Pub Date : 2026-02-23DOI: 10.1109/LGRS.2026.3665522
Jiarun Yang;Xinming Wu;Guangyu Wang
Seismic karsts and channels interpretation is vital for reservoir characterization. These geobodies often coexist in seismic volumes and exhibit similar reflection features, making them challenging to distinguish using existing methods that typically interpret them separately. We propose a multiclass hybrid neural network for simultaneously detecting karsts and channels, offering more efficient and accurate results than independent detection. To address the lack of multiclass labeled data, we developed a workflow to generate a diverse synthetic training dataset. This involves constructing 3-D impedance models characterized by stratigraphic sequences with karsts and channels embedded according to geological and empirical data. To simulate diverse models, key parameters such as velocities, dimensions, and orientations of geobodies are randomly assigned within reasonable ranges based on geological context. Reflectivity models are computed from impedance models and convolved with Ricker wavelets to simulate seismic data, with noise added to enhance realism, yielding diverse seismic volumes with multiclass labels. This dataset is used to train a U-shaped hybrid neural network combining a Swin Transformer encoder with a residual module decoder. The Swin Transformer provides global context awareness and captures long-range dependencies, while the decoder ensures detailed feature restoration. The network performs multiclass segmentation, simultaneously detecting and distinguishing karsts and channels. Field applications demonstrate that the trained model can detect these geobodies with high accuracy and integrity. From the detection, 3-D meshes of the geobodies are constructed to model and analyze their structural geometries. We have made our multiclass training dataset publicly available (https://zenodo.org/records/10781510) for further research on karst and channel interpretation.
{"title":"A Multiclass Training Dataset and Hybrid Neural Network for Simultaneous Karst and Channel Detection","authors":"Jiarun Yang;Xinming Wu;Guangyu Wang","doi":"10.1109/LGRS.2026.3665522","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3665522","url":null,"abstract":"Seismic karsts and channels interpretation is vital for reservoir characterization. These geobodies often coexist in seismic volumes and exhibit similar reflection features, making them challenging to distinguish using existing methods that typically interpret them separately. We propose a multiclass hybrid neural network for simultaneously detecting karsts and channels, offering more efficient and accurate results than independent detection. To address the lack of multiclass labeled data, we developed a workflow to generate a diverse synthetic training dataset. This involves constructing 3-D impedance models characterized by stratigraphic sequences with karsts and channels embedded according to geological and empirical data. To simulate diverse models, key parameters such as velocities, dimensions, and orientations of geobodies are randomly assigned within reasonable ranges based on geological context. Reflectivity models are computed from impedance models and convolved with Ricker wavelets to simulate seismic data, with noise added to enhance realism, yielding diverse seismic volumes with multiclass labels. This dataset is used to train a U-shaped hybrid neural network combining a Swin Transformer encoder with a residual module decoder. The Swin Transformer provides global context awareness and captures long-range dependencies, while the decoder ensures detailed feature restoration. The network performs multiclass segmentation, simultaneously detecting and distinguishing karsts and channels. Field applications demonstrate that the trained model can detect these geobodies with high accuracy and integrity. From the detection, 3-D meshes of the geobodies are constructed to model and analyze their structural geometries. We have made our multiclass training dataset publicly available (<uri>https://zenodo.org/records/10781510</uri>) for further research on karst and channel interpretation.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-02-23","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147440523","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}
Global navigation satellite system reflectometry (GNSS-R) is a promising technique for retrieving soil moisture (SM), with advantages including high spatiotemporal resolution, low-cost, and low-power consumption. Compared to space-borne and airborne platforms, ground-based GNSS-R enables continuous SM monitoring in targeted regions like farmland overextended periods with high resolution. However, reflected GNSS signal penetration depth is affected by rainfall, degrading SM retrieval accuracy during precipitation. SM at a depth of 5–10 cm is a key focus of research in the agricultural field. However, the root mean square error (RMSE) of SM at a depth of 5 cm resolved by GNSS-R can reach 0.15 m3/m3 during rainy weather, which is much higher than the average accuracy level of 0.05 m3/m3 during nonrainy weather. To address this issue, we collected over one year of observational data from a ground-based GNSS-R station deployed within a farmland. In the data processing, reflectance was first calculated from intermediate frequency (IF) data. Subsequently, initial SM was retrieved from the Fresnel reflection coefficients using the Topp empirical model. Analysis revealed that precipitation events induced anomalies in the retrieved reflectance, leading to significant deviations between the GNSS-R derived SM and in situ time domain reflectometry (TDR) measurements. Leveraging this dataset, we proposed a novel ground-based GNSS-R correction algorithm integrating rainfall intensity segmentation with real-time signal-to-noise ratio (SNR) modulation. In situ TDR measurements evaluated the results. The training set RMSE improved to 0.0440 m3/m3, and the test set reached 0.0264 m3/m3.
{"title":"An Improved Ground-Based GNSS-R Soil Moisture Retrieval Algorithm Incorporating Precipitation Effects","authors":"Cheng Qian;Fan Gao;Xinyue Meng;Xiao Li;Nazi Wang;Yunqiao He;Zhenlong Fang;Zhenyao Zhong;Xuejie Wang;Yue Zhu;Lili Jing;Jiqiang Wei;Jilei Mao","doi":"10.1109/LGRS.2026.3665052","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3665052","url":null,"abstract":"Global navigation satellite system reflectometry (GNSS-R) is a promising technique for retrieving soil moisture (SM), with advantages including high spatiotemporal resolution, low-cost, and low-power consumption. Compared to space-borne and airborne platforms, ground-based GNSS-R enables continuous SM monitoring in targeted regions like farmland overextended periods with high resolution. However, reflected GNSS signal penetration depth is affected by rainfall, degrading SM retrieval accuracy during precipitation. SM at a depth of 5–10 cm is a key focus of research in the agricultural field. However, the root mean square error (RMSE) of SM at a depth of 5 cm resolved by GNSS-R can reach 0.15 m3/m3 during rainy weather, which is much higher than the average accuracy level of 0.05 m3/m3 during nonrainy weather. To address this issue, we collected over one year of observational data from a ground-based GNSS-R station deployed within a farmland. In the data processing, reflectance was first calculated from intermediate frequency (IF) data. Subsequently, initial SM was retrieved from the Fresnel reflection coefficients using the Topp empirical model. Analysis revealed that precipitation events induced anomalies in the retrieved reflectance, leading to significant deviations between the GNSS-R derived SM and in situ time domain reflectometry (TDR) measurements. Leveraging this dataset, we proposed a novel ground-based GNSS-R correction algorithm integrating rainfall intensity segmentation with real-time signal-to-noise ratio (SNR) modulation. In situ TDR measurements evaluated the results. The training set RMSE improved to 0.0440 m3/m3, and the test set reached 0.0264 m3/m3.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-02-16","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147299646","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}
Unmanned aerial vehicle (UAV) remote sensing provides an effective solution for monitoring urban infrastructure. Accurate detection of manhole covers in high-resolution UAV imagery is essential for the safe operation and maintenance of underground utility networks. However, detecting manhole covers in complex urban environments remains challenging due to their small size, visual similarity to surrounding structures, and frequent occlusion. To address these challenges, we propose a novel detection model termed manhole cover detection YOLO (MCD-YOLO). First, to exploit the regular geometric structure of manhole covers, we design an EdgeExtract module to enhance the C3k2 block in the backbone network. This module fuses image gradient information and high-frequency features to strengthen the geometric edge representation of manhole covers, thereby improving their discriminability against complex backgrounds. Second, we propose an oriented context interaction (OCI) module that employs multiorientation depthwise separable convolutions to capture both local features and global contextual dependencies, effectively suppressing interference from structurally similar background elements. Finally, we design a distribution-guided localization (DGL) module that dynamically calibrates classification confidence based on the statistical distribution of bounding box regression offsets, significantly reducing high-confidence false positives caused by localization errors under occlusion. Extensive experiments on our self-constructed manhole cover (MHC) dataset and the public VisDrone2019 dataset demonstrate the superior performance of MCD-YOLO.
{"title":"MCD-YOLO: An Improved YOLOv11 Framework for Manhole Cover Detection in UAV Imagery","authors":"Fengchang Li;Shaomei Li;Qing Xu;Zhenyan Yu;Ning You;Liunan Ren","doi":"10.1109/LGRS.2026.3663831","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3663831","url":null,"abstract":"Unmanned aerial vehicle (UAV) remote sensing provides an effective solution for monitoring urban infrastructure. Accurate detection of manhole covers in high-resolution UAV imagery is essential for the safe operation and maintenance of underground utility networks. However, detecting manhole covers in complex urban environments remains challenging due to their small size, visual similarity to surrounding structures, and frequent occlusion. To address these challenges, we propose a novel detection model termed manhole cover detection YOLO (MCD-YOLO). First, to exploit the regular geometric structure of manhole covers, we design an EdgeExtract module to enhance the C3k2 block in the backbone network. This module fuses image gradient information and high-frequency features to strengthen the geometric edge representation of manhole covers, thereby improving their discriminability against complex backgrounds. Second, we propose an oriented context interaction (OCI) module that employs multiorientation depthwise separable convolutions to capture both local features and global contextual dependencies, effectively suppressing interference from structurally similar background elements. Finally, we design a distribution-guided localization (DGL) module that dynamically calibrates classification confidence based on the statistical distribution of bounding box regression offsets, significantly reducing high-confidence false positives caused by localization errors under occlusion. Extensive experiments on our self-constructed manhole cover (MHC) dataset and the public VisDrone2019 dataset demonstrate the superior performance of MCD-YOLO.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-02-12","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147299558","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}
Pub Date : 2026-02-09DOI: 10.1109/LGRS.2026.3662463
Yinpeng Li;Xianghao Liu;Yanqi Wu;Zhuo Jia
Electrical resistivity tomography (ERT) is widely used for near-surface engineering investigations, but volume and shielding effects often blur anomaly geometry and hinder interpretation in conventional inversion sections. This letter introduces a multiscale imaging-segmentation framework, ERTSegNet, that learns an end-to-end mapping from traditional inversion images to binary anomaly masks, thereby improving the interpretability of ERT reconstructions. ERTSegNet integrates Vision Mamba modules into a UNet-style encoder–decoder with dense multiscale skip connections to capture long-range context while preserving local detail, and employs a randomized multiscale training strategy to handle varying electrode configurations. Experiments on synthetic and field data demonstrate accurate anomaly delineation and strong robustness to scale mismatch.
{"title":"Robust Recognition of Anomalous Distribution From Electrical Resistivity Tomography","authors":"Yinpeng Li;Xianghao Liu;Yanqi Wu;Zhuo Jia","doi":"10.1109/LGRS.2026.3662463","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3662463","url":null,"abstract":"Electrical resistivity tomography (ERT) is widely used for near-surface engineering investigations, but volume and shielding effects often blur anomaly geometry and hinder interpretation in conventional inversion sections. This letter introduces a multiscale imaging-segmentation framework, ERTSegNet, that learns an end-to-end mapping from traditional inversion images to binary anomaly masks, thereby improving the interpretability of ERT reconstructions. ERTSegNet integrates Vision Mamba modules into a UNet-style encoder–decoder with dense multiscale skip connections to capture long-range context while preserving local detail, and employs a randomized multiscale training strategy to handle varying electrode configurations. Experiments on synthetic and field data demonstrate accurate anomaly delineation and strong robustness to scale mismatch.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-02-09","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"147299504","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}
Pub Date : 2026-02-06DOI: 10.1109/LGRS.2026.3661420
V. D. Korchuganov;A. A. Duchkov;M. S. Golubeva
Seismic inversion is an established technique for quantitative reservoir characterization, providing estimates of subsurface elastic properties such as acoustic impedance. Since conventional inversion is typically performed independently on each trace, lateral instability of the solution remains a major challenge. A common stabilization strategy relies on horizontal-gradient penalization, which suppresses speckle noise under the assumption of horizontal stratification. However, in complex geological settings with abrupt lateral variations, such approaches may introduce secondary artifacts and oversmoothing. In this study, we propose a dip-guided regularization technique based on structure-tensor total variation (STV). The proposed regularizer incorporates local structural orientation by constructing structure tensors directly from the evolving impedance model and guiding smoothing along dominant geological directions. In contrast to conventional neighbor-trace penalization, this formulation preserves steeply dipping layers and fault-related discontinuities, yielding more stable and geologically consistent inversion results. Unlike existing structure-oriented inversion methods, the proposed approach does not require pre-computation of structural attributes from the seismic volume, as the regularization constraint is updated in situ at each iteration. On synthetic data, the proposed method reduces the RMSE by 53%, increases the correlation coefficient from 0.95 to 0.99, and improves SSIM from 0.83 to 0.89 while preserving sharp layer boundaries. On a field dataset, STV improves the correlation from 0.84 to 0.90 and reduces the RMSE by 19.4%, resulting in enhanced structural fidelity and clearer reservoir compartment delineation.
{"title":"Dip-Guided Poststack Inversion via Structure-Tensor Regularization","authors":"V. D. Korchuganov;A. A. Duchkov;M. S. Golubeva","doi":"10.1109/LGRS.2026.3661420","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3661420","url":null,"abstract":"Seismic inversion is an established technique for quantitative reservoir characterization, providing estimates of subsurface elastic properties such as acoustic impedance. Since conventional inversion is typically performed independently on each trace, lateral instability of the solution remains a major challenge. A common stabilization strategy relies on horizontal-gradient penalization, which suppresses speckle noise under the assumption of horizontal stratification. However, in complex geological settings with abrupt lateral variations, such approaches may introduce secondary artifacts and oversmoothing. In this study, we propose a dip-guided regularization technique based on structure-tensor total variation (STV). The proposed regularizer incorporates local structural orientation by constructing structure tensors directly from the evolving impedance model and guiding smoothing along dominant geological directions. In contrast to conventional neighbor-trace penalization, this formulation preserves steeply dipping layers and fault-related discontinuities, yielding more stable and geologically consistent inversion results. Unlike existing structure-oriented inversion methods, the proposed approach does not require pre-computation of structural attributes from the seismic volume, as the regularization constraint is updated in situ at each iteration. On synthetic data, the proposed method reduces the RMSE by 53%, increases the correlation coefficient from 0.95 to 0.99, and improves SSIM from 0.83 to 0.89 while preserving sharp layer boundaries. On a field dataset, STV improves the correlation from 0.84 to 0.90 and reduces the RMSE by 19.4%, resulting in enhanced structural fidelity and clearer reservoir compartment delineation.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-02-06","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223579","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}
Pub Date : 2026-02-02DOI: 10.1109/LGRS.2026.3657067
{"title":"IEEE Geoscience and Remote Sensing Letters information for authors","authors":"","doi":"10.1109/LGRS.2026.3657067","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3657067","url":null,"abstract":"","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"22 ","pages":"C3-C3"},"PeriodicalIF":4.4,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11370430","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175678","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-02-02DOI: 10.1109/LGRS.2026.3657069
{"title":"IEEE Geoscience and Remote Sensing Letters Institutional Listings","authors":"","doi":"10.1109/LGRS.2026.3657069","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3657069","url":null,"abstract":"","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"22 ","pages":"C4-C4"},"PeriodicalIF":4.4,"publicationDate":"2026-02-02","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11370317","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146175675","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-29DOI: 10.1109/LGRS.2026.3654221
Ming Li;Jiahua Zhang;Jan-Peter Weiss;John J. Braun;William Gullotta;Maggie Sleziak-Sallee
Presents corrections to the paper, Corrections to “Spire Near-Nadir GNSS-R for Sea Ice Detection: First Results”.
对“尖塔近最低点GNSS-R海冰探测：初步结果”的论文进行修正。
{"title":"Corrections to “Spire Near-Nadir GNSS-R for Sea Ice Detection: First Results”","authors":"Ming Li;Jiahua Zhang;Jan-Peter Weiss;John J. Braun;William Gullotta;Maggie Sleziak-Sallee","doi":"10.1109/LGRS.2026.3654221","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3654221","url":null,"abstract":"Presents corrections to the paper, Corrections to “Spire Near-Nadir GNSS-R for Sea Ice Detection: First Results”.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"22 ","pages":"1-3"},"PeriodicalIF":4.4,"publicationDate":"2026-01-29","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"https://ieeexplore.ieee.org/stamp/stamp.jsp?tp=&arnumber=11368646","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146082089","PeriodicalName":null,"FirstCategoryId":null,"ListUrlMain":null,"RegionNum":0,"RegionCategory":"","ArticlePicture":[],"TitleCN":null,"AbstractTextCN":null,"PMCID":"OA","EPubDate":null,"PubModel":null,"JCR":null,"JCRName":null,"Score":null,"Total":0}
Pub Date : 2026-01-28DOI: 10.1109/LGRS.2026.3657665
Fan Yang;Shufan Hu;Huilin Zhou;Yonghui Zhao;Kunwei Feng
High-resolution ground-penetrating radar (GPR) data acquired using the high-frequency antenna enables operators to identify closely spaced targets or slight variations within the subsurface, but with limited penetration depth due to the significant attenuation of electromagnetic waves. To mitigate this inherent tradeoff, we introduce a conditional generative adversarial network (cGAN) with the contrastive unpaired translation (CUT) framework, for reconstructing high-frequency, in other words, high-resolution, GPR images from low-frequency inputs. We use an enhanced U-Net with an attention mechanism as a generator to improve global context modeling. In addition, patch-level contrastive learning is integrated to ensure structural consistency between the low-frequency inputs and high-frequency outputs. After training the network using the shallow parts of a real dual-frequency dataset, we directly reconstruct the high-frequency GPR data from the low-frequency measurements. Results indicate that our proposed method surpasses the CycleGAN-based resolution enhancement method in effectively producing structurally coherent and detailed high-resolution images for both shallow and deep regions. It also demonstrates the generalizability of the proposed method, as the network does not see the data related to deep regions during the training stage. Therefore, our method offers a promising way to reconstruct high-frequency GPR data from low-frequency measurements, significantly enhancing the interpretability of deep subsurface GPR imaging outcomes.
{"title":"High-Frequency GPR Data Reconstruction With Conditional GAN and Contrastive Unpaired Translation","authors":"Fan Yang;Shufan Hu;Huilin Zhou;Yonghui Zhao;Kunwei Feng","doi":"10.1109/LGRS.2026.3657665","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3657665","url":null,"abstract":"High-resolution ground-penetrating radar (GPR) data acquired using the high-frequency antenna enables operators to identify closely spaced targets or slight variations within the subsurface, but with limited penetration depth due to the significant attenuation of electromagnetic waves. To mitigate this inherent tradeoff, we introduce a conditional generative adversarial network (cGAN) with the contrastive unpaired translation (CUT) framework, for reconstructing high-frequency, in other words, high-resolution, GPR images from low-frequency inputs. We use an enhanced U-Net with an attention mechanism as a generator to improve global context modeling. In addition, patch-level contrastive learning is integrated to ensure structural consistency between the low-frequency inputs and high-frequency outputs. After training the network using the shallow parts of a real dual-frequency dataset, we directly reconstruct the high-frequency GPR data from the low-frequency measurements. Results indicate that our proposed method surpasses the CycleGAN-based resolution enhancement method in effectively producing structurally coherent and detailed high-resolution images for both shallow and deep regions. It also demonstrates the generalizability of the proposed method, as the network does not see the data related to deep regions during the training stage. Therefore, our method offers a promising way to reconstruct high-frequency GPR data from low-frequency measurements, significantly enhancing the interpretability of deep subsurface GPR imaging outcomes.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-01-28","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223744","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}
Pub Date : 2026-01-26DOI: 10.1109/LGRS.2026.3657949
Jinrong Shen;Zhenlong Hou;Jikang Wei;Xinyang Zhao;Jiahui Wang
Downward continuation is an effective technique that enhances anomaly resolution and details. However, high-frequency noise is amplified with increasing continuation depth, which adversely affects data interpretation. A novel stable downward continuation network (SDCNet) model has been proposed to enhance the noise robustness and stability of downward continuation. By incorporating fully convolutional networks (FCNs) and Transformer to introduce topographic features and depth information, the model achieves stable downward continuation of magnetic anomaly from a surface to arbitrary target plane or surface, which improves the flexibility of existing intelligent continuation methods. Tests on the synthetic model demonstrate that the proposed method maintains continuation stability while exhibiting noise robustness compared to conventional methods. For noisy data, the maximum reductions in relative error (RE) and root-mean-square error (RMSE) reach up to 78%. Real data application over the Decorah complex in northeastern Iowa, USA, shows that the continuation results align well with the distribution of the rock mass, further validating the effectiveness and practicality of the SDCNet in downward continuation.
{"title":"SDCNet: Stable Downward Continuation of Magnetic Anomalies by Depth Information Fusion","authors":"Jinrong Shen;Zhenlong Hou;Jikang Wei;Xinyang Zhao;Jiahui Wang","doi":"10.1109/LGRS.2026.3657949","DOIUrl":"https://doi.org/10.1109/LGRS.2026.3657949","url":null,"abstract":"Downward continuation is an effective technique that enhances anomaly resolution and details. However, high-frequency noise is amplified with increasing continuation depth, which adversely affects data interpretation. A novel stable downward continuation network (SDCNet) model has been proposed to enhance the noise robustness and stability of downward continuation. By incorporating fully convolutional networks (FCNs) and Transformer to introduce topographic features and depth information, the model achieves stable downward continuation of magnetic anomaly from a surface to arbitrary target plane or surface, which improves the flexibility of existing intelligent continuation methods. Tests on the synthetic model demonstrate that the proposed method maintains continuation stability while exhibiting noise robustness compared to conventional methods. For noisy data, the maximum reductions in relative error (RE) and root-mean-square error (RMSE) reach up to 78%. Real data application over the Decorah complex in northeastern Iowa, USA, shows that the continuation results align well with the distribution of the rock mass, further validating the effectiveness and practicality of the SDCNet in downward continuation.","PeriodicalId":91017,"journal":{"name":"IEEE geoscience and remote sensing letters : a publication of the IEEE Geoscience and Remote Sensing Society","volume":"23 ","pages":"1-5"},"PeriodicalIF":4.4,"publicationDate":"2026-01-26","publicationTypes":"Journal Article","fieldsOfStudy":null,"isOpenAccess":false,"openAccessPdf":"","citationCount":null,"resultStr":null,"platform":"Semanticscholar","paperid":"146223623","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}
京公网安备 11010802042870号
京ICP备2023020795号-1
扫码关注我们
求助内容：
应助结果提醒方式：