Research On Multi-dimensional Information Extraction And Object Detection Algorithms Of Polarized Hyperspecral Images

The polarized hyperspectral imaging has the both advantages of thehyperspectral imaging and the polarized imaging. The polarized hyperspectralimages include polarimetric, spectral, spatial and radiant information. These providea much richer information about objects and scenes than traditional intensity orspectrum images, so the ability of object detection would be increased. Thus thisimaging technique has aroused extensive concern in the last few years. Meanwhile,the research of object detection for polarized hyperspectral imageries can bettermeet the needs of practical application. To detect objects of polarized hyperspectral,this paper does some researches as following.First of all, we study the polarized hyperspectral imaging mechanism and theessential theory of signal deeply. Then the leading factors affecting imaging areanalyzed. Thus we design the imaging experiments, and analyze the polarizationcharacteristics of objects.Second, we research how to extract the information effectively. The polarizedhyperspectral images include multi-dimensional information. To use the acquiredinformation fully, the dissertation extracts polarimetric, spectral and spatialinformation, which compose of polarimetric spectrum data set and polarimetric datacube. Then the polarimetric data cube is fused by different methods.Third, the object detection algorithms are carried out. We study statisticalanomaly detection algorithms and posteriori polarimetric spectral matchingalgorithms in this paper. For the latter algorithm, we utilizes UFCLS to extractposteriori information, then matched filtering technique like OSP and CEMalgorithms are used to detect objects. The ROC curves show that the results of thesecond algorithm are superior to the results of the former. However the formeralgorithm’s precision is subject to the accuracy of posteriori information.Finally, the detection results of different confidence are fused. The results ofpolarimetric spectrum data set and polarimetric data cube may have different beliefvalues. Thus fuzzy integral method is introduced to enhance the detectionperformance. The experiments show that the fuzzy integral results have a higheraccuracy and reliability.