FuzzyVGG: Improving Uncertainty and Interpretability in Satellite Image Classification through Fuzzy Logic Integration: Improving Uncertainty and Interpretability inSatellite Image Classification through Fuzzy LogicIntegration

Nawel Slimani; Ghazala Hcini; Imen Jdey; Monji Kherallah

doi:10.22456/2175-2745.151945

Authors

Nawel Slimani University of Sfax (USS)
Ghazala Hcini University of Sfax (USS) https://orcid.org/0000-0003-3571-417X
Imen Jdey University of Sfax (USS) https://orcid.org/0000-0001-7937-941X
Monji Kherallah University of Sfax (USS)

DOI:

https://doi.org/10.22456/2175-2745.151945

Keywords:

deep learning, satellite image classification, fuzzy logic, uncertainty handling, hybrid models, interpretability

Abstract

Convolutional neural networks (CNNs) have demonstrated impressive performance in image recognition tests due to their deep hierarchical feature extraction capabilities. But when it comes to satellite imagery, which frequently has inherent uncertainties like shifting lighting, unclear object boundaries, and overlapping classes, these conventional models provide serious difficulties. Standard CNN architectures may perform less well as a result of these ambiguities. To address these limitations, this study introduces a novel hybrid approach called Fuzzy VGG, which integrates fuzzy logic into the VGG16 and VGG19 architectures. The primary objective is to enhance model interpretability and robustness in handling the uncertainties present in satellite image data. Our method replaces the conventional dense-softmax output layer with a custom fuzzy classification layer that represents soft class membership using cosine similarity. Our experiments on the UC Merced Land Use dataset show that the integration of fuzzy logic leads to significantly improved performance. The Fuzzy VGG16 model achieved an accuracy of 90%, while Fuzzy VGG19 reached 97% accuracy. This represents a consistent performance gain of 15% over the standard VGG16 (75% accuracy) and VGG19 (80% accuracy) models. These results confirm the effectiveness of our hybrid system in creating a more robust and reliable decision-making mechanism for complex image classification tasks. On a medical dataset, the fuzzy VGG model achieved an overall accuracy of 96%, with macro and weighted averages for precision, recall, and F1-score all equal to 96%.

Downloads

Download data is not yet available.

References

[1] KATKANI, D. et al. A review on applications and utility of remote sensing and geographic information systems in agriculture and natural resource management. International Journal of Environment and Climate Change, v. 12, n. 4, p. 1–18, 2022.

[2] DRITSAS, E.; TRIGKA, M. Remote sensing and geospatial analysis in the big data era: A survey. Remote Sensing, MDPI, v. 17, n. 3, p. 550, 2025.

[3] HCINI, G.; JDEY, I. Boosting hyperspectral image classification with a 3D CNN and vision transformer hybrid architecture. In: IEEE. 2025 11th International Conference on Control, Decision and Information Technologies (CoDIT). [S.l.], 2025. v. 1, p. 2660–2665.

[4] SLIMANI, N.; JDEY, I.; KHERALLAH, M. Performance comparison of machine learning methods based on CNN for satellite imagery classification. In: IEEE. 2023 9th International Conference on Control, Decision and Information Technologies (CoDIT). [S.l.], 2023. p. 185–189.

[5] ZHAO, X. et al. A review of convolutional neural networks in computer vision. Artificial Intelligence Review, Springer, v. 57, n. 4, p. 99, 2024.

[6] KHANAM, R. et al. A comprehensive review of convolutional neural networks for defect detection in industrial applications. IEEE Access, IEEE, 2024.

[7] IGE, A. O.; SIBIYA, M. State-of-the-art in 1D convolutional neural networks: A survey. IEEE Access, IEEE, 2024.

[8] ICHSAN, A.; RIYADI, S.; PARDEDE, D. Analysis of logistic regression regularization in wild elephant classification with VGG-16 feature extraction. Journal of Computer Networks, Architecture and High Performance Computing, v. 6, n. 2, p. 783–793, 2024.

[9] MAGDY, A. et al. Performance enhancement of skin cancer classification using computer vision. IEEE Access, IEEE, v. 11, p. 72120–72133, 2023.

[10] ALI, S.; AGRAWAL, J. Automated segmentation of brain tumour images using deep learning-based model VGG19 and ResNet 101. Multimedia Tools and Applications, Springer, v. 83, n. 11, p. 33351–33370, 2024.

[11] CHOWDHURY, M. T. et al. Classification of satellite images with VGG19 and convolutional neural network (CNN). In: IEEE. 2024 2nd International Conference on Advancement in Computation & Computer Technologies (InCACCT). [S.l.], 2024. p. 397–402.

[12] MITTAL, S. et al. Image classification of satellite using VGG16 model. In: IEEE. 2024 2nd International Conference on Disruptive Technologies (ICDT). [S.l.], 2024. p. 401–404.

[13] ADEGUN, A. A.; VIRIRI, S.; TAPAMO, J.-R. Review of deep learning methods for remote sensing satellite images classification: experimental survey and comparative analysis. Journal of Big Data, Springer, v. 10, n. 1, p. 93, 2023.

[14] AHMAD, M. et al. Hyperspectral image classification—traditional to deep models: A survey for future prospects. IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing, IEEE, v. 15, p. 968–999, 2021.

[15] ZHANG, R.; SHEN, F.; ZHAO, J. A model with fuzzy granulation and deep belief networks for exchange rate forecasting. In: IEEE. 2014 International Joint Conference on Neural Networks (IJCNN). [S.l.], 2014. p. 366–373.

[16] ZHOU, S.; CHEN, Q.; WANG, X. Fuzzy deep belief networks for semi-supervised sentiment classification. Neurocomputing, Elsevier, v. 131, p. 312–322, 2014.

[17] PARK, S. et al. Intra-and inter-fractional variation prediction of lung tumors using fuzzy deep learning. IEEE Journal of Translational Engineering in Health and Medicine, IEEE, v. 4, p. 1–12, 2016.

[18] WANG, L.-X. Generating fuzzy rules from numerical data, with applications. University of Southern California, tech. report, 1991.

[19] TANG, H. H.; AHMAD, N. S. Fuzzy logic approach for controlling uncertain and nonlinear systems: a comprehensive review of applications and advances. Systems Science & Control Engineering, Taylor & Francis, v. 12, n. 1, p. 2394429, 2024.

[20] CHEN, C. P. et al. Fuzzy restricted Boltzmann machine for the enhancement of deep learning. IEEE Transactions on Fuzzy Systems, IEEE, v. 23, n. 6, p. 2163–2173, 2015.

[21] IMAMGULUYEV, R. Integrating fuzzy logic with deep learning: A new approach to explainable artificial intelligence. In: IEEE. 2025 6th International Conference on Mobile Computing and Sustainable Informatics (ICMCSI). [S.l.], 2025. p. 1701–1706.

[22] ZHENG, Y.-J. et al. A Pythagorean-type fuzzy deep denoising autoencoder for industrial accident early warning. IEEE Transactions on Fuzzy Systems, IEEE, v. 25, n. 6, p. 1561–1575, 2017.

[23] DENG, Y. et al. Visual words assignment via information-theoretic manifold embedding. IEEE Transactions on Cybernetics, IEEE, v. 44, n. 10, p. 1924–1937, 2014.

[24] RAHMAT, B. et al. Vehicle license plate image segmentation system using cellular neural network optimized by adaptive fuzzy and neuro-fuzzy algorithms. International Journal of Multimedia and Ubiquitous Engineering, Science and Engineering Research Support Society, v. 11, n. 12, p. 383–400, 2016.

[25] HATRI, C. E.; BOUMHIDI, J. Fuzzy deep learning based urban traffic incident detection. Cognitive Systems Research, Elsevier, v. 50, p. 206–213, 2018.

[26] TABRIZI, P. R. et al. Automatic kidney segmentation in 3D pediatric ultrasound images using deep neural networks and weighted fuzzy active shape model. In: IEEE. 2018 IEEE 15th International Symposium on Biomedical Imaging (ISBI 2018). [S.l.], 2018. p. 1170–1173.

[27] IRTAZA, G. et al. Real-time satellite image classification using convolutional neural networks for earth observation and video feed analysis. Kashf Journal of Multidisciplinary Research, v. 2, n. 03, p. 204–216, 2025.

[28] YUAN, Z. et al. Few-shot remote sensing image scene classification based on metric learning and local descriptors. Remote Sensing, MDPI, v. 15, n. 3, p. 831, 2023.

[29] YOU, H.; GU, J.; JING, W. Multi-label remote sensing image land cover classification based on a multi-dimensional attention mechanism. Remote Sensing, MDPI, v. 15, n. 20, p. 4979, 2023.

[30] KUMARI, N.; MINZ, S. Deep residual SVM: a hybrid learning approach to obtain high discriminative feature for land use and land cover classification. Procedia Computer Science, Elsevier, v. 218, p. 1454–1462, 2023.

[31] ÇIKLAÇANDIR, F. Y.; UTKU, S. Performance comparison of CNN based hybrid systems using UC Merced land-use dataset. Dokuz Eylül Üniversitesi Mühendislik Fakültesi Fen ve Mühendislik Dergisi, Dokuz Eylul University, v. 25, n. 75, p. 725–737.

[32] YE, P. et al. MobileTransNext: Integrating CNN, Transformer, and BiLSTM for image classification. Alexandria Engineering Journal, Elsevier, v. 123, p. 460–470, 2025.

[33] FAYAZ, M. et al. Land-cover classification using deep learning with high-resolution remote-sensing imagery. Applied Sciences, MDPI, v. 14, n. 5, p. 1844, 2024.

[34] CAO, L. A MobileNetV2 model of transfer learning is employed for remote sensing image classification. Advances in Engineering Technology Research, v. 10, n. 1, p. 596–596, 2024.

[35] KARAKUTUK, A. K.; OZDEMIR, O.; SENTURK, S. Optuna-optimized Pythagorean fuzzy deep neural network: A novel framework for uncertainty-aware image classification. Applied Sciences, MDPI, v. 15, n. 20, p. 11097, 2025.