Logo
ScienceStack
Table of Contents
Fetching ...
Sign In

HybridSolarNet: A Lightweight and Explainable EfficientNet-CBAM Architecture for Real-Time Solar Panel Fault Detection

Md. Asif Hossain, G M Mota-Tahrin Tayef, Nabil Subhan

TL;DR

This work tackles the need for accurate, real-time solar panel fault detection on edge devices by proposing HybridSolarNet, a compact EfficientNet-B0 backbone augmented with CBAM. Trained under a strict split-before-augmentation protocol and optimized with focal loss and cosine annealing, the model achieves 92.37% accuracy and a 0.9226 F1 score on the Kaggle Solar Panel Images dataset while occupying only 16.3 MB and delivering 54.9 FPS on GPU. Across 5-fold cross-validation, performance remains stable (±0.41 in accuracy and ±0.39 in F1), and Grad-CAM analyses confirm that the model attends to real defect regions rather than spurious cues. The approach offers a practical, interpretable, and deployment-friendly solution for real-time UAV inspections, with future work targeting embedded platforms and cross-domain robustness.

Abstract

Manual inspections for solar panel systems are a tedious, costly, and error-prone task, making it desirable for Unmanned Aerial Vehicle (UAV) based monitoring. Though deep learning models have excellent fault detection capabilities, almost all methods either are too large and heavy for edge computing devices or involve biased estimation of accuracy due to ineffective learning techniques. We propose a new solar panel fault detection model called HybridSolarNet. It integrates EfficientNet-B0 with Convolutional Block Attention Module (CBAM). We implemented it on the Kaggle Solar Panel Images competition dataset with a tight split-before-augmentation protocol. It avoids leakage in accuracy estimation. We introduced focal loss and cosine annealing. Ablation analysis validates that accuracy boosts due to added benefits from CBAM (+1.53%) and that there are benefits from recognition of classes with imbalanced samples via focal loss. Overall average accuracy on 5-fold stratified cross-validation experiments on the given competition dataset topped 92.37% +/- 0.41 and an F1-score of 0.9226 +/- 0.39 compared to baselines like VGG19, requiring merely 16.3 MB storage, i.e., 32 times less. Its inference speed measured at 54.9 FPS with GPU support makes it a successful candidate for real-time UAV implementation. Moreover, visualization obtained from Grad-CAM illustrates that HybridSolarNet focuses on actual locations instead of irrelevant ones.

HybridSolarNet: A Lightweight and Explainable EfficientNet-CBAM Architecture for Real-Time Solar Panel Fault Detection

TL;DR

This work tackles the need for accurate, real-time solar panel fault detection on edge devices by proposing HybridSolarNet, a compact EfficientNet-B0 backbone augmented with CBAM. Trained under a strict split-before-augmentation protocol and optimized with focal loss and cosine annealing, the model achieves 92.37% accuracy and a 0.9226 F1 score on the Kaggle Solar Panel Images dataset while occupying only 16.3 MB and delivering 54.9 FPS on GPU. Across 5-fold cross-validation, performance remains stable (±0.41 in accuracy and ±0.39 in F1), and Grad-CAM analyses confirm that the model attends to real defect regions rather than spurious cues. The approach offers a practical, interpretable, and deployment-friendly solution for real-time UAV inspections, with future work targeting embedded platforms and cross-domain robustness.

Abstract

Manual inspections for solar panel systems are a tedious, costly, and error-prone task, making it desirable for Unmanned Aerial Vehicle (UAV) based monitoring. Though deep learning models have excellent fault detection capabilities, almost all methods either are too large and heavy for edge computing devices or involve biased estimation of accuracy due to ineffective learning techniques. We propose a new solar panel fault detection model called HybridSolarNet. It integrates EfficientNet-B0 with Convolutional Block Attention Module (CBAM). We implemented it on the Kaggle Solar Panel Images competition dataset with a tight split-before-augmentation protocol. It avoids leakage in accuracy estimation. We introduced focal loss and cosine annealing. Ablation analysis validates that accuracy boosts due to added benefits from CBAM (+1.53%) and that there are benefits from recognition of classes with imbalanced samples via focal loss. Overall average accuracy on 5-fold stratified cross-validation experiments on the given competition dataset topped 92.37% +/- 0.41 and an F1-score of 0.9226 +/- 0.39 compared to baselines like VGG19, requiring merely 16.3 MB storage, i.e., 32 times less. Its inference speed measured at 54.9 FPS with GPU support makes it a successful candidate for real-time UAV implementation. Moreover, visualization obtained from Grad-CAM illustrates that HybridSolarNet focuses on actual locations instead of irrelevant ones.
Paper Structure (22 sections, 6 figures, 5 tables)

This paper contains 22 sections, 6 figures, 5 tables.

Table of Contents

  1. Introduction
  2. Literature Review
  3. Summary of Research Gaps
  4. Methodology
  5. Dataset Preparation and Preprocessing
  6. Proposed Architecture: HybridSolarNet
  7. Baseline Models
  8. Training Configuration
  9. Experimental Results
  10. Test Set Performance
  11. K-Fold Cross-Validation
  12. Error Analysis
  13. ROC and PR Analysis
  14. Efficiency Analysis
  15. Ablation Study
  16. ...and 7 more sections

Figures (6)

  • Figure 1: HybridSolarNet architecture: Input images ($380 \times 380$) processed by EfficientNet-B0, refined by CBAM, and classified via a lightweight head.
  • Figure 2: Split-before-augment pipeline: raw stratified split (70/15/15), training-only augmentation and oversampling, validation/test kept raw for unbiased evaluation.
  • Figure 3: Confusion matrix on the Test Set. Strong diagonal indicates high per-class accuracy.
  • Figure 4: ROC and PR curves. Micro-average AUC $\approx$ 0.99 across six classes.
  • Figure 5: Efficiency benchmarking: FPS, size, and training time.
  • ...and 1 more figures