Few-Shot Change Detection: A General Class-Independent Change Detection Algorithm

In order to solve the problem of detecting single land cover type change under the scenario of multiple land cover type change and the difficulty of labeling large-scale samples, we build the Single SEmantic Change detectiON Dataset (SSECOND) and propose Few-Shot Change Detection (FSCD) network to fulfill the single land cover type change task, and verify the effectiveness of FSCD model on our SSECOND dataset. The goal of the FSCD model is to generate binary masks for a new class of change detection with only a very small number of annotated images. In order to achieve this goal, we propose a two-branch model, one branch is used to extract the features of the changed part of support images, and the other branch is used to extract the features of a pair of images to be detected. By comparing the features to be detected with the features of the changed part, we can judge whether the pixel has changed. We introduce few-shot learning into the field of change detection for the first time, and verify the effectiveness of our proposed model through few-shot experiments on SSECOND dataset. The source code of the proposed FSCD is available at https://github.com/CaoZhijuan-CD/FSCD.