基于耦合模型的古建筑屋顶破损检测研究OA

To address the issues of missed and false detections in large-scale remote sensing images and dense scenes during the inspection of roof damage in ancient buildings,this paper proposes a Mask R-CNN YOLO v8 Pro hybrid detection model.The model integrates Mask R-CNN with an improved YOLO v8 by inputting the ancient building images extracted by Mask R-CNN into YOLO v8,and incorporating a large-scale detection head and shuffle attention(SA)mechanism in the YOLO v8 network to enhance the model's ability to extract small target features and focus on key information.The model was trained and tested on a self-made dataset with comparative experiments.The results show that for detecting four types of roof damage(holes,root erosion,debris,and displaced tiles),Mask R-CNN YOLO v8 Pro achieved an mAP@0.5 of 64.9％,outperforming conventional single detection models.Compared with other coupled models(Mask R-CNN YOLO v5,Mask R-CNN YOLO v8),mAP@0.5 increased by 13.3％and 12.1％,respectively.The study demonstrates that Mask R-CNN YOLO v8 Pro can more effectively extract feature information of small targets and focus on regions of interest in images,thereby reducing instances of missed and false detections.