Non-invasive weight estimation of cultured abalone (Haliotis discus hannai) using image-based calculated volume and morphometric correction

Weight information is essential for grading, growth monitoring, biomass assessment, and harvest decision-making in abalone aquaculture. Image-based morphometric measurement reduces repeated weighing; however, geometric volume models based on external dimensions may retain residual errors due to individual shape variation. This study developed and evaluated interpretable regression models for estimating the weight of cultured abalone ( Haliotis discus hannai ) from image-based morphometric measurements, with a particular focus on the effect of a morphometric correction variable. After quality control based on measured density and morphometric consistency, 490 of 503 individuals were included in the final analysis. Semi-ellipsoid-based calculated volume ( C v ) and a morphometric correction variable ( M c = C v / PA , where PA is projected area) were calculated from total length, body width, thickness, and PA . Four regression models were evaluated using the same five-fold cross-validation scheme. Predictive performance was assessed using root mean square error (RMSE), mean absolute error (MAE), coefficient of determination ( R 2 ), and mean absolute percentage error (MAPE). The C v -based linear model showed RMSE 4.18 g, MAE 3.20 g, R 2 0.969, and MAPE 6.32%. The morphometric correction power model combining C v and M c produced the lowest prediction error, with RMSE 3.50 g, MAE 2.62 g, R 2 0.978, and MAPE 5.17%, reducing RMSE by 16.30% relative to the C v -based model. These results indicate that C v is an effective baseline variable and M c can reduce residual errors associated with individual shape variation. The model may provide useful weight information for non-invasive grading support, growth monitoring, and harvest decision-making under consistent image-acquisition conditions.