将稀疏表示与协同表示结合的高光谱检测.docxVIP

将稀疏表示与协同表示结合的高光谱检测介绍高光谱图像包含数百个连续的狭窄的光谱，这使相比传统方法得到的宽带多谱线的图像而言，它能更精确的识别不同材料，且使潜在的民用和军用应用成为可能。在过去的二十年里，在遥感目标检测方面的团体在寻找识别人造目标或低可能性材料，它有被背景污染的光谱特征。由于高光谱图像的高的光谱分辨率提供的强大的识别，所以它能提供检测这些目标的优点。Over the last two decades, target detection in remote sensing community seeks to discriminate man-made objects or low probability materials, which usually have spectral features contaminated by the background, and HSI offers advantage of detecting these objects due to powerful discrimination provided by its high spectral resolution.最近，对于高光谱图像，一种基于稀疏表示的目标检测算法已经被提出。稀疏表示发现了高光谱图像中的橡素通常可以用一些来自于基础或字典的元素来表示。稀疏表示的系数可以通过L1范数最小化求得，通过确定其系数，这个方法揭示其稀疏表示的识别特点。在参考文献[12]中提到，根据像素的领域部分通常表示为同一类这一事实，一个联合稀疏模型被用来思考在高光谱图像中的空间相关性，进而增加稀疏重建的准确性。在参考文献{15]中提到，一个非局部版本协同表示法，在高光谱图像分类中被介绍。在参考文献[16]中提到，一个协同表示和非常规则的正则化空间[17]的联合分类器被展示。另外，我们当前的工作是利用协同表示来处理异常情况检测。在这里，协同表示意味着所有的原子联合表示一个单个像素点。并且每一个原子有公平的机会来参与表示{除非一个正则化被添加去调整它的重要性），这不同于稀疏表示中的一对一。Recently, a target detection algorithm based on sparse repre-sentation [11,12] has been proposed for HSI. The sparse represen-tation exploits the fact that pixels in HSI can usually be represented using only a few elements from a basis or dictionary [13,14]. The sparse representation coefficients can be solved via an ?1-norm minimization. The algorithm exploits the discriminative characteristics of sparse representation to determine its label. In [12], a joint sparsity model (JSM) was used to consider spatial correlation in HSI to increase the accuracy of sparse reconstruction due to the fact that neighboring pixels usually belong to the same class. In [15], a nonlocal version of collaborative representation, was introduced for HSI classification. In [16], a combination ofcollaborative representation and nearest regularized subspace (NRS) [17] classifier was presented. Besides, our recent work [18] employed collaborative representation for anomaly detection. Here, collaborative representation means all the atoms “collaborate”on the representation of a single pixel, and each atom has th