介绍了一种基于特征层融合的异常检测算法。目前，其他的目标检测算法都需要知道有确定类别标记的样本，而一般的异常检测则是利用统计特征差异分割出图像中不同于背景的点。此方法减少了对先验信息的依赖，但是其结果存在较大虚警。提出的异常检测算法是利用低概率检测算法对高光谱数据先进行特征层融合，再进行分割、提取异常点，其结果降低了虚警和漏警。用这一方法对OMIS系统产生的数据进行了处理，取得了较好的结果。

An anomaly detection approach based on feature fusion is presented in this paper.All the detection algorithms,aside from anomaly detection,require training pixels of the desired class.Anomaly detection is the detection of scene elements that appear unlikely with respect to a probabilistic feature of the scene.The method needs on prior information,but the result has much false alarm.In this paper,we use low probability detection to fuse the data in feature level;then segment the image and detect anomaly elements.The result eliminates much false alarm and improves the detectability.We apply the method to the data produced by OMIS system and achieve satisfying results.