Research On Low-Detection-Rate SAR Image Target Detection And Recognition Theory

Because synthetic aperture radar(SAR)has an all-weather,day/light capability to obtain images,it has become an important tool for earth observation in the military field.As fundamental techniques for SAR image processing and interpretation,the target detection and recognition of a SAR image have irreplaceable function in military applications,such as battlefield reconnaissance,monitoring.With the development of SAR imaging techniques,SAR images have the characteristic of high resolution.Meanwhile,the wide applications of SAR techniques lead to the scenes and electromagnetic environments of SAR imaging more complex,which bring the SAR images different characteristics,such as strong noise,and complex backgrounds.For the high resolution,the statistical properties change and the backgrounds become complex.For the strong noise,the signal to noise ratio decreased.For the complex backgrounds,the targets may be similar as local background.When traditional methods deal with the SAR images with the new characteristics,the performance of SAR ATR systems decreased.In view of the above problems,after surveying and analyzing existing methods,in this dissertation several target detection methods are proposed for the low detection rate(LDR)SAR images in the consideration of accurate modeling of LDR SAR image properties and fully exploiting SAR image information.On this basis,the impact of target detection to recognition is analyzed by the SAR ATR system for the further study on the target recognition in LDR SAR images.The main work and contributions are summarized as follows:1.Based on the study of widely used CFAR target detection method,the problems in detection of the LDR SAR images are analyzed.Then the characteristics and detection difficulties of the LDR SAR images are summarized from three aspects,which are high resolution,strong noise,and complex backgrounds.2.Based on the study of high resolution situation of LDR SAR image,singular value decomposition(SVD)based visual attention method is proposed.The defects of applying the traditional visual attention method to SAR image are analyzed firstly.Then,combining with the pyramid model,the SVD is introduced into visual attention method to bulid the SVD based multilayer structure.In the feature extraction step,a consistency feature and a standard deviation feature is proposed to realize the target detection.The experimental results on different data show the effectiveness of the proposed method.3.Considering that the strong noise situation of LDR SAR image contain rich statistical character,an information theory based and superpixel level target detection method is proposed.The basic cell of SAR images is changed from pixel to patch through the superpixel algorithm firstly.Then by taking advantage of the rich statistical character of patch,an information measurement,including self-information and entropy,is utilized to measure the statistical difference between patches.In this way,the proposed approach is more stable for SAR images with different intensities of speckle noise.4.For the complex scene situation of LDR SAR image,a target detection approach based on the existence probability of target is proposed.Superpixel is the basic unit throughout the approach,which is labeled into each classified scene by texture feature.Then the existence probability of targets is estimated based on the comparison of two saliency depth values which are original and predicted saliency depth.At last,an improved visual attention algorithm,in which the saliency map of scenes are endowed with different weights related to the existence probabilities,derives target detection result.The algorithm enhances the attention for the scene which contains target.Hence,the proposed approach is self-adapting for complex scenes,and the algorithm is widely suitable for different detection missions too,e.g.vehicle,ship,or aircraft detection in related scene of road,harbor,or airport.The experimental results on different data show the effectiveness of the proposed method.5.Based on the research on target detection in LDR image,the influence of the detection on recognition is analyzed by the SAR ATR system.For the high resolution situation,the experimental results show that the false alarms in detection will not impact on the performance of recognition,and multi-feature fusion method is a well research orientation.For the strong noise situation,the recognition method is more stable than CFAR method,and the proposed detection method can improve the performance of SAR ATR system.For the complex scene condition,because the local area could be similar as target in SAR image,the filtering of false alarms should be more emphasized in detection,and the multi-sensor fusion method is a well research orientation.In summary,a series of studies are carried out centered on the problem of LDR SAR image target detection and recognition.This study enriches the theories of SAR image target detection and recognition.It also provides effective methods for information extraction and interpretation of LDR SAR images.