新型高光谱图像快速实时目标检测与分类方法

Novel Fast Real-Time Target Detection and Classification Algorithms for Hyperspectral Imagery

针对逐像元处理的高光谱图像实时线性约束最小方差(LCMV)检测与分类算法计算量大、运行速度慢的问题,在LCMV检测与分类算法的基础上,提出了两种逐行的实时LCMV目标检测与分类算法。首先对LCMV算法进行了因果化,提出了逐行处理的实时因果LCMV(CR-LCMV)检测与分类算法,再利用Woodbury引理,推导出了逐行处理的实时递归因果LCMV(RCR-LCMV)检测与分类算法。实验结果表明:与LCMV检测与分类算法相比,两种新型实时算法均能在不影响检测精度的情况下实时地检测目标与对目标进行分类,且所需的数据存储空间大大降低;与逐像元处理的实时LCMV算法相比,两种新型实时算法可获得几乎与之相同的检测精度,计算复杂度大大降低,实时处理能力更强,算法在运行时间上具有明显的优越性。

The real-time linearly constrained minimum variance （LCMV）detection and classification method for hyperspectral imagery is based on the pixel-by-pixel processing, which has the problems of large amount of computation and slow running speed. Two novel real-time LCMV detection and classification methods based on the LCMV detection and classification method are proposed. Firstly, the LCMV algorithm is carried out causality, a causal real-time LCMV （CR-LCMV） detection and classification method based on the line-by-line processing is proposed. Then, by using Woodbury lemma, a recursive causal real-time LCMV （RCR-LCMV） detection and classification method based on the line-by-line processing is derived. Experimental results show that compared with the traditional LCMV detection and classification algorithm, the two novel real-time algorithms can detect and classify targets in real-time without affecting the detection accuracy, and the required data storage space is greatly reduced. Compared with the real-time LCMV algorithm based on the pixel-by-pixel processing, the real-time processing ability of the two novel real-time algorithms is much strong without affecting the classification accuracy, which has obvious superiority in running time.