Recognition of the gonad of Pacific oysters via object detection

Abstract

Oyster is the largest cultured shellfish in the world, and it has high economic value. The plumpness of the Pacific oyster gonad has important implications for the quality and breeding of subsequent parents. At present, only the conventional method of breaking their shells allows for the observation and study of the interior tissues of Pacific oysters. It is an important task to use computer technology for non-destructive sex detection of oysters and to select mature and full oysters for breeding. In this study, based on the multi-effect feature fusion network R-SINet algorithm, a CF-Net algorithm was designed through a boundary enhancement algorithm to detect inconspicuous objects that appear to be seamlessly embedded in the surrounding environment in nuclear magnetic resonance (NMR) images, effectively solving the problem of difficulty in distinguishing Pacific oyster gonads from background images. In addition, calculations were performed on the segmented gonadal regions to obtain a grayscale value difference map between male and female oysters. It was found that there were significant differences in grayscale values between females and males. This task allows for non-destructive detection of the gender of oysters. Firstly, a small animal magnetic resonance imaging (MRI) system was used to perform MRI on Pacific oysters, and a dataset of oyster gonads was established. Secondly, a gonadal segmentation model was created, and the Compact Pyramid Refinement Module and Switchable Excitation Model were applied to the R-SINet algorithm model to achieve multi-effect feature fusion. Then, the Convformer encoder, Token Reinforcement Module, and Adjacent Transfer Module were used together to form the CF-Net network algorithm, further improving the segmentation accuracy. The experimental results on the oyster gonad dataset have demonstrated the effectiveness of this method. Based on the segmentation results, it is possible to calculate the grayscale values of the gonadal region and obtain the distribution map of the grayscale value difference between male and female oysters. The results can provide a technical methodology for the non-destructive discrimination of oyster gender and later reproduction.
Keywords: pacific oyster gonad, unapparent object detection, target segmentation, deep learning, MRI, R-SINet
DOI: 10.25165/j.ijabe.20241706.8478

Citation: Chen Y F, Yue J, Wang W J, Yang J M, Li Z B. Recognition of the gonad of Pacific oysters via object detection. Int J Agric & Biol Eng, 2024; 17(6): 230–237.