Gao Ce, Wu Yiquan, Hao Xiaohui
College of Electronic and Information Engineering, Nanjing University of Aeronautics and Astronautics, Nanjing 211106, China.
Sensors (Basel). 2020 Dec 28;21(1):144. doi: 10.3390/s21010144.
Target detection in hyperspectral imagery (HSI) aims at extracting target components of interest from hundreds of narrow contiguous spectral bands, where the prior target information plays a vital role. However, the limitation of the previous methods is that only single-layer detection is carried out, which is not sufficient to discriminate the target parts from complex background spectra accurately. In this paper, we introduce a hierarchical structure to the traditional algorithm matched filter (MF). Because of the advantages of MF in target separation performance, that is, the background components are suppressed while preserving the targets, the detection result of MF is used to further suppress the background components in a cyclic iterative manner. In each iteration, the average output of the previous iteration is used as a suppression criterion to distinguish these pixels judged as backgrounds in the current iteration. To better stand out the target spectra from the background clutter, HSI spectral input and the given target spectrum are whitened and then used to construct the MF in the current iteration. Finally, we provide the corresponding proofs for the convergence of the output and suppression criterion. Experimental results on three classical hyperspectral datasets confirm that the proposed method performs better than some traditional and recently proposed methods.
高光谱图像（HSI）中的目标检测旨在从数百个窄连续光谱带中提取感兴趣的目标成分，其中先验目标信息起着至关重要的作用。然而，以往方法的局限性在于仅进行单层检测，这不足以从复杂的背景光谱中准确区分目标部分。在本文中，我们将分层结构引入传统的算法匹配滤波器（MF）。由于MF在目标分离性能方面的优势，即背景成分被抑制而目标得以保留，MF的检测结果被用于以循环迭代的方式进一步抑制背景成分。在每次迭代中，前一次迭代的平均输出被用作抑制标准，以区分当前迭代中被判定为背景的像素。为了更好地从背景杂波中突出目标光谱，对HSI光谱输入和给定的目标光谱进行白化处理，然后用于构建当前迭代中的MF。最后，我们为输出和抑制标准的收敛性提供了相应的证明。在三个经典高光谱数据集上的实验结果证实，所提出的方法比一些传统方法和最近提出的方法表现更好。
