Hyperspectral anomaly detection (HAD) is a research endeavor of high practical relevance within remote sensing scene interpretation. In this work, we propose an unsupervised approach, dual feature extraction network (DFEN) for HAD, to gradually build up ever-greater discrimination between the original data and background. In particular, we impose an end-to-end discriminative learning loss on two networks. Among them, adversarial learning aims to keep the original spectrum while Gaussian constrained learning intends to learn the background distribution in the potential space. To extract the anomaly, we calculate spatial and spectral anomaly scores based on mean squared error (MSE) spatial distance and orthogonal projection divergence (OPD) spectral distance between two latent feature matrices. Finally, the comprehensive detection result is obtained by a simple dot product between two domains to further reduce the false alarm rate. Experiments have been conducted on eight real hyperspectral data sets captured by different sensors over different scenes, which show that the proposed DFEN method is superior to other compared methods in detection accuracy or false alarm rate.

高光谱异常检测（HAD）是遥感场景解释中具有高度实际意义的研究工作。在这项工作中，我们提出了一种无监督的方法，即用于HAD的双特征提取网络（DFEN），以逐步建立原始数据和背景之间越来越大的区分度。特别是，我们在两个网络上施加了端到端的歧视性学习损失。其中，对抗学习旨在保持原始频谱，而高斯约束学习则旨在学习潜在空间中的背景分布。为了提取异常，我们基于两个潜在特征矩阵之间的均方误差（MSE）空间距离和正交投影散度（OPD）光谱距离来计算空间和光谱异常分数。最后，通过两个域之间的简单点积获得全面的检测结果，以进一步降低误报率。对不同传感器在不同场景下捕获的八个真实高光谱数据集进行了实验，表明所提出的DFEN方法在检测精度或误报率方面优于其他比较方法。