Image Processing and Learning group
IPL
The Image Processing and Learning (IPL) research group at Politecnico di Torino conducts advanced research at the intersection of artificial intelligence, signal processing, and imaging, with a strong emphasis on space applications and computer vision. We develop novel machine learning methodologies and signal processing techniques for the analysis, reconstruction, and interpretation of complex data, including images, multi-dimensional signals, 3D data, and graph-structured information. A core focus lies in solving inverse problems in imaging, enabling the recovery of high-quality information from incomplete, noisy, or indirect measurements.
A significant portion of our research is dedicated to Earth observation and space science. The group has extensive expertise in satellite imaging, addressing challenges related to both optical and SAR data. This includes advanced methods for denoising, super-resolution and data fusion across multiple sensors and temporal acquisitions, as well as image segmentation, classification and object detection. Our state-of-the-art super-resolution solutions have achieved top performance in international challenges such as ESA’s PROBA-V and AI4EO competitions, enabling high-resolution products with applications in environmental monitoring, agriculture, and climate analysis. We have also expertise in remote sensing foundation models for satellite image analysis.
The group is also actively engaged in the analysis of space weather phenomena, including the automated detection of coronal mass ejections from solar imagery and the prediction of geomagnetic events through the analysis of time series of solar wind parameters. These efforts contribute to improving our understanding of solar activity and its impact on Earth, with implications for satellite operations and communication systems.
Another major research direction concerns onboard processing and edge intelligence for space systems. IPL develops algorithms and architectures for performing image analysis and data compression directly onboard satellites, reducing the need for data transmission and enabling real-time decision-making. The group has contributed to CCSDS international standards for hyperspectral image compression and has developed AI-based solutions for onboard tasks such as cloud detection, segmentation, anomaly detection and change detection, which meet the stringent computational an energy constraints of space environments.
IPL is also leading research on next-generation space systems. The group has obtained the ERC Advanced Grant project “IntelliSwarm,” which explores high-resolution spatial, spectral, and 3D imaging using satellite swarms. This research leverages neural continuous representations and distributed computing to enable scalable, flexible, and high-performance imaging systems powered by energy-efficient AI models suitable for deployment in space.
Beyond space-specific applications, IPL advances fundamental research in representation learning for scenes in 2D, 3D and more, including graph neural networks and image/video generation using diffusion models, which underpin many of its imaging and inference solutions. These techniques are applied to a wide range of data modalities, such as point clouds and multi-sensor datasets, and support tasks including classification, segmentation, and object detection. A further line of research is devoted to the security of AI-based systems. IPL develops techniques for improving the robustness of neural networks to adversarial perturbations, with applications to robust image classification, trustworthy AI, and safety-critical scenarios.
Our AI research is enable by 30 GPUs 100% owned by the group, as well as privileged access to 24 GPUs in the HPC center at Politecnico di Torino.
The IPL group is participating in several projects funded by the major space companies in Europe, the ESA, the Italian Space Agency and the European Union.
A significant portion of our research is dedicated to Earth observation and space science. The group has extensive expertise in satellite imaging, addressing challenges related to both optical and SAR data. This includes advanced methods for denoising, super-resolution and data fusion across multiple sensors and temporal acquisitions, as well as image segmentation, classification and object detection. Our state-of-the-art super-resolution solutions have achieved top performance in international challenges such as ESA’s PROBA-V and AI4EO competitions, enabling high-resolution products with applications in environmental monitoring, agriculture, and climate analysis. We have also expertise in remote sensing foundation models for satellite image analysis.
The group is also actively engaged in the analysis of space weather phenomena, including the automated detection of coronal mass ejections from solar imagery and the prediction of geomagnetic events through the analysis of time series of solar wind parameters. These efforts contribute to improving our understanding of solar activity and its impact on Earth, with implications for satellite operations and communication systems.
Another major research direction concerns onboard processing and edge intelligence for space systems. IPL develops algorithms and architectures for performing image analysis and data compression directly onboard satellites, reducing the need for data transmission and enabling real-time decision-making. The group has contributed to CCSDS international standards for hyperspectral image compression and has developed AI-based solutions for onboard tasks such as cloud detection, segmentation, anomaly detection and change detection, which meet the stringent computational an energy constraints of space environments.
IPL is also leading research on next-generation space systems. The group has obtained the ERC Advanced Grant project “IntelliSwarm,” which explores high-resolution spatial, spectral, and 3D imaging using satellite swarms. This research leverages neural continuous representations and distributed computing to enable scalable, flexible, and high-performance imaging systems powered by energy-efficient AI models suitable for deployment in space.
Beyond space-specific applications, IPL advances fundamental research in representation learning for scenes in 2D, 3D and more, including graph neural networks and image/video generation using diffusion models, which underpin many of its imaging and inference solutions. These techniques are applied to a wide range of data modalities, such as point clouds and multi-sensor datasets, and support tasks including classification, segmentation, and object detection. A further line of research is devoted to the security of AI-based systems. IPL develops techniques for improving the robustness of neural networks to adversarial perturbations, with applications to robust image classification, trustworthy AI, and safety-critical scenarios.
Our AI research is enable by 30 GPUs 100% owned by the group, as well as privileged access to 24 GPUs in the HPC center at Politecnico di Torino.
The IPL group is participating in several projects funded by the major space companies in Europe, the ESA, the Italian Space Agency and the European Union.
-
Scientific coordinator
-
Research team
Research area
Research topics
- Earth observation and satellite imaging
- Edge processing and intelligence for energy-efficient systems
- Analysis of space weather phenomena
- Artificial intelligence for representation learning, inverse problems, inference and compression
- Next-generation space imaging systems
- Security of AI-based systems