2026
Journal Article MDPI RS Q2 IF 4.2
Costăchioiu, Teodor; Vasilescu, Vlad; Grivei, Alexandru-Cosmin; Griparis, Andreea; Faur, Daniela
In: Remote Sensing, vol. 18, no. 21, 2026, ISSN: 2072-4292.
@article{rs18213655,
title = {SENSERO: A Multi-Scale, Multi-Label Sentinel-2 Benchmark for Romania with Cross-Scale Land-Cover Classification Baselines},
author = {Teodor Costăchioiu and Vlad Vasilescu and Alexandru-Cosmin Grivei and Andreea Griparis and Daniela Faur},
url = {https://www.mdpi.com/2072-4292/18/21/3655},
doi = {10.3390/rs18213655},
issn = {2072-4292},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
journal = {Remote Sensing},
volume = {18},
number = {21},
abstract = {Patch size is an important design parameter in patch-based Earth observation (EO) learning, yet it is rarely treated as an experimental variable because existing benchmarks use a single fixed scale. We introduce SENSERO, a multi-scale, multi-label Sentinel-2 benchmark for Romania consisting of 10,000 geographically fixed locations extracted at eight patch sizes, from 64×64 to 336×336 pixels, at a constant 10 m ground sampling distance, so that all scales share identical patch centres and an identical, spatially disjoint 70/15/15 partition. Each patch provides twelve-band bottom-of-atmosphere reflectance, a rasterised CORINE Land Cover (CLC) 2018 reference map from which five deterministic multi-label annotation layers are derived (three levels of the CLC hierarchy, the BigEarthNet nomenclature, and a scheme adapted to the Romanian landscape). The benchmark task is patch-level multi-label classification; the rasterised reference maps serve to derive and trace the labels, not as a segmentation target. Using ResNet-50 baselines trained from scratch, we find that larger patches bring little benefit at CLC Level 1 but clear gains at Levels 2 and 3 (macro mAP at Level 3 rising from 0.42 at 64 px to 0.61 at 336 px), with per-location gains saturating around 224–256 pixels; that transfer across scales is asymmetric, models trained on wide patches transferring to narrow ones far better than the reverse; and that the choice of nomenclature affects measured performance as much as patch size. Leave-scene-out cross-validation, an ablation of the labelling rule, pretrained ResNet-50 and ViT-S/16 controls and bootstrap confidence intervals show that the scale effect is not attributable to spatial leakage, to the annotation rule, or to the baseline architecture. Comparison with an independent 10 m land-cover product shows that the gain from context is concentrated where CLC labels generalise over heterogeneous ground.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
Patch size is an important design parameter in patch-based Earth observation (EO) learning, yet it is rarely treated as an experimental variable because existing benchmarks use a single fixed scale. We introduce SENSERO, a multi-scale, multi-label Sentinel-2 benchmark for Romania consisting of 10,000 geographically fixed locations extracted at eight patch sizes, from 64×64 to 336×336 pixels, at a constant 10 m ground sampling distance, so that all scales share identical patch centres and an identical, spatially disjoint 70/15/15 partition. Each patch provides twelve-band bottom-of-atmosphere reflectance, a rasterised CORINE Land Cover (CLC) 2018 reference map from which five deterministic multi-label annotation layers are derived (three levels of the CLC hierarchy, the BigEarthNet nomenclature, and a scheme adapted to the Romanian landscape). The benchmark task is patch-level multi-label classification; the rasterised reference maps serve to derive and trace the labels, not as a segmentation target. Using ResNet-50 baselines trained from scratch, we find that larger patches bring little benefit at CLC Level 1 but clear gains at Levels 2 and 3 (macro mAP at Level 3 rising from 0.42 at 64 px to 0.61 at 336 px), with per-location gains saturating around 224–256 pixels; that transfer across scales is asymmetric, models trained on wide patches transferring to narrow ones far better than the reverse; and that the choice of nomenclature affects measured performance as much as patch size. Leave-scene-out cross-validation, an ablation of the labelling rule, pretrained ResNet-50 and ViT-S/16 controls and bootstrap confidence intervals show that the scale effect is not attributable to spatial leakage, to the annotation rule, or to the baseline architecture. Comparison with an independent 10 m land-cover product shows that the gain from context is concentrated where CLC labels generalise over heterogeneous ground.
Proceedings Article IGARSS2026
On the Impact of Patch Size in LLM-Based Image Captioning of Sentinel-2 Imagery
Costachioiu, Teodor; Vasilescu, Vlad; Faur, Daniela
In: IGARSS 2026 - 2026 IEEE International Geoscience and Remote Sensing Symposium, pp. 1396-1400, 2026.
@inproceedings{11730172,
title = {On the Impact of Patch Size in LLM-Based Image Captioning of Sentinel-2 Imagery},
author = {Teodor Costachioiu and Vlad Vasilescu and Daniela Faur},
doi = {10.1109/IGARSS60577.2026.11730172},
year = {2026},
date = {2026-01-01},
urldate = {2026-01-01},
booktitle = {IGARSS 2026 - 2026 IEEE International Geoscience and Remote Sensing Symposium},
pages = {1396-1400},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2025
Proceedings Article
Large Language Models with Retrieval Augmented Generation to streamline Sentinel-2 data Description
Costăchioiu, Teodor; Văduva, Corina; Faur, Daniela
In: IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium, pp. 1022-1026, 2025.
@inproceedings{11242277,
title = {Large Language Models with Retrieval Augmented Generation to streamline Sentinel-2 data Description},
author = {Teodor Costăchioiu and Corina Văduva and Daniela Faur},
doi = {10.1109/IGARSS55030.2025.11242277},
year = {2025},
date = {2025-01-01},
booktitle = {IGARSS 2025 - 2025 IEEE International Geoscience and Remote Sensing Symposium},
pages = {1022-1026},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2014
Proceedings Article
Multitemporal Satellite Image Time Series analysis of urban development in Bucharest and Ilfov areas
Costăchioiu, Teodor; Constantinescu, Rodica; Datcu, Mihai
In: 2014 10th International Conference on Communications (COMM), pp. 1-4, 2014.
@inproceedings{6866702,
title = {Multitemporal Satellite Image Time Series analysis of urban development in Bucharest and Ilfov areas},
author = {Teodor Costăchioiu and Rodica Constantinescu and Mihai Datcu},
doi = {10.1109/ICComm.2014.6866702},
year = {2014},
date = {2014-01-01},
booktitle = {2014 10th International Conference on Communications (COMM)},
pages = {1-4},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2013
Journal Article IEEE TGRS Q1 IF 2.933 (2013)
A Latent Analysis of Earth Surface Dynamic Evolution Using Change Map Time Series
Vaduva, Corina; Costachioiu, Teodor; Patrascu, Carmen; Gavat, Inge; Lazarescu, Vasile; Datcu, Mihai
In: IEEE Transactions on Geoscience and Remote Sensing, vol. 51, no. 4, pp. 2105-2118, 2013.
@article{6357329,
title = {A Latent Analysis of Earth Surface Dynamic Evolution Using Change Map Time Series},
author = {Corina Vaduva and Teodor Costachioiu and Carmen Patrascu and Inge Gavat and Vasile Lazarescu and Mihai Datcu},
doi = {10.1109/TGRS.2012.2219316},
year = {2013},
date = {2013-01-01},
urldate = {2013-01-01},
journal = {IEEE Transactions on Geoscience and Remote Sensing},
volume = {51},
number = {4},
pages = {2105-2118},
abstract = {With a continuous increase in the number of Earth Observation satellites, leading to the development of satellite image time series (SITS), the number of algorithms for land cover analysis and monitoring has greatly expanded. This paper offers a new perspective in dynamic classification for SITS. Four similarity measures (correlation coefficient, Kullback-Leibler divergence, conditional information, and normalized compression distance) based on consecutive image pairs from the data are employed. These measures employ linear dependences, statistical measures, and spatial relationships to compute radiometric, spectral, and texture changes that offer a description for the multitemporal behavior of the SITS. During this process, the original SITS is converted to a change map time series (CMTS), which removes the static information from the data set. The CMTS is analyzed using a latent Dirichlet allocation (LDA) model capable of discovering classes with semantic meaning based on the latent information hidden in the scene. This statistical method was originally used for text classification, thus requiring a word, document, corpus analogy with the elements inside the image. The experimental results were computed using 11 Landsat images over the city of Bucharest and surrounding areas. The LDA model enables us to discover a wide range of scene evolution classes based on the various dynamic behaviors of the land cover. The results are compared with the Corinne Land Cover map. However, this is not a validation method but one that adds static knowledge about the general usage of the analyzed area. In order to help the interpretation of the results, we use several studies on forms of relief, weather forecast, and very high resolution images that can explain the wide range of structures responsible for influencing the dynamic inside the resolution cell.},
keywords = {},
pubstate = {published},
tppubtype = {article}
}
With a continuous increase in the number of Earth Observation satellites, leading to the development of satellite image time series (SITS), the number of algorithms for land cover analysis and monitoring has greatly expanded. This paper offers a new perspective in dynamic classification for SITS. Four similarity measures (correlation coefficient, Kullback-Leibler divergence, conditional information, and normalized compression distance) based on consecutive image pairs from the data are employed. These measures employ linear dependences, statistical measures, and spatial relationships to compute radiometric, spectral, and texture changes that offer a description for the multitemporal behavior of the SITS. During this process, the original SITS is converted to a change map time series (CMTS), which removes the static information from the data set. The CMTS is analyzed using a latent Dirichlet allocation (LDA) model capable of discovering classes with semantic meaning based on the latent information hidden in the scene. This statistical method was originally used for text classification, thus requiring a word, document, corpus analogy with the elements inside the image. The experimental results were computed using 11 Landsat images over the city of Bucharest and surrounding areas. The LDA model enables us to discover a wide range of scene evolution classes based on the various dynamic behaviors of the land cover. The results are compared with the Corinne Land Cover map. However, this is not a validation method but one that adds static knowledge about the general usage of the analyzed area. In order to help the interpretation of the results, we use several studies on forms of relief, weather forecast, and very high resolution images that can explain the wide range of structures responsible for influencing the dynamic inside the resolution cell.
Proceedings Article
Long Satellite Image Time Series: Characterization and Information Content
Costachioiu, Teodor; Lazarescu, Vasile; Datcu, Mihai
In: Proceeding of Big Data from Space, ESA, Frascati, Italy, 2013.
@inproceedings{Costachioiu2013,
title = {Long Satellite Image Time Series: Characterization and Information Content},
author = {Teodor Costachioiu and Vasile Lazarescu and Mihai Datcu},
url = {http://www.congrexprojects.com/docs/default-source/13c10_docs/abstractbook.pdf?sfvrsn=2},
year = {2013},
date = {2013-06-01},
booktitle = {Proceeding of Big Data from Space},
publisher = {ESA},
address = {Frascati, Italy},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Proceedings Article
Costachioiu, Teodor; Vaduva, Corina; Lazarescu, Vasile; Datcu, Mihai
In: MultiTemp 2013 Conference Proceedings, pp. 1–4, IEEE Xplore, 2013.
@inproceedings{dlr88948,
title = {Content extraction of long-term Satellite Image Time Series: spatiotemporal analysis of Bucharest metropolitan area},
author = {Teodor Costachioiu and Corina Vaduva and Vasile Lazarescu and Mihai Datcu},
url = {https://elib.dlr.de/88948/},
year = {2013},
date = {2013-01-01},
booktitle = {MultiTemp 2013 Conference Proceedings},
pages = {1–4},
publisher = {IEEE Xplore},
abstract = {Current earth observation data repositories enable us to extract observations of the same geographical area over the course of time and to create satellite image time series (SITS). Such SITS are data sets of high complexity, embedding spatial, spectral and temporal information.
The rich content of SITS has generated a great interest in the recent years, many SITS analysis methods being developed. Many of these methods are focused towards improving the precision of land use/land cover detection by exploiting the inherent redundancy of SITS data.
Some other methods focus on the evolution of a specific land use/land cover type. Often these methods are applied on SITS that span over a short period of time, typical one or several years. In this paper we focus on a different use scenario of SITS analysis by extracting classes of
dynamic evolutions from long-term satellite image time series. As such we create a SITS of over 100 Landsat datasets, spanning from 1984 to 2011, covering the metropolitan area of Bucharest, Romania – perhaps the longest SITS ever being used. The analysis of this SITS aims to evaluate the spatio-temporal evolutions, focusing on finding classes of temporal dynamics of the main land use/land cover classes of the studied area, such as urban areas, agricultural land, forestry and water bodies, identifying interesting evolutions in the studied area. Furthermore, the extracted dynamic classes are used to establish relationships with known ancillary data
such as CORINE Land Cover vector data sets or statistical datasets for demographics of the studied area.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Current earth observation data repositories enable us to extract observations of the same geographical area over the course of time and to create satellite image time series (SITS). Such SITS are data sets of high complexity, embedding spatial, spectral and temporal information.
The rich content of SITS has generated a great interest in the recent years, many SITS analysis methods being developed. Many of these methods are focused towards improving the precision of land use/land cover detection by exploiting the inherent redundancy of SITS data.
Some other methods focus on the evolution of a specific land use/land cover type. Often these methods are applied on SITS that span over a short period of time, typical one or several years. In this paper we focus on a different use scenario of SITS analysis by extracting classes of
dynamic evolutions from long-term satellite image time series. As such we create a SITS of over 100 Landsat datasets, spanning from 1984 to 2011, covering the metropolitan area of Bucharest, Romania – perhaps the longest SITS ever being used. The analysis of this SITS aims to evaluate the spatio-temporal evolutions, focusing on finding classes of temporal dynamics of the main land use/land cover classes of the studied area, such as urban areas, agricultural land, forestry and water bodies, identifying interesting evolutions in the studied area. Furthermore, the extracted dynamic classes are used to establish relationships with known ancillary data
such as CORINE Land Cover vector data sets or statistical datasets for demographics of the studied area.
The rich content of SITS has generated a great interest in the recent years, many SITS analysis methods being developed. Many of these methods are focused towards improving the precision of land use/land cover detection by exploiting the inherent redundancy of SITS data.
Some other methods focus on the evolution of a specific land use/land cover type. Often these methods are applied on SITS that span over a short period of time, typical one or several years. In this paper we focus on a different use scenario of SITS analysis by extracting classes of
dynamic evolutions from long-term satellite image time series. As such we create a SITS of over 100 Landsat datasets, spanning from 1984 to 2011, covering the metropolitan area of Bucharest, Romania – perhaps the longest SITS ever being used. The analysis of this SITS aims to evaluate the spatio-temporal evolutions, focusing on finding classes of temporal dynamics of the main land use/land cover classes of the studied area, such as urban areas, agricultural land, forestry and water bodies, identifying interesting evolutions in the studied area. Furthermore, the extracted dynamic classes are used to establish relationships with known ancillary data
such as CORINE Land Cover vector data sets or statistical datasets for demographics of the studied area.
Proceedings Article
Semantic analysis of EO-1 hyperion hyperspectral data
Costăchioiu, Teodor; Constantinescu, Rodica; Lăzărescu, Vasile; Datcu, Mihai
In: 2013 5th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS), pp. 1-4, 2013.
@inproceedings{8080676,
title = {Semantic analysis of EO-1 hyperion hyperspectral data},
author = {Teodor Costăchioiu and Rodica Constantinescu and Vasile Lăzărescu and Mihai Datcu},
doi = {10.1109/WHISPERS.2013.8080676},
year = {2013},
date = {2013-01-01},
urldate = {2013-01-01},
booktitle = {2013 5th Workshop on Hyperspectral Image and Signal Processing: Evolution in Remote Sensing (WHISPERS)},
pages = {1-4},
abstract = {In this paper we propose a new method for unsupervised data model discovery in hyperspectral data, by adopting the Latent Dirichlet Allocation (LDA) text modeling tool. The proposed method relies on defining a representation of hyperspectral data that allows LDA analysis, by defining a correspondence between hyperspectral profiles and visual words. As such, we can use Latent Dirichlet Allocation as a method for discovery of latent (intrinsic and natural) meaningful grouping of the spectral bands. Based on the model parameters we propose several use scenarios for hyperspectral data analysis, allowing us to identify semantically meaningful structures in the observed scene. Experiments using the proposed method were carried on an EO-1 Hyperion data set imaging an agricultural area in Romania, acquired in June 2009.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
In this paper we propose a new method for unsupervised data model discovery in hyperspectral data, by adopting the Latent Dirichlet Allocation (LDA) text modeling tool. The proposed method relies on defining a representation of hyperspectral data that allows LDA analysis, by defining a correspondence between hyperspectral profiles and visual words. As such, we can use Latent Dirichlet Allocation as a method for discovery of latent (intrinsic and natural) meaningful grouping of the spectral bands. Based on the model parameters we propose several use scenarios for hyperspectral data analysis, allowing us to identify semantically meaningful structures in the observed scene. Experiments using the proposed method were carried on an EO-1 Hyperion data set imaging an agricultural area in Romania, acquired in June 2009.
2012
Proceedings Article
Semantic Analysis of Satellite Image Time Series
Costachioiu, Teodor; Constantinescu, Rodica; AlZenk, Bashar; Datcu, Mihai
In: Proceedings of the 20th European Signal Processing Conference (EUSIPCO), pp. 2492–2495, EURASIP, Bucharest, Romania, 2012, ISSN: 2076-1465.
@inproceedings{costachioiu2012semantic,
title = {Semantic Analysis of Satellite Image Time Series},
author = {Teodor Costachioiu and Rodica Constantinescu and Bashar AlZenk and Mihai Datcu},
issn = {2076-1465},
year = {2012},
date = {2012-08-01},
urldate = {2012-08-01},
booktitle = {Proceedings of the 20th European Signal Processing Conference (EUSIPCO)},
pages = {2492–2495},
publisher = {EURASIP},
address = {Bucharest, Romania},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Proceedings Article
A semantic framework for data retrieval in large remote sensing databases
Costăchioiu, Teodor; Niţă, Iulian; Lăzărescu, Vasile; Datcu, Mihai
In: 2012 IEEE International Geoscience and Remote Sensing Symposium, pp. 5285-5288, 2012.
@inproceedings{6352416,
title = {A semantic framework for data retrieval in large remote sensing databases},
author = {Teodor Costăchioiu and Iulian Niţă and Vasile Lăzărescu and Mihai Datcu},
doi = {10.1109/IGARSS.2012.6352416},
year = {2012},
date = {2012-01-01},
booktitle = {2012 IEEE International Geoscience and Remote Sensing Symposium},
pages = {5285-5288},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
2011
Proceedings Article
Multiprocessor real time edge detection using FPGA IP cores
Nita, Iulian; Costachioiu, Teodor; Lazarescu, Vasile; Seceleanu, Tiberiu
In: 2011 IEEE 7th International Conference on Intelligent Computer Communication and Processing, pp. 331-334, 2011.
@inproceedings{6047891,
title = {Multiprocessor real time edge detection using FPGA IP cores},
author = {Iulian Nita and Teodor Costachioiu and Vasile Lazarescu and Tiberiu Seceleanu},
doi = {10.1109/ICCP.2011.6047891},
year = {2011},
date = {2011-01-01},
booktitle = {2011 IEEE 7th International Conference on Intelligent Computer Communication and Processing},
pages = {331-334},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Proceedings Article
Costachioiu, Teodor; Lazarescu, Vasile; Datcu, Mihai
In: ISSCS 2011 - International Symposium on Signals, Circuits and Systems, pp. 1-4, 2011.
@inproceedings{5978721,
title = {Classification of scene evolution patterns from Satellite Image Time Series based on spectro-temporal signatures},
author = {Teodor Costachioiu and Vasile Lazarescu and Mihai Datcu},
doi = {10.1109/ISSCS.2011.5978721},
year = {2011},
date = {2011-01-01},
booktitle = {ISSCS 2011 - International Symposium on Signals, Circuits and Systems},
pages = {1-4},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Proceedings Article
Speed-Up of GIS Processing Using Multicore Architectures
Nita, Iulian; Costachioiu, Teodor; Lazarescu, Vasile
In: Murgante, Beniamino; Gervasi, Osvaldo; Iglesias, Andrés; Taniar, David; Apduhan, Bernady O. (Ed.): Computational Science and Its Applications - ICCSA 2011, pp. 293–302, Springer Berlin Heidelberg, Berlin, Heidelberg, 2011, ISBN: 978-3-642-21887-3.
@inproceedings{10.1007/978-3-642-21887-3_23,
title = {Speed-Up of GIS Processing Using Multicore Architectures},
author = {Iulian Nita and Teodor Costachioiu and Vasile Lazarescu},
editor = {Beniamino Murgante and Osvaldo Gervasi and Andrés Iglesias and David Taniar and Bernady O. Apduhan},
url = {https://link.springer.com/chapter/10.1007/978-3-642-21887-3_23},
doi = {https://doi.org/10.1007/978-3-642-21887-3_23},
isbn = {978-3-642-21887-3},
year = {2011},
date = {2011-01-01},
urldate = {2011-01-01},
booktitle = {Computational Science and Its Applications - ICCSA 2011},
pages = {293–302},
publisher = {Springer Berlin Heidelberg},
address = {Berlin, Heidelberg},
abstract = {Today's abilities of gathering and storing data are not matched by the ability to process this vast amount of data. Such examples can be found in the field of remote sensing, where new satellite missions contribute to a continuous downstream of remotely sensed data. Processing of large remotely sensed datasets has a high algorithmic complexity and requires considerable hardware resources. As many of the pixel-level operations are parallelizable, data processing can benefit from multicore technology. In this paper we use Dynamic Data Flow Model of Computation to create a processing framework that is both portable and scalable, being able to detect the number of processing cores and also capable to dynamically allocating tasks in such manner to ensure the balance of the processing load on each core. The proposed method is used to accelerate a water detection algorithm from Landsat TM data and was tested on multiple platforms with different multicore configurations.},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
Today's abilities of gathering and storing data are not matched by the ability to process this vast amount of data. Such examples can be found in the field of remote sensing, where new satellite missions contribute to a continuous downstream of remotely sensed data. Processing of large remotely sensed datasets has a high algorithmic complexity and requires considerable hardware resources. As many of the pixel-level operations are parallelizable, data processing can benefit from multicore technology. In this paper we use Dynamic Data Flow Model of Computation to create a processing framework that is both portable and scalable, being able to detect the number of processing cores and also capable to dynamically allocating tasks in such manner to ensure the balance of the processing load on each core. The proposed method is used to accelerate a water detection algorithm from Landsat TM data and was tested on multiple platforms with different multicore configurations.
2010
Proceedings Article
Costăchioiu, Teodor; Datcu, Mihai
In: 2010 8th International Conference on Communications, pp. 157-160, 2010.
@inproceedings{5509070,
title = {Land cover dynamics classification using multi-temporal spectral indices from satellite image time series},
author = {Teodor Costăchioiu and Mihai Datcu},
doi = {10.1109/ICCOMM.2010.5509070},
year = {2010},
date = {2010-01-01},
booktitle = {2010 8th International Conference on Communications},
pages = {157-160},
keywords = {},
pubstate = {published},
tppubtype = {inproceedings}
}
