| 基于图像处理的烟叶烘烤阶段判别模型优选 |
| |
| 引用本文： | 李增盛,孟令峰,王松峰,高峻,徐小洪,朱先洲,杨超,汪伯军,王爱华,孟霖,刘自畅,杜海娜,刘浩,孙福山.基于图像处理的烟叶烘烤阶段判别模型优选[J].中国烟草学报,2022,28(2):65-76. |
| |
| 作者姓名： | 李增盛 孟令峰 王松峰 高峻 徐小洪 朱先洲 杨超 汪伯军 王爱华 孟霖 刘自畅 杜海娜 刘浩 孙福山 |
| |
| 作者单位： | 1.中国农业科学院烟草研究所，农业部烟草生物学与加工重点实验室，青岛 266101 |
| |
| 基金项目： | 中国农业科学院科技创新工程ASTIP-TRIC03中国烟草总公司重点项目110202102007中国烟草总公司四川省公司科技重点项目SCYC202012中国烟草总公司重庆市公司科技项目B20202NY1335 |
| |
| 摘 要： | 背景和目的 烟叶烘烤阶段的自动判别是建立智能化烟叶烘烤系统的重要环节。为实现烘烤阶段的精确识别和操控，提升烟叶烘烤的精准度。 方法 提取烘烤过程中整夹烟叶图像的11种颜色特征和8种纹理特征，分别对颜色特征和纹理特征进行变量聚类分析，以10为距离，将提取的颜色特征和纹理特征各分为2类。利用相关性分析筛选出每类特征中与烘烤阶段相关性最强的1个特征组成特征子集（R/G、l*、灰度平均和惯性），作为模型输入，分别利用基于遗传算法的支持向量机（GA-SVM）、基于粒子群算法的反向传播（PSO-BP）神经网络和极限学习机（ELM）进行烟叶烘烤阶段的分类识别研究。 结果 以优选后4个图像特征作为模型输入时，所建立的GA-SVM模型的测试集判别准确率为93.27%，PSO-BP神经网络模型的测试集判别准确率为89.35%，ELM模型的测试集判别准确率为85.05%。 结论 基于遗传算法的SVM模型烘烤阶段识别效果优于基于粒子群算法的BP神经网络模型，基于粒子群算法的BP神经网络模型识别效果优于ELM模型。
|
| 关 键 词： | 烤烟 烘烤阶段 图像处理 特征模型 智能烘烤 |
| 收稿时间： | 2021-09-14 |
Selection of optimum discriminant model in tobacco curing stage based on image processing |
| |
| Affiliation: | 1.Institute of Tobacco Research of CAAS, Key Laboratory of Tobacco Biology and Processing, Ministry of Agriculture, Qingdao 266101, China2.Graduate School of Chinese Academy of Agricultural Sciences, Beijing 100081, China3.Liangshan Tobacco Company of Sichuan Province, Xichang, Sichuan 615000, China4.Chongqing Tobacco Science Research Institute, Chongqing 400715, China |
| |
| Abstract: | Background The automatic discrimination in the tobacco curing stage is an important link for the establishment of an intelligent tobacco leaf curing system. This study aims to realize the accurate identification and control in the curing stage and improve the accuracy of tobacco curing. Methods In this study, 11 color features and 8 texture features of the complete tobacco leaf images during the curing process were extracted, then variable cluster analysis of the color features and texture features was carried out, and finally the extracted color feature values and texture features were divided into two categories with a distance of 10. Correlation analysis was performed to filter out one feature the strongest correlation with the curing stage from each categorize of feature to form a feature subset (R/G, l*, gray average and inertia), which was used as model input. Then support vector machine based on genetic algorithm (GA-SVM), the particle swarm algorithm back propagation (PSO-BP) neural network and the extreme learning machine (ELM) were used for classification and recognition in the tobacco curing stage. Results By using the selected four image features as model inputs, the test set discrimination accuracy rates of the established GA-SVM model, the PSO-BP neural network model and the ELM modelwere 93.27%, 89.35%, and 85.05%, respectively. Conclusion The SVM model based on the genetic algorithm in the curing stage has the best recognition, followed by the BP neural network model based on the particle swarm algorithm, and the ELM model ranking last. |
| |
| Keywords: | |
|
| 点击此处可从《中国烟草学报》浏览原始摘要信息 |
|
点击此处可从《中国烟草学报》下载全文 |
|
