Spaceborne synthetic aperture radar (SAR) represents a powerful source of data for enhancing maritime domain awareness (MDA). Wakes generated by traveling vessels hold a crucial role in MDA since they can be exploited both for ship route and velocity estimation and as a marker of ship presence. Even if deep learning (DL) has led to an impressive performance boost on a variety of computer vision tasks, its usage for automatic target recognition (ATR) in SAR images to support MDA is still limited to the detection of ships rather than ship wakes. A dataset is presented in this paper and several state-of-the-art object detectors based on convolutional neural networks (CNNs) are tested with different backbones. The dataset, including more than 250 wake chips, is realized by visually inspecting Sentinel-1 images over highly trafficked maritime sites. Extensive experiments are shown to characterize CNNs for the wake detection task. For the first time, a deep-learning approach is implemented to specifically detect ship wakes without any a-priori knowledge or cuing about the location of the vessel that generated the wake. No annotated dataset was available to train deep-learning detectors on this task, which is instead presented in this paper. Moreover, the benchmarks achieved for different detectors point out promising features and weak points of the relevant approaches. Thus, the work also aims at stimulating more research in this promising, but still under-investigated, field.

Nowadays, there is growing interest in applying Artificial Intelligence (AI) on board Earth Observation (EO) satellites for time-critical applications, such as natural disaster response. However, the unavailability of raw satellite data currently hinders research of lightweight pre-processing techniques and limits the exploration of end-to-end pipelines, which extract insights directly from the source data. To fill this gap, this work presents a novel methodology to automate the creation of datasets for the detection of target events (e.g., warm thermal hotspots) or objects (e.g., vessels) from Sentinel-2 raw data and other multispectral EO pushbroom raw imagery. The presented approach first processes the raw data by applying a pipeline consisting of a spatial band registration and georeferencing of the raw data pixels. Then, it detects the target events by leveraging event-specific state-of-the-art algorithms on the Level-1C products, which are mosaicked and cropped on the georeferenced correspondent raw granule area. The detected events are, finally, re-projected back on the corresponding raw images. We apply the proposed methodology to realize THRawS (Thermal Hotspots in Raw Sentinel-2 data), the first dataset of Sentinel-2 raw data containing warm thermal hotspots. THRawS includes 1090 samples containing wildfires, volcanic eruptions, and 33335 eventfree acquisitions to enable thermal hotspot detection and general classification applications. This dataset and associated toolkits provide the community with both an immediately useful resource to speed up future research on energy-efficient pre-processing algorithms and AI-based end-to-end processing systems on board EO satellites.

In the framework of autonomous spacecraft navigation, this manuscript proposes a novel vision-based terrain relative navigation (TRN) system called FederNet. The developed system exploits a pattern of observed craters to perform an absolute position measurement. The obtained measurements are thus integrated into a navigation filter to estimate the spacecraft state in terms of position and velocity. Recovering crater locations from elevation imagery is not an easy task since sensors can generate images with vastly different appearances and qualities. Hence, several problems have been faced. First, the crater detection problem from elevation images, second, the crater matching problem with known craters, the spacecraft position estimation problem from retrieved matches, and its integration with a navigation filter. The first problem was countered with the robust approach of deep learning. Then, a crater matching algorithm based on geometric descriptors was developed to solve the pattern recognition problem. Finally, a position estimation algorithm was integrated with an Extended Kalman Filter, built with a Keplerian propagator. This key choice highlights the performance achieved by the developed system that could benefit from more accurate propagators. FederNet system has been validated with an experimental analysis on real elevation images. Results showed that FederNet is capable to cruise with a navigation accuracy below 400 meters when a sufficient number of well-distributed craters is available for matching. FederNet capabilities can be further improved with higher resolution data and a data fusion integration with other sensor measurements, such as the lunar GPS, nowadays under investigation by many researchers.

In the framework of maritime surveillance, vessel detection techniques based on spaceborne synthetic aperture radar (SAR) images have promoted extensive applications for the effective understanding of unlawful activities at sea. This paper deals with this topic, presenting a novel approach that exploits a cascade application of a pre-screening algorithm and a discrimination phase. Pre-screening is based on a constant false alarm rate (CFAR) detector, whereas discrimination exploits sub-look analysis (SLA). For the first time, the method has been validated with experiments on multi-frequency (C-, X-, and L-band) SAR images, demonstrating a significant reduction of up to 40% in false alarms within highly congested scenarios, along with a notable enhancement of the receiving operating characteristic (ROC) curves. For future synergic exploitation of multiple SAR missions, the developed dataset, composed of Sentinel-1, SAOCOM, and COSMO-SkyMed images, is comprehensive, having images gathered over the same area with a short time lag (below 15 min). Finally, the diversified processing chains and the results for each mission product and scenario are discussed. Being the first dataset of single-look complex (SLC) SAR multi-frequency data, the present work intends to encourage additional investigation in this promising field of research.

The wakes generated by moving vessels represent relevant patterns in remotely sensed images. They are a marker of ship presence and can be processed to infer route, speed, size, and type of ships. Automatic wake detection can be exploited by law enforcement agencies and local authorities for ensuring a wide range of applications, including maritime traffic surveillance, border control, and protection of marine protected areas. The topic is thus attracting increasing interest from the remote sensing community. This paper contributes in this context presenting a novel approach based on the detection of the keypoints of wake components by Convolutional Neural Networks (CNNs) in electro-optical satellite imagery. The selected approach to deep learning relies on a transfer learning procedure fine-tuning the ImageNet weights. This is performed through an ad-hoc developed dataset realized from Sentinel-2 multi-spectral images and Automatic Identification System (AIS) data in northern Europe. The experimental results confirm the robustness of the proposed method, which is tested against different spectral bands from visible to near-infrared and also by a domain shifting on lower resolution Landsat-9 images. Fractional errors in the positioning of the wake vertex are lower than 10% and the achieved heading accuracy is below 10°. The proposed method is faster than the traditional approaches based on Radon Transform, and due to its lightweight nature, our model can be executed efficiently on edge-AI devices, enabling real-time processing onboard.

Advancements in spaceborne edge computing have facilitated the incorporation of Artificial Intelligence (AI)-powered chips into CubeSats, enabling intelligent data handling and enhanced analytical capabilities with greater operational autonomy. This class of satellites faces stringent energy and memory constraints, necessitating lightweight models typically obtained via compression techniques. This paper addresses model compression through Neural Architecture Search (NAS), enabling computational efficiency and balancing accuracy, size, and latency. More specifically, we design an evolutionary-based NAS framework for onboard processing and evaluate it on both burned-area segmentation and classification tasks. The proposed solution jointly optimizes network architecture and deployment for hardware-specific, resource-constrained platforms, with hardware awareness embedded in the optimization loop to tailor network topologies to the target edge computing chip. The resulting models, designed on CubeSat-class hardware-namely the NVIDIA Jetson AGX \(\text {Orin}^{\text {TM}}\) and Intel® \(\text {Movidius}^{\text {TM}}\) Myriad\(^{\text {TM}}\) X-exhibit a memory footprint below 1 MB, achieving real-time, high-resolution inference in orbit while outperforming handcrafted baselines in terms of latency (3\(\times\) faster) and maintaining competitive mean Intersection over Union (mIoU). Furthermore, on the classification benchmark conducted on an NVIDIA A100-SXM, our approach attained an atthew Correlation Coefficient (MCC) of 0.974-substantially outperforming the baseline EfficientNet-lite0 (0.902) while achieving a \(\times\)47 speedup. These results highlight the framework’s scalability, enabling seamless deployment across the spectrum from resource-constrained edge devices to datacenter-grade accelerators, thereby supporting next-generation on-orbit computing architectures.