Computation and analysis of phenotypic parameters of Scylla paramamosain based on YOLOv11-DYPF keypoint detection

Abstract

Phenotypic parameter measurement of mud crabs is a critical component in aquaculture and scientific research, serving as a key basis for assessing their growth status and quality. To enhance the efficiency of measuring growth parameters such as body dimensions and weight, this study employed the YOLOv11 model for carapace keypoint detection in mud crabs and automated calculation of body length and weight based on detection results. To improve model accuracy, we proposed two modifications to the standard YOLOv11 architecture: a self-developed multi-scale fusion module (SPPF-MP2) replacing the original SPPF module and a lightweight dynamic upsampling module (Dysample) substituting the conventional Upsample module. Results demonstrated that both modules individually contributed to performance enhancement, with a significant 1.8 % improvement in mAP@50–95 and a 0.1 % increase in mAP@50 when combined. During inference, pixel distances between the 12 annotated keypoints were leveraged to compute seven critical body dimensions, with relative errors below 10 % for all measurements. Furthermore, we established seven predictive models to correlate crab dimensions with weight. Among these, the Support Vector Regression (SVR) model exhibited the highest accuracy, achieving a maximum prediction error of 13.74 % and an average error of 4.90 %. This study confirms that YOLO-based visual models for keypoint detection enable high-precision estimation of body dimensions and weight in mud crabs, significantly streamlining the statistical analysis of growth parameters in aquaculture and ecological studies.