Phenological classification of rice crops using ResNet-18 and Sentinel-2 images
Keywords:
precision agriculture, deep learning, vegetation indices, phenological monitoring, multispectral remote sensingAbstract
The classification of rice phenological stages is a critical factor for food security, as it enables the optimization of fertilizer use, irrigation management, and harvest planning. However, traditional field monitoring is costly and difficult to scale. This study proposes an advanced remote sensing method based on the ResNet-18 residual neural network architecture and Sentinel-2 satellite imagery to automatically identify vegetative, reproductive, and maturation stages. The research focused on six municipalities in the state of Rio Grande do Sul, Brazil, during the 2019-2020, 2022-2023, and 2023-2024 agricultural cycles. Methodologically, the network was adapted to process 64 × 64-pixel patches, employing an innovative five-channel input configuration composed of specific vegetation indices (NDVI, GNDVI, EVI, NDRE, and SAVI) instead of conventional spectral bands. This integration enabled the capture of robust biophysical signatures under varying atmospheric conditions. Results obtained from the independent test set demonstrated outstanding performance, with an accuracy of 92.17 %, precision of 92.22 %, recall (sensitivity) of 92.23 %, and an F1-score of 92.15 %. Furthermore, a Kappa index of 88.26 % confirms high agreement and model stability beyond chance. It is concluded that the combination of deep learning and multispectral vegetation indices provides a scalable and highly reliable tool for precision agricultural monitoring. The model proved resilient even in scenarios with limited data availability due to cloud cover, facilitating its implementation in tropical and subtropical regions for real-time data-driven decision-making.
References
G. Xie y S. Niculescu, «Mapping Crop Types Using Sentinel-2 Data Machine Learning and Monitoring Crop Phenology with Sentinel-1 Backscatter Time Series in Pays de Brest, Brittany, France», Remote Sensing, vol. 14, n.o 18, p. 4437, sep. 2022, doi: https://doi.org/10.3390/rs14184437.
A. Nazir et al., «Estimation and Forecasting of Rice Yield Using Phenology-Based Algorithm and Linear Regression Model on Sentinel-II Satellite Data», Agriculture, vol. 11, n.o 10, p. 1026, oct. 2021, doi: https://doi.org/10.3390/agriculture11101026.
G. Tian et al., «An Automatic Method for Rice Mapping Based on Phenological Features with Sentinel-1 Time-Series Images», Remote Sensing, vol. 15, n.o 11, p. 2785, may 2023, doi: https://doi.org/10.3390/rs15112785.
J. Wang et al., «Estimation of rice phenology date using integrated HJ-1 CCD and Landsat-8 OLI vegetation indices time-series images», J. Zhejiang Univ. Sci. B, vol. 16, n.o 10, pp. 832-844, oct. 2015, doi: https://doi.org/10.1631/jzus.B1500087.
T. N. Matongera, O. Mutanga, M. Sibanda, y J. Odindi, «Estimating and Monitoring Land Surface Phenology in Rangelands: A Review of Progress and Challenges», Remote Sensing, vol. 13, n.o 11, p. 2060, may 2021, doi: https://doi.org/10.3390/rs13112060.
J. Segarra, M. L. Buchaillot, J. L. Araus, y S. C. Kefauver, «Remote Sensing for Precision Agriculture: Sentinel-2 Improved Features and Applications», Agronomy, vol. 10, n.o 5, p. 641, may 2020, doi: https://doi.org/10.3390/agronomy10050641.
P. Ebel, A. Meraner, M. Schmitt, y X. X. Zhu, «Multisensor Data Fusion for Cloud Removal in Global and All-Season Sentinel-2 Imagery», IEEE Trans. Geosci. Remote Sensing, vol. 59, n.o 7, pp. 5866-5878, jul. 2021, doi: https://doi.org/10.1109/TGRS.2020.3024744.
R. Abdollahi-Arpanahi, D. Gianola, y F. Peñagaricano, «Deep learning versus parametric and ensemble methods for genomic prediction of complex phenotypes», Genet Sel Evol, vol. 52, n.o 1, p. 12, dic. 2020, doi: https://doi.org/10.1186/s12711-020-00531-z.
Z. Yi, L. Jia, y Q. Chen, «Crop Classification Using Multi-Temporal Sentinel-2 Data in the Shiyang River Basin of China», Remote Sensing, vol. 12, n.o 24, p. 4052, dic. 2020, doi: https://doi.org/10.3390/rs12244052.
A. Htitiou, A. Boudhar, Y. Lebrini, R. Hadria, H. Lionboui, y T. Benabdelouahab, «A comparative analysis of different phenological information retrieved from Sentinel-2 time series images to improve crop classification: a machine learning approach», Geocarto International, vol. 37, n.o 5, pp. 1426-1449, mar. 2022, doi: https://doi.org/10.1080/10106049.2020.1768593.
CONAB, «Calendário de Plantio e Colheita de Grãos no Brasil 2022». CONAB, Brasil, 2022. [En línea]. Disponible en: https://www.gov.br/conab/pt-br/acesso-a-informacao/institucional/publicacoes/arquivos-de-paginas/calendariozplantiozezcolheitazjunz2022.pdf
L. Wang, F. Zhang, Y. Jing, X. Jiang, S. Yang, y X. Han, «Multi-Temporal Detection of Rice Phenological Stages Using Canopy Spectrum», Rice Science, vol. 21, n.o 2, pp. 108-115, mar. 2014, doi: https://doi.org/10.1016/S1672-6308(13)60170-5.
CONAB, «Mapeamentos Agrícolas». 2025. [En línea]. Disponible en: https://portaldeinformacoes.conab.gov.br/mapeamentos-agricolas-downloads.html
M. Pompa-García et al., «NDVI Values Suggest Immediate Responses to Fire in an Uneven-Aged Mixed Forest Stand», Forests, vol. 13, n.o 11, p. 1901, nov. 2022, doi: https://doi.org/10.3390/f13111901.
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Saastal
This work is licensed under a Creative Commons Attribution 4.0 International License.
