草地学报

草地学报 ›› 2024, Vol. 32 ›› Issue (10): 3262-3271.DOI: 10.11733/j.issn.1007-0435.2024.10.027

• 研究论文 • 上一篇    

基于无人机多光谱与纹理特征的饲用大豆地上生物量估算研究

李贵鑫1,2, 安东1,2, 于应文1,2, 沈禹颖1,2,3   

  1. 1. 兰州大学草种创新与草地农业生态系统全国重点实验室, 甘肃 兰州 730020;
    2. 兰州大学草地农业科技学院, 甘肃 兰州 730020;
    3. 兰州大学甘肃庆阳草地农业生态系统国家野外科学观测研究站, 甘肃 庆阳 745000
  • 收稿日期:2024-01-27 修回日期:2024-03-24 发布日期:2024-11-04
  • 通讯作者: 沈禹颖,E-mail:yy.shen@lzu.edu.cn
  • 作者简介:李贵鑫(1998-)，男，汉族，四川绵阳人，硕士研究生，主要从事饲草生产学研究，E-mail:ligx21@lzu.edu.cn
  • 基金资助:
    国家重点研发计划(2022YFD1300803);国家牧草产业技术体系(CARS-34-13)资助

Estimation of Above-Ground Biomass of Forage Soybean Based on UAV Multispectral and Texture Features

LI Gui-xin1,2, AN Dong1,2, YU Ying-wen1,2, SHEN Yu-ying1,2,3   

  1. 1. State Key Laboratory of Herbage Improvement and Grassland Agro-Ecosystems, Lanzhou University, Lanzhou, Gansu Province 730020, China;
    2. College of Pastoral Agriculture Science and Technology, Lanzhou University, Lanzhou, Gansu Province 730020, China;
    3. National Field Scientific Observation and Research Station of Grassland Agro-Ecosystems, Qingyang, Gansu Province 745000, China
  • Received:2024-01-27 Revised:2024-03-24 Published:2024-11-04

PDF

摘要： 本研究利用无人机获取了饲用大豆[Glycine max (L.) Merr.]主要生育时期的多光谱遥感影像，并基于多光谱影像提取的40个纹理特征和光谱反射率构建的12个植被指数，通过多元线性回归(Multiple linear regression,MLR)、人工神经网络(Artificial neural networks,ANN)、随机森林(Random forest,RF)、支持向量机(Support vector regression,SVR)等算法，对饲用大豆营养体生长过程的地上生物量进行了估测。结果表明：450 nm，560 nm，650 nm和730 nm这4个波段的光谱反射率随饲用大豆生育进程而变化，具体表现为分枝前期下降、分枝后期上升至峰值、开花期下降的变化趋势。840 nm波段的光谱反射率表现为分枝期上升并在后期达到峰值、开花期下降的变化趋势。方差(Variance)、对比度(Contrast)、相异性(Dissimilarity)和信息熵(Entropy)等纹理特征在各波段下数值变化基本趋于一致，表现出红边波段最大，红光波段最小。综合来看，基于植被指数与纹理特征作为输入参数的ANN模型对于饲用大豆各生育时期地上部生物量的估测效果最好(R2= 0.71，RMSE = 1.81 t·hm-2)。研究结果可为饲用大豆地上生物量快速精准估测以及高效栽培管理提供技术支撑。

关键词: 无人机多光谱影像, 饲用大豆, 地上生物量, 机器学习, 纹理特征

Abstract: In this study,drones were employed to gather multispectral remote sensing images of the main nutritional growth stages of forage soybeans. Utilizing 40 texture features extracted from the multispectral images and 12 vegetation indices constructed from spectral reflectance,the above-ground biomass of forage soybeans at different growth stages was estimated using algorithms and models such as Multiple Linear Regression (MLR),Artificial Neural Networks (ANN),Random Gorest (RF),and Support Vector Regression (SVR). The findings indicated that the spectral reflectance of the four bands,450 nm,560 nm,650 nm,and 730 nm,changed with the development process of the forage soybeans,displaying a pattern of decrease in the early branching stage,increase to a peak in the late branching stage,and decrease during the flowering stage. The spectral reflectance at the 840 nm band exhibited an increasing trend during the branching stage,reaching a peak in the late branching stage,followed by a decrease during the flowering stage. The dynamic changes in the texture features of Variance,Contrast,Dissimilarity,and Entropy were generally consistent under all the sequences,displaying the biggest value in red-edge band and the smallest value in red band. Overall,the ANN model,using vegetation indices and texture features as input parameters,achieved the best results in estimating the above-ground biomass of forage soybeans across different growth stages (R2= 0.71;RMSE = 1.81 t·hm-2). This research provided technical support for the rapid and accurate estimation of the above-ground biomass of forage soybeans and for efficient cultivation management.

Key words: UAV multispectral images, Forage soybean, Above-ground biomass, Machine learning, Texture features

中图分类号: