An Improved Single Shot Multibox Detector Method Applied in Body Condition Score for Dairy Cows

Area of Science:

• Animal Science
• Computer Vision
• Machine Learning

Background:

• Body Condition Scoring (BCS) is crucial for dairy cow health and productivity.
• Traditional BCS methods are labor-intensive and inefficient for large farms.
• Existing computer vision methods for BCS require further accuracy improvements.

Purpose of the Study:

• To develop a cost-effective and accurate BCS evaluation system for dairy cows.
• To leverage deep learning and machine vision for automated BCS assessment.
• To improve the efficiency and scalability of BCS monitoring in dairy farming.

Main Methods:

• Utilized a dataset of 8972 back-view 2D images of dairy cows captured by network cameras.
• Manually labeled key body parts (tails, pins, rump) for training.
• Developed an improved Single Shot Multi-box Detector (SSD) model inspired by DenseNet and Inception-v4 for tail detection and BCS evaluation.

Main Results:

• Achieved 98.46% classification accuracy and 89.63% location accuracy.
• Demonstrated a fast detection speed of 115 frames per second (fps).
• The improved SSD model has a smaller size (23.1 MB) compared to original SSD and YOLO-v3, reducing hardware costs.

Conclusions:

• The proposed deep learning-based method offers a highly accurate and efficient solution for automated BCS evaluation.
• The system's speed, accuracy, and reduced hardware requirements make it suitable for large-scale dairy farms.
• This approach significantly enhances the monitoring of dairy cow health and metabolic status.