Research on the Effective Use of apriori Information in Image Interpretation

点击翻译按钮获取中文摘要

For the accurate land-use and land-cover analysis of satellite and aerial images, as well as for automatic diagnosis and surgical planning based on medical images, we are aiming at establishing image-interpretation frameworks using a priori knowledge inherent to the subject under consideration. In the study of high-resolution images as well as hyperspectral images, Models with excessive freedom sometimes mislead us to inconsistent results. Sufficient information can be obtained by introducing models with adequately limited degree of freedom, as shown in the following results :Damaged area estimation : The satellite image photographed before disaster was handled as a priori information, and the image photographed after the disaster was compared in nonlinear map algorithm based on the principle of "coincidence enhancement", and change detection of damaged houses was carried out. This system was applied to satellite images of the Bam earthquake of December 26, 2003, and the damaged area w … More as extracted. The processing result was provided to a Japanese field investigation group, and the validity of the detection result was verified via the field study.Sugar content estimation of crops : The flavor of crops is dependent on the fraction of contents such as sugar, amino acid and protein etc. In this study, we tried to estimate these quantities from the hyperspectral image of the crops before the harvest. In the estimation system, we optimized the mathematical operation among large number of wavelength of the hyperspectral images by combining the neural network with the PSO(particle swarm optimization) algorithm. In this estimation, the set of measured values obtained after the harvest was used as a tutorial data.The derivation of human skin extraction index : We analyzed hyperspectral image of short wavelength infrared region, to make use of a priori information effectively in detecting the human victims in the water or lying on the gravel or ground after the disaster. As a result, NHI(normalized human index) was proposed to make the skin tangible by utilizing the spectral information inherent to the human skin.Independent Component Analysis of Periodically Distributed Vegetation Images : As a basic study for the development of high-altitude observation of hyperspectral images of Japanese agricultural fields, we experimentally and theoretically consider the use of a priori knowledge in the effective use of independent component analysis for resolving mixture pixels found in images observed beyond the limitation of physical resolution. Less