Identification of double-yolked duck egg using computer vision

Area of Science:

• Agricultural Science
• Computer Vision
• Biotechnology

Background:

• Double-yolked (DY) eggs are culturally significant but pose hatching challenges in the poultry industry.
• Automated identification of DY eggs can enhance efficiency and reduce economic losses.
• Current methods for egg yolk identification lack precision and speed.

Purpose of the Study:

• To develop and compare two computer vision-based methods for identifying double-yolked (DY) and single-yolked (SY) duck eggs.
• To assess the accuracy and efficiency of Fisher's linear discriminant (FLD) and convolutional neural network (CNN) models for DY egg detection.
• To determine the optimal method for automated DY duck egg identification in the poultry sector.

Main Methods:

• Acquired transmittance images of DY and SY duck eggs using a CCD camera.
• Developed a Fisher's linear discriminant (FLD) model using normalized Fourier descriptors (NFDs) for shape-based classification.
• Implemented a convolutional neural network (CNN) model utilizing preprocessed images for yolk type recognition.

Main Results:

• The FLD model achieved 100% accuracy for SY eggs and 93.2% for DY eggs.
• The CNN model achieved 98% accuracy for SY eggs and 98.8% for DY eggs.
• The CNN model processed images faster (0.12s) than the FLD model (0.20s).

Conclusions:

• Both FLD and CNN models show high potential for automated DY duck egg identification.
• The CNN model offers a slightly faster and more accurate solution for recognizing duck egg yolk types.
• This research provides a foundation for improving efficiency in the poultry industry through advanced computer vision techniques.