基于PIE-Engine的红塔区烟草种植面积提取研究

A Study on the Extraction of Tobacco Planting Area in Hongta District Based on PIE-Engine

摘要: 烟草是云南省的主要经济作物之一，其种植面积和产量信息是农业部门制定相关政策的重要依据。因此，实时、精准和更具成本效益地确定烟草种植面积和监测烟草生长状况的方法极为重要。本文以云南省玉溪市红塔区为研究区域，采用Sentinel-2卫星影像为数据源，利用PIE-Engine遥感云平台，对2021年该研究区的遥感影像进行解译，采用随机森林、支持向量机、神经网络和深度学习四种分类方法分别提取烟草种植面积信息，并进行对比分析获取最优分类算法，为相应的农业生产提供指导，通过研究分析得出深度学习提取结果最为准确，总体精度(OA)达到94.70%，Kappa系数为0.92，提取的烟草面积为1989.36 hm2，最为接近红塔区2021年统计公报中的烟草种植面积，误差仅为3.12%。

Abstract: Tobacco is one of the major cash crops in Yunnan Province, and its planted area and yield infor-mation are an important basis for the agricultural sector to formulate relevant policies. Therefore, a real-time, accurate, and more cost-effective method of determining tobacco acreage and monitoring tobacco growth is extremely important. In this paper, we use Sentinel-2 satellite images as the data source and the PIE-Engine remote sensing cloud platform to interpret the remote sensing images of this study area in 2021 and use four classification methods: random forest, support vector machine, neural network, and deep learning to extract tobacco planting area information respectively, and compare them, and conclude that deep learning The extracted tobacco area was 1989.36 hm2, which was the closest to the tobacco cultivation area in the statistical bulletin of Hongta District in 2021, with an error of only 3.12%.