Fig 1.
Classification of common faults in PV cells.
Table 1.
Comparative analysis of DL algorithms for image processing.
Fig 2.
EfficientNetV2 Efficiency Comparison:
(a) Training Time, (b) Step Time, (c) Parameters, and (d) FLOPs [22].
Fig 3.
Analysis of PV Cell Conditions:
(a) Proportion of Defective vs. Non-defective Cells, (b) Distribution of Cell Types (Monocrystalline vs. Polycrystalline).
Table 2.
Comparative Analysis of Defectivity in Monocrystalline and Polycrystalline PV Cells.
Table 3.
Distribution of PV cell image data for training, validation and test sets.
Fig 4.
Comparative Display of Defective and Non-Defective Monocrystalline PV Cells.
Fig 5.
Comparative Display of Defective and Non-Defective Polycrystalline PV Cells.
Fig 6.
Summary of Data Preparation and Model Configuration Parameters.
Fig 7.
Visualization of Data Augmentation Techniques on PV Cells(a) Shearing, (b) Zooming, (c) Rotation, (d) Flipping.
Fig 8.
Comparison of Original and GradCAM-Enhanced PV Cell Images(a) Original Image, (b) GradCAM Enhancement.
Fig 9.
Grad-CAM Process Flowchart: From Input to Heatmap Visualization.
Table 4.
Key Configuration Parameters and Settings for PV Cell Defect Detection Model.
Table 5.
Comparison of Model Size, Memory Requirements, and Computational Complexity Across EfficientNetV2 Variants.
Table 6.
Comparative Performance Metrics of EfficientNetV2 Models for Photovoltaic Cell Fault Detection.
Table 7.
Comparative Training and Validation Performance of EfficientNetV2 Models During Feature Extraction Stage.
Table 8.
First stage – (Feature Extraction) Training and Validation Metrics at Best Epoch Based on Minimum Validation Loss for EfficientNetV2 Models.
Table 9.
Second Stage – (Fine Tuning) Training and Validation Metrics at Best Epoch Based on Minimum Validation Loss for EfficientNetV2 Models.
Fig 10.
Training and Validation Accuracy Curves During the First Stage (Feature Extraction) for (a) EfficientNetV2B0, (b) EfficientNetV2B2, and (c) EfficientNetV2M Models.
Fig 11.
Training and Validation Loss Curves During the First Stage (Feature Extraction) for (a) EfficientNetV2B0, (b) EfficientNetV2B2, and (c) EfficientNetV2M Models.
Fig 12.
Training and Validation Accuracy Curves During the Second Stage (Fine-Tuning) for (a) EfficientNetV2B0, (b) EfficientNetV2B2, and (c) EfficientNetV2M Models.
Fig 13.
Training and Validation Loss Curves During the Second Stage (Fine-Tuning) for (a) EfficientNetV2B0, (b) EfficientNetV2B2, and (c) EfficientNetV2M Models.
Fig 14.
Confusion Matrices for Binary Classification Performance of (a) EfficientNetV2B0, (b) EfficientNetV2B2, and (c) EfficientNetV2M Models on Photovoltaic Cell Fault Detection.