利用KPCA特征提取的Adaboost红外目标检测

Detection of infrared targets based on Adaboost by feature extraction using KPCA

针对传统红外目标检测算法中存在的不足,提出了一种基于核主成分分析(KPCA)特征提取的Adaboost分类器红外目标检测算法。首先,采用KPCA对目标训练样本进行特征提取,将背景训练样本和待检测样本在概率核空间中向目标样本特征量投影作为它们的特征量;然后,用目标和背景样本特征来训练Adaboost分类器;最后,用此分类器对待检测样本的特征量进行目标检测,并对比分析了支持向量机(SVM)和二次相关滤波器(QCF)的检测算法性能。实验表明,该方法能实现对红外目标较为鲁棒和准确的检测,并且算法中的参数设定具有一定的自适应性。

According to the shortages of conventional infrared target detection method, an algorithm combining kemel principal component analysis （KPCA） and Adaboost classifier was presented. Firstly, KPCA was used to extract the features of the training samples of the target. Then the training samples of the background and the samples to be detected were projected onto the feature vectors of the target in kernel space as the features of themselves respectively. The features of both target and background samples were then used to train the classifier. Finally, the above classifier was applied to detect the target in IR images. The performances of the algorithm presented were compared with support vector machine （SVM） and quadratic correlation filter （QCF）. Experimental results show that the proposed algorithm can achieve a robust and accurate detection and the parameter setting of the detection algorithm has a certain degree of adaptability.