Search

Spatiotemporal satellite image fusion aims to generate remote sensing images satisfying both high spatial and temporal resolution by integrating different satellite imagery datasets with distinct spatial and temporal resolutions. Such fusion technique is crucial for numerous applications that require frequent monitoring at fine spatial and temporal scales spanning agriculture, environment, natural resources, and disaster management. However, existing fusion models have difficulty accommodating abrupt spatial changes in land cover among images and dealing with temporally distant image data. This article proposes a novel multistream spatiotemporal fusion generative adversarial network (STGAN) model for spatiotemporal satellite image fusion that can produce accurate fused images and accommodate substantial temporal differences between the input images. The STGAN employs a conditional generative adversarial network architecture with a multistream input design to better learn temporal features. The generator of STGAN comprises convolutional blocks, a spatial transformer module, a channel attention network, and a U-net module designed to better capture spatial and temporal features from the multiresolution input images. Comprehensive evaluations of the proposed STGAN model have been performed on the Coleambally Irrigation Area and Lower Gwydir Catchment datasets, using both visual inspection and spatial and spectral metrics, including root mean square error, relative dimensionless global error synthesis, spectral angle mapping, structural similarity index measure, and local binary pattern. The experiments show that the proposed STGAN model consistently outperforms existing benchmark models and is capable of generating high-quality fused remote sensing data product of high spatial and temporal resolution.

Given the increasingly severe destruction of wetlands in recent years, research and monitoring for wetland protection are urgently needed. However, wetland monitoring still faces significant challenges such as high manual costs and low image monitoring accuracy, primarily due to the complex composition of wetland terrain and diverse spectral characteristics, especially located in remote suburban areas. With their rich semantic and spectral features, studying wetlands as scenes in remote sensing images and segmenting them effectively can address these challenges. To address the issue of low segmentation accuracy on remote sensing image with GrabCut, this study proposes a remote sensing image segmentation extraction method, named SceneCut, based on feature extraction and multifeature joint graph cuts. The method consists of three parts: feature extraction, feature fusion, and multifeature joint graph cuts. In the first part, image features are calculated using a sliding window approach. In the second part, region covariance calculation and matrix eigenvalue decomposition are used to generate a multichannel feature image from the extracted feature vectors. Finally, in the third part, GrabCut is applied to the multichannel image to perform graph cuts, considering multiple features, and the segmentation result is mapped back to the original image to generate the segmented image. Experimental analysis and field exploration validation of wetland images conducted in this study demonstrate that compared to algorithms that only use RGB features for segmentation or those that do not consider the relationship between multiple features, SceneCut performs well in wetland scene extraction, especially in extracting boundaries with significant blending.

Spatial-temporal fusion algorithms commonly encounter difficulties in effectively striking a balance between the extraction of intricate spatial details and changes over time. To mitigate these problems, we propose a spatiotemporal fusion network for remote sensing based on a global context (GC) attention mechanism. This network comprises an extracting feature network and a difference network. In the difference network, we introduce a GC attention mechanism that focuses on crucial details across various features, assuming different roles in different modules. A GC attention-based U-block (GCAU) employs a U-structured design and integrates GC attention mechanisms into every layer of its architecture. This enables the module to effectively process regions with pronounced spatial heterogeneity and to adeptly capture spatial differential information. A GC layer (GCL) block comprises five interconnected GC attention blocks interspersed with local residuals. During the training process, these residuals assist in rectifying missing data and bolstering feature transfer, thereby enabling the module to more effectively capture temporally dynamic changes in rapidly evolving areas. This allows the module to better capture time-based changes in fast-changing areas. Supervision of both the final output and intermediate difference images is facilitated by a composite loss function, which improves the fusion quality in the temporal, spatial, and visual domains. The model's robustness and superiority are validated through experimental testing on three datasets, accompanied by subjective and objective evaluations, as well as ablation experiments.

Knowledge of sea ice lead and floe density is important for ship navigation, understanding ice pack dynamics, and mapping primary productivity. Furthermore, reliable lead and floe discrimination is essential for the calculation of sea ice freeboard and thickness from satellite altimetry. Despite its value for shipping, the Northwest Passage is commonly excluded from Arctic-wide satellite altimetry sea ice observations owing to the challenges of obtaining accurate retrievals in the region. Here, we map the distribution of leads and floes in the Northwest Passage using Landsat 8 satellite optical imagery, and we use these to evaluate the efficacy of near-coincident classifications from CryoSat-2 satellite radar altimetry. We test different pulse peakiness and stack standard deviation thresholds for discriminating returns from CryoSat-2 into leads and floes and assess the agreement with near-coincident Landsat 8 estimates. CryoSat-2 lead and floe densities are, on average, 14% higher and 45% lower than Landsat 8, respectively. We attribute the difference to a range of factors including the coarser spatial sampling of CryoSat-2 which results in mixed surface types within the radar footprint, off-nadir ranging of leads, and the misidentification of sea ice floes as ambiguous surface type. In particular, we find the poorest agreement in October, when the ice pack is fragile after the summer melt season. We use our findings to develop a bias adjustment to bring CryoSat-2 lead and floe density estimates in line with Landsat 8, reducing the root mean square difference between them from 20% to 5% and 47% to 6%, respectively.

Human-labeled training datasets are essential for convolutional neural networks (ConvNets) in satellite image scene classification. Annotation errors are unavoidable due to the complexity of satellite images. However, the distribution of real-world human-annotated label noises on satellite images and their impact on ConvNets have not been investigated. To fill this research gap, this article, for the first time, collected real-world labels from 32 participants and explored how their annotated label noise affects three representative ConvNets (VGG16, GoogleNet, and ResNet-50) for remote sensing image scene classification. We found that 1) human-annotated label noise exhibits significant class and instance dependence; 2) an additional 1% of human-annotated label noise in training data leads to a 0.5% reduction in the overall accuracy of ConvNets classification; and 3) the error pattern of ConvNet predictions was strongly correlated with that of participant's labels. To uncover the mechanism underlying the impact of human labeling errors on ConvNets, we compared it with three types of simulated label noise: uniform noise, class-dependent noise, and instance-dependent noise. Our results show that the impact of human-annotated label noise on ConvNets significantly differs from all three types of simulated label noise, while both class dependence and instance dependence contribute to the impact of human-annotated label noise on ConvNets. Additionally, the label noise estimation algorithm (confident learning) cannot fully identify label noise. These observations necessitate a reevaluation of the handling of noisy labels, and we anticipate that our real-world label noise dataset would facilitate the future development and assessment of label-noise learning algorithms.

Geochemical data and remote sensing data are two typical exploratory datasets widely employed in various kinds of research, such as mineral development and environmental protection. The fusion of these two types of datasets can provide richer and more accurate information for geoscience analysis. But the existing remote sensing–geochemical data fusion methods have problems; there is a big gap between the two kinds of data in resolution. And there are other limitations, such as low fusion quality, few optional fusion methods, and not suitable for practical production applications. This article proposes a novel convolutional neural network that effectively fuses high-resolution remote sensing data with low-resolution geochemical data, enhancing the quality of geochemical data. A new data augmentation method, which mixes data of different resolutions but with the same mapping relationship, is applied to the fusion process of remote sensing data and geochemical data in this article. Experiments show that this method can improve the quality of fusion results. And transfer learning is applied to the fusion of different elements. Ag and Cr are selected to verify the mobility and generalization of the model. Compared with the existing interpolation methods, the results show that the fusion method based on the model can better characterize the distribution law of geochemical data.

Trees outside forests, in particular on croplands, play a crucial role for food security and climate resilience in the Global South but are threatened by increasing climatic change and human pressures. The dynamics of agroforestry systems and national cropland tree stocks are largely unknown, as currently, no robust monitoring system exists to remotely detect single field trees and track changes at national scales. Here, we present a framework to track cropland trees at the single tree level across multiple years, using a combination of satellite imagery, deep learning, and object-based change classification. The approach matches annual tree center predictions to detect changes, such as individual tree losses from logging or tree mortality events. The slope between annual tree prediction confidence heatmaps is also used to detect areas of gains, with possible applications for monitoring plantation and restoration areas. The framework is designed for PlanetScope nanosatellite imagery, which offers unprecedented opportunities for detailed tree monitoring given the combined high spatial and temporal resolution. PlanetScope imagery, however, also comes with a range of challenges, which are discussed and for which solutions are proposed. We demonstrate the framework by applying it to a national-scale case study of cropland trees in Tanzania from 2018 to 2022, detecting a disappearance of 1 10 500 trees with confidence >0.8.

Change detection has become a crucial topic in the field of remote sensing deep learning due to its extensive application in earth observation. However, real remote sensing images often contain multiple land cover classes with significant intraclass variability and interclass similarity, limiting the performance of change detection in complex scenarios. To address this, we leverage the capabilities of vision foundation models by applying the segment anything model (SAM) to remote sensing change detection, and we name this method dual encoder–decoder SAM (DED-SAM). Specifically, we construct a DED framework, utilizing a small-scale change detection model in both branches to generate mixed prompts including image features, mask prompts, and box prompts. The SAM 2 model is used for fine-grained recognition of dual-temporal images, generating accurate, and stable feature boundaries, which are then used as constraints to generate the final change mask. To validate the effectiveness of DED-SAM across various application scenarios, we conduct quantitative experiments on three public datasets: Levir-CD, SYSU-CD, and CDD, testing its detection capabilities under single change categories, multiple change categories, and seasonal pseudochange interference. The results show that the proposed DED-SAM achieved state-of-the-art F1 scores and IoUs on these three datasets: LEVIR-CD (92.00%, 85.11%), SYSU-CD (84.15%, 72.01%), and CDD (97.72%, 95.47%).

The integration of autonomous aerial vehicles (AAVs) in agriculture has advanced precision farming by enhancing the ability to monitor and optimize agricultural plots. Object detection—critical for identifying crops, pests, and diseases–presents challenges due to data availability and varying environmental conditions. To address these challenges, we propose a Deep Learning framework tailored to agricultural contexts, utilizing domain-specific knowledge from AAV imagery. Our framework uses a knowledge base of visual features and loss values from multiple deep-learning models during the training phase to choose the most effective model for the testing phase. This approach improves model adaptability and accuracy across diverse agricultural scenarios. Evaluated on a comprehensive dataset of AAV-captured images covering various crop types and conditions, our model shows superior performance compared to state-of-the-art techniques. This demonstrates the value of integrating domain knowledge into deep learning for enhancing object detection, ultimately advancing agricultural efficiency, supporting sustainable resource management, and reducing environmental impact.

Pine wilt disease (PWD) poses a severe threat to the health of pine trees and has resulted in substantial losses to global pine forest resources. Due to the minute size of the pathogens and the concealed symptoms of PWD, early detection through remote sensing image technology is essential. However, in practical applications, remote sensing images are easily affected by factors, such as cloud cover and changes in illumination, resulting in significant noise and blurriness in the images. These interference factors significantly reduce the accuracy of existing object detection models. Therefore, this article presents a novel and highly robust methodology for detecting PWD, termed YOLOv8-RD. We synthesized the benefits of residual learning and fuzzy deep neural networks to develop a residual fuzzy module (ResFuzzy), which adeptly filters image noise and refines background features with enhanced smoothness. Simultaneously, we integrated a detail processing module into the ResFuzzy module to enhance the low-frequency detail features transmitted in residual learning. Furthermore, by incorporating the dynamic upsampling operator, our model can dynamically adjust the sampling step size based on the variations in the input feature map during the upsampling process, thereby effectively recovering detail from the feature map. Our model exhibited exceptional robustness to severe noise. When evaluated on a PWD dataset with 100% interference samples at an intensity of 0.07, our model achieved an average precision improvement of 4.9%, 6.3%, 7.3%, and 3.0% compared to four most representative models, making it well suited for PWD detection in interfering environments.

Abstract: Mangroves are crucial ecosystems with significant ecological and economic roles, providing habitat for diverse species, storm protection, and carbon sequestration. While many mangroves worldwide have faced substantial losses, China has seen notable recovery. Understanding the expansion patterns of mangroves necessitates high-resolution, large-scale monitoring of their coverage over time. In this study, we divided coastal zones based on geomorphic and sedimentary environments and generated annual mangrove coverage from 2016 to 2022 using multi-source high-resolution data and a U-Net model. By analyzing morphological changes of mangrove patches in different coastal zones of the Guangxi Beibu Gulf, we summarized expansion patterns and calculated expansion and degradation rates from area changes. Our findings reveal a 1,413 ha increase in mangrove area from 2016 to 2022. Estuarine mangroves experienced an initial increase followed by a decrease, while gulf mangroves showed an initial decrease followed by an increase. Three distinct expansion patterns were identified: (1) regular mangroves expanding parallel to coastlines; (2) mangroves expanding in estuarine areas where freshwater and seawater meet; and (3) independent mangrove patches expanding into surrounding wetlands. In summary, stable and changing mangrove areas were in a ratio of 7:3, with expansion areas at 70.5% and degradation areas at 29.5%. The loss of mangroves in Guangxi Beibu Gulf is primarily due to intense human activities, with minimal contributions from sea level rise and marine dynamics.

Spatio-temporally continuous daily evapotranspiration (ET) is essential for characterizing water and energy exchange and scheduling efficient water management. ET has conventionally been generated using thermal infrared-based models, cloud contamination of satellite data could prohibit accurate estimates of spatial continuous daily ET. Although approaches have been applied to fill these spatial gaps introduced by thermal infrared data, they contain extensive uncertainties and still have gaps. Here, we proposed a two-step reconstruction framework to generate seamless daily ET dataset based on outputs from a soil moisture coupled two-source energy balance (TSEB-SM) model. In these two steps, a deep neural network trained with the outputs of TSEB-SM was used to reconstruct the gaps in daily ET images, which mainly introduced by the missing inputs. Then the remained gaps were filled with reference ET (ETo) directly. The estimated daily ET agrees well with ground measurements across different landcover types with a RMSE of 1.0 mm day−1 and a bias of only 0.2 mm day−1. In terms of spatial distributions and temporal dynamics, the generated daily ET has better consistency with its impacting factors, including the landcover map, land surface temperature, downward solar radiation, etc. Our results suggest that this reconstruction framework can generate reliable seamless daily ET dataset, which has high potential for application in crop water consumption monitoring, crop yield prediction and efficient water management.

Hyperspectral target detection technology is becoming more and more important in remote sensing. Most of the traditional methods of target sensing using spectral information treat hyperspectral image as 2-D matrix regardless of the structure information of hyperspectral image, resulting in insufficient separation of target and background and limited detection accuracy. In order to better utilize the hyperspectral intrinsic structural information, this article proposes a hyperspectral target sensing approach based on tensor adaptive reconstruction cascaded with global and local feature fusion (TRGLF). First, the Tucker decomposition and reconstruction method is utilized to alleviate the influence of noise and other elements in the complex background hyperspectral, which can maintain the intrinsic structural features of the data and improve the background and target separation. The principal components of the factor matrix are determined by a logarithmic singular value summation strategy, and then the energy difference between the spectra to be measured and the previous spectra is calculated to fine-tune the principal components and obtain more appropriate principal component values. Second, on the basis of reconstructed data, a global and local spatial spectral fusion method is adopted to obtain the target of interest. This includes using globally constrained energy minimization to obtain the target of interest and using local differential sensing to further obtain the target boundary. The detection performed on four realistically acquired hyperspectral water surface datasets demonstrate the excellent detection performance of the proposed target detection method.

In this article, we develop a contrastive graph-based active learning pipeline (CGAP) to identify surface water and near-water sediment pixels in multispectral images. The CGAP enhances the graph-based active learning pipeline (Chen et al., 2023), which outperforms methods such as CNN-Unet, support vector machine, and random forest, while requiring less training data. Active learning plays an important role for training data reduction, resulting in an order of magnitude less training data compared with conventional methods and three or more orders of magnitude less compared with CNN-Unet. Our improvements focus on boosting both the pipeline's robustness and efficiency by integrating a feature-embedding neural network prior to graph construction. This neural network, trained using contrastive learning, performs effective data dimension reduction by projecting high-dimensional raw features into a lower dimensional space, thereby facilitating more efficient graph learning. The training process incorporates specialized augmentations to bolster the embedded features' resilience to geometric transformations, varying resolutions, and light cloud cover. Moreover, we develop a Python-based demo, GraphRiverClassifier, that uses the Google Earth Engine and our enhanced pipeline to provide a user-friendly tool for rapid and accurate surface water and sediment analyses and rapid testing of algorithm performances.

Satellites are increasingly adopting onboard AI to optimize operations and increase autonomy through in-orbit inference. The use of deep learning (DL) models for segmentation in hyperspectral imagery offers advantages for remote sensing applications. In this work, we train and test 20 models for multiclass segmentation in hyperspectral imagery, selected for their potential in future space deployment. These models include 1-D and 2-D convolutional neural networks (CNNs) and the latest vision transformers (ViTs). We propose a lightweight 1D-CNN model, 1D-Justo-LiuNet, which outperforms state-of-the-art models in the hypespectral domain. 1D-Justo-LiuNet exceeds the performance of 2D-CNN UNets and outperforms Apple's lightweight vision transformers designed for mobile inference. 1D-Justo-LiuNet achieves the highest accuracy (0.93) with the smallest model size (4563 parameters) among all tested models, while maintaining fast inference. Unlike 2D-CNNs and ViTs, which encode both spectral and spatial information, 1D-Justo-LiuNet focuses solely on the rich spectral features in hyperspectral data, benefitting from the high-dimensional feature space. Our findings are validated across various satellite datasets, with the HYPSO-1 mission serving as the primary case study for sea, land, and cloud segmentation. We further confirm our conclusions through generalization tests on other hyperspectral missions, such as NASA's EO-1. Based on its superior performance and compact size, we conclude that 1D-Justo-LiuNet is highly suitable for in-orbit deployment, providing an effective solution for optimizing and automating satellite operations at edge.

Semantic change detection (SCD) represents a challenging task in the interpretation of remote sensing images (RSIs), with the goal of identifying change regions and extracting semantic information from bitemporal RSIs simultaneously. The recent integration of deep neural networks leveraging multitask learning has shown promise in enhancing SCD performance. However, there is still a challenge in improving SCD performance, specifically in designing a fine-grained network structure that can handle the two subtasks of change region localization and semantic information recognition in parallel. In this context, a novel multitask Siamese network, termed EGMS-Net, is proposed to boost the performance of SCD, which consists of three core components. First, a coarse-to-fine multitask Siamese network is constructed to obtain semantic information and change information at multiple levels. Second, an adaptive change information enhancement method based on spatial-spectral collaborative attention mechanism is proposed, which can assist the accurate localization of change regions without significantly increasing the model parameters. Third, a change information guidance module is developed to strengthen the interaction between multitask branches and reduce the difficulty of network training. Experiments on three benchmark datasets demonstrate that the proposed EGMS-Net outperforms existing state-of-the-art methods in the SCD community.

Accurate building semantic segmentation in remote sensing imagery is essential for urban planning, environmental monitoring, and map creation. While deep learning has achieved significant advancements in this field, precisely segmenting building edges and shadows in complex scenarios remains challenging. Shadows often introduce boundary ambiguities, affecting the local shape and texture information of buildings. Current methods do not fully perceive or utilize shadows. To address these challenges, we propose an advanced high-resolution image segmentation network, high-resolution network, integrated with a shadow-inclusive edge perception module. Our approach involves introducing a shadow-inclusive contour transition module (SCTM) during the feature extraction stage to enhance the features of blurry boundaries. The proposed SCTM and shadow-aware attention module significantly enhance attention maps, improve responses in blurry boundary regions, and increase consistency between predictions and ground truth, setting a new benchmark for building semantic segmentation in remote sensing imagery. This enriched information is then fed into an attention module that concurrently focuses on boundary and channel features, surpassing traditional semantic segmentation methods. We validated our method on three datasets: Massachusetts, WHU, and Inria. Our approach outperformed state-of-the-art methods on the WHU Building Dataset across all metrics, including mIoU, Accuracy, Kappa, and Dice coefficient.

River water body identification plays an important role in flood monitoring, urban planning, Thus, it attracts more interests of studying and investigating, especially based on remote sensing technology, The traditional NDWI (Normalized Difference Water Index) and MNDWI (Modified Normalized Difference Water Index) methods are widely used, However, these methods need manual intervention to select the threshold, In order to achieve automatic water body recognition, deep learning methods, such as CNN, VGG, Unet etc., are applied, Currently there are few works on the water body identification of Qiantang River, Here, one major challenge for high-precision Qiantang water body recognition is the real complex water body features and complicated geological environment, They are the dense distribution of small water bodies in the Qiantang River Basin, large differences in water body nutrition, and the high complexity of surface environments such as mountains and plains, We investigated two traditional and several deep learning methods and found that WatNet was the most effective model for Qiantang River, This model adopts the structure based on encoder-decoder convolutional network, It uses MobileNetV2 as the encoder, which makes it extract more water feature information while being lightweight and uses ASPP module to capture global multi-scale features in deep layers, Experimental results show that the MIoU and OA (Overall Accuracy) can reach 0. 97 and 0. 99 respectively.

Sea ice plays a critical role in influencing various aspects of the atmosphere, ecology, oceans, and even human activities, making the task of sea ice retrieval (SIR) immensely important. Recent advancements have seen the application of machine learning methods based on Global Navigation Satellite System Reflectometry (GNSS-R) technology in the field of sea ice remote sensing. However, the existing algorithms are hindered by an over-reliance on labels, structural redundancies, and limited effectiveness in dealing with sample imbalances. To overcome these challenges, the Anomaly Detection Driven Semi-supervised Multi-task Retrieval Algorithm for Sea Ice based on GNSS-R is proposed in this paper. In the novel approach within the field of SIR, this study introduces the concept of semi-supervised anomaly detection. This method addresses the challenge posed by an over-reliance on labeled data and the issue of sample imbalance. The proposed algorithm is composed of two primary components: the data preprocessing module and the multi-task network structure Sea Ice Multi-Tasks Retrieval Network (SIMTRN) based on GNSS-R. The SIMTRN includes a sea ice detection (SID) module, which employs an implicit deep bootstrap mechanism for enhanced detection capabilities. Additionally, it encompasses a sea ice concentration retrieval (SICR) module that implements an innovative inter-module data feature sharing mechanism. This dual-module approach within a single network structure not only facilitates simultaneous execution of SID and SICR tasks but also effectively addresses and resolves the issue of structural redundancy. Experiments conducted on the datasets illustrate the superior SIR performance of the proposed method.

This paper presents an advanced approach to Human Pose Estimation (HPE) and Semantic Event Classification (SEC), emphasizing the need for sophisticated human skeleton models, context-aware feature extraction, and machine learning techniques for precise event recognition in daily life logs. HPE, crucial in applications like sports analysis and surveillance systems, involves predicting human joint locations from images and videos. Recent deep learning advancements have significantly improved HPE, particularly in crowded scenes and occlusion challenges. Despite many surveys, a comprehensive review of HPE, especially with recent deep learning innovations, is still needed. Our research addresses this by proposing a novel HPE and SEC system. The system begins with preprocessing steps, including converting videos into image sequences, applying sliding window techniques, and converting images to grayscale, then extracting human silhouettes using binary masks. We use the GrabCut algorithm for human detection and perform skeletonization with Hough transform algorithm. Keypoint detection is achieved through pose estimation, and full-body feature extraction includes using OpenPose for movable body parts, the Lucas-Kanade method for a 3D Cartesian view, and Texton Map techniques. Key point features are further characterized using motion histograms, pose landmark visualization and Local Intensity Order Pattern (LIOP) features. The system is optimized with adaptive moment estimations and classified using the XGBoost Classifier. Evaluation on the COCO, UCF50, and YouTube datasets showed classification accuracies of 92.90%, 90.9%, and 91.2%, respectively, demonstrating our approach’s superior performance and effectiveness compared to existing state-of-the-art techniques.

Flooding is one of the most common natural disasters, causing emergencies and significant damage to infrastructure. Amid crises, entities such as the World Food Program employ remotely sensed images, usually via drones, to analyze the situation quickly and plan life-saving interventions. Suitable Computer Vision (CV) tools are necessary to assist experts of first responder teams with the analysis of the image data, increasing their productivity and enabling strategic resource allocation. Many state-of-the-art CV-tools use supervised Deep Learning (DL) techniques, for which labeled training data is needed. In this paper, we present the BlessemFlood21 dataset to support the development of DL-based CV-tools for the task of advanced flood analysis in non-coastal river flood scenarios. In particular, we address the problem of segmenting water from RGB images. The presented resulting dataset consists of high-definition, georeferenced RGB images labeled with precise water masks. The water masks are derived using a proposed semi-automatic human-in-the-loop strategy based on the use of additionally imaged NIR data together with a classical Random Forest approach. We assess the resulting dataset and provide a baseline for future advancements in flood mapping by training and evaluating three different state-of-the-art segmentation models (UNet++, DeepLabV3+, SegFormer-B5). Further, we showcase the potential use of the dataset with several experiments. In particular, we compare with the Floodnet dataset and, we employ the segmentation models trained on our dataset together with a Digital Elevation Model to showcase the potential to estimate flood water levels.

On June 6, 2023, the Kakhovka Dam in Ukraine was damaged, causing a flood that could have had a significant impact on society. A single satellite may not provide sufficient data for timely disaster response. To overcome this, a combination of optical images, a new generation of hydrological monitoring satellite data, and nighttime light data were employed to analyze the disaster. Sentinel-3 provides useful hydrological information for quickly identifying and locating disaster areas, while surface water and ocean topography are able to detect changes in water surface elevation, providing a direct view of the flood's impact on the water area and surface elevation. The datasets both provide information about the extent of the flood area, which is more detailed than that provided by a single source. Furthermore, the NPP-VIIRS data not only reflects the indirect impact of the disaster on the lives and production of the local population, but also provides an intuitive assessment of the damage to the Zaporizhzhya nuclear power plant's power supply. The data showed that the disaster affected an area of about 8 km on either side of the river downstream of the dam. This enables the prediction of the damage and the postdisaster reconstruction strategy from a humanistic perspective. The combination of these three data types enables the specific impact of the disaster to be gauged in terms of its scope, extent, impact on human life, and the postdisaster recovery situation. This provides a scientific reference for the timely formulation and implementation of disaster relief and postdisaster reconstruction measures.

Pansharpening is the process of integrating two heterogeneous remote sensing images to obtain high-resolution multispectral images, which is crucial for downstream tasks. Existing methods utilizing advanced deep-learning techniques are able to achieve good sharpening results. However, the heterogeneity between diverse source images is not sufficiently considered, which in turn results in distortions in the sharpening results. Addressing this gap, we have developed a multibranch pyramid structure, which can build bridges between diverse source images at various scales. It contains three distinct branches, including the PAN branch, the MS branch, and the fusion branch, which efficiently and seamlessly integrates the data flow in distinct branches by means of the pyramid structure. Furthermore, in order to retain more advantageous information, we have developed a specialized adaptive extraction and integration module (AEIM) for each branch, namely, the texture shrinkage adaptive module for the PAN branch, the spectral information consistency module for the MS branch, and the adaptive fusion module for the fusion branch. These AEIMs are specifically designed to cater to diverse sources and distinct stages of the pansharpening process. The adaptive weights they generate can be used to extract and fuse more advantageous information. Ultimately, high-fidelity sharpening outcomes are obtained by minimizing the reconstruction errors at various scales in distinct branches. Extensive experiments show that our methodology surpasses that of representative advanced methods, while maintaining a high level of efficiency. All implementations will be published at MMAPP.

The Arctic Weather Satellite (AWS) is a single instrument mission consisting of a microwave sounder operating in the 54, 89, 183 and 325 GHz bands. The optical design of the instrument consists of a feedcluster directly illuminating a crosstrack scanning mirror, keeping the instrument compact. Due to the simple optics, the beams are not co-aligned and none are in the scanning mirror focal point, leading to beam divergence, asymmetry and scan angle dependent variations in spillover. Using Method of Moment (MoM) simulations of the optics as well as instrument structure, the complete farfield sphere of the instrument can be simulated up to 183 GHz. This paper contains detailed analysis of spillover, beam divergence and intercomparison of physical optics, MoM and nearfield antenna measurements for the mainbeam.

Remote sensing image captioning (RSIC) is a crucial task in interpreting remote sensing images (RSIs), as it involves describing their content using clear and precise natural language. However, the RSIC encounters difficulties due to the intricate structure and distinctive features of the images, such as the issue of rotational ambiguity. The existence of visually alike objects or areas can result in misidentification. In addition, prioritizing groups of objects with strong relational ties during the captioning process poses a significant challenge. To address these challenges, we propose the visual rotated position encoding transformer for RSIC. First of all, rotation-invariant features and global features are extracted using a multilevel feature extraction (MFE) module. To focus on closely related rotated objects, we design a visual rotated position encoding module, which is incorporated into the transformer encoder to model directional relationships between objects. To distinguish similar features and guide caption generation, we propose a feature enhancement fusion module consisting of feature enhancement and feature fusion. The feature enhancement component adopts a self-attention mechanism to construct fully connected graphs for object features. The feature fusion component integrates global features and word vectors to guide the caption generation process. In addition, we construct an RSI rotated object detection dataset RSIC-ROD and pretrain a rotated object detector. The proposed method demonstrates significant performance improvements on four datasets, showcasing enhanced capabilities in preserving descriptive details, distinguishing similar objects, and accurately capturing object relationships.

Semantic segmentation is a fundamental task in remote sensing image analysis. The accurate delineation of objects within such imagery serves as the cornerstone for a wide range of applications. To address this issue, edge detection, cross-modal data, large intraclass variability, and limited interclass variance must be considered. Traditional convolutional-neural-network-based models are notably constrained by their local receptive fields, Nowadays, transformer-based methods show great potential to learn features globally, while they ignore positional cues easily and are still unable to cope with multimodal data. Therefore, this work proposes an adjacent-scale multimodal fusion network (ASMFNet) for semantic segmentation of remote sensing data. ASMFNet stands out not only for its innovative interaction mechanism across adjacent-scale features, effectively capturing contextual cues while maintaining low computational complexity but also for its remarkable cross-modal capability. It seamlessly integrates different modalities, enriching feature representation. Its hierarchical scale attention (HSA) module bolsters the association between ground objects and their surrounding scenes through learning discriminative features at higher level abstractions, thereby linking the broad structural information. Adaptive modality fusion module is equipped by HSA with valuable insights into the interrelationships between cross-model data, and it assigns spatial weights at the pixel level and seamlessly integrates them into channel features to enhance fusion representation through an evaluation of modality importance via feature concatenation and filtering. Extensive experiments on representative remote sensing semantic segmentation datasets, including the ISPRS Vaihingen and Potsdam datasets, confirm the impressive performance of the proposed ASMFNet.

The rapid pace of urbanization underscores the importance of waste monitoring and management in urban planning and environmental conservation. Remote sensing technology enables the aerial observation of terrestrial and marine features, with high-resolution images revealing diverse objects. Deep learning techniques have gained prominence for enhancing waste monitoring precision and efficiency. This article surveys deep learning approaches for waste monitoring in remote sensing images, focusing on relevant datasets. It reviews existing remote sensing datasets, including those from uncrewed aerial vehicles and satellites, for monitoring solid waste and marine debris. Nine publicly available datasets are described in detail, highlighting their origins and applications. The monitoring methods include two kinds of methods: 1) semantic segmentation; and 2) object detection. Semantic segmentation focuses on pixel-level classification and boundary delineation, while object detection targets object-level localization and shape. Representative methods within these categories are explored, and benchmark results from recent studies are summarized to evaluate the performance of various techniques. The discussion addresses current limitations and suggests future research directions, aiming to assist researchers and professionals in environmental monitoring.

Marine ecosystems are significantly threatened by plastic litter, hence, effective detection and monitoring techniques are necessary. The identification and classification of ocean plastic litter is made possible by the development of remote sensing (RS) methods such as optical imaging, thermal infrared (TIR) sensing, hyperspectral imaging (HSI), multispectral imaging (MSI), and synthetic aperture radar (SAR). This article reviews the RS methods highlighting their unique strengths and limitations, emphasizing on their application contexts, data processing requirements, and potential integration for enhanced detection accuracy. It also investigates marine plastic characteristics such as size, specific gravity, spectral characteristics, thermal emissivity, surface roughness, and dielectric properties required for effective ocean plastic litter detection. The study shows different RS data collection methods, platforms used, applications, locations of study, data classes, dataset availability, benefits, and limitations. The raw data obtained by the various RS methods are susceptible to unwanted signals such as atmospheric conditions, cloud cover, and sunglint. Therefore, a detailed review of the different data preprocessing methods for RS data such as atmospheric correction, data fusion, noise reduction, cloud masking, sunglint correction, resampling, and image enhancement are performed. This contributes to enhancing the identification and monitoring of ocean plastic litter. The study further introduces a systematic review of existing ocean plastic litter detection methods based on RS, and shows the detailed process of data collection, preprocessing, data analysis, validation, mapping, and reporting. There are other factors contributing to the effective detection of ocean plastic litter using the RS methods such as sensor resolution, ground truth data, and classification methods which are discussed. Finally, the article investigates the sensor resolution, ground truth data, and classification methods. The study identifies significant gaps in the existing literature and recommends integrating multiple RS methods and optimizing preprocessing techniques to enhance ocean plastic litter detection.

Road extraction from images captured via remote sensing is a pivotal task across multiple domains, encompassing urban planning and intelligent transportation systems. In the realm of high-resolution remote sensing, traditional approaches to road extraction confront obstacles pertaining to reduced accuracy and resilience. This study introduces an innovative methodology for road extraction tailored to high-resolution remote sensing data. The devised algorithm integrates a feature enhancement attention module alongside parallel feature fusion. Specifically, the feature enhancement attention module is introduced to augment the network's capacity in discerning road-related information by analyzing feature maps produced at varying resolutions. Subsequently, during feature map extraction, the parallel feature fusion technique is employed to merge shallow and deep features sharing the same resolution, thus effectively leveraging the strengths of both to enhance the model's precision. Moreover, the network undertakes the computation of correlations among feature maps of differing resolutions as well as the entire feature map, thereby facilitating a holistic grasp of the global structure and semantic information embedded within the image. Experimental evaluations conducted on the CHN6-CUG and Massachusetts datasets substantiate that the proposed approach outperforms prevailing mainstream methods for road extraction in terms of both accuracy and processing speed.

In geospatial image analysis, domain shifts caused by differences between datasets often undermine the performance of deep learning models due to their limited generalization ability. This issue is particularly pronounced in hyperspectral imagery, given the high dimensionality of the per-pixel reflectance vectors and the complexity of the resulting deep learning models. We introduce a semisupervised domain adaptation technique that improves on the adversarial discriminative framework, incorporating a novel multiclass discriminator to address low discriminability and negative transfer issues from which current approaches suffer. Significantly, our method addresses mode collapse by incorporating limited labeled data from the target domain for targeted guidance during adaptation. In addition, we integrate an attention mechanism that focuses on challenging spatial regions for the target mode. We tested our approach on three unique hyperspectral remote sensing datasets to demonstrate its efficacy in diverse conditions (e.g., cloud shadows, atmospheric variability, and terrain). This strategy improves discrimination and reduces negative transfer in domain adaptation for geospatial image analysis.

Multiple modal data fusion can provide valuable and diverse information for remote sensing image segmentation. However, the existing fusion methods often lead to feature loss during the fusion of various modal data, and the complementarity among multimodal features is insufficient. To address these problems, we propose a multimodal gated segmentation network for remote sensing images based on the frequency decomposition. Complementary information from multimodal features is extracted by establishing a long-distance correlation between the low-frequency components of different modal data. In addition, high-frequency detailed features of different modal data are preserved by residual connection. The adaptive gated fusion method is then used to control the information flow between the complementary information and each modality feature map, enabling adaptive fusion between multimodal features. These operations can effectively improve the adaptability of the proposed method in various scenarios and data changes. Extensive experiments demonstrate that the proposed method has good effectiveness, robustness, and generalization and achieved state-of-the-art performance in several remote sensing image semantic segmentation tasks.

Building change detection on remote sensing images is an important approach to monitoring the urban expansion and sustainable development of natural resources. In conventional building change detection tasks, only changed regions between two time phases are typically concerned. The relevance and trend of spatiotemporal changes between multiple time phases are neglected in most cases. In this article, we propose a two-stage continual urbanization detector (COUD) for time series urban building change detection task. The COUD method employs self-supervised pretraining for feature refinement, and performs optimization through temporal distillation approach. Consequently, multitemporal feature extraction and changing regions localization of urban building complexes are conducted. Considering the gap in available dataset for time series change detection task, we produce and release a time series dataset named “TSCD”. Chengdu region of China is selected as the study area in this research, which is partially covered by the proposed TSCD dataset. By applying the proposed COUD method to the selected study area for exploring the changing pattern from 2016 to 2022, a comprehensive analysis is conducted in conjunction with actual planning policies published by the management department. Extensive experimental results confirm the reliability of our proposed method.

Semantic segmentation models experience a significant performance degradation due to domain shifts between the source and target domains. This issue is particularly prevalent in remote sensing imagery, where a semantic segmentation model trained on images from one satellite is tested on images from another. Previous research has often overlooked the role of data augmentation in enhancing a model's adaptability to target domains. In contrast, this article proposes a novel self-training framework that incorporates data augmentation at both the input and feature levels, yielding excellent results. Specifically, we introduce a regularized online self-training framework that effectively addresses the challenges of overconfidence and class imbalance inherent in self-training. Based on this framework, we implement two robust data augmentation strategies at the input and feature levels to facilitate the learning of cross-domain invariant knowledge. At the input level, we employ a large-scale domain mixing strategy, termed multidomain mixing, to enhance the model's generalization capability. At the feature level, we introduce masked feature augmentation, a masking-based perturbation technique applied to the semantic features of the target domain. This approach enhances the consistency of teacher–student network predictions in the target domain feature space, thereby improving the robustness of the model's recognition of target domain features. The integration of the proposed self-training framework with dual-level data augmentation culminates in our innovative self-training-based dual-level data augmentation (STDA) method. Extensive experimental results on the ISPRS semantic segmentation benchmark demonstrate that STDA outperforms existing state-of-the-art methods, showcasing its effectiveness.

In this article, we consider the issue of change detection (CD) for heterogeneous remote sensing images. Existing deep learning-based methods for CD usually utilize square convolution receptive fields, which do not sufficiently exploit the contextual and boundary information in heterogeneous images. To address the aforementioned issue, we propose an enhanced and unsupervised Siamese superpixel-based network for CD in heterogeneous remote sensing images. Our newly proposed method innovatively combines superpixels with the square receptive fields to generate the boundary adherence receptive fields and better capture the contextual information than existing methods only with the regular square receptive fields. Furthermore, we utilize an adaptive superpixel merging module to prevent the oversegmentation of superpixels and strengthen the robustness of our method in terms of superpixel sizes. Experiments based on four real datasets demonstrate that the proposed method achieves higher accuracy than other commonly used CD methods in heterogeneous remote sensing images.

Unmanned Aerial Vehicles (UAVs) equipped with remote sensing (RS) abilities have revolutionized agriculture practices, especially in the area of plant disease detection, with the incorporation of deep learning (DL) techniques. Using the UAVs to capture high-resolution images of agricultural regions, RS information becomes widely accessible for monitoring and identifying crop health and signs of diseases. DL models, specifically convolutional neural networks (CNNs), are employed to analyze these images, enabling automated recognition and detection of plant diseases with enhanced accuracy. This integration of UAV-based RS and DL models facilitates early disease detection, allowing farmers to implement timely interventions, such as targeted pesticide application or crop management strategies, ensuring food security and reducing yield losses. Moreover, this model’s scalability and efficiency improve its value for precision agriculture, optimizing resource usage and promoting sustainable farming practices. This manuscript proposes an Optimal Fuzzy Deep Neural Networks-based Plant Disease Detection and Classification (OFDNN-PDDC) technique on UAV-based Remote Sensed Data. The purpose of the OFDNN-PDDC technique is to proficiently detect and classify the distinct kinds of plant diseases. The OFDNN-PDDC technique follows a three-stage process to enhance plant disease detection performance. At the primary level, the OFDNN-PDDC technique employs an improved ShuffleNetv2 model for learning complex and intrinsic feature patterns on the RS data. Besides, the OFDNN-PDDC technique utilizes a fuzzy restricted Boltzmann machine (FRBM) model to detect plant diseases. Finally, the hyperparameter selection of the OFDNN-PDDC technique is performed by the tent chaotic salp swarm algorithm (TCSSA) model. To illustrate the superior outcomes of the OFDNN-PDDC technique, detailed experiments were accomplished by utilizing APD and CPD datasets. The experimental validation of the OFDNN-PDDC technique portrayed a superior accuracy value of 96.18% and 98.85% over existing techniques under the used datasets.

To address the challenges of object detection in complex remote sensing imagery, where the YOLO backbone network struggles with adaptive learning of feature distributions, leading to insufficient multi-scale feature learning capabilities and low detection accuracy for small and occluded objects, the lightweight Enhanced YOLOv8 with WBiFPN (Weighted Bidirectional Feature Pyramid Network) model is introduced in this paper. This model is designed to enhance multi-scale feature learning performance. It incorporates a feature fusion network based on WBiFPN and introduces the EMA (Efficient Multi-Scale Attention Module) to strengthen the representation of semantic and spatial information, thereby deepening the integration of multi-scale features. The model integrates RepConv (Re-parameterized Convolution) and ConvNeXt C2f in the shallow layers of the backbone network to optimize feature extraction, while the deeper layers include a BoT (Bottleneck Transformer Model) to further enhance multi-scale feature extraction capabilities. To reduce model parameters and computational complexity, the neck network employs a simplified Slim-Neck structure. Experimental results demonstrate that the Enhanced YOLOv8 model exhibits superior performance. Compared to the YOLOv8-n/s/m/l/x series models, the proposed model achieves mean Average Precision (mAP@0.5) of 94.8%, 91.6%, and 82.0% on the NWPU VHR-10, DIOR, and DOTA datasets, respectively, representing improvements of 3.2%, 2.5%, and 2.5%. The average inference speeds are 82 fps, 79 fps, and 76 fps, meeting the real-time requirements of inference. Furthermore, the Enhanced YOLOv8 model outperforms other mainstream models in detection performance.

This comprehensive review delves into the importance of hedgerows in urban green spaces, emphasizing their significant role in sustainable development and providing ecological benefits. Accurate identification and characterization mapping of hedgerows are vital for effective land management, urban planning, and conservation efforts. The article explores the challenges associated with identifying hedgerows in urban environments and the complexities they present for automatic detection. It discusses the limitations of traditional methods and showcases the potential of advances in remote sensing technologies and artificial intelligence (AI) methods, such as deep learning algorithms. Results indicate that deep learning can generally achieve an accuracy of 75% for hedgerow identification. This review article sets out a vision for the future of hedgerow detection and monitoring.

Accurate and rapid postearthquake structural damage assessment is of vital importance for humanitarian relief. Remote sensing techniques have the potential to map large areas with reduced data latency but are limited by several factors, including accuracy (compared to in-situ monitoring campaigns) and data acquisition frequency. Current damage assessment techniques relying on remote sensing data enable rapid assessment in situations where on-site reconnaissance is not possible or desirable. Yet, these techniques rely on different scales, measurement methods, and spatial resolutions, making it difficult to assimilate many different damage products in a homogeneous damage map. Here, we present the results of the U.K.’s Earthquake Engineering Field Investigation Team's remote-sensing-based reconnaissance mission, which was carried out in the aftermath of the series of earthquakes that struck Turkey and Syria in February 2023. We use a set of publicly available damage maps based on synthetic aperture radar, optical imaging, and ground-based reports as well as in-house developed damage products and assess their relative accuracies. We describe the process of supporting on-site reconnaissance planning by creating maps that describe the building stock and diversity of damage in southeast Turkey to assist field survey teams in selecting regions that represent a diverse sample of building typologies and damage levels. Our results show that satellite-based remote sensing damage maps disagree with each other, and extensive validation data are still required to characterize the accuracy of each method at both high and medium resolution. Finally, we provide recommendations for planning and validation of future earthquake response efforts.

In the field of remote sensing image denoising, the current mainstream methods usually only consider using clean or noisy images to guide the network in the training phase. Most of them only apply to specific types of noise, and the denoising effect is not satisfactory enough, with problems such as artifacts and noise residues. In this article, we endeavor to deal with a wide range of noise types, preserving as much detailed information in the image as possible and aiming to address the relevant limitations. Inspired by contrastive learning, we propose a remote sensing image denoising framework based on contrastive learning, named RSID-CR, which constructs positive and negative sample pairs between clean, noisy, and denoised images. Then, we construct a joint loss function consisting of reconstruction loss and contrastive regularization as a guide signal to train the denoising network, such that the denoised image is pushed closer to the clean image and farther away from the noisy image in the feature space. We conduct extensive experiments on two public datasets for five types of noise often present in remote sensing images. In addition, we validate our method using two real noisy remote sensing datasets. The experimental results indicate that our proposed method achieves satisfactory outcomes.

The conversion of agricultural lands, termed “nonagriculturalization,” poses profound threats to food security and ecological stability. Remote sensing image change detection offers an invaluable tool for monitoring this phenomenon. However, most change detection techniques prioritize image comparison over exploiting accumulated vector datasets. Additionally, many current methods are not readily applicable in practical scenarios due to inadequate model generalization capabilities and a scarcity of samples, resulting in a continued reliance on manual intervention for nonagriculturalization detection. In response, this article introduces a novel change detection approach for nonagriculturalization based on the vector data and contrastive learning. Initially, the boundary-constrained simple noniterative clustering algorithm is applied to segment two-phase images under vector data guidance. Samples are then generated using an adaptive cropping method. For early phase image samples, a collaborative validation-based sample annotation framework is employed to optimize and annotate the samples, with the purified high-quality samples serving as the training set for subsequent classification. For later-phase image samples, only those within the cropland vector polygons are retained for prediction. Building on this, a semi-supervised cross-domain contrastive learning framework is proposed for remote sensing scene classification. Ultimately, by integrating nonagriculturalization rules and postprocessing techniques, areas undergoing nonagriculturalization are further detected. Validating our methodology on Wuxi and Yangzhou datasets yielded precision rates of 91.57% and 89.21%, with recall rates of 93.68% and 90.51%, respectively. These outcomes affirm the effectiveness of our method in nonagriculturalization detection, offering robust technical support for research in this domain.

The effective monitoring of eutrophication in inland water bodies is crucial for environmental management and pollution prevention. This study conducts a comprehensive analysis of in situ hyperspectral reflectance data (400–900 nm) and the trophic state index (TSI) obtained from 365 points across ten lakes and reservoirs in Northern Vietnam to propose a trophic classification based on water reflectance spectra features and a TSI estimation model for diagnosis and assessment of lake trophic status. By analyzing the quantity of reflectance peaks and their heights, our study identifies three distinct water reflectance spectra classes corresponding to three trophic levels: mesotrophic to lightly eutrophic, highly eutrophic, and hypertrophic. This classification enables the quick identification of trophic levels directly at the in situ radiometric measurement sites. Our study demonstrates that a logarithmic function of the band ratio, ${{\mathbf{R}}_{\mathbf{rs}}}( {715} )/{{\mathbf{R}}_{\mathbf{rs}}}( {560} )$, is robust for estimating TSI (${{{\bm{R}}}^2}$ = 0.85 and 0.94; root-mean-square error = 5.0 and 3.7 in calibration and validation, respectively), particularly in algal-dominated waters. These findings represent a practical application of hyperspectral remote sensing for effective eutrophication management. They also highlight the potential use of multispectral optical imagery for monitoring eutrophication in tropical regions.

Deep learning methods, especially convolutional neural networks (CNNs) and vision transformers (ViTs), are frequently employed to perform semantic segmentation of high-resolution remotely sensed images. However, CNNs are constrained by their restricted receptive fields, while ViTs face challenges due to their quadratic complexity. Recently, the Mamba model, featuring linear complexity and a global receptive field, has gained extensive attention for vision tasks. In such tasks, images need to be serialized to form sequences compatible with the Mamba model. Numerous research efforts have explored scanning strategies to serialize images, aiming to enhance the Mamba model's understanding of images. However, the effectiveness of these scanning strategies remains uncertain. In this research, we conduct a comprehensive experimental investigation on the impact of mainstream scanning directions and their combinations on semantic segmentation of remotely sensed images. Through extensive experiments on the LoveDA, ISPRS Potsdam, ISPRS Vaihingen, and UAVid datasets, we demonstrate that no single scanning strategy outperforms others, regardless of their complexity or the number of scanning directions involved. A simple, single scanning direction is deemed sufficient for semantic segmentation of high-resolution remotely sensed images. Relevant directions for future research are also recommended.

The extraction of lake water bodies from remote sensing images is vital for water resource management, environmental protection, climate change research, disaster monitoring, and land use planning. This study presents a novel model, AMFF-LWBENet (A Lake Water Body Extraction Network based on ASPP and Multi-scale Feature Fusion), designed to tackle challenges such as insufficient spatial detail, poor edge recognition, and low anti-noise performance found in existing lake water body extraction models. The model utilizes an encoder-decoder architecture, leveraging ResNet50 for downsampling to extract deep features, which are then fused more effectively through the multi-scale dense fusion (MDF) module, incorporating depthwise separable convolution and ASPP to enhance the integration of multi-scale features. Additionally, position and channel correlations in the feature map are captured using DANet, minimizing noise interference at the lake edge and improving segmentation accuracy. Bilinear interpolation is employed to upsample the feature map, and feature fusion is achieved through the cross-layer feature fusion (CFF) module, enabling precise lake water body extraction. The datasets utilized include a self-built W-H dataset, derived from Landsat images collected between 2015 and 2023, as well as the publicly available TP dataset. The AMFF-LWBENet achieved MIoU values of 97.52% on the W-H dataset and 97.5% on the TP dataset, surpassing current state-of-the-art semantic segmentation networks. These results suggest that AMFF-LWBENet is even more effective than many other networks in extracting lake water bodies, offering significantly enhanced support for a wide range of environmental and urban planning applications.

The increasing global population has heightened the demand for food, making it crucial to optimize agricultural practices. As a versatile crop, corn plays a crucial role in global food security, particularly in countries like the United States, Brazil, China, and Argentina, where corn farming is a major economic driver. To meet the growing demand and ensure sustainable practices, innovative technologies such as Unmanned Aerial Vehicles (UAVs) are being increasingly utilized in agriculture. UAVs provide efficient monitoring and sustainable methods, enabling farmers to gather precise data on crop health and yield potential. However, to capture high-resolution images that are essential for accurate analysis, UAVs often need to fly at lower altitudes. This can lead to challenges, such as crop movement caused by the wind generated by the UAV propellers, which can affect the quality of the data collected. On the other hand, images captured at high altitudes present challenges, such as blur and low resolution, which can compromise the extraction of essential characteristics of the crop under analysis. To overcome these limitations, the authors propose the use of some techniques to improve the resolution of the post-flight image of the corn crop. Among them are the classical interpolation techniques such as Nearest Neighbor, Bilinear and Bicubic, as well as Super Resolution (SR) algorithms based on deep learning, such as MuLUT, LeRF and Real-ESRGAN. The metrics used to evaluate the quality of the resulting images were Euclidean, Peak Signal-to-Noise Ratio (PSNR), Structural Similarity Index (SSIM), Jaccard and Learned Perceptual Image Patch Similarity (LPIPS). Experimental results demonstrate significant improvements in the resolution in the images using these SR algorithms compared to traditional interpolation methods, with gains of 364.13%. Although SR techniques have been developed for other purposes, such as improving the resolution of images of people and landscapes, their considerable performance can be observed in agricultural images. These techniques can aid in the early detection of pests, diseases, and nutritional deficiencies, facilitating faster and more effective interventions in corn crops. The results obtained from this study are significant for precision agriculture since increasing the resolution of images can improve the monitoring of plant growth and health, providing faster and more effective interventions. One point to be investigated is the adaptation of these techniques to different types of crops and environmental conditions, in order to expand their potential for applica- tion in different agricultural regions. In future investigations, we hope to refine the accuracy of the proposed approaches, as well as expand the comparisons with other super-resolution algorithms. In addition, tests will be carried out with different datasets, including satellite images, to evaluate the specificity in apply these techniques in different scenarios. This will allow a more comprehensive analysis of the impact of these solutions in the agricultural area.

In post-earthquake reconstruction, the rapid and effective utilization of pre-disaster and post-disaster remote sensing data is crucial. In such scenarios, remote sensing satellite technology demonstrates its unique advantages by quickly and dynamically acquiring high-resolution imagery of extensive areas in earthquake-affected regions, efficiently capturing the immediate post-disaster situation. Change detection technology has become a key tool through these remote sensing images. This technology can automatically identify changes in damaged areas, buildings, and other critical infrastructure through the analysis of pre-disaster and post-disaster remote sensing imagery data, aiding in post-disaster reconstruction efforts. Existing remote sensing change detection methods mainly rely on Convolutional Neural Network (CNN) or Transformer for construction. Still, these methods often fail to fully balance the advantages and disadvantages of these two technologies. They are not specifically optimized for the features of the change detection task (extracting and learning features of the changed area). To address this issue, this paper fully leverages the global information processing capabilities of the Transformer and the local information capture capabilities of CNN, proposing a multi-level feature guided aggregation network model composed of multiple branches that fully integrate the respective strengths of both. The model initially captures global information from the images using the Transformer-based main network. Subsequently, it extracts local information from the images employing a custom multi-scale strip convolution module based on CNN. Subsequently, the global and local information extracted during the encoding phase is further integrated through the feature aggregation network, and the final prediction map is generated using an attention fusion module. In the experimental section, the effectiveness of the proposed algorithm is further validated through comparative experiments conducted on multiple publicly available datasets.

In the field of remote sensing, semantic segmentation of unmanned aerial vehicle (UAV) imagery is crucial for tasks such as land resource management, urban planning, precision agriculture, and economic assessment. Traditional methods use convolutional neural networks (CNNs) for hierarchical feature extraction but are limited by their local receptive fields, restricting comprehensive contextual understanding. To overcome these limitations, we propose a combination of transformer and attention mechanisms to improve object classification, leveraging their superior information modeling capabilities to enhance scene understanding. In this article, we present Swin-based focal axial attention network (SwinFAN), a U-Net framework featuring a Swin transformer as encoder, equipped with a novel decoder that introduces two new components for enhanced semantic segmentation of urban remote sensing images. The first proposed component is a guided focal-axial (GFA) attention module that combines local and global contextual information, enhancing the model's ability to discern intricate details and complex structures. The second component is an innovative attention-based feature refinement head (AFRH) designed to improve the precision and clarity of segmentation outputs through self-attention and convolutional techniques. Comprehensive experiments demonstrate that the accuracy of our proposed architecture significantly outperforms state-of-the-art models. More specifically, our method achieves mean intersection over union (mIoU) improvements of 1.9% on UAVid, 3.6% on Potsdam, 1.9% on Vaihingen, and 0.8% on LoveDA.

In object detection on remote sensing images or aerial images, high-resolution images and low relative area ratio of objects need to be solved. Usually, a high-resolution image should be split and detected separately. Then, the prediction results would be merged as a result of the complete image. Due to the overlap between adjacent image slices, non-maximum suppression (NMS) is used to suppress redundant prediction boxes during the merging process. However, redundant boxes from splitting are different from those generated by detection on a single image. Some special cases cannot be effectively solved by NMS. We have studied these cases and summarized judgment conditions to identify them. Based on NMS, we propose a new method named RODM-NMS for rotated object detection during merging slices of high-resolution images. Cases challenging for NMS are classified into two categories and addressed with specific handling methods. Compared to the traditional NMS, RODM-NMS could achieve a 3.9 improvement in AP50 during merging detection results on $640\times 640$ slices of DOTA by FCOSR. It has low requirements for the overlap gap of slices and shows a significant improvement in performance for weak models. Hence, it is more suitable for computing-constrained mobile platforms such as drones to detect objects on high-resolution images.

Precipitation inversion techniques are crucial in meteorological research, aiming to accurately detect the location and intensity of precipitation events, which is vital for extreme weather response strategies. This study introduces a new infrared precipitation inversion method, residual deep forest, which is based on a deep forest model with two key enhancements. First, a feature optimization filter module is used to ensure high correlation among features in the cascade forest, a component of the deep forest model. This optimization reduces computational burden while maintaining efficiency. Second, embedding a residual structure within the cascade forest creates the residual cascade forest, improving feature expression, information processing, and computational efficiency on high-dimensional, large-scale precipitation data. Experimental results show that residual deep forest surpasses traditional deep forest models and other classical machine learning techniques in precipitation inversion accuracy, especially in identifying high-intensity rainfall. The model achieves an average recall (AR) of 85.97%, a probability of detection of 75.46%, and a false alarm ratio of 49.13%. Ablation experiments demonstrate that the feature optimization filter module reduces training time by 3.56%, testing time by 1.24%, and memory usage by 3.97%, while the residual cascade forest module reduces these metrics by 1.30%, 0.90%, and 2.23%, respectively. The improvements in AR, probability of detection, and false alarm ratio, along with the reduction in computational resources, highlight the model's enhanced efficiency and performance. This method leverages infrared remote sensing to enhance the scope and accuracy of precipitation monitoring, reducing reliance on radar and other ground-based instruments, and significantly improving precipitation inversion accuracy while minimizing spatial and temporal complexity.

With the advancement of remote sensing technologies, multimodal image matching has become increasingly important. Multimodal remote sensing images exhibit significant differences in radiometric and geometric properties due to their distinct imaging principles and being contaminated by various types of noise. Therefore, multimodal image matching remains a challenging problem. This article proposes a novel descriptor called shape-adaptive modality independent region descriptor to achieve robust matching across modalities by leveraging local structural self-similarities within the images. The proposed method first models the noise of each modality and uses a local polynomial approximation–intersection of confidence interval to fit the local texture patterns. It then constructs the descriptor based on regional gradients extracted from local structural masks. Experimental results demonstrate the effectiveness of our approach in extracting structural features from multimodal images with good noise robustness, especially in regions with limited texture, such as mountainous and island areas. The proposed descriptor achieves high accuracy and robustness in multimodal image matching.

Heart rate is a crucial metric in health monitoring. Traditional computer vision solutions estimate cardiac signals by detecting physical manifestations of heartbeats, such as facial discoloration caused by blood oxygenation changes, from subject videos using regression methods. As continuous signals are more complex and expensive to de-noise, this study introduces an alternative approach, employing end-to-end classification models to remotely derive a discrete representation of cardiac signals from face videos. These visual cardiac signal classifiers are trained on discretized cardiac signals, a novel pre-processing method with limited precedent in health monitoring literature. Consequently, various methods to convert continuous cardiac signals into binary form are presented, and their impact on training is evaluated. An implementation of this approach, the temporal shift convolutional attention binary classifier, is presented using the regression-based convolutional attention network architecture. The classifier and a baseline regression model are trained and tested using publicly available and locally collected datasets designed for heart signal detection from face video. The model performance is then assessed based on the heart rate error from the extracted cardiac signals. Results show the proposed method outperforms the baseline on the UBFC-rPPG dataset, reducing cross-dataset root mean square error from 2.33 to 1.63 beats per minute. However, both models struggled to generalize to the PURE dataset, with root mean square errors of 12.40 and 16.29 beats per minute, respectively. Additionally, the proposed approach reduces the computational complexity of model output post-processing, enhancing its suitability for real-time applications and deployment on systems with restricted resources.