Improved YOLOv8 framework for efficient solar panel defect detection

Abstract

The photovoltaic (PV) industry plays a vital role in the renewable energy sector, making efficient fault detection essential to ensure optimal performance and sustainability. While the existing deep learning models have significantly improved defect detection accuracy, their large size and limited feature extraction capabilities reduce detection efficiency and complicate adaptation to varying defect conditions. Therefore, this study proposed the Improved YOLOv8-SEB model, which integrates a Squeeze-and-Excitation Block (SEB) into the YOLOv8 architecture to enhance defect detection in photovoltaic (PV) panels. The SEB dynamically reweights feature maps, allowing the model to emphasize critical features relevant to the task and improve the detection of complex and subtle defects such as “Black Border” and “Broken”. Experimental evaluations using a PV-Multi-Defect dataset show that YOLOv8-SEB achieves a 2.6% increase in mean average precision at IoU 0.5 ([email protected]) and a 3.4% gain in [email protected]:0.95 compared to the baseline YOLOv8, while reducing DFL Loss to 0.16. The improved model demonstrates enhanced precision and recall, with fewer false positives and negatives. Although the integration of SEB introduces additional computational complexity, YOLOv8-SEB remains suitable for real-time deployment on edge devices through potential optimizations. This research offers a scalable and efficient approach for solar panel defect detection, facilitating cost-effective maintenance and supporting the operational stability of PV systems. The proposed model contributes to the advancement of intelligent inspection technologies, helping to extend the lifespan of solar panels and promote the sustainability of renewable energy infrastructures.