Computer vision-based growth prediction and digestive tract assessment in pacific white shrimp (Litopenaeus vannamei)

Abstract

Litopenaeus vannamei is a key economic species in global aquaculture, and monitoring its growth is critical for optimizing feed management, and reducing costs. Traditional manual sampling methods are labor-intensive, error-prone, and can cause stress or injury to shrimp, negatively impacting growth. This study employed computer vision and machine learning to propose an innovative approach for growth prediction and digestive tract assessment in L. vannamei. An improved annotation method was developed to simplify the marking process and reduce weight prediction errors. Furthermore, A new length measurement approach, termed “visual total length”, was introduced to overcome the limitations of traditional measurement techniques. In this study, images of shrimp on feeding trays were analyzed to simulate real aquaculture monitoring conditions, and a prediction system was constructed by combining image segmentation (You Only Look Once v8n-seg, YOLOv8n-seg), classification (YOLOv8n-cls), traditional fitting, and machine learning models (Light Gradient Boosting Machine, LightGBM). The results showed that the improved annotation method also significantly reduced weight prediction errors caused by tail fan area. The visual total length had a high linear correlation with traditional total length (r² = 0.99), effectively allowing it to replace traditional measurements and enhancing applicability in real production environments. The final model achieved an accuracy of over 97 % in predicting length and weight when compared to manual measurements, and over 87 % accuracy in assessing digestive tract fullness. This study provides an efficient and precise method for growth monitoring, laying a solid foundation in future intelligent shrimp aquaculture.