摘要
随着雷达成像技术和高分辨率光栅显示技术的发展和应用,基于船载雷达图像的船只检测成为可能。海上船只检测的主要困难之一是雷达图像中包含固有的海面背景杂波。传统的雷达船只检测方法,如恒虚警率法(CFAR),以杂波分布模型为基础,计算待检测窗口中的信号统计分布来确定自适应阈值,取得了一些成果。但是,当海面背景杂波和船只目标的回波强度在同一数量级,甚至船只目标淹没在海面背景杂波中时,就难以确定一个有效的阈值将船只目标从雷达图像中提取出来。在分析海面背景杂波和船只目标的相关差异性基础上,提出了一种基于船载雷达序列图像的海上船只快速检测方法。该方法首先对相邻两幅图像进行互相关性分析,在两幅图像中的同一位置提取一定尺寸的移动窗口,计算其互相关函数值,窗口移动一个步长,重复操作直至遍布整幅图像,形成一幅由灰度图像互相关函数值组成的相关图像。然后使用概率神经网络模型(PNN模型)来估计相关图像背景杂波的灰度概率密度分布函数(PDF),应用CFAR技术,使用二分法求解一个区分船只和背景噪声的自适应整体阈值,并根据阈值将相关图像二值化,其中大于阈值的像元作为候选的船只目标信息,小于阈值的像元则为海面背景杂波。最后使用连通性8-邻域准则统计各个候选船只目标区域的像元数,并与预先定义的最小船只目标像元数进行比较,偏小的候选船只目标区域作为虚警去除,保留下来的候选船只目标区域即为船只检测结果。研究显示,如果图像序列中没有船只目标信息,则三维相关图像比较平整。相反,如果图像序列中含有船只目标信息,则三维相关图像上有峰值被检测出,通过测量峰值的高度,就能判断存在可能的船只目标。运用X波段船载雷达序列图像对本文提出的海上船只检测方法进行了测试。测试结果表明,该检测方法具有很好的船只检测效果,得到的船只检测结果与目视判别的结果一致。而且该检测算法原理简单,计算速度快,易于实时处理,具有广阔的应用前景。
Effective automatic detection and extraction of maritime ships are of great significance to the safe navigation,harbor surveillance and national defense.This is the prerequisite for further processing such as ship tracking and identification.With the development and implement of the imaging digital radar and the high resolution raster display,ship detection based on shipborne radar imagery becomes feasible.One of the main problems in ship detection is the presence of the inherent sea clutter in the radar image.The traditional method of ship detection with radars such as CFAR(Constant False Alarm Rate) is based on the statistical distribution model of the clutter,in which the self-adapting threshold level is defined automatically by the statistical distribution of the signal in the detection window.It has got some success,but,when the echo of the ship is as strong as that of the surrounding sea clutter,or when the images of the ship embedded in the speckled image of the sea surface,it is difficult to determine a threshold for differentiating a ship in the image.The sea surface consists of a large amount of small scattering objects independent from one another in the time sequential images of the shipborne radar,therefore,it can be treated as random noises without correlation between each other.On the contrary,the ship on the sea is a deterministic target,which is larger than the image resolution of the radar,and thus the correlation between the radar images is great.A ship detection method with time sequential imagery of shipborne radar has been developed based on the difference of correlation between sea surface clutter and ships.In the method,correlation analysis of two consecutive images is made first,in which a moving window of given size is taken at the same position in the radar images to calculate the cross-correlation function value.The window is moved by a pixel and the same process is repeated to cover the entire image.Then a coherence image consisting of the cross-correlation values is produced.Next the PNN(Probabilistic Neural Network) is used to estimate the probabilistic density function of the background clutter of coherence image,and the CFAR technique and dichotomy is used to determine an adapted global threshold for differentiate the ship from background noises.At the same time the coherence image is transformed into its binary image according to the threshold,in which the pixel gray value larger than the threshold is taken as a candidate pixel for the ship and that smaller than the threshold as the background noises.Finally,eight-connectivity checking is used to statistically calculate the image elements of the area of the candidate ship.If the area of a candidate ship is smaller than the predefined minimal ship area,it will be treated as false alarm,and this region could be removed.The remainder is the final result of ship detection.Simulation study shows that when there is no signal of ship target between two consecutive frames of radar images,the 3D(3-dimension) display of coherence image is rather flat.On the contrary,the it has a strong peak.We can judge the existence of a probable ship by measuring the height of the peak.The method proposed has been tested with the X-band shipborne radar imagery.The results show that the method works well,and the ship detection results are consistent with that with the naked eye.And the principle of detection arithmetic is simple and the calculation is fast.It can be operated easily in real-time and has a wider application range.
出处
《海洋学研究》
2009年第4期33-38,共6页
Journal of Marine Sciences
基金
国家海洋局第二海洋研究所基本科研业务费专项资助项目(SZ0737)