Mamba-YOLO-SRC: An Automatic Deep Learning Framework for Respiratory Behavior Detection in the Chinese Giant...

Area of Science:

• Amphibian biology
• Conservation science
• Bio-inspired computing

Background:

• Chinese giant salamanders (Andrias davidianus) exhibit abnormal respiratory behaviors indicating health decline.
• Traditional monitoring is difficult due to their nocturnal, cave-dwelling habits and limitations in manual observation.
• Accurate assessment of pulmonary respiration is vital for captive breeding and conservation of this endangered species.

Purpose of the Study:

• To develop the first automated method for monitoring respiratory behaviors in Chinese giant salamanders.
• To address the challenges posed by the species' elusive nature and limitations of conventional detection methods.
• To provide a reliable tool for advancing research on salamander health and behavior.

Main Methods:

• Proposed Mamba-YOLO-SRC, a hybrid detection framework combining Mamba and YOLO architectures.
• Developed a system to accurately identify four key respiratory behaviors: diving (Dive), head-raising (HeadUP), inhalation (Inhale), and exhalation (Exhale).
• Validated the model's performance using established metrics.

Main Results:

• Achieved a high mean average precision (mAP@0.5) of 0.944.
• Demonstrated strong per-class performance: Dive (0.975), HeadUP (0.925), Exhale (0.948), and Inhale (0.928).
• The Mamba-YOLO-SRC framework proved effective in identifying specific salamander respiratory actions.

Conclusions:

• Mamba-YOLO-SRC offers a feasible and referable technical solution for automated respiratory behavior monitoring.
• This method significantly advances research capabilities for Chinese giant salamanders in both captive and natural environments.
• The system supports improved health assessment and conservation strategies for Andrias davidianus.