Publications

This study addresses imbalanced data challenges in Mineral Prospectivity Mapping (MPM) using geophysical data from Lapland, Finland. It applies data-level imbalanced handling techniques and algorithm-level threshold adjustments to improve model predictions of rare mineral deposits. The performance of four ML models—MLP, RF, DT, and LR—was evaluated. The MLP model achieved the best accuracy (97.13%) on balanced data using synthetic oversampling.

This study explores deep-learning models for automated rice grain classification, which is essential for quality control, cost optimization, and supply chain efficiency. Testing models like ResNet, VGG, EfficientNet, and MobileNet on a dataset of 75,000 images across five rice categories, results showed that EfficientNet achieved the highest accuracy (99.67%), while MobileNet excelled in speed. The study concludes that deep-learning models offer scalable, efficient handling of large, complex data, outperforming traditional machine-learning methods.

This paper presents a new method using a Convolutional Neural Network (CNN) to estimate sea ice concentration in the Baltic Sea by combining Sentinel-1 and AMSR2 data. The CNN architecture retains different resolution inputs without losing valuable information. By using a focal loss function and skip connection, the proposed model achieves a high R2 score of 90.6%.

This study introduces short-term and long-term vessel trajectory prediction methods to enhance maritime traffic management. Tested with Baltic Sea AIS data, the short-term method shows high accuracy, while the long-term method optimizes speed and memory, promising improvements in efficiency and safety.

Presenting a novel CNN model merging SAR and AMSR2 imaging for precise maritime ice-water detection. This fusion enables automatic sea ice chart generation, outperforming uni-modal SAR architecture with 94.60% pixel-wise accuracy and an F1-score of 94.99%.

The paper explores different clustering methods using AIS data to detect two dangerous abnormal behaviors in maritime vessels like dark ships (vessels turning off AIS for illegal activities) and spiral vessel movements. It evaluates K-means, DBSCAN, AP, and GMM techniques using three months of AIS data from the Baltic Sea. Results show that K-means is effective in identifying these threatening events, contributing to improving maritime safety.

In recent years, one of the most widely used tools in the training and evaluation of physicians and surgeons is surgical simulators. Introducing virtual reality and adding haptic technology to the field has made it feasible to develop ever-better surgical simulators. In this regard, we set out to design a brain surgery bipolar electrocoagulation simulator using haptic technology.

Despite significant advances in artificial intelligence and high-speed hardware, computers still do not have the same human capabilities for many cognitive activities. In the human brain, a large part of these activities is performed by the brain neocortex, one of which is detecting anomalies. In this study, we intend to use hierarchical temporal memory to recognize abnormal states in time series. Hierarchical temporal memory is a deep learning method that seeks to mimic the structural and algorithmic features of the brain neocortex. First, we implement this deep learning method, and then we present a solution to improve anomaly detection results. In our solution, in addition to the abnormality score that the network generates, we use the distance between the forecasted values and the actual values. We also change the classifier of this deep learning method to a support vector machine. The model presented achieved an F1 score of 0.350, which, compared to other deep learning methods, was one of the highest scores and the processing speed of this method was ten times faster than other methods.

In this paper, we propose a fuzzy model predictive control method, which can be used in the control of highly nonlinear and complex systems, like chaotic ones. This method only uses the obtained time series of the system and does not require any prior knowledge about the system’s equations.