Automatic Penaeus Monodon Larvae Counting via Equal Keypoint Regression with Smartphones.

Area of Science:

• Aquaculture
• Computer Vision
• Machine Learning

Background:

• Large-scale Penaeus monodon farming relies on purchasing larvae, necessitating accurate counts.
• Current methods for counting small, congested shrimp larvae are challenging and labor-intensive.

Purpose of the Study:

• To develop a simple, efficient, and equipment-free method for accurately counting Penaeus monodon larvae.
• To introduce the Penaeus Larvae Counting Strategy (PLCS) for automated larvae counting.

Main Methods:

• Utilized smartphone-captured images for larvae counting, eliminating the need for specialized equipment.
• Developed a keypoint regression approach treating two keypoint types as equip keypoints.
• Created the Penaeus_1k dataset with 1420 high-resolution images and keypoint annotations.

Main Results:

• The PLCS achieved an average accuracy of 93.79% on a test dataset of 720 images across seven density groups.
• The proposed method outperformed classical density map algorithms in larvae counting.
• Demonstrated the practical efficacy of the PLCS in real-world aquaculture scenarios.

Conclusions:

• The Penaeus Larvae Counting Strategy (PLCS) offers a highly accurate and accessible solution for counting Penaeus monodon larvae.
• Smartphone-based image analysis provides a viable alternative to traditional counting methods in aquaculture.
• The developed dataset and method contribute to advancing automated counting techniques in marine hatchery operations.