光谱特征和纹理特征融合的红茶萎凋含水率检测方法  

Detection Method for Moisture Content of Black Tea Withering Based on the Fusion of Spectral and Texture Features

作　　者：沈跃铖 阮贤萍 郎春晓 冯科达 沈帅 郑航 张诚荣 蒋建东[1] 俞国红[3] SHEN Yuecheng;RUAN Xianping;LANG Chunxiao;FENG Keda;SHEN Shuai;ZHENG Hang;ZHANG Chengrong;JIANG Jiandong;YU Guohong(College of Mechanical Engineering,Zhejiang University of Technology,Hangzhou 310014,China;Shaoxing Shangyu District Agricultural Mechanization Technology Promotion Center,Shaoxing 312300,China;Zhejiang Academy of Agricultural Sciences,Hangzhou 310021,China;Zhejiang Hongming Tea Machine Complete Set Technology Co.,Shaoxing 312300,China)

机构地区：[1]浙江工业大学机械工程学院,浙江杭州310014 [2]绍兴市上虞区农业机械化技术推广中心,浙江绍兴312300 [3]浙江省农业科学院,浙江杭州310021 [4]浙江红茗茶机成套技术有限公司,浙江绍兴312300

出　　处：《茶叶通讯》2024年第3期376-382,共7页Journal of Tea Communication

基　　金：浙江省重点科技研发项目(2023C02010)。

摘　　要：为实现对红茶萎凋含水率的定量检测,研究提出了一种基于谱图特征融合的红茶萎凋含水率检测方法。通过采集不同萎凋时刻的高光谱图像,利用灰度共生矩阵法提取图像的7个纹理特征并与提取到的光谱特征相融合;通过采用PLSR、RF和GBDTR建立含水率预测模型,分析光谱特征、纹理特征以及“光谱特征+纹理特征”对含水率预测的准确性,实现红茶萎凋含水率快速无损检测。结果表明融合后的特征效果均得到较大提升,相比于其他两种模型,PLSR表现良好,在测试集上特征融合后的决定系数R2达0.7932,相较于光谱特征和纹理特征分别提升0.047和0.039。该研究对红茶加工设备的智能化具有重要意义,同时所提出的基于谱图特征融合的模型为红茶萎凋过程的智能检测提供了新思路。In order to realize the quantitative detection of the moisture content of black tea withering,a method for detecting the moisture content of black tea withering based on the fusion of spectral features.The hyperspectral images at different withering moments were collected,and seven texture features were extracted using the gray scale co-occurrence matrix method,and fused with the extracted spectral features,and the moisture content prediction model was established by adopting PLSR,RF and GBDTR to analyze the accuracy of spectral features,texture features,and"spectral features+texture features"in predicting moisture content,the fast and non-destructive detection of moisture content of black tea withering was achieved.The results show that the effect of fused features is greatly improved,and compared with the other two models,PLSR performs well,and the coefficient of determination R²of the fused features on the test set reaches 0.7932,which is 0.047 and 0.039 higher than the spectral features and texture features,respectively.This study is of great significance in the intelligentization of black tea processing equipments,and the proposed model based on spectral feature fusion provides a new idea for the intelligent detection of black tea withering process.

关 键 词：红茶 萎凋 光谱特征 纹理特征 含水率 检测方法 预测模型 

分 类 号：S571.1[农业科学—茶叶生产加工] TS272.52[农业科学—作物学] O657.3[轻工技术与工程—农产品加工及贮藏工程]