基于ENVINet5的高分辨率遥感影像稀疏塑料大棚提取研究

随着设施农业管理要求的提高，需要提取高分辨率遥感影像中大范围、低密度的塑料大棚空间分布信息作为农业管理和资源分配的依据。以浙江省桐庐县为研究区域，利用高分辨率遥感影像数据，对比分析不同机器学习方法提取塑料大棚的效果。ENVINet 5深度学习架构可以克服标签较少的困难，通过语义学习进行塑料大棚提取和面积估算，总体精度和Kappa系数达到97.84%和0.81；U-net深度学习网络的提取结果中，总体精度和Kappa系数为96.22%和0.79，两种深度学习方法均优于利用支持向量机进行塑料大棚提取的结果。研究表明通过深度学习方法提取高分辨率遥感影像中稀疏分布的塑料大棚有很好的效果，可以为农业经济作物管理、规划和气象保障提供支持。

Research on the Sparse Plastic Shed Extraction from High Resolution Images Using ENVINet 5 Deep Learning Method

With the increasing requirements of facility agriculture management， it is necessary to extract the spatial distribution information of plastic greenhouses with large range and low density in high-resolution remote sensing images as the basis for agricultural management and resource allocation. This study takes Tonglu County， Zhejiang Province as the study area， and uses high-resolution remote sensing images to compare and analyze the effect of extracting plastic sheds using different machine learning methods. It was found that the ENVINet5 deep learning architecture could perform plastic shed extraction and area estimation by small-sample semantic learning， and the overall accuracy and kappa coefficient reached 97.84% and 0.81； in the extraction results of U-net deep learning network， the overall accuracy and kappa coefficient were 96.22% and 0.79， which were better than the plastic shed extraction using support vector machine results. This study shows that the extraction of sparsely distributed plastic sheds in high-resolution remote sensing images by deep learning has good results and can provide support for agricultural cash crop management， planning， and weather assurance.