测绘通报 ›› 2018, Vol. 0 ›› Issue (7): 78-82.doi: 10.13474/j.cnki.11-2246.2018.0215

• 技术交流 • 上一篇    下一篇

多源卫星数据的农用地膜信息提取

李佳雨1,2, 王华斌1,2, 王光辉2, 翟浩然2, 韩旻1,2, 程前1,2   

  1. 1. 辽宁工程技术大学, 辽宁 阜新 123000;
    2. 国家测绘地理信息局卫星测绘应用中心, 北京 100048
  • 收稿日期:2017-09-04 出版日期:2018-07-25 发布日期:2018-08-02
  • 作者简介:李佳雨(1992-),男,硕士生,主要从事遥感影像地物提取及生态研究工作。E-mail:sas_ljy@163.com
  • 基金资助:
    国家重点研发计划(2016YFB0501403)

Plastic-mulched Farmland Extraction with Multi-source Satellite Data

LI Jiayu1,2, WANG Huabin1,2, WANG Guanghui2, ZHAI Haoran2, HAN Min1,2, CHENG Qian1,2   

  1. 1. Liaoning Technical University, Fuxin 123000, China;
    2. Satellite Surveying and Mapping Application Center, Beijing 100048, China
  • Received:2017-09-04 Online:2018-07-25 Published:2018-08-02

摘要: 地膜的大量应用使得我国农业生产力快速发展,随着使用量的逐年增长,其对环境的影响已经不容忽视。遥感影像分类技术是农用地膜监测的重要方法。本文基于FSO特征优选和面向对象随机森林分类方法,利用光谱、几何、纹理等特征设计多组试验方案,使用资源三号与Landsat卫星融合影像对甘肃中部地区进行了农用地膜提取。试验结果表明:①多特征融合方法的地膜信息提取精度明显高于单一特征提取方法;②采用FSO特征优选的试验方案的总体分类精度和Kappa系数分别达到90.2%和0.877,明显高于其他方案。这说明特征优选方法在降低特征空间维度的同时减少了特征干扰,有效提升了分类精度。

关键词: 地膜, 面向对象, 特征优选, 随机森林

Abstract: The rapid development of agricultural productivity has been driven by the extensive use of plastic-mulched farmland (PMF).With the increase of PMF usage year by year,its influence on environment cannot be ignored.Remote sensing image classification is an important method for PMF monitoring.Based on FSO feature selection and object-oriented random forest classification method,several sets of experimental schemes were designed using the characteristics of spectrum,geometry and texture,etc.PMF extraction was carried out in the middle area of Gansu province with the sharpening image of ZY-3 and Landsat 8 OLI data.The experimental results show that the extraction accuracy of multi-feature sharpening method is higher than that of single-feature extraction method. The overall accuracy and Kappa coefficients of the experimental scheme using FSO feature optimization are 90.2% and 0.877,respectively,which is obviously higher than other schemes.It indicates that the feature interference is decreased by the feature optimization method while feature space dimension is reduced,and the classification accuracy is improved.

Key words: plastic-mulched farmland, object-oriented, feature selection, random forest

中图分类号: