摘要
为研究将区域协方差算子(regional covariance descriptors,RCD)用于高分辨率遥感图像中的车辆检测,提出了两种新的图像特征提取方法。针对原始RCD方法未能利用图像的二值信息的情况,提出了一种基于二值统计的局部特征模式(local feature pattern,LFP)方法。针对原始RCD方法中卷积核简单、方向单一的问题,设计了一种多尺度、多方向的正弦函数卷积核,提出了一种基于Gabor卷积核的RCD(G)方法。对比了各种方法用于车辆分类的准确率,结果表明,LFP方法的准确率和原始RCD方法差不多,而RCD(G)方法的准确率比原始RCD方法提高了1. 35%,准确率达到95. 25%。另外,LFP+RCD(G)方法的准确率比LFP+原始RCD方法高1. 75%,达到了96. 65%。
In order to study the use of regional covariance descriptors(RCD)for vehicle detection in high-resolution remote sensing images,two new image feature extraction methods are proposed.A binary feature-based local feature pattern(LFP)method is constructed for the original RCD method that does not use the binary information of the image.Aiming at the problem that the original RCD method has simple convolution kernel and single direction,a multi-scale,multi-directional sine function convolution kernel is designed and a RCD(G)method based on Gabor convolution kernel is constructed.Comparing various methods for the accuracy of vehicle classification,the accuracy of the LFP method is similar to that of the original RCD method,and the accuracy of the RCD(G)method is improved by 1.35%compared to the original RCD method,and the accuracy of the LFP+RCD(G)method is 1.75%higher than that of the LFP+original RCD method,reaching 96.65%.
作者
阳理理
陈雪云
陈家华
YANG Li-li;CHEN Xue-yun;CHEN Jia-hua(College of Electric Engineering,Guangxi University,Nanning 530004,China)
出处
《广西大学学报（自然科学版）》
CAS
北大核心
2018年第5期1794-1802,共9页
Journal of Guangxi University（Natural Science Edition）
基金
国家自然科学基金资助项目(61661006)
