The Research Of Hyperspectral Abundance Estimation Model Based On Spectral Characteristics

Hyperspectral remote sensing has been widely used in more and more fields nowadays, such as the oil spilling analysis and chlorophyll estimation in green plants. But the existence of mixed pixels has always been the obstacle to the study of this problem. To decompose the mixed pixels people always turn to the traditional method of LSM (Least Squares Method) now. But its main drawback is that it involves a large amount of matrix operations, especially regarding to the huge dimension of hyperspectral images, and it will take much time. This is not suitable for the situation which requires a real-time answer.Motivated by this, in this thesis we have developed a new model of endmember abundance estimation which referred to as AEMSC (Abundance Estimation Model based on Spectral Characteristics). The main contributions of the thesis are summarized as follows:Firstly, we build the model by calculating NSAM (normalized SAM) and NSID (normalized SID). With the simulated endmembers and mixed pixels, we get the results of NSAM and NSID. So we can plot the point between abundance and NSAM/NSID. And then, we get the AEMSC through curve fitting.Secondly, to test and verify the accuracy of the model, we conduct our oil slick experiment. With the data we compose the mixed pixels, and choose different pixels to conduct the model which turns out different error rate. So we get the variance trend of error rate with different pixels which can be used to choose appropriate points in application.Finally, we further conduct its application in the real hyperspectral oil spilling images which comes from Peng-lai19-3C platform. After preprocessing, we conduct the model if only there are a small amount of pixels to be decomposed. And with the model we get the abundance in all pixels which we compares with the results through LSM. The results of simulation experiments, and theory demonstrate that the proposed model AEMSC outperforms LSM in terms of efficiency. In specific, the time can be reduced by33.55%at most.