基于组合特征的航母目标识别方法  

A Recognition Method for Target of Aircraft Carriers Based on Combination Features

作　　者：张润鑫 武文波[1] 陈瑞明[1] ZHANG Runxin;WU Wenbo;CHEN Ruiming(Beijing Institute of Space Mechanics ＆ Electricity, Beijing 100094, China)

机构地区：[1]北京空间机电研究所,北京100094

出　　处：《航天返回与遥感》2018年第2期126-132,共7页Spacecraft Recovery & Remote Sensing

摘　　要：为了解决较高分辨率条件下的可见光航天遥感图像航母目标识别问题,文章提出了一套完整的识别算法。该算法通过基于最大类间方差算法(Otsu算法)的双阈值分割方法获得舰船目标,然后通过形态学闭运算去除目标图像内部的空洞,对获得的目标图像进行优化。结合航母的形态特点,文章提出了描述目标的舰首宽度比编码特征,该特征能够显著的描述航母和其他舰船之间的区别。通过长宽比和舰首宽度比编码构成组合特征向量对提取的舰船目标进行描述。最后文章采用k折交叉验证方法划分训练样本集合和测试样本集合,采用最小距离分类器进行目标识别。所有的算法都在Matlab软件上进行仿真实验,实验结果表明,相比于其他的舰船描述特征,文章提出的组合特征对航母的识别率更好。For the problem of recognizing the aircraft carrier target with higher resolution in space remote sensing, a complete set of recognition algorithms is proposed in the paper. The ship target is obtained by Double Threshold Segmentation method based on Otsu algorithm, and then the target image is removed the holes inside it and optimized by implementing morphological operations. Considering the aircraft carrier’s morphological characteristics, the length to width ratio coding is proposed, which can describe obviously the difference between an aircraft carrier and other types of vessels. The combined feature vector is obtained using the aspect ratio and the length to width ratio, and then is used to describe the target. Finally, the K-fold cross validation method is used to divide the training samples set and the test samples set, and then A Minimum Distance Classifier is adopted for target recognition. All the algorithms are performed by Matlab simulation software, and the results show that the combined feature proposed in this paper is better than other features in identifying the aircraft carriers.

关 键 词：双阈值分割 航母目标识别 舰首宽度比编码 最小距离分类器 航天遥感 

分 类 号：TP751[自动化与计算机技术—检测技术与自动化装置]