Deep Convolutional Neural Networks for Detecting Solar Cell Defects

Solar cell defect detection is a crucial assurance for the dependably functioning of solar panels. The artificial feature selection required, the excessive number of training parameters, and the subpar detection performance for particular defect kinds are some of the issues with the machine learning defect detection approach based on computer vision technology. Deep convolutional neural network (DCNN)-based fault detection for solar cells is proposed. This method builds a deep network with three convolution layers, one pooling layer, one fully connected layer, and one output layer using solar cell photos as the input and a distinguishing defect category as the detection goal. The parameter number optimization technique, the parameter adjustment algorithm, and measures to address the overfitting issue are proposed during the training of network parameters. The experimental findings on the dataset demonstrate that the DCNN method's detection accuracy for solar cell faults can reach more than 97% and its F value can reach 0.690. This method's real-time detection speed can satisfy the needs of practical production, and its detection accuracy in each defect category is higher than that of previous approaches.