Yazar "Peker, Murat" seçeneğine göre listele

Listeleniyor 1 - 5 / 5
  • Küçük Resim Yok
    Öğe
    AI-Assisted Fungal Infection Detection Using Impedimetric Odour Biosensors
    (Erciyes Üniversitesi, 2025) Soylu, Efe Kayra; Safi, Soukaina; Atalay, Mustafa Altay; Peker, Murat
    Yeast infections have been widely recognized and if no quick and accurate treatment method is applied, they can be very dangerous and might even turn into death. In comparison with old-fashioned diagnostic solutions such as culturing, which takes around one to three days to reveal yeast infections, rapid and effective treatment is often not initiated. In the current study a novel method is offered involving the extraction of yeast fungal strain identification in a rapid, cost-effective, and accurate way. Through the application of agelatin-based hydrogel coating that represents the way in which odor receptors attach to cells a sensing concept for impedimetric odor was constructed. The hydrogel was further improved by adding glycerol for its structural stability and graphite powder for its better conductivity. The process of making a sensor involved applying the modified hydrogel to wires made of copper. The sensor was then exposed to the odor molecules from culture tests of Candida albicans, Candida glabrata, and Candida tropicalis, which were placed in a controlled environment. Changes in impedance took place, and these measurements were analyzed using a Random Forest machine learning algorithm that helped to get 94% classification success. This new testing process may lead to a revolution in the era of clinical diagnostics. It will enable speediness, simplicity, as well as precision in the detection of yeast fungal infections, which, in turn, will decrease health risks leading to unnecessary treatment costs by approved drug companies.
  • Küçük Resim Yok
    Öğe
    Comparative Analysis of ORB-SLAM2 and ORB-SLAM3 Under Visual Image Degradations
    (Institute of Electrical and Electronics Engineers Inc., 2025) Kiziloz, Dursun Alp; Peker, Murat
    This paper presents a comprehensive performance comparison between ORB-SLAM2 and ORB-SLAM3 systems under various image degradation scenarios. The freiburg3_sitting_halfsphere sequence from the TUM RGB-D dataset, which includes dynamic scenes, served as the experimental baseline. Three controlled variations of this sequence were generated by independently applying Gaussian blurring, image cropping, and contrast reduction to the original dataset. The algorithms were evaluated in a controlled Docker-based environment using RGB-D inputs. The evaluation focused on several key performance metrics including FPS, map point density per keyframe (MP/KF), Absolute Trajectory Error (ATE), and the x, y, and z components of the estimated trajectory. The findings reveal that ORB-SLAM3 generally outperformed ORB-SLAM2 in terms of tracking accuracy and FPS most scenarios, demonstrating its enhanced robustness, especially under conditions of blur and low contrast. However, in the cropped dataset both systems suffered a notable drop in performance. ORB-SLAM2 exhibited slightly better localization accuracy along the x-axis in cropped dataset, indicating that severe field-of-view reduction can diminish the advantages offered by ORB-SLAM3's architectural enhancements. This study uniquely provides an axis-wise trajectory evaluation of degraded visual data, offering insights into the robustness and limitations of modern feature-based SLAM systems in visually impaired environments. © 2025 IEEE.
  • Küçük Resim Yok
    Öğe
    Data-Adaptive Censoring for Short-Term Wind Speed Predictors Based on MLP, RNN, and SVM
    (Ieee-Inst Electrical Electronics Engineers Inc, 2022) Sarp, Ali Ogun; Menguc, Engin Cemal; Peker, Murat; Guvenc, Buket Colak
    This study introduces novel short-term wind speed predictors based on multilayer perceptron (MLP), recurrent neural network (RNN), and support vector machine (SVM) by combining them with the data-adaptive censoring (DAC) strategy. Taking into account the multistep ahead prediction mode, we design a DAC strategy based on the least mean square (LMS) algorithm, which iteratively obtains a new training dataset consisting of the most informative input-output wind data from all training set for MLP, RNN, and SVM structures. This enables us to censor less informative training data with high accuracy and thereby the training costs of the MLP, RNN, and SVM are reduced without a considerably adverse effect on their prediction performances in testing processes. The conducted simulation results on real-life large-scale short-term wind speed data verify the mentioned attractive features of the proposed predictors.
  • Küçük Resim Yok
    Öğe
    Horizontal attention convolution layer for stereo matching
    (Institute of Electrical and Electronics Engineers Inc., 2021) Emlek A.; Peker, Murat
    Obtaining a disparity map with stereo matching is one of the most important research topics in areas such as image processing and computer vision. Disparity maps are frequently used by autonomous systems that need depth information of the environment. Recently, high accuracy disparity maps have been obtained with end-to-end deep learning. In this study, a horizontal attention-based convolution layer has been proposed in order to better extract the sequential information in the horizontal plane in the rectified stereo images in methods based on deep learning. The proposed structure has been applied to the DispNetC network, which has been widely used in the literature, and has increased the performance of the network. On the other hand, the proposed method have a very low effect on the network's runtime. The results obtained are shown on the Scene Flow dataset. The codes of the study are available at the following address: https://github.com/aemlek/HADN.
  • Küçük Resim Yok
    Öğe
    P3SNet: Parallel Pyramid Pooling Stereo Network
    (Ieee-Inst Electrical Electronics Engineers Inc, 2023) Emlek, Alper; Peker, Murat
    In autonomous driving and advanced driver assistance systems (ADAS), stereo matching is a challenging research topic. Recent work has shown that high-accuracy disparity maps can be obtained with end-to-end training with the help of deep convolutional neural networks from stereo images. However, many of these methods suffer from long run-time for real-time studies. Therefore, in this paper, we introduce P3SNet, which can generate both real-time results and competitive disparity maps to the state-of-the-art. P3SNet architecture consists of two main modules: parallel pyramid pooling and hierarchical disparity aggregation. The parallel pyramid pooling structure makes it possible to obtain local and global information intensively from its multi-scale features. The hierarchical disparity aggregation provides multi-scale disparity maps by using a coarse-to-fine training strategy with the help of the costs obtained from multi-scale features. The proposed approach was evaluated on several benchmark datasets. The results on all datasets showed that the proposed P3SNet achieved better or competitive results while having lower runtime. The code is available at https://github.com/aemlek/P3SNet.