Comparison of Deep Learning Methods for Jellyfish Aurelia Aurita over Surface Detection Utilizing Augmentation

KASAPBAŞI M. C., ŞANVER U.

2026 IEEE Ural-Siberian Conference on Biomedical Engineering, Radioelectronics and Information Technology, USBEREIT 2026, Yekaterinburg, Rusya, 14 - 15 Mayıs 2026, (Tam Metin Bildiri)

  • Yayın Türü: Bildiri / Tam Metin Bildiri
  • Doi Numarası: 10.1109/usbereit70063.2026.11580809
  • Basıldığı Şehir: Yekaterinburg
  • Basıldığı Ülke: Rusya
  • Anahtar Kelimeler: CNN, comparative deep learning analysis, jellyfish detection
  • İstanbul Ticaret Üniversitesi Adresli: Evet

Özet

Jellyfish detections are important not only for assessing environmental conditions, population dynamics, or early warning systems, but also for the industrial and economic aspects of fisheries, as well as human and tourism safety. Another challenge is above-surface imagery due to sunlight reflections on the sea surface. To address this, we curated a new above-surface jellyfish image dataset that includes cases with surface reflections. This paper represents a comparison assessment of pre-trained deep learning models for jellyfish automatic detection, a critical mission in marine observation and ecological protection. Six leading convolutional neural network (CNN) architectures, including DenseNet201, EfficientNet-B0, MobileNetV2, ResNet50, AlexNet, and ShuffleNet, were compared based on classification of many metrics and computational expense. Among these, ResNet50 and DenseNet201 performed best in terms of accuracy, precision, sensitivity (recall), and Dice Similarity Coefficient. To further enhance model performance, Stochastic Gradient Method was also employed as an optimization technique. Beyond the classification metrics, training time was recorded and analyzed for each model to assess computational cost and feasibility of real-world applicability. The results indicate that ResNet50 offers the best trade-off between detection accuracy and computational speed and is the most desirable tool for efficient and reliable jellyfish surveillance systems.