利用结构纹理分解的海洋舰船目标检测

Warship target detection algorithm based on cartoon-texture decomposition

针对卫星遥感海洋可见光图像,提出一种基于结构纹理分解的舰船目标检测算法.首先对图像进行结构纹理分解,将舰船目标分解到纹理分量中,而其他目标分解到结构分量中;为了进一步消除纹理分量中其他目标的细节对舰船检测的干扰,利用结构分量中陆地、云块以及海洋背景等的先验信息去除其在纹理分量中的干扰纹理;然后在纹理分量中利用Gabor滤波增强舰船目标并进行检测.实验结果和数据分析表明,该算法能够有效地提高检测率,同时降低了虚警率.

Based on cartoon-texture decomposition, we propose a scheme for detection of warship targets in optical remote sensing images. First, an image is decomposed into a structure component and a texture component. The structure component contains objects of large sizes such as sea, lands and clouds that possiblly appear, whiIe the texture component contains objects of comparablly small sizes such as warship targets and details of other objects. In order to reduce the influence of the details on the detection of warships in the texture component, we remove the details of lands, clouds and sea from the texture component by using the structural information of these objects. Finally, the ship targets are enhanced by using Gabor filtering. Experimental results and data analysis show that the new algorithm efficiently improves the detection rate and significantly reduces the false rate.