Xinjiang Agricultural Sciences ›› 2020, Vol. 57 ›› Issue (5): 932-939.DOI: 10.6048/j.issn.1001-4330.2020.05.018

Previous Articles     Next Articles

Research on Object Recognition Based on UAV Multispectral Image

WEI Qing, ZHANG Baozhong, WEI Zheng   

  1. State Key Laboratory of Simulation and Regulation of Water Cycle in River Basin, China Institute of Water Resources and Hydropower Research/National Center for Efficient Irrigation Engineering and Technology Research, Beijing 100038, China
  • Received:2019-12-20 Published:2020-04-23
  • Correspondence author: ZHANG Baozhong(1981-),male,Taiyuan City,Shanxi Province, professorate senior engineer, Research areas include principle and technology of water-saving irrigation,(E-mail)zhangbaozhong333@163.com
  • Supported by:
    The National Key R & D Program Project (2017YFC0403202); Excellent Youth Science Foundation (51822907); Fundamental Research Business Expenses of China Institute of Water Resources and Hydropower Research (01881910)

基于无人机多光谱影像的地物识别

魏青, 张宝忠, 魏征   

  1. 中国水利水电科学研究院流域水循环模拟与调控国家重点实验室国家节水灌溉北京工程技术研究中心,北京 100038
  • 通讯作者: 张宝忠(1981-),男,山西太原人,教授级高工,研究方向为节水灌溉理论与技术,(E-mail)zhangbaozhong333@163.com
  • 作者简介:魏青(1995-),女,河南商丘人,硕士研究生,研究方向为节水灌溉理论与技术,(E-mail)weiqing1221@126.com
  • 基金资助:
    国家重点研发计划(2017YFC0403202);优秀青年科学基金(51822907);中国水利水电科学研究院基本科研业务费专项(01881910)

Abstract: 【Objective】 In view of the lack of timeliness of farmland information acquisition and the difficulty of grasping basic farmland information in time, in this project, the UAV multi-spectral images acquired in May and June 2018 were used to study the extraction of some farmland types in Daxing experimental base in Beijing. 【Method】Firstly, the species of interest were identified, and the temporal and spectral characteristics of the image were analyzed. Then, the normalized vegetation index NDVI, normalized green-blue difference index NGBDI, modified ratio vegetation index MSR and red-band reflectance were determined as the optimal classification features, and the image was segmented by threshold based on spectral variables. The decision tree classification method based on visual interpretation was used to realize the classification of land features and extract the planting area. The method was validated by selecting the ground survey data based on visual interpretation. 【Result】 The results showed that the decision tree classification method based on temporal and spectral characteristics had good effect and the method was applicable to extracting wheat, fruit trees and big shed with errors of 10.68%, 6.06% and 16.48%, respectively. Besides, the area extraction error was within 17%, so we can safely say that UAV multi-spectral remote sensing image has certain applicability for ground object recognition. 【Conclusion】 The advantages of UAV in low cost and high efficiency provide reference for timely access to farmland information.

Key words: UAV; multi-spectral remote sensing; temporal and spectral features; terrain recognition; decision tree classification

摘要: 【目的】利用2018年5和6月获取的无人机多光谱影像对北京市大兴试验基地的部分农田进行地物类型提取研究。【方法】确定感兴趣地物种类,对影像进行时相与光谱特征分析,然后确定归一化植被指数NDVI、归一化绿蓝差异指数NGBDI、修正型比值植被指数MSR和红边波段反射率可以作为最优分类特征,通过基于光谱变量阈值分割的决策树分类法,实现地物分类,并提取种植面积,选取基于目视解译的地面调查数据进行方法验证。【结果】基于时相与光谱特征的决策树分类方法有较好效果,该方法用于小麦、果树和大棚的提取,误差值分别为10.68%、6.06%和16.48%,面积提取误差在17%以内,对无人机多光谱遥感影像进行地物识别具有一定的适用性。【结论】无人机低成本、高效率的优势为农田信息及时获取提供参考。

关键词: 无人机, 多光谱遥感, 时相与光谱特征, 地物识别, 决策树分类法

CLC Number: