Institution profile

Thales

Industry researcheurope · fr
Official website
Research library62linked papers
Opportunities0open roles
Selected work

Representative Papers

Using Shadows in Circular Synthetic Aperture Sonar Imaging for Target Analysis

Jan 23, 2026

Circular synthetic aperture sonar (CSAS) enhances azimuth resolution and coverage but sacrifices seafloor shadow information—critical for target recognition and 3D reconstruction—due to its omnidirectional imaging. This work presents the first systematic approach to recover and exploit shadows from CSAS data: multi-view images are generated via sub-aperture filtering, clear shadows are extracted using a fixed-focus shadow enhancement (FFSE) technique, and an interactive visualization interface facilitates manual segmentation. Subsequently, a space-carving method infers the 3D shape of targets from shadow contours. By moving beyond the conventional reliance on intensity-only CSAS imagery, the proposed framework significantly improves target identification and demonstrates the efficacy and potential of shadow information for underwater 3D reconstruction.

1 citationsRead paper

Hybrid Quantum CNN for Cross-Sensor Spaceborne Volcanic Thermal Activity Recognition Worldwide

Jul 29, 2026

This work addresses the limitations of classical deep learning models in on-orbit processing of global multi-source satellite thermal anomaly data—specifically their poor cross-sensor generalization, heavy reliance on large annotated datasets, and incompatibility with onboard computational constraints. To overcome these challenges, the authors propose a hybrid quantum AlexNet architecture that integrates a parameterized quantum circuit (PQC) as a trainable layer embedded within a classical convolutional backbone. By leveraging quantum embedding to map high-level image features into a high-dimensional Hilbert space, the model substantially enhances feature discriminability and environmental robustness. Experimental results demonstrate that, compared to purely classical counterparts, the proposed approach achieves higher accuracy and superior cross-domain generalization in volcanic thermal anomaly detection across diverse sensors, while significantly reducing both the number of trainable parameters and the required volume of training data.

0 citationsRead paper
Recent publications

Latest Papers

Hybrid Quantum CNN for Cross-Sensor Spaceborne Volcanic Thermal Activity Recognition Worldwide

Jul 29, 2026

This work addresses the limitations of classical deep learning models in on-orbit processing of global multi-source satellite thermal anomaly data—specifically their poor cross-sensor generalization, heavy reliance on large annotated datasets, and incompatibility with onboard computational constraints. To overcome these challenges, the authors propose a hybrid quantum AlexNet architecture that integrates a parameterized quantum circuit (PQC) as a trainable layer embedded within a classical convolutional backbone. By leveraging quantum embedding to map high-level image features into a high-dimensional Hilbert space, the model substantially enhances feature discriminability and environmental robustness. Experimental results demonstrate that, compared to purely classical counterparts, the proposed approach achieves higher accuracy and superior cross-domain generalization in volcanic thermal anomaly detection across diverse sensors, while significantly reducing both the number of trainable parameters and the required volume of training data.

0 citationsRead paper

Volcanic Clouds Detection through QCNN and Geostationary Satellite Multispectral Imagery

Jul 29, 2026

Accurate identification of volcanic clouds remains challenging due to their spectral similarity to meteorological clouds, highly variable eruption morphologies, and the limited spectral resolution of geostationary satellites. This work proposes a novel hybrid quantum-classical convolutional architecture for volcanic cloud detection in remote sensing, integrating quantum convolutional neural networks (QCNNs) with classical CNNs to classify SEVIRI multispectral imagery into categories containing volcanic ash, SO₂, or their mixtures. Two QCNN variants—employing 2 and 4 qubits, respectively—are evaluated experimentally. Results demonstrate that the proposed approach effectively identifies volcanic cloud scenarios and outperforms purely classical models, thereby validating the practical potential and feasibility of quantum machine learning in Earth observation applications.

0 citationsRead paper