Abstract: The traditional camera source identification method using sensor pattern noise has a low degree of automation. The deep learning method is directly used for identification, and the feature extraction stage is more inclined to learn the features related to the image content, rather than the camera "fingerprint" feature. In order to solve the above problems, this paper designs a digital image camera source identification architecture based on the Siamese network, and the two network branches share the weight. A deep residual network with an attention mechanism is used for feature extraction to suppress the influence of image content on the entire camera source identification task. The comparative experiment results show that the method proposed in this paper improves the accuracy of camera source identification to a certain extent.