CattleNet-XAI: An explainable CNN framework for efficient cattle weight estimation.

Area of Science:

• Agricultural Science
• Computer Science
• Machine Learning

Background:

• Manual cattle weight estimation is inaccurate and labor-intensive.
• Traditional regression models struggle with complex image data for weight prediction.
• Automated methods are needed for efficient and precise livestock management.

Purpose of the Study:

• To develop an efficient and explainable framework (CattleNet-XAI) for automated cattle weight estimation.
• To compare the performance of a custom Convolutional Neural Network (CNN) against other models.
• To enhance the accuracy of weight prediction using advanced image processing and deep learning.

Main Methods:

• Developed CattleNet-XAI, a custom CNN framework with advanced image preprocessing.
• Utilized YOLOv5 for feature extraction in traditional machine learning models.
• Trained and evaluated multiple models including CNNs, EfficientNetB3, Random Forest, and Linear Regression.
• Measured performance using Mean Absolute Error (MAE), Mean Squared Error (MSE), and Root Mean Squared Error (RMSE).

Main Results:

• The custom CNN model (3Conv3Dense variation) achieved superior accuracy.
• Achieved a Mean Absolute Error (MAE) of 18.02 kg and Root Mean Squared Error (RMSE) of 19.85 kg.
• Demonstrated significant improvement over traditional machine learning and other CNN models.

Conclusions:

• Deep learning, particularly CNNs, offers a highly accurate and automated solution for livestock weight estimation.
• CattleNet-XAI provides an effective and explainable approach to modern cattle management.
• Automated weight estimation enhances farm management, health assessment, and productivity optimization.