Remote sensing image change detection is one of the research focuses in the field of remote sensing. At present, most of them are deep learning methods, which mainly use single channel or siamese network to extract features, which can effectively extract change features. However, with the increasing resolution of remote sensing images, the single feature extraction method is susceptible to the influence of irrelevant details, which leads to the insufficient segmentation ability of the changed and unchanged regions in the detection results. Therefore, this paper proposed a fire-new triple-channel regionenhancement change detection network to enhance feature extraction capability from multiple perspectives. Firstly, the method constructed a triple-channel region-enhancement encoder, and used three feature extraction channels to extract similarity information, comprehensiveness information and difference information in a directional manner. At every scales of the encoder, region-separation enhancement modules were able to augment channel 2 with channels 1 and 3, which was beneficial to obtain better effect of changing region segmentation. Secondly, it designed a layer interaction-guidance fusion decoder, and used the interactive guidance between higher-level and lower-level features. So the decoder fused effectively the different-level features by the mutual guidance between high-level features and low-level features. Finally, it used an adaptive weight based on information entropy, which gave more attention to high entropy regions, to optimize loss function. Then, the new loss function improved the training process of the network model. The results of experiment on common datasets show that this network achieves better detection accuracy than the contrast networks.