Morphometric Information for Yangtze Finless Porpoises Using Detection-Guided SAM2 Segmentation with UAV Imagery

Morphometric information provides quantitative descriptors of cetacean size and shape and may support future assessments of individual condition and population status when combined with appropriate biological calibration. However, conventional contact-based measurements are difficult to apply to free-ranging Yangtze finless porpoises (Neophocaena asiaeorientalis). UAV imagery offers a non-contact means of acquiring porpoise morphometric data, but automated workflows for converting UAV observations into reliable body-surface measurements remain limited. To address this gap, an oriented detection-guided SAM2 morphometric workflow, termed ODG-SAM2-Morph, was developed to automatically extract body-surface morphometric parameters from UAV imagery. The workflow integrates YOLO26-OBB for oriented target localization, SAM2 for prompt-guided body-surface segmentation, and differentiated morphometric extraction strategies for complete-body and partial-body samples. Guided by oriented detections, the small SAM2 model with the R-Box + 1FG prompt generated body-surface masks with mean IoU, mean Dice, Precision, and Recall values of 0.824, 0.901, 0.883, and 0.932, respectively. For complete-body samples, automatically extracted body-length and body-width parameters showed preliminary agreement with manual measurements. For 114 partial-body samples, the automatically extracted maximum visible body width agreed reasonably well with manual measurements, with a MAE of 3.32 cm and a MAPE of 11.93%. Continuous-sequence analysis further showed that the representativeness of maximum visible body width depended on trunk exposure, contour clarity, body posture, and inter-frame stability. By converting UAV observations into quantitative image-derived morphometric measurements, ODG-SAM2-Morph provides a practical basis for non-contact morphometric monitoring of Yangtze finless porpoises under natural survey conditions.