Citation

Lim, Jia Min and Lim, Kian Ming and Lee, Chin Poo and Lim, Jit Yan (2026) Few-Shot Photovoltaic Fault Classification via ProtoNet Angular Metric Learning and Automated Augmentation. In: 3rd IEEE International Conference on Energy and Electrical Engineering, EEE 2026, 26 June 2026 - 27 June 2026, Nanchang.

Text 5.pdf - Published Version Restricted to Repository staff only Download (550kB)

Abstract

Reliable fault diagnosis in photovoltaic (PV) systems is essential for maximizing energy yield, yet the scarcity of labeled data for rare defect types limits the effectiveness of traditional deep learning. While few-shot learning offers a viable path for data-scarce scenarios, existing frameworks often rely on manually tuned augmentation pipelines and distance metrics that are sensitive to global thermal intensity variations. In this paper, we propose AA-ShuffleNet framework optimized for fine-grained PV fault classification. We leverage TrivialAugment, an automated stochastic augmentation policy, to enhance feature robustness without human-tuned hyperparameters. For feature extraction, we employ a lightweight ShuffleNetV2 backbone, leveraging channel shuffling to capture subtle textural details in high-resolution UAV thermal imagery. Furthermore, we adopt a prototypical network framework that performs classification in an angular embedding space using cosine similarity, enabling magnitudeinvariant feature matching that is robust to varying thermal gradients. Experimental results on the PVF-10 dataset with 10 fault categories show that the proposed approach consistently outperforms existing methods under various few-shot learning settings, offering an efficient and scalable solution for autonomous photovoltaic system diagnostics.

Item Type:	Conference or Workshop Item (Paper)
Uncontrolled Keywords:	Angular Metric Learning, Few-Shot Learning, Photovoltaic Fault Classification
Subjects:	Q Science > QA Mathematics > QA71-90 Instruments and machines > QA75.5-76.95 Electronic computers. Computer science
Divisions:	Faculty of Information Science and Technology (FIST)
Depositing User:	Ms Rosnani Abd Wahab
Date Deposited:	28 Oct 2026 07:19
Last Modified:	28 Oct 2026 07:20
URII:	http://shdl.mmu.edu.my/id/eprint/16945

Downloads

Edit (login required)