Mehmet Kaya

In this study, the design and control of a mobile gardening robot based on image processing and artificial intelligence were implemented. The system, running on Raspberry Pi 4B, can distinguish different plant species and weeds using a YOLOv5-based object detection and classification model; it performs tasks such as cutting weeds, watering useful plants, and moving autonomously while avoiding obstacles. Experiments conducted in a pot environment demonstrated that the robot can successfully perform plant detection, classification, and maintenance functions. With these features, the developed robot offers a low-cost and sustainable solution for small and medium-scale agricultural businesses. The mobile gardening robot designed within the scope of the study was experimentally tested for image processing-based plant recognition, autonomous plant orientation, plant cutting, and watering operations for different scenarios. When the results obtained were examined, it was seen that the developed deep learning-based plant detection model achieved an 82% mAP@50 success rate, while the plant classification model achieved a 92.9% mAP@50 success rate.

Medical Visual Question Answering (Med-VQA) systems have the potential to support medical image interpretation and clinical decision-making processes. However, the “black-box” nature of existing systems and low-resolution constraints limit the transparency of model decisions, hindering clinical applicability. This work proposes a high-resolution holistic framework called robust and efficient layer-integrated gradients (RE-LIG) to enhance reliability and explainability in Med-VQA systems. The proposed architecture is built upon three key components: (1) high-resolution visual encoding: the PubMedCLIP encoder is scaled to high-resolution using dynamic positional embedding interpolation to capture fine details. (2) Multimodal semantic fusion: clinical questions solved by BioLinkBERT and visual features obtained by PubMedCLIP are aligned through a coattention mechanism. (3) Explainability framework: to counter the noisy nature of classical gradient methods, the RE-LIG algorithm, which combines noise tunneling and layer-based integration strategies, has been integrated into the system. Extensive experiments conducted on the SLAKE dataset demonstrate the proposed framework’s success in primarily increasing model faithfulness. Quantitative analyses demonstrate that the RE-LIG method achieves a + 28.9% higher explanation fidelity (RE-LIG AOPC = 0.3180 vs. Vanilla IG = 0.2467, Bootstrap 95% CI [0.262–0.375], Wilcoxon p < 0.001) compared to standard gradient approaches. While achieving this gain in explainability, competitive performance with state-of-the-art (SOTA) models was achieved without compromising diagnostic performance (80.77% overall accuracy, 87.61% closed-ended, and 77.34% open-ended performance). Ablation studies confirm that the integrated noise reduction mechanisms shift the model’s focus from background noise to actual pathological boundaries. The findings demonstrate that explainability is not merely a visual aid for clinical confidence but a measurable and verifiable requirement.

Climate, and especially cities with hot climatic conditions, directly impact human life. In this study, hourly datasets from the central meteorological station in Diyarbakır city center for the years 1990–2022 were utilized. These data were analyzed using RayMan Pro-2.1 software, and Physiological Equivalent Temperature values were derived. The obtained Physiological Equivalent Temperature values were analyzed using the SARIMAX model implemented on a machine learning infrastructure to uncover the changes between 2022 and 2050. According to the results obtained, the Physiological Equivalent Temperature value, which was 15.42 °C in 1990 in real terms, increased by 21.3% to 18.66 °C in 2022. According to the SARIMAX model predictions, Physiological Equivalent Temperature values in 2022 are estimated to rise to 21.42 °C by 2050, reflecting an increase of 14.79%. The aim of this study is to examine the temporal variations in human bioclimatic comfort values and provide a foundation for future predictions. This will contribute to the development of urban master plans by local and administrative authorities.

As optical access networks continue to evolve toward higher capacity, longer reach, and increased user density, accurately predicting transmission performance has become increasingly complex. Conventional physics-based models often struggle to capture the nonlinear and stochastic behavior of modern passive optical networks (PONs), particularly under diverse operating conditions. In this study, a hybrid deep learning (DL) framework is proposed for the prediction of key performance indicators, including Q-factor, receiver sensitivity, and bit error rate (BER), in asymmetric 160/80 Gbps TWDM-PON systems, which is the target capacity by ITU-T G.989.1 specifications. The proposed approach integrates Gradient Boosting Regression and Multi-Layer Perceptron models within an ensemble learning structure to enhance robustness and predictive accuracy. A synthetic dataset comprising 1000 samples was generated to emulate realistic transmission scenarios with variations in distance, power level, and noise conditions for both upstream and downstream channels. Experimental results demonstrate strong agreement between the proposed DL-based predictions and conventional optical simulation outcomes, while the proposed predictions achieve superior adaptability and reduced computational complexity. High coefficients of determination (R² > 0.94) and low error metrics confirm the effectiveness of the framework, highlighting its potential as a fast and reliable alternative to traditional performance evaluation methods in next-generation optical access networks.

Diagnosing diseases from medical images and reporting them at the paragraph level is a significant challenge for deep learning-based autonomous systems. Existing work primarily focuses on achieving high accuracy, often paying less attention to the computational cost of training and testing. The goal of this work is to build a low computational cost and high-performance hybrid encoder-decoder architecture capable of producing autonomous medical reports. On the encoder side of our architecture, called FAST-MRG, features are extracted from images with a transformer-based encoder enriched with distillation techniques, while on the decoder side, a generative pre-training transformer generates paragraph-level text using the extracted features. Numerical analysis with word matching evaluation metrics, temporal analysis and observational analysis were performed to measure the success of the architecture. Our hybrid encoder-decoder architecture was trained and tested using chest X-ray images and reports from the Indiana University Chest X-ray collection dataset. The FAST-MRG architecture achieved scores of 0.373, 0.226 and 0.332 on the Bleu-1, Meteor and Rouge evaluation metrics, respectively. It also has an average time efficiency of 66% compared to previous work using similar GPU environments. The study demonstrates through experiments that meaningful reports are produced that can support doctors in diagnosis and treatment processes. In the study, the results are presented not only with measurable average values but also with a density distribution graph and the test results are analyzed in depth. With its low runtime and high performance, the proposed architecture can serve as a basis for future work.

Background/Objectives: Early and accurate diagnosis of chest diseases is a critical challenge in clinical practice, particularly in scenarios where multiple pathologies may coexist. While deep learning-based medical image analysis has shown promising results, most existing studies rely on unimodal data and fixed-scale datasets, limiting their generalizability and clinical relevance. In this study, we present a comprehensive comparative analysis of unimodal and multimodal deep learning models for multi-label chest disease classification using chest X-ray images and associated clinical metadata. Methods: A total of twelve models were developed based on three widely used convolutional neural network architectures—ResNet50, EfficientNetB3, and DenseNet121—under both unimodal (image-only) and multimodal (image + clinical data) configurations. To systematically investigate the impact of data scale, experiments were conducted on two distinct versions: the Random Sample of NIH Chest X-ray Dataset and the NIH Chest X-ray Dataset, containing 5606 and 121,120 samples, respectively. Model performance was evaluated using label-based Area Under the Receiver Operating Characteristic Curve (AUROC) metrics. Results: Experimental results demonstrate that multimodal fusion consistently outperforms unimodal approaches across all architectures and data scales, with more pronounced improvements observed in large-scale settings. Furthermore, increasing data volume leads to improved generalization and reduced performance variance, particularly for rare pathologies. Conclusions: These findings highlight the effectiveness of multimodal, multi-label learning in enhancing diagnostic accuracy and support the development of robust clinical decision support systems for chest disease assessment.