Photovoltaic Cell Surface Defect Detection via Subtle Defect Enhancement and Background Suppression

Area of Science:

• Materials Science
• Electrical Engineering
• Computer Vision

Background:

• Photovoltaic (PV) cells are crucial for solar energy, but manufacturing defects like cracks and interruptions reduce efficiency.
• Existing defect detection methods struggle with subtle surface flaws and background noise.
• Accurate defect identification is vital for optimizing PV cell performance and reliability.

Purpose of the Study:

• To develop an advanced PV Cell Surface Defect Detector (PSDD) for precise identification and localization of subtle defects.
• To enhance the feature representation of fine defects and suppress background noise during detection.
• To improve the overall accuracy and robustness of defect detection in PV cells.

Main Methods:

• Proposed a PV Cell Surface Defect Detector (PSDD) incorporating a Subtle Feature Refinement Module (SFRM) and a Background Noise Suppression Block (BNSB).
• SFRM refines fine-grained features by rearranging spatial information and using attention mechanisms to highlight defect-related channels.
• BNSB employs a dual-path strategy with a Background-Aware Module (BAM) and residual structure for multi-scale feature fusion and noise reduction.

Main Results:

• The proposed PSDD achieved superior performance in detecting subtle surface defects on PV cells.
• PSDD demonstrated the highest mAP50 score of 93.6% on the PVEL-AD dataset, outperforming existing methods.
• The integrated SFRM and BNSB modules effectively enhanced feature representation and suppressed background noise.

Conclusions:

• The novel PSDD, with its SFRM and BNSB components, offers a significant advancement in PV cell surface defect detection.
• This method effectively addresses the challenges posed by subtle defects and complex background noise.
• The improved detection accuracy holds potential for enhancing the quality control and efficiency of photovoltaic manufacturing.