Shrimp phenotypic data extraction and growth abnormality identification method based on instance segmentation

Abstract

Shrimp phenotypic characteristics encompass externally or partially internally observable traits like body length, width, color, and morphology. These traits are vital for assessing shrimp health, growth, and environmental adaptability. The traditional manual sampling method is labor-intensive and inadequately reflects the overall growth status of cultured shrimp. Therefore, this paper proposes a shrimp phenotypic data extraction and growth abnormality identification method based on instance segmentation. Firstly, overhead images of shrimp are collected to form a dataset. Secondly, an improved YOLOv8 algorithm is utilized for instance segmentation of the shrimp. Then, shrimp phenotypic data is extracted through image processing techniques and reference object transformation. Finally, the Kolmogorov-Smirnov and Shapiro-Wilk tests are used to assess the growth and development status of the shrimp groups, while individual growth abnormalities are identified using the Z-value algorithm and the Interquartile Range method. Additionally, this paper optimizes shrimp farming processes such as grading, selective breeding, and feeding based on phenotypic data. On the self-built dataset, the proposed method achieves an average precision of instance segmentation of 99.5 % and an F1 score of 99.3 %. Comparing the shrimp lengths calculated by our model with manual measurements, the absolute error in shrimp length fitting is 4.63 mm, with a relative error of 4.58 %. The accuracy of detecting abnormal individuals in shrimp is 91 %. The results indicate that the proposed method can effectively evaluate shrimp growth status and identify abnormal growth patterns.