Selected work

Representative Papers

A Modelling and Evaluation Framework for EuroCrops-Driven Sentinel-2 Crop Segmentation

May 30, 2026

This study addresses challenges in crop semantic segmentation using Sentinel-2 imagery and EuroCrops field annotations, including label heterogeneity, domain shift, and limited cross-regional generalization. To tackle these issues, the authors develop a configurable data processing and evaluation framework that harmonizes multi-source vector labels into aligned multispectral image–mask pairs. They train a four-level U-Net architecture with Group Normalization to segment ten crop classes, leveraging all ten Sentinel-2 spectral bands and optimizing a composite loss function combining class-weighted cross-entropy and Dice loss. Experimental results show a mean Intersection over Union (mIoU) of 0.7665 and pixel accuracy of 0.8693 on an internal test set. The first systematic cross-regional evaluation reveals strong transfer performance for dominant crops like maize and wheat, yet highlights limited generalization for minority classes and single-temporal inputs.

0 citationsRead paper

Lake Detection and Water Quality Estimation in Sentinel-2 Data

May 23, 2026

Inland water bodies are becoming increasingly scarce and complex to manage, necessitating efficient and automated monitoring approaches. This study systematically evaluates the performance of three machine learning models against the conventional NDWI thresholding method for lake detection using Sentinel-2 satellite imagery and introduces an interpretable color-mapping scheme tailored to water quality indices. Experimental results demonstrate that the selected optimal model excels in accuracy, robustness, and practical applicability. Furthermore, the proposed visualization technique substantially enhances the interpretability of water quality assessments and strengthens their utility in supporting informed decision-making.

0 citationsRead paper

Texture Regenerating and Grafting Using Genome-Driven Neural Cellular Automata

May 13, 2026

This work proposes a genome-driven neural cellular automata (NCA) framework to address the limitations of conventional multi-texture synthesis methods, which lack self-repair capabilities and flexible composition mechanisms. By initializing specific genomic channels during inference, the proposed approach enables autonomous regeneration of damaged regions and seamless grafting of heterogeneous textures without requiring retraining. The method transcends the constraints of static texture synthesis by supporting dynamic composition, high-quality generation of complex textures, and efficient self-repair of corrupted areas. Consequently, it significantly enhances the robustness and scalability of texture synthesis systems.

0 citationsRead paper

Detection of Adversarial Attacks in Robotic Perception

Mar 30, 2026

This work addresses the vulnerability of deep neural networks to adversarial attacks in robotic semantic segmentation, which poses significant risks to safety-critical systems. To mitigate this issue, the authors propose a dedicated adversarial attack detection method tailored to robotic perception scenarios. By integrating semantic segmentation architectures with an adversarial example detection mechanism, the approach overcomes the limitation of existing robustness research that predominantly focuses on image classification. Experimental results demonstrate that the proposed method effectively identifies and defends against adversarial attacks targeting semantic segmentation models, thereby substantially enhancing the safety and robustness of robotic systems operating in real-world environments.

0 citationsRead paper

Foundational Models and Federated Learning: Survey, Taxonomy, Challenges and Practical Insights

Sep 05, 2025

The integration of federated learning (FL) with foundation models (FMs) lacks a systematic survey and unified taxonomic framework, hindering collaborative modeling over distributed private data. Method: We propose the first comprehensive, lifecycle-oriented taxonomy for FL–FM integration, unifying technical paradigms—including self-supervised learning, fine-tuning, knowledge distillation, and transfer learning—and systematically categorizing 42 representative methods. Our analysis draws on bibliometric and technical reviews of 250 core papers selected from an initial pool of 4,200+. Contribution/Results: We introduce a three-dimensional evaluation framework assessing scalability, efficiency, and complexity. The resulting self-consistent knowledge system provides reproducible technical roadmaps and practical guidelines—particularly for high-privacy domains such as healthcare—thereby advancing both theoretical understanding and real-world deployment of privacy-preserving foundation model learning.

0 citationsRead paper

Recent publications

Latest Papers

A Modelling and Evaluation Framework for EuroCrops-Driven Sentinel-2 Crop Segmentation

May 30, 2026

This study addresses challenges in crop semantic segmentation using Sentinel-2 imagery and EuroCrops field annotations, including label heterogeneity, domain shift, and limited cross-regional generalization. To tackle these issues, the authors develop a configurable data processing and evaluation framework that harmonizes multi-source vector labels into aligned multispectral image–mask pairs. They train a four-level U-Net architecture with Group Normalization to segment ten crop classes, leveraging all ten Sentinel-2 spectral bands and optimizing a composite loss function combining class-weighted cross-entropy and Dice loss. Experimental results show a mean Intersection over Union (mIoU) of 0.7665 and pixel accuracy of 0.8693 on an internal test set. The first systematic cross-regional evaluation reveals strong transfer performance for dominant crops like maize and wheat, yet highlights limited generalization for minority classes and single-temporal inputs.

0 citationsRead paper

Lake Detection and Water Quality Estimation in Sentinel-2 Data

May 23, 2026

Inland water bodies are becoming increasingly scarce and complex to manage, necessitating efficient and automated monitoring approaches. This study systematically evaluates the performance of three machine learning models against the conventional NDWI thresholding method for lake detection using Sentinel-2 satellite imagery and introduces an interpretable color-mapping scheme tailored to water quality indices. Experimental results demonstrate that the selected optimal model excels in accuracy, robustness, and practical applicability. Furthermore, the proposed visualization technique substantially enhances the interpretability of water quality assessments and strengthens their utility in supporting informed decision-making.

0 citationsRead paper

Texture Regenerating and Grafting Using Genome-Driven Neural Cellular Automata

May 13, 2026

This work proposes a genome-driven neural cellular automata (NCA) framework to address the limitations of conventional multi-texture synthesis methods, which lack self-repair capabilities and flexible composition mechanisms. By initializing specific genomic channels during inference, the proposed approach enables autonomous regeneration of damaged regions and seamless grafting of heterogeneous textures without requiring retraining. The method transcends the constraints of static texture synthesis by supporting dynamic composition, high-quality generation of complex textures, and efficient self-repair of corrupted areas. Consequently, it significantly enhances the robustness and scalability of texture synthesis systems.

0 citationsRead paper

Detection of Adversarial Attacks in Robotic Perception

Mar 30, 2026

This work addresses the vulnerability of deep neural networks to adversarial attacks in robotic semantic segmentation, which poses significant risks to safety-critical systems. To mitigate this issue, the authors propose a dedicated adversarial attack detection method tailored to robotic perception scenarios. By integrating semantic segmentation architectures with an adversarial example detection mechanism, the approach overcomes the limitation of existing robustness research that predominantly focuses on image classification. Experimental results demonstrate that the proposed method effectively identifies and defends against adversarial attacks targeting semantic segmentation models, thereby substantially enhancing the safety and robustness of robotic systems operating in real-world environments.

0 citationsRead paper

Foundational Models and Federated Learning: Survey, Taxonomy, Challenges and Practical Insights

Sep 05, 2025

The integration of federated learning (FL) with foundation models (FMs) lacks a systematic survey and unified taxonomic framework, hindering collaborative modeling over distributed private data. Method: We propose the first comprehensive, lifecycle-oriented taxonomy for FL–FM integration, unifying technical paradigms—including self-supervised learning, fine-tuning, knowledge distillation, and transfer learning—and systematically categorizing 42 representative methods. Our analysis draws on bibliometric and technical reviews of 250 core papers selected from an initial pool of 4,200+. Contribution/Results: We introduce a three-dimensional evaluation framework assessing scalability, efficiency, and complexity. The resulting self-consistent knowledge system provides reproducible technical roadmaps and practical guidelines—particularly for high-privacy domains such as healthcare—thereby advancing both theoretical understanding and real-world deployment of privacy-preserving foundation model learning.

0 citationsRead paper