Machine learning for extracting morphological phenotypic traits and estimating weight in largemouth bass

The acquisition of phenotypic data is fundamental for the genetic breeding of aquatic animals. To enhance the efficiency of obtaining phenotypic data for largemouth bass (LMB), we developed a phenotype extraction that employed the key points detection Real-Time Multi-Person Pose Estimation (RTMPose) model for rapid, efficient, and accurate measurement of morphological phenotypic data from captured images. 717 images were respectively acquired from the side view and top view and constructed the corresponding datasets. After training and validating the image dataset using the RTMPose model, 15 morphological traits data (total body length, body length, body height, jaw length, head length, head height, trunk length, pectoral fin length, dorsal fin length, anal fin length, caudal fin length, caudal peduncle length, caudal peduncle height, distance between eyes, and body thickness) were measured. All traits achieved the MAPE of less than 3%, and most traits exhibited R 2 values exceeding 94%. Subsequently, multiple machine learning models were constructed to predict body weight (BW) and carcass weight (CW) of LMB. Among them, multiple linear regression, ridge regression, lasso regression, and support vector regression yielded similarly high predictive performance, For BW, these four models achieved R 2 values of approximately 0.943, with MAE ranging from 8.167 to 8.338 g, RMSE between 11.903 and 12.036 g, and MAPE from 1.957% to 1.979%. For CW, the R 2 was around 0.880, with MAE values between 8.597 and 8.681 g, RMSE ranging from 11.388 to 11.568 g, and MAPE between 3.291% and 3.328%. And SHAP analysis identified body height as the key morphological trait influencing both BW and CW in the sampled population. Our findings demonstrate that integrating keypoint detection with machine learning enables efficient and accurate phenotyping and weight prediction in fish, highlighting its practical value for intelligent aquaculture and breeding applications.