Shrimp Larvae Counting Based on Improved YOLOv5 Model with Regional Segmentation

Area of Science:

• Aquaculture Technology
• Computer Vision
• Machine Learning

Background:

• Accurate shrimp larvae counting is vital for shrimp farming success.
• Traditional counting methods are challenged by the small size and high density of larvae.
• Existing automated methods often struggle with dense populations.

Purpose of the Study:

• To develop an advanced algorithm for accurately counting densely packed shrimp larvae.
• To improve upon existing You Only Look Once (YOLOv5) models for small object detection.
• To enhance the efficiency and accuracy of shrimp larvae enumeration in aquaculture.

Main Methods:

• An enhanced You Only Look Once version 5 (YOLOv5) model was developed.
• Incorporated C2f and convolutional block attention modules to improve small shrimp recognition.
• Utilized a regional segmentation approach with stitching and deduplication to prevent double counting.

Main Results:

• The proposed algorithm demonstrated superior performance compared to other shrimp counting techniques.
• Achieved an accuracy exceeding 98% for counting high-density shrimp larvae in large quantities.
• The regional segmentation and deduplication strategy effectively addressed challenges of overlapping detections.

Conclusions:

• The enhanced YOLOv5 algorithm offers a highly accurate and efficient solution for shrimp larvae counting.
• This method significantly advances automated counting capabilities in aquaculture.
• The developed technique is robust for managing large-scale, high-density shrimp populations.