Hyperspectral anomaly detection is a hot topic in remote sensing applications. Most of the conventional detectors are based on the Reed–Xiaoli (RX) method and assumedly targets and backgrounds follow a Gaussian distribution in which two problems exist: the outliers in the Gaussian distribution statistics limit the detection accuracy of RX method, and the larger proportions between the backgrounds and anomaly targets account for the higher false alarm rate. In this letter, a new hyperspectral anomaly detection method is proposed, which can solve the two problems mentioned above. The new method includes two improved ideas. First, third- and fourth-order moments are used as statistical features to improve the outlier peak values and highlight the targets. Second, the adaptive cosine estimation as the structural assumption for the RX method is used to suppress the backgrounds for anomalous targets. Experiments on real hyperspectral data sets suggest that our proposed method could not only effectively decrease the impact of background statistics but also improve the detection ability of such outlier values. Furthermore, comparative experimental results revealed that the proposed method achieves higher detection rates with lower false alarm rates.

中文翻译：

---

一种基于高阶统计量和自适应余弦估计器的高光谱异常检测新方法

高光谱异常检测是遥感应用中的热门话题。大多数常规检测器基于 Reed-Xiaoli (RX) 方法，假设目标和背景遵循高斯分布，存在两个问题：高斯分布统计中的异常值限制了 RX 方法的检测精度，以及较大的比例背景和异常目标之间的误报率较高。在这封信中，提出了一种新的高光谱异常检测方法，可以解决上述两个问题。新方法包括两个改进的想法。首先，使用三阶和四阶矩作为统计特征来改善异常峰值并突出目标。第二，自适应余弦估计作为 RX 方法的结构假设用于抑制异常目标的背景。对真实高光谱数据集的实验表明，我们提出的方法不仅可以有效降低背景统计的影响，还可以提高此类异常值的检测能力。此外，比较实验结果表明，所提出的方法以较低的误报率实现了更高的检测率。