Robust Crayfish Size Estimation Under Complex Poses: A Hybrid YOLO-LSTM Framework With Eye Distance Regression

Accurate crayfish (Procambarus clarkii) size estimation under complex poses (e.g., curled bodies or occlusions) remains challenging for aquaculture automation. This study presents a Robust Crayfish Size Estimation (RCSE) framework that innovatively combines YOLO-based detection with LSTM regression, which introduces eye distance as a novel pose-invariant biometric feature. The two-stage system employs a YOLOv11-checkpoint for primary eye detection (98.2% precision), with automatic activation of the CBAM-enhanced YOLOv11-CBAM-checkpoint (98.99% mAP50) upon detecting eye-count anomalies, positional deviations, or bounding box conflicts-critical for handling complex pose scenarios. The YOLOv11n-seg model achieves 99.50% mAP50 in body segmentation, while our LSTM network fuses eye distance with supplementary features (pincer length, head dimensions) to attain R² = 0.897 in size estimation-a 23.6% improvement over traditional methods. This work establishes a new vision-based paradigm for crustacean biometrics, demonstrating exceptional robustness to morphological variations and directly applicable to automated grading systems in precision aquaculture.