ICD: VHR-Oriented Interactive Change-Detection Algorithm

In recent years, deep learning has become the mainstream development direction in change-detection field, and its accuracy speed have also reached a high level. However, method based on cannot predict all change areas accurately, application is limited due to local prediction defects. For this reason, we propose an interactive network (ICD) for very resolution (VHR) convolution neural network. The integrates positive- negative-click information distance layer of network, users can correct defects by adding clicks. We carried out experiments open source dataset WHU LEVIR-CD. By clicks, their F1-scores reach 0.920 0.912, respectively, which are 4.3% 4.2% higher than original To better evaluate correction ability set evaluation indices—click-correction ranges, suitable evaluating carry above models. results show that clicks effectively improve result accuracy.