基于光谱技术的穴盘苗品质分类算法研究  

Quality Classification Algorithm of Plug Seedling Based on Spectral Technology

作　　者：汤来锋 桑伢员 姬江涛 许威广 杨旺 Tang Laifeng;Sang Yayuan;Ji Jiangtao;Xu Weiguang;Yang Wang(School of Information Engineering,Xinjiang University of Science and Technology,Aksu 843000,China;College of Agricultural Equipment Engineering,Henan University of Science and Technology,Luoyang 471003,China)

机构地区：[1]新疆理工学院信息工程学院,新疆阿克苏843000 [2]河南科技大学农业装备工程学院,河南洛阳471003

出　　处：《农机化研究》2025年第8期243-250,共8页Journal of Agricultural Mechanization Research

基　　金：国家自然科学基金项目(51975186)。

摘　　要：穴盘苗经培育后难免会出现同一穴盘内不同钵苗的品质良莠不齐的情况,如果未经品质分级并将劣质苗剔除的过程,直接将钵苗机械化移栽至大田,会对后续农产品的产量和品质等产生负面影响。为此,以番茄穴盘苗为研究对象,基于光谱技术和机器学习方法,对试验样本进行快速无损检测并实现了分类识别。首先,分别采集样本的生理生化指标和光谱数据,综合品质分级的各项标准将各样本所属的品级与光谱数据对应;其次,采用多元散射校正(MSC)和竞争性自适应重加权法(CARS)对采集到的光谱数据进行预处理和降维,消除了部分冗杂数据并提升了数据的可靠性;最后,分别采用随机森林和BP神经网络算法建立了穴盘钵苗的品质分类模型,达到了通过光谱数据对穴盘苗进行品质分类的目的。通过对分类模型评价指标的对比可知:MSC+CARS+BP神经网络结合的算法所构建的分类模型识别效果最佳,其准确率达到了98.2%,调和平均数为98.5%,可为今后进行钵苗品质的快速无损检测、实现选择性移栽健壮苗提供技术参考。It is inevitable that the quality of different pot seedlings in the same pot will be uneven after the cultivation of pot seedlings.If the pot seedlings are mechanically transplanted to the field environment without the process of quality classification and the removal of inferior seedlings,it will have a negative impact on the yield and quality of subsequent agricultural products.Therefore,took the tomato plug seedling as the research object,and based on the spectral technology and machine learning method,the test samples were rapidly nondestructive testing and classification recognition was realized.Firstly,the physiological and biochemical indexes and spectral data of samples were collected respectively,and the grade of each sample was corresponding to the spectral data according to The comprehensive quality classification criteria.Secondly,Multiple Scattering Correction(MSC)and Competitive Adaptive Reweighting(CARS)were used to preprocess and reduce the dimension of the acquired spectral data,which eliminated some redundant data and improves the reliability of the data.Finally,the quality classification model of plug seedlings was established by using random forest and BP neural network algorithm respectively,and the purpose of quality classification of plug seedlings was realized by spectral data.By comparing the evaluation indexes of the classification model,it can be concluded that the classification model constructed by the algorithm combined with MSC+CARS+BP neural network had the best recognition effect,its accuracy rate was 98.2%,and the mean of reconciliation was 98.5%,which can provide technical reference for rapid nondestructive testing of pot seedling quality in the future,in order to achieve selective transplanting of robust seedlings.

关 键 词：穴盘苗 光谱检测 机器学习 随机森林 BP神经网络 

分 类 号：O433.4[机械工程—光学工程]