Umut Gunes Sefercik
Current Organizations / Institutes
The organizations/institutes from the scholar's latest article
Previous Organizations / Institutes
The organizations / institutes from the scholar's other articles in the past ten years
Research Subjects
Research Interests
Umut Gunes Sefercik has a total of 41 co-authors. Here are the top co-authors by number of shared publications:
- Can Atalay
Department of Geomatics Engineering, Zonguldak Bulent Ecevit University, Zonguldak 67100, Türkiye
6 shared publications - Mertcan Nazar
Gebze Teknik Üniversitesi
6 shared publications - Gürcan Büyüksalih
Institute of Marine Sciences and Management, Istanbul University, Istanbul, Türkiye
5 shared publications - Ismail Colkesen
Department of Geomatics Engineering, Gebze Technical University, Gebze, Kocaeli, Türkiye
5 shared publications Karsten Jacobsen
Leibniz University Hannover, Institute of Photogrammetry and Geoinformation, Hannover
5 shared publications- Mehmet Alkan
Department of Geomatic Engineering, Faculty of Civil Engineering, Yıldız Technical University, Istanbul, Türkiye
4 shared publications
Latest Publications
- Open Access
- 27 April 2026
- Published by Springer Nature in Precision Agriculture
- Vol. 27 (3) , 56
Evaluating transformer- and CNN-based semantic segmentation models for sunflower inflorescence identification using a UAV RGB orthomosaic
Purpose: Monitoring sunflower inflorescence development is critical for yield assessment and precision crop management. While convolutional neural networks (CNNs) have shown promise in UAV-based crop segmentation, the behavior and practical implications of recent vision transformer architectures for inflorescence-level identification and spatial pattern analysis remain insufficiently explored. This study aims to systematically evaluate transformer-based and CNN-based models for sunflower inflorescence detection and to assess their capability for field-scale spatial characterization. Methods: A high-resolution UAV orthomosaic was used to evaluate state-of-the-art transformer-based models (SegFormer, Dense Prediction Transformer (DPT), and UPerNet) and CNN-based models (U-Net, DeepLabv3+, and PSPNet). A controlled experimental framework was adopted, in which spatially disjoint training and testing subsets were extracted from the same production field to capture realistic within-field heterogeneity. All models were evaluated using standard performance metrics, including accuracy, precision, recall, F-score, and IoU. Beyond model-level performance comparison, targeted ablation analyses were conducted to examine the influence of key methodological choices, including loss function selection, patch overlap, and data augmentation strategies. In addition, explainable AI analysis (Grad-CAM) and computational cost assessments were performed. Results: Among the 15 evaluated model configurations, the DPT model with the Twins-PCPVT-Base encoder achieved the highest segmentation performance (F-score: 0.946, IoU: 0.897) and demonstrated the most stable behavior across validation and spatially disjoint testing subsets. Explainable AI analysis using Grad-CAM revealed distinct attention patterns between transformer- and CNN-based models, while computational cost analysis highlighted trade-offs between segmentation accuracy and efficiency. To enhance agronomic relevance, object-based segmentation outputs were aggregated into field-scale spatial representations using complementary inflorescence-derived indicators describing inflorescence abundance and size. In addition, a weighted head area index (WHAI) was further introduced to integrate count- and area-based information, providing a balanced, image-derived spatial descriptor of within-field variability in inflorescence development. Conclusions: Taken together, the results indicate that transformer-based semantic segmentation, when integrated with object-level spatial indicators, enables consistent and interpretable field-scale characterization of within-field variability in sunflower inflorescence development, thereby enhancing the agronomic relevance of UAV-based image analysis for precision agriculture applications.
- 27 April 2026
- Published by Springer Nature in Precision Agriculture
- Vol. 27 (3) , 56
Open AccessEvaluating transformer- and CNN-based semantic segmentation models for sunflower inflorescence identification using a UAV RGB orthomosaic
Purpose: Monitoring sunflower inflorescence development is critical for yield assessment and precision crop management. While convolutional neural networks (CNNs) have shown promise in UAV-based crop segmentation, the behavior and practical implications of recent vision transformer architectures for inflorescence-level identification and spatial pattern analysis remain insufficiently explored. This study aims to systematically evaluate transformer-based and CNN-based models for sunflower inflorescence detection and to assess their capability for field-scale spatial characterization. Methods: A high-resolution UAV orthomosaic was used to evaluate state-of-the-art transformer-based models (SegFormer, Dense Prediction Transformer (DPT), and UPerNet) and CNN-based models (U-Net, DeepLabv3+, and PSPNet). A controlled experimental framework was adopted, in which spatially disjoint training and testing subsets were extracted from the same production field to capture realistic within-field heterogeneity. All models were evaluated using standard performance metrics, including accuracy, precision, recall, F-score, and IoU. Beyond model-level performance comparison, targeted ablation analyses were conducted to examine the influence of key methodological choices, including loss function selection, patch overlap, and data augmentation strategies. In addition, explainable AI analysis (Grad-CAM) and computational cost assessments were performed. Results: Among the 15 evaluated model configurations, the DPT model with the Twins-PCPVT-Base encoder achieved the highest segmentation performance (F-score: 0.946, IoU: 0.897) and demonstrated the most stable behavior across validation and spatially disjoint testing subsets. Explainable AI analysis using Grad-CAM revealed distinct attention patterns between transformer- and CNN-based models, while computational cost analysis highlighted trade-offs between segmentation accuracy and efficiency. To enhance agronomic relevance, object-based segmentation outputs were aggregated into field-scale spatial representations using complementary inflorescence-derived indicators describing inflorescence abundance and size. In addition, a weighted head area index (WHAI) was further introduced to integrate count- and area-based information, providing a balanced, image-derived spatial descriptor of within-field variability in inflorescence development. Conclusions: Taken together, the results indicate that transformer-based semantic segmentation, when integrated with object-level spatial indicators, enables consistent and interpretable field-scale characterization of within-field variability in sunflower inflorescence development, thereby enhancing the agronomic relevance of UAV-based image analysis for precision agriculture applications.
- Open Access
- 23 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 17004-17020
Developing and Validating RGB UAV-Based NDVI Estimation Maps in a Heterogeneous Landscape Using Multiple Machine and Deep Learning Algorithms
Technological advances in unmanned aerial vehicle (UAV) systems have enhanced their spectral capabilities and enabled the acquisition of multispectral data for real-time NDVI mapping. However, the high hardware cost of multispectral sensors and the additional requirements of radiometric calibration limit their widespread use. To overcome these limitations, this study proposes innovative artificial intelligence (AI)-based approaches to estimate NDVI from low-cost RGB UAV imagery. The primary objective of this research is to generate NDVI maps using only RGB UAV data and to evaluate the performance of different machine learning (ML) and deep learning (DL) algorithms. Unlike previous studies in the literature that typically relied on a single image or a single model for NDVI estimation, eight models were employed in this study, including Linear Regression, Extreme Gradient Boosting, Light Gradient Boosting, CatBoost, Multi-Layer Perceptron (MLP), Pix2Pix, and two stacking ensemble models, to generate georeferenced NDVI maps from RGB UAV imagery. In addition to focusing on individual classes, the models were tested over a heterogeneous area containing various land use and land cover (LULC) classes, and their performance was evaluated across all classes using accuracy metrics such as the R² and RMSE, based on reference NDVI maps derived from multispectral UAV imagery. In conclusion, ML-based RGB-derived NDVI mapping, with CatBoost achieving the highest accuracy (RMSE = 0.09, R² = 0.83), offers a reliable and cost-effective alternative for vegetation monitoring; notably, stacking models showed no significant advantage over single models (p > 0.05, Kruskal–Wallis and Dunn's post hoc).
- 23 April 2026
- Published by Institute of Electrical and Electronics Engineers (IEEE) in IEEE Journal of Selected Topics in Applied Earth Observations and Remote Sensing
- Vol. 19 (19391404) , 17004-17020
Open AccessDeveloping and Validating RGB UAV-Based NDVI Estimation Maps in a Heterogeneous Landscape Using Multiple Machine and Deep Learning Algorithms
Technological advances in unmanned aerial vehicle (UAV) systems have enhanced their spectral capabilities and enabled the acquisition of multispectral data for real-time NDVI mapping. However, the high hardware cost of multispectral sensors and the additional requirements of radiometric calibration limit their widespread use. To overcome these limitations, this study proposes innovative artificial intelligence (AI)-based approaches to estimate NDVI from low-cost RGB UAV imagery. The primary objective of this research is to generate NDVI maps using only RGB UAV data and to evaluate the performance of different machine learning (ML) and deep learning (DL) algorithms. Unlike previous studies in the literature that typically relied on a single image or a single model for NDVI estimation, eight models were employed in this study, including Linear Regression, Extreme Gradient Boosting, Light Gradient Boosting, CatBoost, Multi-Layer Perceptron (MLP), Pix2Pix, and two stacking ensemble models, to generate georeferenced NDVI maps from RGB UAV imagery. In addition to focusing on individual classes, the models were tested over a heterogeneous area containing various land use and land cover (LULC) classes, and their performance was evaluated across all classes using accuracy metrics such as the R² and RMSE, based on reference NDVI maps derived from multispectral UAV imagery. In conclusion, ML-based RGB-derived NDVI mapping, with CatBoost achieving the highest accuracy (RMSE = 0.09, R² = 0.83), offers a reliable and cost-effective alternative for vegetation monitoring; notably, stacking models showed no significant advantage over single models (p > 0.05, Kruskal–Wallis and Dunn's post hoc).
- Open Access
- 27 March 2026
- Published by Afyon Kocatepe Universitesi Fen Ve Muhendislik Bilimleri Dergisi in Afyon Kocatepe University Journal of Sciences and Engineering
- Vol. 26 (2) , 393-407
Improving Digital Twin and Virtual Reality Performance in Challenging Land Classes by UAV and TLS Data Fusion
The creation of realistic three-dimensional (3D) models of objects has become a subject of significant interest and attention in both academic and industrial contexts. The generation of a virtual representation of physical reality within a digital environment can be achieved by utilizing high-quality and precise 3D textured mesh models derived from aerial photos obtained through the latest technological advancements, including unmanned aerial vehicle (UAV) systems. Optical UAV systems have become an indispensable tool in 3D modeling; however, due to operating on the basis of passive remote sensing, their 3D modeling performance may be decreased in dense vegetation, forest understory, and structures, including bridges, viaducts, and underpasses, in which some parts cannot be comprehensively visualized. In this study, a solution based on the fusion of different techniques is proposed to enhance precise 3D digital twin generation, hence improving virtual reality (VR) performance in challenging land classes. Accordingly, a methodology has been developed for combining optical UAV data collected in the above-mentioned areas, where 3D modeling performance may be degraded, with terrestrial laser scanning (TLS) data collected during the field survey in the same areas to achieve high geometric accuracy and generate 3D digital twins that are highly realistic in terms of geometric and spectral features. The proposed methodology was successfully employed to generate realistic digital twins of objects with high accuracy, achieving a geometric accuracy of 3.8 cm as the average root mean square error (RMSE). Finally, high-accuracy digital twins were successfully integrated into the VR environment using the Unity game engine. This resulted in a realistic and interactive VR experience, allowing users to explore the environment and view the precise 3D digital twins in high detail. This approach enhances the interpretability of spatial data and facilitates user engagement with complex spatial information within an immersive virtual world.
- 27 March 2026
- Published by Afyon Kocatepe Universitesi Fen Ve Muhendislik Bilimleri Dergisi in Afyon Kocatepe University Journal of Sciences and Engineering
- Vol. 26 (2) , 393-407
Open AccessImproving Digital Twin and Virtual Reality Performance in Challenging Land Classes by UAV and TLS Data Fusion
The creation of realistic three-dimensional (3D) models of objects has become a subject of significant interest and attention in both academic and industrial contexts. The generation of a virtual representation of physical reality within a digital environment can be achieved by utilizing high-quality and precise 3D textured mesh models derived from aerial photos obtained through the latest technological advancements, including unmanned aerial vehicle (UAV) systems. Optical UAV systems have become an indispensable tool in 3D modeling; however, due to operating on the basis of passive remote sensing, their 3D modeling performance may be decreased in dense vegetation, forest understory, and structures, including bridges, viaducts, and underpasses, in which some parts cannot be comprehensively visualized. In this study, a solution based on the fusion of different techniques is proposed to enhance precise 3D digital twin generation, hence improving virtual reality (VR) performance in challenging land classes. Accordingly, a methodology has been developed for combining optical UAV data collected in the above-mentioned areas, where 3D modeling performance may be degraded, with terrestrial laser scanning (TLS) data collected during the field survey in the same areas to achieve high geometric accuracy and generate 3D digital twins that are highly realistic in terms of geometric and spectral features. The proposed methodology was successfully employed to generate realistic digital twins of objects with high accuracy, achieving a geometric accuracy of 3.8 cm as the average root mean square error (RMSE). Finally, high-accuracy digital twins were successfully integrated into the VR environment using the Unity game engine. This resulted in a realistic and interactive VR experience, allowing users to explore the environment and view the precise 3D digital twins in high detail. This approach enhances the interpretability of spatial data and facilitates user engagement with complex spatial information within an immersive virtual world.
- 25 March 2026
- Published by Turkish Journal of Remote Sensing in Turkish Journal of Remote Sensing
- Vol. 8, 1-12
Detection of reclamation-induced ground deformation with GACOS-corrected StaMPS PSI: Case study Hatay Airport
Many coastal cities around the globe have initiated construction projects in reclaimed areas with the objective of constructing airports and expanding air transportation services. The looseness of soil structures in reclaimed lands can result in deformations in infrastructure and structures above. It is therefore crucial to implement effective detection and monitoring procedures to identify and assess such deformations. In this study, 151 Sentinel-1A single-look complex (SLC) images, obtained between 2 December 2017 and 29 January 2023, were employed for the purpose of detecting reclamation-induced ground deformation at Hatay Airport using persistent scatterer interferometry (PSI), a multi-temporal interferometric synthetic aperture radar (MT-InSAR) method, through the Stanford Method for Persistent Scatterers (StaMPS) software. Furthermore, the atmospheric phase delay contributions were corrected through the implementation of the Generic Atmospheric Correction Online Service for InSAR (GACOS) data using the Toolbox for Reducing Atmospheric InSAR Noise (TRAIN). The average line-of-sight (LOS) deformation velocity across the area was determined to be between -6.25 mm/year and 9.02 mm/year, with fluctuations observed. The deformation trend observed on the northern and southern sides exhibited disparate directionalities with respect to the LOS. Areas exhibiting a deficiency in PS points relative to the remainder of the dataset may be attributed to the presence of low coherence signals. The application of StaMPS PSI for the monitoring of ground deformation demonstrated an overall accuracy of sub-centimeter measurements in a reclaimed airport area. Overall, the application of MT-InSAR techniques, particularly StaMPS PSI combined with GACOS corrections, enabled the robust detection and monitoring of ground deformation patterns with remarkable precision. This advanced approach is essential for maintenance planning, safety management, and risk mitigation in critical infrastructures such as airports. Early identification of deformation allows for timely intervention, helping to safeguard both structural integrity and operational continuity, thus contributing to the long-term safety and resilience of these strategically important facilities.
- 25 March 2026
- Published by Turkish Journal of Remote Sensing in Turkish Journal of Remote Sensing
- Vol. 8, 1-12
Detection of reclamation-induced ground deformation with GACOS-corrected StaMPS PSI: Case study Hatay Airport
Many coastal cities around the globe have initiated construction projects in reclaimed areas with the objective of constructing airports and expanding air transportation services. The looseness of soil structures in reclaimed lands can result in deformations in infrastructure and structures above. It is therefore crucial to implement effective detection and monitoring procedures to identify and assess such deformations. In this study, 151 Sentinel-1A single-look complex (SLC) images, obtained between 2 December 2017 and 29 January 2023, were employed for the purpose of detecting reclamation-induced ground deformation at Hatay Airport using persistent scatterer interferometry (PSI), a multi-temporal interferometric synthetic aperture radar (MT-InSAR) method, through the Stanford Method for Persistent Scatterers (StaMPS) software. Furthermore, the atmospheric phase delay contributions were corrected through the implementation of the Generic Atmospheric Correction Online Service for InSAR (GACOS) data using the Toolbox for Reducing Atmospheric InSAR Noise (TRAIN). The average line-of-sight (LOS) deformation velocity across the area was determined to be between -6.25 mm/year and 9.02 mm/year, with fluctuations observed. The deformation trend observed on the northern and southern sides exhibited disparate directionalities with respect to the LOS. Areas exhibiting a deficiency in PS points relative to the remainder of the dataset may be attributed to the presence of low coherence signals. The application of StaMPS PSI for the monitoring of ground deformation demonstrated an overall accuracy of sub-centimeter measurements in a reclaimed airport area. Overall, the application of MT-InSAR techniques, particularly StaMPS PSI combined with GACOS corrections, enabled the robust detection and monitoring of ground deformation patterns with remarkable precision. This advanced approach is essential for maintenance planning, safety management, and risk mitigation in critical infrastructures such as airports. Early identification of deformation allows for timely intervention, helping to safeguard both structural integrity and operational continuity, thus contributing to the long-term safety and resilience of these strategically important facilities.
- Open Access
DInSAR ve MT-DInSAR Teknolojileri ile Afet Erken Uyarı, Tespit, İzleme ve Yönetimi
Günümüzde, afetlerin engellenmesi veya hızlı müdahale edilmesi amacıyla afet erken uyarı, tespit, izleme ve yönetme üzerine birçok bilimsel çalışma yapılmaktadır. 21. yy ile beraber bu çalışmalarda, uzaktan algılama teknolojilerinden elde edilen verilerin kullanımı vazgeçilmez hale gelmiştir. Özellikle son on yılda, yüksek çözünürlük ve geniş kapsama alanı kabiliyetine sahip bazı uydulara ait verilerin ücretsiz sunulması ve insansız hava aracı teknolojisinde yaşanan gelişmeler afet erken uyarı, tespit, izleme ve yönetme faaliyetlerinde uzaktan algılama verilerinin daha efektif kullanımını beraberinde getirmiştir. Bu doğrultuda kullanılan uzaktan algılama teknolojilerinin başında interferometrik yapay açıklıklı radar (InSAR) gelmektedir. Yapay açıklıklı radar (SAR) teknolojisinin üç boyutlu (3B) tasvir yöntemi olan InSAR, hedef alanda yüksek kalitede dijital yüzey modellerinin ve deformasyon haritalarının üretimine olanak verir. InSAR, deprem, heyelan, tasman, volkanik aktivite vb. afetlerin merkez üssü, etki miktarı ve yayılım alanı gibi önemli metrikleri hızlı elde edebilme imkanı sunmaktadır. InSAR teknolojisi ile afet tespit, izleme ve yönetim çalışmalarında diferansiyel InSAR (DInSAR) ve çok zamanlı DInSAR (MT-DInSAR) yöntemleri kullanılmaktadır. MT-DInSAR yönteminin en sık tercih edilenleri ise kısa baz uzunluğu altkümesi (SBAS), sürekli saçıcılar interferometrisi (PSI) ve SAR tomografi (TomoSAR)’dır. Bu derlemede, afet izleme çalışmalarında kullanılan InSAR teknikleri ele alınmış ve literatürde deprem, heyelan, tasman, volkanik aktivite ve sel konularında yapılmış önemli çalışmalarda InSAR kullanım gerekçeleri ve ulaşılan sonuçlar sunulmuştur.
Open AccessDInSAR ve MT-DInSAR Teknolojileri ile Afet Erken Uyarı, Tespit, İzleme ve Yönetimi
Günümüzde, afetlerin engellenmesi veya hızlı müdahale edilmesi amacıyla afet erken uyarı, tespit, izleme ve yönetme üzerine birçok bilimsel çalışma yapılmaktadır. 21. yy ile beraber bu çalışmalarda, uzaktan algılama teknolojilerinden elde edilen verilerin kullanımı vazgeçilmez hale gelmiştir. Özellikle son on yılda, yüksek çözünürlük ve geniş kapsama alanı kabiliyetine sahip bazı uydulara ait verilerin ücretsiz sunulması ve insansız hava aracı teknolojisinde yaşanan gelişmeler afet erken uyarı, tespit, izleme ve yönetme faaliyetlerinde uzaktan algılama verilerinin daha efektif kullanımını beraberinde getirmiştir. Bu doğrultuda kullanılan uzaktan algılama teknolojilerinin başında interferometrik yapay açıklıklı radar (InSAR) gelmektedir. Yapay açıklıklı radar (SAR) teknolojisinin üç boyutlu (3B) tasvir yöntemi olan InSAR, hedef alanda yüksek kalitede dijital yüzey modellerinin ve deformasyon haritalarının üretimine olanak verir. InSAR, deprem, heyelan, tasman, volkanik aktivite vb. afetlerin merkez üssü, etki miktarı ve yayılım alanı gibi önemli metrikleri hızlı elde edebilme imkanı sunmaktadır. InSAR teknolojisi ile afet tespit, izleme ve yönetim çalışmalarında diferansiyel InSAR (DInSAR) ve çok zamanlı DInSAR (MT-DInSAR) yöntemleri kullanılmaktadır. MT-DInSAR yönteminin en sık tercih edilenleri ise kısa baz uzunluğu altkümesi (SBAS), sürekli saçıcılar interferometrisi (PSI) ve SAR tomografi (TomoSAR)’dır. Bu derlemede, afet izleme çalışmalarında kullanılan InSAR teknikleri ele alınmış ve literatürde deprem, heyelan, tasman, volkanik aktivite ve sel konularında yapılmış önemli çalışmalarda InSAR kullanım gerekçeleri ve ulaşılan sonuçlar sunulmuştur.
- Open Access
- 19 May 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-M-6, 67-72
NDVI Prediction with RGB UAV Imagery Utilizing Advanced Machine Learning Regression Models
The ever-evolving technology has significantly affected the sensors used in UAV cameras and has played an important role in the expansion of the application areas of hobbyist and commercial UAVs. In particular, UAVs with multispectral (MS) cameras, which have the potential to detect a wide range of spectral information, are widely used in many popular research areas such as precision agriculture and forestry. However, despite their advanced capabilities, the high cost of these technologies limits their accessibility for basic users. In this study, the agricultural potential of RGB UAVs, which have a much wider user base due to their lower cost, was investigated by predicting the Normalized Difference Vegetation Index (NDVI), which is widely preferred for plant classification, growth and health monitoring. In the literature, RGB camera-based NDVI prediction studies involving machine learning and deep learning algorithms have focused on the correlation of the results with the reference data (R2) or the model accuracy of the algorithms used. The approaches applied have generally been tested on single photographs or solely on vegetation areas. In this study, using the MS UAV NDVI map as reference, a comprehensive evaluation approach was applied where each pixel of the NDVI prediction maps produced by categorical boosting (CatBoost), light gradient boosting machine (LightGBM) and a stacking ensemble learning model obtained from the combination of both algorithms, whose performance in NDVI estimation has not been tested extensively before. The models were tested in an urban area with numerous buildings and a large study area with dense vegetation. The performance of the NDVI maps was analyzed using R2, Root Mean Square Error (RMSE), Normalized Median Absolute Deviation (NMAD) and Standard Deviation (STD) metrics. As a result of the comprehensive analysis, it was found that the models performed similarly in general, but the LightGBM model was slightly behind the others. The considerable results around 0.81–0.83 as R2 and ~0.09 as RMS and STD clearly showed that RGB cameras can be a lower-cost alternative solution for generating NDVI maps in agricultural studies when supported by machine learning models.
- 19 May 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-M-6, 67-72
Open AccessNDVI Prediction with RGB UAV Imagery Utilizing Advanced Machine Learning Regression Models
The ever-evolving technology has significantly affected the sensors used in UAV cameras and has played an important role in the expansion of the application areas of hobbyist and commercial UAVs. In particular, UAVs with multispectral (MS) cameras, which have the potential to detect a wide range of spectral information, are widely used in many popular research areas such as precision agriculture and forestry. However, despite their advanced capabilities, the high cost of these technologies limits their accessibility for basic users. In this study, the agricultural potential of RGB UAVs, which have a much wider user base due to their lower cost, was investigated by predicting the Normalized Difference Vegetation Index (NDVI), which is widely preferred for plant classification, growth and health monitoring. In the literature, RGB camera-based NDVI prediction studies involving machine learning and deep learning algorithms have focused on the correlation of the results with the reference data (R2) or the model accuracy of the algorithms used. The approaches applied have generally been tested on single photographs or solely on vegetation areas. In this study, using the MS UAV NDVI map as reference, a comprehensive evaluation approach was applied where each pixel of the NDVI prediction maps produced by categorical boosting (CatBoost), light gradient boosting machine (LightGBM) and a stacking ensemble learning model obtained from the combination of both algorithms, whose performance in NDVI estimation has not been tested extensively before. The models were tested in an urban area with numerous buildings and a large study area with dense vegetation. The performance of the NDVI maps was analyzed using R2, Root Mean Square Error (RMSE), Normalized Median Absolute Deviation (NMAD) and Standard Deviation (STD) metrics. As a result of the comprehensive analysis, it was found that the models performed similarly in general, but the LightGBM model was slightly behind the others. The considerable results around 0.81–0.83 as R2 and ~0.09 as RMS and STD clearly showed that RGB cameras can be a lower-cost alternative solution for generating NDVI maps in agricultural studies when supported by machine learning models.
- Open Access
- 19 May 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-M-6, 265-272
Quality Comparison of Free-of-Charge Global DSMs: ALOS AW3D30, TanDEM-X 12 m, TanDEM-X EDEM, TanDEM-X 90 m, and Sentinel-1
Digital Surface Model (DSM), with the height of the visible surface, is the most detailed digital 3D cartographic representation of Earth visible surface including all natural and man-made objects with X, Y planimetric coordinates and altitude Z. DSMs are frequently used in many areas such as agriculture, forestry, architecture, archaeology, disaster monitoring, especially in mapping, and this situation brings with it a high demand. In order to meet these demands, many DSM generation techniques have been developed based on the use of remote sensing technologies. Satellite technologies are used in the generation of DSMs of large areas such as provinces, countries, and continents due to the limited availability of airborne technologies. In this context, the global DSM concept, which has the potential to cover the whole world, has emerged with the beginning of the 2000s. Global DSMs have advantages and disadvantages against each other depending on the sensing principle and properties of the images used for their generation. This study aims to compare the qualities of the mostly preferred free global DSMs worldwide, produced with different sensing principles and image properties, with reference model-based visual and statistical analyses. In the analyses, the qualities of the commercial (free for scientific use) TanDEM-X 12 m (TDM12) and the free ALOS World 3D 30 m (AW3D30), TanDEM-X 30 m edited DEM (TDM EDEM) and TanDEM-X 90 m (TDM90) global DSMs were evaluated using a photogrammetric reference DSM with 5 m grid. Sentinel-1 DSM was also generated with 30 m point spacing and included in the investigations due to availability of free scenes worldwide which can be used for the generation of DSMs globally. The results demonstrated that the quality of the evaluated DSMs was TDM12 > TDM EDEM > AW3D30 > Sentinel-1 > TDM90. Generated height error maps which illustrate the pixel-based height differences between evaluated and reference DSMs revealed all the incoherent areas.
- 19 May 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-M-6, 265-272
Open AccessQuality Comparison of Free-of-Charge Global DSMs: ALOS AW3D30, TanDEM-X 12 m, TanDEM-X EDEM, TanDEM-X 90 m, and Sentinel-1
Digital Surface Model (DSM), with the height of the visible surface, is the most detailed digital 3D cartographic representation of Earth visible surface including all natural and man-made objects with X, Y planimetric coordinates and altitude Z. DSMs are frequently used in many areas such as agriculture, forestry, architecture, archaeology, disaster monitoring, especially in mapping, and this situation brings with it a high demand. In order to meet these demands, many DSM generation techniques have been developed based on the use of remote sensing technologies. Satellite technologies are used in the generation of DSMs of large areas such as provinces, countries, and continents due to the limited availability of airborne technologies. In this context, the global DSM concept, which has the potential to cover the whole world, has emerged with the beginning of the 2000s. Global DSMs have advantages and disadvantages against each other depending on the sensing principle and properties of the images used for their generation. This study aims to compare the qualities of the mostly preferred free global DSMs worldwide, produced with different sensing principles and image properties, with reference model-based visual and statistical analyses. In the analyses, the qualities of the commercial (free for scientific use) TanDEM-X 12 m (TDM12) and the free ALOS World 3D 30 m (AW3D30), TanDEM-X 30 m edited DEM (TDM EDEM) and TanDEM-X 90 m (TDM90) global DSMs were evaluated using a photogrammetric reference DSM with 5 m grid. Sentinel-1 DSM was also generated with 30 m point spacing and included in the investigations due to availability of free scenes worldwide which can be used for the generation of DSMs globally. The results demonstrated that the quality of the evaluated DSMs was TDM12 > TDM EDEM > AW3D30 > Sentinel-1 > TDM90. Generated height error maps which illustrate the pixel-based height differences between evaluated and reference DSMs revealed all the incoherent areas.
- Open Access
- 19 May 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-M-6, 259-264
Analysis of Coherence Between Linear and Cubic SBAS Displacement Models in Deformation Monitoring
Land reclamation projects are a common feature of coastal cities around the world and have been carried out for a number of different purposes. The rapid growth of the global population and the concomitant increase in socioeconomic developments have resulted in a corresponding rise in demand for air transportation services. The construction of airports on reclaimed areas may result in an increased susceptibility to ground subsidence, which could potentially lead to damage to the infrastructure and safety concerns for the public. Consequently, traditional ground deformation methodologies have been extensively employed in small-scale applications, although their spatial resolution is restricted and they require a considerable input of labor and time. In order to monitor ground deformation at reclaimed airports, interferometric synthetic aperture radar (InSAR) technology has become an invaluable tool. Furthermore, the small baseline subset (SBAS) method, a multi-temporal InSAR (MT-InSAR) approach, was developed for the periodic monitoring of ground deformation with sub-centimeter accuracy. This was achieved through the utilization of a SAR dataset with small baselines, thereby enhancing the quality of interferograms. The present study comprises a coherence analysis of linear and cubic SBAS displacement models, with ground deformation monitoring at Hatay Airport based on 151 Sentinel-1A single-look complex (SLC) SAR images acquired between 2 December 2017 and 29 January 2023. Consequently, the mean deformation velocity and the cumulative deformation values were determined in the satellite line-of-sight (LOS) direction. Cumulative vertical deformation maps were generated for both the linear and cubic SBAS displacement models, with values ranging from −142.71 mm to 60.11 mm and from −224.84 mm to 69.9 mm, respectively. The coherence analysis yielded a standard deviation of ±8.525 mm after bias elimination, for the differences in cumulative vertical deformation between the linear and cubic SBAS displacement models. A cumulative vertical deformation difference map was ultimately constructed, with values ranging from −42 mm to 42 mm.
- 19 May 2025
- Published by Copernicus GmbH in The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences
- Vol. XLVIII-M-6, 259-264
Open AccessAnalysis of Coherence Between Linear and Cubic SBAS Displacement Models in Deformation Monitoring
Land reclamation projects are a common feature of coastal cities around the world and have been carried out for a number of different purposes. The rapid growth of the global population and the concomitant increase in socioeconomic developments have resulted in a corresponding rise in demand for air transportation services. The construction of airports on reclaimed areas may result in an increased susceptibility to ground subsidence, which could potentially lead to damage to the infrastructure and safety concerns for the public. Consequently, traditional ground deformation methodologies have been extensively employed in small-scale applications, although their spatial resolution is restricted and they require a considerable input of labor and time. In order to monitor ground deformation at reclaimed airports, interferometric synthetic aperture radar (InSAR) technology has become an invaluable tool. Furthermore, the small baseline subset (SBAS) method, a multi-temporal InSAR (MT-InSAR) approach, was developed for the periodic monitoring of ground deformation with sub-centimeter accuracy. This was achieved through the utilization of a SAR dataset with small baselines, thereby enhancing the quality of interferograms. The present study comprises a coherence analysis of linear and cubic SBAS displacement models, with ground deformation monitoring at Hatay Airport based on 151 Sentinel-1A single-look complex (SLC) SAR images acquired between 2 December 2017 and 29 January 2023. Consequently, the mean deformation velocity and the cumulative deformation values were determined in the satellite line-of-sight (LOS) direction. Cumulative vertical deformation maps were generated for both the linear and cubic SBAS displacement models, with values ranging from −142.71 mm to 60.11 mm and from −224.84 mm to 69.9 mm, respectively. The coherence analysis yielded a standard deviation of ±8.525 mm after bias elimination, for the differences in cumulative vertical deformation between the linear and cubic SBAS displacement models. A cumulative vertical deformation difference map was ultimately constructed, with values ranging from −42 mm to 42 mm.
- Open Access
- 26 March 2025
- Published by Turkish Journal of Remote Sensing and GIS in Turkish Journal of Remote Sensing and GIS
- Vol. 6 (1) , 57-72
Ayçiçeği Çiçek Tablalarının Mask R-CNN Derin Sinir Ağı ile RGB ve Multispektral İHA Ortomozaiklerinden Örnek Segmentasyonu
Ayçiçeği, ülkemiz için önemli bir yağlı tohum kaynağı olup, büyük ve kendine özgü çiçek tablasıyla diğer tarımsal bitkilerden ayrılır. Ayçiçeği tablasının doğru bir şekilde tespit edilmesi, verim tahmini ve sürdürülebilir üretim planlaması açısından çok önemlidir. Bu çalışmanın temel amacı, derin öğrenme tabanlı Mask R-CNN modelinin RGB ve multispektral İHA ortomozaiklerinden ayçiçeği çiçek tablasını tespit etme performansının değerlendirilmesidir. Bu amaçla, Sakarya'nın Arifiye ilçesindeki çalışma alanı üzerinde İHA hava fotoğrafları toplanmış ve yüksek çözünürlüklü ortomozaikler üretilmiştir. ResNet-50 omurgalı Maske R-CNN modeli, RGB ve multispektral İHA ortomozaiklerinden manuel olarak etiketlenmiş veri kümeleri üzerinde eğitilmiş ve test edilmiştir. Sonuçlar, modelin multispektral verilerle daha yüksek doğruluk elde ettiğini, tahmin edilen ve gerçek tespitler arasındaki örtüşmeyi ölçen 0,50 birleşim üzerinde kesişim eşiğinde %94,74 Ortalama Hassasiyete ve %96,73 F1 Skoruna ulaştığını göstermektedir. RGB verileriyle, model en iyi performansı 0,75 eşiğinde göstererek %78,19 F1 Skoru ve %63,50 Ortalama Hassasiyet elde etmiştir. Bu bulgular, multispektral İHA verilerinin ayçiçeği çiçek tablalarının tespit edilmesinde daha etkili olduğunu, RGB İHA verilerinin ise daha iyi lokalizasyon doğruluğu sağladığını göstermektedir.
- 26 March 2025
- Published by Turkish Journal of Remote Sensing and GIS in Turkish Journal of Remote Sensing and GIS
- Vol. 6 (1) , 57-72
Open AccessAyçiçeği Çiçek Tablalarının Mask R-CNN Derin Sinir Ağı ile RGB ve Multispektral İHA Ortomozaiklerinden Örnek Segmentasyonu
Ayçiçeği, ülkemiz için önemli bir yağlı tohum kaynağı olup, büyük ve kendine özgü çiçek tablasıyla diğer tarımsal bitkilerden ayrılır. Ayçiçeği tablasının doğru bir şekilde tespit edilmesi, verim tahmini ve sürdürülebilir üretim planlaması açısından çok önemlidir. Bu çalışmanın temel amacı, derin öğrenme tabanlı Mask R-CNN modelinin RGB ve multispektral İHA ortomozaiklerinden ayçiçeği çiçek tablasını tespit etme performansının değerlendirilmesidir. Bu amaçla, Sakarya'nın Arifiye ilçesindeki çalışma alanı üzerinde İHA hava fotoğrafları toplanmış ve yüksek çözünürlüklü ortomozaikler üretilmiştir. ResNet-50 omurgalı Maske R-CNN modeli, RGB ve multispektral İHA ortomozaiklerinden manuel olarak etiketlenmiş veri kümeleri üzerinde eğitilmiş ve test edilmiştir. Sonuçlar, modelin multispektral verilerle daha yüksek doğruluk elde ettiğini, tahmin edilen ve gerçek tespitler arasındaki örtüşmeyi ölçen 0,50 birleşim üzerinde kesişim eşiğinde %94,74 Ortalama Hassasiyete ve %96,73 F1 Skoruna ulaştığını göstermektedir. RGB verileriyle, model en iyi performansı 0,75 eşiğinde göstererek %78,19 F1 Skoru ve %63,50 Ortalama Hassasiyet elde etmiştir. Bu bulgular, multispektral İHA verilerinin ayçiçeği çiçek tablalarının tespit edilmesinde daha etkili olduğunu, RGB İHA verilerinin ise daha iyi lokalizasyon doğruluğu sağladığını göstermektedir.
- Open Access
- 9 August 2024
- Published by Frontiers Media SA in Frontiers in Marine Science
- Vol. 11, 1388704
Comparative analyses for determining shallow water bathymetry potential of multispectral UAVs: case study in Tavşan Island, Sea of Marmara
Recently, the use of unmanned aerial vehicles (UAVs) in bathymetric applications has become very popular due to the rapid and periodic acquisition of high spatial resolution data that provide detailed modeling of shallow water body depths and obtaining geospatial information. In UAV-based bathymetry, the sensor characteristics, imaging geometries, and the quality of radiometric and geometric calibrations of the imagery are the basic factors to achieve most reliable results. Digital bathymetric models (DBMs) that enable three-dimensional bottom topography definition of water bodies can be generated using many different techniques. In this paper, the effect of different UAV imaging bands and DBM generation techniques on the quality of bathymetric 3D modeling was deeply analyzed by visual and statistical model-based comparison approaches utilizing reference data acquired by a single-beam echosounder. In total, four different DBMs were generated and evaluated, two from dense point clouds derived from red–green–blue (RGB) single-band and multispectral (MS) five-band aerial photos, and the other two from Stumpf and Lyzenga empirical satellite-based bathymetry (SDB) adapted to UAV data. The applications were performed in the Tavşan Island located in Istanbul, Turkey. The results of statistical model-based analyses demonstrated that the accuracies of the DBMs are arranged as RGB, MS, Lyzenga, and Stumpf from higher to lower and the standard deviation of height differences are between ±0.26 m and ±0.54 m. Visual results indicate that five-band MS DBM performs best in identifying the deepest areas.
- 9 August 2024
- Published by Frontiers Media SA in Frontiers in Marine Science
- Vol. 11, 1388704
Open AccessComparative analyses for determining shallow water bathymetry potential of multispectral UAVs: case study in Tavşan Island, Sea of Marmara
Recently, the use of unmanned aerial vehicles (UAVs) in bathymetric applications has become very popular due to the rapid and periodic acquisition of high spatial resolution data that provide detailed modeling of shallow water body depths and obtaining geospatial information. In UAV-based bathymetry, the sensor characteristics, imaging geometries, and the quality of radiometric and geometric calibrations of the imagery are the basic factors to achieve most reliable results. Digital bathymetric models (DBMs) that enable three-dimensional bottom topography definition of water bodies can be generated using many different techniques. In this paper, the effect of different UAV imaging bands and DBM generation techniques on the quality of bathymetric 3D modeling was deeply analyzed by visual and statistical model-based comparison approaches utilizing reference data acquired by a single-beam echosounder. In total, four different DBMs were generated and evaluated, two from dense point clouds derived from red–green–blue (RGB) single-band and multispectral (MS) five-band aerial photos, and the other two from Stumpf and Lyzenga empirical satellite-based bathymetry (SDB) adapted to UAV data. The applications were performed in the Tavşan Island located in Istanbul, Turkey. The results of statistical model-based analyses demonstrated that the accuracies of the DBMs are arranged as RGB, MS, Lyzenga, and Stumpf from higher to lower and the standard deviation of height differences are between ±0.26 m and ±0.54 m. Visual results indicate that five-band MS DBM performs best in identifying the deepest areas.