薛宇飞 张军 张萍 李宇宸

摘  要：为探索快速、实时、精准和更具成本效益的烟草面积和产量监测方法，选取云南省德宏傣族景颇族自治州芒市为研究区域，采用Sentinel-2多光谱数据为数据源，对2020年3月份云南省德宏州芒市烟草、林地、水体等地物光谱特征和植被指数、红边指数进行了分析，应用面向对象的方法精准提取烤烟种植区域信息，提取的烟草面积为3 873.34 hm2。精度评价结果表明，提取的总体精度达到94.38%，Kappa系数为0.93，可满足烟叶生产管理的实际需求。认为以Sentinel-2多光谱影像为数据源结合面向对象的方法可以有效实现烟草种植信息的精准提取，进而为县域范围内烟草生产管理提供数据支撑。

关键词：烟草;Sentinel-2;面向对象分类;随机森林分类;精准提取

Abstract： In order to explore a fast， real-time， accurate and more cost-effective monitoring method for tobacco area and yield， Mang City， Dehong Dai Jingpo Autonomous Prefecture， Yunnan Province was selected as the study area， and Sentinel-2 multispectral data was used as the data source to analyze the spectral characteristics and vegetation index and red edge index of tobacco， woodland， water bodies and other features in Mangshi， Dehong Dai Jingpo Autonomous Prefecture， Yunnan Province， in March 2020. The object-oriented method was applied to accurately extract information on flue-cured tobacco planting area， and the extracted tobacco area was 3 873.34 ha. The accuracy evaluation results showed that the overall accuracy of extraction reached 94.38%， and the Kappa coefficient was 0.93， which could meet the actual needs of tobacco production management. It is concluded that the combination of Sentinel-2 multispectral images as the data source and the object-oriented approach can effectively realize the accurate extraction of tobacco planting information， and then provide data support for tobacco production management within the county.

Keywords： tobacco; Sentinel-2; object-oriented classification; random forest classification; accurate extraction

煙草是我国重要的经济作物之一。烟草的生长发育对光、热、水等自然环境条件反应敏感，烟叶生产周期长、种植劳动强度大、用工投入多、技术要求高[1]。快速、实时、准确和定位化获取烟草生长信息是实现烟叶精益生产、推动现代烟草农业发展的基础[2]。目前烟草面积核查、产量估计仍是基于实地踏勘和人工测量估算，并逐级上报进行统计汇总，中间过渡环节、人为因素干扰较多，且造成大量人力、物力和财力浪费[3-4]。

遥感在监测和区分不同类型的植被以及反演植被健康情况和作物成熟状况方面发挥出显著作用[5-7]，已在农作物监测领域得到广泛应用。吕小艳等[8]论述了应用遥感监测烟草病虫害、长势和产量的可行性，但受限于诸多因素，利用遥感在大尺度上对烟田面积精确提取面临挑战。……