Andrzej Kucik
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
Andrzej Kucik has a total of 9 co-authors. Here are the top co-authors by number of shared publications:
- Andreas Stokholm
DTU Space, National Space Institute, Technical University of Denmark (DTU)
5 shared publications - Sine Munk Hvidegaard
DTU Space, National Space Institute, Technical University of Denmark (DTU), Elektrovej, building 327, Kgs. Lyngby, 2800, Denmark
4 shared publications Nicolas Longépé
ϕ-lab, European Space Research Institute (ESRIN), The European Space Agency (ESA), Largo Galileo Galilei, 1, Frascati RM, 00044, Italy
3 shared publicationsAnna Naemi Willer
1 shared publicationDaniel Olesen
1 shared publication- Gabriele Meoni
European Space Agency
1 shared publication
Latest Publications
- Open Access
- 20 January 2025
- Published by Copernicus GmbH
Observing auroras with dynamic vision sensors
Auroras are a faint space weather light phenomenon caused by the interaction between the solar wind and the Earth’s magnetic field. This interaction can lead to processes that increase the energy of charged particles in the magnetosphere, enabling the particles to enter the ionosphere and cause diffuse or discrete auroras. Often, the images of auroras are captured using exposure times of 1-2 seconds to collect sufficient light. However, prolonged exposure times also intensify other light sources, such as urban- or moonlight, thus, dark night conditions are preferred. Long exposure times are also incapable of capturing fast dynamics, though some studies have carried out high-speed imaging with up to 160 frames per second (160Hz). Instead of traditional cameras, we propose to utilise the emerging optical technology, Dynamic Vision Sensors (DVS), that possess high dynamic ranges (110 - 120 dB) and sampling rates. DVS is a biologically-inspired silicon retina that detects negative or positive brightness change on a logarithmic scale similar to human eyes. In addition, pixels can independently adjust to lighting scenarios with an adaptable brightness threshold and without a static upper limit or constant frame rate. If no change occurs, no information is registered, providing a variable data rate and low power consumption.Here, we present the first DVS observations of auroras in 5kHz, captured in March 2023 in Tromsø, Norway. We extract and interpret information based on reconstructed brightness and event frames. Further, we derive the incoming photon flux by mimicking a photometer. In all, we show that DVS is capable of observing auroras in challenging urban- and moonlight conditions while preserving the high temporal resolution that could enable a paradigm shift for aurora monitoring.
- 20 January 2025
- Published by Copernicus GmbH
Open AccessObserving auroras with dynamic vision sensors
Auroras are a faint space weather light phenomenon caused by the interaction between the solar wind and the Earth’s magnetic field. This interaction can lead to processes that increase the energy of charged particles in the magnetosphere, enabling the particles to enter the ionosphere and cause diffuse or discrete auroras. Often, the images of auroras are captured using exposure times of 1-2 seconds to collect sufficient light. However, prolonged exposure times also intensify other light sources, such as urban- or moonlight, thus, dark night conditions are preferred. Long exposure times are also incapable of capturing fast dynamics, though some studies have carried out high-speed imaging with up to 160 frames per second (160Hz). Instead of traditional cameras, we propose to utilise the emerging optical technology, Dynamic Vision Sensors (DVS), that possess high dynamic ranges (110 - 120 dB) and sampling rates. DVS is a biologically-inspired silicon retina that detects negative or positive brightness change on a logarithmic scale similar to human eyes. In addition, pixels can independently adjust to lighting scenarios with an adaptable brightness threshold and without a static upper limit or constant frame rate. If no change occurs, no information is registered, providing a variable data rate and low power consumption.Here, we present the first DVS observations of auroras in 5kHz, captured in March 2023 in Tromsø, Norway. We extract and interpret information based on reconstructed brightness and event frames. Further, we derive the incoming photon flux by mimicking a photometer. In all, we show that DVS is capable of observing auroras in challenging urban- and moonlight conditions while preserving the high temporal resolution that could enable a paradigm shift for aurora monitoring.
- Open Access
- 5 September 2023
- Published by Copernicus GmbH
- Vol. 2023, 1-25
Reply to review
We investigate how different Convolutional Neural Network (CNN) U-Net models specialised in addressing partial labelling tasks related to mapping Sea Ice Concentration (SIC) can improve performance. We use Sentinel-1 SAR images and human-labelled ice charts as the reference to train models that benefit from advantages gained from different model optimisation objectives by utilising a multistage inference scheme. We find our multistage model inference approach that apply a classification (CrossEntropy or Earth Mover's Distance squared) optimised model to separate open water, intermediate SIC and fully covered ice in conjunction with a regression (Mean Square Error or Binary CrossEntropy) optimised model, that assigns specific intermediate classes, to perform the best. To evaluate the models we introduce several specific metrics illustrating the performance in key areas, such as the separation of macro classes, intermediate class, and an accuracy metric better encapsulating uncertainties in the reference data. We achieve R2-score of ~93 %, similar to state-of-the-art in the literature (Kucik and Stokholm, 2023). However, our models exhibit significantly better open water and 100 % SIC detections. The multistage synergises high open water and fully covered sea ice accuracies achieved with classification optimised objectives with good intermediate class performance obtained by regressional loss functions. In addition, our findings indicate that the number of classes that the intermediate concentrations are compressed into does not influence the result significantly it is the loss function used to optimise the model that assigns the specific intermediate class to have the largest impact.
- 5 September 2023
- Published by Copernicus GmbH
- Vol. 2023, 1-25
Open AccessReply to review
We investigate how different Convolutional Neural Network (CNN) U-Net models specialised in addressing partial labelling tasks related to mapping Sea Ice Concentration (SIC) can improve performance. We use Sentinel-1 SAR images and human-labelled ice charts as the reference to train models that benefit from advantages gained from different model optimisation objectives by utilising a multistage inference scheme. We find our multistage model inference approach that apply a classification (CrossEntropy or Earth Mover's Distance squared) optimised model to separate open water, intermediate SIC and fully covered ice in conjunction with a regression (Mean Square Error or Binary CrossEntropy) optimised model, that assigns specific intermediate classes, to perform the best. To evaluate the models we introduce several specific metrics illustrating the performance in key areas, such as the separation of macro classes, intermediate class, and an accuracy metric better encapsulating uncertainties in the reference data. We achieve R2-score of ~93 %, similar to state-of-the-art in the literature (Kucik and Stokholm, 2023). However, our models exhibit significantly better open water and 100 % SIC detections. The multistage synergises high open water and fully covered sea ice accuracies achieved with classification optimised objectives with good intermediate class performance obtained by regressional loss functions. In addition, our findings indicate that the number of classes that the intermediate concentrations are compressed into does not influence the result significantly it is the loss function used to optimise the model that assigns the specific intermediate class to have the largest impact.
- Open Access
- 19 August 2023
- Published by Copernicus GmbH
- Vol. 2023, 1-25
Comment on egusphere-2023-976
We investigate how different Convolutional Neural Network (CNN) U-Net models specialised in addressing partial labelling tasks related to mapping Sea Ice Concentration (SIC) can improve performance. We use Sentinel-1 SAR images and human-labelled ice charts as the reference to train models that benefit from advantages gained from different model optimisation objectives by utilising a multistage inference scheme. We find our multistage model inference approach that apply a classification (CrossEntropy or Earth Mover's Distance squared) optimised model to separate open water, intermediate SIC and fully covered ice in conjunction with a regression (Mean Square Error or Binary CrossEntropy) optimised model, that assigns specific intermediate classes, to perform the best. To evaluate the models we introduce several specific metrics illustrating the performance in key areas, such as the separation of macro classes, intermediate class, and an accuracy metric better encapsulating uncertainties in the reference data. We achieve R2-score of ~93 %, similar to state-of-the-art in the literature (Kucik and Stokholm, 2023). However, our models exhibit significantly better open water and 100 % SIC detections. The multistage synergises high open water and fully covered sea ice accuracies achieved with classification optimised objectives with good intermediate class performance obtained by regressional loss functions. In addition, our findings indicate that the number of classes that the intermediate concentrations are compressed into does not influence the result significantly it is the loss function used to optimise the model that assigns the specific intermediate class to have the largest impact.
- 19 August 2023
- Published by Copernicus GmbH
- Vol. 2023, 1-25
Open AccessComment on egusphere-2023-976
We investigate how different Convolutional Neural Network (CNN) U-Net models specialised in addressing partial labelling tasks related to mapping Sea Ice Concentration (SIC) can improve performance. We use Sentinel-1 SAR images and human-labelled ice charts as the reference to train models that benefit from advantages gained from different model optimisation objectives by utilising a multistage inference scheme. We find our multistage model inference approach that apply a classification (CrossEntropy or Earth Mover's Distance squared) optimised model to separate open water, intermediate SIC and fully covered ice in conjunction with a regression (Mean Square Error or Binary CrossEntropy) optimised model, that assigns specific intermediate classes, to perform the best. To evaluate the models we introduce several specific metrics illustrating the performance in key areas, such as the separation of macro classes, intermediate class, and an accuracy metric better encapsulating uncertainties in the reference data. We achieve R2-score of ~93 %, similar to state-of-the-art in the literature (Kucik and Stokholm, 2023). However, our models exhibit significantly better open water and 100 % SIC detections. The multistage synergises high open water and fully covered sea ice accuracies achieved with classification optimised objectives with good intermediate class performance obtained by regressional loss functions. In addition, our findings indicate that the number of classes that the intermediate concentrations are compressed into does not influence the result significantly it is the loss function used to optimise the model that assigns the specific intermediate class to have the largest impact.
- Open Access
- 27 June 2023
- Published by Copernicus GmbH in EGUsphere
- Vol. 2023, 1-25
AI4SeaIce: Task Separation and Multistage Inference CNNs for Automatic Sea Ice Concentration Charting
We investigate how different Convolutional Neural Network (CNN) U-Net models specialised in addressing partial labelling tasks related to mapping Sea Ice Concentration (SIC) can improve performance. We use Sentinel-1 SAR images and human-labelled ice charts as the reference to train models that benefit from advantages gained from different model optimisation objectives by utilising a multistage inference scheme. We find our multistage model inference approach that apply a classification (CrossEntropy or Earth Mover's Distance squared) optimised model to separate open water, intermediate SIC and fully covered ice in conjunction with a regression (Mean Square Error or Binary CrossEntropy) optimised model, that assigns specific intermediate classes, to perform the best. To evaluate the models we introduce several specific metrics illustrating the performance in key areas, such as the separation of macro classes, intermediate class, and an accuracy metric better encapsulating uncertainties in the reference data. We achieve R2-score of ~93 %, similar to state-of-the-art in the literature (Kucik and Stokholm, 2023). However, our models exhibit significantly better open water and 100 % SIC detections. The multistage synergises high open water and fully covered sea ice accuracies achieved with classification optimised objectives with good intermediate class performance obtained by regressional loss functions. In addition, our findings indicate that the number of classes that the intermediate concentrations are compressed into does not influence the result significantly it is the loss function used to optimise the model that assigns the specific intermediate class to have the largest impact.
- 27 June 2023
- Published by Copernicus GmbH in EGUsphere
- Vol. 2023, 1-25
Open AccessAI4SeaIce: Task Separation and Multistage Inference CNNs for Automatic Sea Ice Concentration Charting
We investigate how different Convolutional Neural Network (CNN) U-Net models specialised in addressing partial labelling tasks related to mapping Sea Ice Concentration (SIC) can improve performance. We use Sentinel-1 SAR images and human-labelled ice charts as the reference to train models that benefit from advantages gained from different model optimisation objectives by utilising a multistage inference scheme. We find our multistage model inference approach that apply a classification (CrossEntropy or Earth Mover's Distance squared) optimised model to separate open water, intermediate SIC and fully covered ice in conjunction with a regression (Mean Square Error or Binary CrossEntropy) optimised model, that assigns specific intermediate classes, to perform the best. To evaluate the models we introduce several specific metrics illustrating the performance in key areas, such as the separation of macro classes, intermediate class, and an accuracy metric better encapsulating uncertainties in the reference data. We achieve R2-score of ~93 %, similar to state-of-the-art in the literature (Kucik and Stokholm, 2023). However, our models exhibit significantly better open water and 100 % SIC detections. The multistage synergises high open water and fully covered sea ice accuracies achieved with classification optimised objectives with good intermediate class performance obtained by regressional loss functions. In addition, our findings indicate that the number of classes that the intermediate concentrations are compressed into does not influence the result significantly it is the loss function used to optimise the model that assigns the specific intermediate class to have the largest impact.
- Open Access
- 12 April 2023
- Published by Springer Nature in Scientific Reports
- Vol. 13 (1) , 1-10
AI4SeaIce: selecting loss functions for automated SAR sea ice concentration charting
For maritime navigation in the Arctic, sea ice charts are an essential tool, which still to this day is drawn manually by professional ice analysts. The total Sea Ice Concentration (SIC) is the primary descriptor of the charts and indicates the fraction of ice in an ocean surface area. Naturally, automating the SIC chart creation is desired. However, the optimal representation of the corresponding machine-learning task is ambivalent and discussed in the community. In this study, we explore the representation with either regressional or classification objectives, each with two different (weighted) loss functions: Mean Square Error and Binary Cross-Entropy, and Categorical Cross-Entropy and the Earth Mover’s Distance, respectively. While all models achieve good results they differ as the regression-based models obtain the highest numerical similarity to the reference charts, whereas the classification-optimised models generate results more visually pleasing and consistent. Rescaling the loss functions with inverse class weights improves the performance for intermediate classes at the expense of open water and fully-covered sea ice areas.
- 12 April 2023
- Published by Springer Nature in Scientific Reports
- Vol. 13 (1) , 1-10
Open AccessAI4SeaIce: selecting loss functions for automated SAR sea ice concentration charting
For maritime navigation in the Arctic, sea ice charts are an essential tool, which still to this day is drawn manually by professional ice analysts. The total Sea Ice Concentration (SIC) is the primary descriptor of the charts and indicates the fraction of ice in an ocean surface area. Naturally, automating the SIC chart creation is desired. However, the optimal representation of the corresponding machine-learning task is ambivalent and discussed in the community. In this study, we explore the representation with either regressional or classification objectives, each with two different (weighted) loss functions: Mean Square Error and Binary Cross-Entropy, and Categorical Cross-Entropy and the Earth Mover’s Distance, respectively. While all models achieve good results they differ as the regression-based models obtain the highest numerical similarity to the reference charts, whereas the classification-optimised models generate results more visually pleasing and consistent. Rescaling the loss functions with inverse class weights improves the performance for intermediate classes at the expense of open water and fully-covered sea ice areas.
- 1 June 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2020-2030
Investigating Spiking Neural Networks for Energy-Efficient On-Board AI Applications. A Case Study in Land Cover and Land Use Classification
Spiking neural networks have been attracting the interest of researchers due to their potential energy efficiency. This feature makes them appealing for applications on board CubeSats or small Earth observation satellites, given their strict energy consumption requirements. However, the performance of spiking neural networks in terms of the accuracy on the space-scene classification datasets, and their benefits with respect to the energy efficiency still remain to be demonstrated. This work is a preliminary investigation on deploying spiking neural networks to land cover and land use classification problems. To train a spiking model, a VGG-16-based artificial neural network has been trained on EuroSAT RGB benchmark dataset. The parameters of this network were then used to initialise a spiking model, which was optimised by fine-tuning the connection weights and the synaptic filters parameters. By using the mean neuron activations, and the number of time-steps and their width as proxies for the energy consumption of the models, this study shows different trade-offs between accuracy, and energy efficiency, when comparing a spiking model to the deep learning approach. Moreover, some additional input data preprocessing strategies are investigated as a method of further enhancement of the energy benefits of the spiking models.
- 1 June 2021
- Published by Institute of Electrical and Electronics Engineers (IEEE)
- p. 2020-2030
Investigating Spiking Neural Networks for Energy-Efficient On-Board AI Applications. A Case Study in Land Cover and Land Use Classification
Spiking neural networks have been attracting the interest of researchers due to their potential energy efficiency. This feature makes them appealing for applications on board CubeSats or small Earth observation satellites, given their strict energy consumption requirements. However, the performance of spiking neural networks in terms of the accuracy on the space-scene classification datasets, and their benefits with respect to the energy efficiency still remain to be demonstrated. This work is a preliminary investigation on deploying spiking neural networks to land cover and land use classification problems. To train a spiking model, a VGG-16-based artificial neural network has been trained on EuroSAT RGB benchmark dataset. The parameters of this network were then used to initialise a spiking model, which was optimised by fine-tuning the connection weights and the synaptic filters parameters. By using the mean neuron activations, and the number of time-steps and their width as proxies for the energy consumption of the models, this study shows different trade-offs between accuracy, and energy efficiency, when comparing a spiking model to the deep learning approach. Moreover, some additional input data preprocessing strategies are investigated as a method of further enhancement of the energy benefits of the spiking models.