Image-based quantification of chicken carcass size using deep learning and image processing to support screening in...

Area of Science:

• Agricultural Science
• Computer Vision
• Food Safety

Background:

• Poultry inspection in Miyazaki Prefecture faces challenges due to high volume, potentially impacting inspector burden and reliability.
• Current visual inspection methods for chicken carcasses may be subjective and labor-intensive.

Purpose of the Study:

• To develop and validate an image-based artificial intelligence (AI) method for supporting the screening of nonconforming chicken carcasses.
• To assess the potential of AI-derived carcass measurements in identifying conditions like emaciation and wooden breast (WB).

Main Methods:

• Utilized You Only Look Once version 8 (YOLOv8) for geometric correction and detection of chicken carcasses and shackles from video footage.
• Calculated carcass area and a novel thickness index (aspect ratio) using reference scales and estimated line speed.
• Correlated AI-derived indices with visual assessments by a former poultry inspector.

Main Results:

• The YOLOv8 model achieved high accuracy (mAP50 = 0.995) in carcass detection.
• Carcass area and thickness index were found to be distinct, complementary features capturing different aspects of conformation.
• AI-derived parameters showed potential for classifying nonconforming carcasses when applied to scatterplots.

Conclusions:

• The developed image-based method shows promise as a supplementary tool for poultry inspection, potentially reducing inspector workload.
• This AI approach can contribute to standardizing inspection quality and advancing automated poultry inspection systems.