Solar Panel Thermal Images

Our study provides fault detection of solar panels using resource-efficient UAVs (Unmanned Air Vehicle) combined with image processing and deep learning methods to facilitate the maintenance and repair of the solar panels. Our method provide technical details of faulty panels along with their specific fault types on an individual panel basis which facilitates quick and precise identification of faulty panels. We compared the classification performance of several deep learning and visual transformer models on our novel data set in terms of accuracy, precision, recall and f1-score metrics. The Swin Transformer model outperformed all other models on the test data with 0.8381 accuracy, 0.8383 precision, 0.8381 recall and 0.8368 f1-score

本研究借助资源高效型无人机（Unmanned Air Vehicle，UAV）结合图像处理与深度学习技术，开展太阳能光伏板故障检测工作，以助力光伏板的运维与检修。本研究方法可针对单块光伏板，输出故障面板的技术细节及其具体故障类型，从而实现故障面板的快速精准识别。本研究基于自建的新型数据集，从准确率、精确率、召回率与F1值四个评估指标出发，对比了多款深度学习模型与视觉Transformer模型的分类性能。其中，Swin Transformer模型在测试集上的表现优于其余所有模型，其准确率达0.8381、精确率达0.8383、召回率达0.8381、F1值达0.8368。