地理空间信息2025，Vol.23Issue(2)：11-14,40,5.DOI:10.3969/j.issn.1672-4623.2025.02.003
流形学习和卷积神经网络的农作物精细分类
Crops Fine Classification Based on Manifold Learning and Convolutional Neural Network
摘要
Abstract
A basic problem in precision agriculture is how to accurately extract different crops from airborne hyperspectral remote sensing imag-es of an agricultural scene.Aiming at the above problem,we proposed a hyperspectral image classification framework for crop fine extraction.Firstly,we introduced a manifold learning method named local covariance matrix,which was often applied to target detection in computer vision,for dimensionality reduction and feature extraction of agricultural hyperspectral images,combining principal component analysis.Then,we designed a lightweight convolution neural network to classify terrain objects.Taking Fanglu tea farm and Longkou farmland as examples,we carried out supervised classification experiments of small samples.Based on convolutional neural network,the front end can effectively extract advanced features of input data and continuously optimize the feature extraction performance,while the back end can effectively mine the nonlinear mapping between advanced features and classification categories and avoid overfitting.The results show that the proposed method achieved the best classification accuracy for all kinds of crops,and the average classification accuracy of crops is increased by 3.63%and 4.75%,respectively,compared with the suboptimal method.关键词
高光谱遥感/局部协方差矩阵/卷积神经网络/农作物分类/流形学习Key words
hyperspectral remote sensing/local covariance matrix/convolutional neural network/crop classification/manifold learning分类
天文与地球科学引用本文复制引用
时会省,朱文军,李贺颖..流形学习和卷积神经网络的农作物精细分类[J].地理空间信息,2025,23(2):11-14,40,5.基金项目
河南省重点研发与推广专项(科技攻关)项目(212102310414). （科技攻关）
