International journal of applied earth observation and geoinformation : ITC journal最新文献

{"title":"A global analysis of SAR altimetry signals over different landcover types","authors":"Maximilian Eitel,&nbsp;Michael Schmitt","doi":"10.1016/j.jag.2025.105000","DOIUrl":"10.1016/j.jag.2025.105000","url":null,"abstract":"<div><div>Satellite radar altimeters were originally designed for water applications, but their echoes over land surfaces remain less well understood. In this study we analyze how Sentinel-3 (S3) synthetic aperture radar (SAR) altimetry waveforms respond to different surface types and what physical characteristics are encoded in the signal. To probe this, we conduct classification experiments with a feature-enhanced one-dimensional convolutional neural network (1D-CNN) and analyze its performance. Since surface type information is relevant for climate, hydrology, and biodiversity applications, understanding these signal responses shows to what extent altimetric waveforms may provide consistent class-specific information despite their large elliptical footprint and heterogeneous landscapes. This study investigates the response of Sentinel-3 altimetry waveforms to different land cover types by employing a 1D-CNN to extract land cover information, complemented by a visual analysis of waveform patterns in relation to surface structures. Our results show that information about the underlying surface is embedded in the signals and can be extracted. They further reveal the sensitivity of Sentinel-3 altimetry to variations in land cover. By enhancing our 1D-CNN model with shape-based and contextual features, it effectively captures surface characteristics despite the large altimeter footprint. An ablation study highlights the complementary role of these features, as their removal negatively impacts performance. The best-performing 1D-CNN achieves a macro-averaged F1 (Macro-F1) score of 0.57 and an overall accuracy of 0.67, outperforming both a random forest and a dummy baseline. The classification includes six surface types: Tree, Shrub, Grass, Crop, Bare/Sparse Vegetation, and Water. Although some misclassification occurs, particularly in transition zones and among classes with similar vegetation structures and soil properties, the model provides valuable insights into systematic waveform behavior, highlighting the potential of SAR altimetry signals to capture broad surface characteristics.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105000"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145659109","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"An automated method for estimating the depth of melt ponds using ICESat-2 LiDAR point cloud data: application to surface melt of Arctic sea ice","authors":"Xiaoyi Shen ,&nbsp;Haili Li ,&nbsp;Chang-Qing Ke","doi":"10.1016/j.jag.2025.105033","DOIUrl":"10.1016/j.jag.2025.105033","url":null,"abstract":"<div><div>The depth of melt ponds upon Arctic sea ice is a critical indicator for understanding the state of sea ice and the progression of climate change. The advent of the ICESat-2 lidar altimeter offers a promising avenue to monitor melt ponds, and existing studies are capable of detecting melt pond location and depth. Here we propose a novel approach that incorporates ICESat-2 photon positioning and elevation uncertainties. By converting the photon elevation distribution along the satellite track into a two-dimensional image where horizontal and vertical axes represent track distance and elevation, we employ image classification techniques to distinguish the surface and bottom of melt ponds. This provides a fully automated method for identifying melt ponds, and then the estimation of their depths. Converting photon information into images effectively simplifies the computation. We tested this method on 100 randomly distributed melt pond samples of varying sizes and depths. Out of these, 97 melt ponds were successfully identified. A comparison with manually annotated data revealed an average absolute bias of 5 cm and a correlation coefficient of 0.86, outperforming other methods. This approach can detect large and deep melt ponds with widths greater than 17 m and depths ranging from 0.3 to 2 m, which can be further used to monitor the melt ponds in the whole Arctic. It facilitates the acquisition of more detailed melt pond depth information, which is crucial for quantifying the surface melt of Arctic sea ice.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105033"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791197","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Remote sensing of a gantry-equipped facility: Optimizing accuracy by integrating SfM photogrammetry and laserscan computer graphics through fixed base model geometry","authors":"Eise W. Nota ,&nbsp;Brechtje A. van Amstel ,&nbsp;Wiebe Nijland ,&nbsp;Marcel C.G. van Maarseveen ,&nbsp;Maarten G. Kleinhans","doi":"10.1016/j.jag.2026.105098","DOIUrl":"10.1016/j.jag.2026.105098","url":null,"abstract":"<div><div>Gantry-equipped facilities are used in a wide range of research fields, for instance geomorphology. Such setups allow for frequent and comprehensive measurements from sensors mounted on gantries or in fixed overhead position. These measurements often require high spatial and relative accuracy because of small scale morphology. Our objective is to develop high-performance data processing methods for our 20 by 3 m laboratory facility that is used for physical scale experiments on coastal morphology. We compare new and existing methods through quantification of spatial and relative errors in DEMs. Our new data processing methods incorporate SfM photogrammetry and computer graphics using a “base model” of our facility under idealized conditions. With this base model, we are able to successfully align all sensor outputs, creating large and geometrically consistent timeseries of orthomosaics and DEMs. Furthermore, the base model geometry provides for a gridded laserscan processing method, incorporating fixed extrinsic, intrinsic and distortion parameters of the sensors along the gantry. Compared to SfM photogrammetrically constructed DEMs, this new method shows greatly improved relative accuracy (0.74 mm compared to 0.97–4.18 mm) and spatial accuracy (2.13 mm compared to 2.39–8.25 mm). Moreover, we show that variations of extrinsic parameters of laserscan equipment along the gantry survey are significant, while these parameters are often assumed to be constant. Our improved methods may open up new research questions in experimental geomorphology and can be applied to other facilities, research fields and industries.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105098"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146006417","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Moving vehicles tracking from satellite video data based on spatiotemporal high-order relation learning and reasoning","authors":"Ziyuan Feng ,&nbsp;Xianfeng Zhang ,&nbsp;Bo Zhou ,&nbsp;Miao Ren ,&nbsp;Xiaobo Zhi","doi":"10.1016/j.jag.2025.105015","DOIUrl":"10.1016/j.jag.2025.105015","url":null,"abstract":"<div><div>Tracking moving vehicles in satellite videos presents several challenges, including complex background interference and the difficulty of detecting small targets. Most existing multiple object tracking (MOT) methods utilize convolutional models to capture local semantics or self-attention mechanisms to address global semantics for moving target detection. However, these methods tend to struggle with small and visually similar targets, making them particularly vulnerable to complex background interference, which often results in a large number of false positives and missed detections. Furthermore, many current approaches rely on the Hungarian matching algorithm or other intricate, unlearnable association optimization methods to achieve effective tracking once relevant information is gathered. This reliance often yields suboptimal outputs from the network models. To tackle these issues, this article presents an end-to-end graph network based on spatiotemporal high-order relation learning and reasoning for vehicle tracking in satellite video. The representation module of spatial high-order relations is designed to capture the spatial high-order relations between moving vehicles and their local environments, as well as global key references. Meanwhile, the temporal semantic reasoning module focuses on analyzing the evolution of these spatial high-order relations over time, thereby constructing the spatiotemporal high-order connections among the targets of interest and ensuring the continuous and stable detection of moving vehicles. Ultimately, a graph network based on spatiotemporal high-order relation reasoning is developed to perform learnable associations of target information across video frames, achieving a globally optimal solution to the tracking problem. Comparative experiments on the SatVideoDT, CGSTL, and ShuangQing-1 satellite video datasets demonstrate that the proposed method effectively enables end-to-end tracking of moving vehicles, attaining state-of-the-art performance across most evaluation metrics. On the SatVideoDT dataset, the model achieves a Multiple Object Tracking Accuracy (MOTA) of 65.1% and an Identity F1 Score (IDF1) of 70.9%. The proposed network model holds significant promise for the automated interpretation of satellite video data. The code is available at https://github.com/zsspo/GHOST-R.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105015"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145759385","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"AI in soil moisture remote sensing","authors":"Carsten Montzka ,&nbsp;Luca Brocca ,&nbsp;Hao Chen ,&nbsp;Narendra N. Das ,&nbsp;Antara Dasgupta ,&nbsp;Mehdi Rahmati ,&nbsp;Thomas Jagdhuber","doi":"10.1016/j.jag.2025.105011","DOIUrl":"10.1016/j.jag.2025.105011","url":null,"abstract":"<div><div>Soil moisture, a pivotal component of the hydrological cycle, exerts a profound influence on land surface exchange processes, but its spatial variability poses challenges for large-scale field observations, increasing reliance on satellite-based retrievals. However, spaceborne estimates face limitations due to model uncertainties and sensor-related constraints. Recent advances in artificial intelligence (AI) offer promising alternatives to traditional methods by enabling data-driven estimation of soil moisture without strong physical assumptions. Thus, a critical review of emerging AI-based soil moisture retrieval methods with respect to their advantages and disadvantages is vital to ensure the best utilization of such tools for soil moisture sensing, especially with novel sensors and data constantly being generated.</div><div>In this comprehensive review, we furnish the first structured overview of AI methods and their applications in soil moisture retrievals from remote sensing. AI is able to enhance soil moisture retrieval by learning complex (highly nonlinear) relationships between satellite observations and ground reference data, to support time series reconstruction by filling gaps in data sets, to estimate subsurface soil moisture conditions from surface signals and auxiliary inputs, to enable spatial scaling by translating soil moisture estimates across different resolutions using multi-resolution data, to predict temporal dynamics as a soil moisture forecast, and to contribute to broader assessments of the water cycle and beyond by integrating soil moisture with further hydrological variables. Future directions for each method are also identified to address the scientific challenges of soil moisture retrieval and help focus the research community on the key open questions in the new era of rapidly expanding AI applications.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105011"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145738415","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Sim-to-real image to image translation for remote sensing fine-grained ship images using generative diffusion models","authors":"Zhiming Deng,&nbsp;Baixin Ai,&nbsp;Tianyu Zhang,&nbsp;Cheng Wei,&nbsp;Xibin Cao","doi":"10.1016/j.jag.2025.104996","DOIUrl":"10.1016/j.jag.2025.104996","url":null,"abstract":"<div><div>Advances in artificial intelligence have enabled the automation of many remote sensing tasks; however, performance remains constrained by dataset quality, especially for fine-grained ship classification in remote sensing imagery, where publicly available datasets suffer from class imbalance and sample scarcity. To address these issues, we propose a novel simulation-to-real style-transfer pipeline for fine-grained ship imagery, comprising three modules: the Simulation Image Generation (SIG) module, the Conditional Image Generation (IG) module, and the Wake Inpainting (WI) module. In the SIG module, we construct an optical remote sensing imaging system capable of producing high-resolution simulated images containing fine-grained ship objects. To overcome the loss of detailed features inherent in global style-transfer methods, the IG module introduces the SPAM-ControlNet algorithm, which generates fine-grained ship images with accurate characteristics. In the WI module, we generate the inpainting region at the stern based on the ship wake model, then apply Stable Diffusion Inpainting to synthesize realistic wake patterns, thereby harmonizing the generated ship objects with the ocean background. This pipeline enables the synthesis of seamless, high-resolution remote sensing images populated with detailed ship objects. Building on this pipeline, we also release a hybrid dataset, FGSCR-SR-12, which combines real and synthetic images across 12 ship classes to mitigate long-tail distribution challenges caused by scarce classes. All code and the FGSCR-SR-12 dataset are publicly available at <span><span>https://github.com/Slimyer/SPAM-Controlnet</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 104996"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145738349","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Towards onboard thermal hotspots segmentation with raw multispectral satellite imagery","authors":"Cristopher Castro Traba ,&nbsp;David Rijlaarsdam ,&nbsp;Jian Guo ,&nbsp;Roberto Del Prete ,&nbsp;Gabriele Meoni","doi":"10.1016/j.jag.2026.105095","DOIUrl":"10.1016/j.jag.2026.105095","url":null,"abstract":"<div><div>The rapid spread and destructive nature of wildfires and volcanic activity have intensified the need for low latency detection systems. The growing intensity and frequency of globally distributed thermal hotspots have driven the development of satellite-based detection solutions. Conventional approaches rely on ground-based processing, which limits low latency capabilities due to revisit times over ground stations and data handling requirements. This work proposes the first onboard payload processing pipeline for segmentation of thermal hotspots in raw multispectral satellite imagery. The pipeline leverages the Near InfraRed (NIR) and Short-Wave InfraRed (SWIR) spectral bands, and the combination of onboard Artificial Intelligence (AI) and raw imagery significantly reduces the delay between image acquisition and event detection. Furthermore, we present Segmentation of Thermal Hotspots in Raw Sentinel-2 data (SegTHRawS), the first publicly available dataset for thermal hotspot segmentation in raw multispectral satellite imagery. The segmentation model employed is a Fully Convolutional Network (FCN) derived from U-Net, named ResUnet-S2, designed for fast on-device inference. This model achieved an Intersection over Union (IoU) of 0.988 and an F-1 score of 0.986 on SegTHRawS, with its detection and generalization capabilities validated using an external thermal hotspot segmentation dataset. The proposed pipeline was verified on CubeSat-compatible hardware, achieving an end-to-end execution, from image acquisition to event detection, in 1.45 s, faster than the image acquisition process, and consuming a peak power of 4.05 W. These results demonstrate the potential of onboard processing solutions for minimizing the detection latency of current approaches, particularly for thermal hotspot segmentation, using edge computing satellite hardware.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105095"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145977789","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Geospatial machine learning model for limestone suitability assessment","authors":"Idowu Jane Bada ,&nbsp;Michael Adeyinka Oladunjoye ,&nbsp;Moruffdeen Adedapo Adabanija","doi":"10.1016/j.jag.2025.105055","DOIUrl":"10.1016/j.jag.2025.105055","url":null,"abstract":"<div><div>Accurate assessment of limestone quality and investment potential requires advanced techniques due to spatial variability in thickness and geochemical composition which traditional exploration methods cannot capture efficiently. This study integrates Machine Learning (ML) and Geographic Information Systems (GIS) to optimize limestone exploration. The analysis used limestone and overburden thickness, X-ray Fluorescence (XRF), and Atomic Absorption Spectroscopy (AAS) data from 23 core samples with multiple ML classifiers: Decision Tree (DT), Logistic Regression (LR), XGBoost, Support Vector Machine (SVM), and Random Forest (RF). Predictive ability was evaluated using Accuracy, F1 Scores, Precision-Recall (P-R) curves, Area under P-R curves (AUC-PR), and Feature importance Analysis. Features were validated using a non-parametric approach. Predicted datasets of the selected classifier were subjected to limestone classification criteria and integrated into a GIS to generate predictive Limestone Suitability Index (LSI) map. DT and LR models showed 100 % accuracy, XGBoost performed poorly at 60 %, and SVM and RF had moderate performance (80 %). The F1-scores of 1.00 for LR and DT, 0.71 for SVM and RF, and 0.45 for XGBoost indicate prediction reliability differences. RF and SVM achieved balanced precision-recall (0.65–0.80), with RF attaining a higher AUC_PR (0.871) than SVM (0.643). The non-parametric validation of the features identified RF as most suitable. The LSI map based on RF outputs, categorized the area into high, medium, and low potential zones with high potential zones characterized by thick, CaO rich limestone beds (16.0–34.1 m, CaO ≥ 50 %, SiO2 ≤ 8 %). This made ML, specifically, RF an essential tool for limestone resource<!--> <!-->evaluation.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105055"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145926086","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"Few-Shot change detection in optical and SAR remote sensing images for disaster response","authors":"Di Wang ,&nbsp;Guorui Ma ,&nbsp;Xiao Wang ,&nbsp;Ronghao Yang ,&nbsp;Yongxian Zhang","doi":"10.1016/j.jag.2026.105100","DOIUrl":"10.1016/j.jag.2026.105100","url":null,"abstract":"<div><div>Few-shot change detection in optical and Synthetic Aperture Radar images is a critical task for disaster monitoring. offering significant application value in complex scenarios with extremely limited labeled samples. However, the randomness of disasters causes a notable data distribution shift between public datasets and real disaster scenarios. With only a few annotated image pairs, existing methods struggle to effectively fuse features from heterogeneous images, leading to severe performance degradation. To address this challenge, we propose a Dual-Stage Training framework for Change Detection (DSTCD), specifically designed for few-shot scenarios involving fewer than 20 labeled image pairs. DSTCD first undergoes source task pre-training on a heterogeneous image registration dataset. Subsequently, in the target task stage, it leverages task guided feature transfer module to transfer the structural and semantic features of image registration to the change detection task. This mechanism significantly enriches the feature representations under few-shot conditions, enabling accurate identification of affected areas. To validate its performance, we conducted comparative and ablation studies against eleven state-of-the-art methods on four public datasets covering both urban expansion and water expansion scenarios. Experimental results demonstrate that DSTCD achieves a significant performance lead. Its average F1-score surpasses the second-best method by 6.98% in urban expansion scenarios and by 13.09% in water expansion scenarios, proving its superior performance and strong multi-scenario adaptability. Furthermore, robustness analysis of varying training sample sizes and real-world disaster application validation further confirm the method’s practicality and robustness for data-scarce disaster monitoring tasks. The code of the proposed method will be made available at <span><span>https://github.com/Lucky-DW/DSTCD</span><svg><path></path></svg></span>.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105100"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145957348","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}

{"title":"High–low frequency feature fusion network for pavement crack segmentation in complex environments","authors":"Huazhong Jin,&nbsp;Guofeng Wu,&nbsp;Yi Cen","doi":"10.1016/j.jag.2025.105048","DOIUrl":"10.1016/j.jag.2025.105048","url":null,"abstract":"<div><div>For routine road maintenance, accurate and efficient crack detection holds significant research value and practical importance. With advances in computer vision, image processing-based crack detection has achieved promising results. However, in complex environments, factors such as illumination and weather introduce noise interference, especially in high-resolution crack images where fine fracture features show heightened complexity and diversity. This often leads to degraded detection accuracy in such challenging scenarios. To address this, we propose FE-SegNeXt, a convolutional neural network segmentation framework tailored for high-resolution crack detection in complex environments. The algorithm leverages distinct characteristics of high- and low-frequency features in crack images: average pooling separates these features, while a dedicated Frequency Collaborative Enhancement Module (FCEM) independently enhances high- and low-frequency components before fusing multi-band information. This design significantly improves the model’s ability to extract subtle cracks from high-resolution images under noisy conditions. Additionally, we introduce a Locally Enhanced Feed-Forward Network (LE-FFN) to amplify perception of weak crack signals in local regions, further refining fine-grained feature extraction. Experimental results on the public datasets Sun520, Rain365, and BJN260 demonstrate that the proposed method achieves F1-scores of 61.55%, 62.57%, and 60.31%, respectively, outperforming existing crack detection algorithms.</div></div>","PeriodicalId":73423,"journal":{"name":"International journal of applied earth observation and geoinformation : ITC journal","volume":"146 ","pages":"Article 105048"},"PeriodicalIF":8.6,"publicationDate":"2026-02-01","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"145791191","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}