Özge Aslan Yıldız

Seasonal classification from satellite imagery is a crucial remote sensing task for monitoring vegetation dynamics, agricultural processes, environmental change, and climate-related spatial patterns. However, limited labeled data, regional variability, and the high computational cost of training large-scale networks from scratch make developing generalizable models for this task challenging. This study proposes a hybrid framework based on transfer learning for seasonal classification using monthly satellite imagery from all 81 provinces of Türkiye. Eight ImageNet pretrained deep learning models were evaluated as feature extractors, and the extracted representations were evaluated using seven machine learning classifiers. The selection of feature extractors and classifiers was performed using 10-fold cross-validation repeated five times. Among the feature extractor models, the combination of ConvNeXt and Multi-Layer Perceptron produced the best result with a mean accuracy of 0.698 and a mean F1-score of 0.696. As a result of the comparative analysis, ConvNeXt, Vision Transformer, and Swin Transformer were selected as the three most successful feature extractors, and Multi-Layer Perceptron was selected as the final classifier. The hybrid model, constructed by concatenating the features of these three selected feature extractors, achieved a mean accuracy of 0.725 and a mean F1-score of 0.724. After fine-tuning, the proposed hybrid model achieved the highest performance with a mean accuracy of 0.740 and a mean F1-score of 0.742. The results demonstrate that combining pretrained representations and fine-tuning improved seasonal classification in satellite imagery. The proposed framework offers an effective and computationally practical approach for seasonal classification, providing a promising foundation for future environmental monitoring and agricultural remote sensing applications.

Dial-a-Ride problem (DARP) is a specialized variant of the vehicle routing problem that focuses on designing optimal routes for users who specify pick-up and delivery requests between origins and destinations. This problem involves determining the most effective and efficient routes using limited vehicles and resources to offer responsive door-to-door transportation services for users with specific needs. Logistics service providers require systems that can deliver optimal solutions to these combinatorial problems within a reasonable time. Recently, there has been a significant increase in the use of artificial intelligence optimization algorithms such as meta-heuristics or learning-based approaches to solve such problems. Among the learning-based approaches, reinforcement learning has gained prominence for routing and scheduling tasks, owing to its ability to adaptively learn from complex state spaces and dynamically changing environments. In this study, a novel transformer-based deep reinforcement learning method is proposed to solve the Dial and Ride problem for a single service vehicle. In the proposed model, we adopt a modified transformer architecture instead of employing traditional linear layers, we integrate convolutional layers. To validate our approach, we conduct comprehensive experiments comparing our method against four well-known metaheuristic algorithms and a Deep Q-Network algorithm. The results indicate that proposed approach outperforms these techniques in terms of shorter total travel distances. Additionally, the proposed method is tested on a real-world scenario generated in the Buyukdere neighborhood of Eskisehir. The results demonstrate that the proposed method makes it possible to solve the problem within a reasonable time. This study confirms that the proposed deep reinforcement learning method can effectively address the Dial-a-Ride Problems.

The growth of the urban population and the increase in e-commerce activities have resulted in challenges for last-mile delivery. On the other hand, electric vehicles (EVs) have been introduced to last-mile delivery as an alternative to fossil fuel vehicles. Electric vehicles (EVs) not only play a pivotal role in reducing greenhouse gas emissions and air pollution but also contribute significantly to the development of more energy-efficient and environmentally sustainable urban transportation systems. Within these dynamics, the Electric Vehicle Routing Problem (EVRP) has begun to replace the Vehicle Routing Problem (VRP) in last-mile delivery. While classic vehicle routing ignores fueling, both the location of charging stations and charging time should be included in the Electric Vehicle Routing Problem due to the long recharging time. This study addresses the Capacitated EVRP (CEVRP) with a novel Q-learning algorithm. Q-learning is a model-free reinforcement learning algorithm designed to maximize an agent’s cumulative reward over time by selecting optimal actions. Additionally, a new dataset is also published for the EVRP considering field constraints. For the design of the dataset, real geographical positions have been used, located in the province of Eskisehir, Türkiye. It also includes environmental information, such as streets, intersections, and traffic density, unlike classical EVRP datasets. Optimal solutions are obtained for each instance of the EVRP by using the mathematical model. The results of the proposed Q-learning algorithm are compared with the optimal solutions of the presented dataset. Test results show that the proposed algorithm provides remarkable advantages in obtaining routes in a shorter time for EVs.

Software development for robotic systems is traditionally performed based on simulations, manual code implementation, and testing. However, this software development approach can cause safety issues in some scenarios, including multiple robots sharing a workspace. When different robots are executing individual planned tasks, they may collide when not adequately coordinated. Safety problems related to coordination between robots may not be encountered during testing, depending on timing, but may occur during the system’s operation. In this case, formal verification methods can provide a more reliable means to ensure the safety of robotic systems. This paper uses the formal method of model checking for the safety verification of multiple industrial robot manipulators with path conflicts. We give comparative results of two model-checking tools applied to a system with two robot manipulators. Whole workflows, from requirement specification to testing, are presented.