Research on Chengdu Ma Goat Recognition Based on Computer Vison.

Area of Science:

• Agricultural Science
• Computer Vision
• Animal Genetics

Background:

• The Chengdu ma goat is a vital local breed in China, crucial for genetic resource protection.
• Current breeding and protection methods are rudimentary, lacking technological support and hindering effective conservation.
• Small-scale, primitive farming practices limit the potential for large-scale breeding and protection of Chengdu ma goats.

Purpose of the Study:

• To develop an automated individual recognition method for Chengdu ma goats.
• To address the limitations of current breeding and protection strategies through intelligent means.
• To lay the foundation for precision feeding and management based on individual goat identification.

Main Methods:

• Construction of a novel Chengdu ma goat dataset for object detection and classification.
• Implementation of an improved TPH-YOLOv5 based detection algorithm for accurate goat localization in dense, variable-scale environments.
• Integration of a self-supervised learning module into a classifier to enhance performance without additional labeled data or computational load.

Main Results:

• The developed method accurately recognizes individual Chengdu ma goats in real-world indoor barn settings.
• The TPH-YOLOv5 based algorithm effectively handles high-density scenes and significant variations in goat target scale.
• The self-supervised learning module improved classification accuracy without increasing computational costs.

Conclusions:

• The proposed automatic individual recognition system is effective for Chengdu ma goats in indoor breeding environments.
• This technology offers a labor-saving and scalable solution for livestock management and conservation.
• The system enables precision feeding strategies by facilitating identification based on sex and age, contributing to improved animal husbandry.