Background whitened target detection algorithm for hyperspectral imagery

Hyperspectral remotely sensed imagery has undergone rapid advancements recently. Hyperspectral sensors collect surface information with hundreds of channels which results in hundreds of co-registered images. To process this huge amount of data without information of the scene is a great challenge, especially for anomaly detection. Several methods are devoted to this problem, such as the well-known RX algorithm and highmoment statistics approaches. The RX algorithm can detect all anomalies in a single image but it cannot discriminate them. On the other hand, the high-moment statistics approaches use criteria such as Skewness and Kurtosis to find the projection directions recursively, so it is computationally expensive. In this paper, we propose an effective algorithm for anomaly detection and discrimination extended from RX algorithm, called Background Whitened Target Detection Algorithm (BWTDA). It first models the background signature with Gaussian distribution and applies whitening process. After the process, the background will be indepenent-identical-distributed Gaussian in all spectral bands. Then apply Target Detection Process (TDP) to search for potential anomalies automatically and Target Classification Process (TCP) for classifying them individually. The experimental results show that the proposed method can improve the RX algorithm by discriminating the anomalies and outperforming the original high-moment statistics approach in terms of computational time.