草地学报 ›› 2020, Vol. 28 ›› Issue (5): 1427-1435.DOI: 10.11733/j.issn.1007-0435.2020.05.030

• 技术研发 • 上一篇    下一篇

BP神经网络在环青海湖地区天然草地估产研究中的应用

于璐1,2,3, 王迅1,2,3, 柴沙驼1,2,3, 刘书杰1,2,3   

  1. 1. 青海大学畜牧兽医科学院, 青海 西宁 810016;
    2. 青海省高原放牧家畜动物营养与饲料科学重点实验室, 青海 西宁 810016;
    3. 青海省牦牛工程技术研究中心, 青海 西宁 810016
  • 收稿日期:2020-03-18 修回日期:2020-06-04 发布日期:2020-09-19
  • 通讯作者: 王迅
  • 作者简介:于璐(1994-),女,黑龙江哈尔滨人,硕士研究生,主要从事草地遥感研究,E-mail:1131892090@qq.com
  • 基金资助:
    国家重点研发计划课题(2018YFD0502301);国家自然科学基金项目(41461081,31660673)资助

Application of BP Neural Network in Natural Grassland Yield Estimation in the Area Around Qinghai Lake

YU Lu1,2,3, WANG Xun1,2,3, CHAI Sha-tuo1,2,3, LIU Shu-jie1,2,3   

  1. 1. Qinghai Academy of Animal Husbandry and Veterinary Sciences in Qinghai University, Qinghai University, Xining, Qinghai Province 810016, China;
    2. Key Laboratory of Plateau Grazing Animal Nutrition and Feed Science of Qinghai Province, Qinghai University, Xining 810016, Qinghai Province, China;
    3. Yak Engineering Technology Research Center of Qinghai Province, Xining, Qinghai Province 810016, China
  • Received:2020-03-18 Revised:2020-06-04 Published:2020-09-19

摘要: 为构建一种对不同草地类型与时期的天然草场高精度产草量估测模型,快速获取环青海湖区域月际产草量数据、减少数据处理和反复建模过程中的工作量,本试验基于高分卫星影像数据,以草地类型、植被指数及实测草场产草量为训练样本,构建了环青海湖地区天然牧草产草量神经网络预估模型。结果显示:以2-6-1为构架的人工神经网络模型,目标误差、学习次数设定为0.003,800时,产草量模型预测值与实测值呈高度相关(R2=0.743,RMSE=58.531 g·m-2),达到实际估产需求,证明本文人工神经网络模型对环青海湖区域天然牧草产草量估测的可行性与适用性。

关键词: 天然草地, 产草量, 环青海湖, BP-ANN模型, 植被指数, 高分卫星遥感

Abstract: This paper aims to build a high-precision estimation model for natural grassland with different grassland types and periods,quickly obtain the monthly grass yield data around Qinghai Lake,and reduce the workload in data processing and repeated modeling. Based on high-resolution satellite image data and taking grassland type,vegetation index and measured pasture yield as training samples,this experiment constructed a neural network prediction model of natural forage yield in the area around Qinghai Lake. Results showed that when the artificial neural network model of the architecture was 2-6-1,the target error and the number of learning set was 0.003 and 800 respectively,the predicated yield was highly correlated to the measured values (R2=0.743,RMSE=58.531 g·m-2),which achieved actual yield estimation demand. The artificial neural network model provided in this study is applicable to estimate the natural grass yield of Qinghai lake area.

Key words: Natural grassland, Grass yield, Qinghai Lake, BP-ANN model, Vegetation index, High-resolution satellite remote sensing

中图分类号: