http://dspace.beu.edu.tr:8080/xmlui/handle/123456789/15607 2026-04-23T06:58:24Z 2026-04-23T06:58:24Z ÇİFTÇİ, Bahar TEKİN, Ramazan http://dspace.beu.edu.tr:8080/xmlui/handle/123456789/15726 2025-08-21T06:57:42Z 2024-01-01T00:00:00Z

Deep Learning Based Offline Handwritten Signature Recognition ÇİFTÇİ, Bahar; TEKİN, Ramazan In our digitalized world, the need for reliable authentication methods is steadily increasing. Biometric authentication methods are divided into two main categories: physiological and behavioral. While physiological biometrics include features such as face, iris, and fingerprint, behavioral biometrics encompass dynamics such as gait, speech, and signature. Most of these methods require specialized equipment, whereas signatures can be easily obtained without additional tools, making them ideal for verifying the legality of documents. Although manual signature recognition is effective, it is resource-intensive, slow, and susceptible to errors. With advancements in technology, the need to automate the signature recognition process to enhance accuracy and efficiency has become increasingly important. Based on this need, in this study, five different DL techniques (GoogLeNet, MobileNet-V3 Large, Inception-V3, ResNet50 and EfficientNet-B0) are used to classify signature images with detailed analyses. DL methods have outperformed traditional techniques by leveraging the power of CNNs to automatically learn and extract complex features from signature data. The dataset used consists of a total of 12,600 images belonging to 420 individuals, each contributing 30 original signatures. The dataset is divided into training, validation, and test sets in different proportions to analyze classification performance. The pre-trained DL models were fine-tuned to optimize their parameters for the signature dataset. The results demonstrate that DL models achieve high accuracy in signature classification, with the GoogLeNet and Inception-V3 models reaching an accuracy of 98.77% at a 20% test rate. The study also highlights the impact of different test rates on model performance.

2024-01-01T00:00:00Z YILMAZ ACAR, Züleyha http://dspace.beu.edu.tr:8080/xmlui/handle/123456789/15724 2025-08-21T06:54:44Z 2024-01-01T00:00:00Z

A Machine Learning Prediction Model for Myelitis and Multiple Sclerosis Based on Fourier Transform Features from MRI Images YILMAZ ACAR, Züleyha Abstract Myelitis is a neurodegenerative disease positioned in the spinal cord, with multiple sclerosis (MS) being a common subtype. Radiological indicators enable the diagnosis of these diseases. This study proposes a classification framework to detect myelitis, MS, and healthy control (HC) groups using magnetic resonance imaging (MRI) images. The feature extraction step involves applying the fast Fourier transform (FFT) to MRI images. FFT is important because it converts spatial data into the frequency domain, making it easier to identify patterns and abnormalities that indicate these diseases. Then, statistical features (mean, minimum, maximum, standard deviation, skewness, kurtosis, and total energy) are extracted from this frequency information. These features are then used to train support vector machine (SVM), k-nearest neighbor (KNN), and decision tree algorithms. In multi-class classification (myelitis vs. MS vs. HC), the proposed method achieves a classification accuracy of 99.31% with SVM, with average precision, recall, and F1-score values of 99.27%, 99.21%, and 99.24%, respectively, indicating effective classification across all classes. In the binary class classification (HC vs. MS, MS vs. myelitis, HC vs. myelitis), the SVM achieves an outstanding classification accuracy of 99.36%, 99.71%, and 100% respectively. This study highlights the efficiency of FFT-based feature extraction in forming detection patterns for classifying HC, MS, and myelitis classes.

2024-01-01T00:00:00Z ÇİMEN, Ömür KESKİN, S.Nilay http://dspace.beu.edu.tr:8080/xmlui/handle/123456789/15722 2025-08-21T06:51:51Z 2024-01-01T00:00:00Z

Determining the Suction Capacity of Compacted Clays with Fuzzy-Set Theory ÇİMEN, Ömür; KESKİN, S.Nilay Abstract Water suction capacity is an important parameter affecting soil's swelling properties and volumetric change. The water suction capacity is determined through timeconsuming laboratory experiments. However, this has random errors due to the heterogeneous and anisotropic structure of the soil sample together with the error caused by the operator made the experiment. Solving such an estimation problem including error can be easily achieved using fuzzy-set theory. In this study, we use fuzzy-set theory to predict the suction capacity of compacted clayey soils. For this reason, the engineering properties of clayey soil (plasticity index, dry density, initial water content, and suction capacity) are partitioned into fuzzy subsets, and fuzzy rules are formed. Later, a computer program in the Fortran language is written to estimate the suction capacity of compacted clayey soil from these properties. It is shown that there is a good similarity between the results of the tests and the proposed fuzzy logic model.

2024-01-01T00:00:00Z KARADENIZ, Alper Talha http://dspace.beu.edu.tr:8080/xmlui/handle/123456789/15720 2025-08-21T06:48:58Z 2024-01-01T00:00:00Z

Automatic Classification of Melanoma Skin Cancer Images with Vision Transform Model and Transfer Learning KARADENIZ, Alper Talha Melanoma is one of the most aggressive and lethal forms of skin cancer. Therefore, early diagnosis and correct diagnosis are very important for the health of the patient. Cancer diagnosis is made by field experts and this increases the possibility of error. Today, with the developing deep learning technology, it has been seen that automatic detection of Melanoma skin cancer can be performed with high accuracy by computer systems. One of the latest technologies developed in the field of deep learning is the Vision Transformer (ViT) model. This model was produced by Google and has achieved very successful results in the field of classification. This study aims to detect melanoma skin cancer with high accuracy using the ViT model. In the study, the melanoma skin cancer dataset consisting of 9600 training and 1000 test images in the Kaggle library was used. In order to use the data set more effectively, some pre-processing methods were first applied. Model performance was evaluated using the transfer learning approach together with the ViT model on this data set. Training and experimental testing of the model was carried out with Python language on the Colab platform. As a result of the experimental studies carried out on the test data set, it was seen that the model reached 93.5% accuracy rate. This rate is competitive and promising when compared to existing models in the literature.

2024-01-01T00:00:00Z