Target Detection Based On Multi-band And Multi-polarization SAR Image Fusion

Synthetic Aperture Radar (SAR) is an all-weather; all-time high resolutionmicrowave remote sensing imaging radar used widely in modern military area toobtain important information. Target detection is the premise condition for targetrecognition, it uses target feature to extracted target from the image; which occupies avery important position in contemporary military application.Constant False Alarm Rate (CFAR) technology is the most important means inradar automatic detection system to control the false alarm rate. However, thebackground clutter modeling and complex object types in SAR image bring a greatdifficult; so many traditional CFAR algorithms cannot obtain accurate results.Therefore, this paper proposes to improve the accuracy of testing results firstly.Thus, as the camouflage net shielding is a common means for againstreconnaissance and surveillance, so how to detect the camouflaged targets has apositive significant in practical applications. But during the interpretation process ofSAR camouflaged targets, due to the SAR images cannot provide the differencesinformation of echo waveform in spectral reflectance and polarization characteristics,and the detection method by decomposition the laser echo waveform’s characteristicsare not applicable, therefore we need to find an image detection method which issuitable for SAR echo’s characteristic.This paper put forward own target detection algorithms for single band/polarization, multi bands/ polarizations, gives an almost perfect algorithms library forSAR image target detection and algorithm performance evaluation system. Specificcontents include as following:1. Existing CFAR algorithms are adopted global modeling, which using thesame distribution model to estimate the background clutter and detect thewhole area; but practical ground covers complex types, and different groundarea has its most suitable backgrounds model, which leading the used modelis not fit in some regional, making higher loss of CFAR and bringing downthe test performance. So the paper presented an algorithm which judged theareas according to the different characteristics of background, such asstatistical variance and mean ratio. In this way, CFAR detector could select the distribution model on the basis of the regional type automatically and getthe best detection results.2. Camouflage netting is usually used to against monitoring and reconnaissance,and SAR pictures of different bands and polarizations can provide synergyinformation of targets covered by the camouflage net. This paper provides amulti-bands SAR target detection method based on decision fusion, usedimproved Neyman Pearson rule: regarding the probability of target detectionand effective detection ratio as performance indexes, to select the mosteffective single-band or polarization resource; scale filter was used toevaluate per target pixel, to reduce the redundant information which camefrom single channel; evaluating the improved algorithm and the original oneby using effective detection ratio as performance index of fusion detectionalgorithm.3. To implement engineering application, the registration algorithm warehouseis established, including the mainstream classical algorithms and the newalgorithms proposed by us. Besides, the objective evaluation index system ofmultifrequency and multipolarization SAR image target dection algorithmsis given. The relational algorithms and evaluation indexes implemented byMatlab are embedded into SAR image fusion software system developed byour team. At last, we make a full evaluation for all algorithms in theregistration algorithm warehouse.