陆江明,范婷婷,穆青爽,康志龙.融合光谱与纹理特征的龙井茶等级无损识别[J].,2021,37(3):301-307.
融合光谱与纹理特征的龙井茶等级无损识别
Nondestructive Identification of Longjing Tea Grade by Fusing Spectral and Textural Feature
投稿时间:2020-09-02  
DOI:10.13982/j.mfst.1673-9078.2021.3.0826
中文关键词:  无损识别  高光谱成像  支持向量机  光谱特征  纹理特征  数据融合  模型优化
英文关键词:nondestructive identification  hyperspectral imaging  SVM  spectral feature  texture feature  data fusion  model optimization
作者简介:陆江明(1995-),男,硕士研究生,研究方向:高光谱图像处理 通讯作者:康志龙(1971-),男,博士,副研究员,研究方向:高光谱图像和半导体光电子学
基金项目:国家自然科学基金项目(61401307);河北省高等学校科学技术研究项目(ZD2018045);天津市企业科技特派员项目(18JCTPJC57500)
作者单位
陆江明 (河北工业大学电子信息工程学院,天津 300401) 
范婷婷 (河北工业大学电子信息工程学院,天津 300401) 
穆青爽 (河北工业大学电子信息工程学院,天津 300401) 
康志龙 (河北工业大学电子信息工程学院,天津 300401) 
AuthorInstitution
LU Jiang-ming (Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, China) 
FAN Ting-ting (Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, China) 
MU Qing-shuang (Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, China) 
KANG Zhi-long (Electronics and Information Engineering, Hebei University of Technology, Tianjin 300401, China) 
摘要点击次数: 20
全文下载次数: 32
中文摘要:
      龙井茶等级快速无损识别具有重要意义。本研究以六个等级龙井茶为实验对象,应用高光谱成像技术,分别建立基于光谱特征、纹理特征及融合特征的支持向量机(Support Vector Machine,SVM)识别模型。首先采用标准正态变量变换(Standard Normal Variate,SNV)对光谱进行归一化处理,提取光谱特征,建立SVM光谱模型;然后通过T分布和随机近邻嵌入(T-Distributed Stochastic Neighbour Embedding,T-SNE)算法将高维高光谱数据映射到低维空间,选取特征图像。应用灰度共生矩阵(Gray-Level Co-Occurrence Matrix,GLCM),提取纹理特征,建立SVM图像模型;最后将光谱特征和纹理特征进行数据级融合,建立SVM混合模型。数据显示,光谱模型预测集识别率为91.11%,图像模型预测集识别率为75.42%,混合模型预测集识别率为95.14%。结果表明,与仅使用光谱或纹理信息建模相比,结合光谱和纹理特征可以提高模型识别的准确率。为进一步提高混合模型精度,引入人工蜂群(Artificial Bee Colony,ABC)算法,迭代优化SVM模型的惩罚因子 和核函数宽度 ,得到最优模型,预测集准确率可达98.61%。本研究为改进龙井茶叶快速无损评估技术提供了一种可靠的方法。
英文摘要:
      The rapid and nondestructive identification of Longjing tea grade was of great significance. In this study, support vector machine (SVM) model was respectively established by using hyperspectral imaging technology with six levels of Longjing tea, based on spectral features, texture features and fusion features. First, standard normal variable (SNV) was used to normalize the spectra, extract the spectral features, and establish the SVM spectral model. Then, the high-dimensional hyperspectral data was mapped to the low-dimensional space through the T-distributed and stochastic neighbor embedding (T-SNE) algorithm, and feature images were selected. Gray-level Co-occurrence matrix (GLCM) was applied to extract texture features and establish a SVM image model. Finally, spectral features and texture features were fused at the data level to establish a SVM mixture model. The results showed that the recognition rate of predictive sets based on spectral model was 91.11%, the recognition rate of predictive sets based on image model was 75.42% and the recognition rate of predictive sets based on mixed model was 95.14%. It illustrated that compared with modeling using only spectral or texture information, combining spectral and texture features can improve the accuracy of identification. In order to further improve the performance of the mixed model, artificial bee colony (ABC) algorithm was introduced to iteratively optimize the penalty factor C and kernel function width g of the SVM model, construct the optimal model, and the accuracy of the prediction sets can be reached 98.61%. The study provides a reliable method to improve the rapid nondestructive assessment technology of Longjing tea.
查看全文  查看/发表评论  下载PDF阅读器
关闭