分波段Transformer特征提取在近红外光谱数据分类中的应用  

Application of Segmented Transformer Feature Extraction in Near Infrared Spectral Data Classification

作　　者：李永生[1] 郝贤伟[1] 向澍 时艺丹 厉小润[2] Li Yongsheng;Hao Xianwei;Xiang Shu;Shi Yidan;Li Xiaorun(China Tobacco Zhejiang Industrial Co.,Ltd.,Hangzhou 310008,Zhejiang,China;College of Electrical Engineering,Zhejiang University,Hangzhou 310027,Zhejiang,China;Ocean College,Zhejiang University,Zhoushan 316000,Zhejiang,China)

机构地区：[1]浙江中烟工业有限责任公司,浙江杭州310008 [2]浙江大学电气工程学院,浙江杭州310027 [3]浙江大学海洋学院,浙江舟山316000

出　　处：《激光与光电子学进展》2024年第13期444-451,共8页Laser & Optoelectronics Progress

基　　金：国家自然科学基金项目(62171404);浙江大学—浙江中烟联合实验室项目(KYY5100120001)。

摘　　要：针对近红外光谱维度高、受噪声扰动大,以及不同省份样本间光谱相似性高的特点,提出一种分波段Transformer特征提取网络,并将其应用于烟叶产地识别,以提高分类准确性。首先,根据近红外光谱的一维结构特点,设计了嵌入层,利用一维卷积实现特征的压缩。其次,通过滑动窗将数据按照波段维度分成3部分,并改进了Transformer结构进行光谱特征的提取,避免波段数过多导致计算效率低下。最后,为适应光谱数据的特点,设计了多层一维卷积的回归头,用于样本产地预测。为验证所提算法的有效性,以分类准确率、精确率、召回率等指标与其他算法进行性能比较,进行了多个对比实验,实验结果验证了模型中各个结构的优越性。该模型能够有效地利用光谱结构,实现特征提取和噪声压制,成功完成了烟叶产地识别任务。A segmented Transformer feature extraction network is proposed to address the challenges posed by near infrared spectroscopy,including high dimensionality,susceptibility to noise interference,and high spectral similarity among samples from different provinces.This network is applied to tobacco leaf origin identification to enhance classification accuracy.First,based on the onedimensional structure of near infrared spectroscopy,an embedding layer is designed to compress features using onedimensional convolution.Second,the data is divided into three parts along the spectral dimension using sliding windows.The Transformer architecture is improved to extract spectral features,addressing the issue of computational inefficiency caused by a large number of spectral bands.Last,to adapt to the characteristics of spectral data,a regression head with multiple layers of onedimensional convolution is designed to predict the origin of the samples.To validate the effectiveness of the proposed algorithm,several comparative experiments are conducted,comparing classification accuracy,precision,recall,and other metrics with other algorithms.The superiority of each structure in the model is verified.The experimental results demonstrate that the proposed model effectively utilizes the spectral structure for feature extraction and noise suppression,successfully accomplishing the task of tobacco leaf origin identification.

关 键 词：近红外光谱 近红外光谱数据分类 特征提取 一维卷积 TRANSFORMER 

分 类 号：O657.3[理学—分析化学]