Target Recognition And Location Change Analysis System Based On Remote Sensing Image

With the progress of aerospace technology and satellite remote sensing technology,high-resolution remote sensing image has been developed rapidly in the direction of multiangle,omni-directional,all-weather and high-precision.The characteristic information contained in the image has high research value in national security,resource investigation,environmental monitoring,urban planning and management and other major fields.This paper aims to implement a target recognition and location change analysis system based on high resolution remote sensing image based on deep learning target detection algorithm and traditional image processing algorithm.The system consists of two sub modules: typical target recognition and location change analysis.Aiming at the requirement of electric tower detection under large field of view visible light image,this paper designs a large field of view target detection algorithm based on fast RCNN.By designing the size and sliding step of the sliding window,any electric tower in the original image can be completely included in a small clipping image.On this basis,NMS algorithm is used to post process the electric tower detection result set,The incomplete and redundant detection frames are eliminated to obtain the optimal prediction frame.After testing,the recognition rate of the typical target recognition module for the electric tower under visible light image is more than83%.Aiming at the requirements of electric tower and electric line detection in highresolution SAR images,considering the lack of texture information of targets in SAR images,this paper designs a electric tower and electric line detection algorithm based on traditional image technology,which can remove most of the noise in SAR images through a series of operations,such as image binarization,expansion and corrosion,de aggregation noise and so on.On this basis,with the help of prior knowledge based on electric tower line trend judgment,the electric tower line path generation algorithm is designed and implemented.After testing,the recognition rate of typical target recognition module for electric tower and electric line under SAR image is more than92%.In order to solve the problem of large ratio of length to width and rotation angle of aircraft and ship in high-resolution visible light image,this paper proposes an aircraft and ship detection algorithm based on rotation frame.Combined with the ROI pooling layer information of various sizes,the algorithm can classify and regress the rotating bounding box,and achieve higher quality bounding box prediction.After testing,the recognition rate of the location change analysis module for different types of images is more than 86%.In order to analyze the change of aircraft and ship location in high-resolution image sequences,an image alignment algorithm based on feature point matching strategy is proposed.By introducing the idea of image pyramid,a feature point matching algorithm based on orb is designed and implemented.By using the mapping relationship between feature points and positions in sequence images,the overdetermined equation is established,the homography matrix is solved,and the alignment of target positions in front and back sequence images is realized by affine transformation.On this basis,the positioning changes of aircraft and ships are analyzed.