基于U-Net和超像素分割的烟株自动提取分析  

Automatic extraction and analysis of tobacco plants based on U-Net and superpixel segmentation

作　　者：杨威 黄亮[1,2] 陈裕汉 夏炎 常军 YANG Wei;HUANG Liang;CHEN Yuhan;XIA Yan;CHANG Jun(Faculty of Land Resource Engineering,Kunming University of Science and Technology,Kunming 650093,China;Surveying and Mapping Geo-Informatics Technology Research Center on Plateau Mountains of Yunnan Higher Education,Kunming 650093,China;West Yunnan University of Applied Sciences,Dali 671006,Yunnan,China;The First Geodetic Surveying Brigade of Ministry of Natural Resources,Xi’an 710054,China)

机构地区：[1]昆明理工大学国土资源工程学院,昆明市650093 [2]云南省高校高原山区空间信息测绘技术应用工程研究中心,昆明市650093 [3]滇西应用技术大学,云南省大理市671006 [4]自然资源部第一大地测量队,西安市710054

出　　处：《烟草科技》2022年第4期20-27,共8页Tobacco Science & Technology

基　　金：国家自然科学基金项目“南方山地城镇建设用地分布与变化的坡度梯度效应研究”(41961039);云南省应用基础研究计划面上项目“基于全卷积神经网络的多源遥感影像变化检测”(2018FB078)。

摘　　要：为解决田间烟株自动识别和计数问题,基于U-Net和SLIC超像素分割,建立了一种烟株自动识别与计数的方法。首先通过训练语义分割网络U-Net提取烟田面积;然后构建过绿差值指数(Excess Green Difference Index,EGDI)去除杂草并提取烟株覆盖面;再使用简单线性迭代聚类(Simple Linear Iterative Clustering,SLIC)算法对烟草覆盖面进行分割;最后将分割出的烟株进行自动标记和计数。结果表明,采用U-Net网络对烟田面积提取得到的平均交并比(Mean Intersection over Union,MIoU)达到98.24%。运用该方法提取烟株的平均总体精度为99.21%,平均准确度为93.42%,表明该方法对烟株提取和计数具有可行性。To automatically recognize and count tobacco plants in fields,an automatic tobacco plant recognizing and counting method was proposed based on U-Net and Simple Linear Iterative Clustering(SLIC)superpixel segmentation.Firstly,the area of tobacco fields was extracted by training semantic segmentation network U-Net,then excess green difference index(EGDI)algorithm was constructed to remove any weeds and to extract tobacco plant coverage,and further the tobacco plant coverage was segmented using the SLIC algorithm.Finally,the segmented tobacco plants were automatically labeled and counted.The results showed that the Mean Intersection over Union(MIoU)obtained by the U-Net network for tobacco field extraction reached 98.24%.Average overall accuracy of 99.21%and average precision of 93.42%were obtained by using this method to extract tobacco plants,which indicated the feasibility of this method for large-scale tobacco plant extraction and counting.

关 键 词：烟株 自动识别 深度学习 超像素分割 简单线性迭代聚类 

分 类 号：S476[农业科学—农业昆虫与害虫防治]