基于高分遥感影像与面向对象的林地识别分类提取研究  

Research on forest land recognition,classification and extraction based on high-resolution remote sensing images and object orienting

作　　者：刘依扬 吕勇[1] LIU Yiyang;LYU Yong(Central South University of Forestry and Technology,Changsha 410004,Hunan,China)

机构地区：[1]中南林业科技大学,湖南长沙410004

出　　处：《湖南林业科技》2025年第1期24-30,共7页Hunan Forestry Science & Technology

基　　金：浙江省林业科技计划“建德市2023年森林资源变化遥感即时监测预警研究”。

摘　　要：针对传统软件人工地物判读效率低、分类提取复杂、自动化程度较低等问题,本研究以浙江省建德市大洋镇为研究区,在第三次全国国土资源调查数据的基础上,基于GF-1、ZY-3等卫星遥感影像,采用国产“简译”自动解译软件中的多尺度分割和面向对象分类的方法,运用最小距离法和深度学习法开展林地地类的分类识别与提取。研究结果表明:最小距离算法和深度学习算法的总体精度分别为72.06%和81.86%,Kappa系数分别为0.6275和0.7582;相较于最小距离算法,基于MobileNetV3网络结构的深度学习分类算法精度更高,在速度和准确性方面更为平衡高效,能较好地满足大范围林地快速提取与分类的需求。Aiming at the problems of low efficiency of manual feature interpretation,complicated classification and extraction,and low degree of automation by traditional software,the classification recognition and extraction of forest land types in Dayang Town,Jiande City,Zhejiang Province were carried out by the methods of multi-scale segmentation and object-oriented classification in the domestic“Easy Interpretation”automatic interpretation software,the minimum distance method and the deep learning method on the basis of the data of the third national land resources survey and satellite remote sensing images such as GF-1 and ZY-3.The results showed that the overall accuracy of the minimum distance algorithm and the deep learning algorithm were 72.06%and 81.86%,respectively,and the Kappa coefficients were 0.6275 and 0.7582,respectively.Compared with the minimum distance algorithm,the deep learning classification algorithm based on the structure of the MobileNetV3 network was more accurate,more balanced and efficient in terms of speed and accuracy,and it could better meet the needs of rapid extraction and classification of large-scale forest land.

关 键 词：林地地类 简译 面向对象分类 高分辨率遥感影像 深度学习 

分 类 号：S791.27[农业科学—林木遗传育种]