Detecting PV module defects at a single glance

Researchers in China have developed a novel deep learning model to detect defects in photovoltaic panels.

The approach leverages high-resolution visible light imaging to identify defects using an algorithm based on the deep learning framework You Only Look Once (YOLO), which makes predictions with just a single forward pass through the neural network.

The new method also incorporates a hybrid attention transformer into the cross-stage partial bottleneck, combined with two convolutional backbones.

“Building on the foundation of YOLOv8, we introduce a novel HAT-C2f neural network module and redesign the backbone component,” the researchers explained. “In the neck section, the conventional C2f module is replaced with the RepNCSPELAN4 architecture—an efficient layer design with specified channel sizes, repetitions, and convolutions—and an SKAttention mechanism is added before the detection head.”

The HAT-C2f module enhances the backbone’s ability to extract fine image details, while integrating RepNCSPELAN4 into the neck improves feature aggregation, allowing the system to detect objects of varying sizes. Adding SKAttention to the detection head enables the model to adapt to different scales.

The model, named YOLO-HRS, was evaluated on 6,500 labeled visible-light images from the data science competition platform and online community Kaggle, categorized into four classes: clean, dust, cracked, and bird droppings. Around 80% of the images were used for training, and the remaining 20% for validation. YOLO-HRS was tested against earlier YOLO models and state-of-the-art object detection algorithms. Ablation studies, in which individual components of the model were evaluated independently, were also conducted.

The analysis showed that the YOLO-HRS achieved 86.87% precision, 84.6% recall, a mean average precision (mAP) at an intersection over union (IoU) 0.5 of 88.98%, and [email protected]:0.95 of 77.08%.

Ablation studies demonstrated substantial performance improvements in object detection, while comparisons with other models showed YOLO-HRS outperformed them. For instance, at [email protected], only YOLOX approached its performance with 85.59%, while RT-DETR, Faster-RCNN, NanoDet, and RetinaNet scored 79.34%, 66.29%, 64.16%, and 69.54%, respectively.

Compared to the baseline YOLOv8, YOLO-HRS showed a 3% improvement in [email protected].

“In summary, the model was experimentally validated using visible-light images of PV panels, confirming its high reliability and precision,” the team concluded. “YOLO-HRS provides accurate defect detection in visible-light PV panels, offering a more dependable solution for real-world applications.”

Looking ahead, the researchers plan to further optimize and extend YOLO-HRS.

“First, we will refine the model architecture and test it on diverse datasets, including infrared and electroluminescence images, to evaluate its applicability across different scenarios. Second, we aim to develop lightweight structures, including novel down-sampling and feature extraction methods, to improve the balance between accuracy and model size. Finally, exploring cross-domain applications and self-supervised learning techniques could reduce dependence on large annotated datasets,” they said.

The new method was introduced in “A novel deep learning model for defect detection in photovoltaic panels using visible light imaging,” published in Engineering Applications of Artificial Intelligence. The research was conducted by scientists from China’s Zhejiang University of Finance and Economics and Hangzhou Dianzi University.

Popular content

China strengthens intellectual property protection in solar industry

05 January 2026 Chinese authorities are planning to implement a series of policy measures to introduce stronger intellectual property protection to the country's phot...