Comparison of Biocode Based Machine Learning and Segmentation Model for Automated Prawn Size Prediction for Real Prawn Farm

The aquaculture industry has witnessed a significant surge in technological integration over the past decade, leveraging advancements in the Internet of Things (IoT) and Artificial Intelligence (AI) to enhance production efficiency. Monitoring the growth of aquatic cultures, such as fishes and prawns, plays a pivotal role in optimizing production. However, traditional methods, notably in prawn aquaculture in Asia, still heavily rely on stress-inducing techniques like netting and manual measurements. This paper proposes a novel approach using YOLO-based detection algorithm with biocode reference point based on the ArUco design to predict prawn size based on the length and are dimension to analysed the improvement in predicting the size. The propose solution is compared extensively using statistical tests against segmentation model with and without fine-tuning. Initial results show that both YOLOv8 segmentation and YOLOv8 with Segment Anything Model (SAM) give similar results with the actual mean length of 15.7 cm and predicted mean lengths are 20.5 cm and 20.8 cm respectively. However, after fine-tuning the parameter, the accuracy of predicted results improved significantly with 15.8 cm for YOLOv8 segmentation and 16 cm for YOLOv8 with SAM. The mean absolute error of 0.93 cm for YOLOv8 segmentation and 1.10 cm for YOLOv8 with SAM, and the width of 0.48 cm and 0.78 cm respectively.