Lightweight segmentation method and system for unmanned aerial vehicle road slope crack and medium

The invention belongs to the technical field of image processing, and particularly relates to an unmanned aerial vehicle-oriented road slope crack lightweight segmentation method and system and a medium, and the method comprises the steps: firstly collecting road slope real-time video data through an unmanned aerial vehicle; then, the collected real-time video data are preprocessed, and the preprocessed real-time video data are generated; constructing a multi-scale deformable convolution crack semantic segmentation model, obtaining preprocessed real-time video sample data of the slope, and inputting the data into the multi-scale deformable convolution crack semantic segmentation model for training; and finally, transmitting the preprocessed real-time video data to the trained multi-scale deformable convolution crack semantic segmentation model, outputting a segmentation result, and identifying crack information existing in the slope according to the segmentation result. According to the invention, the problem