Evaluation of Eating Posture Detection Artificial Intelligence Models Trained on Different Supervised Datasets and...

Area of Science:

• Agricultural Science
• Artificial Intelligence
• Animal Behavior

Background:

• Accurate monitoring of broiler feeding behavior is crucial for optimizing production efficiency and animal welfare.
• Object detection models using Artificial Intelligence (AI) offer a potential solution for automated behavior analysis.
• The specific content of supervised datasets significantly influences the performance of AI models.

Purpose of the Study:

• To investigate how supervised dataset composition affects the accuracy of an AI model for detecting eating postures (EP) in broilers.
• To evaluate the effectiveness of including similar postures (SEP) and varying numbers of individuals in the dataset.
• To develop and validate a prototype tool for real-time broiler feed intake estimation.

Main Methods:

• Trained an object detection AI model using datasets with varying content related to eating postures (EP) and similar postures (SEP).
• Analyzed the impact of the number of individuals in EP and SEP per image on detection accuracy.
• Developed a prototype tool for real-time feed intake estimation and conducted validation trials.

Main Results:

• Increased individuals in EP did not consistently improve AI model accuracy; varied individual counts in still images enhanced detection.
• Including both EP and SEP, especially in the same image, significantly improved detection accuracy.
• Balanced proportions of images with different numbers of individuals in EP and SEP were necessary for optimal performance.
• The prototype tool achieved a mean absolute error (MAE) of 4.3g and a mean absolute percentage error (MAPE) of 13.3% for feed intake estimation.

Conclusions:

• Dataset composition, including the inclusion of similar postures and balanced individual representation, is critical for accurate AI-based broiler behavior detection.
• The developed tool demonstrates high accuracy for real-time feed intake estimation in broilers.
• These findings contribute to advancing precision livestock farming through AI-driven behavioral monitoring.