This PDF file contains the front matter associated with SPIE Proceedings Volume 13196, including the Title Page, Copyright information, Table of Contents, and Conference Committee information.

We propose a two-step spatial enhancement procedure for the two 100m thermal infra red (TIR) bands of LandSat 8/9, captured by its TIR spectrometer (TIRS), approached as a problem of fusion of heterogeneous data, or multimodal fusion. The fusion algorithm is guided by the statistical similarity between the TIR and visible and near infra red (VNIR) and short wave infra red (SWIR) bands provided at 30m by the operational land imager (OLI). In the first step, hyper-sharpening is applied from 100m scale to 30m scale (3:10 scale ratio): the two TIRS bands are spatially enhanced by means of two linear combination of the 30m VNIR+SWIR bands, devised to maximize the correlation with each thermal band at its native 100m scale. In the second step, the thermal bands, previously hyper-sharpened at 30m, are pansharpened through the 15m Panchromatic (PAN) band of OLI. The proposed approach is compared to plain 100m-to-15m pansharpening carried out uniquely by means of the Pan image of OLI. Both visual evaluations and statistical indexes measuring the radiometric and spatial consistency at the three scales are provided and discussed. The superiority of the two-step approach is undoubtedly highlighted.

Cloud detection is a crucial aspect of remote sensing with numerous research papers focusing on improving precision. Recent advancements have enhanced precision using extensive networks, albeit at the expense of longer processing times. Implementation of these networks often demands powerful graphics processing units. The University of the Bundeswehr Munich has been developing the ATHENE-1 Lower Earth Orbit (LEO) satellite for the Seamless Radio Access Network for Internet of Space (SeRANIS) mission with constrained onboard data storage and processing capacity. Therefore, implementing large-scale models with millions of parameters on ATHENE-1 is impractical and risks operational failure. We utilize open-source labeled data for pre-launch training, considering cross-platform performance challenges and the potential for optimized parameter configuration after launch. To address the two aspects of the satellite mission, we focus on implementing small-sized networks with fewer parameter configurations. Cloud detection is a fundamental step in the ATHENE-1 onboard processing pipeline of the remote sensing imagery. It determines the cloud percentage in an image and assists in decision-making regarding captured imagery. Images heavily obscured by clouds can be promptly discarded and remaining onboard images will be used to generate cloud masks. This processing step results in the optimal usage of onboard data storage and downlink passes required for the ATHENE-1 satellite. The research uses UNet architecture for cloud detection, analyzing how reducing model size and parameters affects performance.

Infrared (IR) imaging sensors designed to acquire the 0.9 to 14 micrometers wavelength band offer unique advantages over the daylight cameras for a multitude of consumer, industrial and defense applications. However, IR images lack natural color information and can be quite challenging to interpret without sensor specific training. As a result, transforming IR images into perceptually realistic color images is a valuable research problem with a substantial potential for commercial value. Recently, various research works that use deep neural networks to colorize single mode (near or thermal) infrared images have been reported. In this paper, we present a novel convolutional auto-encoder architecture that takes multiple images captured with different imaging modes (near IR, thermal IR and low-light) to perform colorization using the visual cues that exist in all imaging modes. We present visual results demonstrating that using multiple IR imaging modes improves the overall visual quality of the results.

The Medium Wavelength Infrared (MWIR) band is optimal for detecting high-intensity targets, e.g. fires. Infrared target detection and object recognition in the MWIR spectrum pose challenges for small satellites with resource constraints. The SeRANIS experimental satellite mission, under development at the University of the Bundeswehr Munich, features an MWIR sensor with a variable Field of View (FoV), but with a limited spatial resolution. This research aims to detect Hypersonic Glide Vehicle (HGV) traveling at speeds of approximately up to 7km/s in the upper atmosphere. The 1280*1024-pixel resolution offers a variable footprint ranging from 37km to 465km diagonally, facilitating the detection of intense heat signatures emitted by Hypersonic Glide Vehicles (HGVs). Combining the sensor’s spatial resolution, small target size, and high speed increases the complexity of developing detection methods. Additionally, the experimental satellite mission relies solely on a single MWIR band, increasing the complexity of precise target recognition. To address these concerns, this research proposes a three-part method: (a) hot spot detection, (b) computation and application of radiance threshold, and (c) filtering out static hot spots. However, the lack of a dataset for HGV detection poses a significant challenge for developing dedicated techniques. Therefore, this research introduces a strategy to create synthetic datasets, to replicate realistic sensor movement in orbit, inserting stationary targets with a range of radiance, and simulating moving targets with varying trajectories and radiance to validate the proposed methodology.

Earth observation (EO) plays a crucial role in creating and sustaining a resilient and prosperous society that has far reaching consequences for all life and the planet itself. Remote sensing platforms like satellites, airborne platforms, and more recently dones and UAVs are used for EO. They collect large amounts of data and this needs to be downlinked to Earth for further processing and analysis. Bottleneck for such high throughput acquisition is the downlink bandwidth. Data-centric solutions to image compression is required to address this deluge. In this work, semantic compression is studied through a compressed learning framework that utilizes only fast and sparse matrix-vector multiplication to encode the data. Camera noise and a communication channel are the considered sources of distortion. The complete semantic communication pipeline then consists of a learned low-complexity compression matrix that acts on the noisy camera output to generate onboard a vector of observations that is downlinked through a communication channel, processed through an unrolled network and then fed to a deep learning model performing the necessary downstream tasks; image classification is studied. Distortions are compensated by unrolling layers of NA-ALISTA with a wavelet sparsity prior. Decoding is thus a plug-n-play approach designed according to the camera/environment information and downstream task. The deep learning model for the downstream task is jointly fine-tuned with the compression matrix and the unrolled network through the loss function in an end-to-end fashion. It is shown that addition of a recovery loss along with the task dependent losses improves the downstream performance in noisy settings at low compression ratios.

There is a natural distortion effect that seen from long-range imaging systems occurs in hot airs causes to see defocusing images. It is known as heat haze, heat scintillation, mirage or atmospheric turbulence. In order to mitigate these effects, a set of steps are applied, and in an image fusion step, model-based restoration technique which is DT-CWT (Dual Tree Complex Wavelet Transform), with directional selectivity and shift-invariance properties, against other wavelet transforms, is used. In this work, both pixel-based and region-based image fusion methods are used for restoration. More than one distorted input was used in order to get one cleared image and then video result is also obtained from more than one output frame by shifting successive input frames. Both image and video results are compared with human-based visual system and FR (Full Reference)/ NR (Non-Reference) quality metrics. For comparison, the algorithm that bult-in our camera is also used with our pixel-based and region-based restoration methods for three different distances of scenes.

Hyperspectral imaging instruments can collect hundreds of spectral bands for the same area on the Earth surface which enables extracting extended fine information of the area of interest. Due to limited bandwidth from the satellite to Earth ground station, hyperspectral image classification and object detection on-board have gained significant interest due to their importance in the analysis and interpretation of the content of the image opening new applications based on hyperspectral images. The processing and analysis of images onboard reduces the data volume to be transmitted to the ground station and allows real-time decisions to be made onboard. In this paper, a high-performance embedded system is considered to run target detection in hyperspectral images. The work considers several deep-learning models and their deployment in a Jetson Orin Nano to run object detection. It takes into account the tradeoff between throughput, model complexity, and accuracy. The model is quantized to 8-bits to explore the utilization of INT8 operations of the Jetson Orin GPU, to achieve real-time performance.

Accurate identification of building footprints from high-resolution satellite imagery is crucial for urban planning and disaster response. This paper investigates building detection methodologies using the Mask R-CNN framework and its variants, aiming to address challenges such as accurate boundary pixel classification and reducing false positives. Two WorldView-3 datasets, including the SpaceNet Building Detection Dataset and a dataset on Prato, Italy, are utilized for analysis. Augmentation techniques, such as NDVI and Sobel edge detection features, and evaluation metrics such as F1-score and Average Precision are employed to assess model performance. Findings reveal the superiority of the Point Rend Mask R-CNN in detecting medium and large buildings in densely populated urban environments. Notably, Point Rend and the use of NDVI and Sobel demonstrate substantial improvements compared to other methods for building detection. This investigation provides insights into the efficacy of Mask R-CNN framework and its variants for advancing building footprint delineation across various applications.

Advancements in satellite missions have dramatically improved the monitoring of vegetation and agricultural activities through high-resolution Satellite Image Time Series (SITS), providing enhanced insights into crop dynamics and boundary identification. However, traditional UNet-based Convolutional Neural Networks (CNNs), though effective for crop mapping, often struggle to capture the full spatio-temporal complexities inherent in these datasets, particularly when it comes to detecting less distinct boundaries. To address these challenges, a novel attention-based residual 3D UNet architecture has been developed, incorporating a spatial-temporal attention mechanism that enhances the networks ability to represent spatial and temporal features. This attention mechanism is strategically implemented in the decoder, where it gathers information from both the encoder and the previous layer within the decoder. This dual-source integration allows the model to focus more effectively on relevant crop boundaries during training, assigning greater weight to these crucial areas while reducing the emphasis on non-crop regions. The residual 3D UNet architecture adeptly handles the intricate spatial-spectral-temporal correlations present in SITS, enabling more accurate and simultaneous modelling of both spatial and temporal information. The proposed method is evaluated on an area with small-scale crop fields in Germany using Sentinel-2 SITS data collected over several months, this approach demonstrated superior performance in boundary detection compared to existing state-of-the-art methods, particularly in scenarios where boundaries are less clearly defined.

To date, careful data treatment workflows and statistical detectors are used to perform hyperspectral image (HSI) detection of any gas contained in a spectral library which is often expanded with physics-models to incorporate different spectral characteristics. Generally, surrounding evidence or known gas-release parameters are used to provide confidence in or confirm detection capability, respectively. This makes quantifying detection performance difficult as it is nearly impossible to develop absolute ground truth for gas target pixel presence in collected HSI. Consequently, developing and comparing new detection methods, especially machine learning (ML) based methods, is beholden to subjectivity in derived detection map quality. In this work, we demonstrate the first use of transformer-based paired neural networks (PNNs) for one-shot gas target detection for multiple gases while providing quantitative classification and detection metrics for their use on labeled data. Terabytes of training data are generated from a database of long-wave infrared (LWIR) HSI obtained from historical Mako sensor campaigns over Los Angles. By incorporating labels, singular signature representations, and a model development pipeline, we can tune & select PNNs to detect multiple gas targets which are not seen in training on a quantitative basis. We additionally assess our test set detections using interpretability techniques widely employed for ML-based predictors, but less common on detection methods relying on learned latent spaces.

Radar sounders are active sensors that operate by transmitting electromagnetic waves toward the subsurface of a target area and capturing the reflected signals. The reflected signals are then processed to create images or maps of the target subsurface. Supervised deep learning techniques have emerged as powerful tools for radar sounder data analysis. However, they require a significant amount of labeled data, which is challenging given both the difficulty of acquiring such data in specific subsurface environments and the cost of labeling them with complex photointerpretation procedures. Recently, some methods have been proposed to segment the radargrams by employing deep complex models or pretrained networks. However, these methods may lead to models that are too complex for the problem and computationally inefficient. Thus, we present a computationally efficient u2net model that combines u2net and octave convolution. This combination offers the advantage of a deep efficient architecture with rich multi-scale features while keeps computational and memory requirements relatively low. The results show the efficiency and adaptability of our model to the availability of limited labeled data and its generalization capabilities when augmented with additional data. Furthermore, our proposed model significantly reduces the number of parameters used compared to existing methods for radar sounder segmentation.

Various natural disasters occur on the earth. In Japan, heavy rains and earthquakes have caused particularly severe damage. We focus on landslides caused by them. This study proposes a landslide detection method using synthetic aperture radar (SAR). SAR uses microwave observations, and microwaves are reflected according to the properties of materials on the earth’s surface. In addition, microwave amplitude and phase information can be obtained, and these are used for various analyses. They are often used to detect disasters, mostly to detect changes caused by disasters. For example, change detection by differential reflection intensity, analysis of terrain variation by phase difference, and detection of material by properties of polarization. Therefore, multiple SAR data are required for disaster detection. However, in the event of a disaster, rapid detection of the damaged area is necessary. For this reason, this study investigates a method for detecting the damaged area from a single SAR data. As a research method, instance segmentation is conducted using YOLOv8. The SAR data used in the experiments were obtained for the Noto Peninsula earthquake. This disaster occurred on January 1st in 2024 in the Noto region of Ishikawa Prefecture and caused extensive damage. Images of landslide areas were obtained from SAR data, annotated and trained instance segmentation by YOLOv8 to evaluate test performance.

The introduction of new generation hyperspectral satellite sensors, combined with advancements in deep learning methodologies, has significantly enhanced the ability to discriminate detailed land-cover classes at medium-large scales. However, a significant challenge in deep learning methods is the risk of overfitting when training networks with small labeled datasets. In this work, we propose a data augmentation technique that leverages a guided diffusion model. To effectively train the model with a limited number of labeled samples and to capture complex patterns in the data, we implement a lightweight transformer network. Additionally, we introduce a modified weighted loss function and an optimized cosine variance scheduler, which facilitate fast and effective training on small datasets. We evaluate the effectiveness of the proposed method on a forest classification task with 10 different forest types using hyperspectral images acquired by the PRISMA satellite. The results demonstrate that the proposed method outperforms other data augmentation techniques in both average and weighted average accuracy. The effectiveness of the method is further highlighted by the stable training behavior of the model, which addresses a common limitation in the practical application of deep generative models for data augmentation.

Change Detection (CD) approaches for hyperspectral images (HSI) are mainly unsupervised and hierarchically extract the endmembers to determine the multiple change classes but require many parameters to set manually. Recently, HSI CD has been approached with DL methods because of their capacity to learn features of changes automatically, but they require a huge amount of labeled data for weakly or fully supervised training. They mostly perform binary CD only and do not fully exploit the spectral information. Accordingly, we propose an unsupervised DL CD method to identify multiple change classes in bi-temporal HSIs, inspired by a sparse autoencoder for spectral unmixing. The proposed method learns the endmembers of the unchanged class and the various classes of change by solving an unmixing problem with a Convolutional Autoencoder (CAE) trained in an unsupervised way using unlabeled patches sampled from the difference of the bi-temporal HSIs. The spectral unmixing problem is solved by applying three constraints to the CAE: a sparsity l₂₁-norm constraint that forces the model to learn non-redundant information, a non-negativity constraint, and the sum-to-one constraint. After the training, we process the difference image with the trained Autoencoder to extract the abundance maps of the various change types being derived from the endmembers learned by the model during the training. A Change Vector Analysis approach detects the changed areas that are clustered with an X-means approach using the change abundances to obtain a multi-class change map. We obtained promising results by testing the proposed method on bi-temporal Hyperion images acquired on Benton County, Washington, USA, in May 2004 and May 2007, and bi-temporal PRISMA images acquired on an area close to Vienna in April 2020 and September 2021 that show the changes in crop fields.

With the drastic increase of Remote Sensing payloads both in terms of satellite count and advancements of sensing technologies and resolution, automatic matching and registration of Synthetic Aperture Radar (SAR)-Optical images for mission-critical applications with time efficiency and high accuracy is a must. Optical and SAR sensors utilise distinct imaging techniques. As a result, the process of matching images from these two modalities is not only less precise but also requires a significant amount of time. In this study, we introduce an innovative algorithm that combines deep learning methods with the traditional Gabor Jet Model. This is achieved by employing cross entropy loss function to calculate the anticipated shift. The encoder-decoder structure is used to map non-linear dependencies and maintain both local and global data. Current state-of-the-art (SoTA) methods focus on either the spatial or temporal domain. However, our approach integrates both the spatial and temporal domains, preserving both global and local characteristics for comparison. Experimental results show that our proposed algorithm attains pixel-level precision and surpasses the current SoTA methods.

In recent years, water-related accidents caused by torrential rain have been occurring frequently. Visual search for persons requiring rescue is challenging from coast or riverbank. Due to water currents and underwater topography, search from boat is also difficult. This research aims to develop a safe, wide area and accurate target search method using point cloud data from drone. The authors focused on a LiDAR system called Airborne Laser Bathymetry (ALB) which is specialized for underwater observation. A green laser ALB, In particular, has capability to obtain underwater topography data because it is equipped with not only near-infrared laser used in conventional land surveying but also green visible laser for observing in relatively shallow water. The purpose of this study is to make it possible to identify the water surface, underwater topography, and underwater floating objects such as algae from green laser ALB point cloud data using machine learning methods. For machine learning, I use Pointnet⁺⁺, a network effective for point cloud processing, and SVM (Support Vector Machine), specialized for two class classification. The Pointnet⁺⁺ addresses the limitations of the previously used Pointnet by sampling local features based on point cloud distance and density for learning. In proposed method, Pointnet⁺⁺ is used to input three-dimensional coordinates X, Y and Z and extract three classes: water surface, underwater topography, and floating objects. Then, by inputting the Z-axis coordinate data and backscatter data (Intensity) into the SVM, it becomes possible to detect persons requiring rescue from among the floating objects.

Hyperspectral images are constantly improving their spectral resolution and spatial resolution，so utilizing hyperspectral images for object detection has become a research hotspot in the field of hyperspectral remote sensing. Anomaly detection is the main way to achieve object detection. Abnormal targets in hyperspectral images are usually composed of a few pixels (or even sub-pixel) that are clearly different from the surrounding background pixels. Compared with the background, abnormal targets have two characteristics: spectral anomaly and spatial anomaly. Traditional hyperspectral image anomaly detection methods only utilize spectral anomalies and ignore spatial anomalies between pixels. Hyperspectral images can be represented by third-order tensors, where the first two orders of the third-order tensor are used to represent the spatial dimension of the image (i.e. the height and width of the image), and the third order is used to represent the spectral dimension. Therefore, tensor decomposition can simultaneously represent the spatial and spectral features of anomalous targets. This paper proposes a new anomaly detection method based on tensor decomposition and information entropy. This method is mainly divided into three steps. Firstly, a third-order tensor is used to represent the cube of the detected hyperspectral image, and the Tucker decomposition of the third-order tensor is applied to the detected hyperspectral image. Secondly, the background information in the detected hyperspectral image is removed using information entropy, and the remaining feature components are reconstructed into the hyperspectral image. Thirdly, the RX algorithm is used to detect anomalies in the reconstructed hyperspectral image. Compared with methods based on spectral anomalies, this method has better detection efficiency.

Acceleration of image segmentation using deep learning methods on satellite imagery has become ubiquitous in various applications areas such as land cover classification, disaster monitoring and vegetation detection. However, the increase in satellite image resolution and large data volume required for remote sensing applications has resulted in a substantial increase in computational resource usage and demand for real-time processing. This paper investigates quantum computing as a novel approach to meet these computational demands, exploiting its parallel processing strengths. We evaluate hybrid quantum models (COQCNN, MQCNN, FQCNN) against classical CNN and U-Net architectures in remote sensing classification. Although COQCNN and MQCNN underperformed, FQCNN reached 53.26% accuracy, outperforming the classical CNN by 8%. Despite quicker convergence, quantum models struggle with complex feature segmentation, a task where U-Net excels. This study highlights quantum convolutions as a potential path to enhance convergence while addressing challenges like noise from multiple quantum channels affecting accuracy.

Image geo-localization estimates an image's global position by comparing it with a large-scale image database containing known positions. This localization technology can serve as an alternative positioning method for unmanned aerial vehicles (UAV) in situations where a global position system is unavailable. Feature-based image-matching methods typically involve descriptors constructed from pixel-level key points in the images. The number of descriptors in one image can be substantial. Filtering and comparing these large quantities of descriptors for image matching would be quite time-consuming. Due to the large scale of satellite images, matching them with aerial images using this method can be challenging to achieve in real-time. Thus, this paper proposes a semantic matching-based approach for real-time image geo-localization. The types, quantities, and geometric information of objects in satellite images are extracted and used as sematic-level descriptors. The sematic-level descriptors of an aerial image captured by UAV are extracted by an object recognition model. The quantity of semantic-level descriptors is orders of magnitude less than pixel-level descriptors. The location of the aerial image can be rapidly determined by matching the semantic-level descriptors between the aerial image and satellite images. In the experiments, the speeds of matching an aerial image with satellite images using the semantic matching and a feature-based matching method were 0.194 seconds per image and 125.68 seconds per image, respectively. Using semantic matching methods is 648 times faster than using feature matching methods. The results demonstrate that the proposed semantic matching methods have the potential for real-time image geo-localization.

The Mediterranean agriculture sector is vulnerable to abiotic and biotic stresses of diverse nature, with wheat (Triticum aestivum. L.) being by area extension one of the most important crops. In this sense, it is especially important to identify the most appropriate tools for monitoring the impact of drought and fungal diseases on wheat yield. The purpose of our research is to develop a prediction model for crop monitoring and phenotyping purposes using RGB Vegetation Indexes (VIs), through ground-acquired RGB images. The current study assessed forty advanced winter wheat accessions at four experimental sites in NE Spain: Briviesca, Ejea de los Caballeros, Sos del Rey Católico and Tordómar, receiving using 15 VIs computed from RGB (Red-Green-Blue) bands. The sites received full/moderate irrigation support or rainfed only. For each date and treatment data subset, VIs or their associated Principal Components (PC) was analyzed. In light of this, an interesting range separation between the treated and untreated groups at several different sites was observed. Furthermore, the use of sprinkler irrigation resulted in minor fungal pressure due to its lower likelihood of fungal dispersion. As a conclusion, we realized that the variations in RGB VIs over the growing season can be used as user-friendly, time-efficient and cost-effective tools to distinguish different growth stages/phenologies and largely for disease prediction in order to assess fungicide treatment efficacy among wheat varieties, across different experimental sites. Moreover, differences in weather and other site differences, such as irrigation method or a lack of, seem to have impacts on fungicide pressure.

We introduce a novel method for updating 3D geospatial models, specifically targeting occlusion removal in large-scale maritime environments. Traditional 3D reconstruction techniques often face problems with dynamic objects, like cars or vessels, that obscure the true environment, leading to inaccurate models or requiring extensive manual editing. Our approach leverages deep learning techniques, including instance segmentation and generative inpainting, to directly modify both the texture and geometry of 3D meshes without the need for costly reprocessing. By selectively targeting occluding objects and preserving static elements, the method enhances both geometric and visual accuracy. This approach not only preserves structural and textural details of map data but also maintains compatibility with current geospatial standards, ensuring robust performance across diverse datasets. The results demonstrate significant improvements in 3D model fidelity, making this method highly applicable for maritime situational awareness and the dynamic display of auxiliary information.

Monitoring uncooperative vessels without transponders is of strategical interest both for the civil and military world. Ship detection lacks the ability to discern full situational knowledge of the vessel. However, moving ships generate wakes containing significant information – from current and possibly past position, heading, and speed, to vessel size and hull class. UEIKAP (Unveil and Explore the In-depth Knowledge of earth observation data for maritime Applications) is a project founded by the Italian Ministry of University and Research, and its objective is to develop a deep learning-based solution for wake detection in optical and synthetic aperture radar (SAR) spaceborne remote imagery. A dataset of real and simulated imagery is under development and will be used to train a landmark-based detection model able to exploit the characteristic features of ship wakes. This is accompanied by an in-depth sea characterization and meteo-marine conditions study, which is used to properly discriminate sea surface clutter for the objects of interest. All the results will be validated by test campaign at sea. This manuscript goes over the different types of data used to obtain the aforementioned contextual knowledge for project UEIKAP, from Automatic Identification System (AIS) data providers to sources of local meteo-marine information. Indications are provided regarding the integration of these inomogeneous data sources with the deep learning-based wake detection architecture. Information on the methods of the first data gathering campaign, held in July 2024 in Venice, is provided, accompanied by observations and preliminary results gathered from the experience.

Radar Sounder (RS) data contain information on subsurface geology and are analyzed mostly with automatic techniques on single-pass acquisitions. A few preliminary studies on multitemporal RS data acquired on the cryosphere focus on possible advances in ad-hoc data acquisition strategy and ice-sheet monitoring, such as the percolation zone. However, challenges related to data corregistration and the inherent characteristics of the target hinder the multitemporal analysis. This paper analyzes bi-temporal radargrams with partially overlapping footprints (thus showing information from neighboring geographical areas) in the cryosphere and defines a strategy to estimate candidate changes. The paper proposes projecting and locally corregistering the radargram pairs at different depths to identify the expected changes due to glacier displacement and snow accumulation. Comparing the corregistered radargrams, we identify candidate unexpected changes in the ice sheet morphology that glaciologists should further validate. The proposed method is validated on several radargram pairs acquired by MCoRDS-3 in Antarctica in 2014 and 2016.

This article presents an elementary change detection algorithm designed using a synchronous model of computation (MoC) aiming at efficient implementations on parallel architectures. The change detection method is based on a 2D-first-order autoregressive ([2D-AR(1)]) recursion that predicts one-lag changes over bitemporal signals, followed by a high-parallelized spatial filtering for neighborhood training, and an estimated quantile function to detect anomalies. The proposed method uses a model-based on the functional language paradigm and a well-defined MoC, potentially enabling energy and runtime optimizations with deterministic data parallelism over multicore, GPU, or FPGA architectures. Experimental results over the bitemporal CARABAS-II SAR UWB dataset are evaluated using the synchronous MoC implementation, achieving gains in detection and hardware performance compared to a closed-form and well-known complexity model over the generalized likelihood ratio test (GLRT). In addition, since the one-lag AR(1) is a Markov process, its extension for a Markov chain in multitemporal (n-lags) analysis is applicable, potentially improving the detection performance still subject to high-parallelized structures.

SAR image interpretation requires expertise. A lot of SAR to optical research studies have been conducted to assist SAR image interpretation. However, existing studies have arbitrarily selected the polarizations of SAR data to be translated. This study explores SAR polarization combinations for Very-High-Resolution (VHR) SAR to optical conversion. Through comprehensive quantitative and qualitative analysis of experiments and results, we find the optimal SAR polarization combination for each SAR configuration. We used multiple image quality metrics including PSNR, SSIM, LPIPS, and FID for thorough quantitative analysis. We also propose a modified conditional Brownian Bridge Diffusion Model (BBDM) that can leverage all four polarizations of Quad polarization configured SAR. This study provides evidence for which polarization combinations should be selected in future SAR to optical research according to the polarization configuration of the SAR system.

An innovative deep learning-based solution to address motion blur in radar images, resulting from radar platform or target movement, is presented in this work. Leveraging Convolutional Neural Network (CNN), the proposed method learns a mapping from blurred to deblurred images, while a separate CNN estimates the point spread function (PSF) of the motion blur. This estimated PSF is then used to reconstruct deblurred images, optimising the reconstruction process by integrating the input image, estimated PSF, and ground truth relationship into the training loss term. Trained on a comprehensive dataset of simulated blurred and deblurred radar images, generated from a numerical imaging model, the model exhibits exceptional performance, outperforming state-of- the-art methods across varying degrees and lengths of blur. Specifically, testing on 6,410 images yields mean squared error (MSE) and structural similarity index (SSIM) scores of 0.0086 and 0.9398, respectively. Additionally, validation on experimental measurements showcases promising results. This comprehensive evaluation underscores the effectiveness and versatility of the proposed approach, offering significant advancements in radar image processing for various applications such as target detection, recognition, surveillance, and navigation.

Ground deformation can be detected by processing SAR (Synthetic Aperture Radar) phase data acquired in different periods. However, due to the characteristics of SAR, it is difficult to determine the direction of ground deformation as the distance change between the satellite and the ground surface is observed. Therefore, on-site field observation is required since SAR observation results differ from the actual amount of ground deformation. This study aims to estimate ground deformation over a wide area using satellite SAR data, understand the disaster situation quickly, and reduce secondary damage risks caused by on-site field observation. In this paper, Interferometric SAR (InSAR) analysis is applied to estimate ground deformation caused by Kumamoto earthquake in 2016 from C-band SAR data on Sentinel-1 satellite. 2.5-dimensional analysis is conducted by combining the InSAR analysis results of the ascending and descending orbits, and the direction of ground deformation caused by earthquake is visualized using displacement vectors. Furthermore, changes in land cover classification, which classifies land based on surface vegetation and geology is performed by using time-series analysis based on machine learning techniques from optical sensor images obtained from Sentinel-2. The results show that the accurate understanding of the damage situation over a wide area is very effective in terms of estimating landslides and speeding up disaster response, such as evacuation.

Non-contact and non-invasive target detection in scattering media is a crucial task for advancements in biomedical imaging and environmental measurements. A novel sensing approach which combination of Michelson interferometer and ghost imaging (MIGI) is designed and developed for reconstructing the real image of target sample in scattering media. The limitations of optical interferometry, such as the need of scanning time for two-dimension measurements, and the inability of ghost imaging to capture cross-sectional images, can be effectively mitigated through their combination. However, this technique necessitates numerous illuminations of random structured light in ghost imaging to reconstruct the image of the target with high quality, thereby extending the measurement time. To address this issue and facilitate real-time measurement in MIGI, we reconstructed the real image close to the target sample being measured by passing degraded images obtained through short-term measurement through deep learning. This approach significantly reduces the number of measurements required to obtain a clear image in the simulation by 90%. In practical experiments, the number of measurements needed to achieve an equivalent structural similarity index method (SSIM) value is reduced. This paper discusses analysis of measurement time and SSIM based on the values of the dataset used for model construction.

In the framework of the Horizon-Europe project “Instantaneous Infrastructure Monitoring by Earth Observation (IIMEO)” the objective is to design, implement and demonstrate key technological factors of a future satellite-based Earth Observation (EO) system capable of providing functions necessary for instantaneous monitoring of infrastructures in near real time. The system will implement a tiled acquisition of multitemporal SAR images over a railway infrastructure and perform near real-time change-obstacle detection at every new acquisition within one hour after the satellite passes over the area. The tile-based obstacle change-detection multitemporal system is explained in detail.

Mediterranean regions are characterized by increasing fire frequency and severity during recent years, threatening ecosystems, infrastructures, and human wellbeing. Greece has been particularly affected with devastating casualties including environmental and socioeconomic losses. Here, we examined five distinct wildfires that took place in Greece during summers 2021 and 2023 with the overall objective to delineate the burned area and quantify possible built-up losses. More specifically, we focused on fire events that occurred in Northern Peloponnese, the Island of Evia, and Eastern Attica during summer 2021, and the recent wildfires at the Rhodes Island and the region of Evros during summer 2023. These regions cover different topographic, land cover and bioclimatic characteristics with also various built-up densities and wildland-urban interfaces. Optical remote sensing observations from the Copernicus Sentinel 2 and the Landsat satellite missions were used to calculate spectral indices such as the Normalized Burn Ratio (NBR) and the dNBR (i.e., the difference between pre- and post-fire NBR) and estimate the total burned area and burn severity. Specific urban features that were affected (i.e., roads, building, etc.) were estimated using auxiliary geospatial information from two openly available datasets (Open Street Maps and Microsoft Building Footprint). Exploring the increasingly available satellite imagery offers novel insights into several natural hazards, including wildfires, providing timely estimates of possible infrastructure losses and supporting decision making.

Laser speckle imaging techniques have become widespread in many areas where non-invasive remote measurements are needed. For example, medical and microbiological fields. This technique is applicable for monitoring the behavioral activity of microorganisms. In the current study, using experiments with fungi and bacteria, we compare signal and image processing algorithms for analyzing microorganism’s activity by laser speckle imaging techniques and demonstrate the advantages of the proposed method: sensitive sub-pixel correlation algorithm. The obtained results could allow to propose a technology for faster detection of bacterial and fungal growth in the culture medium. They could also be used to speed up the determination of antibacterial and antifungal susceptibility results.