Detection Of Hidden Target Suspected Based On Hyperspectral Remote Sensing

With advances in camouflage technology and the development ofcamouflage hyperspectral remote sensing technology, based on the ori ginalmulti-spectral remote sensing technology visible and hidden no longer meet therequirements of modern camouflage target detection technique, used alonehyperspectral image target detection although spectral data to reflect The basicinformation of the target, but the detection accuracy is difficult to furtherimprove. Object of this research is the use of hidden manner protectivecamouflage net goal by laboratory hyperspectral image data acquired by animaging spectrometer, biochemical analysis between plant leaves and thebackground vegetation spectral camouflage nets in the image data andquantitative parameters of the inversion information differences, featureselection, feature space and the decision to establish the classification ofhyperspectral images based on spectral function and plants characteristicbiochemical parameters, the joint detection of vegetation in the backgroundhidden targets.This thesis carried out the following three aspects of research work:First, basic research conducted plants characteristic biochemical parametersof plants quantitative inversion. Respectively quantitative inversion of severalcommonly used methods of analysis, and the results of experiments comparingvarious methods of inversion. The results can be seen by the target and thebackground of these types of quantitative data obtained by the inversion methodhas some differences. Among them, the vegetation index calculation is simple,applicability, and some vegetation index on the target and backgrounddifferences are significant, can be used as a feature to detect targets.Second, the spectral characteristics of hyperspectral image analysis features,easy to focus on selected target detection algorithm for optimal feature bandcombination from hyperspectral data cube. This paper uses Tabu Search featureselection algorithm, the evaluation function algorithm, neighborhood function,such as standards and related parameters relate the results of feature selection.Through in-depth analysis of the algorithm, the algorithm found the randomnessof the initial solution to bring the final results of feature selection is often verydifferent, given the taboo proposed algorithm based on a priori knowledge of theinitial detection of the target solution, the results obtained in this way isrelatively concentrated, are stable, compared with the random effect given totaboo search algorithm is better than the initial solution. By tabu search algorithm based on a priori knowledge of the spectral characteristics wereobtained optimal combination of image, ready for subsequent testing.Finally, the paper based on the analysis target detection theory and Fishercriterion decision function based on the design and application of the decisionfunction, respectively, based on image spectral characteristics, based on leafbiochemical parameters features, image-based spectral characteristics and leafbiochemical parameters characteristic combination of three cases, experimentswere carried out by detection of an image based on the spectral characteristicsand the effect of leaf bound biochemical characteristics than the amount usedalone feature detection. Finally, based on the above research and algorithms,completed a background of vegetation suspected hidden object detectionexperiments, to the purpose of this topic.