| 高光谱成像的非烟物质分类识别研究 |
| |
| 引用本文： | 李智慧,梅吉帆,李辉,李嘉康,卢敏瑞,王芳,张腾健,堵劲松,洪伟龄,徐大勇.高光谱成像的非烟物质分类识别研究[J].中国烟草学报,2022,28(3):81-88. |
| |
| 作者姓名： | 李智慧 梅吉帆 李辉 李嘉康 卢敏瑞 王芳 张腾健 堵劲松 洪伟龄 徐大勇 |
| |
| 作者单位： | 1.中国烟草总公司郑州烟草研究院，河南郑州市高新区枫杨街2号 450001 |
| |
| 基金项目： | 福建中烟工业有限责任公司科技项目“卷烟产品及原料高光谱特征分析与应用技术研究”D2020248 |
| |
| 摘 要： | 目的 利用高光谱成像技术和机器学习方法对烟叶中的非烟物质进行分类识别。 方法 使用可见—近红外高光谱成像技术，采用归一化（Normalization）、标准正态变化（SNV）、多元散射校正（MSC）、一阶导数（FD）、卷积平滑（SG）对光谱数据进行预处理，通过连续投影变换（SPA）和主成分载荷（PCA loadings）进行特征波长选择，并应用随机森林（RF）、Softmax和支持向量机（SVM）建立分类模型。 结果 SNV为最佳光谱预处理方法，SPA选择特征波长建立的SVM模型为最优模型，训练集和测试集正确率分别为99.82%和99.47%。 结论 高光谱成像技术结合SPA-SVM模型可以有效分类识别烟叶中的非烟物质。
|
| 关 键 词： | 高光谱成像 非烟物质 连续投影算法 特征波长 支持向量机 分类 |
| 收稿时间： | 2021-08-19 |
Research on classification and recognition of non-tobacco related material (NTRM) based on hyperspectral imaging technology |
| |
| Affiliation: | 1.Zhengzhou Tobacco Research Institute of CNTC, Zhengzhou 450001, China2.Fujian Wuyi Leaf Tobacco Co., Ltd., Nanping 354000, China3.Technology Center, China Tobacco Fujian Industrial Co., Ltd., Xiamen 361021, China |
| |
| Abstract: | The NTRM in tobacco leaves were recognized and classified using hyperspectral imaging technology and machine learning. Hyperspectral imaging system was used to collect spectral data of 400~1000 nm band tobacco leaves and NTRM. Five preprocessing methods were used to preprocess the original spectra, and standard normal variate (SNV) was selected as the best preprocessing method. Successive projection algorithm (SPA) and Principal component analysis loadings (PCA loadings) were used to screen out 6 characteristic wavelengths. Random forest (RF), Softmax and support vector machine (SVM) were employed to establish identification models based on characteristic wavelength and full spectrum. The results showed that SVM model of the full spectrum had the best recognition results, and the recognition accuracy of samples in the calibration set and test set were 100% and 99.6%, respectively. SPA method was superior to PCA loadings algorithm, and identification rates of SPA-SVM model calibration set and test set were 99.82% and 99.47% respectively. Hyperspectral imaging combined with SPA-SVM model demonstrate the efficient classification and recognition of NTRM in tobacco leaves. |
| |
| 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 |
|
