摘要
本文通过动态采集小麦的可见/近红外漫反射光谱(600~1 600 nm),结合偏最小二乘法(PLS)和BP神经网络(BP-ANN)建模方法建立小麦的蛋白质、水分、湿面筋、硬度指数的定量预测模型,利用决定系数(R^(2))、均方根误差(RMSE)和验证集标准偏差与预测标准偏差的比值(RPD)作为评价指标进行验证,比较分析了3种不同预处理方式对建立的PLSR模型的影响,同时分析比较了4种不同训练算法对建立的BP-ANN模型的影响。2种不同的建模方法,虽有不同但都取得了较好的预测结果。试验表明,利用不同的预处理方法可以有效提高实验结果。利用深度学习的方法对小麦品质建模,在水分和硬度指数这2个品质参数上有效地提高了预测精度。结果表明利用动态光谱建立小麦品质模型是切实可行的,对近红外在线检测技术有一定的参考价值。
The dynamic visible/near infrared spectra of wheat were obtained by diffuse reflectance method(600~1 600 nm).The regression models of protein, moisture, wet gluten and hardness index of wheat were established by partial least squares(PLS) and BP neural network(BP-ANN). The coefficient of determination(R^(2)), root mean square error of cross validation and Ratio of Standard Deviation of the Validation set to Standard Error of Prediction(RPD) was used as evaluation indexes to verify the PLS model. At the same time, the influence of four different training algorithms on the BP-ANN model was analyzed and compared. Although the two different modeling methods were different, they have achieved good prediction results. The results indicated that different pretreatment methods could effectively improve the test results, and the deep learning method could effectively improve the prediction accuracy of water content and hardness index. It was proved that the dynamic spectrum model was feasible, and it had a certain reference value for near infrared on-line detection technology.
作者
周星宇
姜洪喆
蒋雪松
沈飞
何学明
张祎
莫晓嵩
Zhou Xingyu;Jiang Hongzhe;Jiang Xuesong;Shen Fei;He Xueming;Zhang Yi;Mo Xiaosong(College of Mechanical and Electronic Engineering,Nanjing Forestry University,Nanjing 210037;School of Food Science and Engineering,Nanjing University of Finance and Economics,Nanjing 210037;Jiangsu Grain and Oil Quality Monitoring Center,Nanjing 210037)
出处
《中国粮油学报》
CAS
CSCD
北大核心
2022年第3期157-162,共6页
Journal of the Chinese Cereals and Oils Association
基金
国家自然科学基金面上项目(31772061)
江苏省现代农机装备与技术示范推广项目(NJ2020-20)。
关键词
近红外
小麦
蛋白质
near infrared
wheat
protein
