改进时间卷积网络的红壤有机质高光谱预测模型

doi:10.3964/j.issn.1000-0593(2023)09-2942-10

摘要
参考文献
相关文章 (15)

全文: PDF (13857 KB)
输出: BibTeX | EndNote (RIS)

摘要：针对现有卷积神经网络土壤有机质（SOM）预测模型用在小样本数据集存在建模效果差，预测精度不够高等问题，为更加精准预测土壤SOM含量，以广西国有黄冕林场和国有雅长林场采集的206个土壤样品为研究对象，提出了一种改进时间卷积网络（SATCN）的红壤有机质高光谱预测模型。对土壤样品进行Savitaky-Golay（SG）平滑以及一阶微分（1DR）、二阶微分（2DR）、标准正态变量（SNV）和多元散射校正（MSC）四种变换，对比分析长短记忆网络（LSTM）、偏最小二乘回归（PLSR）和支持向量机（SVM）在不同光谱预处理下的建模效果，结果表明，采用SG处理后的光谱一阶微分预处理方法，建模效果最好；在时间卷积网络（TCN）架构上，采用浅层网络结构，在TCN残差结构中加入自注意力层，提高模型特征学习能力；每个卷积核权重加入L2正则化，防止过拟合；选取一阶微分作为光谱预处理，建立ResNet-13、VGGNet-7、时间卷积网络（TCN）和改进时间卷积网络（SATCN）四种模型，对比分析四种模型建模效果，以及SATCN模型在不同网络深度下模型建模效果。结果表明，在一阶微分光谱预处理的情况下，浅层SATCN模型建模效果优于深层模型；SATCN模型中的自注意力残差结构，不仅能够强化光谱序列重要特征，模型特征学习能力和预测精度也有显著提高。相比于CNN、TCN等建模方法，提出的SATCN模型建模效果最好，拥有更高的精确度和极好的模型估测能力，验证集的决定系数（R²）为0.943，均方根误差（RMSE）为3.042 g·kg^-1，相对分析误差（RPD）为4.273。综上所述，SOM含量的最佳预测模型是采用SG平滑后一阶微分光谱预处理基础上建立的SATCN预测模型，对广西林地土壤有机质含量进行更加了精准预测。

关键词：土壤；高光谱；有机质；自注意力机制；时间卷积网络

Abstract：The existing convolutional neural network soil organic matter (SOM) prediction models suffer from low modeling effectiveness and low prediction accuracy under small sample data sets. In order to predict the content of organic matter in the soil more accurately, this paper proposes a hyperspectral prediction model of red soil organic matter with an improved Self Attention Temporal Convolutional Network (SATCN) using 206 collected soil samples as the research object. In this paper, after Savitaky-Golay (SG) smoothing is performed on soil samples, four transformations are performed: first-order differential (1DR), second-order differential (2DR), standard normal variable (SNV) and multivariate scattering correction (MSC). The modeling effects of Long Short-Term Memory (LSTM), Partial Least Squares Regression (PLSR) and Support Vector Machine (SVM) under different spectral preprocessing were compared and analyzed. The results show that the first-order differential preprocessing method of the spectrum after SG processing has the best modeling effect. A shallow network structure is applied in the temporal convolutional network (TCN) architecture, a self-attention layer is added to the TCN residual structure to improve the model feature learning capability, and L2 regularization is added to each convolutional kernel weight to prevent overfitting. First-order differentiation is selected as the spectral preprocessing, and four models of ResNet-13, VGGNet-7, TCN and improved temporal convolutional network (SATCN) are constructed to compare and analyze the modeling effects of the four models, as well as the modeling effects of SATCN models at different network depths. The results show that the shallow SATCN modeling is better than the deep model in the case of first-order differential spectral preprocessing; the self-attention residual structure in the SATCN model not only enhanced the important features of the spectral sequence, but also significantly improved the model feature learning ability and prediction accuracy. Compared with modeling methods such as CNN and TCN, the proposed SATCN model has higher accuracy and excellent model estimation capability with a coefficient of determination (R²) of 0.943, root mean square error (RMSE) of 3.042 g·kg^-1, and relative analysis error (RPD) of 4.273 for the validation set. In summary, the best budget SOM content of this paper model is the SATCN prediction model based on the first-order differential spectral preprocessing after SG smoothing, which provides a more accurate prediction of soil organic matter in Guangxi woodlands.

Key words：Soil; Hyperspectral; Organic matter; Self-attentive mechanism; Time-convolutional network

收稿日期: 2022-05-16 修订日期: 2022-07-31

中图分类号:

TP399

基金资助: 国家自然科学基金项目（61662017），广西自然科学基金项目（2018GXNSFAA281235）资助

通讯作者: 王宇 E-mail: 2007002@glut.edu.cn

作者简介: 邓昀，1980年生，桂林理工大学教授 e-mail: 574359451@qq.com

引用本文:

邓昀，牛照文，冯琦尧，王宇. 改进时间卷积网络的红壤有机质高光谱预测模型[J]. 光谱学与光谱分析, 2023, 43(09): 2942-2951.
DENG Yun, NIU Zhao-wen, FENG Qi-yao, WANG Yu. A Novel Hyperspectral Prediction Model of Organic Matter in Red Soil Based on Improved Temporal Convolutional Network. SPECTROSCOPY AND SPECTRAL ANALYSIS, 2023, 43(09): 2942-2951.

链接本文:

https://www.gpxygpfx.com/CN/10.3964/j.issn.1000-0593(2023)09-2942-10 或 https://www.gpxygpfx.com/CN/Y2023/V43/I09/2942

[1] Munson S A, Carey A E. Applied Geochemistry, 2004, 19(7): 1111.
[2] Yan L, Escobar M S, Kaneko H, et al. Chemometrics and Intelligent Laboratory Systems, 2017, 167: 139.
[3] Lin Z D, Wang Y B, Wang R J, et al. Journal of Applied Spectroscopy, 2017, 84(3): 529.
[4] WANG Hai-feng, ZHANG Zhi-tao, CHEN Jun-ying, et al(王海峰, 张智韬, 陈俊英, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2018, 34(14): 124.
[5] Shen L, Gao M, Yan J, et al. Remote Sensing, 2020, 12(7): 1206.
[6] Padarian J, Minasny B, McBratney A B. Geoderma Regional, 2019, 16: e00198.
[7] TAN Ai-ling, WANG Xiao-si, CHU Zhen-yuan, et al(谈爱玲, 王晓斯, 楚振原, 等). Food and Fermentation Industry(食品与发酵工业), 2020, 46(23): 213.
[8] ZHONG Liang, GUO Xi, GUO Jia-xin, et al(钟亮, 郭熙, 国佳欣, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2021, 37(1): 203.
[9] YU Lei, HONG Yong-sheng, ZHOU Yong, et al(于雷, 洪永胜, 周勇, 等). Transactions of the Chinese Society of Agricultural Engineering(农业工程学报), 2016, 32(13): 95.
[10] Galvao R K H, Araujo M C U, José G E, et al. Talanta, 2005, 67(4): 736.
[11] Li R, Zheng S, Duan C, et al. Remote Sensing, 2020, 12(3): 582.
[12] Mei X, Pan E, Ma Y, et al. Remote Sensing, 2019, 11(8): 963.
[13] Li R, Yin B, Cong Y, et al. Sensors, 2020, 20(21): 6271.
[14] WEI Dan-ping, ZHENG Guang-hui(魏丹萍, 郑光辉). Spectroscopy and Spectral Analysis(光谱学与光谱分析), 2022, 42(2): 517.
[15] GUO Jia-xin, ZHU Qing, ZHAO Xiao-min, et al(国佳欣, 朱青, 赵小敏, 等). Journal of Applied Ecology(应用生态学报), 2020, 31(3): 863.
[16] SHI Lei, ZHANG Xin-qian, TAO Yong-cai, et al(石磊, 张鑫倩, 陶永才, 等). Journal of Chinese Computer Systems(小型微型计算机系统), 2019, 40(7): 1486.