Abstract: In response to the problems of false positives and false negatives in the current illegal building image recognition and image change detection, this paper proposes a method for detecting illegal building using UAV (Unmanned Aerial Vehicle) images, innovatively using the YCrCb color space to perform differential operations and thresholding on the three components separately, and combining image classification algorithms to exclude image changes caused by non-construction activities. Finally, by using structural similarity detection to evaluate the similarity of image change regions in new and old-time phases, the method identifies the image change regions caused by illegal building. Compared with the current mainstream non-single type violation building detection methods based on drone images, the detection time of 4032×3024 resolution image is reduced by 40.9% to 4.08 seconds, and the accuracy is improved by 15.3% to reach 86.3%, with no missed detections. The results show that this method can effectively and quickly identify areas where illegal building is underway, which is beneficial for city managers to discover and handle these illegal activities in a timely manner, thus improving the efficiency and response speed of urban management.

Key words: illegal building detection, UAV image comparison, YCrCb color space, classification filtering