查询结果:   史杨,王儒敬,汪玉冰.基于卷积神经网络和近红外光谱的土壤有机碳预测模型[J].计算机应用与软件,2018,35(10):147 - 152,266.
中文标题
基于卷积神经网络和近红外光谱的土壤有机碳预测模型
发表栏目
人工智能与识别
摘要点击数
613
英文标题
SOIL ORGANIC CARBON PREDICTION BASED ON CONVOLUTIONAL NEURAL NETWORKS AND NEAR INFRARED SPECTROSCOPY
作 者
史杨 王儒敬 汪玉冰 Shi Yang Wang Rujing Wang Yubing
作者单位
中国科学院合肥智能机械研究所 安徽 合肥 230031 中国科学技术大学自动化系 安徽 合肥 230027    
英文单位
Institute of Intelligent Machines, Chinese Academy of Sciences, Hefei 230031, Anhui, China Department of Automation, University of Science and Technology of China, Hefei 230027, Anhui, China    
关键词
近红外光谱 卷积神经网络 土壤有机碳 建模方法
Keywords
NIR spectroscopy Convolutional neural network Soil organic carbon Modeling approach
基金项目
国家自然科学基金项目(31671586)
作者资料
史杨,博士生,主研领域:模式识别,农业智能系统。王儒敬,研究员。汪玉冰,副研究员。 。
文章摘要
利用土壤近红外光谱间接预测土壤成分与传统的实验室分析方法相比,具有节约时间、成本低等优点,而建立准确的预测模型至关重要。提出将深度卷积神经网络模型用于大尺度范围的土壤有机碳含量的预测中。设计五种不同深度的卷积神经网络模型,使用覆盖欧洲23国的、土壤种类多样的数据集对模型进行训练。将其预测结果与传统的主成分回归、支持向量回归等线性建模方法进行对比。实验表明,使用包含6~7个卷积层的卷积神经网络预测有机碳含量的均方根误差可以达到9.69 g/kg,比其他线性建模方法预测大尺度土壤有机碳更准确。
Abstract
Compared with traditional laboratory analysis method, the method of indirect prediction of soil composition by near infrared spectroscopy has the advantages of time saving and low cost. It is essential to establish the accurate prediction model. We applied deep convolutional neural network model to predict soil organic carbon content in large scale. We designed five convolutional neural network models with different depth. A dataset covering 23 European countries and various soil types was adopted to train these models. The prediction results were compared with the traditional linear modeling methods such as principal component regression and support vector regression. The experimental results show that the root mean square error of prediction of organic carbon content can reach 9.69 g/kg by using convolutional neural network with 6~7 convolutional layers, which is more accurate than that of linear modeling methods in predicting soil organic carbon in large scale.
下载PDF全文