| 邱薇纶1，周燕舞2，石孟良2.基于数据融合策略植物油光谱模式的识别[J].中国油脂,2023,48(7):.[QIU Weilun1, ZHOU Yanwu2, SHI Mengliang2.Spectral pattern recognition of vegetable oils based on data fusion strategy[J].China Oils and Fats,2023,48(7):.] |
| 基于数据融合策略植物油光谱模式的识别 |
| Spectral pattern recognition of vegetable oils based on data fusion strategy |
| |
| DOI： |
| 中文关键词: 植物油 光谱 数据融合 特征提取 识别 |
| 英文关键词:vegetable oil spectral data fusion feature extraction recognition |
| 基金项目:湖南省自然科学基金面上项目（2023JJ30221） |
|
| 摘要点击次数: 1349 |
| 全文下载次数: 1349 |
| 中文摘要: |
| 为实现对不同植物油的快速无损分类识别，探究数据融合技术在提升光谱分类模型精度方面的可行性与应用价值，借助衰减全反射-傅里叶变换红外光谱技术、表面增强拉曼光谱技术结合多源数据融合技术，开展了对7种共计180份植物油样本的分类识别。基于单一光谱模型、数据层融合模型和特征层融合模型，比较了Bayes判别分析（BDA）和多层感知器神经网络（MLP）两种化学计量学方法在区分各样本时的差异，同时考察了主成分分析、广义最小平方、最大似然、主轴因式分解4种算法在特征提取方面的差异。结果表明，光谱数据融合在识别植物油方面具有显著的优势，BDA模型对各样本的区分能力强于MLP模型，相较于其他3种算法，主成分分析在油样特征提取方面展现了较为理想的结果。基于PCA特征提取的特征层融合BDA模型为最佳识别模型，以此实现了180份植物油样本100%的准确区分，同时对5种品牌花生油达到了100%的准确区分，实现了对各样本“种类-品牌”的两级识别分类工作。 |
| 英文摘要: |
| In order to achieve the rapid and non-destructive recognition of different vegetable oils, and explore the feasibility and application value of data fusion technology in improving the classification accuracy of spectral models, a total of 180 vegetable oil samples from 7 kinds were recognized and classified by attenuated total reflectance-Fourier transform infrared spectroscopy, surface-enhanced Raman spectroscopy and multi-source data fusion. The differences of Bayes discriminant analysis （BDA）and multilayer perceptron neural network（MLP） in classifying all samples were compared and discussed based on single model, data layer fusion model and feature layer fusion model. Besides, the differences of principal component analysis, generalized least square, maximum likelihood and principal axis factorization in feature extraction were investigated. The results showed that the spectral data fusion had significant advantages in recognizing vegetable oils. The ability of BDA model to distinguish each sample was more predominant than that of MLP. Compared with another three algorithms, principal component analysis showed the more positive results in extracting features. The BDA model based on feature layer fusion from PCA feature extraction was considered as the optimal model,and it could achieve 100% accurate differentiation of 180 samples and 100% accurate differentiation of 5 brands of peanut oil, realizing the two-level recognition and classification of samples from kind to brand. |
| 查看全文 查看/发表评论 下载PDF阅读器 |
| 关闭 |
