Yazar "Turkdamar, Mehmet Ugur" seçeneğine göre listele

Listeleniyor 1 - 2 / 2
  • Küçük Resim Yok
    Öğe
    DSHFS: a new hybrid approach that detects structures with their spatial location from large volume satellite images using CNN, GeoServer and TileCache
    (Springer London Ltd, 2024) Tasyurek, Murat; Turkdamar, Mehmet Ugur; Ozturk, Celal
    Satellite images, widely used in recent years and entered the field of remote sensing technology, continuously record image data in databases worldwide. These image data, which are continually obtained, changing, and have a large volume, fall into the big data category. On the other hand, CNN-based techniques, a sub-branch of artificial intelligence, have been widely used to classify and segment image data in recent years. The scope of this study, firstly, it was tried to determine spatial positions and building objects from large-volume satellite images with classical CNN methods. However, classical CNN models could not process the data even in ECW format, one of the most compressed satellite imagery forms. To overcome these problems, a new approach called DSHFS has been proposed. It uses hybrid CNN, GeoServer, and TileCache techniques related to different disciplines to detect structures from large-volume satellite images and their spatial locations. In the proposed DSHFS approach, large-volume satellite imagery is first published in WMS format with GeoServer, which has an open-source strategy. Then, the data are converted into small images containing coordinate information with the TileCache system in size 256 x 256. Finally, the building objects in these images are detected by CNN models. In the proposed DSHFS approach, the actual locations of the detected structures on the earth are calculated using the location information of the image presented as input to the CNN model. In order to examine the performance of the proposed DSHFS approach, satellite imagery covering 12.5 GB in the computer system in ECW format, which corresponds to an area of approximately 17.200 km(2) of Kayseri province, was used. In the proposed DSHFS approach, Faster R-CNN, MobileNet, and YOLO models are used as the CNN model. When the proposed DSHFS approach is examined according to the F1 score, DSHFS Faster R-CNN, MobileNet, and YOLO obtain F1 scores of 0.961, 0.964, and 0.910, respectively. When evaluating the computational efficiency of the proposed approaches, it was found that DSHFS Faster R-CNN, MobileNet, and YOLO took 512.72, 188.20, and 99.35 s, respectively, to identify the structures and their locations in the image. DSHFS YOLO approach detected approximately 2 times faster than MobileNet and approximately 5 times faster than Faster R-CNN. When the proposed DSHFS approach is generally examined, it detects the building objects from the satellite image and their actual positions on the earth in approximately 0.13 s.
  • Küçük Resim Yok
    Öğe
    Transfer learning and fine-tuned transfer learning methods' effectiveness analyse in the CNN-based deep learning models
    (Wiley, 2023) Ozturk, Celal; Tasyurek, Murat; Turkdamar, Mehmet Ugur
    Object detection is a type of application that includes computer vision and image processing technologies, which deal with detecting, tracking, and classifying desired objects in images. Computer vision is a field of artificial intelligence that enables computers and systems to derive information from digital images and take action or suggestions based on that information. CNN is one of the current methods of object detection due to its ease of use and GPU-supported parallel working features. Due to the aim of completing deep learning model training quickly or due to insufficient dataset, many studies using the transfer learning method are carried out in fields such as medicine, agriculture, and weapons. However, there are very few studies that use the fine-tuning method and compare transfer learning in terms of effectiveness. By paying attention to the balanced distribution of the data, approximately 100 images of each chess piece type were included in the analysis and a dataset of at least 1000 images was created. The without transfer learning fine-tune, fine-tuned transfer learning, transfer learning, fully supervised learning (FSL) and weakly supervised learning (WSL) applied models performances compared. Experimental results show that the fine-tuned transfer learning applied YOLO V4 model produces more accurate results than the other models in FSL and the transfer learning applied Faster R-CNN model produces more accurate results than the other models in WSL.