联合空间预处理与谱聚类的协同稀疏高光谱异常检测

发布时间：2018-05-09 14:16

本文选题：遥感 + 高光谱图像　；参考：《光学学报》2017年04期

【摘要】：针对利用稀疏表示进行高光谱图像异常目标检测效率不高的问题,基于高光谱图像成像原理和图像结构,充分利用高光谱图像的空间特性和光谱特性,并在它们之间建立协同处理机制,提出了联合空间预处理与谱聚类的协同稀疏高光谱图像异常目标检测算法。该算法首先对高光谱图像空间特性进行分析,并结合光谱特性进行空间预处理,使得处理后的高光谱图像更易于异常目标的检测;利用建立在谱图划分思想基础上的谱聚类方法进行波段子集划分,谱聚类方法具有收敛于全局最优解、聚类速度快的特点;利用提出的新的空间和光谱协同稀疏差异指数方法对每个子集进行异常目标检测,该协同稀疏方式充分考虑了高光谱图像的空间特性和光谱特性,通过对每个波段子集检测结果进行叠加,得到最终异常检测结果。利用真实的AVIRIS高光谱图像和合成的高光谱图像对算法进行仿真实验和结果分析,结果表明该算法具有稳健性,同时检测精度高,虚警率低。
[Abstract]:Based on the principle and structure of hyperspectral image, the spatial and spectral characteristics of hyperspectral image are fully utilized to solve the problem of using sparse representation to detect abnormal targets in hyperspectral images. A cooperative processing mechanism is established between them, and an algorithm for detecting anomalous targets in hyperspectral images is proposed, which combines spatial preprocessing and spectral clustering. The algorithm firstly analyzes the spatial characteristics of hyperspectral images and carries out spatial preprocessing combined with spectral characteristics to make the processed hyperspectral images easier to detect abnormal targets. The spectral clustering method based on the idea of spectral partition is used to partition the band subset. The spectral clustering method has the characteristics of converging to the global optimal solution and fast clustering speed. A new spatial and spectral synergistic sparse difference index method is proposed to detect abnormal targets for each subset. The spatial and spectral characteristics of hyperspectral images are fully considered in the cooperative sparse method. By superposing the detection results of each band subset, the final anomaly detection results are obtained. Real AVIRIS hyperspectral images and synthetic hyperspectral images are used to simulate and analyze the algorithm. The results show that the algorithm is robust and has high detection accuracy and low false alarm rate.
【作者单位】：哈尔滨工程大学计算机科学与技术学院;大庆师范学院机电工程学院;哈尔滨工程大学信息与通信工程学院;
【基金】：国家自然科学基金(61571145) 黑龙江省博士后基金(LBH-Z14062)
【分类号】：TP391.41

【相似文献】