目的 高分辨率遥感图像中，靠岸舰船检测有着广泛的应用前景，其主要难点在于舰船与港口陆地在空间上紧邻，在颜色和纹理特征上相似，舰船与港口陆地难以分割。针对这种情况，本文利用港口岸线平直的几何特点和靠岸舰船多为舷靠的停泊特点，提出一种基于投影分析的靠岸舰船检测方法。方法 首先，对原始图像进行预处理，利用k-means聚类算法与区域生长算法相结合的方式得到海陆分割图像，利用sobel算子与Otsu分割结合的方式获取边缘图像；然后，通过改进的Hough变换提取直线特征，结合港岸几何特性定位港口岸线；再将海陆分割后的二值图像向沿岸线和垂直岸线两个方向进行投影，根据沿岸线方向投影形态确定和分离并靠舰船，根据垂直岸线方向的投影形态定位舰船目标；最后，利用舰船尺寸、长宽比、最小外接矩形占空比特征去除虚警。结果 本方法在15个港口场景不同分辨率的遥感图像测试集上，整体检测率达到85.4%,虚警率达17.2%；限定分辨率范围在2-4米的情形下，检测率提高到93.5%，虚警率降低至5.3%。结论 本文方法简单有效，无需港口先验信息，适用于多尺度和多方向的靠岸舰船目标检测任务，对不同类型舰船形态差异具有鲁棒性，且能够分离并靠舰船。
Inshore Ship Detection in High Resolution RemoteSensing Image Using Projection Analysis
Lei Zhang,Xing Hong,Yuehuan Wang,Bin Zhou(School of Automation,HuazhongSUniversitySof Science and Technology,Wuhan;Beijing Aerospace Automatic Control Institute,Beijing)
Objective In high resolution remote sensing images, inshore ship detection has broad application prospects such as ocean surveillance, fisheries management and military reconnaissance, etc. However, unlike the ship detection under the pure sea background, inshore ship detection is much more challenging considering complex background of the port. The main difficulty of inshore ship detection is that the ship and the dock are adjacent in space and similar in color and texture features, which makes it difficult to distinguish them. There"s not many effective methods on this task. The existing methods can be mainly divided into three types: the first type is based on the template matching but prior geographical information of the port is needed, which is often not easy to obtain. The second type is based on the ship contour method while the robustness is low and it is difficult to detect side-by-side ships. The third type is based on local features of the ship, which often assume that the ship has a "V"-shaped bow and is powerless to other ships. Other than the existing methods, we propose a method for inshore ship detection using the projection analysis. Our method is based on the observation that the shoreline of dock is typically straight and inshore ships are usually anchored along the ship"s rail. Method First of all, The original image is preprocessed by two sibling approaches: one that segments the sea and land and another extracts edges. For sea-land segmentation, the k-means clustering algorithm and region growing algorithm are combined together to improve the segmentation quality, which is considerable significant for our method. Meanwhile, the original image is processed to a gradient image by sobel operator and then the gradient image is segmented to an edge image with Otsu algorithm. Secondly, an improved Hough transform is conducted on the edge image to extract the straight lines, among which are dock shorelines. In order to remove interference lines, we assume that all extracted lines of dock shorelines should only lie on the border of water. Then, we start to search ships on both two sides of the located dock shorelines. Taking one dock shoreline for example, we project the sea-land segmentation image perpendicular to the dock shoreline direction and get a projection curve. If there is a ship anchored along the dock shoreline, the projection curve shape is convex. Otherwise, the curve shape is flat. Furthermore, we can locate the ship with a bounding box by analyzing the curve shape conveniently. In order to separate the side-by-side ships, we conduct another projection in the dock shoreline direction and separate side-by-ships by analyzing the peak and valley of the projection curve. Finally, we remove false alarms using features of ship size, aspect ratio and duty ratio. Result We have chose randomly 292 high resolution remote sensing images of 15 different scenes in the Google Earth to test our method. The test images have a total of 962 ships, consisting of 139 aircraft carriers, 794 destroyers and 29 civilian ships. The resolution of the images ranges from 1 meter and 5.5 meters. Thus, the ships on these images are variant in scale, orientation and even brightness. Our method has correctly detected 822 of the 962 ships, including 134 aircraft carriers, 666 destroyers and 22 civilian ships. These are 85.4% of the total detection rate, 96.4% of the aircraft carrier detection rate and 83.9% destroyer detection rate. Meanwhile, we have 171 false alarm target and false alarm rate of 17.2%. Results show that if the resolution is limited from 2 meters to 4 meters, the total detection rate grows up to 93.5% and false alarm rate decreases to 5.3% respectively. However, it’s also need to be mentioned that our method is sensitive to the quality of sea-land segmentation, which is key to extract the straight line features of the dock shorelines. So the detection rate is compromised on very complex background as shown in this article. Conclusion Our method is simple and effective for inshore ship detection task. No prior information of the harbors is needed. It’s suitable for detection of inshore ships in variable resolutions and directions. It is robust to ship shapes and side-by-side ships can be detected as well. Since our method is sensitive to the quality of sea-land segmentation, more powerful segmentation algorithm may help and this is our following research direction.