Date of Award
2020
Document Type
Dissertation
Degree Name
Doctor of Philosophy (PhD)
Department
Electrical and Computer Engineering
Committee Chair
W. David Pan
Committee Member
Dongsheng Wu
Committee Member
Earl Wells
Committee Member
Huaming Zhang
Committee Member
Seong-Moo Yoo
Subject(s)
Hyperspectral imaging., Machine learning., Image compression., Image processing--Digital techniques.
Abstract
Hyperspectral imaging technology has found many useful applications in various domains such as remote sensing. Data compression allows for efficient storage and transmission of massive hyperspectral image datasets. In this dissertation, we study efficient predictive coding schemes for lossless compression of hyperspectral images. We use machine learning techniques to improve the following two key components of the predictive coding process: (i) accurate pixel value prediction, and (ii) more efficient entropy coding of the prediction errors (residues). To this end, we propose an adaptive filtering framework based on concatenated neural networks, which are capable of extracting both spatial and spectral correlations for accurate pixel value prediction. The neural networks have the unique feature of not requiring pre-training before functioning as an adaptive filter. A shallow network with only two hidden layers has been found to be effective in reducing prediction errors, with low computational complexity. For efficient entropy coding of prediction residuals using Golomb-Rice codes, we propose a universal coding parameter estimation method using deep belief networks, without relying on any assumption on the probability distribution of the input data. We formulate the problem of selecting the best coding parameter for a given input sequence as a supervised pattern classification problem. Simulations on the synthesized data and hyperspectral image datasets show that we can achieve significantly more accurate estimation of the coding parameters than three state-of-the-art methods. More extensive simulations on standard CCSDS test datasets show that the proposed methods achieve significant improvements over several other state-of-the-art lossless compression methods, including standard compressors such as ESA, CCSDS-122, and CCSDS-123.
Recommended Citation
Jiang, Zhuocheng, "Efficient predictive lossless hyperspectral image compression using machine learning" (2020). Dissertations. 208.
https://louis.uah.edu/uah-dissertations/208