Invariant Attribute Profiles: A Spatial-Frequency Joint Feature Extractor for Hyperspectral Image Classification

Abstract: Up to the present, an enormous number of advanced techniques have been developed to enhance and extract the spatially semantic information in hyperspectral image processing and analysis. However, locally semantic change, such as scene composition, relative position between objects, spectral variability caused by illumination, atmospheric effects, and material mixture, has been less frequently investigated in modeling spatial information. As a consequence, identifying the same materials from spatially different scenes or positions can be difficult. In this paper, we propose a solution to address this issue by locally extracting invariant features from hyperspectral imagery (HSI) in both spatial and frequency domains, using a method called invariant attribute profiles (IAPs). IAPs extract the spatial invariant features by exploiting isotropic filter banks or convolutional kernels on HSI and spatial aggregation techniques (e.g., superpixel segmentation) in the Cartesian coordinate system. Furthermore, they model invariant behaviors (e.g., shift, rotation) by the means of a continuous histogram of oriented gradients constructed in a Fourier polar coordinate. This yields a combinatorial representation of spatial-frequency invariant features with application to HSI classification. Extensive experiments conducted on three promising hyperspectral datasets (Houston2013 and Houston2018) demonstrate the superiority and effectiveness of the proposed IAP method in comparison with several state-of-the-art profile-related techniques. The codes will be available from the website: https://sites.google.com/view/danfeng-hong/data-code.

• The paper proposes Invariant Attribute Profiles (IAPs) as a novel spatial-frequency joint feature extractor designed to obtain invariant features for hyperspectral image classification.
• Experiments demonstrate that IAPs significantly outperform state-of-the-art methods, achieving substantial improvements in accuracy metrics on challenging hyperspectral datasets like Houston 2013 and 2018.
• This research enhances accurate land use and land cover classification critical for remote sensing and earth observation applications, with future potential for integration with deep learning frameworks.

Invariant Attribute Profiles: A Spatial-Frequency Joint Feature Extractor for Hyperspectral Image Classification

The paper "Invariant Attribute Profiles: A Spatial-Frequency Joint Feature Extractor for Hyperspectral Image Classification" introduces an innovative approach for enhancing hyperspectral image (HSI) classification by addressing significant challenges in the extraction of invariant features from these complex datasets. Previous methods have largely neglected the variability induced by local semantic changes in spatial domains, which can obscure material identification when the scene composition changes or when spectral variabilities arise due to environmental factors.

To overcome these limitations, the authors propose a novel feature extraction strategy termed invariant attribute profiles (IAPs). The IAPs operate jointly in the spatial and frequency domains, thereby capturing more robust features that are invariant to changes in the geographic and frequency behaviors of the spectral data. Their method involves the employment of isotropic filter banks and Fourier transformations, combined with spatial aggregation techniques (superpixel segmentation), to model the invariant features efficiently.

The research highlights how IAPs robustly model various changes by implementing a continuous histogram of oriented gradients (HOG) in a Fourier polar coordinate system. This approach addresses both spatial and spectral variabilities, offering a comprehensive feature representation that significantly improves HSI classification accuracy.

Key findings of their experiments are underscored by substantial improvements in classification performance metrics—such as overall accuracy (OA), average accuracy (AA), and Kappa coefficient (κ)—on challenging datasets, including those from the Houston 2013 and 2018 IEEE GRSS data contests. These results are compared against state-of-the-art methods, with IAPs consistently outperforming various attribute profile-related techniques, especially in scenarios involving complex scene variability.

The practical implications of this work manifest in accurate land use and land cover classification, crucial for applications in remote sensing and earth observation. Theoretically, the research fosters advancement in feature extraction technology, offering methodological insights into handling spatial-frequency domain challenges.

Future developments speculated within the paper include supervised or semi-supervised approaches for end-to-end learning integration, potentially harnessing deep learning frameworks to enhance automatic feature extraction and classification systems further. Such advancements could propel the efficacy of HSI analysis, particularly in dynamically complex environments where invariant feature extraction is paramount. This paper makes a solid contribution to the domain, proposing solutions that could bridge existing gaps between spectral analysis and advanced machine learning methodologies in remote sensing applications.