便携式茶芽品质参量多光谱检测装置及系统

打开文本图片集

中图分类号：S24；TP73 文献标志码：A 文章编号：1008-0864（2025）12-0108-15

Abstract：Aimingatthe problemsof loweficiency，highcostanddiffculty inon-site implementation existing in traditional teaqualitydetectionmethods，aportablemultispectral detectiondeviceand system for teabud quality parameters was developed.The hardware of the device integrated a multispectral sensing unit with 9 characteristic wavelengths，acamera module andaglobal positioning system（GPS）module.Byadopting amaster-slave computer collaborativearchitecture，the in-situsynchronous collction and wireless transmissionof multispectral reflectance， images，and geographicallocation informationof teabudswererealized.The software system was basedonamultitask deep neural network，which estimated chlorophyls，polyphenols and freeamino acids content of tea bud using thecollected multi-bandreflectancedataanddisplayedtheresultsinrealtime.Theresultsshowedthatrelative errors ofmulti-band reflectance ofthe device’smultispectral modulewere within 4% and 8% under sunny and cloudy conditions，respectively；the averagecoefficient of determination（ Rz ）ofthe multi-band reflectance cross-calibration model was O.820，with anaverage root mean square error（RMSE）of O.O28.Basedon the corrcted reflectance data of teabuds，the constructed multi-task modelachieved high estimation accuracy for chlorophyllandfree amino acid in tea buds.The model' s R2 was O.77，and the relative percentage deviation（RPD） was greater than 2.0.While the estimation accuracy for tea polyphenolswasrelatively low，and the model’s R2 was O.51，the RMSE was 0.70% ，the RPD was 1.42.The research results provided portable and intellgent technical support for rapid and non-destructive monitoringof teabudqualityin teagardenson site，and hadpositivesignificanceforpromoting theprecise management and digital upgrading of tea production.

Keywords：portable detection device；multispectral；tea shoot quality；multi-task learning；inversion model; monitoring

中国是茶叶的起源地和最大生产国，深厚的茶文化底蕴与庞大的产业规模使其在全球茶业中占据核心地位1。（剩余21880字）