Research On Recognition Of Damaged Road After Earthquake Based On UAV Image

China's natural disasters have occurred frequently,and the Wenchuan earthquake in 2008 brought about a terrible loss.Reconstruction of road information that is in urgent need of damage after the disaster.However,direct field research is not only dangerous but also inefficient.Remote sensing technology has become an important means of post-earthquake rescue and assessment,especially for low-altitude remote sensing drone systems,because of its timeliness,synchronization,and ground conditions.The emergency response speed is fast and the operation ability is strong.The obtained drone image can timely reflect the damage information after the disaster.Therefore,this paper studies the damage road identification method based on UAV image.The research work and results are as follows:(1)The data set of damaged roads is established in this study.At present,there is no public data set of damaged roads.This paper collects and collates the postearthquake UAV images.Because there are few damaged road images,the data enhancement method is studied,and the data set of damaged roads in VOC2007 format is produced for training of Faster RCNN target detection network.(2)A damage detection method based on the target detection algorithm Faster RCNN is proposed.A corrupted road image dataset in VOC2007 format was produced,and the network was trained using the dataset.Through the trained model,the detected images are predicted.The experimental results show that the damage detection method based on Faster RCNN is feasible,the average recall rate can reach 63.9%,the accuracy rate can reach 100%,and the number of false detections is 0,but the missed detection The situation is more serious.Using the trained model to detect the detected image can achieve real-time detection of the damaged road.(3)An object-oriented damage road recognition method based on road vector file is proposed.The segmentation of the existing road vector file based on the existing road vector file will be generated for further processing.Segmentation uses chessboard segmentation and multi-scale segmentation to create band ratios based on damaged road image features,and uses Assign algorithm and fuzzy classification algorithm to extract uncorrupted roads,vegetation,and damaged roads.The result analysis shows that the above method can effectively identify the road information of high-resolution UAV image damage.