Date of Award

2024

Document Type

Thesis

Degree Name

Master of Science (MS)

Department

Computer Science

Committee Chair

Tathagata Mukherjee

Committee Member

Manil Maskey

Committee Member

Chaity Banerjee Mukherjee

Research Advisor

Tathagata Mukherjee

Subject(s)

Materials--Analysis, Machine learning, Hyperspectral imaging

Abstract

Hyperspectral unmixing is essential in determining material compositions in mixed samples with a larger number of endmembers, but achieving precise decom- position remains challenging due to the limited work done, the presence of spectral variability, and endmember complexities. This study proposes many machine learning methods and techniques to identify the material composition in mixed samples (composites). Other methods to solve this problem mainly focus on a very small number of endmembers; our approach works with many end members, improving the model’s robustness. The proposed methods effectively tackle spectral variability, high dimensionality, and the presence of similar endmembers by integrating a multilayer perceptron model with Fast Fourier Transform feature extraction of the derivatives features, an original sampling technique, and a weighted loss function. The findings illustrate the robustness of the model in different datasets with the importance of using synthetic and real data for training the underlying models. The results show that the proposed methods improve the performance, with applications to material characterization and land cover mapping.

Available for download on Friday, June 12, 2026

Share

COinS