pith. sign in

arxiv: 2304.09730 · v1 · submitted 2023-04-19 · 💻 cs.CV

Hyperspectral Image Analysis with Subspace Learning-based One-Class Classification

Sertac Kilickaya , Mete Ahishali , Fahad Sohrab , Turker Ince , Moncef Gabbouj This is my paper

Pith reviewed 2026-05-24 09:51 UTC · model grok-4.3

classification 💻 cs.CV
keywords hyperspectral image classificationone-class classificationsubspace learningcurse of dimensionalityland use classificationdimensionality reductionimbalanced dataremote sensing
0
0 comments X

The pith

Subspace learning methods for one-class classification map hyperspectral data to a lower-dimensional space optimized for the task without separate dimensionality reduction.

A machine-rendered reading of the paper's core claim, the machinery that carries it, and where it could break.

The paper investigates subspace learning approaches that are optimized specifically for one-class classification and applies them to hyperspectral image analysis. These methods produce a reduced feature space directly suited to describing a single target class from high-dimensional spectral bands. The framework is intended to address both the exponential growth in required training samples as band count increases and the typical class imbalance in land-use mapping problems. Experiments on hyperspectral datasets are presented to show that the integrated subspace one-class pipeline can operate effectively on such data.

Core claim

Subspace learning methods for one-class classification can be applied directly to hyperspectral images so that the learned lower-dimensional representation is optimized for the classification task, removing the need for a prior band-selection or dimensionality-reduction stage while accommodating the imbalanced label distributions common in land-use and land-cover problems.

What carries the argument

Subspace learning one-class classifiers that map high-dimensional spectral vectors into a feature space optimized for single-class data description.

If this is right

  • No separate band selection or dimensionality reduction preprocessing step is required before classification.
  • The method directly accommodates the high number of spectral bands and the imbalanced class distributions typical of land-use mapping.
  • One-class classifiers trained only on the target class can still produce usable decision boundaries for hyperspectral scenes.
  • The same subspace optimization can be reused across different hyperspectral sensors or acquisition conditions without redesigning a reduction stage.

Where Pith is reading between the lines

These are editorial extensions of the paper, not claims the author makes directly.

  • The same integrated subspace approach could be tested on other high-dimensional remote-sensing modalities such as multispectral or LiDAR point clouds.
  • If the learned subspaces prove stable, they might serve as a starting point for transfer to new geographic regions with limited labeled target samples.
  • The framework leaves open whether the subspace dimensions carry interpretable spectral meaning that domain experts could inspect.

Load-bearing premise

Subspace learning procedures can be formulated to produce lower-dimensional representations that remain effective for hyperspectral data when the objective is one-class classification rather than multi-class separation.

What would settle it

On standard hyperspectral benchmark datasets, the subspace one-class classifiers would need to show accuracy or F1 scores no better than conventional pipelines that first apply separate dimensionality reduction before one-class classification.

Figures

Figures reproduced from arXiv: 2304.09730 by Fahad Sohrab, Mete Ahishali, Moncef Gabbouj, Sertac Kilickaya, Turker Ince.

Figure 1
Figure 1. Figure 1: One-class Classification of HSI Images using S-SVDD Approach [PITH_FULL_IMAGE:figures/full_fig_p002_1.png] view at source ↗
read the original abstract

Hyperspectral image (HSI) classification is an important task in many applications, such as environmental monitoring, medical imaging, and land use/land cover (LULC) classification. Due to the significant amount of spectral information from recent HSI sensors, analyzing the acquired images is challenging using traditional Machine Learning (ML) methods. As the number of frequency bands increases, the required number of training samples increases exponentially to achieve a reasonable classification accuracy, also known as the curse of dimensionality. Therefore, separate band selection or dimensionality reduction techniques are often applied before performing any classification task over HSI data. In this study, we investigate recently proposed subspace learning methods for one-class classification (OCC). These methods map high-dimensional data to a lower-dimensional feature space that is optimized for one-class classification. In this way, there is no separate dimensionality reduction or feature selection procedure needed in the proposed classification framework. Moreover, one-class classifiers have the ability to learn a data description from the category of a single class only. Considering the imbalanced labels of the LULC classification problem and rich spectral information (high number of dimensions), the proposed classification approach is well-suited for HSI data. Overall, this is a pioneer study focusing on subspace learning-based one-class classification for HSI data. We analyze the performance of the proposed subspace learning one-class classifiers in the proposed pipeline. Our experiments validate that the proposed approach helps tackle the curse of dimensionality along with the imbalanced nature of HSI data.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit. Tearing a paper down is the easy half of reading it; the pith above is the substance, this is the friction.

Referee Report

2 major / 1 minor

Summary. The manuscript proposes using subspace learning methods optimized for one-class classification (OCC) on hyperspectral images (HSI). The approach maps high-dimensional spectral data to a lower-dimensional space tailored for OCC, eliminating the need for separate band selection or dimensionality reduction steps. It targets the curse of dimensionality and label imbalance in land use/land cover classification, with the abstract claiming that experiments validate the pipeline's effectiveness for HSI data.

Significance. If substantiated, the work could simplify HSI analysis pipelines by integrating representation learning with OCC, which is relevant for applications with high spectral dimensionality and imbalanced classes. The positioning as a pioneer study in subspace learning-based OCC for HSI may stimulate related research, though the significance depends on demonstrating gains from the joint optimization rather than from the OCC component alone.

major comments (2)
  1. [Abstract] Abstract: The central claim that 'our experiments validate that the proposed approach helps tackle the curse of dimensionality along with the imbalanced nature of HSI data' lacks any reported dataset sizes, quantitative metrics, baseline comparisons, or ablation against a separate DR (e.g., PCA) + identical OCC pipeline. This evidence is load-bearing for the claim that the subspace optimization itself addresses the issues without separate DR.
  2. [Abstract] Abstract (paragraph on the proposed framework): The assertion that 'subspace learning methods can be directly optimized for one-class classification such that the resulting lower-dimensional representation is effective for hyperspectral data without any separate band selection or dimensionality reduction step' is not accompanied by any description of the optimization procedure, objective function, or how the subspace is learned specifically for OCC, preventing assessment of whether the integration is substantive.
minor comments (1)
  1. [Abstract] The abstract would be strengthened by naming the specific subspace learning methods investigated and the HSI datasets used, even at a high level.

Simulated Author's Rebuttal

2 responses · 0 unresolved

We thank the referee for the constructive feedback on the abstract. We address each major comment below and will revise the manuscript to improve clarity and support for the claims.

read point-by-point responses

  1. Referee: [Abstract] Abstract: The central claim that 'our experiments validate that the proposed approach helps tackle the curse of dimensionality along with the imbalanced nature of HSI data' lacks any reported dataset sizes, quantitative metrics, baseline comparisons, or ablation against a separate DR (e.g., PCA) + identical OCC pipeline. This evidence is load-bearing for the claim that the subspace optimization itself addresses the issues without separate DR.

    Authors: We agree that the abstract would benefit from additional context to support the validation claim. In the revision, we will add concise references to dataset characteristics (number of bands and samples), key quantitative metrics from the experiments, and a note that ablations against separate dimensionality reduction plus OCC are included in the results section. This will help substantiate the claim while respecting abstract length limits. revision: yes

  2. Referee: [Abstract] Abstract (paragraph on the proposed framework): The assertion that 'subspace learning methods can be directly optimized for one-class classification such that the resulting lower-dimensional representation is effective for hyperspectral data without any separate band selection or dimensionality reduction step' is not accompanied by any description of the optimization procedure, objective function, or how the subspace is learned specifically for OCC, preventing assessment of whether the integration is substantive.

    Authors: The subspace learning methods referenced are existing techniques optimized via one-class objectives (such as minimizing the volume of a hypersphere around target samples in the learned subspace). We will revise the abstract to include a brief mention of this optimization for OCC. Complete details on the procedures, objective functions, and integration are provided in the methodology section along with citations to the source papers. revision: yes

Circularity Check

0 steps flagged

No significant circularity; derivation is self-contained empirical validation

full rationale

The paper proposes subspace learning methods optimized for one-class classification as a joint framework for HSI data, stating that this avoids separate dimensionality reduction steps and suits imbalanced labels. The abstract and provided text contain no equations, no fitted parameters renamed as predictions, and no self-citations invoked as load-bearing uniqueness theorems. The central claim rests on experimental validation of the pipeline rather than any derivation that reduces by construction to its own inputs or definitions. No self-definitional loops, ansatz smuggling, or renaming of known results are present in the given material. This is the normal case of a methodological proposal whose performance claims are externally falsifiable via the reported experiments.

Axiom & Free-Parameter Ledger

0 free parameters · 1 axioms · 0 invented entities

Review performed on abstract only; no explicit free parameters, invented entities, or non-standard axioms are stated. The central claim rests on the domain assumption that subspace optimization for OCC transfers effectively to spectral data.

axioms (1)
  • domain assumption Subspace learning methods exist that map high-dimensional data to a lower-dimensional space optimized specifically for one-class classification.
    Invoked when the abstract states there is no separate dimensionality reduction procedure needed.

pith-pipeline@v0.9.0 · 5817 in / 1197 out tokens · 51826 ms · 2026-05-24T09:51:59.514441+00:00 · methodology

discussion (0)

Sign in with ORCID, Apple, or X to comment. Anyone can read and Pith papers without signing in.

Reference graph

Works this paper leans on

18 extracted references · 18 canonical work pages

  1. [1]

    Sparse representation based band selection for hyperspectral images

    Shuangjiang Li and Hairong Qi. Sparse representation based band selection for hyperspectral images. In IEEE International Conference on Image Processing (ICIP) , pages 2693–2696. IEEE, 2011

    work page 2011

  2. [2]

    A new sparsity-based band selection method for target detection of hyperspectral image

    Kang Sun, Xiurui Geng, and Luyan Ji. A new sparsity-based band selection method for target detection of hyperspectral image. IEEE Geoscience and Remote Sensing Letters , 12(2):329– 333, 2015

    work page 2015

  3. [3]

    Srl-soa: self- representation learning with sparse 1d-operational autoencoder for hyperspectral image band selection

    Mete Ahishali, Serkan Kiranyaz, Iftikhar Ahmad, and Moncef Gabbouj. Srl-soa: self- representation learning with sparse 1d-operational autoencoder for hyperspectral image band selection. In IEEE International Conference on Image Processing (ICIP) , pages 2296–2300. IEEE, 2022

    work page 2022

  4. [4]

    Y. Yuan, G. Zhu, and Q. Wang. Hyperspectral band selection by multitask sparsity pursuit. IEEE Transactions on Geoscience and Remote Sensing , 53(2):631–644, 2015

    work page 2015

  5. [5]

    Zhang, B

    W Sun, L. Zhang, B. Du, W. Li, and Y. M. Lai. Band selection using improved sparse subspace clustering for hyperspectral imagery classification. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 8(6):2784–2797, 2015

    work page 2015

  6. [6]

    Effi- cient hierarchical-pca dimension reduction for hyperspectral imagery

    Abhishek Agarwal, Tarek El-Ghazawi, Hesham El-Askary, and Jacquline Le-Moigne. Effi- cient hierarchical-pca dimension reduction for hyperspectral imagery. In IEEE International Symposium on Signal Processing and Information Technology , pages 353–356. IEEE, 2007

    work page 2007

  7. [7]

    Similarity-based unsupervised band selection for hyperspectral image analysis

    Qian Du and He Yang. Similarity-based unsupervised band selection for hyperspectral image analysis. IEEE Geoscience and Remote Sensing Letters , 5(4):564–568, 2008

    work page 2008

  8. [8]

    An efficient method for supervised hyper- spectral band selection

    He Yang, Qian Du, Hongjun Su, and Yehua Sheng. An efficient method for supervised hyper- spectral band selection. IEEE Geoscience and Remote Sensing Letters , 8(1):138–142, 2010

    work page 2010

  9. [9]

    Band selection for hyperspectral image classification using mutual information

    Baofeng Guo, Steve R Gunn, Robert I Damper, and James DB Nelson. Band selection for hyperspectral image classification using mutual information. IEEE Geoscience and Remote Sensing Letters, 3(4):522–526, 2006

    work page 2006

  10. [10]

    A fast volume-gradient-based band selection method for hyperspectral image

    Xiurui Geng, Kang Sun, Luyan Ji, and Yongchao Zhao. A fast volume-gradient-based band selection method for hyperspectral image. IEEE Transactions on Geoscience and Remote Sensing, 52(11):7111–7119, 2014

    work page 2014

  11. [11]

    Semisupervised hyperspectral band selection via spectral–spatial hypergraph model

    Xiao Bai, Zhouxiao Guo, Yanyang Wang, Zhihong Zhang, and Jun Zhou. Semisupervised hyperspectral band selection via spectral–spatial hypergraph model. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing , 8(6):2774–2783, 2015

    work page 2015

  12. [12]

    Support vector data description

    David MJ Tax and Robert PW Duin. Support vector data description. Machine learning , 54:45–66, 2004

    work page 2004

  13. [13]

    Subspace support vector data description

    Fahad Sohrab, Jenni Raitoharju, Moncef Gabbouj, and Alexandros Iosifidis. Subspace support vector data description. In International Conference on Pattern Recognition (ICPR) , pages 722–727. IEEE, 2018

    work page 2018

  14. [14]

    Veganzons, and Borja Ayerdi

    Manual Grana, Miguel A. Veganzons, and Borja Ayerdi. Hyperspectral remote sensing scenes. http://www.ehu.eus/ccwintco/index.php/Hyperspectral_Remote_Sensing_Scenes. [On- line; accessed January 20, 2022]

    work page 2022

  15. [15]

    Baumgardner, Larry L

    Marion F. Baumgardner, Larry L. Biehl, and David A. Landgrebe. 220 band aviris hyper- spectral image data set: June 12, 1992 indian pine test site 3. https://purr.purdue.edu/ publications/1947/1, Sep 2015. [Online; accessed January 20, 2022]. 8

    work page 1992

  16. [16]

    Graph-embedded subspace support vector data description

    Fahad Sohrab, Alexandros Iosifidis, Moncef Gabbouj, and Jenni Raitoharju. Graph-embedded subspace support vector data description. Pattern Recognition, 133:108999, 2023

    work page 2023

  17. [17]

    Nonlinear projection trick in kernel methods: An alternative to the kernel trick

    Nojun Kwak. Nonlinear projection trick in kernel methods: An alternative to the kernel trick. IEEE Transactions on Neural Networks and Learning Systems , 24(12):2113–2119, 2013

    work page 2013

  18. [18]

    Ellipsoidal sub- space support vector data description

    Fahad Sohrab, Jenni Raitoharju, Alexandros Iosifidis, and Moncef Gabbouj. Ellipsoidal sub- space support vector data description. IEEE Access, 8:122013–122025, 2020

    work page 2020