Abstract: To address the problem that existing deep learning image tampering detection models can hardly make use of the tampering trace features extracted from the shallow layer of the network, resulting in poor detection and low localization accuracy, an image tampering detection algorithm based on the improved Ringed Residual U-Net (RRU-Net) is proposed. Firstly, a hierarchical supervision strategy is used to design a tampering fusion localization module to output the model in layers, so that the deep and shallow feature information is fully fused and the sensitivity of the model to the texture and edge information of the shallow layers is improved. Then, the loss function is improved by adding the parameter β to the binary cross-entropy, and then the total loss is measured by the loss of the different layers. Finally, group normalization is used in the model to speed up the convergence of the network while avoiding overfitting. The experimental results on the CSAIA and Columbia datasets showed an increase in F1 values of 0.08 and 0.072 compared to RRU-Net, respectively. It shows that the algorithm has high detection accuracy and can effectively locate tampered areas.