高级检索

    基于近红外光谱的连续小波变换与卷积注意力模块建立秦艽的定性分析模型

    Establishment of Qualitative Analysis Model of Gentiana Macrophylla by Continuous Wavelet Transform and Convolution Attention Module Based on Near Infrared Spectrometry

    • 摘要: 针对近红外光谱的处理研究大多聚焦于对原始的一维光谱信号直接进行卷积特征抽取,为了更加全面地挖掘光谱数据中的信息,提高分类模型的建模效果,提出了连续小波变换与卷积注意力模块建立秦艽定性分析模型的方法。采用连续小波变换将一维的信号转换为二维图像表现形式,以得到的小波时频图作为光谱特征,建立具有注意力机制的秦艽近红外光谱的卷积神经网络定性分析模型Att-GoogleNet,并通过翻转、对比度增强以及加入高斯噪声来扩充数据集实现数据增强,提高模型的泛化能力。结果表明:对207个秦艽样品的产地进行分析,Att-GooogleNet模型的分类准确率为99.6%,准确率、精确率、召回率、特异度、F1分数均优于传统机器学习模型。

       

      Abstract: Most of the research on near infrared spectrum processing focuses on directly extracting convolutional features from the original one-dimensional spectral signals, in order to comprehensively explore the information in the spectral data and improve the modeling effect of classification models, a method for establishment of qualitative analysis model of Gentiana macrophylla by continuous wavelet transform and convolution attention module was proposed. The continuous wavelet transform was used to convert one-dimensional signals into two-dimensional image representations, and the obtained wavelet time-frequency maps were used as spectral features. A convolutional neural network qualitative analysis model Att-GoogleNet for the near infrared spectra of Gentiana macrophylla with attention mechanism was established, and the dataset was expanded by flipping, contrast enhancement, and adding Gaussian noise to achieve data augmentation and improve the generalization ability of the model. As shown by the results, this method was analyzed the origin of 207 samples of Gentiana macrophylla, and the classification accuracy of the Att-GouogleNet model was 99.6%, the accuracy, precision, recall, specificity and F1-Score were better than those of traditional machine learning models.

       

    /

    返回文章
    返回