基于空谱联合异常度的高光谱异常目标检测

随着高光谱图像技术的不断发展,光谱分辨率和空间分辨率不断提高,相比于其他遥感图像,能够获得更为精细的光谱特征。这为地物的高精度分类、解混和目标检测等研究领域提供了理论平台,其中由于高光谱异常目标检测技术不需要地物的先验信息,更符合实际应用的需求。针对现有的大多数高光谱异常目标检测算法只关注目标和背景在光谱信息方面的差异,而忽略两者空间信息的差异,导致检测精度不高的问题,提出了一种基于空谱联合异常度的高光谱异常目标检测算法。该算法不需要假设图像的背景模型,建立在滑动双窗口的基础上,提出了光谱异常度和空间异常度两个概念。在光谱异常度计算中,考虑了波段间的非线性特征,采用光谱角匹配的核函数方法进行检测,基于双窗口模型的基础上逐个计算中心像元与局部背景像元的核光谱角并设置阈值来获得中心像元的光谱异常度;在空间异常度的计算中,由于物质在空间方面的聚类特性,通过构建像元点的空间窗模型能够得到代表像元类别的图像块灰度向量,同时求解不同像元之间图像块灰度向量的欧式距离并设置阈值来获得中心像元的空间异常度;最后将中心像元的光谱异常度与空间异常度进行加和则可得到中心像元的空谱联合异常度,基于滑动双窗口模型对整幅图像的像元进行逐个检测,即可得到图像的异常检测结果。采用AVIRIS的三组真实高光谱数据对所提算法进行仿真实验,并与传统的RX算法、LRX算法和KRX算法进行对比研究,结果表明本文算法具有较好的检测效果,与KRX算法相比,运行速度具有较大幅度的提升。

Space-Spectrum Joint Anomaly Degree for Hyperspectral Anomaly Target Detection

With the continuous development of hyperspectral image technology,spectral resolution and spatial resolution are continuously improved,and finer spectral features can be obtained compared to other remote sensing images.This provides a theoretical platform for the research fields of high-precision classification,demixing and target detection of features,in which the hyperspectral anomaly target detection technology does not require a priori information of the features,which is more in line with the needs of practical applications,and has become a research hotspot.For most hyperspectral anomaly target detection algorithms,only focusing on the spectral difference between the target and the backgroundand neglecting the difference in spatial structure between the two to result in low detection accuracy,a space-spectrum joint anomaly degree for hyperspectral anomaly detectionalgorithm is proposed.The algorithm does not need to assume the background model of the image,based on the sliding double window,and proposes two concepts of the spectral anomaly and spatial anomaly.In the calculation of spectral anomaly,the nonlinear characteristics between the bands are considered,and the kernel function method based on spectral angle matching is used for detection.Based on the two-window model,the nuclear spectral angles of the central pixel and the local background pixel are calculated one by one and set.The threshold value is used to obtain the spectral anomaly of the central pixel;in the calculation of the spatial anomaly,due to the spatial clustering property of the matter,the image block gray vector representing the pixel class can be obtained by constructing the spatial window model of the pixel point.At the same time,the Euclidean distance of the image block gray vector between different pixels is solved and the threshold is set to obtain the spatial anomaly of the central pixel.Finally,the spectral anomaly of the central pixel and the spatial anomaly are summed to obtain the center.The spatial anomaly joint anomaly of the pixel,based on the sliding double window model,detects the pixels of the whole image one by one,and the abnormal detection result of the image can be obtained.The simulation results of the proposed algorithm are carried out by using three sets of real hyperspectral data of AVIRIS,and compared with the traditional RX algorithm,LRX algorithm and KRX algorithm.The results show that the proposed algorithm has better detection effect,compared with KRX algorithm.The running speed has a large increase.