Shi Hao, Zhang Qingjun, Bian Mingming, Wang Hangyu, Wang Zhiru, Chen Liang, Yang Jian
Department of Electronics, Tsinghua University, Beijing 100084, China.
Beijing Institute of Spacecraft System Engineering, Beijing 100094, China.
Sensors (Basel). 2018 Feb 12;18(2):563. doi: 10.3390/s18020563.
With the rapid development of remote sensing technologies, SAR satellites like China's Gaofen-3 satellite have more imaging modes and higher resolution. With the availability of high-resolution SAR images, automatic ship target detection has become an important topic in maritime research. In this paper, a novel ship detection method based on gradient and integral features is proposed. This method is mainly composed of three steps. First, in the preprocessing step, a filter is employed to smooth the clutters and the smoothing effect can be adaptive adjusted according to the statistics information of the sub-window. Thus, it can retain details while achieving noise suppression. Second, in the candidate area extraction, a sea-land segmentation method based on gradient enhancement is presented. The integral image method is employed to accelerate computation. Finally, in the ship target identification step, a feature extraction strategy based on Haar-like gradient information and a Radon transform is proposed. This strategy decreases the number of templates found in traditional Haar-like methods. Experiments were performed using Gaofen-3 single-polarization SAR images, and the results showed that the proposed method has high detection accuracy and rapid computational efficiency. In addition, this method has the potential for on-board processing.
随着遥感技术的快速发展，像中国的高分三号卫星这样的合成孔径雷达（SAR）卫星拥有更多的成像模式和更高的分辨率。随着高分辨率SAR图像的可得性，自动船舶目标检测已成为海洋研究中的一个重要课题。本文提出了一种基于梯度和积分特征的新型船舶检测方法。该方法主要由三个步骤组成。首先，在预处理步骤中，采用一个滤波器来平滑杂波，并且可以根据子窗口的统计信息自适应地调整平滑效果。因此，它在实现噪声抑制的同时能够保留细节。其次，在候选区域提取中，提出了一种基于梯度增强的海陆分割方法。采用积分图像法来加速计算。最后，在船舶目标识别步骤中，提出了一种基于类Haar梯度信息和Radon变换的特征提取策略。该策略减少了传统类Haar方法中找到的模板数量。使用高分三号单极化SAR图像进行了实验，结果表明所提出的方法具有较高的检测精度和快速的计算效率。此外，该方法具有进行机载处理的潜力。
