Kernel RX Algorithm Of Anomaly Detection And Multiple Classifier Fusion Technology

Hyperspectral remote sensing image data contains abundant spatial imageinformation and spectral radiation information. It is widely used in environmentalmonitoring, civilian areas, military field and natural resource exploration. This thesismain content is anomaly target detection and classification algorithm of hyperspectralimage. The structure of this thesis is as follows:(1) The chapter introduces the application background and research significanceof hyperspectral remote sensing image, and summarizes situation of the researches athome and abroad. At the same time, it describes the current research situation ofanomaly detection and classification technology.(2) In this paper, we introduce the traditional RX algorithm of anomaly detectiontechnology, and present several kinds of modified RX anomaly detection algorithms.According to by comparing these algorithms, we know the advantages anddisadvantages of them.(3) The traditional RX algorithm and a single kernel RX algorithm of furtherresearch, a weighted combination kernel RX algorithm is proposed in this paper. It notonly has the limitation of the Gaussian kernel function, but also the globalperformance of polynomial kernel function. The weighted combination kernelfunction has spectral characteristics and space characteristics on structural dimension,so, it can be more efficient access to land feature information. The hyperspectral dataprovided by the Institute Changchun Optical Precision Machinery and Physics,Chinese Academy of Sciences and AVIRIS data, which is captured in the Salinasvalley of California Indicate the effectiveness of the method.(4) One kind of multiple classifier fusion technology based on Anomalydetection is proposed in the fourth chapter. AdaBoost algorithm and SVM algorithmare two kinds of classification algorithms. They combine with RX anomaly detectionalgorithm to classify of target. Then, the two kinds of classification resultsfuse together by using decision level fusion technology. Using RX anomaly detectiontechnology in order to detect most parts of anomaly targets and ruling out mostbackground points. The experimental result shows that the algorithm not only canimprove the accuracy of classification, but also improve efficiency of classification. In order to accord with the practical application, the experiment reduces manualintervention. So, in the selection of training samples, we choose spectral fingerprintsas the training sample.