Deep Learning Based Egg Fertility Detection

Area of Science:

• Agricultural Science
• Computer Vision
• Artificial Intelligence

Background:

• Fertile egg recognition is crucial for efficient poultry farming.
• Current methods for fertility assessment can be labor-intensive and time-consuming.
• Automated systems can improve accuracy and speed in egg fertility determination.

Purpose of the Study:

• To implement deep learning (DL) for fertile egg recognition from incubator images.
• To classify chicken eggs by segmentation and fertility status using a Mask R-CNN approach.
• To develop a single DL model for detection, classification, and segmentation of fertile and infertile eggs.

Main Methods:

• Utilized a Mask R-CNN-based deep learning model.
• Applied the model to images from incubator settings.
• Evaluated performance using Average Precision (AP) and Intersection over Union (IoU) metrics.
• Tested with datasets of 5 and 18 fertile eggs.

Main Results:

• Achieved accurate identification of fertile eggs with an optimal IoU threshold of 0.7.
• The system correctly determined all fertile eggs by the third day in both test datasets.
• Demonstrated the capability of a single DL model for detection, classification, and segmentation.

Conclusions:

• The developed DL system offers a highly successful and efficient solution for fertile egg recognition.
• The Mask R-CNN approach provides accurate segmentation and classification of egg fertility.
• Future work should focus on optimizing camera and lighting for enhanced segmentation performance.