基于稀疏表示分步重构算法的高光谱目标检测

摘要

针对传统稀疏表示重构算法在高光谱目标检测中表现出运算速度慢的问题，提出了分步重构算法（Two Steps Reconstruction，TSR）。该方法先求得 K 个与待测像元最相似的字典原子，然后用这些原子线性表示待测像元以求解稀疏向量，舍弃了传统重构算法的迭代求解的方式，直接通过求解逆矩阵，简化了运算过程，使运算速度大幅提高。本文给出了方法的具体过程并将其与传统方法及其改进方法进行比较。实验结果表明，TSR 在保证检测精度不下降的同时能够大幅提升运算速度。%Aiming at the efficiency problem of traditional reconstruction algorithm for target detection in hyperspectral image, a two steps reconstruction algorithm was proposed. K nearest atoms with text pixel were calculated for representing the text pixel to calculate sparse vectors. This method solves the problem by calculating inverse matrix instead of iteration, which simplifies the process. The specific process of the method is given and the method is compared with the traditional and improved ones. Experimental results show that this method can improve the computation speed without decreasing the accuracy of detection.