Comparison of Sub-Models and Examination of Factors Enhancing Performance in a Deep Learning Model Adaptable to Embedded Systems G m l Sistemlere Uyumlanabilen Derin grenme Modelinin Altmodel Karsilastirmasi ve Performansi Artiran Etkenlerin Incelenmesi

Yilmaz C., Celebi A. T.

33rd IEEE Conference on Signal Processing and Communications Applications, SIU 2025, İstanbul, Türkiye, 25 - 28 Haziran 2025, (Tam Metin Bildiri) identifier

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/siu66497.2025.11112168
  • Basıldığı Şehir: İstanbul
  • Basıldığı Ülke: Türkiye
  • Anahtar Kelimeler: cycar, deep learning, object detection, yolo
  • Kocaeli Üniversitesi Adresli: Evet

Özet

In this study, the performance of the YOLOv7 model, which can be implemented on embedded systems and FPGAs, was examined for detecting desired targets using civilian and military unmanned aerial vehicles. The performance of its sub-models, YOLOv7-X and YOLOv7-Tiny, was evaluated at different altitudes. The study discusses possible improvements to enable the YOLOv7-Tiny model, which achieved the highest performance, to detect targets from altitudes of up to 500 meters. Initially, the Cycar dataset was used for training To enhance detection at higher altitudes, dataset augmentation was applied, and the DOTA dataset was also utilized. It was observed that the numerical increase in dataset samples significantly contributed to the results