Automated Defect Detection in Solar Cell Images Using Deep Learning Algorithms

Abstract

This research study introduces a unique method that makes use of a wide range of deep learning (DL) techniques for automated flaw identification in solar cell images. The research paper investigates how well 24 distinct convolutional neural network (CNN) architectures - Residual network (ResNet), densely connected convolutional networks (DenseNet), visual geometry group (VGG), Inception, mobile network (MobileNet), Xception, SqueezeNet, and AlexNet - classify solar cells into defected and non-defective categories. This study is interesting since it does a thorough assessment of a wide variety of models and concentrates on high-performance architectures and lightweight models that may be used in contexts with limited resources. The research paper performed our studies using a balanced and well-curated dataset of 3,102 images of solar cells with a range of common faults. MobileNetV2 and Xception demonstrated excellent performance in defect identification, with accuracy rates of 99.95% and 99.29% respectively, with minimal validation losses. This study demonstrates the potential of efficient models such as MobileNetV2 for real-world use in solar energy generation. It also provides a detailed comparison of several DL models. The results suggest that the inclusion of these models might significantly enhance quality control systems, offering a reliable and efficient method for detecting flaws in solar cells.