CoChat

Background Reconstruction via 3D-Transformer Network for Hyperspectral Anomaly Detection

Open access
Wu, Ziyu, Wang, Bin · 2023 · Remote Sensing

The DOI matches its Crossref record; 49 of the 52 references it cites were found in a registry or an index; 2 of the 24 sources its claims cite were checked for support.

Abstract

Recently, autoencoder (AE)-based anomaly detection approaches for hyperspectral images (HSIs) have been extensively proposed; however, the reconstruction accuracy is susceptible to the anomalies and noises. ... To solve the above issues, this paper proposes a background reconstruction framework via a 3D-transformer (3DTR) network for anomaly detection in HSIs. The experimental results on both synthetic and real hyperspectral datasets demonstrate that the proposed 3DTR network is able to effectively detect most of the anomalies by comprehensively considering the spatial correlations among pixels and the spectral similarity among spectral bands of HSIs. ...

Overview

CoChat Summary

TL;DR

The paper proposes a 3D-transformer (3DTR) background-reconstruction framework for hyperspectral anomaly detection, combining spatial and spectral transformer modules in series with pre-detection and similarity-based patch generation. Experiments on two synthetic and six real hyperspectral datasets report that the proposed method generally outperforms traditional, model-based, and AE-based detectors, with fewer false alarms.

Key findings

  • On the first synthetic dataset with Gaussian noise, the proposed method achieved AUC(D,F) of 0.9987 ± 0.0005 at 30 dB, 0.9975 ± 0.0016 at 25 dB, and 0.9944 ± 0.0039 at 20 dB, higher than RX, CRD, GTVLRR, LSCTV, AE, RGAE, and GAED in Table 3.
  • On the second synthetic dataset with weak signals, the proposed method achieved AUC(D,F) of 0.9946 ± 0.0019 at 30 dB, 0.9939 ± 0.0038 at 25 dB, and 0.9912 ± 0.0066 at 20 dB, again outperforming the compared detectors in Table 5.
  • On six real datasets, the proposed method obtained the highest AUC(D,F) values in Table 7: Texas Coast-1 0.9946, Texas Coast-2 0.9991, Pavia 0.9972, HYDICE 0.9959, San Diego 0.9958, and Hyperion 0.9996.
  • Generalization evaluation reported validation-set AUC of 1.0000 and testing-set AUC of 0.9991, with the five validation anomalies detected faultlessly; ablation favored connecting the two TR modules in series (C4/M4) over single-module or parallel/merged alternatives.

Methods

  • Two synthetic datasets were generated from AVIRIS San Diego airport imagery: 224 initial bands, 186 retained after removing low-SNR and water-vapor bands; a 100 × 100 subimage was trimmed from a 400 × 400 image. Anomalies were implanted via a linear mixing model, with embedding proportions f = [0.2, 0.4, 0.6, 0.8] for the first synthetic dataset and f = [0.05, 0.1, 0.2, 0.4] for the second; anomaly blocks were 1 × 1, 2 × 1, 1 × 2, and 2 × 2.
  • The proposed method uses coarse pre-detection to remove potential anomalies (N_a = 300), a patch-generating method selecting N_p = 100 similar pixels around each center pixel, and a 3DTR network with spatial and spectral TR modules connected in series.
  • Evaluation used 3D ROC curves, 2D ROC curves, and AUC metrics including AUC(D,F), AUC(F,τ), AUC(D,τ), AUC_BS, AUC_TD, AUC_SNPR, AUC_TD-BS, and AUC_OD; noise experiments used 20 replicates per noise level. Real experiments used six datasets: Texas Coast-1, Texas Coast-2, Pavia, HYDICE, San Diego, and Hyperion.

Limitations

  • The paper states that the proposed method’s training time is relatively high, though test time is within an acceptable range; it notes training time can be significantly reduced by replacing CPU with GPU.
  • The paper states that some background pixels are inevitably removed by the loose pre-detection threshold, but argues this does not affect detection performance because the network can reconstruct removed background pixels from similar reserved pixels.
  • Evident from the text: the synthetic anomalies are generated by linear mixing and random placement, which may not fully represent real hyperspectral anomaly variability; the real-data evaluation relies on AUC and detection maps without reporting formal statistical significance tests beyond mean ± std for the noise experiments.

Model-written summary. Check it against the paper before relying on it.

Integrity signals

Registry

Crossref · 10.3390/rs15184592

Cites

49 of 52 references verified0 citation errors · 3 not checkable

Statistics

Nothing to recompute.

Verification

Claims the paper makes

1 Valid · 0 Not valid · 11 Not checked · 74 Own results and interpretations

42 more claims

Create a free account to see all 86 claims and to verify the 3 references still open.

See all claims — free account
  • It may state two separate claims as one. 34
  • Probably not a substantial claim. 7
  • The quote only partly states this claim. 2

Create a free account to see these claims.

Self-check

Percentages, sums and reported differences re-computed from the paper’s own numbers.

Each contradiction between two passages listed here was confirmed by an independent check.

1 of 1

Citations

Cites

Show all 52 references

Research works printed but not linked: 1

  1. Goodfellow, I., Pouget-Abadie, J., Mirza, M., Xu, B., Warde-Farley, D., Ozair, S., Courville, A., and Bengio, Y. (2014). Generative adversarial nets. Adv. Neural Inf. Process. Syst., 2672–2680.

Cited by

See all 13

Metadata

Authors
Wu, Z., & Wang, B.
Affiliations
Fudan University (CN)
Corresponding author
Wang, Bin
Venue
Remote Sensing
Publisher
MDPI AG
Published
2023-09-18
Volume
15
Issue
18
Pages
4592
ISSN
2072-4292 (ISSN-L)
OpenAlex
W4386849902
Cited by
13
References
52
Citations by year
2026: 3 · 2025: 7 · 2024: 3
FWCI
1.43
Type
journal article
Language
en
Open access
Locations
  • Remote Sensing (publishedVersion, cc-by)
  • DOAJ (DOAJ: Directory of Open Access Journals) (submittedVersion, cc-by-sa)
Funders
  • National Natural Science Foundation of China (61971141, 2022YFB3903404)
  • National Key Research and Development Program of China (61971141, 2022YFB3903404)
Topics
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Remote-Sensing Image Classification
  • Physical Sciences › Computer Science › Computer Vision and Pattern Recognition › Satellite Image Processing and Photogrammetry
  • Physical Sciences › Engineering › Media Technology › Advanced Image Fusion Techniques
Keywords
hyperspectral anomaly detection, background reconstruction, 3D transformer, hyperspectral imagery, autoencoder-based anomaly detection, spatial correlation, spectral similarity, patch generation, false alarm reduction

About these checks

Registry, retraction, Cites and statistics results are computed, not judged, with one exception: a printed reference that no index lists is read once by a model before it counts as a citation error. Claims are extracted by a model and kept only when their quote appears verbatim in the text; each cited source is then checked, with a model’s help, for whether it says what the claim relies on. Those verdicts can be wrong — read the claim’s quote and the cited source before relying on one.

The claim verdicts this report counts are experimental: their accuracy is not yet measured.

A language model drew the claims from the paper on 2026-10-10, under version 3 of the claim definition (model deepseek-v4.1-flash, prompt extract-v3).

A second model labelled each claim on 2026-10-10 (model jev-1.13-20260917, definition revision gold-guideline-v0.2).

Statistics and arithmetic are recomputed, not judged.

Confidence bands are shown only for fields calibrated on our annotated test set. None is calibrated yet, so no band is shown.

What counts as a claim ›How we measure confidence ›How we check this ›

Are you an author, or do you see a problem with this page? Request a review

How to cite

Cite CoChat for the checks on this page, and the paper for its own findings.

This page
CoChat. (2026). Citation check: Background Reconstruction via 3D-Transformer Network for Hyperspectral Anomaly Detection. https://app.cochat.ai/papers/10.3390/rs15184592
The paper
Wu, Z., & Wang, B. (2023). Background Reconstruction via 3D-Transformer Network for Hyperspectral Anomaly Detection. Remote Sensing. https://doi.org/10.3390/rs15184592

Data sources

Bibliographic data: Semantic Scholar, OpenAlex, Crossref.

Contains information from the Semantic Scholar Academic Graph Datasets, made available under the ODC Attribution License (ODC-BY 1.0).

Semantic Scholar’s data is described in: Kinney, R., et al. (2023). The Semantic Scholar Open Data Platform. arXiv:2301.10140.