Individual dairy cow identification based on lightweight convolutional neural network.

Area of Science:

• Computer Vision
• Artificial Intelligence
• Animal Science

Background:

• Traditional livestock identification methods are slow and require large models, limiting practical farm application.
• Accurate individual cow recognition is crucial for farm management and data collection.

Purpose of the Study:

• To develop a lightweight deep learning model for efficient and accurate individual cow identification in complex backgrounds.
• To improve upon existing deep learning architectures for livestock recognition.

Main Methods:

• Utilized Alexnet as a skeleton for a lightweight convolutional neural network, enhanced with multiscale convolutions and a BasicBlock for stability.
• Incorporated an improved inception module and attention mechanism for multi-scale feature extraction.
• Collected side-view images of 13 cows for experimental validation.

Main Results:

• Achieved 97.95% accuracy in individual cow identification.
• Demonstrated a single training time of 6 seconds, significantly faster than original Alexnet.
• The model has the smallest parameter size (6.51 MB) compared to Vgg16, Resnet50, Mobilnet V2, and GoogLenet.

Conclusions:

• The proposed lightweight model overcomes limitations of traditional methods, offering high accuracy, speed, and efficiency.
• The method is suitable for real-world farming environments with complex image backgrounds.
• Provides a valuable reference for individual animal identification using deep learning.