Yazar "Erdogmus, Pakize" seçeneğine göre listele
Listeleniyor 1 - 10 / 10
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe Artificial Neural Network-Based 4-D Hyper-Chaotic System on Field Programmable(2020) Koyuncu, Ismail; Alcin, Murat; Erdogmus, Pakize; Tuna, MuratIn this presented study, a 4-D hyper-chaotic system newly proposed to the literature, has been implemented as Multi-Layer Feed-Forward Artificial Neural Network-based on FPGA chip with 32-bit IEEE-754-1985 floating-point number standard to be utilized in real time chaos-based applications. In the first step of the study, 4-D hyper-chaotic system has been numerically modeled on FPGA using Dormand-Prince numeric algorithm. In the second step,the data set (4X10,000) obtained from Matlab-based numeric model has been divided into two parts as training data set (4X8,000) and test data set (4X2,000) to create ANN-based 4-D hyper-chaotic system. A Multi-Layer Feed-Forward ANN structure with 4 inputs and 4 outputs has been constructed for ANN-based 4-D hyper-chaotic system. This structure has only one hidden layer and there are 8 neurons having Tangent Sigmoid activation function used as the activationfunction in each neuron.2.58E-07 Mean Square Error (MSE) value has been obtained from the training of ANN-based 4-D hyper-chaotic system. In the third step, after the successful training of ANN-based 4-D hyper-chaotic system, the design of ANN-based 4-D hyper-chaotic system has been carried out on FPGA by taking the bias and weight values of the ANN structure as reference. In this step, at first, Matlab-based Feed-Forward Multi-Layer 4X8X4 network structure has been coded in Very High Speed Integrated Circuit Hardware Description Language (VHDL) to be implemented on FPGA chips. Then, the bias and weight values of the ANN structure has been converted from decimal number system to floating-point number standard and these converted values have been embedded into the network structure.In the last step, the ANN-based 4-D hyper-chaotic system designed on FPGA has been synthesized and tested using Xilinx ISE Design Suite. The chip statistics have been given after the Place&Route process carried out for the Virtex XC6VHX255T-3FF1155 FPGA chip. The maximum operating frequency of ANN-based 4-D hyper-chaotic system on FPGA has been obtained as 240.861 MHZ.Öğe Evaluation of Classification Performance of New Layered Convolutional Neural Network Architecture on Offline Handwritten Signature Images(Mdpi, 2024) Ozkan, Yasin; Erdogmus, PakizeWhile there are many verification studies on signature images using deep learning algorithms in the literature, there is a lack of studies on the classification of signature images. Signatures are used as a means of identification for banking, security controls, symmetry, certificates, and contracts. In this study, the aim was to design network architectures that work very fast in areas that require only signature images. For this purpose, a new Si-CNN network architecture with existing layers was designed. Afterwards, a new loss function and layer (Si-CL), a novel architecture using Si-CL as classification layer in Si-CNN to increase the performance of this architecture, was designed. This architecture was called Si-CNN+NC (New Classification). Si-CNN and Si-CNN+NC were trained with two datasets. The first dataset which was used for training is the C-Signatures (Classification Signatures) dataset, which was created to test these networks. The second dataset is the Cedar dataset, which is a benchmark dataset. The number of classes and sample numbers in the two datasets are symmetrical with each other. To compare the performance of the trained networks, four of the most well-known pre-trained networks, GoogleNet, DenseNet201, Inceptionv3, and ResNet50, were also trained with the two datasets with transfer learning. The findings of the study showed that the proposed network models can learn features from two different handwritten signature images and achieve higher accuracy than other benchmark models. The test success of the trained networks showed that the Si-CNN+NC network outperforms the others, in terms of both accuracy and speed. Finally, Si-CNN and Si-CNN+NC networks were trained with the gold standard dataset MNIST and showed superior performance. Due to its superior performance, Si-CNN and Si-CNN+NC can be used by signature experts as an aid in a variety of applications, including criminal detection and forgery.Öğe An experimental comparison of the widely used pre-trained deep neural networks for image classification tasks towards revealing the promise of transfer-learning(Wiley, 2022) Kabakuş, Abdullah Talha; Erdogmus, PakizeThe easiest way to propose a solution based on deep neural networks is using the pre-trained models through the transfer-learning technique. Deep learning platforms provide various pre-trained deep neural networks that can be easily applied for image classification tasks. So, Which pre-trained model provides the best performance for image classification tasks? is a question that instinctively comes to mind and should be shed light on by the research community. To this end, we propose an experimental comparison of the six popular pre-trained deep neural networks, namely, (i) VGG19, (ii) ResNet50, (iii) DenseNet201, (iv) MobileNetV2, (v) InceptionV3, and (vi) Xception by employing them through the transfer-learning technique. Then, the proposed benchmark models were both trained and evaluated under the same configurations on two gold-standard datasets, namely, (i) CIFAR-10 and (ii) Stanford Dogs to benchmark them. Three evaluation metrics were employed to measure performance differences between the employed pre-trained models as follows: (i) Accuracy, (ii) training duration, and (iii) inference time. The key findings that were obtained through the conducted a wide variety of experiments were discussed.Öğe GENDER ESTIMATION WITH CONVOLUTIONAL NEURAL NETWORKS USING FINGERTIP IMAGES(2020) Sırma, Kerem; Erdogmus, PakizeBringing several innovations to our daily life, the importance of artificial intelligence technology hasbeen increasing day by day and has created new fields for researchers. Gender classification is also animportant research topic in the field of artificial intelligence. Studies on gender prediction from face,body, and even fingerprint images have been done. Also, today, biometric recognition systems havereached levels that can determine people's fingerprints, face, iris, palm prints, signature, DNA, andretina. In this study, various models were trained and tested on gender classification from fingertipimages. In the, a ready dataset was not used and finger images were collected from more than 200people. Rotation, cutting, and background reduction are applied to the collected images and madeready for the training. 4 different network models were set in the fieldwork. Data augmentation andtransfer learning were used in these models. Working in a limited area, the model we created hasachieved high-performance results, for all that the quality and angles of each image are different. Themodel proposed in this study has a performance rate of 86.39%.Öğe Generative Networks and Royalty-Free Products(2020) Özkan, Yasin; Erdogmus, PakizeIn recent years, with the increasing power of computers and Graphics Processing Units (GPUs), vast variety ofdeep neural networks architectures have been created and realized. One of the most interesting and generativetype of the networks are Generative Adversarial Networks (GANs). GANs are used to create things such asmusic, images or a film scenerio. GANs consist of two networks working simultaneously. Generative networkcaptures data distribution and discriminative network estimates the probability of the Generative Networkoutput, coming from training data of discriminative network. The objective is to both maximizing the generativenetwork products reality and minimize the discriminative network classification error. This procedure is aminimax two-player game. In this paper, it has been aimed to review the latest studies with GANs, to gather therecent studies in an article and to discuss the possible issues with royalty free products created by GANs. Withthis aim, from 2018 to today, the studies on GANs have been gathered to the citation numbers. As a result, therecent studies with GANs have been summarized and the potential issues related to GANs have been submitted.Öğe Hibrit Çok-Amaçlı Rüzgar Güdümlü Optimizasyon Algoritması(2020) Ay, Fethiye Sultan Özpehlivan; Erdogmus, PakizeTemel anlamda optimizasyon, bir veya birden fazla problemin belirli koşullar altındaki en iyi çözümlerini bulmaişlemidir. Günümüzde bu problemlerin çözümü için klasik yöntemler ve sezgisel yöntemler kullanılmaktadır.Sezgisel yöntemlerden biri olan Rüzgar Güdümlü Optimizasyon algoritması, rüzgarın atmosfer içerisindekihareketini temel alarak atmosferik dinamik eşitlikten yararlanan tek amaçlı optimizasyon problemlerine çözümarayan bir algoritmadır.Bu çalışmada çok-amaçlı optimizasyon problemlerinin çözümü için Rüzgar Güdümlü Optimizasyon algoritmasıyeniden düzenlenmiştir. Çok-amaçlı optimizasyon problemlerinde elde edilen sonuçların gerçek sonuçlara nekadar yakınsadığı ve bu sonuçların ne kadar çeşitli olduğu kullanılan yöntemlerin performansı hakkında bilgivermektedir. Baskın olmayan sıralama, ağırlıklı toplam, normal sınır kesişimi gibi metotlar çok-amaçlıoptimizasyon problemlerinde sıklıkla kullanılan yaklaşımlardır. Bu yaklaşımlardan bazıları çeşitlilik açısından önplana çıkarken bazılarının ise en iyi sonuca daha iyi yakınsadığı gözlenmiştir. Bu çalışmanın temel amacı eldeedilen çözümleri hem çeşitlilik hem de yakınsama açısından en iyi hale getirmektir.Bu amaç kapsamında baskın olmayan sıralama ve adaptif ızgara yaklaşımları bir arada kullanılarak yeni bir hibrityaklaşım geliştirilmiştir. Daha iyi bir yakınsama için baskın olmayan sıralama, çeşitlilik için adaptif ızgarayaklaşımı bir arada kullanılmıştır. Geliştirilen bu hibrit yaklaşım test problemleri ve doğrusal olmayan denklemsistemlerinde test edilerek sonuçları literatürde iyi bilinen Baskın Olmayan Sıralamalı Genetik Algoritma (NSGAII) ve Çok-Amaçlı Parçacık Sürü Optimizasyonu (MOPSO) algoritmaları ile karşılaştırılmıştır. Deneysel sonuçlarincelendiğinde çeşitlilik ve yakınsama performansı açısından geliştirilen hibrit yaklaşımın kabul edilebilir olduğugözlenmiştir.Öğe A novel handwritten Turkish letter recognition model based on convolutional neural network(Wiley, 2021) Kabakus, Abdullah Talha; Erdogmus, PakizeConvolutional neural networks have provided state-of-the-art solutions for many subfields of computer vision. While there exist many studies in the literature for several languages, studies for handwritten Turkish character recognition lack in the research field. To this end, we propose a novel handwritten Turkish letter recognition model based on a convolutional neural network. Since, to the best of our knowledge, there do not exist any publicly available handwritten Turkish letters datasets, we constructed a handwritten Turkish letters dataset that consists of 25,875 samples. To compare the performance of the proposed model with the related work, three state-of-the-art models, namely, VGG19, InceptionV3, and Xception, were utilized through the transfer learning technique. When these models were evaluated on the handwritten Turkish letter dataset, the proposed model's accuracy was calculated as high as 96.07% which was higher than the benchmark models. To measure the generalization ability of the proposed model, it was evaluated on a gold standard dataset, namely, EMNIST, and has achieved an accuracy of 80.54% which was higher than the benchmark models. Finally, the proposed model was trained and evaluated on the EMNIST dataset and it has achieved an accuracy of 94.61% which outperformed the related work.Öğe A Novel Hybrid Image Segmentation Method for Detection of Suspicious Regions in Mammograms Based on Adaptive Multi-Thresholding (HCOW)(Ieee-Inst Electrical Electronics Engineers Inc, 2021) Toz, Guliz; Erdogmus, PakizeSuspicious region segmentation is one of the most important parts of CAD systems that are used for breast cancer detection in mammograms. In a CAD system, there can be so many suspicious regions determined for a mammogram because of the complex structure of the breast. This study proposes a hybrid thresholding method to use in the CAD systems for efficient segmentation of the mammograms and reducing the number of the suspicious regions. The proposed method provides fully-automatic segmentation of the suspicious regions. This method is based on determining an adaptive multi-threshold value by using three different techniques together. These techniques are Otsu multilevel thresholding, Havrda & Charvat entropy, and w-BSAFCM algorithm that was developed by the authors of this paper for image clustering applications. In the proposed method, segmentation of a mammogram is performed on two sub-images obtained from that mammogram, the pectoral muscle and the breast region to prevent any information loss. The method was tested on 55 mass-mammograms and 210 non-mass mammograms of the mini-MIAS database, and it was compared with Shannon, Renyi, and Kapur entropy methods and with some of the related studies from the literature. The segmentation results of the tests were evaluated in terms of the number of suspicious regions, the number of correctly detected masses, and the performance measure parameters, accuracy, false-positive rate, specificity, volumetric overlap, and dice similarity coefficient. According to the evaluations, it was shown that the proposed method can both successfully locate the mass regions and significantly reduce the number of the non-mass suspicious regions on the mammograms.Öğe The promise of convolutional neural networks for the early diagnosis of the Alzheimer?s disease(Pergamon-Elsevier Science Ltd, 2023) Erdogmus, Pakize; Kabakus, Abdullah TalhaAlzheimer's Disease (AD) is one of the most devastating neurologic disorders, if not the most, as there is no cure for this disease, and its symptoms eventually become severe enough to interfere with daily tasks. The early diagnosis of AD, which might be up to 8 years before the onset of dementia symptoms, comes with many promises. To this end, we propose a novel Convolutional Neural Network (CNN) as a cheap, fast, yet accurate solution. First, a gold-standard dataset, namely DARWIN, that was proposed for the detection of AD through handwriting and consisted of 1D features, was used to generate the 2D features, which were yielded into the proposed novel model. Then, the proposed novel model was trained and evaluated on this dataset. According to the experimental result, the proposed novel model obtained an accuracy as high as 90.4%, which was higher than the accuracies obtained by the state-of-the-art baselines, which covered a total of 17 widely-used classifiers.Öğe Question answering system with text mining and deep networks(Springer Heidelberg, 2024) Ardac, Hueseyin Avni; Erdogmus, PakizeQuestion answering systems are capable of responding to user inquiries using natural language. These systems analyze questions utilizing natural language processing methods and retrieve responses from appropriate data sources using information retrieval techniques. Additionally, text mining and deep network techniques can enhance the effectiveness of question answering systems by providing more accurate and relevant information. In this study, we developed question answering models employing text mining and deep networks. We trained a pre-existing English BERT-base model with the Stanford Question Answering Dataset (SQuADv1.1) utilizing various hyperparameters and fine-tuning values. Our training yielded impressive results with an F1 score of 88.13 and an Exact Match (EM) rate of 80.74, outperforming previous studies in the field. An improvement study was conducted on the Turkish History Question Answering Dataset (THQuADv1.0), which led to the update of the dataset to THQuADv2.0 by adding questions regarding the units of D & uuml;zce University. The pre-trained Turkish BERTurk-base model received training with the THQuADv2.0 dataset utilizing the successful hyperparameters and fine-tuning values obtained in the English model. As a consequence of the training, we developed the BERTDuQuA (BERT D & uuml;zce University Question Answering) model for answering Turkish questions. The BERTDuQuA model demonstrated exceptional performance, achieving an F1 score of 87.10 and an EM of 76.90.
