Pith. sign in

REVIEW 3 major objections 6 minor 34 references

Fusing JEPA semantic embeddings with EfficientNet features classifies thermal PV faults more accurately with half the parameters of the prior best model.

Reviewed by Pith at T0; open to challenge. T0 means a machine referee read the full paper against a public rubric. the ladder, T0–T4 →

Challenge this review Re-run · record.json Download PDF Read on arXiv ↗

T0 review · grok-4.5

2026-07-13 04:42 UTC pith:P3OVZGRS

load-bearing objection Solid applied fusion paper: JEPA + EfficientNetV2-S beats GEPFNet on two public IR PV sets with ~47% fewer parameters; ImageNet pretraining is load-bearing but measured. the 3 major comments →

arxiv 2607.09205 v1 pith:P3OVZGRS submitted 2026-07-10 eess.IV cs.CV

Joint-Embedding Predictive Architecture for Solar PV Panel Fault Classification

Seyyedhamid Azimidokht , Mehdi Monemi , Abdelhak Kharbouch , Farid Hamzehaghdam , Mehdi Rasti , Jamshid Aghaei , Emil Kurvinen This is my paper
classification eess.IV cs.CV
keywords Joint-Embedding Predictive ArchitectureJEPAphotovoltaic fault classificationinfrared thermographyEfficientNetV2self-supervised learningmultibranch fusionsolar PV inspection
verification ladder T0 review T1 audit T2 compute T3 formal T4 reserved

The pith

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

Thermal infrared images of solar panels show faults mainly as subtle temperature patterns with little texture, so fully supervised models can miss them and class imbalance makes the problem harder. This paper shows that a Joint-Embedding Predictive Architecture (JEPA) pretrained on a large natural-image set can supply transferable semantic features for those images, and that pairing those features with a standard convolutional backbone yields better classification than either alone. The resulting multibranch network, JEFFNet, is trained with a staged freeze-then-unfreeze schedule and tested on two public IR datasets for both fine-grained multiclass labels and a practical healthy-versus-faulty decision. It matches or exceeds the previous state-of-the-art while using only about half as many parameters. The result matters because large solar farms need automated, non-contact inspection that is both accurate and light enough to deploy at scale.

Core claim

Combining self-supervised JEPA semantic representations (from a Vision Transformer context encoder pretrained on ImageNet-1k via StoP-JEPA) with supervised EfficientNetV2-S convolutional features, fused after projection and trained with a staged fine-tuning schedule, produces a more accurate and parameter-efficient classifier for thermal IR PV faults than prior single-branch or equivariant fusion networks on the PVF-10 and ISM benchmarks.

What carries the argument

JEFFNet: a multibranch network that runs the same IR image through a frozen-then-unfrozen JEPA context encoder (global-average-pooled patch embeddings) and an EfficientNetV2-S backbone in parallel, projects and transforms both streams, concatenates them, and classifies with weighted cross-entropy.

Load-bearing premise

That ImageNet-1k JEPA pretraining already supplies useful semantic features for low-texture thermal PV images, so that a simple freeze-then-unfreeze schedule and concatenation with a convolutional branch are enough to transfer them.

What would settle it

Retrain JEFFNet from scratch with a JEPA encoder that never saw ImageNet (or any large natural-image set) and only the solar IR data; if the multiclass F1 on PVF-10 and ISM then falls back to the ablation levels near 65–70 percent, the claimed transfer benefit disappears.

Watch this falsifier — get emailed when new claim-graph text bears on it.

Editorial analysis

A structured set of objections, weighed in public.

Desk editor's note, referee report, simulated authors' rebuttal, and a circularity audit.

Referee Report

3 major / 6 minor

Summary. The paper proposes JEFFNet, a multibranch model that fuses a StoP-JEPA-pretrained ViT-B/16 context encoder (ImageNet-1k) with an EfficientNetV2-S branch for thermal IR photovoltaic fault classification. Representations are projected, refined, concatenated, and classified under weighted cross-entropy with a staged freeze-then-unfreeze schedule. Evaluation on PVF-10 (10-class and derived 2-class) and InfraredSolarModules (12-class and derived 2-class) reports macro F1 of 93.21 / 97.53 (PVF-10) and 72.60 / 94.69 (ISM), with 108.6M parameters versus 205.91M for GEPFNet. Ablations (Table I) isolate pretraining domain, fine-tuning schedule, and multibranch fusion; multi-run means and standard deviations are given in Tables II–V; code and public datasets are cited.

Significance. If the reported gains hold under fair comparison, the work is a useful systems contribution for IR PV inspection: it shows that latent-space JEPA features can be productively combined with a compact CNN under staged adaptation, yields competitive or better macro metrics on two public benchmarks, and roughly halves parameters relative to the recent GEPFNet baseline. Strengths that support credibility include multi-run reporting, class-imbalance-aware loss and macro metrics, public code, and an ablation that explicitly measures the load-bearing role of ImageNet StoP-JEPA pretraining versus in-domain JEPA training. The novelty is primarily architectural and empirical rather than theoretical; the practical value is parameter-efficient transfer for low-texture thermal imagery.

major comments (3)
  1. Tables II and IV: the headline multiclass F1 gains over GEPFNet (PVF-10: 93.21±1.24 vs 92.47±0.87; ISM: 72.60±1.96 vs 71.32±0.72) have substantially overlapping standard deviations. The abstract and §I-B state that JEFFNet “improves over GEPFNet” by 0.74% and 1.28% F1 without a paired significance test or confidence interval on the difference. Please either (i) report a paired test over the same 10 seeds/splits, or (ii) temper the multiclass superiority language to “competitive with a large parameter reduction,” while retaining stronger claims only where separation is clear (e.g., PVF-10 2-class F1 97.53±0.35 vs 95.01±0.58).
  2. Table I is a single representative run under one fixed split, while Tables II–V report mean±std over 10 runs. The staged schedule and multibranch fusion are load-bearing for the design story (rows 2–5: Frozen 85.02 F1 82.70 → Staged 93.86/93.69 → JEFFNet 94.77/93.84). Please re-run the ablation under the same multi-seed protocol as the main tables, or clearly mark Table I as exploratory and avoid using single-run deltas as primary evidence for staged training and fusion.
  3. §III-D: binary healthy/faulty metrics are obtained by collapsing multiclass logits rather than training a dedicated binary model. This is disclosed, but the contribution text and abstract present 2-class F1/accuracy as primary results. Either add a short experiment with a binary-trained head (same backbone) to validate the proxy, or reframe the 2-class numbers strictly as a post-hoc screening evaluation derived from the multiclass model.
minor comments (6)
  1. §II opening: typo “prelimiaries”; §I-B: “thestagedfine-tuning” and “non-stagedtraining” need spacing.
  2. Fig. 4 caption is only “The training and validation loss curves for proposed model.” Specify dataset, task (e.g., PVF-10 10-class), and which stage of staged training is shown.
  3. Tables IV–V: Coat Lite Small is listed with all “–”; either drop the row or note why it was not re-evaluated on ISM.
  4. Eqs. (6)–(14): projection/scaler dimensions and dropout rates are not stated in the text (only in free-parameter spirit). A short table or appendix listing hidden sizes would aid reproducibility beyond the GitHub link.
  5. §III-A: ISM images are 24×40 and resized to 224×224; a brief note on interpolation and whether upsampling artifacts were checked would help readers of low-resolution IR work.
  6. Related work cites GEPFNet as Information Fusion 2026; ensure the bibliographic entry matches the version used for the reported numbers so the comparison remains auditable.

Circularity Check

0 steps flagged

No significant circularity: purely empirical multibranch fusion evaluated on external public IR datasets against independent baselines.

full rationale

The paper's central claim is an engineering result (staged fusion of ImageNet-pretrained StoP-JEPA ViT-B context-encoder embeddings with EfficientNetV2-S features yields higher macro-F1 than GEPFNet on PVF-10 and ISM while using fewer parameters). All reported numbers are obtained by supervised training/evaluation on the two public thermal-IR datasets under fixed or stratified splits, with multi-run means and standard deviations; the 2-class scores are simply post-hoc merges of the multiclass outputs. ImageNet-1k StoP-JEPA pre-training is an external, publicly released checkpoint; the ablation (Table I) explicitly measures its contribution rather than hiding it. There are no fitted constants re-labeled as predictions, no uniqueness theorems, no self-citation chains that force the architecture, and no renaming of known results. Self-citations (I-JEPA, StoP-JEPA, a JEPA tutorial by overlapping authors) supply only background and are not load-bearing for the performance claims. The derivation chain is therefore self-contained against external benchmarks.

Axiom & Free-Parameter Ledger

4 free parameters · 4 axioms · 1 invented entities

Empirical deep-learning paper. Load-bearing choices are standard ML hyperparameters, the assumption that ImageNet-pretrained StoP-JEPA embeddings remain useful for thermal IR, and the architectural decision to fuse by projection+concatenation. No new physical entities or free constants fitted to the scientific target; the free parameters are ordinary training knobs.

free parameters (4)
  • staged learning rates (new layers / JEPA encoder) = 3e-4 / 1e-5
    Set by hand to 3e-4 and 1e-5; performance depends on this schedule.
  • freeze duration = 10 epochs
    First 10 of 100 epochs keep JEPA frozen; chosen without exhaustive search reported.
  • projection / scaler / classifier hidden dimensions and dropout
    Architectural widths and dropout rates of the fusion modules are design choices that affect capacity and regularization.
  • class weights for weighted cross-entropy = inverse frequency
    Inverse-frequency weights computed on the training split; standard but still data-dependent.
axioms (4)
  • domain assumption StoP-JEPA context encoder pretrained on ImageNet-1k yields transferable semantic patch embeddings for thermal IR PV images.
    Central to the ablation (Table I rows 1 vs 2) and to all subsequent claims; not proved, only empirically supported on the two datasets.
  • ad hoc to paper Global-average-pooled ViT patch tokens and EfficientNetV2-S GAP features are complementary and can be usefully fused by linear projection + concatenation.
    Architectural premise of JEFFNet (Eqs. 7–14); alternatives (attention fusion, gating) are left to future work.
  • domain assumption Macro-averaged F1 / recall / MCC on the chosen stratified splits are the appropriate figures of merit under class imbalance.
    Standard for imbalanced multi-class inspection; used throughout Tables II–V.
  • ad hoc to paper Binary healthy/faulty metrics obtained by collapsing multiclass logits are a valid proxy for a dedicated binary detector.
    Stated in §III-D; no separate binary models are trained.
invented entities (1)
  • JEFFNet (JEPA-EFFicientNet multibranch architecture) independent evidence
    purpose: Fuse StoP-JEPA semantic embeddings with EfficientNetV2-S convolutional features for IR PV fault classification.
    Named architecture introduced in the paper; independent evidence is the reported public-code experiments on two datasets.

pith-pipeline@v1.1.0-grok45 · 23309 in / 3041 out tokens · 43267 ms · 2026-07-13T04:42:11.607800+00:00 · methodology

0 comments
read the original abstract

The rapid expansion of solar photovoltaic (PV) systems has increased the need for reliable and scalable fault classification, as manual inspection is impractical at scale. Thermal infrared (IR) imaging provides a non-contact solution for identifying PV faults; however, accurate classification remains challenging due to class imbalance, limited texture information, and subtle thermal differences. In this work, we investigate the applicability of Joint-Embedding Predictive Architecture (JEPA) for thermal IR PV fault classification across various scenarios and propose JEFFNet (JEPA-EFFicientNet), a multibranch architecture that combines JEPA-based self-supervised representation learning with EfficientNetV2-S-based supervised convolutional feature extraction. JEFFNet fuses semantic representations from a JEPA-pretrained Vision Transformer with convolutional features from EfficientNetV2-S, enabling complementary feature learning. JEFFNet is evaluated on two public thermal IR datasets, PVF-10 and InfraredSolarModules (ISM), for both multiclass and derived binary (healthy/faulty) classification. On PVF-10, JEFFNet achieves an F1-score of $93.21$ and an accuracy of $94.33$ in the 10-class task, and an F1-score of $97.53$ and an accuracy of $96.41$ in the derived 2-class task. On ISM, JEFFNet achieves an F1-score of $72.60$ and an accuracy of $83.88$ in the 12-class task, and an F1-score of $94.69$ and an accuracy of $94.78$ in the derived 2-class task. JEFFNet also uses only 108.6M parameters versus 205.91M for GEPFNet, a 47.2\% reduction. These results demonstrate that combining self-supervised semantic and supervised convolutional features provides an effective, parameter-efficient solution for thermal IR PV fault classification. The source code is publicly available at https://github.com/Azimi2kht/JEFFNet

Figures

Figures reproduced from arXiv: 2607.09205 by Abdelhak Kharbouch, Emil Kurvinen, Farid Hamzehaghdam, Jamshid Aghaei, Mehdi Monemi, Mehdi Rasti, Seyyedhamid Azimidokht.

Figure 1
Figure 1. Figure 1: Architecture of the proposed JEFFNet fusion model. The input thermal IR image is processed by two parallel branches: a JEPA-pretrained ViT encoder, which produces patch-level semantic feature vectors that are aggregated by global average pooling, and an EfficientNetV2-S backbone, which extracts spatial convolutional feature maps followed by global average pooling. The resulting branch representations are t… view at source ↗
Figure 2
Figure 2. Figure 2: Detailed structure of the projection, scaling, and classifier modules used in JEFFNet. The projector maps each branch representation into a shared feature space using a fully connected layer followed by layer normalization and GELU activation. The scaler further refines the projected representation using two fully connected layers with intermediate normalization and activation. The classifier head maps the… view at source ↗
Figure 3
Figure 3. Figure 3: Class distributions of the datasets. Subfigure (a) shows the PVF-10 dataset, where the class symbols are: A: string short circuit, B: break, C: shadow, D: bottom dirt, E: debris cover, F: short circuit panel, G: substring open circuit, H: junction box heat, I: hot cell, and J: healthy panel. Subfigure (b) shows the ISM dataset, where the class symbols are: A: Diode-Multi, B: Soiling, C: Hot-Spot-Multi, D: … view at source ↗
Figure 4
Figure 4. Figure 4: The training and validation loss curves for proposed model. patches of size 16 × 16. The JEPA encoder was initialized from the ImageNet-1k pretrained StoP-JEPA ViT-B checkpoint released by the authors of StoP-JEPA [23]. For downstream PV fault classification, we adopted a staged fine-tuning strategy. During the training, data augmentation was applied to improve robustness and reduce overfitting. The augmen… view at source ↗
Figure 5
Figure 5. Figure 5: Confusion matrix of the best-performing JEFFNet run on the PVF-10 test set for the 10-class classification task. Each cell shows the sample count and corresponding row-wise percentage, with color intensity indicating the class-wise prediction rate. semantically rich datasets which is not fully exploited using existing solar datasets such as PVF-10 and ISM. To evaluate this, the same StoP-JEPA configuration… view at source ↗

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

34 extracted references · 4 linked inside Pith

  1. [1]

    Renewables 2025: Analysis and forecasts to 2030,

    International Energy Agency, “Renewables 2025: Analysis and forecasts to 2030,” International Energy Agency, Tech. Rep., October 2025, available at: https://iea.blob.core.windows.net/assets/76ad6eac- 2aa6-4c55-9a55-b8dc0dba9f9e/Renewables2025.pdf. [Online]. Available: https://www.iea.org/reports/renewables-2025

    2025

  2. [2]

    Energy and AI,

    ——, “Energy and AI,” International Energy Agency, Tech. Rep., April 2025, available at: https://iea.blob.core.windows.net/assets/ de9dea13-b07d-42c5-a398-d1b3ae17d866/EnergyandAI.pdf. [Online]. Available: https://www.iea.org/reports/energy-and-ai

    2025

  3. [3]

    Trends in Photovoltaic Applications 2025,

    IEA PVPS Task 1, “Trends in Photovoltaic Applications 2025,” IEA Photovoltaic Power Systems Programme (PVPS), Tech. Rep., October 2025, available at: https://iea-pvps.org/wp-content/uploads/ 2025/10/IEA-PVPS Trends 2025-.pdf. [Online]. Available: https: //iea-pvps.org/trends reports/trends-2025

    2025

  4. [4]

    Photovoltaic (pv) systems – requirements for testing, documentation and maintenance – part 3: Photovoltaic modules and plants – outdoor infrared thermography,

    “Photovoltaic (pv) systems – requirements for testing, documentation and maintenance – part 3: Photovoltaic modules and plants – outdoor infrared thermography,” International Electrotechnical Commission (IEC), Geneva, Switzerland, Tech. Rep. IEC TS 62446-3:2017, 2017

    2017

  5. [5]

    Standards for Photovoltaic Energy Systems,

    Sample, T., et al., “Standards for Photovoltaic Energy Systems,” European Commission, Joint Research Centre (JRC), Tech. Rep. JRC127142, 2022. [Online]. Available: https://publications.jrc.ec.europa. eu/repository/handle/JRC127142

    2022

  6. [6]

    Failures of Photovoltaic modules and their Detection: A Review,

    M. Waqar Akram, G. Li, Y. Jin, and X. Chen, “Failures of Photovoltaic modules and their Detection: A Review,”Applied Energy, vol. 313, p. 118822, 2022. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/S0306261922002677

    2022

  7. [7]

    Assessment of photovoltaic module failures in the field,

    International Energy Agency Photovoltaic Power Systems Programme (IEA PVPS), Task 13, “Assessment of photovoltaic module failures in the field,” IEA PVPS, Technical Report, 2017, available from IEA PVPS website. [Online]. Available: https://iea-pvps.org/key-topics/ report-assessment-of-photovoltaic-module-failures-in-the-field-2017/

    2017

  8. [8]

    Self-supervised learning from images with a joint-embedding predictive architecture,

    M. Assran, Q. Duval, I. Misra, P. Bojanowski, P. Vincent, M. Rabbat, Y. LeCun, and N. Ballas, “Self-supervised learning from images with a joint-embedding predictive architecture,” inProceedings of the IEEE/CVF conference on computer vision and pattern recognition, 2023, pp. 15 619– 15 629

    2023

  9. [9]

    A Novel Defect Detection Method for Photovoltaic Panels: Enhancing Visible Light Images Recognition Efficiency Through Extended Deep Learning Models,

    L. Zhu, J. Cheng, and G. Liu, “A Novel Defect Detection Method for Photovoltaic Panels: Enhancing Visible Light Images Recognition Efficiency Through Extended Deep Learning Models,”Available at SSRN 5022044

  10. [10]

    Deep learning for solar pv fault classification using RGB imaging and comparison of preprocessing techniques,

    M. E. Ramachandran, G. B. Balachandran, P. Velladurai, and A. Rajaku- mar, “Deep learning for solar pv fault classification using RGB imaging and comparison of preprocessing techniques,”Solar Energy, vol. 301, p. 113959, 2025

    2025

  11. [11]

    Automatic classification of defective photovoltaic module cells in electroluminescence images,

    S. Deitsch, V. Christlein, S. Berger, C. Buerhop-Lutz, A. Maier, F. Gall- witz, and C. Riess, “Automatic classification of defective photovoltaic module cells in electroluminescence images,”Solar Energy, vol. 185, pp. 455–468, 2019

    2019

  12. [12]

    Automated defect detection and localization in photovoltaic cells using semantic segmentation of electroluminescence images,

    J. Fioresi, D. J. Colvin, R. Frota, R. Gupta, M. Li, H. P. Seigneur, S. Vyas, S. Oliveira, M. Shah, and K. O. Davis, “Automated defect detection and localization in photovoltaic cells using semantic segmentation of electroluminescence images,”IEEE Journal of Photovoltaics, vol. 12, no. 1, pp. 53–61, 2021

    2021

  13. [13]

    Automatic detection and analysis of photovoltaic modules in aerial infrared imagery,

    S. Dotenco, M. Dalsass, L. Winkler, T. W¨ urzner, C. Brabec, A. Maier, and F. Gallwitz, “Automatic detection and analysis of photovoltaic modules in aerial infrared imagery,” in2016 IEEE Winter Conference on Applications of Computer Vision (WACV). IEEE, 2016, pp. 1–9

    2016

  14. [14]

    A machine learning framework to identify the hotspot in photovoltaic module using infrared thermography,

    M. U. Ali, H. F. Khan, M. Masud, K. D. Kallu, and A. Zafar, “A machine learning framework to identify the hotspot in photovoltaic module using infrared thermography,”Solar Energy, vol. 208, pp. 643–651, 2020

    2020

  15. [15]

    Photovoltaic defect classification through thermal infrared imaging using a machine learning approach,

    C. Dunderdale, W. Brettenny, C. Clohessy, and E. E. Van Dyk, “Photovoltaic defect classification through thermal infrared imaging using a machine learning approach,”Progress in Photovoltaics: Research and Applications, vol. 28, no. 3, pp. 177–188, 2020

    2020

  16. [16]

    Thermal inspection of photovoltaic modules with deep convolutional neural networks on edge devices in auv,

    M. Le, D. Le, and H. H. T. Vu, “Thermal inspection of photovoltaic modules with deep convolutional neural networks on edge devices in auv,”Measurement, vol. 218, p. 113135, 2023. 12

    2023

  17. [17]

    Remote anomaly detection and classification of solar photovoltaic modules based on deep neural network,

    M. Le, D. K. Nguyen, V.-D. Dao, N. H. Vu, H. H. T. Vuet al., “Remote anomaly detection and classification of solar photovoltaic modules based on deep neural network,”Sustainable Energy Technologies and Assessments, vol. 48, p. 101545, 2021

    2021

  18. [18]

    Fault diagnosis of photovoltaic modules using deep neural networks and infrared images under algerian climatic conditions,

    N. Kellil, A. Aissat, and A. Mellit, “Fault diagnosis of photovoltaic modules using deep neural networks and infrared images under algerian climatic conditions,”Energy, vol. 263, p. 125902, 2023

    2023

  19. [19]

    An image is worth 16x16 words: Transformers for image recognition at scale,

    A. Dosovitskiy, L. Beyer, A. Kolesnikov, D. Weissenborn, X. Zhai, T. Unterthiner, M. Dehghani, M. Minderer, G. Heigold, S. Gellyet al., “An image is worth 16x16 words: Transformers for image recognition at scale,”arXiv preprint arXiv:2010.11929, 2020

    Pith/arXiv arXiv 2010

  20. [20]

    Benchmarking Vision Transformers and CNNs for Thermal Photovoltaic Fault Detection with Explainable AI Validation,

    S. Aksoy, “Benchmarking Vision Transformers and CNNs for Thermal Photovoltaic Fault Detection with Explainable AI Validation,”arXiv preprint arXiv:2509.07039, 2025

    Pith/arXiv arXiv 2025

  21. [21]

    GEPFNet: A group equivariant feature extraction with parallel fusion neural network for solar photovoltaic fault classification,

    J. Guo, B. K. Ng, C.-T. Lam, and P. H. Abreu, “GEPFNet: A group equivariant feature extraction with parallel fusion neural network for solar photovoltaic fault classification,”Information Fusion, vol. 129, p. 104014, 2026. [Online]. Available: https: //www.sciencedirect.com/science/article/pii/S1566253525010760

    2026

  22. [22]

    Tutorial on Joint Embedding Predictive Architectures (JEPA): Foundations, Applications, and Future Directions,

    M. Monemi, M. Chinipardaz, M. Rasti, M. Bennis, and M. Latva- Aho, “Tutorial on Joint Embedding Predictive Architectures (JEPA): Foundations, Applications, and Future Directions,”Authorea Preprints, 2025

    2025

  23. [23]

    Stochastic positional embeddings improve masked image modeling,

    A. Bar, F. Bordes, A. Shocher, M. Assran, P. Vincent, N. Ballas, T. Darrell, A. Globerson, and Y. LeCun, “Stochastic positional embeddings improve masked image modeling,”arXiv preprint arXiv:2308.00566, 2023

    Pith/arXiv arXiv 2023

  24. [24]

    A-jepa: Joint-embedding predictive architecture can listen,

    Z. Fei, M. Fan, and J. Huang, “A-jepa: Joint-embedding predictive architecture can listen,”arXiv preprint arXiv:2311.15830, 2023

    Pith/arXiv arXiv 2023

  25. [25]

    Self-supervised pre-training with joint- embedding predictive architecture boosts ECG classification performance,

    K. Weimann and T. O. Conrad, “Self-supervised pre-training with joint- embedding predictive architecture boosts ECG classification performance,” Computers in Biology and Medicine, vol. 196, p. 110809, 2025

    2025

  26. [26]

    Imagenet: A large-scale hierarchical image database,

    J. Deng, W. Dong, R. Socher, L.-J. Li, K. Li, and L. Fei-Fei, “Imagenet: A large-scale hierarchical image database,” in2009 IEEE Conference on Computer Vision and Pattern Recognition, 2009, pp. 248–255

    2009

  27. [27]

    Efficientnetv2: Smaller models and faster training,

    M. Tan and Q. Le, “Efficientnetv2: Smaller models and faster training,” inInternational conference on machine learning. PMLR, 2021, pp. 10 096–10 106

    2021

  28. [28]

    Efficientnet: Rethinking model scaling for convolutional neural networks,

    ——, “Efficientnet: Rethinking model scaling for convolutional neural networks,” inInternational conference on machine learning. PMLR, 2019, pp. 6105–6114

    2019

  29. [29]

    Squeeze-and-excitation networks,

    J. Hu, L. Shen, and G. Sun, “Squeeze-and-excitation networks,” in Proceedings of the IEEE conference on computer vision and pattern recognition, 2018, pp. 7132–7141

    2018

  30. [30]

    PVF-10: A high-resolution unmanned aerial vehicle thermal infrared image dataset for fine-grained photovoltaic fault classification,

    B. Wang, Q. Chen, M. Wang, Y. Chen, Z. Zhang, X. Liu, W. Gao, Y. Zhang, and H. Zhang, “PVF-10: A high-resolution unmanned aerial vehicle thermal infrared image dataset for fine-grained photovoltaic fault classification,”Applied Energy, vol. 376, p. 124187, 2024

    2024

  31. [31]

    Infrared solar module dataset for anomaly detection,

    M. Millendorf, E. Obropta, and N. Vadhavkar, “Infrared solar module dataset for anomaly detection,” inProc. Int. Conf. Learn. Represent, vol. 3, 2020

    2020

  32. [32]

    Comparing two k-category assignments by a k-category correlation coefficient,

    J. Gorodkin, “Comparing two k-category assignments by a k-category correlation coefficient,”Computational biology and chemistry, vol. 28, no. 5-6, pp. 367–374, 2004

    2004

  33. [33]

    Anomaly detection for PV modules using Multi-modal data fusion in aerial inspections,

    L. S. Pinho, T. D. Sousa, C. D. Pereira, and A. M. Pinto, “Anomaly detection for PV modules using Multi-modal data fusion in aerial inspections,”IEEE Access, 2025

    2025

  34. [34]

    DSeq-JEPA: Discriminative Sequential Joint-Embedding Predictive Architecture,

    X. He, S. Sakai, S. Chandhok, S. Beery, K. Yuan, N. Padoy, T. Hasegawa, and L. Sigal, “DSeq-JEPA: Discriminative Sequential Joint-Embedding Predictive Architecture,”arXiv preprint arXiv:2511.17354, 2025

    arXiv 2025