Multi-Feature Fusion and Optimization for Micropterus salmoides Tracking and Body Length Monitoring in Complex...

Area of Science:

• Aquaculture technology
• Computer vision
• Animal biometrics

Background:

• Traditional manual fish measurement causes stress and errors.
• Continuous, non-contact monitoring is needed for intensive aquaculture systems.
• Accurate body length data is crucial for growth monitoring and management.

Purpose of the Study:

• To develop an improved YOLOv8-based multi-target tracking framework for non-contact body length measurement of Micropterus salmoides.
• To enhance system stability and accuracy in challenging aquaculture environments (high density, variable lighting, turbidity).
• To overcome limitations of traditional manual measurement methods.

Main Methods:

• Utilized a monocular vision geometric measurement framework with camera calibration and refraction correction.
• Integrated multi-scale feature enhancement, lightweight re-identification (ReID), and a hybrid data association mechanism (IoU, cosine similarity, motion consistency).
• Employed the Hungarian algorithm for optimal matching and dynamic threshold adjustment with exponential moving-average for identity consistency.

Main Results:

• Achieved 82.7% mAP50 at 88 FPS, with MOTA of 76.9% and IDF1 of 81.5% (3.2% and 5.3% higher than baselines).
• Reduced identity switches (IDSW) by 56.2% (from 89 to 39).
• Attained a body length error (BLE) of 5.2 ± 1.8% (MAE = 1.35 cm), a 38.8% improvement over the baseline, meeting aquaculture tolerance requirements.

Conclusions:

• The improved YOLOv8 framework significantly enhances non-contact fish body length measurement accuracy and stability in recirculating aquaculture systems.
• The integrated detection and tracking optimizations are essential for achieving application-level accuracy and reducing measurement errors.
• This technology offers a promising solution for precise, stress-free monitoring of fish growth in intensive aquaculture.