Hyperspectral Abnormal Target Detection Based on Extended Multi-attribute Profile and Fast Local RX Algorithm

In order to further improve the speed and accuracy of hyperspectral abnormal target detection, a fast anomaly target detection method based on extended multi-attribute profiles and improved Reed-Xiaoli is proposed. Extended multi-attribute Profiles are extracted from the original hyperspectral images by mathematical morphological transformations. Moreover, a novel fast local Reed-Xiao algorithm is also proposed. Iteratively update inverse matrix of covariance using matrix inverse lemma, thereby reducing the computational complexity of the Mahalanobis distance. The combination of extended multi-attribute profiles and fast local Reed-Xiaoli detector effectively utilizes the spectral information and spatial information of hyperspectral images, it greatly improves the detection accuracy and reduce the running time. Experimental results on three real data sets show the AUC value of the algorithm in this paper is 0.996 7, 0.985 6 and 0.981 6 respectively. The operation time is 21.218 1 s, 15.192 8 s and 32.337 9 s respectively. The proposed method has obvious advantages in detection accuracy and speed, and has good practical value.