Scopus İndeksli Yayınlar Koleksiyonu / Scopus Indexed Publications Collection
Permanent URI for this collectionhttps://hdl.handle.net/11147/7148
Browse
16 results
Search Results
Conference Object Citation - Scopus: 1Pedestrian Equipment Anomaly Detection With Computer Vision in Warehouses(Avestia Publishing, 2024) Elçi,T.; Ünlü,M.Z.; Kantar,D.; Türker,A.Y.; Güney,H.; Ustaoğlu,A.The rapid growth of the logistics sector in recent years caused the expansion of warehouse areas and the increase in the number of equipment used. With the increase in these activities, the possibility of work accidents in warehouses also increases. In defiance of this situation, it has been determined that a real-time prediction system of pedestrian and equipment interaction is needed to ensure in-warehouse reliability. This system should address the urgent need to reduce the risk of work accidents and focus on the overall goal of reducing the possibility of work accidents in warehouse environments. To overcome this challenge, we propose a comprehensive Warehouse Anomaly Detection and Control System consisting of object detection, object tracking, action detection, and alarm classification components which will play an important role in increasing work safety in warehouse environments. YOLOv7 (You Only Look Once version 7) is a deep learning model that detects objects quickly and accurately in a single network pass. The deep learning-based Deep SORT algorithm used for object tracking provides a dynamic understanding of the warehouse environment by continuously storing these identified problems in real-time. The action detection part of this system is designed to identify and analyze actions and movements, recognizing anomalies and potential risks. In this part, the speed of pedestrians and equipment are detected utilization of 3D bounding boxes of objects and perspective transformation. The possible accident risks are measured using the intersection percentage of these areas, the magnitude of speed, the direction of the motion vector of pedestrian and equipment, and the distances between objects. Alert levels can be considered as encounter, near-miss, and emergency. Using this system in warehouses will reduce the risk of possible work accidents that may even result in death. © 2024, Avestia Publishing. All rights reserved.Conference Object Citation - Scopus: 19Thquad: Turkish Historic Question Answering Dataset for Reading Comprehension(Institute of Electrical and Electronics Engineers Inc., 2021) Soygazi,F.; Çiftçi,O.; Kök,U.; Cengiz,S.Question answering(QA) is a field in natural language processing and information retrieval, it aims to give answers to the questions using natural language. In this paper, we present the Turkish question answering dataset, which is THQuAD and baseline results with contextualized word embeddings. THQuAD consists of two different datasets one of them is TQuad on Turkish Islamic Science history within the scope of Teknofest 2018 "Artificial Intelligence competition", the second dataset on Ottoman history within the scope of Teknofest 2020 "Dogal Dil íçleme Yarismasi" prepared by us. THQuAD is a reading comprehension dataset, consisting of questions, answers, and passages. Our objective is to give an answer to a specific question by understanding the passage and extracting the answer from this passage. We generate contextualized word embeddings from pre-trained Turkish Bert, Electra, Albert language models after fine-tuning on different hyperparameters with neural networks. © 2021 IEEEArticle Citation - Scopus: 20Estrus Detection and Dairy Cow Identification With Cascade Deep Learning for Augmented Reality-Ready Livestock Farming(Multidisciplinary Digital Publishing Institute (MDPI), 2023) Arıkan, İ.; Ayav, T.; Seçkin, A.Ç.; Soygazi, F.Accurate prediction of the estrus period is crucial for optimizing insemination efficiency and reducing costs in animal husbandry, a vital sector for global food production. Precise estrus period determination is essential to avoid economic losses, such as milk production reductions, delayed calf births, and disqualification from government support. The proposed method integrates estrus period detection with cow identification using augmented reality (AR). It initiates deep learning-based mounting detection, followed by identifying the mounting region of interest (ROI) using YOLOv5. The ROI is then cropped with padding, and cow ID detection is executed using YOLOv5 on the cropped ROI. The system subsequently records the identified cow IDs. The proposed system accurately detects mounting behavior with 99% accuracy, identifies the ROI where mounting occurs with 98% accuracy, and detects the mounting couple with 94% accuracy. The high success of all operations with the proposed system demonstrates its potential contribution to AR and artificial intelligence applications in livestock farming. © 2023 by the authors.Article Label-Free Retraining for Improved Ground Plane Segmentation(Springer, 2022) Uzyıldırım, Furkan Eren; Özuysal, MustafaDue to increased potential applications of unmanned aerial vehicles over urban areas, algorithms for the safe landing of these devices have become more critical. One way to ensure a safe landing is to locate the ground plane regions of images captured by the device camera that are free of obstacles by deep semantic segmentation networks. In this paper, we study the performance of semantic segmentation networks trained for this purpose at a particular altitude and location. We show that a variation in altitude and location significantly decreases network performance. We then propose an approach to retrain the network using only a new set of images and without marking the ground regions in this novel training set. Our experiments show that we can convert a network’s operating range from low to high altitudes and vice versa by label-free retraining.Article Citation - Scopus: 1Performance Analysis and Feature Selection for Network-Based Intrusion Detection With Deep Learning(Türkiye Klinikleri, 2022) Caner, Serhat; Erdoğmuş, Nesli; Erten, Yusuf MuratAn intrusion detection system is an automated monitoring tool that analyzes network traffic and detects malicious activities by looking out either for known patterns of attacks or for an anomaly. In this study, intrusion detection and classification performances of different deep learning based systems are examined. For this purpose, 24 deep neural networks with four different architectures are trained and evaluated on CICIDS2017 dataset. Furthermore, the best performing model is utilized to inspect raw network traffic features and rank them with respect to their contributions to success rates. By selecting features with respect to their ranks, sets of varying size from 3 to 77 are assessed in terms of classification accuracy and time efficiency. The results show that recurrent neural networks with a certain level of complexity can achieve comparable success rates with state-of-the-art systems using a small feature set of size 9; while the average time required to classify a test sample is halved compared to the complete set.Conference Object Citation - WoS: 3Citation - Scopus: 4Deep Learning Based Segmentation Pipeline for Label-Free Phase-Contrast Microscopy Images(IEEE, 2020) Ayanzadeh, Aydın; Yalçın Özuysal, Özden; Okvur, Devrim Pesen; Önal, Sevgi; Töreyin, Behçet Uğur; Ünay, DevrimThe segmentation of cells is necessary for biologists in the morphological statistics for quantitative and qualitative analysis in Phase-contrast Microscopy (PCM) images. In this paper, we address the cell segmentation problem in PCM images. Deep Neural Networks (DNNs) commonly is initialized with weights from a network pre-trained on a large annotated data set like ImageNet have superior performance than those trained from scratch on a small dataset. Here, we demonstrate how encoder-decoder type architectures such as U-Net and Feature Pyramid Network (FPN) can be improved by an alternative encoder which pre-trained on the ImageNet dataset. In particular, our experimental results confirm that the image descriptors from ResNet-18 are highly effective in accurate prediction of the cell boundary and have higher Intersection over Union (IoU) in comparison to the classical U-Net and require fewer training epochs.Conference Object Citation - Scopus: 2Yara İyileşmesi Mikroskopi Görüntü Serilerinin Otomatik Analizi - Bir Ön-çalışma(IEEE, 2020) Mayalı, Berkay; Şaylığ, Orkun; Yalçın Özuysal, Özden; Pesen Okvur, Devrim; Töreyin, Behçet Uğur; Ünay, DevrimCollective cell analysis from microscopy image series is important for wound healing research. Computer-based automation of such analyses may help in rapid acquisition of reliable and reproducible results. In this study phase -contrast optical microscopy image series of an in-vitro wound healing essay is manually delineated by two experts and its analysis is realized, traditional image processing and deep learning based approaches for automated segmentation of wound area are developed and their perlOrmance comparisons are carried out.Conference Object Citation - Scopus: 1A Preliminary Study on Cell Motility Analysis From Phase-Contrast Microscopy Image Series(IEEE, 2020) Kayan, Emre; Kavuşan, Tarık; Önal, Sevgi; Pesen Okvur, Devrim; Yalçın Özuysal, Özden; Töreyin, Behçet Uğur; Ünay, DevrimAnalyses of morphology, polarity, and motility of cells is important for cell biology research such as metastatic and invasive capacity of cells, wound healing, and embryonic development. Automation of such analyses using image series of phase-contrast optical microscopy, which allows label-free imaging of live cells in their living environment, is a need. With this purpose, in this study image series of a cell motility experiment is manually annotated, and an automation algorithm realizing motion and shape analyses of cells using the annotated data is developed. In addition, due to the low number of annotated data at hand, a U-Net based solution is devised for automated segmentation of the cells and its performance is evaluated.Conference Object Citation - WoS: 2Citation - Scopus: 3Impact of Variations in Synthetic Training Data on Fingerprint Classification(IEEE, 2019) İrtem, Pelin; İrtem, Emre; Erdoğmuş, NesliCreating and labeling data can be extremely time consuming and labor intensive. For this reason, lack of sufficiently large datasets for training deep structures is often noted as a major obstacle and instead, synthetic data generation is proposed. With their high acquisition and labeling complexity, this also applies to fingerprints. In the literature, a number of synthetic fingerprint generation systems have been proposed, but mostly for algorithm evaluation purposes. In this paper, we aim to analyze the use of synthetic fingerprint data with different levels of degradation for training deep neural networks. Fingerprint classification problem is selected as a case-study and the experiments are conducted on a public domain database, NIST SD4. A positive correlation between the synthetic data variation and the classification rate is observed while achieving state-of-the-art results.Article Citation - WoS: 2Citation - Scopus: 2Improving Outdoor Plane Estimation Without Manual Supervision(Springer, 2022) Uzyıldırım, Furkan Eren; Özuysal, MustafaRecently, great progress has been made in the automatic detection and segmentation of planar regions from monocular images of indoor scenes. This has been achieved thanks to the development of convolutional neural network architectures for the task and the availability of large amounts of training data usually obtained with the help of active depth sensors. Unfortunately, it is much harder to obtain large image sets outdoors partly due to limited range of active sensors. Therefore, there is a need to develop techniques that transfer features learned from the indoor dataset to segmentation of outdoor images. We propose such an approach that does not require manual annotations on the outdoor datasets. Instead, we exploit a network trained on indoor images and an automatically reconstructed point cloud to estimate the training ground truth on the outdoor images in an energy minimization framework. We show that the resulting ground truth estimate is good enough to improve the network weights. Moreover, the process can be repeated multiple times to further improve plane detection and segmentation accuracy on monocular images of outdoor scenes.