Research On Anomaly Detection Oriented Technologies Of Multisource Remote Sensing Imagery Fusion

There are much more multisource, multi-resolution, multi-sensor remote sensing images as the developing of remote sensing technology. These images contain both redundant information and complementary information. In this case, it turn to be a hot topic in these years that how to utilize multisource images effectively to obtain comprehensive and accurate interpretation results. However, image fusion methods are often applied based on the subsequent applications. Target recognition is one of the most important technologies of hyperspectral images. Anomaly detection is a valuable way of searching targets whose spectral characteristics are not known. The accuracy of anomaly detection is often limited, because the spatial resolution of hyperspectral image is relatively low. Therefore, this thesis discusses anomaly detection oriented technologies of multisource images fusion, to increase detection results with the aid of panchromatic images with high spatial resolution.Firstly, this thesis discusses the fusion methods of hyperspectral images and panchromatic images on pixel level. After the comparison of classical methods such principle component analysis, wavelet transformation, Schmidt orthogonal transformation and ARSIS(Amé1ioration de la Résolution Spatiale par Injection de Structures) Concept, the best method is studied more deeply. An improved method based on the ARSIS Concept is proposed. The proposed method is evaluated in image-oriented method and in application-oriented method, and its effectiveness is verified.Secondly, this thesis discusses the anomaly detection methods of hyperspectral images. Basic principles of anomaly detection and classical algorithms are studied, and their limitation is analyzed. Then a novel algorithm is proposed, which is based on elliptically-contoured model and finite mixture model. Simulation data and real data are both used in experiments. The detection results, ROC curves and computation durations are compared to other classical methods to test the effectiveness of the proposed method. Experiments show that the proposed method has higher detection rate and costs less computation time than other classical methods.Last, this thesis discusses the anomaly detection methods of multisource images. Detection results on pixel-level fusion and decision-level fusion are both taken into consideration. On pixel level, the proposed methods in the above two sections are utilized; on decision level, linear consensus theory and DempsterShafer evidence theory are used to fuse the detection results of multisource images. Datasets in different resolution are used to test these fusion methods. The results in different situations are analyzed, and suggestion of choosing fusion methods in various circumstances is given.