A high-precision measuring phenotypic characteristics method for juvenile fish

• Developed YOLOv11-CCPose by integrating SMAFormer and ContextGuideFPN with YOLOv11. • Proposed a novel multi-segment measurements method to measure fish with different bending shapes • Integrated a juveniles biomass estimation system with high-precision detection. • It achieved average measurement errors of 1.36%, 3.04%, and 4.21% for length, height, and weight. The rapid swimming and frequent posture changes of juvenile fish have brought great difficulties to the accurate measurement phenotypic characteristics (e.g., length, height, and weight) based on machine vision. This study proposed a novel measurement method for phenotypic characteristic parameters of juvenile spotted knifejaw ( Oplegnathus punctatus ), which combined keypoint detection (YOLOv11-CCPose), multi-segment measurement and weight estimation model. Firstly, ContextGuideFPN module was incorporated in YOLOv11-CCPose to enhance the feature extraction capability for juveniles. And the feature extraction module (C3K2) was improved through the integration of SMAFormer and Convolutional Gated Linear Units (CGLU), significantly improving the accuracy of identifying features. Compared with baseline model, YOLOv11-CCpose achieved improvements of 3.2 % in precision, 2.7 % in recall, 1.8% in [email protected], and 1.6 % in [email protected]:0.95. Therefore, it markedly strengthening the multi-scale feature extraction capacity for fast-swimming juveniles with various postures. Secondly, a novel multi-segment measurements method was proposed to reduce the measurement errors of the length and height of juveniles in various bending postures. And a WBP weight estimation model was proposed to estimate weight using dual features of length and height. Finally, a juveniles biomass estimation system with high-precision detection was developed and verified. The maximum measurement errors of the system for length, height, and weight were 3.22%, 4.56%, and 6.00%, respectively, while the average errors were 1.36% for length, 3.04% for height, and 4.21% for weight. It proved that the juveniles biomass estimation system was feasible under real aquaculture conditions and achieves high detection accuracy.