Solar Cell Busbars Surface Defect Detection Based on Deep Convolutional Neural Network

Abstract

Defect detection of the solar cell surface with texture and complicated background is a challenge for solar cell manufacturing. The classic manufacturing process relies on human eye detection, which requires many workers without a steady and good detection effect. In order to solve the problem, a visual defect detection method based on a new deep convolutional neural network (CNN) is designed in this paper. First, we develop a CNN model by adjusting the depth and width of the model. Then, the optimal CNN model structure is developed by comparing the performance of different depth and width combinations. This research focuses on finding a way to distinguish defects in solar cells from the background texture of busbars and fingers. The characteristics of solar cell color images are analyzed. We find that defects exhibited different distinguishable characteristics in various structures. The deep CNN model is constructed to enhance the discrimination capacity of the model to distinguish between complicated texture background features and defect features. Finally, some experimental results and K-fold cross-validation show that the new deep CNN model can detect solar cell surface defects more effectively than other models. The accuracy of defect recognition reaches 85.80%. In solar cell manufacturing, such an algorithm can increase the productivity of solar cell manufacturing and make the manufacturing process smarter.