Photovoltaic Cell Health Assessment Using Image

Abstract

With the beginning of 21st century, stimulation of improving energy efficient policies increased the public interest towards renewable energy especially solar energy through Photovoltaic Systems. It is because solar energy is noiseless and pollution free. This interest opened the research gate of achieving high optimal performance of these solar systems. In this work, classification of the PV cells with respect to health i.e. healthy or faulty is proposed. A dataset of 2624 images of PV modules (containing both healthy and faulty images) is used in this proposed work. Faulty PV cells contained shadow effect, cracked PV cells and dust contamination. The results of this work are obtained using three approaches. First approach includes extraction of hand crafted features from the original and augmented dataset and then training artificial neural network for binary and multi classification based on the extracted features of healthy and faulty solar panels. Second approach includes the transfer learning using pre-trained convolutional neural networks for the classification problem. Third approach includes the designing of customized convolutional neural network architectures to classify the PV cell dataset with respect to their health status. 94.12% accuracy for binary classification, 89.20% accuracy for binary classification with 0.5 threshold and 83.29% accuracy for multi classification is achieved as best results using the third approach.