Methodologies for earth observation with hyperspectral analysis and multimodal image fusion

Student thesis: Doctoral Thesis


Remote sensing has been one of the most common approaches to acquire relevant information for Earth observation, using both active and passive devices. With different sensors and platforms, various aspects of the Earth's surface can be observed for analysis. Facing the considerable multimodal heterogeneous remote sensing data, how to effectively extract the key information to support the varying needs from different applications has become a big challenge. Due to different characteristics of the remote sensing datasets acquired from various sensors and conditions, effective technologies are demanded for accurate and efficient data interpretations. In this thesis, the object identification and pixel level classification task on the multimodal remote sensing data is particularly focused, especially the Hyperspectral imagery (HSI), Multispectral Instrument (MSI) and the Syntenic Aperture Radar (SAR) to facilitate different Earth observation tasks. Although HSI has rich spectral information to enable the high discrimination ability for subtle spectral differences on materials, it suffers from different sources of noise and highly correlated spectral information. Despite of various approaches which have been proposed for denoising/smoothing and data reduction, the efficacy is still affected even using the corrected dataset (the data with the water absorption bands and noise bands discarded), especially when the training samples are limited and unbalanced. In this thesis, A signal decomposing technology is introduced into HSI for more effective feature extraction and improved data classification, where a superpixelwise multiscale Prophet model (SMP) is proposed for noise-robust feature extraction and effective classification of the HSI. The prophet model can deeply dig into the complex latent structures of HSI and extract features with enlarged interclass diversity and improved intraclass similarity. Firstly, the first three principal components of the HSI are extracted for implementing the superpixelwise segmentation, where pixels are grouped into regions with adaptively determined sizes and shapes. Secondly, a multiscale prophet model is utilized to extract the multiscale informative trend components from the average spectrum of each superpixel. Taking the multiscale trend signal as the input feature, the HSI data are further classified superpixelwisely, which is further refined by majority voting based decision fusion. Experiments on three publicly available datasets have fully validated the efficacy and robustness of the proposed approach, when benchmarked with several state-of-the-art classifiers, including some typical spatial-spectral methods and deep learning classifiers. In addition, both quantitative and qualitative assessment has validated the efficacy of our approach in noise-robust classification of HSI even with limited training samples, especially in classifying uncorrected data (without pre-filtering the water-absorption and noisy bands). The proposed SMP method focuses on the multiscale noise removal in spectral domain, which shows limited performance in spatial noise filter. In order to explore spatial noise robust features whilst reducing the data dimension, a novel effective and efficient feature extraction framework is proposed for the HSI, namely Multiscale 2D singular spectrum analysis (2D-SSA) with principal component analysis (2D-MSSP). This method investigates the multiscale strategy in the spatial domain by combining the dimension reduction in the spectral domain. First, multiscale 2D-SSA is applied to exploit the multiscale spatial features in each spectral band of HSI via extracting the varying trends in different windowing scales. Taking the extracted trend signals at each scale level as the input features, the principal component analy
Date of Award30 May 2022
Original languageEnglish
Awarding Institution
  • University Of Strathclyde
SponsorsUniversity of Strathclyde
SupervisorMalcolm Macdonald (Supervisor), Jinchang Ren (Supervisor), Paul Murray (Supervisor) & Wenzhi Liao (Supervisor)

Cite this