Leakage detection is a critical measure to ensure the proper operation of hazardous waste landfills. Traditional manual visual inspections are inefficient, and hazardous waste can affect the health of workers.A method based on instance segmentation is proposed to detect leakage. The method uses a camera mounted on an inspection robot to capture video images, the YOLOSeg-EViT network performs instance segmentation. Experimental results show that the algorithm achieves a mean average precision of 75.7%, and when detecting with a 1 280×720 pixel resolution video, the detection is 54.4 frames per second. On an edge computing platform, the algorithm reaches mean average precision 75.1% and 33.1 frames per second. The leakage detection algorithm effectively identifies leakage, improves the detection efficiency, and prevents workers from being exposed to hazardous substances, making it highly practical.