Whether the surface of the solar panel is intact or not plays a decisive role in its power generation efficiency. The traditional manual detection method, infrared penetration detection method and machine vision detection method all have their own shortcomings. Due to the difficulty of obtaining outdoor solar panel images, small number of samples, high sample similarity and other problems, most of the deep learning algorithms can not well complete the outdoor solar panel defect detection task. In view of the particularity of this task, a defect detection method based on the improved DenseNet network model is proposed. The DenseNet basic network model is selected, L2 regularization is added to the model, the Batch Normalization layer is adjusted to solve the over-fitting problem, and the activation function ReLU function with the SELU function is replaced, which can better alleviate the problem of gradient disappearance and strengthen the robustness of the network. In the final experiment, the accuracy of the training set is 93%, and the accuracy of the test set is 87%. It can effectively detect and distinguish different degrees of damage to the battery board.