Automatic detection of visual faults on photovoltaic modules using deep ensemble learning network

Abstract

The present study proposes an ensemble-based deep neural network (DNN) model for autonomous detection of visual faults such as glass breakage, burn marks, snail trail, and discoloration, delamination on various photovoltaic modules (PVM). The proposed technique utilizes an image dataset captured by RGB (Red, Green, Blue) camera mounted on an unmanned aerial vehicle (UAV). In the first step, the images are preprocessed by deriving spatial and frequency domain features, such as discrete wavelet transform (DWT), texture, grey level co-occurrence matrix (GLCM), fast Fourier transform (FFT), and grey level difference method (GLDM). The processed images are inserted as input in the proposed ensemble-based deep neural network (DNN) model in order to detect any visual faults on the photovoltaic modules (PVM). The performance of the proposed model is evaluated through classification accuracy, receiver operating characteristic (ROC) curve, and confusion matrix. The results demonstrate that the proposed ensemble-based deep neural network (DNN) model, along with the random forest classifier, achieved a classification accuracy of 99.68% for detecting visual faults on the PV modules. To verify the performance and robustness of the proposed model, we compare our model’s results to those of various classification approaches described in the literature. The suggested approach is compatible with the commercial unmanned aerial vehicle (UAV) embedded flight management system for fault detection.

Keywords

• Photovoltaic modules
• Autonomous fault detection
• Deep neural network
• Ensemble learning
• Random forest
• Unmanned aerial vehicle