Early-stage fertilised egg viability detection based on machine vision.

Area of Science:

• Agricultural Science
• Biotechnology
• Computer Vision

Background:

• Early detection of fertilised egg viability is crucial for efficient incubation and hatchery management.
• Current methods for assessing egg viability can be labor-intensive and lack high-throughput capabilities.
• Challenges exist in accurately identifying multiple eggs and ensuring consistent detection during early incubation stages.

Purpose of the Study:

• To develop and validate an automated system for intelligent recognition and high-throughput detection of fertilised egg viability.
• To investigate factors influencing image consistency in transillumination imaging of eggs.
• To establish an optimal timeframe for early-stage viability detection using image analysis.

Main Methods:

• A monochrome camera captured transillumination images of eggs throughout incubation.
• Image processing techniques including median filtering, sharpening, and segmentation were applied to extract egg regions.
• Feature extraction (grayscale and texture) and classification models (Logistic Regression, XGBoost, LightGBM, CNN) were employed for viability assessment.

Main Results:

• The Convolutional Neural Network (CNN) model achieved 99% accuracy for viable embryos on day 8.
• Optimal detection was achieved on day 6 with 95% accuracy, outperforming manual inspection.
• The system demonstrated high precision, recall, and F1 scores for both viable and non-viable eggs.

Conclusions:

• Automated viability detection of fertilised eggs is feasible using monochrome cameras and advanced image processing/classification models.
• This technology offers a robust solution for high-throughput, early-stage screening, supporting automated incubation systems.
• The proposed method enhances efficiency and accuracy in hatchery operations, enabling timely interventions.