%0 Journal Article %T 基于深度学习的光伏电池缺陷自动分类系统
Deep Learning-Based Automatic Classification System for Photovoltaic Cell Defects %A 李研 %J Artificial Intelligence and Robotics Research %P 648-661 %@ 2326-3423 %D 2024 %I Hans Publishing %R 10.12677/airr.2024.133066 %X 随着太阳能光伏技术的稳步提升，其性能和稳定性等指标会直接影响系统的工作效率。但是在光伏电池的生产中，常常会出现各种缺陷，这些缺陷可能导致功率输出降低、寿命缩短，甚至严重影响系统的安全和可靠性。传统的光伏电池缺陷分类方法主要依赖于人工提取特征和手动分类，存在效率低、准确性不高的问题。为了解决这一问题，本研究提出了一种基于深度学习的光伏电池缺陷分类方法。本文选择了ResNet-18、ResNet-34、ResNet-50和VGG-16四种不同的深度学习模型作为研究对象，并对不同的模型进行针对性改进，再通过细致的实验设置和严格的评估标准，比较这些改进后的模型在光伏电池缺陷分类任务上的效果。从模型的准确率、损失函数、收敛速度等多个维度进行比较分析，以找出最适合光伏电池缺陷分类任务的网络模型。为光伏电池制造和运行过程中的质量控制和故障诊断提供重要支持。未来，将进一步优化算法性能，探索更加高效和精确的深度学习模型，并结合实际应用需求，推动该技术在光伏电池领域的广泛应用与推广。
With the steady improvement of solar photovoltaic technology, its performance and stability and other indicators will directly affect the efficiency of the system. However, in the production of photovoltaic cells, there are often various defects, which may lead to reduced power output, shortened life, and even seriously affect the safety and reliability of the system. Traditional defect classification methods of photovoltaic cells mainly rely on manual feature extraction and manual classification, which has low efficiency and low accuracy. In order to solve this problem, a deep learning-based defect classification method for photovoltaic cells is proposed in this study. In this paper, four different deep learning models, ResNet-18, ResNet-34, ResNet-50 and VGG-16, were selected as research objects, and targeted improvements were made to the different models. Then, through careful experimental Settings and strict evaluation criteria, the effects of these improved models on the task of photovoltaic cell defect classification were compared. The accuracy, loss function and convergence rate of the model are compared and analyzed to find out the most suitable network model for defect classification of photovoltaic cells. It provides important support for quality control and fault diagnosis during the manufacturing and operation of photovoltaic cells. In the future, the algorithm performance will be further optimized to explore more efficient and accurate deep learning models, and combined with practical application needs, to promote the wide application and promotion of this technology in the field of photovoltaic cells. %K 缺陷分类， %K 深度学习， %K 光伏电池， %K 图像分类
Defect Classification %K Deep Learning %K Photovoltaic Cells %K Image Classification %U http://www.hanspub.org/journal/PaperInformation.aspx?PaperID=95458