Automatic detection of scratches on chicken carcass in slaughter factory using image processing and machine learning

Abstract

Importance of the work: Ensuring the quality of chicken products is paramount, as low-quality chicken not only impacts consumer health but also results in time and financial losses. Automation can improve production by reducing labor costs and increasing efficiency.
Objectives: To develop a software program to process video images acquired on a chicken carcass production line and categorize each carcass based on identified scratches.
Materials & Methods: The treatment consisted of counting the scratches on chickens to categorize them in terms of meat quality. The developed software used machine learning and model building to create models to detect scratches on the chicken carcass. The software identified and marked groups of pixels defined by the patterns on the image. Results were compared to those from manual assessment by an experienced operator.
Results: The software achieved 95% accuracy and directly processed videos without the need for pre-processing, such as background removal. The carcass categories were determined based on the number of scratches based on calibration with the software. Each of the three mislabeled chickens had a significant scratch, but no large scratches are observed. The grade of the chicken was directly affected by these scratches, with important factors being poor detection and the absence of a large scratch pattern.
Main finding: The program could be improved by incorporating calibration from both sides of the carcass. This would improve the adaptability of the developed system. It should be possible to develop a system using visual image data to automatically sort chickens on the production line into different categories.