Abstract
The photovoltaic (PV) industry is essential to global renewable energy generation; however, inefficient and accurate fault detection in PV panels remains challenging. Conventional approaches, such as manual inspection and I–V curve tracing, are labor-intensive and inefficient, while infrared (IR) imaging with unmanned aerial vehicles (UAVs) provides scalability but limited accuracy in detecting small or complex defects. In this study, a lightweight real-time detection model, TA-YOLOv11, is proposed for UAV-based IR PV panel defect identification. Edge details in UAV-captured IR images are enhanced through image sharpening, an improved two-orientation attention aggregation (I-TOAA) block is employed for cross-layer feature fusion and noise suppression, and an adaptive threshold focal loss (ATFL) function is introduced to adjust loss weights for hard-to-detect targets. The model is deployed on a Jetson Orin Nano edge computing platform, achieving balanced accuracy and real-time performance at 18 frames per second (FPS) for large-scale PV inspections. Experimental results indicate that TA-YOLOv11 surpasses the baseline YOLOv11 model, with precision, recall, and mean average precision (mAP) improved by 3.0%, 3.3%, and 1.4%, respectively, confirming its effectiveness for automated industrial PV defect detection.
| Original language | English |
|---|---|
| Article number | 5050412 |
| Journal | IEEE Transactions on Instrumentation and Measurement |
| Volume | 74 |
| DOIs | |
| State | Published - 2025 |
| Externally published | Yes |
UN SDGs
This output contributes to the following UN Sustainable Development Goals (SDGs)
-
SDG 7 Affordable and Clean Energy
Keywords
- Infrared (IR) image
- object detection
- sensor data processing
- solar panel
Fingerprint
Dive into the research topics of 'A Lightweight Model for Infrared Photovoltaic Panel Defect Detection via UAV-Based Inspection Systems'. Together they form a unique fingerprint.Cite this
- APA
- Author
- BIBTEX
- Harvard
- Standard
- RIS
- Vancouver