Nondestructive estimation method of live chicken leg weight based on deep learning

Area of Science:

• Agricultural Science
• Computer Vision
• Biomedical Engineering

Background:

• Phenotype determination, specifically leg weight, is crucial for broiler breeding.
• Noninvasive methods are preferred to minimize animal stress and damage.
• Accurate leg weight estimation supports selective breeding programs.

Purpose of the Study:

• To develop and validate a noninvasive system for estimating broiler chicken leg weight.
• To improve the accuracy and efficiency of leg weight measurement in live broiler chickens.
• To provide technical support for phenotype determination in commercial broiler breeding.

Main Methods:

• Development of a broiler leg weight estimation system integrating computed tomography (CT) acquisition equipment and a weight-estimation model.
• Implementation of an improved You-Only-Look-Once (YOLOv5) segmentation algorithm (YOLO-MCLW) with a segmentation head, multiscale attention, and atrous spatial pyramid pooling for accurate leg segmentation from CT scans.
• Utilizing a random forest fitting network to estimate leg weight from extracted morphological parameters of the segmented leg images.

Main Results:

• The developed system achieved an average absolute error of 7.27 g and an average percentage error of 4.82% in estimating broiler leg weight.
• The prediction coefficient of determination (R²) for broiler chicken legs reached 88.98%.
• The segmentation model achieved a high intersection over union (IoU) of 95.45% and processed 37.04 images per second.

Conclusions:

• The proposed noninvasive system effectively estimates broiler chicken leg weight with high accuracy and efficiency.
• The integration of an improved YOLOv5 model and random forest network offers a robust solution for automated leg weight determination.
• This technology provides valuable support for phenotype determination and selective breeding in the broiler industry.