Recently, autoencoder (AE)-based hyperspectral anomaly detection methods have demonstrated excellent performance on hyperspectral images (HSIs). The AE can simultaneously reconstruct both the anomaly targets and the background, but the lack of prior information limits the ability to detect anomalies. This study proposes a novel hyperspectral anomaly detection method based on a guided AE to reduce the feature representation for the anomaly targets. First, a multilayer AE network with skip connections is proposed to fully extract the abundant latent features from HSIs and enhance the expressive ability of the network. The reconstructed HSI can be obtained by the proposed AE network. Second, to suppress the anomaly targets in the obtained reconstructed HSI and better represent background features, a guided module based on a guided image is added to the network to reduce the feature representation of the anomaly targets by providing feedback information. Moreover, the guided image is calculated using a proposed spectral similarity method that uses the local spatial features of the HSI. Finally, we use the reconstruction error as a performance metric and compare the results of our proposed method with other state-of-the-art methods on six real-world HSIs. The results demonstrate the effectiveness and superiority of the proposed method.

Hyperspectral anomaly detection is a challenging task due to the inaccurate evaluation of background statistics and the contamination of anomaly pixels. In this paper, we propose an effective hyperspectral anomaly detection algorithm based on local joint subspace process and support vector machine (SVM). The method mainly consists of three steps. At first, in the local joint subspace process, we combine Mahalanobis distance detection and spectral angular distance detection of subspace projection by local sliding change dual windows. With this local joint subspace process, the spatial-spectral information is jointly utilized and the local subspace score is also obtained. Then, to further improve the detection probability, we propose the support vector score by classifying the dimensionally reduced hyperspectral image data through a trained SVM model. Finally, we obtain the final detection result from the linear combination of above scores. Compared with alternative methods, the proposed method has shown a superior performance on both synthetic and real-world datasets.

Anomaly detection (AD) in hyperspectral target detection is of particular interest because no prior knowledge of ground object spectra is required. However, it is difficult to utilize the salient features of hyperspectral image (HSI) and mitigate the effects of noise in hyperspectral AD, which greatly limits the detection performance. Here, we report a strategy to implement hyperspectral AD by visual attention model and background subtraction with adaptive weight. Through band selection method, the most discriminating bands are selected as the input images for subsequent processing. Then, the hyperspectral visual attention model is introduced, for the first time, into hyperspectral AD for extracting the salient feature map of the input images. Furthermore, the background subtraction process that can reduce the background and noise in the salient feature map is developed via curvature filter. Using this operation, the initial anomaly area map is obtained. Finally, incorporating with the spectral information, an adaptive weight map is applied to the initial anomaly area map to further suppress the background. In the experiment, the proposed method is compared with seven other state-of-the-art methods on synthetic and real-world HSI. Most importantly, the results demonstrate that the proposed method is effective and performs better than alternative methods. We believe that this method can open a new avenue of visual processing methods for hyperspectral AD.