摘要
【目的】机器学习模型在农作物分类研究领域有着较高精度,但如何利用历史样本用于当前时间的作物分类是一个难点。迁移学习的核心思想在于找到已有数据与新数据之间的相似性,文章旨在探索迁移学习方法使用历史样本进行作物分类的可靠性。【方法】该文以嫩江市为研究区域,基于实地采样数据与遥感数据,用随机森林(Random Forest,RF)分类器,结合多种遥感指数,对2020—2021年嫩江市玉米与大豆种植区域进行分类;利用动态时间规整方法,以2020—2021年实地采样数据生成2022年的分类样本,用RF对2022年嫩江市的玉米与大豆种植区域进行分类。【结果】(1)对2020—2021年玉米与大豆种植区域进行分类,RF的平均总体精度达到97.8%。(2)对动态时间规整方法生成的2022年玉米与大豆种植区域进行分类,RF的总体精度达到87.5%。【结论】基于迁移学习的作物识别方法达到较高精度,具有实践意义,可提高历史时期样本的利用效率。
[Purpose]Machine learning reveals relatively high precision in crop classification,but how to use historical samples for current crop classification remains problem.The main idea of transfer learning is to find the similarity between existing data and new data,therefore this paper will explore the feasibility of transfer learning for crop classification using historic crop samples.[Method]Based on the sample dataset,Random Forest(RF)classifier was used to classify the maize and soybean planting areas in Nenjiang from 2020 to 2021,combined with a variety of remote sensing indices;Dynamic Time Warping(DTW)method was used to generate samples for 2022 based on field sample dataset from 2020 to 2021,and Random Forest classifier was used to classify the maize and soybean in Nenjiang in 2022.[Result](1)The mean overall accuracy of RF was 97.8%for the year 2020 and 2021.(2)Based on the field sample dataset in 2022 generated from 2020 and 2021 by DTW,the overall accuracy of RF reached 87.5%.[Conclusion]The method of crop classification based on transfer learning can achieve high precision and has practical significance,and can improve the utilization efficiency of historical samples.
作者
吴禹瑨
李禹萱
宋茜
任超
冷佩
Wu Yujin;Li Yuxuan;Song Qian;Ren Chao;Leng Pei(College of Geomatics and Geoinformation,Guilin University of Technology,Guilin 541006,Guangxi,China;State Key Laboratory of Efficient Utilization of Arid and Semi-arid Arable Land in Northern China/Institute of Agricultural Resources and Regional Planning,Chinese Academy of Agricultural Sciences,Beijing 100081,China)
出处
《中国农业信息》
2023年第4期1-10,共10页
China Agricultural Informatics
基金
科技创新2030——重大项目课题“多模式协同的农情反演、预测及智能计算”(2021ZD0113704)
中央级公益性科研院所基本科研业务费专项资金“东北农作物‘一张图’制图研究”(1610132021010)
关键词
迁移学习
作物分类
机器学习
遥感指数
transfer learning
crop classification
machine learning
remote sensing indices