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.
1. Introduction
Hyperspectral imaging (HSI) captures hundreds of narrow spectral bands, allowing for precise material identification. However, two major hurdles persist: spectral variability (the same material appearing different due to lighting) and spatial/semantic change (objects appearing at different angles or positions). Existing tools like Attribute Profiles (APs) are effective at capturing object shape and size but are highly sensitive to rotation and shift. If a building is oriented differently in various parts of a scene, standard APs produce inconsistent feature vectors, leading to classification errors.
2. Methodology
The core innovation of this work is the Invariant Attribute Profile (IAP), a joint spatial-frequency extractor that ignores irrelevant geometric variations. The framework consists of two primary components:
2.1 Spatial Invariant Features (SIFs)
SIFs are constructed in the Cartesian coordinate system using isotropic filters—kernels that maintain the same appearance from any direction—to extract robust convolutional features. These features are then aggregated using SLIC superpixel segmentation, which groups pixels into semantically meaningful objects rather than treating them as isolated points.
2.2 Frequency Invariant Features (FIFs)
FIFs mathematically enforce invariance by transforming image gradients into a Fourier polar coordinate system. Because spatial rotation corresponds to a circular shift in the phase of a Fourier transform, the authors use self-steerable Fourier bases to neutralize rotation effects. This yields three types of frequency features: magnitude, absolute rotation-invariant features, and relative rotation-invariant features.
3. Experimental Results
The IAP method was evaluated against state-of-the-art techniques, such as Extinction Profiles (EEPs), across three benchmark datasets: Pavia University, Houston 2013, and Houston 2018.
- Houston 2018: IAPs achieved an Overall Accuracy (OA) of 83.68%, a substantial improvement over baseline spectral features (65.85%) and EEPs (75.09%).
- Houston 2013: The method improved accuracy for challenging urban classes like "Commercial" and "Highway" by at least 10% compared to other profile-based methods.
- Visual Quality: Classification maps produced by IAPs are noticeably cleaner, with fewer "salt-and-pepper" noise artifacts and sharper boundaries between land-cover types.
4. Conclusion
Invariant Attribute Profiles provide a robust solution for HSI classification by addressing the limitations of traditional spatial feature extractors regarding rotation and shift. While highly effective for structured urban objects, future research will focus on extending these invariance principles into end-to-end deep learning architectures to better handle the chaotic textures of irregular natural classes like grass or scattered trees.
Summary updated