Study On Algorithms Of Local Anomaly Detection For Hyperspectral Remote Sensing Imagery

With the advance of spatial-resolution and spectrum-resolution of hyperspectral remote sensing imagery, the technology of target detection is rapidly developed. The anomaly detector can detect targets whose signatures are spectrally distinct from their surroundings without prior knowledge, and has become a hot spot in the field of target detection. To detect the weak anomaly targets, the research on hyperspectral local anomaly detection algorithm is done in image space and spectrum space. The main work of this paper is as follow.Firstly, several typical anomaly detections are implemented, and the comparison and analysis is given. Through the experiment on actual data, the difference on global detection between statistical model and linear mixed model is analyzed, the problem on weak anomaly detection is summarized, and the conceptions of local anomaly detection on image space and spectrum space are shown.Secondly, aiming at image space, a local anomaly detection approach based on center projection is proposed. In the anomaly detection, we almost focus on the isolate targets. To solve the detection problem for the roads and edges of ground which have a strong anomaly degree, an approach based on center projection is developed. With the consideration of space relation among the objects, it can detect the small anomalies, and the false alarm of roads is reduced.Finally, aiming at spectrum space, according to the characteristic of the neighbor of spectrum space,combined the mechanism of anomaly detection, a principle of division in spectrum space is proposed. According to the principle, the K-Mean cluster algorithm is adopted. The performance is validated by the actual data, and the selection of class number is studied. At last, the difference of the performance of local anomaly detection between image space and spectrum space is compared.