Yazar "Turan, Salih Can" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim
    Yayın
    A deep learning approach to document recovery: high performance with denoiseU-net
    (Afyon Kocatepe University, 2025) Turan, Salih Can; Çıplak, Zeki; Sarıkaş, Ali; Yıldız, Kazım
    Image denoising, a crucial task in image processing, has consistently faced challenges despite ongoing research efforts. In this study, a dataset was created by extracting 20,000 images from 60 public sources, including scanned or digitized documents. Each image was verified to contain at least one of the following: plain text, image, table, or mathematical expression. Common types of noise, including random black and white pixels, Gaussian blur, gray areas, speckle noise, random directional lines, Poisson noise, and salt-and-pepper noise, were applied to the images. To create the test set, each of the seven types of noise was individually added to 500 images excluded from the dataset, resulting in a balanced test set of 3,500 images. The complete dataset consists of 23,000 images, with a training-to-test ratio of 5:1. Specifically, our proposed DenoiseU-Net model aims to recover noisy scanned documents and performs effectively across various content types, such as tables, images, mathematical equations, and text. Experimental results show that the average precision, recall, and F1-score of DenoiseU-Net on the test set are 99.36%, 99.59%, and 99.48%, respectively. In addition to these evaluation results, the average SSIM and PSNR values, which are commonly used parameters to assess image quality, were obtained as 0.9657 and 40.28 dB, respectively. The primary objective of this study is not to demonstrate superior performance over state-of-the-art (SOTA) methods, but rather to evaluate how deep learning models, such as the proposed DenoiseU-Net, perform on medium-scale or small-scale datasets in practical scenarios.
  • Küçük Resim
    Yayın
    Recognition of leaf diseases in hazelnut orchards with drone imagery by deep learning models: effectiveness of U-Net model variants
    (Institute of Electrical and Electronics Engineers Inc., 2025) Boyar, Tülin; Turan, Salih Can; Çıplak, Zeki; Yıldız, Seyit Gazi; Sarıkaş, Ali; Yıldız, Kazım; Demir, Önder
    This paper presents a novel semantic segmentation approach for the binary classification of diseased and healthy regions in hazelnut orchards, addressing a significant gap in existing research by focusing on tree-level analysis. Utilizing video imagery captured by unmanned aerial vehicles (UAVs), this study investigates the challenges associated with drone-based image acquisition, such as environmental conditions, physical obstructions, battery life, and image stability. A unique image dataset specifically designed for tree-based disease detection is created. Various deep learning models, including seven U-Net variants, are employed to tackle the complexities of image processing and segmentation. The U-Net++ model is identified as the most effective for this task, demonstrating strong performance in distinguishing affected areas. However, further optimization of hyperparameters and expansion of the dataset are recommended to enhance model robustness and generalization. This research underscores the critical need for additional studies on UAV-based tree-level disease detection and highlights the substantial potential of deep learning in agricultural technology for mitigating production losses through early and accurate identification of stressed regions.