Lightweight Convolutional Network For Automated Photovoltaic Defect Detection

As the World moves towards renewable energy, photovoltaic modules are a fundamental option due to their green nature. However, the manufacturing process of solar cells is complex and vulnerable to discrepancies which can impact the overall performance of the system. Although human-led inspection is seen as the de-facto quality inspection protocol, issues pertaining to bias, cost and time can make it an expensive process. To this effect, this paper focuses on the development of a custom convolutional architecture that is lightweight, hence deployable within manufacturing facilities to assist with defective solar cell inspection. In addition, to address the issue of data scarcity, representative data augmentations are producing tailored towards enhancing the model's generalizability. The high efficacy of the proposed CNN and proposed augmentations can be gauged by the fact that 98% F1 score was achieved overall.