基于级联分类器的红外弱小目标快速检测  

Fast detection of small infrared targets based on cascade classification

作　　者：许庆晗[1] 金立左[1] 费树岷[1] 

机构地区：[1]东南大学自动化学院,南京210096

出　　处：《高技术通讯》2012年第7期772-777,共6页Chinese High Technology Letters

基　　金：航空科学基金（20080169003,20115169016）资助项目.

摘　　要：为克服弱小目标检测在大尺寸图像上存在时间开销大的问题，提出了一种多级处理的检测算法。首先在整幅图像上搜索局部灰度极值作为候选目标位置，随后对候选目标位置的邻域进行特征提取，最后在特征空间中进行分类。针对小目标特性，提出了一种灰度分布特征，以此特征对候选目标位置进行特征提取，进而提出了一种加权逻辑斯蒂回归算法，用其在特征空间建立分类器，将检测问题转化为二分类问题。实验结果表明，该算法对低信噪比图像可以达到较高的检测率和较低的虚警率，并具有较好的实时性。To solve the problem that small infrared target detection in large images takes much time, a multi-stage algo- rithm is proposed, which first of all considers local extrema of gray intensity in a whole image as candidate targets, then extracts the features in the neighborhood of the candidates, and finally identifies true targets in the feature space. According to the features of small targets, a feature of gray intensity distribution is presented, with which candidate targets are extracted, and then an algorithm named weighted logistic regression is proposed. Consequent- ly, the detection problem is converted to a binary classification problem in the feature space. The experimental re- sults show that the proposed algorithm has the good performance for real-time target detection with low SNR images.

关 键 词：红外小目标 检测 灰度分布特征 加权逻辑斯蒂回归 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]