Research On Multi-target Intelligent Detection Algorithm Based On SAR Image Of Complex Scene

Synthetic aperture radar(SAR)has a very wide application prospect in the fields of aerospace,earth observation and military reconnaissance.How to detect the target of interest from complicated scenes is very important for battlefield reconnaissance and military command.Due to the difficulty in multi-target detection of complex scene,SAR images in complex scene are studied in this paper,and finally an intelligent algorithm for multi-target detection of SAR images is proposed.The main research contents and achievements are as follows:(1)The fitting of clutter distributions in different backgrounds is studied.Six different statistical distributions are used to fit the different background images of Sandia database,and the best statistical model is judged combined with three fitting precision evaluation indexes.The statistical models of low vegetation,high vegetation,urban buildings,sea clutter and other regions are given.Finally,a variety of statistical model base is established.(2)CFAR detection methods are studied,and an intelligent iterative CFAR detection algorithm based on statistical model is proposed in this paper.Due to the limitation of automatic censoring CFAR detector in image application,the background prior knowledge is added to the local CFAR detection,according to the region of the background and the background statistical model base,the best statistical model is chosen.While considering the other target samples mixed in the clutter window makes the parameter estimation not accurate,an intelligent iterative CFAR detection algorithm is proposed.CFAR detection is carried out several times to obtain more accurate detection results.Finally,a global ranking method is proposed to solve the global threshold when the censoring depth is calculated.Experiments show that adding this global sort initialization algorithm can greatly reduce the number of iteration and improve the efficiency of detection.The experimental results show that the proposed detector is better than the auto-censoring detector and the local CFAR detector.(3)The target detection of SAR image reconstruction is studied.Firstly,Due to the deficiency of reconstructed image quality evaluation index,three new indexes are proposed and the validity of the index is verified.Secondly,because of the different mechanism between SAR image and optical image,a nonlocally centralized sparse representation reconstruction algorithm for SAR images is proposed,and the evaluation results show that the proposed algorithm is better than traditional PPB algorithm in image denoising.The iterative CFAR detection is performed on the reconstructed images.Experimental results show that the speckle noise in SAR images can be reduced by the proposed method,which is beneficial to target detection.Compared with the original image detection results,detection probability and quality factors have greatly improved.(4)The algorithm of EF feature detection is studied.Aiming at the large number of false alarms in CFAR detection image,it is proposed to detect the CFAR detection images by EF feature detection to suppress the false alarm.Then the original image detection and the reconstructed image detection are merged to enrich the target edge information.Finally,the detection algorithm proposed in this paper is applied to multi-target detection in complex scene.Not only multiple or multi-class targets in the image can be detected,but also the target in the same image can be detected for optional purpose.