Remote sensing image-based photovoltaic panel extraction via deep learning
Solar energy,represented by the photovoltaic industry,is considered one of the safest and cleanest renewable energy sources for the future. The utilization of high-resolution remote sensing images for extracting and monitoring the land distribution of photovoltaic panels holds immense significance for the development and supervision of the photovoltaic industry. This paper utilized high-resolution remote sensing images from Ordos in 2022 to create a sample set for photovoltaic panel extraction. Five commonly used semantic segmentation models were constructed,namely fully convolutional network (FCN),pyramid scene parsing network (PSPNet),DeepLabV3+,U-Net,and U-Net++,so as to perform semantic segmentation on photovoltaic panels. By comparing extraction accuracy and visual effects of different models,an automatic identification model was set for photovoltaic panel extraction. The results demonstrate that the U-Net++model achieves the highest accuracy,with a pixel accuracy of 96％,an F1 of 95％,and an intersection-over-union (IoU) of 90.4％ in the test set.
deep learningsemantic segmentationphotovoltaic power generationhigh-resolution remote sensing imagesOrdos
万方数据
维普
