A Laboratory-created Dataset And A Study For Hyperspectral Unmixing

Hyperspectral imaging is a continuously growing field of study for its rich spectral information.It can be applied in a wide range of fields,such as agriculture,geology,mil-itary and environment.The combination of two-dimensional image and rich information in the spectral dimension provides solutions to many problems that cannot be solved by traditional RGB imaging or multispectral imaging.However,because of factors such as the low spatial resolution of the spectral imaging devices,and the diversity of materials in a scene,an observed pixel may contain several materials and affect the accuracy of further target analysis.Spectral unmixing plays an important role in hyperspectral data processing.A significant amount of effort has been made in the past decade to solve the spectral unmixing problem.The major work of this thesis consists of the following three aspects.(1)Spectral unmixing has been extensively studied and a variety of unmixing al-gorithms have been proposed in the literature.However,the lack of publicly available datasets with ground truth makes it difficult to evaluate and compare the performance of unmixing algorithms in a quantitative and objective manner.Most of the existing works rely on the use of numerical synthetic data and an intuitive inspection of the results of real data.To alleviate this dilemma,this study design three experimental scenes,includ-ing printed checkerboards to mimic linear model,mixed quartz sands to mimic intimated model,and reflection with a vertical board to mimic bilinear model.A dataset is then created by imaging these scenes with the hyperspectral camera in our laboratory,pro-viding 36 mixtures with more than 1.3×10~5pixels with 256 wavelength bands ranging from 400 to 1000 nm.The experimental settings are strictly controlled so that pure ma-terial spectral signatures and material compositions are known.Some typical linear and nonlinear unmixing algorithms are also tested with this dataset and lead to meaningful results.(2)Nonlinear spectral unmixing is an important and challenging problem in hyper-spectral image processing.Classical nonlinear algorithms are usually derived based on specific assumptions on the nonlinearity and its generalization capability is limited.In recent years,deep learning shows its advantage in addressing general nonlinear problems.However,existing ways of using deep neural networks for unmixing are always supervised method,they need a training set with known ground truth,which is often generated by other unmixing approaches.In this study,we develop a novel blind hyperspectral un-mixing scheme based on a deep autoencoder network.Both encoder and decoder of the network are carefully designed so that we can conveniently extract estimated endmem-bers and abundances simultaneously from the nonlinearly mixed data without training data with ground truth.Experimental results validate the proposed scheme and show its superior performance.(3)Hyperspectral image is often affected by shadows caused by vegetation,electric utility poles,and buildings.The shadowed regions of a hyperspectral image affect the accuracy of classification and target detection algorithms.Therefore,removing shadows is important for both enhancing the interpretability of the data and further target anal-ysis.Shadow removal approaches based on spectral unmixing have been proposed in the literature using the linear mixture model.However,objects that produce shadows may also introduce light scattering,and the higher-order interactions of photons can cause nonlinearity.Using linear mixing model is improper.This work integrates the nonlinear hyperspectral unmixing in the unmixing-based shadow removal,and the effects of apply-ing typical nonlinear algorithms within the approach are investigated.The usefulness of nonlinear unmixing in hyperspectral shadow removal is verified based on the results of applications to both laboratory-created real data and actual airborne data.