ScatteredLiquidComponentAnalysisApproach

此仓库为论文 T. Liu, C. Zhou, C. Fang, H. Zhu, Y. Li and J. Wu, "A Scattered Liquid Component Analysis Approach Based on Spectral Visual Encoding and Fusion," in IEEE Sensors Journal, doi: 10.1109/JSEN.2023.3336797. 源码

项目背景

主要提出了一种从多谱线特征层融合角度提出了基于格拉姆角场（Gramian angular field）与多尺度卷积神经网络（multi-scale convolutional neural network）的光谱特征提取方法，使用格拉姆角场将一维光谱信号变换为二维图像增强了信号特征，使用多尺度卷积神经网络提取图像特征，利用特征向量回归计算金属离子浓度。

A spectral feature extraction method based on the Gramian Angular Field (GAF) and a multi-scale convolutional neural network (CNN) is primarily proposed from the perspective of fusing features across multiple spectral lines. The Gramian Angular Field is employed to transform one-dimensional spectral signals into two-dimensional images, enhancing signal characteristics. The multi-scale convolutional neural network is then used to extract features from the images, and feature vectors are utilized for the regression calculation of metal ion concentrations.

配合./doc中的中文论文原稿和T. Liu, C. Zhou, C. Fang, H. Zhu, Y. Li and J. Wu, "A Scattered Liquid Component Analysis Approach Based on Spectral Visual Encoding and Fusion," in IEEE Sensors Journal, doi: 10.1109/JSEN.2023.3336797. 正式发表论文参考

目录结构

./data

• 此目录主要存放原始数据和各种处理后的数据
• 以及各种处理脚本
• process_2d_new.py 为批量将目标库中的csv转为格拉姆角场（CNN训练中只使用目标库数据） ref_2d_process.py 处理参考库中的csv转为格拉姆（对比测试用，实际训练中未使用）

./cnn

• 此目录存放神经网络代码

./gaf_show

• 此目录存放格拉姆角场可视化文件

./pretreatment

• 此目录下为光谱矫正和原始光谱误差等预处理文件