A PHENOTYPIC MEASUREMENT AND WEIGHT PREDICTION MODEL OF PELTEOBAGRUS FULVIDRACO BASED ON COMPUTER VISION

To improve the efficiency of phenotypic data acquisition of yellow catfish (Pelteobagrus fulvidraco Richardson), we developed a simple phenotype acquisition device, which can quickly, efficiently and accurately measure the phenotypic characteristic parameters of fish in the collected images through the YOLOv8 network. In this study, a total of 1752 image data of yellow catfish were collected. Following the training and validation of the image set using the YOLOv8 network, we completed the measurement of 11 phenotypic characteristics in 584 yellow catfish, including body area, head area, total length, body length, head length, body height, head height, body width, head width, abdominal area, and pectoral fin length. The results showed an average relative error of approximately 1% for length-related phenotypic traits and the average relative error for measuring area-related phenotypic traits was around 3%. Moreover, the time required for measuring all phenotypic traits was within 1s. Furthermore, correlation analysis, path analysis, and regression analysis were performed on the obtained phenotype and body weight data for yellow catfish. The results showed that the head height had the greatest effect on the body weight of yellow catfish, followed by head width, body area, body length, abdominal area, and pectoral fin length. A multiple model was established to fit the weight based on these six traits, and the largest correlation index (0.948) was obtained, indicating that these traits were important traits associated with the body weight of yellow catfish. This study estabolishs a rapid method to measure the phenotypic data and provides a basis for the creation of novel varieties of yellow catfish.