基于高光谱成像的烟丝中梗签分类识别研究
Research on Classification and Recognition of Stem Sticks in Shredded Cut Tobacco Based on Hyperspectral Imaging
陶发展 1杨栋 1洪伟龄 2苏子淇 3付主木 1林志平2
作者信息
- 1. 河南科技大学 信息工程学院,河南 洛阳 471023
- 2. 福建中烟工业有限责任公司,福建 厦门 361004
- 3. 中国烟草总公司郑州烟草研究院 烟草工艺重点实验室,河南 郑州 450001
- 折叠
摘要
针对烟丝中梗签分类识别检测问题,利用高光谱成像技术,结合机器学习方法对烟丝中掺杂的梗签进行分类快速识别.首先,基于短波近红外高光谱成像技术,采用标准正态变化(SNV)对烟丝和梗签光谱数据进行预处理,消除光谱散射和漫反射的影响,减少检测干扰信息.然后,利用连续投影算法(SPA)进行特征波长选择,融合极端梯度提升(XGBoost)算法,提出了 1 种基于 XGboost算法的烟丝中梗签分类模型.最后,采用Mean-shift均值漂移算法和形态学梯度算法相结合的方式作为后处理方法,对模型的分类结果进行后处理,实现烟丝中梗签的快速智能检测.结果表明:建立的 SNV-SPA-XGBoost分类模型,训练集和测试集准确率分别达到 100%和 99.32%,经后处理后对 A1(1.0～1.5 cm)、A2(0.5～1.0 cm)、A3(<0.5 cm)梗签检测准确率分别达 100%、95.50%和 86%.
Abstract
Regarding the classification,recognition,and detection of stem sticks in cut tobacco leaves,this article uses hyperspectral imaging technology and combines machine learning methods to classify and rapidly identify adulterated stem sticks in cut tobacco leaves.Firstly,based on shortwave near-infrared hyperspectral imaging technology,standard normal variate(SNV)transformation is applied to preprocess the spectral data of both cut tobacco leaves and stem sticks.The goal is to eliminate the effects of spectral scattering and reflectance,reducing various sources of interference.Subsequently,a feature wavelength selection is conducted using the successive projections algorithm(SPA),integrate the extreme gradient boosting(XGBoost)algorithm,proposed a stem stick classification model in shredded cut tobacco based on the XGBoost algorithm.Finally,a post-processing method is employed to achieve intelligent detection of stem sticks within cut tobacco leaves,combining the mean-shift mean-shift algorithm and morphological gradient algorithm.The classification results of the model are post-processed using this approach.The results demonstrate that the established SNV-SPA-XGBoost classification model achieves accuracy rates of 100%for the training set and 99.32%for the test set.After post-processing,the accuracy rates for detecting A1(1.0～1.5 cm),A2(0.5～1.0 cm),and A3(<0.5 cm)stem sticks reach 100%,95.50%,and 86%respectively.
关键词
高光谱成像/机器学习/梗签/连续投影算法/XGBoost/分类Key words
hyperspectral imaging/machine learning/stem sticks/successive projections algorithm/XGBoost/classification引用本文复制引用
基金项目
国家自然科学基金项目(62371182)
河南省高校科技创新人才计划项目(23HASTIT021)
河南省科技研发计划联合基金(225200810007)
河南省科技研发计划联合基金(222103810036)
中国烟草总公司科技项目(110202202010)
出版年
2024
国家科技期刊平台
NSTL
万方数据