Remote Sensing Target Detection And Recognition Based On Deep Learning And Image Enhancement

With the development and advancement of remote sensing technology,remote sensing images are more and more widely used in the fields of surface classification,reconnaissance detection,environmental management and national defense construction.The innovation of computer vision research and the generation of massive data bring more opportunities and challenges to the refined processing of images.In this paper,the problem of detecting and identifying typical targets such as airplanes and ships in remote sensing images is adopted.The deep neural network model is adopted,and the detection,identification and positioning are also used for regression processing,which can effectively shorten the training time and improve the processing efficiency.An image enhancement method based on edge detection is proposed,which can better highlight the target features and further improve the detection accuracy.The specific research work is as follows:We studied the biological mechanism and modeling solution process of neural network.For the solution of feedforward neural network model,the idea of back propagation algorithm is analyzed.The main data sets used in the current target detection field are also summarized,including traditional optical data sets and optical remote sensing.Datasets and SAR remote sensing datasets,and an evaluation index for image target detection.The method of target detection and recognition using remote regression image in remote sensing image is explored.Feature extraction,feature recognition and target localization are unified through regression structure,and candidate region generation is reduced by image segmentation,and network is optimized.The loss function is constructed,the target region is determined by the non-maximum suppression method,the training model is accelerated by the GPU,and the results are compared and analyzed.A preprocessing method for SAR remote sensing image enhancement is proposed.The target features are characterized by Sobel operator and gray scale expansion.The threshold threshold is then set to process the image,and the depth convolutional neural network is adaptively improved,so that the model can be more good recognition feature extraction.