Research On Image Registration Of Small Unmanned Aerial Vehicle Based Monitoring Cultivated Land Changes In The Hills And Mountains

Numerous different ideas,models and change detection methods have been proposed for monitoring cultivated land changes in mountainous terrain of southwest and southeast China.However,how to design an efficient land cover change monitoring or detection system that ensures a high detection rate remains a critical and challenging step.Note that the unchanged and changed structures in the difference image will directly affect the quality of the change results.In addition,the images acquired by small unmanned aerial vehicle?UAV?can contain scale changes,noises,geometric distortions and illumination changes at different temporal scales.To address aforementioned problems,a robust set of change detection framework for monitoring land cover change in mountainous terrain is proposed by using multi-temporal remote sensing images,which contains the following contributions.?i?Dual feature descriptor?DFD?is defined for measuring global and local discrepancies between two data point sets,and a deterministic annealing scheme is employed to control the balance of the DFD.?ii?Multi-scale feature descriptor constructed by layer formed applying pre-trained VGG?Visual Geometry Group?and neighbouring structure descriptor?NSD?.?iii?Dynamic inlier selection:at the early stage of registration,the rough transformation is quickly determined by the most reliable feature points.After which the registration details are optimized by increasing the number of feature points.?iv?The double constraints for L₂-minimizing estimate?₂L E?based energy optimization is formulated to calculate a reasonable position in the reproducing kernel Hilbert space.?v?Fuzzy C-Means?FCM?classifier is adopted to generate a similarity matrix between image pair of geometric correction process,and a robust change map is produced through feature similarity analysis.The study was carried out in the five land conservation regions of Sichuan,Guizhou,Yunnan,Guangxi,and Guangdong.These regions have a variety of land cover types including cropland?Slope gradients range is 5°³5°?,building-up,forest,etc.The data set contains a total of 500 image pairs that are acquired by small UAV.Extensive experiments on data set are conducted.Experimental results show that the proposed method provides better performances in most cases after comparing with five state-of-the-art image registration methods and four state-of-the-art change detection methods.