A photovoltaic cell defect detection model capable of topological knowledge extraction

Area of Science:

• Materials Science
• Renewable Energy Engineering
• Computer Vision

Background:

• The global shift to clean energy drives solar adoption, but traditional defect detection struggles with noisy photovoltaic cell electroluminescence (EL) images.
• Existing object detection models are insufficient for the unique challenges presented by EL imaging in solar cells.

Purpose of the Study:

• To develop an advanced defect detection model for photovoltaic cells that overcomes limitations of traditional methods.
• To enhance the accuracy of defect identification in solar cell EL images through novel feature extraction and topological analysis.

Main Methods:

• Implemented a multi-scale dynamic context-based feature extraction for capturing local texture and structural defect information.
• Introduced a centralized feature pyramid structure with spatial semantics to model visual centers and relate local/global features.
• Utilized a spatial semantic knowledge extraction strategy to build feature topology and uncover defect correlations.

Main Results:

• The developed model effectively captures fine-grained local features by combining static and dynamic contexts.
• The spatial semantic pyramid structure improved the representation of defect characteristics by elucidating feature relationships.
• The feature enhancement strategy led to precise defect detection by mapping features to a higher-order representation.

Conclusions:

• The proposed defect detection model, integrating topological knowledge, significantly enhances the analysis of photovoltaic cell EL images.
• This advancement supports the reliable quality control of solar cells, crucial for the accelerating clean energy transition.