alleviate the problem of gradient disappearance and make defect extraction more fully; at the same time, a batch normalization layer and Relu layer are added after each convolution layer to prevent the loss of defect details; then a dual attention mechanism is introduced. To enhance target features and suppress irrelevant features,improving the overall detection accuracy of the model. Finally, two different networks are used to be compared with the method in this paper. The experimental results show that the network can obtain more detailed feature information, which further improves the accuracy of solar cell image segmentation.