Automatic Detection of Photovoltaic Module Cells using Multi-Channel Convolutional Neural Network

摘要

Due to the complexity production of photovoltaic (PV) module cells, it is easy to generate defects such as broken grid, open weld and hidden crack in many processes. Based on artificial feature extraction method is time-consuming, low recognition rate, the traditional convolutional neural network (CNN) relys on a single channel to extract image feature is not sufficient, this paper proposes a method of multi-channel convolutional neural network (MCCNN) to detect the defects in PV module cells, multi-channel has the scale of different image size, it is able to extract the image feature from different scale, The features are fused on the fully connected layer, finally through the Random Forest (RF) classifier to classify, it can improve the accuracy of recognition. Simulation results show that the MCCNN can quickly and accurately identify the PV module cells defect and defect categories, and accurately mark it on the original image.