Identifying defective solar cells in electroluminescence images using deep feature representations

Identifying defective solar cells in electroluminescence images using deep feature representations Alaa S. Al‐Waisy1, Dheyaa Ahmed Ibrahim1, Dilovan Asaad Zebari2, Shumoos Hammadi3, Hussam Mohammed4, Mazin Abed Mohammed5 and Robertas Damaševičius6 1 Computer Engineering Techniques Department, Information Technology College, Imam Ja’afar Al-Sadiq University, Baghdad, Iraq 2 Department of Computer Science, College of Science, Nawroz University, Duhok, Kurdistan Region, Iraq 3 Computer Science Department, Al-Ma’aref University College, Ramadi, Anbar, Iraq 4 Computer Center, University of Anbar, Ramadi, Anbar, Iraq 5 Information systems Department, College of Computer Science and Information Technology, University of Anbar, Ramadi, Anbar, Iraq 6 Department of Software Engineering, Kaunas University of Technology, Kaunas, Lithuania ABSTRACT Submitted 17 February 2022 Accepted 4 May 2022 Published 19 May 2022 Corresponding author Robertas Damaševičius, Academic editor Vimal Shanmuganathan Additional Information and Declarations can be found on page 17 DOI 10.7717/peerj-cs.992 Copyright 2022 Al‐Waisy et al. Distributed under Creative Commons CC-BY 4.0 Electroluminescence (EL) imaging is a technique for acquiring images of photovoltaic (PV) modules and examining them for surface defects. Analysis of EL images has been manually performed by visual inspection of images by experts. This manual procedure is tedious, time-consuming, subjective, and requires deep expert knowledge. In this work, a hybrid and fully-automated classification system is developed for detecting different types of defects in EL images. The system fuses the deep feature representations extracted from two different deep learning models (Inception-V3 and ResNet50) to form more discriminative feature vectors. These feature vectors are then fed into the classifier layer to assign them into one of different types of defects. A large-scale, challenging solar cells dataset composed of 2,624 EL images was used to assess the performance of the proposed system in both the binary classification (functional vs defective) task and multi-class classification (functional, mild, moderate, and severe) task. The proposed system has managed to detect the correct defect type with less than 1 s per image with an accuracy rate of 98.15% and 95.35% in the binary classification and multi-classification task, respectively. Subjects Artificial Intelligence, Computer Vision, Neural Networks Keywords Electroluminescence imaging, Solar cells, Photovoltaics, Defect recognition, Deep learning INTRODUCTION People and governments are taking steps to decrease consumption of fossil fuels used in transportation and power plants and increase the use of green (renewable) energy sources (Okewu et al., 2017). Solar energy systems are a renewable energy source and have gained wide attention in recent years, especially in microgrids (Ayodele et al., 2019). Solar power plants have been developed around the world, resulting in the activation of large-scale manufacturing facilities that create solar energy components (Demirci, Beşli & How to cite this article Al‐Waisy AS, Ibrahim DA, Zebari DA, Hammadi S, Mohammed H, Mohammed MA, Damaševičius R. 2022. Identifying defective solar cells in electroluminescence images using deep feature representations. PeerJ Comput. Sci. 8:e992 DOI 10.7717/ peerj-cs.992 Gümüşçü, 2021). One of the most important and sensitive components is the solar panels which should be protected from damage. Solar cell damage is primarily caused by environmental exposure or during the manufacturing process of solar panels. Solar panels are often shielded from environmental impacts such as rain, wind, and snow by an aluminum frame and a layer of glass lamination on the outside. Although these safeguards are in place, they are not always effective in preventing mechanical damages such as when the PV module is dropped during installation, the effect of falling tree limbs, hail, or heat stress. Apart from that, manufacturing defects such as improper soldering or defective wires are also able to result in damaged PV modules. Such faults have a negative impact on solar module quality, increase the amount of electricity lost, and considerably decrease the efficiency of solar panels (Li et al., 2019; Deitsch et al., 2019). Therefore, the solar cell should be used carefully to avoid damaged that might affect its performance. The electroluminescence (EL) imaging is a technique that provide an images of photovoltaic (PV) modules and examining them to provide insights into a range of some defects on the surface of PV modules (i.e. damaged cells) (Buerhop et al., 2018). Consequently, it is vital to monitor the state of solar modules and to replace or repair any units that are found to be defective to ensure that solar power plants operate at their greatest efficiency (Akram et al., 2019). The EL image examination manually is an expensive and time-consuming task. It requires sufficient experience and knowledge of the subject, and is only possible on a small scale. Autonomous PV inspection is important when working on a wide scale (Bedrich et al., 2018; Bedrich et al., 2018). Therefore, the main objective of this work is to propose an automated trainable model that can help to identify different defects of solar cell using the advance deep learning techniques. To achieve, the goal, a hybrid and fully automated supervised classification system for the automated detection of different defects in EL images of solar cells is developed. The proposed classification system depends on the feature representations extracted from two deep learning approaches (Inception-V3 and ResNet50). The contributions of this study can be summarized as follows.
A hybrid and fully automated supervised classification system for automated identification of different faults in EL images of solar cells is developed. This is the first work to develop such a solution for distinguishing between EL images of functional and defective solar cells using powerful and discriminative feature representations obtained from two deep learning models (Inception-V3 and ResNet50).
A training strategy supported by training tricks (dropout method, data augmentation) is employed to avoid overfitting and enhance generalization capability of deep learning models. The remainder of this paper is structured as follows: “Related Work” includes briefly review for the current state-of-the-art related works. “Proposed System” provides an overview of the proposed system. The experimental results of the proposed system are presented in “Experimental Results”. Finally, the conclusions and future research directions are reported in the “Conclusions and Future Work”. Al‐Waisy et al. (2022), PeerJ Comput. Sci., DOI 10.7717/peerj-cs.992 2/18 Figure 1 An overview of the proposed hybrid and fully-automated classification system for detecting different types of defects in EL images of solar cells. Figures of cells are (...truncated)