Researchers from the University of Sannio and the European Space Agency have developed a Quantum Machine Learning (QML) model, QSpeckleFilter, for Synthetic Aperture Radar (SAR) speckle filtering. The model uses quantum algorithms to improve the quality of SAR images, addressing computational complexities. The team’s work highlights the potential of Quantum Convolutional Neural Networks (QCNNs) in enhancing visual data representation in computer vision. The QSpeckleFilter model preprocesses the dataset and filters the SAR speckle, demonstrating its superiority over other state-of-the-art methods. This advancement opens new possibilities for Earth Observation applications.
Quantum Machine Learning for SAR Speckle Filtering
Francesco Mauro, Alessandro Sebastianelli, Maria Pia Del Rosso, Paolo Gambacand, and Silvia Liberata Ullo from the Engineering Department at the University of Sannio, Benevento, Italy, and the European Space Agency, Frascati, Italy, have developed a Quantum Machine Learning (QML) model for Synthetic Aperture Radar (SAR) speckle filtering. The model, named QSpeckleFilter, uses quantum algorithms to address computational complexities and improve the quality of SAR images. The researchers’ work demonstrates the potential of Quantum Convolutional Neural Networks (QCNNs) in enhancing visual data representation in computer vision through quantum-inspired operations.
The Challenge of Speckle Noise in SAR Images
SAR has revolutionized Earth monitoring, providing detailed insights into terrestrial surface use and cover regardless of weather conditions. However, SAR imagery quality is often compromised by speckle, a granular disturbance that poses challenges in producing accurate results without suitable data processing. The need for filtering SAR images to mitigate the effects of speckle is crucial for obtaining more reliable and actionable insights.
The Potential of Quantum Machine Learning
The revolutionary potential of Quantum Machine Learning (QML) in the domain of despeckling opens to possible investigations aiming to assess the advantages that quantum algorithms can offer over classical approaches. QCNNs signify a groundbreaking advancement in deep learning, merging principles of quantum computing into convolutional neural networks (CNNs). QCNNs aim to elevate visual data representation in computer vision through quantum-inspired operations.
The QSpeckleFilter Model
The QSpeckleFilter model preprocesses the dataset and then filters the SAR speckle. The proposed method can overcome others in the State-of-the-Art (SOTA) as demonstrated in the research. The model uses quantum convolution to preprocess the input data, which are then used to train a modified version of the speckle denoiser proposed in previous work by the same authors.
Background and Methodology
The ongoing challenge of mitigating speckle noise in SAR images has prompted the development of various filtering methods. These methods can be broadly categorised into two main groups: Single-product speckle filtering and Multitemporal speckle filtering. Within these categories, approaches can further be classified based on the underlying principles: Statistical methods and AI-based methods. The researchers opted to build upon the method proposed by Sebastianelli et al., as it already yielded superior metric values on the test set compared to the SOTA QSpeckleFilter.
Data and Methods
The researchers used a long timeseries of GRD Sentinel-1 images without logarithmic scaling from Google Earth Engine (GEE), applying temporal averaging to reduce speckle influence and generating speckle noise using a Gamma distribution. The final dataset comprises 2637 Sentinel-1 acquisitions without speckle and 2637 corresponding acquisitions with speckle. The generated dataset was preprocessed using the Quanvolutional operator, which yielded more informative feature maps, enhancing the overall capability of the model.
Summary
The QSpeckleFilter model represents a significant advancement in the field of SAR speckle filtering. By harnessing the power of quantum algorithms and machine learning, the researchers have developed a method that improves the quality of SAR images, opening new avenues for Earth Observation (EO) applications.
“QSpeckleFilter: a Quantum Machine Learning approach for SAR speckle filtering” is an article authored by F. Mauro, Alessandro Sebastianelli, Maria Pia Del Rosso, Paolo Gamba, and Silvia Liberata Ullo. The article was published on February 2, 2024, and can be accessed through the digital object identifier (DOI) https://doi.org/10.48550/arxiv.2402.01235. The source of the article is arXiv (Cornell University).
arXiv (Cornell University)