基于边缘线分析与聚合通道特征的港口舰船检测  

Docked Ship Detection Based on Edge Line Analysis and Aggregation Channel Features

作　　者：黎经元 厉小润[1] 赵辽英[2] Li Jingyuan;Li Xiaorun;Zhao Liaoying(College of Electrical Engineering,Zhejiang University,Hangzhou,Zhejiang 310027,China;Institute of Computer Application Technology,Hangzhou Dianzi University,Hangzhou,Zhejiang 310018,China)

机构地区：[1]浙江大学电气工程学院,浙江杭州310027 [2]杭州电子科技大学计算机应用技术研究所,浙江杭州310018

出　　处：《光学学报》2019年第8期217-226,共10页Acta Optica Sinica

基　　金：国家自然科学基金(61671408);教育部联合基金(6141A02022350)

摘　　要：针对可见光学遥感图像港口舰船检测过程中,人造目标造成检测结果准确率低、虚警率高的问题,提出了一种基于边缘线梯度特征定位和聚合通道特征的舰船检测方法。基于多尺度多结构元素形态学滤波实现海陆分割;并结合遥感图像中港口的矩形形状特点,定义边缘梯度正切角和港口凹凸度特征以对港口进行定位,获取港口感兴趣区域集合。提取舰船目标的聚合通道特征,并通过聚合通道特征构建的样本训练库和AdaBoost算法完成分类器的训练,利用训练完成后的分类器完成舰船目标的最终判别确认。实验结果表明该算法相较于传统的HOG特征和Haar特征,检测效果良好,准确率和召回率得到较大的提升。Aiming at the problems of low accuracy and high false alarm rate caused by artificial targets in the process of optical remote sensing image docked ship detection.This paper proposes a new method based on edge line gradient features and aggregation channel features for docked ship detection.The multi-structural and multiscale element morphological filters are used to realize the division of sea and land.According to the rectangular shape characteristics of the port in remote sensing images,the edge gradient tangent angle and the port concave and convex features are defined to locate the port,obtaining collection of port region of interest.The aggregation channel features of ships will be extracted and used to train the classifier for the docked ships by AdaBoost algorithm.The trained classifier is used to confirm the real ships in the port.Compared with traditional HOG feature and Haar feature,the proposed algorithm has better detection effect,and its precision and recall rate are greatly improved.

关 键 词：机器视觉 光学遥感图像 港口舰船检测 边缘线梯度特征 聚合通道特征 ADABOOST算法 

分 类 号：TP753[自动化与计算机技术—检测技术与自动化装置]