基于GLRT的光学卫星遥感图像舰船目标检测

传统的CFAR检测应用到光学卫星遥感图像舰船目标检测中时不能对黑极性目标进行判断,针对此提出改进的基于广义似然比检验(Generalized Likelihood Ratio Test,GLRT)的舰船目标检测算法。该算法采用滑动窗口检测形式,在假设背景和目标灰度均服从高斯分布的前提下,通过GLRT判断背景窗口与目标窗口是否同分布来检测目标,兼顾了目标黑白两种极性的情况。算法实现中对图像进行了分块检测,并通过形态学处理对检测结果进行了目标聚类。采用SPOT5与CBERS实测数据进行实验,验证了海背景服从高斯分布的假设。典型数据检测结果表明,该算法可以检测黑极性目标,且相比CFAR虚警率更低,大量数据计算ROC曲线的结果以及比CFAR检测少约40%的耗时进一步表明该算法性能更优。

Ship Target Detection from Optical Satellite Remote Sensing Image based on GLRT

The technology in optical satellite remote sensing develops quickly,which is important to seek how to detect the ship target from large amounts of optical satellite remote sensing images.Aiming at the problem that traditional CFAR detection cannot detect the black polar target,a ship target detection method based on GLRT(Generalized Likelihood Ratio Test) is proposed.Like CFAR detection,this method uses a sliding window to detect target.Assuming the intensity of the background and target follow Gaussian distribution,detects the target through the judge that the distribution in the background window and target window is the same by the GLRT theory,which considers both the white and black polar target.Detection is operated in the image blocks and morphological operation is used for target cluster.The images used in Experiments come from SPOT5 and CBERS.Assume that the intensity of the background follows Gaussian distribution is validated.The result of detection in two images shows that this method can detect the black polar target and have fewer false alarms than CFAR detection.The ROC curve and the cost time that is 40% less than CFAR detection,and further show that this method is more superior.