What question did this study set out to answer?

Assess how CNNs recognize faults in photovoltaic systems and identify shortcut learning dependencies.

March 7, 2026AI0 citationsOpen Access

Diagnosing Shortcut Learning in CNN-Based Photovoltaic Fault Recognition from RGB Images: A Multi-Method Explainability Audit

BDBogdan DiaconuConstantin Brâncuși University of Targu Jiu

Puntos clave

Assess how CNNs recognize faults in photovoltaic systems and identify shortcut learning dependencies.
Conduct an explainability audit using five CNN architectures on the Kaggle PV Panel Defect Dataset.
Evaluate model performance and decision-making with LIME, occlusion sensitivity, and Integrated Gradients.
Analyze faithfulness of model attributions and how they relate to background context.
ResNet50 shows the highest predictive performance among the architectures tested.
EfficientNetB0 exhibits the most consistent evidence for faithfulness in model explanations.
Class vulnerability is noted, particularly for Clean and damaged classes, highlighting reliance on contextual cues.

Resumen

Convolutional neural networks (CNNs) can achieve high accuracy in photovoltaic (PV) fault recognition from RGB imagery, yet their decisions may rely on shortcut cues induced by heterogeneous backgrounds, viewpoints, and class imbalance. This work presents a multi-method explainability audit on the Kaggle PV Panel Defect Dataset (six classes), comparing five architectures (Baseline CNN, VGG16, ResNet50, InceptionV3, EfficientNetB0). Explanations are obtained with LIME superpixel surrogates (reported together with kernel-weighted surrogate fidelity), occlusion sensitivity (quantified via IoU@Top10% against consistent proxy masks, Shannon entropy, and Hoyer sparsity), and Integrated Gradients evaluated by deletion–insertion faithfulness and a Faithfulness Gap. While ResNet50 yields the best predictive performance, EfficientNetB0 shows the most consistent faithfulness evidence and stable panel-centered attributions. The analysis highlights class-dependent vulnerability to context cues, especially for the Clean and damaged classes, and supports using quantitative explainability diagnostics during model selection and dataset curation to mitigate shortcuts in vision-based PV monitoring.

Preguntar a la IA

Me gusta

Guardar

Ver artículo completo