高光谱遥感技术通过成像光谱仪记录带有地物光谱信息的辐射信号, 获得包含光谱信息和空间信息的三维高光谱图像, 在光谱解混、 图像分类、 目标检测等方面取得了广泛的应用。 近年来, 随着遥感技术的发展及人们对获取目标准确位置的需求逐渐加大, 目标检测取得了较快的发展。 根据是否提前掌握目标光谱作为先验信息, 目标检测分为光谱匹配检测和异常检测。 光谱匹配检测需要目标光谱作为先验信息, 通常检测精度较高、 效果较好。 而异常检测不需要先验信息, 应用范围更广, 但是检测精度通常低于光谱匹配检测。 由于实际应用中缺少完备且实用的光谱库, 先验信息的获取较为困难, 不需要先验信息的异常检测成为研究的热点。 针对异常检测与光谱匹配检测相比精度较低的问题, 提出一种基于近似后验信息的高光谱异常检测算法。 首先利用矩阵分解算法对原始高光谱图像数据进行矩阵分解, 得到纯净的背景矩阵与包含噪声的异常矩阵。 舍弃异常矩阵, 将得到的背景矩阵作为近似背景信息。 然后计算图像所有像元光谱向量与背景矩阵中均值向量的马氏距离对图像进行初始异常检测, 得到初始异常像元, 将初始异常像元光谱取均值作为近似目标信息。 最后将近似背景信息与近似目标信息作为先验信息, 进行正交子空间投影得到最终的异常检测算法。 将本算法作用于图像中所有像元, 得到对整幅图像的异常检测结果。 为证明本算法的优良效果, 采用一组仿真数据和一组AVIRIS真实高光谱数据进行实验, 并与RX, LRX和LSMAD算法进行对比。 实验表明, 无论是从定性的角度还是定量的角度来看, 该算法能够有效抑制噪声, 在信噪比较低的情况下仍然可以有效地检测出图像中的异常目标, 检测精度较高并且对检测效率的影响不大, 取得了较好的检测效果。

Hyperspectral Remote Sensing technology records radiation signals with spectral information of objects through imaging spectrometers to obtain three-dimensional hyperspectral images containing spectral information and spatial information. It has been widely used in spectral unmixing, image classification, and target detection. In recent years, with the development of remote sensing technology and the increasing demand for accurate location of targets, target detection has achieved rapid development. According to whether the target spectrum is grasped in advance as a priori information, target detection is divided into spectrum matching detection and anomaly detection. Spectrum matching detection requires the target spectrum as a priori information, and usually has higher detection accuracy and better results. The anomaly detection does not require prior information and has a wider application range, but the detection accuracy is usually lower than that of spectral matching detection. Due to the lack of a complete and practical spectral library in practical applications, it is difficult to obtain prior information, and anomaly detection that does not require prior information has become a research hotspot. This paper proposes an Approximate Posterior Information-based Hyperspectral Anomaly Detection Algorithm. First, the matrix decomposition algorithm is used to decompose the original hyperspectral image data to obtain a pure background matrix and an anomaly matrix containing noise. The anomaly matrix is discarded, and the obtained background matrix is used as approximate background information. Then calculate the Mahalanobis distance between the spectral vector of all the pixels in the image and the mean vector in the background matrix to perform initial anomaly detection on the image to obtain the initial anomaly. Finally, the approximate background information and approximate target information are used as prior information, and orthogonal subspace projection is performed to obtain the final anomaly detection algorithm. Applying this algorithm to all the pixels in the image, we get the anomaly detection result for the whole image. In order to prove the excellent effect of this algorithm, a group of simulation data and a group of AVIRIS real hyperspectral data were used for experiments, and compared with RX, LRX, LSMAD algorithms. Experiments show that the algorithm can effectively suppress noise, both from a qualitative perspective and a quantitative perspective, and can still effectively detect anomalous targets in the image when the signal-to-noise ratio is relatively low. The effect of detection efficiency is small, and good detection results have been achieved.