Real-Time Jellyfish Classification and Detection Based on Improved YOLOv3 Algorithm.

Area of Science:

• Marine Biology
• Computer Vision
• Image Processing

Background:

• Jellyfish outbreaks pose significant threats to marine ecosystems, fisheries, and coastal industries globally.
• Effective monitoring systems are crucial for mitigating the impacts of jellyfish blooms.
• Current optical detection methods for jellyfish are underdeveloped, necessitating advancements in technology.

Purpose of the Study:

• To develop an efficient and accurate jellyfish detection method using digital image processing and deep learning.
• To enhance underwater image quality for improved jellyfish detection.
• To optimize the YOLOv3 algorithm for real-time jellyfish identification.

Main Methods:

• Applied underwater image preprocessing techniques: prior defogging, adaptive histogram equalization, and multi-scale retinal enhancement.
• Created a dataset of 2141 images featuring seven jellyfish species and fish.
• Utilized and optimized the YOLOv3 algorithm with its Darknet53 feature extraction network for real-time detection.
• Incorporated label smoothing and cosine annealing learning rate methods during model training.

Main Results:

• Image preprocessing significantly improved underwater image quality, aiding detection.
• The optimized YOLOv3 algorithm achieved improved jellyfish detection accuracy while maintaining real-time performance.
• The developed system demonstrates a strong foundation for real-time underwater jellyfish optical imaging and monitoring.

Conclusions:

• Advanced image processing and optimized deep learning models are effective for underwater jellyfish detection.
• The study provides a robust framework for developing real-time monitoring systems to manage jellyfish populations.
• This research contributes to mitigating the negative impacts of jellyfish blooms on marine environments and human activities.