Adaptation of a ResNet-based deep learning model for predicting chemical properties of soils in central Russia using spectroscopy data
https://doi.org/10.55648/1998-6920-2026-20-2-93-106
Abstract
The paper presents adaptation of a deep learning model based on residual neural network ResNet with CBAM attention modules for predicting nitrogen, phosphorus and potassium content in soils of central Russia using visible and near-infrared spectroscopy. The model pre-trained on the global LUCAS library was fine-tuned on a local sample of 100 soil samples from Voronezh, Kursk, Rostov regions and Krasnodar Krai. A partial layer freezing strategy was applied. After fine-tuning, the coefficient of determination of 0.91 and the ratio of performance to deviation of 3.01 were achieved for nitrogen, and 0.56 and 2.40 for phosphorus, respectively. The possibility of using portable spectrometers in precision farming systems is shown.
About the Authors
Ilya Evgenievich Novoselov
Federal State Autonomous Educational Institution of Higher Education "National Research Tomsk State University"
Russian Federation
Leading Software Engineer, Cognitive Robotics LLC
Russian Federation
Leading Software Engineer, Cognitive Robotics LLC
Olga Evgenievna Baklanova
Federal State Autonomous Educational Institution of Higher Education "National Research Tomsk State University"
Russian Federation
Candidate of Physical and Mathematical Sciences, Associate Professor of the Institute of Applied Mathematics and Computer Science, Tomsk State University
Russian Federation
Candidate of Physical and Mathematical Sciences, Associate Professor of the Institute of Applied Mathematics and Computer Science, Tomsk State University
References
1. Shin S.K., Lee S.J., Park J.H. Prediction of Soil Properties Using Vis-NIR Spectroscopy Com-bined with Machine Learning: A Review [Prediction of Soil Properties Using Vis-NIR Spectros-copy Combined with Machine Learning: A Review]. Sensors, 2025, vol. 25, article 5045.
2. Sirotkin V.V., Vasyukov S.V., Usmanov B.M. Izuchenie pochvennykh parametrov na osnove polevykh spektrometricheskikh dannykh [Study of soil parameters based on field spectromet-ric data]. Vestnik Udmurtskogo universiteta. Seriya: Biologiya. Nauki o Zemle, 2020, vol. 30, no. 1, pp. 71–82.
3. Viscarra Rossel R.A., Behrens T. Using data mining to model and interpret soil diffuse reflec-tance spectra [Using data mining to model and interpret soil diffuse reflectance spectra]. Geoderma, 2010, vol. 158, no. 1-2, pp. 46–54.
4. de Santana F.B., de Souza A.M., Poppi R.J. Visible and near Infrared Spectroscopy Coupled to Random Forest to Quantify Some Soil Quality Parameters [Visible and near Infrared Spec-troscopy Coupled to Random Forest to Quantify Some Soil Quality Parameters]. Spectrochim. Acta Part A Mol. Biomol. Spectrosc., 2018, vol. 191, pp. 454–462.
5. Chen S., Li S., Ma W. et al. Rapid determination of soil classes in soil profiles using vis–NIR spectroscopy and multiple objectives mixed support vector classification [Rapid determination of soil classes in soil profiles using vis–NIR spectroscopy and multiple objectives mixed sup-port vector classification]. European Journal of Soil Science, 2019, vol. 70, no. 1, pp. 42–53.
6. Tóth G., Jones A., Montanarella L. (eds.). LUCAS Topsoil Survey. Methodology, data and re-sults [LUCAS Topsoil Survey. Methodology, data and results]. Luxembourg, Publications Of-fice of the European Union, 2013. 141 p. (EUR 26102).
7. Kattenborn T., Leitloff J., Schiefer F., Hinz S. Review on Convolutional Neural Networks (CNN) in Vegetation Remote Sensing [Review on Convolutional Neural Networks (CNN) in Vegetation Remote Sensing]. ISPRS Journal of Photogrammetry and Remote Sensing, 2021, vol. 173, pp. 24–49.
8. He K., Zhang X., Ren S., Sun J. Deep Residual Learning for Image Recognition [Deep Residual Learning for Image Recognition]. 2016 IEEE Conference on Computer Vision and Pattern Recognition (CVPR), Las Vegas, NV, USA, 2016, pp. 770–778.
9. Ng W., Minasny B. Convolutional neural network for simultaneous prediction of several soil properties using visible/near-infrared, mid-infrared, and their combined spectra [Convolutional neural network for simultaneous prediction of several soil properties using visible/near-infrared, mid-infrared, and their combined spectra]. Geoderma, 2019, vol. 352, pp. 251–267.
10. Woo S., Park J., Lee J.-Y., Kweon I.S. CBAM: Convolutional Block Attention Module [CBAM: Convolutional Block Attention Module]. ArXiv, 2018, art. 1807.06521.
11. Wang X., Zhang M.-W., Zhou Y.-N. et al. Simultaneous estimation of multiple soil properties from vis-NIR spectra using a multi-gate mixture-of-experts with data augmentation [Simultane-ous estimation of multiple soil properties from vis-NIR spectra using a multi-gate mixture-of-experts with data augmentation]. Geoderma, 2024, vol. 445, article 116888.
12. Zhong L., Guo X., Xu Z., Ding M. Soil properties: Their prediction and feature extraction from the LUCAS spectral library using deep convolutional neural networks [Soil properties: Their prediction and feature extraction from the LUCAS spectral library using deep convolutional neural networks]. Geoderma, 2021, vol. 402, article 115366.
13. Novoselov I.E., Baklanova O.E. Programmnaya realizatsiya modeli glubokogo obucheniya dlya prognozirovaniya svoystv pochvy na osnove dannykh spektroskopii [Software implemen-tation of a deep learning model for predicting soil properties based on spectroscopy da-ta]. Vestnik Rossiyskogo novogo universiteta. Seriya: Slozhnye sistemy: modeli, analiz, uprav-lenie, 2025, no. 3, pp. 125–139.
14. Buryak Zh.A., Gafurov A.M. Otsenka spektral'no-otrazhatel'nykh svoystv erodirovannykh ag-ropochv Respubliki Tatarstan [Assessment of spectral-reflective properties of eroded agricul-tural soils of the Republic of Tatarstan]. Regional'nye geosistemy, 2025, vol. 49, no. 3, pp. 517-532.
15. Gorokhova I.N., Khitrov N.B. Raspoznavanie kamenistykh, peschanykh i karbonatnykh s poverkhnosti pochv na yuge Privolzhskoy vozvyshennosti (Volgogradskaya oblast') po kos-micheskim snimkam [Recognition of stony, sandy, and calcareous surface soils in the south of the Volga Upland (Volgograd region) from satellite images]. Pochvovedenie, 2023, no. 11, pp. 1340–1356.
Review
For citations:
Novoselov I.E., Baklanova O.E. Adaptation of a ResNet-based deep learning model for predicting chemical properties of soils in central Russia using spectroscopy data. The Herald of the Siberian State University of Telecommunications and Information Science. 2026;20(2):93-106. (In Russ.) https://doi.org/10.55648/1998-6920-2026-20-2-93-106
Views:
44
JATS XML
Email this article 
