DEEP LEARNING-BASED AUTOMATED DEFECT DETECTION IN SOLAR CELL IMAGES
Thota Prathap¹, Dipak Kumar Yadav¹, Aklesh Mishra¹
Under the guidance of Mr. G. Lakpathi, Assistant Professor
¹Department of Computer Science & Engineering, Guru Nanak Institute of Technology
Affiliated to JNTUH-Hyderabad, Ranga Reddy District-501506, Telangana, India
Email: {22831A05K1, 22831A05K6, 22831A05K7}@gniindia.org
Abstract—This paper presents an automated deep learning-based methodology for detecting defects in solar cell images using the Xception convolutional neural network architecture. Solar energy production is critically dependent on the quality and efficiency of individual solar cells; defects such as micro-cracks, scratches, and surface irregularities can substantially reduce energy output and panel longevity. Traditional manual inspection methods are labor-intensive, inconsistent, and prone to human error, limiting scalability in industrial settings. The proposed system leverages depthwise separable convolutions inherent to the Xception architecture to extract complex, hierarchical features from high-resolution solar cell images, enabling precise differentiation between defective and non-defective cells while maintaining computational efficiency suitable for resource-constrained environments. A balanced and well-curated dataset of electroluminescence (EL) solar cell images spanning diverse defect categories was employed for training and validation. The pipeline incorporates rigorous preprocessing steps including image normalization, resizing to 299×299 pixels, and data augmentation (rotation, flipping, zooming, and brightness adjustment) to improve generalization and mitigate overfitting. Experimental results demonstrate that the Xception model achieves 93% overall classification accuracy across 11 defect classes, with performance metrics—precision, recall, and F1-score—confirming strong reliability for automated quality control in solar panel manufacturing.
Keywords—Solar Cell Defect Detection, Deep Learning, Xception Architecture, Convolutional Neural Networks, Transfer Learning, Electroluminescence Imaging, Image Classification, Quality Control.
