Spectral Unmixing Based On A Hybrid GORT Model

As the complexity of target and its environment, the variability of imaging environment,the limitations of sensor performance and other factors, the obtained spectra often show amixed state. It greatly constrains the range and depth of optical remote sensing application,so, in recent years, spectral unmixing and its application become an important research topicof remote sensing.At present, most of the studies for the mixed spectra began from spatial characteristicsof the image, it did not reflect the physical mechanism of the mixed spectra, and it was oftenconfined to such a situation—the mixed spectra of that the target and the environment isadjacent in the same horizontal plane, this is not appropriate for the situation that the targetand the environment in the same vertical plane. In the actual situation, the latter situationsuch as natural target (soil) or non-natural target (artificial target) covered by trees with acertain degree also widely exists. In this paper, spectral unmixing for the second situation isdone to get the target reflectance, on the one hand, it does a more detailed research onvegetation physical models, and on the other hand, it promotes the practical applicationvalue of optical remote sensing.This paper studied on the natural target (soil) or non-natural target (artificial target)covered by discrete sparse vegetation canopy, the target and environment were in the samevertical plane.For the mixed type of canopy geometry and distribution, in this paper, we firstdescribed the characteristics and application objects of a variety of canopy BRDF models, aswell as the research status of some commonly used inversion methods, and then weselected the GO-RT hybrid model for the forward modeling, the hybrid model consistsof geometrical optics model and radiative transfer model, it combines both the advantagesand is connected by the between-gap probability and within-gap probability.In addition, for the purpose of this research—spectral unmixing, after sensitivityanalysis for a series of parameters to be inversed in the forward vegetation physicalmodel, we chose the multi-stage inversion method to inverse all parameters, and at last wegot the target reflectance with lower error, where the inversion algorithm usingparticle swarm optimization.