eleni charou | National Centre of Scientific Research "DEMOKRITOS" (original) (raw)

Papers by eleni charou

Research paper thumbnail of Building block extraction and classification by means of Markov random fields using aerial imagery and LiDAR data

Proceedings of SPIE, Oct 26, 2016

Building detection has been a prominent area in the area of image classification. Most of the res... more Building detection has been a prominent area in the area of image classification. Most of the research effort is adapted to the specific application requirements and available datasets. Our dataset includes aerial orthophotos (with spatial resolution 20cm), a DSM generated from LiDAR (with spatial resolution 1m and elevation resolution 20 cm) and DTM (spatial resolution 2m) from an area of Athens, Greece. Our aim is to classify these data by means of Markov Random Fields (MRFs) in a Bayesian framework for building block extraction and perform a comparative analysis with other supervised classification techniques namely Feed Forward Neural Net (FFNN), Cascade-Correlation Neural Network (CCNN), Learning Vector Quantization (LVQ) and Support Vector Machines (SVM). We evaluated the performance of each method using a subset of the test area. We present the classified images, and statistical measures (confusion matrix, kappa coefficient and overall accuracy). Our results demonstrate that the MRFs and FFNN perform better than the other methods.

Research paper thumbnail of Procedural 3D Terrain Generation using Generative Adversarial Networks

arXiv (Cornell University), Oct 13, 2020

Research paper thumbnail of A Deep Learning approach to Hyperspectral Image Classification using an improved Hybrid 3D-2D Convolutional Neural Network

In recent years, the task of Hyperspectral Image (HSI) classification has appeared in various fie... more In recent years, the task of Hyperspectral Image (HSI) classification has appeared in various fields, including Remote Sensing. Meanwhile, the evolution of Deep Learning, and the prevalence of the Convolutional Neural Network (CNN) has revolutionized the way unstructured, especially visual, data are processed. 2D CNN have proved highly efficient in exploiting the spatial information of images, but in HSI classification, data contain both spectral and spatial features. To make use of these characteristics, many variations of a 3D CNN have been proposed, but a 3D Convolution comes at a high computational cost. A fusion of 3D and 2D convolutions decreases processing time by distributing spectral-spatial feature extraction across a lighter, less complex model. An enhanced Hybrid network architecture is proposed alongside a data preprocessing plan, with the aim of achieving a significant improvement in classification results. Four benchmark datasets (Indian Pines, Pavia University, Salinas and Data Fusion 2013 Contest) are used to compare the model to other hand-crafted or deep learning architectures. It is demonstrated that the proposed network outperforms state-of-the-art approaches in terms of classification accuracy, while avoiding some commonly used, computationally expensive design choices.

Research paper thumbnail of A dynamic model for environmentally safe shipping through the Aegean Sea

Aegean Sea is an extremely sensitive marine area anticipating a catastrophic event to occur any t... more Aegean Sea is an extremely sensitive marine area anticipating a catastrophic event to occur any time now, owing both to hazardous vessel crossing its waters and the significant rise of the intensive traffic. This paper aims to present a probabilistic Bayesian model predicting the probability of a collision, contact or grounding occurrence in the Aegean Sea. The model takes into account the dynamic information of the navigation area and the prevailing weather conditions. The training of the network was performed using the data of the historical accident database of the Marine Rescue Coordination Centre, the National Meteorological Office of Greece and the Aminess database. The whole study has been run within the framework of the AMINESS project.

Research paper thumbnail of Fast and efficient land-cover classification of multispectral remote sensing data using artificial neural network techniques

Abstract A time and memory efficient methodology for supervised and unsupervised land-cover class... more Abstract A time and memory efficient methodology for supervised and unsupervised land-cover classification of multispectral remote sensing (MRS) data based on artificial neural network (ANN) techniques is presented. The proposed methodology first performs a vector ...

Research paper thumbnail of Generative Adversarial Networks for Data Augmentation in Hyperspectral Image Classification

Research paper thumbnail of Integration of Intelligent Techniques for Environmental Data Processing

Springer eBooks, 2003

In this paper, some aspects of the usefulness of intelligent techniques in environmental data pro... more In this paper, some aspects of the usefulness of intelligent techniques in environmental data processing are discussed. The capabilities of neural networks in improving memory requirements for storage of environmental data and the increase in processing speed are analyzed. Finally, a software package for processing multi-source (geophysical, geochemical, satellite, etc.) data using various neural, fuzzy, multimodular, pattern-recognition and image processing algorithms is presented.

Research paper thumbnail of Comparative analysis of classification techniques for building block extraction using aerial imagery and LiDAR data

ABSTRACT Building detection has been a prominent area in the area of image classification. Most o... more ABSTRACT Building detection has been a prominent area in the area of image classification. Most of the research effort is adapted to the specific application requirements and available datasets. In this paper we present a comparative analysis of different classification techniques for building block extraction. Our dataset includes aerial orthophotos (with spatial resolution 20cm), a DSM generated from LiDAR (with spatial resolution 1m and elevation resolution 20 cm) and DTM (spatial resolution 2m) from an area of Athens, Greece. The classification methods tested are unsupervised (K-Means, Mean Shift), and supervised (Feed Forward Neural Net, Radial-Basis Functions, Support Vector Machines). We evaluated the performance of each method using a subset of the test area. We present the classified images, and statistical measures (confusion matrix, kappa coefficient and overall accuracy). Our results demonstrate that the top unsupervised method is the Mean Shift that performs similarly to the best supervised methods.

Research paper thumbnail of Class-Wise Principal Component Analysis for hyperspectral image feature extraction

arXiv (Cornell University), Apr 9, 2021

This paper introduces the Class-wise Principal Component Analysis, a supervised feature extractio... more This paper introduces the Class-wise Principal Component Analysis, a supervised feature extraction method for hyperspectral data. Hyperspectral Imaging (HSI) has appeared in various fields in recent years, including Remote Sensing. Realizing that information extraction tasks for hyperspectral images are burdened by data-specific issues, we identify and address two major problems. Those are the Curse of Dimensionality which occurs due to the highvolume of the data cube and the class imbalance problem which is common in hyperspectral datasets. Dimensionality reduction is an essential preprocessing step to complement a hyperspectral image classification task. Therefore, we propose a feature extraction algorithm for dimensionality reduction, based on Principal Component Analysis (PCA). Evaluations are carried out on the Indian Pines dataset to demonstrate that significant improvements are achieved when using the reduced data in a classification task.

Research paper thumbnail of Spatial Resolution Enhancement of Remote Sensing Mine Images using Deep Learning Techniques

arXiv (Cornell University), Jul 17, 2020

Deep learning techniques are applied so as to increase the spatial resolution of Sentinel-2 satel... more Deep learning techniques are applied so as to increase the spatial resolution of Sentinel-2 satellite imagery, depicting the Amynteo lignite mine in Ptolemaida, Greece. Resolution enhancement by factors 2 and 4 as well as by factors 2 and 6 using Very-Deep Super-Resolution (VDSR) and DSen2 networks, respectively, provides fairly well results on Amynteo lignite mine images. Particularly, the aim of this research is to super-resolve (i) Sentinel-2 bands from 10m/pixel and 20m/pixel to 5m/pixel, that is even greater than the sensor's resolution, using VDSR network and (ii) Sentinel-2 lower-resolution bands from 20m/pixel and 60m/pixel to 10m/pixel using DSen2 network.

Research paper thumbnail of Generating Elevation Surface from a Single RGB Remotely Sensed Image Using Deep Learning

Remote Sensing, Jun 22, 2020

Research paper thumbnail of Super-Resolution Techniques in Photogrammetric 3D Reconstruction from Close-Range UAV Imagery

Heritage

Current Multi-View Stereo (MVS) algorithms are tools for high-quality 3D model reconstruction, st... more Current Multi-View Stereo (MVS) algorithms are tools for high-quality 3D model reconstruction, strongly depending on image spatial resolution. In this context, the combination of image Super-Resolution (SR) with image-based 3D reconstruction is turning into an interesting research topic in photogrammetry, around which however only a few works have been reported so far in the literature. Here, a thorough study is carried out on various state-of-the-art image SR techniques to evaluate the suitability of such an approach in terms of its inclusion in the 3D reconstruction process. Deep-learning techniques are tested here on a UAV image dataset, while the MVS task is then performed via the Agisoft Metashape photogrammetric tool. The data under experimentation are oblique cultural heritage imagery. According to results, point clouds from low-resolution images present quality inferior to those from upsampled high-resolution ones. The SR techniques HAT and DRLN outperform bicubic interpolat...

Research paper thumbnail of Sentinel-2 Images at 2.5m Spatial Resolution via Deep-Learning: A Case Study in Zakythnos

2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP)

Research paper thumbnail of Sentinel-2 and SPOT-7 Images in Machine Learning Frameworks for Super-Resolution

Research paper thumbnail of Comparative study of visual feature extraction methods for building retrieval on urban databases

2015 6th International Conference on Information, Intelligence, Systems and Applications (IISA), 2015

In a great variety of applications, such as real-time robot localization and visual navigation, a... more In a great variety of applications, such as real-time robot localization and visual navigation, architectural design, 3D city reconstruction, building recognition in urban environments is required. In this work a comparative study of visual feature extraction methods for building retrieval on urban databases is performed. To this end, a database of 183 building facades taken at two different regions near the center of Athens was created. Five feature extraction methods are used in order to capture visual properties. Features between images are then matched using RANSAC algorithm and a visual transformation is calculated. Comparative retrieval results are provided and discussed.

Research paper thumbnail of A maritime data analytics platform for policy recommendation

2015 6th International Conference on Information, Intelligence, Systems and Applications (IISA), 2015

The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the signif... more The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the significant increase of the traffic of tankers from and to the Black Sea that pass through narrow straits formed by the 1600 Greek islands. Reducing the risk of a ship accident is therefore vital to all socio-economic and environmental sectors. This paper presents a maritime data analytics platform for policy recommendation. The online tool focuses on extracting aggregated vessel risks using spatiotemporal analysis of multilayer information: vessel trajectories, vessel data, meteorological data, bathymetric/ hydrographic data as well as information regarding environmentally important areas (e.g. protected high-risk areas, etc.). The web interface enables user-friendly spatiotemporal queries at the front-end, while a series of data mining functionalities extracts aggregated statistics regarding: (a) marine risks and accident probabilities for particular areas (b) trajectories clustering information (c) general marine statistics (vessel types, etc.). Employing the above tool, if-then-scenarios can be constructed and resulting statistics and diagrams can provide essential arguments at maritime policy makers.

Research paper thumbnail of Long-Term Marine Traffic Monitoring for Environmental Safety in the Aegean Sea

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2015

The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the signif... more The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the significant increase of the traffic of tankers from and to the Black Sea that pass through narrow straits formed by the 1600 Greek islands. Reducing the risk of a ship accident is therefore vital to all socioeconomic and environmental sectors. This paper presents an online long-term marine traffic monitoring work-flow that focuses on extracting aggregated vessel risks using spatiotemporal analysis of multilayer information: vessel trajectories, vessel data, meteorological data, bathymetric / hydrographic data as well as information regarding environmentally important areas (e.g. protected high-risk areas, etc.). A web interface that enables user-friendly spatiotemporal queries is implemented at the frontend, while a series of data mining functionalities extracts aggregated statistics regarding: (a) marine risks and accident probabilities for particular areas (b) trajectories clustering information (c) general marine statistics (cargo types, etc.) and (d) correlation between spatial environmental importance and marine traffic risk. Towards this end, a set of data clustering and probabilistic graphical modelling techniques has been adopted.

Research paper thumbnail of Use of Sentinel-1 data for Maritime Domain Awareness: Preliminary results

In this work the utility of Synthetic Aperture Radar (SAR) data acquired from the SAR instrument ... more In this work the utility of Synthetic Aperture Radar (SAR) data acquired from the SAR instrument on-board the recently launched ESA's Sentinel-1 to Maritime Domain Awareness (MDA) is studied. The limitations of the currently used Automatic Identification System (AIS) for ships is presented. The use of space based SAR sensors for MDA is discussed and the recently available ESA's Sentinel1 satellite is considered in more detail. An automatic methodology for ship detection from Sentinel-1 is presented and applied in the environmentally important sea straits between Kythera and Elafonisos islands, Southern Greece. Preliminary results show that Sentinel1 has a significant role to play in MDA.

Research paper thumbnail of Land Cover Semantic Segmentation Using ResUNet

arXiv (Cornell University), Oct 13, 2020

In this paper we present our work on developing an automated system for land cover classification... more In this paper we present our work on developing an automated system for land cover classification. This system takes a multiband satellite image of an area as input and outputs the land cover map of the area at the same resolution as the input. For this purpose convolutional machine learning models were trained in the task of predicting the land cover semantic segmentation of satellite images. This is a case of supervised learning. The land cover label data were taken from the CORINE Land Cover inventory and the satellite images were taken from the Copernicus hub. As for the model, U-Net architecture variations were applied. Our area of interest are the Ionian islands (Greece). We created a dataset from scratch covering this particular area. In addition, transfer learning from the BigEarthNet dataset [1] was performed. In [1] simple classification of satellite images into the classes of CLC is performed but not segmentation as we do. However, their models have been trained into a dataset much bigger than ours, so we applied transfer learning using their pretrained models as the first part of out network, utilizing the ability these networks have developed to extract useful features from the satellite images (we transferred a pretrained ResNet50 into a U-Res-Net). Apart from transfer learning other techniques were applied in order to overcome the limitations set by the small size of our area of interest. We used data augmentation (cutting images into overlapping patches, applying random transformations such as rotations and flips) and cross validation. The results are tested on the 3 CLC class hierarchy levels and a comparative study is made on the results of different approaches.

Research paper thumbnail of Chapter 7 Monitoring of Transnational Lakes Using Geomatic Techniques: A Case Study for Prespa Lakes

Research paper thumbnail of Building block extraction and classification by means of Markov random fields using aerial imagery and LiDAR data

Proceedings of SPIE, Oct 26, 2016

Building detection has been a prominent area in the area of image classification. Most of the res... more Building detection has been a prominent area in the area of image classification. Most of the research effort is adapted to the specific application requirements and available datasets. Our dataset includes aerial orthophotos (with spatial resolution 20cm), a DSM generated from LiDAR (with spatial resolution 1m and elevation resolution 20 cm) and DTM (spatial resolution 2m) from an area of Athens, Greece. Our aim is to classify these data by means of Markov Random Fields (MRFs) in a Bayesian framework for building block extraction and perform a comparative analysis with other supervised classification techniques namely Feed Forward Neural Net (FFNN), Cascade-Correlation Neural Network (CCNN), Learning Vector Quantization (LVQ) and Support Vector Machines (SVM). We evaluated the performance of each method using a subset of the test area. We present the classified images, and statistical measures (confusion matrix, kappa coefficient and overall accuracy). Our results demonstrate that the MRFs and FFNN perform better than the other methods.

Research paper thumbnail of Procedural 3D Terrain Generation using Generative Adversarial Networks

arXiv (Cornell University), Oct 13, 2020

Research paper thumbnail of A Deep Learning approach to Hyperspectral Image Classification using an improved Hybrid 3D-2D Convolutional Neural Network

In recent years, the task of Hyperspectral Image (HSI) classification has appeared in various fie... more In recent years, the task of Hyperspectral Image (HSI) classification has appeared in various fields, including Remote Sensing. Meanwhile, the evolution of Deep Learning, and the prevalence of the Convolutional Neural Network (CNN) has revolutionized the way unstructured, especially visual, data are processed. 2D CNN have proved highly efficient in exploiting the spatial information of images, but in HSI classification, data contain both spectral and spatial features. To make use of these characteristics, many variations of a 3D CNN have been proposed, but a 3D Convolution comes at a high computational cost. A fusion of 3D and 2D convolutions decreases processing time by distributing spectral-spatial feature extraction across a lighter, less complex model. An enhanced Hybrid network architecture is proposed alongside a data preprocessing plan, with the aim of achieving a significant improvement in classification results. Four benchmark datasets (Indian Pines, Pavia University, Salinas and Data Fusion 2013 Contest) are used to compare the model to other hand-crafted or deep learning architectures. It is demonstrated that the proposed network outperforms state-of-the-art approaches in terms of classification accuracy, while avoiding some commonly used, computationally expensive design choices.

Research paper thumbnail of A dynamic model for environmentally safe shipping through the Aegean Sea

Aegean Sea is an extremely sensitive marine area anticipating a catastrophic event to occur any t... more Aegean Sea is an extremely sensitive marine area anticipating a catastrophic event to occur any time now, owing both to hazardous vessel crossing its waters and the significant rise of the intensive traffic. This paper aims to present a probabilistic Bayesian model predicting the probability of a collision, contact or grounding occurrence in the Aegean Sea. The model takes into account the dynamic information of the navigation area and the prevailing weather conditions. The training of the network was performed using the data of the historical accident database of the Marine Rescue Coordination Centre, the National Meteorological Office of Greece and the Aminess database. The whole study has been run within the framework of the AMINESS project.

Research paper thumbnail of Fast and efficient land-cover classification of multispectral remote sensing data using artificial neural network techniques

Abstract A time and memory efficient methodology for supervised and unsupervised land-cover class... more Abstract A time and memory efficient methodology for supervised and unsupervised land-cover classification of multispectral remote sensing (MRS) data based on artificial neural network (ANN) techniques is presented. The proposed methodology first performs a vector ...

Research paper thumbnail of Generative Adversarial Networks for Data Augmentation in Hyperspectral Image Classification

Research paper thumbnail of Integration of Intelligent Techniques for Environmental Data Processing

Springer eBooks, 2003

In this paper, some aspects of the usefulness of intelligent techniques in environmental data pro... more In this paper, some aspects of the usefulness of intelligent techniques in environmental data processing are discussed. The capabilities of neural networks in improving memory requirements for storage of environmental data and the increase in processing speed are analyzed. Finally, a software package for processing multi-source (geophysical, geochemical, satellite, etc.) data using various neural, fuzzy, multimodular, pattern-recognition and image processing algorithms is presented.

Research paper thumbnail of Comparative analysis of classification techniques for building block extraction using aerial imagery and LiDAR data

ABSTRACT Building detection has been a prominent area in the area of image classification. Most o... more ABSTRACT Building detection has been a prominent area in the area of image classification. Most of the research effort is adapted to the specific application requirements and available datasets. In this paper we present a comparative analysis of different classification techniques for building block extraction. Our dataset includes aerial orthophotos (with spatial resolution 20cm), a DSM generated from LiDAR (with spatial resolution 1m and elevation resolution 20 cm) and DTM (spatial resolution 2m) from an area of Athens, Greece. The classification methods tested are unsupervised (K-Means, Mean Shift), and supervised (Feed Forward Neural Net, Radial-Basis Functions, Support Vector Machines). We evaluated the performance of each method using a subset of the test area. We present the classified images, and statistical measures (confusion matrix, kappa coefficient and overall accuracy). Our results demonstrate that the top unsupervised method is the Mean Shift that performs similarly to the best supervised methods.

Research paper thumbnail of Class-Wise Principal Component Analysis for hyperspectral image feature extraction

arXiv (Cornell University), Apr 9, 2021

This paper introduces the Class-wise Principal Component Analysis, a supervised feature extractio... more This paper introduces the Class-wise Principal Component Analysis, a supervised feature extraction method for hyperspectral data. Hyperspectral Imaging (HSI) has appeared in various fields in recent years, including Remote Sensing. Realizing that information extraction tasks for hyperspectral images are burdened by data-specific issues, we identify and address two major problems. Those are the Curse of Dimensionality which occurs due to the highvolume of the data cube and the class imbalance problem which is common in hyperspectral datasets. Dimensionality reduction is an essential preprocessing step to complement a hyperspectral image classification task. Therefore, we propose a feature extraction algorithm for dimensionality reduction, based on Principal Component Analysis (PCA). Evaluations are carried out on the Indian Pines dataset to demonstrate that significant improvements are achieved when using the reduced data in a classification task.

Research paper thumbnail of Spatial Resolution Enhancement of Remote Sensing Mine Images using Deep Learning Techniques

arXiv (Cornell University), Jul 17, 2020

Deep learning techniques are applied so as to increase the spatial resolution of Sentinel-2 satel... more Deep learning techniques are applied so as to increase the spatial resolution of Sentinel-2 satellite imagery, depicting the Amynteo lignite mine in Ptolemaida, Greece. Resolution enhancement by factors 2 and 4 as well as by factors 2 and 6 using Very-Deep Super-Resolution (VDSR) and DSen2 networks, respectively, provides fairly well results on Amynteo lignite mine images. Particularly, the aim of this research is to super-resolve (i) Sentinel-2 bands from 10m/pixel and 20m/pixel to 5m/pixel, that is even greater than the sensor's resolution, using VDSR network and (ii) Sentinel-2 lower-resolution bands from 20m/pixel and 60m/pixel to 10m/pixel using DSen2 network.

Research paper thumbnail of Generating Elevation Surface from a Single RGB Remotely Sensed Image Using Deep Learning

Remote Sensing, Jun 22, 2020

Research paper thumbnail of Super-Resolution Techniques in Photogrammetric 3D Reconstruction from Close-Range UAV Imagery

Heritage

Current Multi-View Stereo (MVS) algorithms are tools for high-quality 3D model reconstruction, st... more Current Multi-View Stereo (MVS) algorithms are tools for high-quality 3D model reconstruction, strongly depending on image spatial resolution. In this context, the combination of image Super-Resolution (SR) with image-based 3D reconstruction is turning into an interesting research topic in photogrammetry, around which however only a few works have been reported so far in the literature. Here, a thorough study is carried out on various state-of-the-art image SR techniques to evaluate the suitability of such an approach in terms of its inclusion in the 3D reconstruction process. Deep-learning techniques are tested here on a UAV image dataset, while the MVS task is then performed via the Agisoft Metashape photogrammetric tool. The data under experimentation are oblique cultural heritage imagery. According to results, point clouds from low-resolution images present quality inferior to those from upsampled high-resolution ones. The SR techniques HAT and DRLN outperform bicubic interpolat...

Research paper thumbnail of Sentinel-2 Images at 2.5m Spatial Resolution via Deep-Learning: A Case Study in Zakythnos

2022 IEEE 14th Image, Video, and Multidimensional Signal Processing Workshop (IVMSP)

Research paper thumbnail of Sentinel-2 and SPOT-7 Images in Machine Learning Frameworks for Super-Resolution

Research paper thumbnail of Comparative study of visual feature extraction methods for building retrieval on urban databases

2015 6th International Conference on Information, Intelligence, Systems and Applications (IISA), 2015

In a great variety of applications, such as real-time robot localization and visual navigation, a... more In a great variety of applications, such as real-time robot localization and visual navigation, architectural design, 3D city reconstruction, building recognition in urban environments is required. In this work a comparative study of visual feature extraction methods for building retrieval on urban databases is performed. To this end, a database of 183 building facades taken at two different regions near the center of Athens was created. Five feature extraction methods are used in order to capture visual properties. Features between images are then matched using RANSAC algorithm and a visual transformation is calculated. Comparative retrieval results are provided and discussed.

Research paper thumbnail of A maritime data analytics platform for policy recommendation

2015 6th International Conference on Information, Intelligence, Systems and Applications (IISA), 2015

The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the signif... more The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the significant increase of the traffic of tankers from and to the Black Sea that pass through narrow straits formed by the 1600 Greek islands. Reducing the risk of a ship accident is therefore vital to all socio-economic and environmental sectors. This paper presents a maritime data analytics platform for policy recommendation. The online tool focuses on extracting aggregated vessel risks using spatiotemporal analysis of multilayer information: vessel trajectories, vessel data, meteorological data, bathymetric/ hydrographic data as well as information regarding environmentally important areas (e.g. protected high-risk areas, etc.). The web interface enables user-friendly spatiotemporal queries at the front-end, while a series of data mining functionalities extracts aggregated statistics regarding: (a) marine risks and accident probabilities for particular areas (b) trajectories clustering information (c) general marine statistics (vessel types, etc.). Employing the above tool, if-then-scenarios can be constructed and resulting statistics and diagrams can provide essential arguments at maritime policy makers.

Research paper thumbnail of Long-Term Marine Traffic Monitoring for Environmental Safety in the Aegean Sea

The International Archives of the Photogrammetry, Remote Sensing and Spatial Information Sciences, 2015

The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the signif... more The Aegean Sea is characterized by an extremely high marine safety risk, mainly due to the significant increase of the traffic of tankers from and to the Black Sea that pass through narrow straits formed by the 1600 Greek islands. Reducing the risk of a ship accident is therefore vital to all socioeconomic and environmental sectors. This paper presents an online long-term marine traffic monitoring work-flow that focuses on extracting aggregated vessel risks using spatiotemporal analysis of multilayer information: vessel trajectories, vessel data, meteorological data, bathymetric / hydrographic data as well as information regarding environmentally important areas (e.g. protected high-risk areas, etc.). A web interface that enables user-friendly spatiotemporal queries is implemented at the frontend, while a series of data mining functionalities extracts aggregated statistics regarding: (a) marine risks and accident probabilities for particular areas (b) trajectories clustering information (c) general marine statistics (cargo types, etc.) and (d) correlation between spatial environmental importance and marine traffic risk. Towards this end, a set of data clustering and probabilistic graphical modelling techniques has been adopted.

Research paper thumbnail of Use of Sentinel-1 data for Maritime Domain Awareness: Preliminary results

In this work the utility of Synthetic Aperture Radar (SAR) data acquired from the SAR instrument ... more In this work the utility of Synthetic Aperture Radar (SAR) data acquired from the SAR instrument on-board the recently launched ESA's Sentinel-1 to Maritime Domain Awareness (MDA) is studied. The limitations of the currently used Automatic Identification System (AIS) for ships is presented. The use of space based SAR sensors for MDA is discussed and the recently available ESA's Sentinel1 satellite is considered in more detail. An automatic methodology for ship detection from Sentinel-1 is presented and applied in the environmentally important sea straits between Kythera and Elafonisos islands, Southern Greece. Preliminary results show that Sentinel1 has a significant role to play in MDA.

Research paper thumbnail of Land Cover Semantic Segmentation Using ResUNet

arXiv (Cornell University), Oct 13, 2020

In this paper we present our work on developing an automated system for land cover classification... more In this paper we present our work on developing an automated system for land cover classification. This system takes a multiband satellite image of an area as input and outputs the land cover map of the area at the same resolution as the input. For this purpose convolutional machine learning models were trained in the task of predicting the land cover semantic segmentation of satellite images. This is a case of supervised learning. The land cover label data were taken from the CORINE Land Cover inventory and the satellite images were taken from the Copernicus hub. As for the model, U-Net architecture variations were applied. Our area of interest are the Ionian islands (Greece). We created a dataset from scratch covering this particular area. In addition, transfer learning from the BigEarthNet dataset [1] was performed. In [1] simple classification of satellite images into the classes of CLC is performed but not segmentation as we do. However, their models have been trained into a dataset much bigger than ours, so we applied transfer learning using their pretrained models as the first part of out network, utilizing the ability these networks have developed to extract useful features from the satellite images (we transferred a pretrained ResNet50 into a U-Res-Net). Apart from transfer learning other techniques were applied in order to overcome the limitations set by the small size of our area of interest. We used data augmentation (cutting images into overlapping patches, applying random transformations such as rotations and flips) and cross validation. The results are tested on the 3 CLC class hierarchy levels and a comparative study is made on the results of different approaches.

Research paper thumbnail of Chapter 7 Monitoring of Transnational Lakes Using Geomatic Techniques: A Case Study for Prespa Lakes