Optimized YOLOv5 for Real-Time Surface Defect Detection in Solar Cells

Abstract

The rapid growth of renewable energy has increased the demand for efficient solar cell production, 
where surface defects significantly impact performance and longevity. Solar cell defect detection is 
critical to ensure quality, yet traditional methods often fail to meet modern manufacturing demands. 
The problem lies in accurately identifying defects like "Mono" or "Poly" on solar cell surfaces, which 
can reduce efficiency if undetected. Traditional systems primarily rely on manual inspection or basic 
image processing techniques, involving human experts visually examining cells or using rule-based 
algorithms to detect anomalies. These methods are time-consuming, prone to human error, and lack 
scalability, with manual inspections often missing subtle defects and automated systems struggling with 
complex patterns, leading to a detection accuracy as low as 70% in some cases. The need for an 
automated, accurate, and scalable solution is evident to enhance production quality and reduce costs in 
the solar industry. The proposed system is a Django-based web application integrated with an optimized 
YOLOv5 model, enabling users to upload solar cell images and receive defect predictions with 
probabilities, such as identifying a "Mono" defect with a 0.80526405 probability. It leverages OpenCV 
for image preprocessing (resizing to 32x32, HSV conversion, flipping), Keras for model inference, and 
Matplotlib for visualization, displaying results via base64 encoding on a user-friendly interface. The 
system includes user authentication, file handling, and prediction workflows, achieving higher accuracy 
and automation compared to traditional methods. Its significance lies in streamlining defect detection, 
reducing manual effort, and improving scalability for large-scale solar cell production, potentially 
increasing detection accuracy to over 90%. This solution bridges the gap between machine learning and 
industrial applications, offering a cost-effective, reliable tool for quality assurance in renewable energy 
manufacturing.