An Improved Method for Broiler Weight Estimation Integrating Multi-Feature with Gradient Boosting Decision Tree.

Area of Science:

• Agricultural Engineering
• Computer Vision
• Animal Science

Background:

• Broiler weighing is crucial for the poultry industry.
• Current camera-based weight estimation methods struggle with older birds, leading to increased errors.
• A need exists for more accurate and robust broiler weight estimation systems.

Purpose of the Study:

• To develop a novel framework for accurate broiler weight estimation using depth images.
• To overcome the limitations of existing methods that focus on younger broilers.
• To improve the accuracy and robustness of weight estimation for broilers of all ages.

Main Methods:

• Instance segmentation of depth images using Mask R-CNN.
• Feature extraction using Customized Resnet50 (C-Resnet50) and artificial features.
• Fusion of extracted features using a feature fusion module.
• Weight estimation employing Gradient Boosting Decision Tree (GBDT).

Main Results:

• The framework effectively segments individual broilers from depth images, even with complex backgrounds.
• Feature fusion and GBDT integration enhance estimation accuracy and robustness.
• Achieved a Mean Absolute Error (MAE) of 0.093 kg and R² of 0.707 on 63-day-old bantam chickens.

Conclusions:

• The proposed framework significantly improves broiler weight estimation accuracy, especially for older birds.
• It offers a more robust and economically viable alternative to traditional weighing methods.
• This advancement has the potential to enhance broiler farming management and efficiency.