UDUğur Demiroğlu

Uğur Demiroğlu

*This is an algorithmically generated author record based on publicly available information.

Current Organizations / Institutes

The organizations/institutes from the scholar's latest article


Uğur Demiroğlu has a total of 6 co-authors. Here are the top co-authors by number of shared publications:


Latest Publications

  • Benchmarking Deep Learning Models for Breast Cancer Detection: A Comparison of Vision Transformers and CNNs

    by Uğur Demiroğlu, Bilal Şenol

    Breast cancer is a major global health issue, and accurate early detection is critical for improving patient outcomes. Deep learning-based image classification techniques have shown remarkable success in medical imaging, particularly convolutional neural networks (CNNs) and transformer-based models. This study evaluates and compares the performance of Vision Transformers (ViTs) with well-established CNN architectures, including AlexNet, ResNet-50, and VGG-19, for breast cancer image classification. The research aims to investigate whether ViTs can outperform conventional deep learning models in this domain and to analyze their strengths and limitations. The study utilizes a publicly available breast cancer dataset comprising 9,248 images categorized into benign, malignant, and normal classes. The dataset is preprocessed by resizing all images to 224×224 pixels, normalizing pixel intensity values, and applying data augmentation techniques. All models are trained under the same conditions using 80% of the data for training, 10% for validation, and 10% for testing. Performance evaluation is conducted based on accuracy, precision, recall, and F1-score metrics. Experimental results indicate that ResNet-50 achieves the highest classification accuracy (93.62%), outperforming the other models in terms of overall performance. AlexNet, despite having the smallest parameter count, delivers competitive accuracy (88.32%) while being computationally efficient. VGG-19, known for its depth, achieves 87.51% accuracy but has the highest computational cost. ViTs, although promising, achieve a lower accuracy of 87.46%, suggesting that transformer-based architectures may require larger datasets and further optimization to surpass traditional CNNs in medical image classification tasks. This study highlights that CNN-based models, particularly ResNet-50, remain the most effective approach for breast cancer classification in the given dataset. However, ViTs present a potential alternative, and future research should explore hybrid models integrating both CNN and transformer-based architectures to enhance classification performance.


  • Open Access

    Multi-Objective Step Response Shaping via the Fractional-Order Proportional-Integral-Derivative Controller

    by Uğur Demiroğlu, Bilal Şenol

    A well-known problem in control system design and analysis is the shaping of the unit step reference response of a system to produce desired transient characteristics for various system references. The necessity of having fast, accurate, and stable control systems for a large number of practical applications has created the need for advanced control methods. In this regard, the development of fractional-order controllers has received considerable attention from the control community. Many papers and books on the topic of fractional-order systems have been published, which also include the usefulness of fractional calculus in the area of controllers. The fractional order proportional integral derivative controller is proven to be versatile, and its design can be obtained for any given target step response. A sufficiently large number of response characteristics, such as performance, phase margin, immunity to plant modeling, and robustness, can be adjusted by means of five tuning parameters. The control strategy of this paper focuses on developing a fractional order proportional integral derivative controller, which aims at overcoming the infeasibility of the controller to satisfy the conflicting goals of go-to speed and settling time in the traditional PID controller. The controller design has two main goals: one is to satisfy system stability, while the other is tuning the overshoot and the settling time. In this direction, the genetic algorithm is implemented. The results are presented through an illustrative example.

  • Open Access

    Optimizing PI Controller for Stability and Overshoot in Step Response Using GA and PSO Techniques, A Comparative Study

    by Bilal Şenol, Uğur Demiroğlu

    In this paper, we present a comprehensive and in-depth investigation on the optimization of Proportional-Integral (PI) controller tuning for achieving stability and desired overshoot in the step response. The main objective of this study is to compare the effectiveness of Genetic Algorithm (GA) and Particle Swarm Optimization (PSO) techniques in finding the optimal parameters for the PI controller. The PI controller is a widely used control algorithm that plays a crucial role in many industrial processes. Its tuning greatly affects the system's performance, particularly in terms of stability and overshoot. Therefore, finding the optimal tuning parameters is of utmost importance. To address this optimization problem, we propose the utilization of two popular metaheuristic algorithms, GA and PSO. These algorithms are known for their ability to efficiently search through large solution spaces and find near-optimal solutions. By applying these algorithms to the PI controller tuning problem, we aim to determine which technique yields better results in terms of stability and overshoot tuning. In our comparative study, we provide a detailed explanation of both GA and PSO algorithms, focusing on their working principles and mathematical formulations. We also describe how these algorithms can be applied to the PI controller tuning problem. Furthermore, we highlight the key differences between GA and PSO, shedding light on their strengths and weaknesses. To assess the performance of GA and PSO, we conduct several experiments using different benchmark functions and step response models. We measure the stability and overshoot metrics for various parameter settings obtained through GA and PSO. By thoroughly analyzing the obtained results, we draw meaningful conclusions regarding the effectiveness of each technique. Our findings demonstrate that both GA and PSO exhibit promising results in optimizing PI controller tuning. These observations provide valuable insights and guidelines for choosing the appropriate algorithm based on specific control requirements. In conclusion, this comparative study is thought to contribute to the field of control systems engineering by offering a comprehensive analysis of GA and PSO techniques in the context of PI controller tuning. By highlighting their strengths and weaknesses, it is aimed to provide researchers and practitioners with valuable information for making informed decisions when optimizing control parameters for stability and overshoot reduction purposes.

  • Open Access

    Integrating Pretrained Deep Neural Networks with Traditional Classification Techniques for Enhanced Oral Cancer Diagnosis

    by Bilal Şenol, Uğur Demiroğlu

    The purpose of this research is to present a hybrid approach to the classification of oral cancer images. This approach combines traditional classification methods such as Support Vector Machines (SVM), K-Nearest Neighbors (KNN), and Decision Trees with advanced feature extraction from pretrained deep neural networks (GoogleNet and MobileNetV2). Through the use of the suggested method, features are extracted from the deep learning models, resulting in the formation of a robust hybrid model that enhances diagnostic accuracy. The hybrid model achieves a classification accuracy of 90.01% with Quadratic SVM, which represents a 22.36% improvement over solo deep learning models. Comparative analyses indicate the tremendous performance advantages that the hybrid model has achieved. The findings highlight the potential of merging contemporary deep learning skills with older methods in order to improve the accuracy and dependability of medical picture categorization, particularly in the diagnostic process for oral cancer.


  • Open Access

    Evaluating Vision Transformer Models for Breast Cancer Detection in Mammographic Imaging

    by Uğur Demiroğlu, Bilal Şenol

    Breast cancer is a leading cause of mortality among women, with early detection being crucial for effective treatment. Mammographic analysis, particularly the identification and classification of breast masses, plays a crucial role in early diagnosis. Recent advancements in deep learning, particularly Vision Transformers (ViTs), have shown significant potential in image classification tasks across various domains, including medical imaging. This study evaluates the performance of different Vision Transformer (ViT) models—specifically, base-16, small-16, and tiny-16—on a dataset of breast mammography images with masses. We perform a comparative analysis of these ViT models to determine their effectiveness in classifying mammographic images. By leveraging the self-attention mechanism of ViTs, our approach addresses the challenges posed by complex mammographic textures and low contrast in medical imaging. The experimental results provide insights into the strengths and limitations of each ViT model configuration, contributing to an informed selection of architectures for breast mass classification tasks in mammography. This research underscores the potential of ViTs in enhancing diagnostic accuracy and serves as a benchmark for future exploration of transformer-based architectures in the field of medical image classification.


  • Open Access

    Diagnosis of the Skin Cancer by Vision Transformers

    by Uğur Demiroğlu

    Skin cancer, one of the most frequent cancers, requires early identification for efficient treatment and better survival. Early diagnosis relies on precise and quick skin lesion categorization into benign and malignant categories. This work uses Vision Transformers (ViTs) to classify skin cancer photos by modeling long-range relationships and capturing complicated visual patterns. ViTs, originally created for natural language processing, have showed great potential in picture classification tasks because to their self-attention processes, outperforming CNNs. A public collection of 270 skin lesion images—240 malignant and 30 benign—was used in this study. Preprocessing included scaling and normalizing the dataset to 384x384x3 and splitting it into 80% training and 20% testing sets. Transfer learning optimised a pre-trained ViT model for this job. To improve accuracy and avoid overfitting, hyperparameters were carefully selected for network training. Parallel computing accelerated training to 30 minutes and 20 seconds. Vision Transformers classify medical images well, according to the study. The ViT model outperformed numerous other methods with 98.15% accuracy on the test set. A confusion matrix analysis showed great sensitivity in detecting malignant lesions and low misclassification of benign patients. These findings show that ViTs can capture detailed medical picture aspects, making them useful for dermatological diagnoses. This study shows that Vision Transformers can improve diagnosis accuracy and lays the groundwork for their use in other medical imaging fields. In the battle against skin cancer and other illnesses that need early diagnosis, ViTs' scalability, efficiency, and precision are important. Future research might integrate ViTs with other deep learning architectures to improve robustness and flexibility. This study adds to the data supporting sophisticated AI in medical diagnostics and lays the groundwork for automated, reliable, and efficient healthcare solutions.


  • Open Access

    Feature selection and classification of brain midline tumors using tiny vision transformers and neighborhood component analysis

    by Uğur Demiroğlu, Bilal Şenol

    Brain midline tumors pose significant diagnostic challenges due to their complex structure and variability in presentation. In this study, we propose a robust classification framework for detecting brain midline tumors using image classification and feature selection techniques. The images, categorized into positive and negative classes, were processed using the Google Tiny Model Vision Transformer without any additional training. Feature extraction was performed by utilizing the head layer of the Tiny Model Vision Transformer, which yielded 1,000 features from the fully connected layer based on the pre-trained weights of the network. These features were initially classified using classical machine learning classifiers such as support vector machines, k-nearest neighbors and decision trees. To improve classification accuracy and reduce computational costs, Neighborhood Component Analysis was applied as a feature selection method. Neighborhood Component Analysis selected the top 460 most informative features from the initial set of 1,000 features. These selected features were subsequently used for classification, and the performance was compared with the results obtained using the full feature set. The comparative analysis revealed that the Neighborhood Component Analysis-based feature selection significantly enhanced classification accuracy and reduced processing time without sacrificing model reliability. The findings demonstrate that combining Tiny Model Vision Transformers with Neighborhood Component Analysis is an effective approach for brain midline tumor classification, offering a balance between accuracy and computational efficiency. This method holds promise for improving early diagnosis and aiding clinical decision-making, making it a valuable tool in medical image analysis and brain tumor detection.

  • Preprocessed Vision Transformers and Classical Classifiers in Diagnosing Skin Diseases

    by Bilal Şenol, Uğur Demiroğlu

    Vision Transformers (ViTs) are the state-of-the-art deep learning technology in medicine. ViTs require a large number of parameters, so they need a relatively large dataset for learning. This is currently possible due to the digitization of healthcare. As a comparison, we also use classical classifiers, which are characterized by a relatively low number of input data. In clinical practice, high-resolution images such as those from dermoscopy, confocal microscopy, reflectance confocal microscopy, and Raman spectroscopy are used to diagnose skin diseases. ViTs have potential in clinical practice. The advantage of the model over convolutional neural networks is that they do not use convolutional operations. Preprocessed images from a dataset were classified experimentally using five ViTs models of various sizes and respective classical classifiers. Comparative experiments were conducted also on preprocessed dermatoscopic images from another dataset. This article introduces an artificial intelligence method for identifying various skin conditions. The dataset contains images that are classified into 5 categories: normal, melanoma, arsenic, psoriasis, and eczema. During the study, skin images underwent initial processing using the Adaptive Histogram Equalization (AHE) technique, which enhanced the contrast to reveal important details. Following this preprocessing, features were obtained from the images using ViTs, renowned for their ability to capture intricate visual information. These extracted features were then utilized in conjunction with traditional machine learning classifiers, resulting in accurate diagnosis of the skin conditions being studied. The findings emphasize the effectiveness of combining ViTs with classical classifiers in tasks related to medical image classification.


  • Tumor Detection by Classification of Brain MRI Images Using the Vision Transformers

    by Uğur Demiroğlu

    The interplay between applied mathematics and artificial intelligence is pivotal for advancing both fields. AI fundamentally relies on statistical and mathematical techniques to derive models from data, thus enabling computers to improve their performance over time. Classification of brain MRI images for tumor detection has improved significantly with the advent of machine learning and deep learning techniques. Classical classifiers such as Support Vector Machines (SVM), Tree, and k-Nearest Neighbors (k-NN) have been widely used in conjunction with feature extraction methods to improve the accuracy of tumor detection in MRI scans. Recent studies have shown that classical classifiers can effectively analyze features extracted from MRI images, which can lead to improved diagnostic capabilities. Feature extraction is a critical step in the classification process. Classification of brain MRI images using Vision Transformers (ViT) represents a significant advancement in medical imaging and tumor detection. ViT leverages the transformer architecture, which is highly successful in natural language processing, to effectively process visual data. This approach allows capturing long-range dependencies within images and enhances the ability of the model to distinguish complex patterns associated with brain tumors. Recent studies have demonstrated the effectiveness of ViT in various classification tasks, including medical imaging. In our study, the classification accuracy of the dataset from the ViT network was 78.26%. In order to increase tumor detection performance, features of ViT network were extracted and given to classical classifiers, and 81.9% accuracy was achieved in Tree classifier. As a result, classification of brain MRI images using ViT represents a new approach with the strengths of deep learning and traditional machine learning methods, namely feature extraction and classification in classical classifiers.


  • Analyzing complex fractional order systems physical phenomena in IOPI controller design

    by Bilal Şenol, Uğur Demiroğlu

    The Fractional Complex Order Plant model, which has lately gained popularity in applied physics and control systems, is the main subject of this study. The major contribution of this study to the literature is the discussion of the physical phenomena of complex plant models and how they affect the stability and robustness of the systems. Because the Fractional Complex Order Plant model is the most general mathematical form, other plant models covering the Integer Order Plant and the Fractional Order Plant can be easily created with this benefit. The proposed approach using the classical Proportional Integral controller which is recalled as the Integer‐Order PI controller in this paper gives the calculation equations of the physical alterations of plants having integer, fractional, and complex orders. Along with the visuals and with the aid of simulations, the consequences of the parameters on the system are described. Additionally, the advantages and disadvantages of the proposed controller designs for each of the three plant species are discussed.