首页 > , Vol. , Issue () : -
Objective Hyperspectral anomaly detection is used to identify the pixels which have significant spectral contrast with their surrounding pixels, and plays a valuable role in military and civilian fields due to the characteristic that the priori spectral information is not required. The existing local contrast based methods usually adopt dual rectangular windows scheme for hyperspectral anomaly detection, however, they empirically set the size of dual window, which limits their generalization capability. Method To address above issue, a hyperspectral anomaly detection method via combining adaptive window saliency detection and improved superpixel segmentation is proposed in this paper. To reduce the computation complexity of the proposed method, an adversarial autoencoder is firstly introduced to reduce the dimension of the hyperspectral image. Secondly, the compressed hyperspectral image is segmented by improved superpixel segmentation. The existing spectral distance measurements used in the superpixel segmentation is effective under the condition that the relationship between the spectral value and the intensity of each pixel is linear, however, the condition cannot be guaranteed in practical applications. To solve this problem, the improved superpixel segmentation adopts the orthogonal projection divergence to measure the spectral distance. Afterwards, an adaptive window based saliency detection algorithm is proposed and used to obtain the initial detection results. Specifically, the size of the inner window is adaptively determined by the superpixels, which ensures that the pixels belonging to the same inner window are homogeneous, and the outer window can be obtained by enlarging the inner window with fixed size. Finally, to reduce the false alarm rate, the domain transform recursive filter and thresholding operation are employed to optimize the initial detection results. Result The comparisons between the orthogonal projection divergence and three common spectral distance measurements (euclidean distance, spectral angular mapping and spectral information divergence) in terms of AUC show that the orthogonal projection divergence based method gets the highest score on 5 datasets. The comparisons between the adaptive window and traditional manual setting dual window in terms of AUC show that the adaptive window based method gets the highest score on 5 datasets. To validate the overall performance of the proposed method, the comprehensive comparisons between proposed method and 7 state-of-the-art methods on 5 public datasets are implemented. To be specific, the subjective comparisons show that the anomalous pixels detected by proposed method are more precise and have stronger contrast with background regions, the objective comparisons demonstrate that the overall detection accuracy of proposed method is highest and separability between the anomalous pixels and background pixels of proposed method is best. Conclusion In summary, three conclusions can be derived from this paper. First of all, the improved superpixel segmentation algorithm can improve the segmentation results and the proposed adaptive window scheme can improve the performance of saliency detection. Secondly, the proposed method has excellent detection accuracy, false alarm rate and separability between the anomalous pixels and background pixels. Finally, the overall performance of the proposed method is superior to state-of-the-art methods.