Analyzing Data Modalities for Cattle Weight Estimation Using Deep Learning Models.

Area of Science:

• Agricultural Science
• Computer Vision
• Machine Learning

Background:

• Accurate cattle weight estimation is crucial for effective livestock management.
• Traditional methods are labor-intensive and may lack precision.
• Advancements in AI and sensor technology offer new possibilities for automated estimation.

Purpose of the Study:

• To investigate the impact of various data modalities on cattle weight estimation accuracy.
• To evaluate the performance of deep learning models with integrated data sources.
• To identify optimal data combinations for robust weight prediction.

Main Methods:

• Collected a novel cattle dataset with RGB, depth, segmentation, and combined modalities.
• Utilized a vision-transformer-based zero-shot model for segmentation and feature extraction.
• Applied three baseline deep learning models for comparative analysis.
• Assessed performance using Mean Absolute Error (MAE), Root Mean Squared Error (RMSE), Mean Absolute Percentage Error (MAPE), and R-squared (R2).

Main Results:

• Different data modalities show varying impacts on weight estimation accuracy.
• Combined modalities generally yield more robust and precise predictions compared to single modalities.
• Deep learning models benefit significantly from the integration of diverse data sources.

Conclusions:

• The study provides insights into the effectiveness of different data modalities for cattle weight estimation.
• Findings support the use of combined data sources to enhance precision in livestock management systems.
• This research facilitates informed decision-making for optimizing automated cattle monitoring.