Yazar "Dinc, Baris" seçeneğine göre listele
Listeleniyor 1 - 4 / 4
Sayfa Başına Sonuç
Sıralama seçenekleri
Öğe A Novel Hybrid Optic Disc Detection and Fovea Localization Method Integrating Region-Based Convnet and Mathematical Approach(Springer, 2023) Dinc, Baris; Kaya, YasinOptic disc (OD), fovea, and blood vessels are major anatomical structures in a fundus image. In retinal image processing, automated detection of structural parts is crucial in analyzing image patterns and abnormalities caused by eye diseases such as glaucoma, macular edema, and diabetic retinopathy. This paper presents a robust and efficient OD detection and fovea localization method integrating region-based deep convolutional neural networks and a mathematical approach. The proposed model consists of two stages: In the first stage, we generated multiple OD region proposals and then detected the OD based on the boundary box with the highest score using Faster R-CNN. In the second stage, we calculated the localization of the fovea employing a mathematical model considering the coordinates of the predicted OD region. We used four publically available fundus image databases, ORIGA-light, DRIVE, DIARET-DB1, and MESSIDOR, to evaluate our model. The proposed hybrid model was trained with 70% images of the ORIGA-light database and tested on 30% images of ORIGA-light and all images of the other databases. To show the robustness of the model, all databases were divided into two parts, as normal and diseased samples. The presented model achieved a reliable and flexible performance in detecting the OD, with overall IoU results of 88.5, 75.5, 84.4, and 86.8% on ORIGA-light, DRIVE, DIARET-DB1, and MESSIDOR databases, respectively. Moreover, the average fovea localization results in terms of IoU were 58.1, 66.1, 71, and 73.2% on ORIGA-light, DRIVE, DIARET-DB1, and MESSIDOR databases, respectively. The experimental tests demonstrate that the proposed approach achieves promising results for both normal and diseased images.Öğe A Two-Stage Region Proposal Optimization for Efficient Bird Detection in Deep Learning Pipelines(Wiley, 2025) Dinc, Baris; Kaya, YasinRecent advancements in deep learning have significantly improved object detection performance in computer vision. Nevertheless, region-based convolutional neural networks (R-CNNs) often suffer from computational inefficiencies due to the generation of an excessive number of candidate regions, many of which are redundant or irrelevant. This paper introduces a novel two-stage preprocessing strategy to optimize the region proposal phase in R-CNN-based object detection systems, specifically focusing on bird species recognition. The proposed approach effectively filters out high-frequency noise in non-object regions and reduces the total number of region proposals without compromising detection accuracy. Experimental evaluations conducted on five benchmark bird datasets demonstrate that our method increases the proportion of region proposals with an Intersection over Union (IoU) greater than 0.5 from 80.95% to 86.11%. Furthermore, the number of positive proposals increases by 330% during training and 726% during testing, while the number of redundant proposals is reduced by 55.48%. Moreover, the proposed model reduces the average time required to generate region proposals per image by up to 70%, significantly enhancing computational efficiency. Additionally, to analyze the proposed model in various real-world scenarios, it was evaluated on the publicly available Eastern Cottontail Rabbits dataset, which serves as a challenging benchmark. The proportion of predicted bounding boxes with IoU greater than 0.5 also increased by 8.66%, indicating a notable improvement in localization accuracy. To observe the impact of the proposed method on single-stage object detectors, five bird datasets were unified into a multiclass object detection dataset and evaluated using YOLOv8. The proposed approach improved precision by approximately 18% and recall by 42%. The results validate the effectiveness of the proposed preprocessing framework in improving both the efficiency and accuracy of object detection systems, making it well suited for deployment in parallel and distributed computing environments.Öğe HBDFA: An intelligent nature-inspired computing with high-dimensional data analytics(Springer, 2024) Dinc, Baris; Kaya, YasinThe rapid development of data science has led to the emergence of high-dimensional datasets in machine learning. The curse of dimensionality is a significant problem caused by high-dimensional data with a small sample size. This paper proposes a novel hybrid binary dragonfly algorithm (HBDFA) in which a distance-based similarity evaluation algorithm is embedded before the dragonfly algorithm (DA) searching behavior to select the most discriminating features. The two-step feature selection mechanism of HBDFA enables the method to explore the feature space reduced by the distance-based similarity evaluation algorithm. The model was evaluated on two datasets. The first dataset contained 200 reports from 4 evenly distributed categories of Daily Mail Online: COVID-19, economy, science, and sports. The second dataset was the publicly available Spam dataset. The proposed model is compared with binary versions of four popular metaheuristic algorithms. The model achieved an accuracy rate of 96.75% by reducing 66.5% of the top 100 features determined on the first dataset. Results on the Spam dataset reveal that HBDFA gives the best classification results with over 95% accuracy. The experimental results show the superiority of HBDFA in searching high-dimensional data, improving classification results, and reducing the number of selected features.Öğe Improved object detection approach employing adaptive padding and Viola-Jones algorithm(PeerJ Inc, 2025) Dinc, Baris; Kaya, YasinIn recent years, significant advancements have been achieved in the field of object detection (OD) thanks to the rapid developments in deep learning (DL) methodologies. Although single-stage detectors have accelerated OD, the presence of high-frequency noise and irrelevant details makes these models sensitive to background noise. This challenge causes various difficulties, such as increased false-positive rates in the training process and computational costs. To address these limitations, we present a novel approach that improves the object detector's detection capability by reducing high-frequency noise in the input image. Specifically, the proposed model comprises a novel adaptive padding (AP) mechanism and a region of interest (RoI) detector to provide a balance between RoI generation and OD. To improve the ability to detect multi-scale objects within complex scenes, the model is separately trained with a RoI detector that combines a set of weak candidate regions into a single frame and then enhances its flexibility using the AP mechanism. To create a distinct contrast between the object regions and the background, a low-pass filter is finally applied to the input image. The proposed model was used for You Only Look Once version 8 (YOLOv8) and tested with 351 images of six bird species. It is essential to note that the proposed model outperformed the baseline YOLOv8, achieving superior detection performance in terms of accuracy, classification, and localization loss, while minimizing the high-frequency noise generated by redundant sub-images. The model increased the mean average precision calculated at an intersection over union (IoU) threshold of 0.50 (mAP50) from 0.954 to 0.984 and the mAP50-95 from 0.633 to 0.781. The proposed approach was also evaluated under domain shift conditions and indicated strong robustness across different data distributions. Moreover, the region proposal-based detection results showed more balanced and accurate classification results, with a reduction in the misclassified sample rate. Finally, validation with the state-of-the-art YOLOv12 further demonstrated superior performance, with gains exceeding 3% in mAP50 and 20% in mAP50-95, underscoring the scalability and robustness of the proposed approach.
