Research Of Fast Target Detection Algorithms In High Resolution SAR Images

As synthetic aperture radar(SAR) has the ability to acquire images under all-time and all-weather conditions, it has become a significant way for remote sensing based earth observation. In the military field, SAR has penetrating ability and can probe various kinds of targets effectively. In the civil field, it is applied for weather forecast, disaster monitoring and the research of terrain and landform widely.As the resolution of SAR system improves, it contributes to get more refined target information. But difficults are brought to SAR target detection by the high resolution SAR image: the statistical properties change, the scenes become very complex, and the data amount is huge.SAR target detection is the critical phase of targets recognition and is a hotspot of SAR image interpretation. The performance of target detection algorithms has a major influence on target recognition. The most frequently used SAR target detection method is constant false alarm rate(CFAR) detetion. But, CFAR detection methods based on the pixel has several disadvantages as follows: target fractures, many false alarms and slow velosity of the algorithm.In view of the above problems, after surveying and analyzing existing superpixel methods and CFAR detection methods based on the pixel, CFAR detection method based on the superpixel is researched for high resolution SAR images.The main contents in this thesis are as follows:(1) The fast CFAR detection algorithm based on the pixel is researched. From the algorithm and the hardware, fast CFAR detection methods based on the pixel are summarized. The connections of several clutter satistical models are analysed.(2) A clutter statistical model based on the superpixel, the compound Gamma distrbution, is researched in this thesis. When a superpixel changes into a pixel, the compound Gamma distribution is degraded to the Gamma distribution. The theoretical analysis and the proof of this two clutter satatistical models are made in detail in this thesis. We prove they have high goodness of fit by calculating the KL values and a modified chi-square test. Besides, after several common superpixel methods are analysed, the performance of SLIC is better than other superpixel methods. So SLIC is chosen in this thesis.(3) A moving window based on the superpixel is researched. Combining the moving window with clutter statistical models above, CFAR detection algorithm based on the superpixel is achieved. In order to adapt to improvement of the resolution of SAR images increasingly and improve the real-time of this algorithm further, OpenMP is utilised to achieve the parallel algorithm so that it can make full use of CPU resource and reduce the running time. Experiments show that the detection performance of this algorithm is effective, and its velosity is fast.