Development of a real-time cattle lameness detection system using a single side-view camera.

Area of Science:

• Computer Vision
• Machine Learning
• Animal Science

Background:

• Dairy farming faces challenges in early detection of cattle lameness.
• Traditional methods for lameness detection are labor-intensive and subjective.
• Advancements in AI offer potential for automated livestock health monitoring.

Purpose of the Study:

• To investigate the application of deep learning for automated cattle lameness detection.
• To compare the performance of Mask-RCNN (Detectron2) and YOLOv8 for cattle detection and tracking.
• To evaluate various machine learning algorithms for lameness classification based on extracted features.

Main Methods:

• Utilized Mask-RCNN (Detectron2) and YOLOv8 for object detection and cattle tracking.
• Employed image processing for feature extraction from cattle mask regions.
• Applied machine learning algorithms including AdaBoost, SVM, Decision Trees, and Random Forests for lameness classification.

Main Results:

• Detectron2 achieved 98.98% accuracy in cattle detection, and the tracking module reached 99.50% accuracy.
• AdaBoost demonstrated the highest lameness classification accuracy at 77.9%.
• Other algorithms like Decision Trees (75.32%), SVM (75.20%), and Random Forest (74.9%) also showed effective performance.

Conclusions:

• The proposed deep learning system successfully enables automated cattle lameness detection.
• This technology has the potential to significantly improve dairy farm management and animal welfare.
• The study highlights the value of computer vision in livestock health monitoring.