基于深度学习的高分辨率遥感影像光伏用地提取

近年来我国光伏产业发展迅猛,随之也产生了诸多用地问题,通过遥感技术提取光伏用地,监测光伏用地分布与用地状况,对于光伏产业健康发展具有重要意义。本文提出一套基于深度学习方法的高分辨率遥感影像光伏用地自动提取方法,该方法利用GF-1等卫星影像和Google Earth影像构建光伏用地样本,基于ResNeSt-50作为骨干网络的DeepLab V3+模型实现深度学习语义分割算法,并结合计算机图形学方法对深度学习结果进行后处理,实现了面向高分辨率遥感影像较通用的且高精度的光伏用地自动提取。该方法的深度学习模型验证精度mIoU值达0.899 2,提取结果具有良好的边缘精度且具有广泛的适用性,支持GF-1、ZY-3、GF-6、GF-2和Google Earth等影像。

Photovoltaic land extraction from high-resolution remote sensing images based on deep learning method

In recent years, the rapid development of China’s photovoltaic industry has caused a lot of land use problems. Extracting photovoltaic land through remote sensing technology and monitoring the distribution and status of photovoltaic land are of great significance for the healthy development of the photovoltaic industry. This research proposes a set of automatic photovoltaic land extraction methods for high-resolution remote sensing images. Satellite images such as GF-1 and Google Earth images are used to construct photovoltaic land samples. A deep learning semantic segmentation algorithm based on the ResNeSt-50 and DeepLab V3+model is proposed. The deep learning results are post-processed with computer graphic methods and the general-purpose and high-precision automatic extraction of photovoltaic land for high-resolution remote sensing images are achieved. The proposed deep learning model has a verification accuracy of mIoU of 0.899 2, and the extraction results have good edge accuracy. The method has wide applicability, and supports images such as GF-1, ZY-3, GF-6, GF-2, and Google Earth images.