With the growing demand for renewable energy, solar panel quality control has become critical in photovoltaic industries. This paper proposes LPV-YOLO – a lightweight detection network combining Ghost modules, attention mechanisms, and spatial pyramid pooling to achieve efficient solar panel defect identification.
1. Network Architecture
The LPV-YOLO framework consists of:
- Backbone: GhostMConv and C3MGhost modules reduce parameters while maintaining feature extraction capability
- Neck: SE attention enhances channel-wise feature interaction
- Head: Modified detection heads with MSSPPF for multi-scale fusion
2. Key Improvements
2.1 Lightweight Modules
The Ghost module generates feature maps through linear operations:
$$r_s = \frac{c \cdot k^2}{\frac{1}{s} \cdot c \cdot k^2 + \frac{s-1}{s} \cdot d^2} \approx s$$
2.2 Attention Pyramid Pooling
MSSPPF integrates SimAM attention with spatial pyramid pooling:
$$e_t = \frac{4(\hat{\sigma}^2 + \lambda)}{(t – \hat{\mu})^2 + 2\hat{\sigma}^2 + 2\lambda}$$
2.3 Channel Attention
SE block recalibrates channel weights:
$$X_{out} = \sigma\left(\frac{1}{E}\right) \otimes X$$
3. Experimental Results
3.1 Dataset Preparation
| Defect Type | Original | Augmented |
|---|---|---|
| Crack | 4235 | 14444 |
| Hotspot | 4094 | 14444 |
3.2 Performance Comparison
| Model | mAP(%) | Params(M) | Size(MB) |
|---|---|---|---|
| YOLOv5s | 94.4 | 7.23 | 13.7 |
| LPV-YOLO | 93.8 | 3.71 | 7.4 |
The proposed method achieves 70.42 FPS with minimal accuracy loss, making it suitable for solar panel inspection drones and edge devices.
4. Defect Detection Metrics
Critical evaluation metrics for solar panel inspection:
$$mAP = \frac{1}{n}\sum_{i=1}^n AP_i$$
$$AP = \int_0^1 P(R)dR$$
Experimental results demonstrate superior performance in detecting various solar panel defects including micro-cracks, hot spots, and electrical failures.
5. Implementation Details
| Parameter | Value |
|---|---|
| Input Size | 640×640 |
| Batch Size | 16 |
| Learning Rate | 0.001 |
The network optimization focuses on maintaining detection accuracy while reducing computational complexity, crucial for large-scale solar farm inspections.
6. Conclusion
LPV-YOLO demonstrates efficient solar panel defect detection with:
- 49% parameter reduction compared to baseline
- 93.8% mAP on multi-defect dataset
- Real-time performance (70.42 FPS)
This approach provides a practical solution for automated solar panel quality control, significantly reducing manual inspection costs while ensuring reliable defect identification.