Satellite Image Gudided Airborne Object Matching Methods

Airborne object matching guided by images captured using a remote sensing satellite is used in airborne object identification;airborne object images and the provided satellite reference image are used.In particular,when an object is in an unknown area,the unmanned aerial vehicle(UAV)first obtains the satellite reference image from the satellite and uses the object matching method to locate the object in the airborne object image based on the satellite reference image.Because airborne objects are acquired in complex environments that cannot resolve the three issues that arise due to operational demands,robust and accurate object matching guided by satellite images is still challenging.First,because the satellite reference and airborne object images are acquired from different views,the issue of view difference arises.Second,the satellite reference and airborne object images are acquired at different heights;this results in a large-scale morphological difference.Third,because the satellite reference and the airborne object images are taken at different times,variations in the ground cover exist.Research has shown that point-,set-,and domain-matching methods are advantageous in solving the aforementioned issues.In object matching tasks,these methods can be used to address the problems associated with the differences in view,large-scale morphology,and ground cover.However,in practical application,the performance of the point matching method decreases in heterogeneous object matching tasks.Therefore,the set matching method is ineffective in complex airborne environments,and achieving an effective region division in the presence of several interferences and noise is challenging when using the domain matching method.These methods therefore have room for improvement in terms of adaptability.In this regard,this study examines the matching of airborne objects guided by satellite images from the following three aspects: point matching,set matching,and region matching.A point set region three-level fusion object matching method is developed on this foundation to address the three aforementioned issues.Further,to address the issue arising from the differences in the gray area of two images taken using different sensors,a point matching method based on phase-congruency keypoint-response constraints is proposed to obtain the set of corresponding keypoints from the satellite reference image and the airborne object images.This method performs the spatial transformation of the satellite reference image and the airborne object images.First,the autoencoder-weighted fusion method is used to obtain the gray alignment image,which makes the satellite reference image and the object images obtained from different sensors comparable.Second,the keypoint responses of two images are calculated using the phase consistency theory,which forms the foundation of the point matching method that is based on phase-congruency keypoint-response constraints.The distinguishing feature of this method is that it considers the invariance of phase information in the weak texture images and uses the phase information to construct the keypoint response,which accurately reflects the distinguishability and the matching ability of the keypoints.Finally,to optimize the keypoint detection and feature descriptor model,a method that computes the phasecongruency keypoint-response loss function is used on the two gray aligned images.This allows for the generation of keypoints and descriptors that are similar.The aforementioned point matching method with similar keypoints and descriptors reduces the occurrence of outliers in the corresponding keypoint set,thereby guaranteeing accurate set matching.According to the experimental results,the proposed method shows superior performance as compared to the existing methods and addresses the issue of view difference,a common problem in object matching.To address the problem of large-scale morphological differences,a diffeomorphic set matching method based on an adaptive multiscale kernel is proposed to address the problem of inaccuracy of set matching caused by outliers and multiscale variation of the corresponding keypoint set.This method determines the spatial transformation relationship between the corresponding keypoints set and obtains the spatially aligned images corresponding to the satellite reference and airborne object images.This eliminates the scale morphology difference between the two images and the method thus forms a basis for calculating the similarity between the satellite reference and airborne object images.The corresponding point set is first divided into multiple keypoint subsets using a probability mixture model,and the optimal spatial transformation model is then determined using the comparison results of multiple keypoint subsets.This classification method reduces the influence of abnormal corresponding keypoints on set matching.Second,the scale of the kernel of the diffeomorphism is adaptively adjusted on each subset by calculating the minimum error between the estimated probability density of each subset and its actual density.This results in consistency between the scale of the diffeomorphism kernel and the scale of the spatial transformation of the corresponding keypoint set and ensures that the most optimal spatial transformation model is identified.The experimental results demonstrate the superiority of the improved set matching over the existing set matching methods and its ability to address the issue of large-scale morphology differences in object matching.A region matching method based on region-division fusing-phase information is proposed to address the issue of ground cover differences in airborne object matching guided by satellite images.This method resolves the problems of region division and similarity calculation caused by interference and noise in airborne object images and determines the optimal matching outcome by assessing the similarity of region matching.The phase information image is first generated using the binary encoding of phase information before fusing the phase information and original images to create the advanced image,thereby increasing the information on the edge and angle.Second,the maximal stable region detection method based on probability statistics,which increases multi-instantaneous observation data,is applied to the advanced image to divide the region,thereby addressing the issue of probability statistics bias caused by small sample data.Third,the diffeomorphic set matching method is used to compute the similarity of region matching to obtain the optimal matching result.The experimental results demonstrate that the improved region matching method is superior to the existing region matching techniques and resolves the issue of ground cover differences in object matching.Point-,set-and domain-matching methods are optimized to meet the task requirements of airborne object matching guided by satellite images.On this basis,a three-level fusion-image object-matching method for point set regions is developed by cascading the three methods together.First,the corresponding keypoints set from the satellite reference image and the airborne object images are obtained using a point matching method based on phase-congruency keypoint-response constraints.Then,the diffeomorphic set matching method based on an adaptive multiscale kernel is used to determine the spatial transformation relationship between the corresponding keypoints set and align the spatial location of the satellite reference and airborne object images.Finally,the similarity between the satellite reference image and the airborne object images is computed to evaluate the matching result between the two images using the region matching method that utilizes region-division fusing-phase information.The experimental results show that proposed image matching method is superior to the existing object matching methods and effectively addresses the issues related arising from the difference in view,large-scale morphology,and ground cover in object matching.