基于层次化分类器的遥感图像飞机目标检测

在大量航空航天遥感图像中，快速发现和统计飞机目标并对其进行准确定位，在军事和民用方面均具有重要意义。结合遥感图像特点，针对飞机目标的特征，文章设计了一种基于层次化的分类器的遥感图像飞机目标检测方法。首先用基于哈尔（Haar）特征的底层AdaBoost分类器快速去除大部分非目标区域；然后用基于梯度方向直方图（Histogram of Oriented Gradient，HOG）特征的顶层支持向量机（Support Vector Machine，SVM）分类器进行精细检测。在分辨率为1m的遥感图像数据集上的实验结果表明，层次化分类器在保证较高检测率的前提下，大大降低了虚警率，可以有效解决遥感图像飞机检测问题。

Aircraft Detection in Remote Sensing Image Based on Hierarchical Classifiers

Quick finding and counting aircraft objects and acquiring their accurate positions in a number of space and aviation remote sensing images, are of great significance both in military and civilian applica-tions.According to the characteristics of remote sensing images and the features of aircraft objects, a new method is designed for aircraft detection in remote sensing images which makes use of the hierarchical classi-fiers. Firstly, bottom AdaBoost classifiers based on Haar features are used to quickly filter out most of the non-aircraft areas.Then top SVM （Support Vector Machine） classifiers based on HOG （Histogram of Oriented Gradient） features are applied for fine recognition.The experimental results in remote sensing image dataset with the resolution of 1m show that hierarchical classifiers can greatly reduce the false alarm rate with high detection rate, so this method can effectively solve the problem of aircraft detection in remote sensing image.