This PDF file contains the front matter associated with SPIE Proceedings Volume 12267, including the Title Page, Copyright information, Table of Contents, and Conference Committee Page.

With the continuous improvement of the number and capability of micro-nano satellites, on-board intelligent data processing becomes a necessary configuration. The constellation with hundreds micro-nano satellites has the ability to high-frequency detection of ship targets and realizes continuous awareness of the global ocean, which is of great significance in maritime rescue, waterway management and combating illegal fishing. In this paper, a fast on-board ship detection method for panchromatic image is proposed. Firstly, GPU (graphics processing unit) of commercial devices is used to form high performance and low power computing capability on the micro-nano satellite. Then, according to the characteristics of ship targets, a convolutional neural network based on lightweight model is designed to quickly obtain accurate number and location information of ship targets. The algorithm deployed on micro-nano satellite can transform massive remote sensing data into target slices, greatly reduce the pressure of satellite-ground data transmission and improve the application efficiency of remote sensing data. We test our method on a dataset of more than 90 panchromatic images. The results show that the detection rate of this algorithm is better than 0.95, and the average processing speed for an image block of 1024 × 1024 pixel is less than 0.2 seconds, which has a wide application prospect.

Satellite image positioning and registration is a key technology for geostationary Earth observation satellites. Our work can provide powerful prior information for the positioning and registration of Earth observation images. We simulate the process of geostationary orbit satellites observing the space region with discrete detector array, generate star observation sequence images considering various situation, and study the high precision subpixel star centroid extraction algorithm. Under the condition that the star trajectory passes through the pixel, the proposed algorithm achieves the extraction error of less than 0.5 pixels.

Near-space remote sensing image registration is an important foundation of near-space image processing. For large image jitter distortion, geometric and atmospheric distortion of its image, we propose a two-step method based on deep neural networks, which includes a coarse-to-fine registration process. We construct a near-space image registration dataset which is captured from Google Maps and hot air balloon platforms, etc. For obtaining candidates, the coarse alignment stage applies classical geometric validation methods to a corresponding set of pre-trained deep features. The fine alignment network is based on pyramidal feature extraction and optical flow estimation to realize local flow field inference from coarse to fine. We construct a regularization layer for each level to ensure smoothness. Applying our method to our synthetic dataset, the experimental result shows that it has a competitive result that is evaluated based on the root mean square error, peak signal to noise ratio and structural similarity.

This article investigates basic preprocessing techniques to improve classification accuracy in the context of Automatic Target Recognition (ATR) of non-cooperative targets in Synthetic Aperture Radar (SAR) images. Preprocessing techniques are considered in synthetic data providing different inputs to a model-based classification algorithm. Experiments with preprocessing techniques such as area reduction, morphological transformations, and speckle filtering were run using ten target classes of the SAMPLE dataset. The classification is performed in measure data using scattering centers as features. The results reveal that the original image without any preprocessing techniques reached the best classification performance. However, investigations with other classifiers that use different features may benefit from such preprocessing techniques.

The presence of noise, displacement of points, and empty spots in a raw Light Detection and Ranging (LiDAR) point cloud are common phenomena caused by reflective surfaces or objects. Typical approaches to solve this problem are either avoid or cover the reflective areas or to manually remove the erroneous data in post processing. This can help clean the point cloud structure but will cause sparsity issues. To combat this, in this paper, we introduce a two-step process to perform point cloud restoration. Instead of removing noise, this approach can restore the points to the closest surface which they may belong to. Next, to fill out empty spots, we introduce a technique called point cloud inpainting, which involves interpolating points in 2D then mapping it back to 3D for flat surfaces. The point cloud then becomes more photorealistic and easier to use for other computer vision tasks.

In this article we present a combination of marked point processes with convolutional neural networks applied to remote sensing. While point processes allow modeling interactions between objects via priors, classical methods rely on contrast measures that become unreliable as objects of interest and context become more diverse. We propose learning likelihood measures using convolutional neural networks to make these measures more versatile and resilient. We apply our method to the detection of vehicles in satellite images.

We are studying in-orbit real-time object detection for remote sensing satellites. Due to the small object size of remote sensing images, it is hard to achieve high detection accuracy, especially for resource-constrained spacecraft computers. Lightweight object detection models such as YOLO and SSD are feasible choices to achieve acceptable detection speed on board. This study proposes an accuracy-improvement method for the lightweight neural networks with an upscaling ratio estimator without retraining the model. The estimator exploits a scaling ratio that determines how much the image should be resized. With our scaling estimator, we have achieved 10.09% higher accuracy than the original YOLOv4-Tiny models with a 40% detection speed overhead.

This paper proposes a novel coarse-to-fine ice-block falls detection approach based on the YOLO-V4 network and a postprocessing strategy by considering the illumination properties (i.e., adjacent distance, direction, area ratio of ice-block and shadows) of the considered ice-block targets. The proposed approach mainly consists of two steps: 1) Coarse detection of ice-block falls based on the YOLO-V4 network. 2) Extraction of the illumination properties of ice-block targets, and refine the initial detection results based on the post-processing strategy. By taking the edge of the Boreum Planum in Mars Arctic as a research region where presents frequent ice-block falls activity, the HiRISE (High Resolution Imaging Science Experiment) image was used to verify the reliability of the proposed approach. Note that in this work we only focused on the ice-block targets whose length and width are larger than 0.5m (2 pixels in the HiRISE image). Final obtained experimental results confirmed the effectiveness of the proposed approach for identifying the ice-block falls activity over large Martian areas at both local and global scales.

Storing and processing Remote Sensing (RS) images require large amounts of memory space and computing resources. Consequently, RS images are compressed and stored in various compression formats, such as JPEG2000. However, the processing of RS images for machine interpretation and understanding still necessitates the deployment of an image decompression stage in its entirety, followed by a computationally demanding image analysis pipeline. The image analysis stage is commonly composed of machine learning techniques, such as Deep Convolutional Neural Network (DCNN) models. Classification of remote sensing images is among the most common image analysis tasks. In the scope of this paper, we propose a sub-band image based classification method for the Remote Sensing Scene Classification (RSSC) task in the JPEG2000 compressed domain. The proposed approach exploits the already available sub-band image coefficients to classify RS images without needing for full decompression. Our study shows that our method increases the high frequency information in the LL sub-band and allows the image to contain more detail, leading to improved classifier performance while taking advantage of the partial decompression method.

Object detection from remote sensing images has been performed on the ground. Recently, on-board object detection has been studied only to show its feasibility with single-stage detectors. However, highly accurate models such as two stage detectors are compute intensive so that they are too slow to run on power-constrained on-board computers. In this paper, we propose a speed-up method for two-stage detectors. Two-stage detectors extract features and ROIs(Region of Interest) in the first stage and then classify them at the second stage. This structure gives high accuracy but induces large inference latency. In remote sensing images from satellites, object size is small relative to the whole image. Based on this characteristic, we propose to exclude features related to the large objects in the first stage. To verify our concept, we have selected various R-CNN models as two-stage object detectors. We have implemented our methods on two NVIDIA Jetson boards. We have achieved 1.8x speed up in inference latency with 5% accuracy drop with the small object dataset.

The Bulgarian coastal zone of the Black Sea is known with large number of active landslides. The factors that contribute to their activations are both endogenous and exogenous. The goal of this paper is to summarize the efforts made in researching the temporal behavior of several landslides located in the NE Bulgaria region using data from synthetic aperture radar (SAR) processed by the differential interferometric SAR (DInSAR) method. In it presented are results for monitoring several active landslides in the area of circus Dalgya Yar based on the mentioned data and technology. The results are compared with displacements in GNSS points that are located in geodetic network around landslides circus.

The number of forest fires is growing exponentially with globalization negative impacts and industry evolution. The firefighters are unable to attend fire sources in the desired elapse time. Hence a huge number of forests are destroyed yearly. The statics demonstrate horrible prediction in a time interval of less than ten years. Necessary action and evolution plans must be established to save the globe from an invasive destruction due to the disappear of green areas and consequent disequilibrating ecosystem effects. The obvious idea is to take advantage of current evolution in informatic systems and robotic field, to develop a distance controllable device to scan areas classified as high risk in the vulnerable season (hot season). The first step is to design a machine learning accurate approach to detect fire area on pictures acquired by probable drone or intelligent systems, responsible of the scanning task. Through literature, several approaches were developed treating pictures that are more with afront view of the flames. Training a machine learning algorithm with such pictures with huge areas of flames is feasible. Nonetheless, treating aerial images is not a very easy approach. A deep analysis of the chosen feature engineering technique and machine learning model is required. The current paper accesses the performance of wavelet-based feature extraction technique within different traditional clustering techniques and ranking methods. The results were accessed using different metrics, to show the effectiveness of the approach, namely sensitivity specificity, precision, recall, f-measure, and g-mean.

In this paper, we propose a processing chain jointly employing Sentinel-1 and Sentinel-2 data aimed at monitoring changes in the status of the vegetation cover by integrating the four visible and near infrared (VNIR) bands with the three red-edge (RE) bands of Sentinel-2, approximately spanning the gap between Red and NIR bands (700 nm – 800 nm) with bandwidths of 15/20 nm and 20 m pixel spacing. The RE bands will be sharpened to 10 m and the resulting 7-bands, 10 m fusion product will be integrated with polarimetric features calculated from the Interferometric Wide (IW) Ground Range Detected (GRD) product of Sentinel-1, available at 10 m pixel spacing. Key point of the fusion of optical bands is the correction of atmospheric path radiance before fusion is accomplished through modulation of the interpolated band by a sharpening term achieved through the hyper-sharpening paradigm. Whenever surface reflectance data are available, haze estimation and correction can be skipped. Hyper-sharpening of Sentinel-2 multispectral (MS) bands and modulation-based integration of Sentinel-1 polarimetric synthetic aperture radar (SAR) features are applied on a multitemporal dataset acquired before and after a recent fire event.

Many Change Detection (CD) methods exploit the bi-temporal multi-modal data derived by multiple sensors to find the changes effectively. State-of-the-Art CD methods define features with a common domain between the multi-modal data by normalizing input images or ad hoc feature extraction/selection methods. Deep Learning (DL) CD methods automatically learn features with a common domain during the training or adapt the features derived by multi-modal data. However, CD methods focusing on multi-sensor multi-frequency SAR data are still poorly investigated. We propose a DL CD method that exploits a Cycle Generative Adversarial Network (CycleGAN) to automatically learn and extract multi-scale feature maps in a domain common to the input multifrequency multi-sensor SAR data. The feature maps are learned, during unsupervised training, by generators that aim to transform the input data domain into the target one while preserving the semantic information and aligning the feature domain. We process the multi-sensor multi-frequency SAR data with the trained generators to produce bi-temporal multi-scale feature maps that are compared to enhance changes. A standard-deviation-based feature selection is applied to keep only the most informative comparisons and reject the ones with poor change information. The multi-scale comparisons are used for a detail preserving CD. Preliminary experimental results conducted on bi-temporal SAR data acquired by Cosmo-SkyMed and SAOCOM on the urban area of Milan, Italy, in January 2020 and August 2021 provided promising results.

Comparing with the multispectral remote sensing image, hyperspectral image (HSI) has higher spectral resolution, a near continuous spectral signature, thus can represent fine spectral variations that occurred in the temporal domain. This allows more spectral changes to be detected, especially major changes that reflected on the overall spectral signature (associating with the abrupt land-cover transitions), as well as subtle changes that reflect only on a portion of the spectral signature (associating with the change of physicochemical properties of the land-cover classes). Currently, there are some available hyperspectral change detection (CD) data sets. However, they have the following drawbacks. First, there is a lack of diversity in the data source; all data sets were created using the Hyperion sensor mounted on the EO-1 satellite. Second, these data sets mainly concentrate on the river and agriculture scenes, which lose their diversity for representing different land-covers. In this paper, we construct three new change detection data sets by using the multitemporal images acquired by the China’s new generation of hyperspectral satellites, i.e., OHS, GF-5 and ZY1-02D. These data sets present various event-driven land-cover changes, such as new building construction, crop replacements, and the expansion of energy facilities. Then a novel unsupervised hyperspectral change detection approach is proposed based on the intrinsic image decomposition (IID). Experimental results confirmed the effectiveness of the proposed approach in terms of higher overall accuracy by comparing with the reference techniques.

This work addresses the problem of hyperspectral data compression and the evaluation of the reconstruction quality for different compression rates. Data compression is intended to transmit the enormous amount of data created by hyperspectral sensors efficiently. The information loss due to the compression process is evaluated by the complex task of spectral unmixing. We propose an improved 1D-Convolutional Autoencoder architecture with different compression rates for lossy hyperspectral data compression. Furthermore, we evaluate the reconstruction by applying metrics such as SNR and SA and compare them to the spectral unmixing results.

Hyperspectral image classification is one of the most researched topics within hyperspectral analysis. Its importance is determined by its immediate outcome, a classified image used for planning and decision-making processes within a variety of engineering and scientific disciplines. Within the last few years, researchers have solved this task employing self-supervised learning to learn robust feature representations to alleviate the dependency on large amounts of labels required by supervised deep learning. Aiming to learn representations for hyperspectral classification purposes, several of these works use dimensionality reduction that could exclude relevant information during feature learning. Moreover, they are based on contrastive instance learning that requires a large memory bank to store the result of pairwise feature discriminations, which represents a computational hurdle. To overcome these challenges, the current approach performs self-supervised cluster assignments between sets of contiguous bands to learn semantically meaningful representations that accurately contribute to solving the hyperspectral classification task with fewer labels. The approach starts with the pre-processing of the data for self-supervised learning purposes. Subsequently, the self-supervised band-level learning phase takes the preprocessed image patches to learn relevant feature representations. Afterwards, the classification step uses the previously learned encoder model and turns it into a pixel classifier to execute the classification with fewer labels than awaited. Lastly, the validation makes use of the kappa coefficient, and the overall and average accuracy as well-established metrics for assessing classification results. The method employs two benchmark datasets for evaluation. Experimental results show that the classification quality of the proposed method surpasses supervised learning and contrastive instance learning-based methods for the majority of the studied data partition levels. The construction of the most adequate set of augmentations for hyperspectral imagery also indicated the potential of the results to further improve.

The task of remote sensing image scene classification (RSISC), which aims at classifying remote sensing images into groups of semantic categories based on their contents, has assumed an important role in a wide range of applications such as urban planning, natural hazards detection, environmental monitoring, vegetation mapping or geospatial object detection. During the past years, the research community focusing on RSISC tasks has shown significant effort to publish diverse datasets as well as to propose different approaches. Recently, almost all proposed RSISC systems are based on deep learning models, which proves powerful and outperform traditional approaches using image processing and machine learning. In this paper, we also leverage the power of deep learning technologies, evaluate a variety of deep neural network architectures and indicate main factors affecting the performance of a RSISC system. Given the comprehensive analysis, we propose a deep learning based framework for RSISC, which makes use of a transfer learning technique and a multihead attention scheme. The proposed deep learning framework is evaluated on the NWPU-RESISC45 benchmark dataset and achieves a classification accuracy of up to 92.6% and 94.7% with two official data split suggestions (10% and 20% of entire the NWPU-RESISC45 dataset for training). The achieved results are very competitive and show potential for real-life applications.

The application of deep neural networks to remote sensing imagery is often constrained by the lack of ground-truth annotations. Adressing this issue requires models that generalize efficiently from limited amounts of labeled data, allowing us to tackle a wider range of Earth observation tasks. Another challenge in this domain is developing algorithms that operate at variable spatial resolutions, e.g., for the problem of classifying land use at different scales. Recently, self-supervised learning has been applied in the remote sensing domain to exploit readily-available unlabeled data, and was shown to reduce or even close the gap with supervised learning. In this paper, we study self-supervised visual representation learning through the lens of label efficiency, for the task of land use classification on multi-resolution/multi-scale satellite images. We benchmark two contrastive self-supervised methods adapted from Momentum Contrast (MoCo) and provide evidence that these methods can be perform effectively given little downstream supervision, where randomly initialized networks fail to generalize. Moreover, they outperform out-of-domain pretraining alternatives. We use the large-scale fMoW dataset to pretrain and evaluate the networks, and validate our observations with transfer to the RESISC45 dataset.

With the new generation of satellite technologies, the archives of remote sensing (RS) images are growing very fast. To make the intrinsic information of each RS image easily accessible, visual question answering (VQA) has been introduced in RS. VQA allows a user to formulate a free-form question concerning the content of RS images to extract generic information. It has been shown that the fusion of the input modalities (i.e., image and text) is crucial for the performance of VQA systems. Most of the current fusion approaches use modalityspecific representations in their fusion modules instead of joint representation learning. However, to discover the underlying relation between both the image and question modality, the model is required to learn the joint representation instead of simply combining (e.g., concatenating, adding, or multiplying) the modality-specific representations. We propose a multi-modal transformer-based architecture to overcome this issue. Our proposed architecture consists of three main modules: i) the feature extraction module for extracting the modality-specific features; ii) the fusion module, which leverages a user-defined number of multi-modal transformer layers of the VisualBERT model (VB); and iii) the classification module to obtain the answer. In contrast to recently proposed transformer-based models in RS VQA, the presented architecture (called VBFusion) is not limited to specific questions, e.g., questions concerning pre-defined objects. Experimental results obtained on the RSVQAxBEN and RSVQA-LR datasets (which are made up of RGB bands of Sentinel-2 images) demonstrate the effectiveness of VBFusion for VQA tasks in RS. To analyze the importance of using other spectral bands for the description of the complex content of RS images in the framework of VQA, we extend the RSVQAxBEN dataset to include all the spectral bands of Sentinel-2 images with 10m and 20m spatial resolution. Experimental results show the importance of utilizing these bands to characterize the land-use land-cover classes present in the images in the framework of VQA. The code of the proposed method is publicly available at https://git.tu-berlin.de/rsim/multimodal- fusion-transformer-for-vqa-in-rs.

Radar Sounders (RSs) are active sensors widely used for planetary exploration and Earth observation that probe the subsurface in a non-intrusive way by acquiring vertical profiles, called radargrams. Radargrams contain information on subsurface geology and are analyzed with neural networks for segmentation and target detection. However, most of these methods rely on supervised training, which requires a large amount of labeled data that is hard to retrieve. Hence, a need emerges for a novel method for unsupervised radargram segmentation. This paper proposes a novel method for unsupervised radargram segmentation by analyzing semantically meaningful features extracted from a deep network trained with a contrastive logic. First, the network (encoder) is trained using a pretext task to extract meaningful features (query). Considering a dictionary of possible features (keys), the encoder training loss can be defined as a dictionary look-up problem. Each query is matched to a key in a large and consistent dictionary. Although such a dictionary is not available for RS data, it is dynamically computed by extracting meaningful features with another deep network called the momentum encoder. Secondly, deep feature vectors are extracted from the encoder for all radargram pixels. After the feature selection, the feature vectors are binarized. Since pixels of the same class are expected to have similar feature vectors, we compute the similarity between the feature vectors to generate a cluster of pixels for each class. We applied the proposed method to segment radargrams acquired in Greenland by the MCoRDS-3 sensor, achieving good overall accuracy.

Subsurface tile drainage pipes provide agronomic, economic and environmental benefits. By lowering the water table of wet soils, they improve the aeration of plant roots and ultimately increase the productivity of farmland. They do however also provide an entryway of agrochemicals into subsurface water bodies and increase nutrition loss in soils. For maintenance and infrastructural development, accurate maps of tile drainage pipe locations and drained agricultural land are needed. However, these maps are often outdated or not present. Different remote sensing (RS) image processing techniques have been applied over the years with varying degrees of success to overcome these restrictions. Recent developments in deep learning (DL) techniques improve upon the conventional techniques with machine learning segmentation models. In this study, we introduce two DL-based models: i) improved U-Net architecture; and ii) Visual Transformer-based encoder-decoder in the framework of tile drainage pipe detection. Experimental results confirm the effectiveness of both models in terms of detection accuracy when compared to a basic U-Net architecture. Our code and models are publicly available at https: //git.tu-berlin.de/rsim/drainage-pipes-detection.

Radar sounders (RS) provide information on subsurface targets for planetary investigations. Several simulation techniques have been developed to support the RS design and the data interpretation. Each technique has different properties and modeling capabilities, achieving different trade-offs between accuracy and computational requirements. The state-of-the-art RS simulation techniques include: i) numerical methods, such as the Finite- Difference Time-Domain (FDTD) technique, which allows the modelling of small-scale scattering phenomena at the cost of high computational requirements; ii) facet modeling and ray-tracing based methods, such as the multi-layered coherent RS (MCS) technique, which requires less computational resources than FDTD, allowing the modeling of large-scale scattering phenomena. Recently an integrated simulation methodology has been presented, for simulating small-scale scattering phenomena in large scenarios. However, this methodology was designed for modeling only surface scattering. In this paper, we propose a method that extends the capabilities of the integrated methodology to model both large and small-scale roughness in a multi-layer scenario. The proposed method uses the FDTD technique to evaluate the effects associated with small-scale roughness in terms of i) scattering phenomena associated with the layers and ii) power losses associated with the signal transmitted through a rough layer. To recursively apply scattering and transmission to multiple layers of the subsurface, a coherent ray-tracing method is used. We experimentally assessed the effectiveness of the proposed methodology on three-layer models by integrating the effect of roughness imposed on the layers and in transmission through them.

Radar Sounders (RSs) are sensors operating in the nadir-looking geometry (with HF or VHF bands) by transmitting modulated electromagnetic (EM) pulses and receiving the backscattering response from different subsurface targets. Recently, convolutional neural network (CNN) architectures were established for characterizing RS signals under the semantic segmentation framework. In this paper, we design a Fast Fourier Transform (FFT) based CNN-Transformer encoder to effectively capture the long-range contexts in the radargram. In our hybrid architecture, CNN models the high-dimensional local spatial contexts, and the Transformer establishes the global spatial contexts between the local spatial ones. To overcome Transformer complex self-attention layers by reducing learnable parameters; - we replace the self-attention mechanism of the Transformer with unparameterized FFT modules as depicted in FNet architecture for Natural Language Processing (NLP). The experimental results on the MCoRDS dataset indicate the capability of the CNN-Transformer encoder along with the unparameterized FFT modules to characterize the radargram with limited accuracy cost and by reducing the time consumption. A comparative analysis is carried out with the state-of-the-art Transformer-based architecture.

In this paper proposed is the surface deformations in the area of the nuclear power plant (NPP) Kozloduy to be analyzed using a time series of remotely sensed SAR data from the Sentinel-1 mission of ESA. It should be noted that this is the only way to obtain reliable information about the motions at regular intervals since the zone has restricted access. The justification for carrying out this study is– first the mentioned radar instrument was purposely built for study those phenomena at large scales and second the provision of the data and the software for their processing are obtained free of charge from the ESA repositories. It also should be underlined that the improved spatial resolution of the sensor compared with similar ones (e.g. ERS-1/2, Envisat) and the short revisiting time (6/12 days) allow production of information with better quality compared to the results obtained up to this moment using remotely sensed data only. In this research to produce the interferometric maps that reveal the overall ground stability of the studied region the short baseline subsets (SBAS) approach was adopted using the DInSAR technique as main processing method. Those maps can serve the needs of the competent local authorities as well as to the plant operator in order to increase the safety of the population living in the area.

The detection and monitoring of landslides in dam reservoirs are essential for the maintenance and sustainable use of the infrastructure. Sentinel-1 datasets, freely available from European Space Agency (ESA), and Synthetic Aperture Radar Interferometry (InSAR) processing methods provide multitemporal surface change information in terms of deformation and enable engineering geological analyses. The main purpose of this study is to investigate the deformations within the large landslides based on InSAR measurements in the reservoir of the Kalekoy dam, which was built in Bingol Province in the Eastern part of Türkiye and filled in 2018. The region is prone to landslides at various magnitudes. For this purpose, the Looking into Continents from Space with Synthetic Aperture Radar (LiCSAR) products were used to apply the Small-Baseline Subsets (SBAS) method for the deformation extraction from Sentinel-1 time series datasets. The activities within the landslides and their surroundings were detected from the deformation maps produced for three periods between December 2015- December 2017 (pre-dam), December 2018-January 2020 (post-dam) and within 2018 (December 2017 - December 2018). Based on the time series analysis results, it was observed that the slope movements increased significantly in 2018, the time of the filling of the reservoir, and in the post-dam periods; and their patterns have also altered.

This paper focused on the applied remote surveillance practices of surface displacements for the subsidence and uplift risk assessments of the largest oil and gas fields and pipelines in Kazakhstan and Azerbaijan.

In case of InSAR applications for Tengiz Oilfield in Kazakhstan, SBAS-InSAR remote sensing technique followed by 3D and 2D decompositions showed the continuous subsidence at the Tengiz oilfield with increasing velocity. 3D and 2D decompositions of LOS measurements to vertical movement showed that the Tengiz Oil Field 2018-2020 continuously subsided with the maximum annual vertical deformation velocity around 70 mm. The vertical deformation confirmed typical patterns of subsidence caused by oil extraction. However, detected east-west and north-south horizontal movements at the Tengiz field clearly indicated that the study area crossed by seismic faults is also affected by natural tectonic processes.

In case of InSAR applications for pipelines, both PS-InSAR and SBAS-InSAR techniques showed continuous subsidence in the kilometer range of 13-70 crossing two seismic faults. The ground uplift deformations were observed in the pipeline kilometer range of 0-13. Although both PS-InSAR and SBAS measurements were highly consistent in deformation patterns and trends along pipelines, they showed differences in the spatial distribution of ground deformation classes and noisiness of produced results. SBAS showed better performance than PS-InSAR along buried petroleum and gas pipelines in the following aspects: the complete coverage of the measured points, significantly lower dispersion of the results, continuous and realistic measurements and higher accuracy of ground deformation rates against the GPS historical measurements.

In case of InSAR applications for Absheron oil and gas fields, PS-InSAR showed the existence of ground deformation processes observed for the period of 2015-2017 with three hotspots of highest subsidence rates and three hotspots of highest uplift rates in oil and gas fields. The determined maximum displacement rates of subsidence and uplift processes were −26 mm/y and +23 mm/y, respectively. However spatial density analysis of deformation velocity presented the natural patterns of uplift and subsidence tectonic processes. This allowed to determine that two oil and gas fields hold a higher probability of being affected by man-made oil and gas exploration activities, whereas the one oil field is affected by both natural and man-made processes.

Global outgoing longwave radiation variation is examined for a 6-year (2014-2019) period using reanalysis and FengYun 3B/3C OLR products: the interpolated OLR analyses and the visible and infrared radiometer OLR on FengYun 3 series satellite platforms. Tropical Pacific and Niño key regions are essential range for monitoring El Niño events, which are chosen to investigate the spatial-temporal correlation, seasonal evolution and statistical difference with a variety source of OLR anomalies. Results indicate that monthly OLR anomalies can monitor the Niño variation. OLR anomalies are associated with enhanced amplitude in the EP El Niño, while OLR anomalies exhibit stronger intensity than the CP El Niño index. For FengYun 3B/3C the monthly satellite products have a root mean squared error of 10.94 W/m² and 12.73 W/m² as compared for the interpolated OLR products.

Ship detection in remote sensing images is important for maritime surveillance. With the rapid development of earth observation technology, high-resolution imaging satellites can provide more observational information. In the face of massive remote sensing data, object-level annotation requires a lot of time and manpower. Weakly supervised object detection is trained using only image-level annotations, thus reducing the requirement for object-level annotations. However, there are still some problems in the detection of weakly supervised ships in remote sensing images, because of the complex, dense distribution and diverse scale characteristics of the ship environment. We propose a weakly supervised object detection method that combines Transformer with weakly supervised learning for ship detection in remote sensing images. First, Proposal Clustering Learning (PCL) for weakly supervised object detection is used as the baseline to detect ships, and the network is continuously refined for better detection performance. Second, the prior location and size information is added to the features of the proposal through the transformer module. This additional information can be used as an important basis for judging whether the proposal is optimal, thereby improving the detection performance. To evaluate the effectiveness of our method, extensive experiments are conducted on a complex dataset of large-scene remote sensing ships. Experimental results show that our method achieves better detection performance than other methods.

Earth observation satellites can improve the accuracy of image registration by sensing the stars and determining their centroids, as is applied on GOES-16 meteorological satellite of the United States and Himawari 8 of Japan. In this paper, we propose an improved method to accurately extract the subpixel centroids of the stars observed by a single-line detector array with high spatiotemporal resolution. We firstly design a three-way screening method to accurately and quickly detect the position of the star in a long image sequence. Secondly, we estimate the time when the star passes the center of the detector by finding the peak of the energy curve of the observed star images. Then, we fit the trajectory of the star according to the angular velocity in the field of view of the geostationary satellite. Finally, the subpixel centroid of the star in a certain time can be obtained using the fitted trajectory. To verify our method, we simulate the star images of a single-line detector array for geostationary Earth observation with different magnitudes and different sensor parameters. We carried out extensive experiments on the simulated data. Experimental results show that our star centroid extraction method can accurately detect the observed stars and extract their centroids.

Remote operating and autonomous systems are widely applied in various fields, and the development of technology for human machine interface and communication is strongly demanded. In order to overcome the limitations of the conventional keyboard and tablet devices, various vision sensors and state-of-the-art artificial intelligence image processing techniques are used to recognize hand gestures. In this study, we propose a method for recognizing a reference sign language using auto labeled AI model training datasets. This study can be applied to the remote control interfaces for drivers to vehicles, person to home appliances, and gamers to entertainment contents and remote character input technology for the metaverse environment.

The accurate segmentation of the leaf area on scanned digital images plays a crucial role in the automated evaluation of its morphological characteristics. We propose here a new algorithm for extracting leaf area from the digital images based on a combination of a parametric description of shadow and background areas in the color space by support vector data description (SVDD) and the structure transfer filtering method based on the gamma-normal probabilistic model. The combination of these methods allows us to consider color information as well as sharp changes in image intensity at the edges of a leaf.

For decades, the design of remote sensors has to make trade-offs among many characteristics such as the field of view (FOV), spatial resolution, spectral resolution, radiometric resolution, and the number of bands. It’s inevitable to weaken some characteristics to enhance others. Moreover, these problems lead to using multi-sources of remote sensing data in practical projects where a single sensor can’t meet the relevant requirements. The Airborne Dual-mode High-resolution Hyperspectral Imager (ADHHI) provides a new solution to the above limitations by the technology of multi-camera stitching. In this way, many excellent but conflicting characteristics can be separated into different imaging sub-systems, and are combined together during data processing. Supported by related processing algorithms and software, ADHHI embeds many excellent characteristics into one system, such as high spatial resolution, high spectral resolution, and high radiometric resolution. Firstly, this paper picks some common imaging sensors to illustrate the problem of conflicting characteristics. Secondly, we introduce the camera structure and sensor parameters of ADHHI. Then, we sketch out the data processing workflow and elaborate on relevant principles and results of the whole geometric correction, such as homo-spectral stitching and hetero-spectral registration. Finally, this airborne hyperspectral imager’s advantages and application prospects are concluded.

Publisher's Note: This paper, originally published on 26 October 2022, was replaced with a corrected/revised version on 30 November 2023. If you downloaded the original PDF but are unable to access the revision, please contact SPIE Digital Library Customer Service for assistance.