Spectroscopy and Spectral Analysis, Volume. 42, Issue 9, 2848(2022)
Hyperspectral Visualization of Citrus Leaf Moisture Content Based on CARS-CNN
Figures & Tables(10)
Fig. 2. Spectra after different pretreatments
(a): Raw spectra; (b): Pretreated by SNV; (c): Pretreated by MSC; (d): Pretreated by SG
Fig. 5. Regression curves of different models
(a): Raw spectra+CARS+CNN; (b): Raw spectra+CARS+PLSR; (c): SNV+PCA+RF; (d): SNV+PCA+SVR
Table 1. Distribution statistics of water content in citrus leaves(%)
View tableView in ArticleTable 1. Distribution statistics of water content in citrus leaves(%)
样本量 最小值 最大值 平均值 方差 500 19.59 83.05 62.22 0.93
Table 2. CNN model parameter setting
View tableView in ArticleTable 2. CNN model parameter setting
网络层 模型参数 输入层 全波段为256×1, CARS筛选后为29×1, PCA提取后为10×1 卷积层C1 卷积核大小为1×3×16, 步长为1 最大池化层S1 最大池化, 步长为1, 经过特征选择大小设置为1×1, 全波段大小设置为1×2 卷积层C2 卷积核大小为1×3×32, 步长为1 最大池化层S2 最大池化, 步长为1, 经过特征选择大小设置为1×1, 全波段大小设置为1×2 卷积层C3 卷积核大小为1×3×64, 步长为1 最大池化层S3 最大池化, 步长为1, 经过特征选择大小设置为1×1, 全波段大小设置为1×2 全连接层F5 32个神经元 输出层 1个神经元, 输出柑橘叶片含水量预测值
Table 3. Forecast results of different models
View tableView in ArticleTable 3. Forecast results of different models
模型 选择特征
波段数据预
处理训练集 测试集 RMSE RMSE CNN 全波段 原始数据 0.934 3 0.024 9 0.915 9 0.028 6 SNV 0.998 6 0.003 5 0.876 1 0.034 0 MSC 0.964 8 0.017 5 0.866 2 0.037 7 SG 0.245 3 0.084 3 0.202 5 0.085 8 PCA 原始数据 0.997 8 0.004 3 0.699 1 0.055 7 SNV 0.997 5 0.004 3 0.696 9 0.063 0 MSC 0.999 4 0.002 3 0.726 3 0.044 3 SG 0.999 7 0.001 4 0.169 6 0.104 7 CARS 原始数据 0.967 9 0.016 3 0.947 0 0.021 4 SNV 0.998 1 0.004 2 0.891 5 0.032 3 MSC 0.985 5 0.011 3 0.895 4 0.032 5 SG 0.238 3 0.088 1 0.172 1 0.077 7
Table 4. Comparison of prediction results of different models
View tableView in ArticleTable 4. Comparison of prediction results of different models
模型 训练集 测试集 RMSEC RMSEC 原始数据+CARS+PLSR 0.879 4 0.033 9 0.858 1 0.034 7 SNV+PCA+RF 0.947 8 0.022 3 0.746 2 0.040 9 SNV+PCA+SVR 0.643 6 0.055 3 0.612 6 0.065 7 原始数据+CARS+CNN 0.967 9 0.016 3 0.947 0 0.021 4 原始数据+全波段+CNN 0.934 3 0.024 9 0.915 9 0.028 6
Tools
Get Citation
Copy Citation Text
. Hyperspectral Visualization of Citrus Leaf Moisture Content Based on CARS-CNN[J]. Spectroscopy and Spectral Analysis, 2022, 42(9): 2848
Paper Information
Category: Research Articles
Received: Jul. 28, 2021
Accepted: Oct. 26, 2021
Published Online: Nov. 17, 2022
The Author Email:
© Copyright 2018-2021 | Chinese Laser Press.
All Rights Reserved 沪ICP备15018463号-20