Finding the Flies in the Sky Using Advanced Deep Learning Algorithms


Tokat M., Bedir S., REŞİT C.

2nd International Conference on Forthcoming Networks and Sustainability in the AIoT Era, FoNeS-AIoT 2024, İstanbul, Türkiye, 27 - 29 Ocak 2024, cilt.1035 LNNS, ss.333-344, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Cilt numarası: 1035 LNNS
  • Doi Numarası: 10.1007/978-3-031-62871-9_26
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Sayfa Sayıları: ss.333-344
  • Anahtar Kelimeler: convolutional neural network, Drone detection, Resnet50, VGG16, YOLOv8
  • Yıldız Teknik Üniversitesi Adresli: Evet

Özet

Image processing is a concept that allows us to perform operations on images by extracting meaningful features from them. Deep learning algorithms are frequently used in image processing. Although there are many algorithms for image classification, in this study, Convolutional Neural Network based deep learning algorithms VGG16, ResNet50 and YOLOv8 (You Only Look Once) are used. After traditional algorithms such as Support Vector Machine, Linear Regression, KNN or Decision Tree, deep learning algorithms have been developed that perform very well in image processing. As we can see from this study, we can say that the YOLO algorithm has recently gone one step further. In the dataset used, there are a total of 1359 images with and without drone images. The images were pretrained on imagenet and used. When we compare these fine-tuned algorithms, the accuracy of VGG16 was %92.74, ResNet50 was %91.06 and YOLOv8 was %95.4. It was determined that YOLOv8 performed better according to the dataset used.