EBBA-detector: An effective detector for defect detection in solar panel EL images with unbalanced data

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Vision

Background:

• Solar panel defect detection is vital for quality control.
• Challenges include varied defect sizes, background noise, and imbalanced data.

Purpose of the Study:

• To propose an effective solar panel defect detection model.
• To address limitations of existing methods in handling diverse defect characteristics and data distributions.

Main Methods:

• Introduced the EBBA-Detector, featuring an Enhanced Bidirectional Feature Pyramid Network (EBFPN) and a Balanced-Attention Module (B-A Module).
• Developed a Scaled Dynamic Focal Loss (SDFL) function to manage imbalanced datasets.
• Validated the model on a large-scale electroluminescence (EL) dataset.

Main Results:

• Achieved a mean Average Precision (mAP) of 89.85% on the EL dataset.
• Demonstrated superior performance in detecting multiple defect categories compared to other models.
• Showcased effectiveness in detecting small defects and handling background interference.

Conclusions:

• The EBBA-Detector provides an efficient and accurate solution for solar panel defect detection.
• The proposed framework effectively handles small targets, large targets, and imbalanced data distributions.
• This work contributes to advancing automated quality control in solar panel manufacturing.