In recent years, deep learning has become the mainstream development direction in the field of change detection, and its accuracy and speed have also reached high levels. However, a change detection method based on deep learning cannot accurately predict all change areas, and its applicability is limited due to local prediction defects. For this reason, we propose an interactive change detection (ICD) network for very-high-resolution (VHR) remote sensing images based on a deep convolutional neural network. Positive and negative click information is integrated in the distance layer of the change detection network, and users can correct prediction defects by adding clicks. We carry out experiments on the open source WHU dataset and the homemade GTMAP-ICD dataset. By adding clicks, the F1 coefficients achieved on these dataset can reach 0.833 and 0.819, respectively, which are 5.8% and 5.3% higher than those of the original network. To better evaluate the correction ability of clicks, we propose an evaluation index, the click correction range, which is suitable for the evaluation of clicks, and we carry out experiments on the above models. The results show that the method of adding clicks can effectively correct prediction defects and improve the accuracy of the obtained results.