UAV Image Positioning And Change Detection Method Based On Satellite Remote Sensing Image

Remote sensing change detection is to detect the changed targets from two remote sensing images acquired at different times.Both satellites and UAV platforms provide amount of remote sensing image data.However,in most cases,there is no historical archive data obtained by UAV for specific research areas.Thus,in these cases,change detection can only be applied on the current UAV images and historical archived remote sensing images.The previous researches has the following two limitations.Firstly,the big difference of resolution between UAV images and satellite remote sensing images,results in low matching accuracy with previous algorithms.Secondly,amount of model parameters are required with the known change detection algorithms based on deep learning and attention mechanisms.In order to avoid the above problems,this thesis has carried out of the researches on UAV image positioning and change detection methods based on satellite remote sensing images.The main content includes the following three sections.(1)The simulation analysis of the on-orbit remote sensing satellites has been realized and the corresponding software system has been developed.The thesis completed the calculation of the sub-satellite point trajectory,the calculation of the coverage area and the analysis of the satellite transit.Then,the data list of the transit satellite over the target area with the set of time period can be analyzed.The simulation experiments analyzed the transit time of Gaofen-1,2 and 3 satellites from the realized model.The results proved the realized model achieved the similar results compared with the results from commercial software STK,which indicating the efficiency of the model.(2)A UAV image matching and positioning algorithm based on similarity approximation is proposed.Big differences of spatial resolution between UAV images and satellite images results in low matching accuracy with the traditional algorithms.Then,in order to avoid the problem,the thesis proposed a method based on similarity approximation.The method applies Gaussian filtering on UAV images before applying the current matching algorithms,which improves the similarity between UAV images and satellite images,and ultimately improves the accuracy of the matching algorithm.(3)A remote sensing image change detection method based on siamese convolution network and lightweight dual attention module is proposed.The proposed lightweight dual attention module has the advantages of higher detection accuracy with fewer parameters.The method has been test on both the CDD dataset and the OSCD dataset.It has been performed well and the overall detection accuracy reached 98.44% and95.18%,respectively.Then,in order to further reducing the parameters,the model was pruned using the network slimming method.With the model applied,the model performance only slightly decreased(the F1 score decreased by 0.68%)with 66% of model parameters reducing.Moreover,a change detection method combining the fullscale connection and the SMRNC＿LDA model was proposed to further improve the change detection accuracy.The overall detection accuracy of this method on the CDD dataset and LEVIR-CD dataset reached 98.98% and 99.25%,respectively.