欢迎访问林草资源研究

林业资源管理 ›› 2023›› Issue (1): 141-152.doi: 10.13466/j.cnki.lyzygl.2023.01.017

• 技术应用 • 上一篇    下一篇

基于面向对象的吉林一号遥感影像湿地植被群落分类

谢文春1(), 李强峰1(), 李艳春1, 吴振山2, 杨正凡2   

  1. 1.青海大学 农牧学院,西宁 810016
    2.乌兰县自然资源局和林业草原局,青海 乌兰 817199
  • 收稿日期:2022-12-24 修回日期:2023-02-12 出版日期:2023-02-28 发布日期:2023-05-05
  • 通讯作者: 李强峰(1973-),男,青海西宁人,教授,主要研究方向为林业3S技术。Email:1324985124@qq.com
  • 作者简介:谢文春(1993-),男,四川攀枝花人,硕士研究生,主要研究方向为林业3S技术。Email:1435368854@qq.com
  • 基金资助:
    2021年第二批林业改革基金(青财资环字〔2021〕1732号)

Object-Oriented Classification of Wetland Vegetation Community in Jilin-1 Remote Sensing Image

XIE Wenchun1(), LI Qiangfeng1(), LI Yanchun1, WU Zhenshan2, YANG Zhengfan2   

  1. 1. College of Agriculture and Animal Husbandry,Qinghai University,Xining 810016,China
    2. Wulan County Natural Resources Bureau and Forestry and Grassland Bureau,Wulan,Qinghai 817199,China
  • Received:2022-12-24 Revised:2023-02-12 Online:2023-02-28 Published:2023-05-05

摘要:

利用遥感技术提取湿地植被群落组成与分布,对湿地的建设具有重要意义。以青海乌兰都兰湖国家湿地公园为研究区,利用吉林一号遥感影像,通过影像分割,特征优化,选择KNN与RF分类模型,对都兰湖湿地植被群落进行划分,并验证分类精度。结果表明,根据ESP 2工具提供的分割尺度,植被群落面向对象分类的最优分割尺度为18,植被与非植被区域分割尺度为32和85。地物类型划分方面,仅利用影像波段信息及相关指数的阈值不能精确提取地物类别,需要结合影像几何特征和纹理特征提高分类精度,利用特征空间优化工具对61个影像特征进行优化,最终筛选出了40个影像特征并用于分类。根据混淆矩阵分类精度评价结果,KNN算法分类结果优于RF,其中KNN总体分类精度为81.80%,Kappa系数为0.79;RF总体分类精度为72.59%,Kappa系数为0.68。根据分类结果,都兰湖湿地植被覆盖率为44.41%,植被群落的组成及分布特征可以为湿地生态建设及管理提供依据。

关键词: 都兰湖, 植被群落分类, 面向对象, 影像特征, 分类模型

Abstract:

The use of remote sensing technology to extract the composition and distribution of wetland vegetation communities is of great significance to the construction of wetlands.Taking Qinghai Ulandulan Lake National Wetland Park as the research area,using the number of Jilin No.1 remote sensing images,KNN and RF classification models were selected through image segmentation and feature optimization,the vegetation community of Dulan Lake wetland was divided,and the classification accuracy was verified.The results showed that according to the segmentation scale provided by ESP 2 tool,the optimal segmentation scale for object-oriented classification of vegetation communities was 18,and the segmentation scales of vegetation and non-vegetation areas were 32 and 85,respectively.For character type division,only using the threshold of image band information and related index could not accurately extract the feature category,it was necessary to combine the image geometric features and texture features to improve the classification accuracy,use the feature space optimization tool to optimize 61 image features,and finally screen out 40 image features,and use them for classification.According to the confusion matrix classification accuracy evaluation results,the classification results of KNN algorithm were better than RF,among which the overall classification accuracy of KNN was 81.80%,the Kappa coefficient was 0.79,the overall classification accuracy of RF was 72.59%,and the Kappa coefficient was 0.68.According to the classification results,the vegetation coverage rate of Dulan Lake wetland was 44.41%,and the composition and distribution characteristics of vegetation communities in the results could provide a basis for wetland ecological construction and management.

Key words: Lake Dulan, vegetation community classification, object-oriented, image features, classification model

中图分类号: