A Research On The Key Technology Of Quickly Detecting Suspicious Changed Land-use Based On UAV Images

The most stringent land management system is executed in our country and it is of great importance to update the land-use information timely,accurately and objectively.For the traditional UAV land monitoring,it is necessary to control,correct and stitch,which takes lots of time and efforts and is inefficient.This paper is devoted to the research on the rapid discovery key technology of detecting suspicious changed land-use without stitching UAV images.The main contents are as follows:(1)A large-scale image feature extraction algorithm based on overlapping meshing is proposed.The large-scale imgae of the monitoring area is divided into tiles with a small overlapped sub-area based on the automatic overlapping meshing,the keypoints are adjusted based on the relationship between each tile and image,and the lossless extraction of keypoints is realized.The database of massive large-scale image keypoints is constructed based on R*Tree index,which lays a foundation for the fast matching of the keypoints between the new and old temporal image.(2)A mapping model and an algorithm of fast matching among the keypoints are researched.The mapping model from the keypoints of the new temporal images to the keypoint database of the large-scale old temporal image is established based on the UAV POS information.The fast matching between the new temporal images and the old is achieved based on R*Tree index,Kd-Tree model,and RANSAC algorithm.The feasibility of the algorithm is proved by visualizing the experimental results,which lays a foundation for the rapid discovery of the new temporal images with suspicious changed land-use.(3)An algorithm for the rapid discovery of suspicious changed land-use with merging keypoint matching,image registration and image fingerprinting is proposed.Firstly,the new temporal images without obvious change are excluded based on the keypoint matching ratio.Secondly,an image registration algorithm is used to confirm that there is a new temporal image with large area change,partition the new temporal image with successful registration,and registrate every partition with the old temporal reading after successful registration,for the failed registration partition,analyzing the reason to determine whether the partition is changed or suspicious.Finally,determine the successfully registrated partitions whether there is a change in the sub-image or there may be changed by the Hamming Distance,which is used to measure the similarity of image fingerprint.