EVALUATION OF THE U-NET ARCHITECTURE FOR SEMANTIC SEGMENTATION OF NATURAL LAND COVERS IN RGB SATELLITE IMAGES
EVALUATION OF THE U-NET ARCHITECTURE FOR SEMANTIC SEGMENTATION OF NATURAL LAND COVERS IN RGB SATELLITE IMAGES
DOI:
https://doi.org/10.23854/07199562.2025613.ruizKeywords:
semantic segmentation, U-Net, automated cartography, satellite imagery, geospatial artificial intelligenceAbstract
This study evaluates the feasibility of applying convolutional neural networks, specifically the U-Net architecture, for the semantic segmentation of natural land covers in RGB satellite images from the DeepGlobe dataset. The research is part of the binational COMIXTA project between the Military Geographic Institute of Chile (IGM) and the Agustín Codazzi Geographic Institute of Colombia (IGAC), aimed at strengthening cartographic
methodologies based on artificial intelligence. Two versions of the model were trained: one without an explicit
validation set and another using a simple validation strategy with an 80/20 data split and an early stopping
mechanism. The results show that the model without validation suffered from overfitting, reaching artificially high metrics (IoU increased to 0.83), while the model with validation produced more conservative but generalizable
predictions (IoU equivalent to 0.42). Qualitative evaluation revealed systematic errors in the “water” class due to data imbalance. Techniques such as mixed precision training, robust normalization, and GELU activation were used to improve training efficiency and stability. The implementation was carried out in an accessible computing environment (NVIDIA T1000 GPU), demonstrating that these methodologies can be replicated in public institutions with limited resources. This work establishes a solid technical foundation for future extensions toward multiclass models, integration of multispectral imagery, and large-scale automated artographic production.
Downloads
References
Bengio, Y. (2012). Practical recommendations for gradient-based training of deep architectures. En G. Montavon, G. B. Orr, & K.-R. Müller (Eds.), Lecture notes in computer science: Vol. 7700. Neural networks: Tricks of the trade (2ª ed., pp. 437–478). Springer. https://doi.org/10.1007/978-3-642-35289-8_26
Buda, M., Maki, A., & Mazurowski, M. A. (2018). A systematic study of the class imbalance problem in convolutional neural networks. Neural Networks, 106, 249–259. https://doi.org/10.1016/j.neunet.2018.07.011
Demir, I., Koperski, K., Lindenbaum, D., Pang, G., Huang, J., Basu, S., Hughes, F., Tuia, D., & Raskar, R. (2018). DeepGlobe 2018: A challenge to parse the Earth through satellite images. En Proceedings of the IEEE Conference on Computer Vision and Pattern Recognition Workshops (CVPRW) (pp. 172–181). IEEE. https://doi.org/10.1109/CVPRW.2018.00031
Everingham, M., van Gool, L., Williams, C. K. I., Winn, J., & Zisserman, A. (2010). The pascal visual object classes (VOC) Challenge. International Journal of Computer Vision, 88(2), 303–338.
Gómez, C., White, J. C., & Wulder, M. A. (2016). Optical remotely sensed time series data for land cover classification: A review. ISPRS Journal of Photogrammetry and Remote Sensing, 116, 55–72. https://doi.org/10.1016/j.isprsjprs.2016.03.008
Hendrycks, D., & Gimpel, K. (2016). Gaussian Error Linear Units (GELUs). ArXiv. https://arxiv.org/abs/1606.08415
Ioffe, S., & Szegedy, C. (2015). Batch normalization: Accelerating deep network training by reducing internal covariate shift. En Proceedings of the 32nd International Conference on Machine Learning (ICML) (Vol. 37, pp. 448–456). PMLR.
Kedron, P., Frazier, A. E., Goodchild, M. F., & Li, W. (2021). Reproducibility and replicability: A new hope for quantitative geography. Annals of the American Association of Geographers, 111(5), 1271-1274. https://doi.org/10.1080/24694452.2020.1863548
Li, X., He, Y., & Chen, Z. (2020). U-Net based deep learning for deforestation detection in Amazon rainforest using Sentinel-2 imagery. Remote Sensing Letters, 11(12), 1085-1094.
Lin, T.-Y., Goyal, P., Girshick, R., He, K., & Dollár, P. (2020). Focal loss for dense object detection. IEEE Transactions on Pattern Analysis and Machine Intelligence, 42(2), 318–327. https://doi.org/10.1109/TPAMI.2018.2858826
Loshchilov, I., & Hutter, F. (2019). Decoupled weight decay regularization. En International Conference on Learning Representations (ICLR).
Maggiori, E., Tarabalka, Y., Charpiat, G., & Alliez, P. (2017). Convolutional neural networks for large-scale remote-sensing image classification. IEEE Transactions on Geoscience and Remote Sensing, 55(2), 645–657. https://doi.org/10.1109/TGRS.2016.2612821
Micikevicius, P., Narang, S., Alben, J., Diamos, G., Elsen, E., Garcia, D., Ginsburg, B., Houston, M., Kuchaiev, O., Venkatesh, G., & Wu, H. (2018). Mixed precision training. En International Conference on Learning Representations (ICLR).
Ronneberger, O., Fischer, P., & Brox, T. (2015). U-Net: Convolutional networks for biomedical image segmentation. En N. Navab, J. Hornegger, W. M. Wells, & A. F. Frangi (Eds.), Lecture Notes in Computer Science: Vol. 9351. Medical Image Computing and Computer-Assisted Intervention – MICCAI 2015 (pp. 234–241). Springer. https://doi.org/10.1007/978-3-319-24574-4_28
Sokolova, M., & Lapalme, G. (2009). A systematic analysis of performance measures for classification tasks. Information Processing & Management, 45(4), 427–437. https://doi.org/10.1016/j.ipm.2009.03.002
Srivastava, N., Hinton, G., Krizhevsky, A., Sutskever, I., & Salakhutdinov, R. (2014). Dropout: A simple way to prevent neural networks from overfitting. Journal of Machine Learning Research, 15(56), 1929–1958.
Tuia, D., Volpi, M., Copa, L., & Kanevski, M. (2011). A survey of active learning algorithms for supervised remote sensing image classification. IEEE Journal of Selected Topics in Signal Processing, 10(8), 1325-1337.
Volpi, M., & Tuia, D. (2017). Fully convolutional networks for semantic segmentation of aerial images. En 2017 IEEE International Geoscience and Remote Sensing Symposium (IGARSS) (pp. 3452-3455). IEEE.
Zhang, F., Liu, C., & Wang, L. (2019). Water body segmentation in urban areas from high-resolution images using a U-Net deep network. Remote Sensing, 11(21), 2530.
Zhu, X. X., Tuia, D., Mou, L., Xia, G. S., Zhang, L., Xu, F., & Fraundorfer, F. (2017). Deep learning in remote sensing: A comprehensive review and list of resources. IEEE Geoscience and Remote Sensing Magazine, 5(4), 8–36. https://doi.org/10.1109/MGRS.2017.2762307
Downloads
Published
How to Cite
Issue
Section
License
Copyright (c) 2025 Revista Geográfica de Chile Terra Australis
This work is licensed under a Creative Commons Attribution 4.0 International License.
Copyright notice:
Revista Geográfica de Chile Terra Australis retains the economic rights (copyright) of the works published in the Revista Geográfica de Chile Terra Australis. The reuse of the content is allowed under a license:
CC BY
Recognition
This license allows others to distribute, remix, tweak and build upon your work, even for commercial purposes, as long as you are acknowledged as the author of the original creation. This is the most helpful license offered. It is recommended for maximum dissemination and use of the materials subject to the license.
