Chick embryo development assessment and fertility detection using pixel-wise hyperspectral image analysis and deep...

Area of Science:

• Agricultural Science
• Biotechnology
• Imaging Technology

Background:

• Efficient poultry production relies on accurate early fertility and embryonic development assessment.
• Traditional methods for evaluating egg fertility and development are often destructive or labor-intensive.

Purpose of the Study:

• To develop and validate a non-destructive hyperspectral imaging (HSI) system for real-time fertility and embryonic development monitoring in poultry eggs.
• To compare the performance of various classification algorithms for HSI data analysis.

Main Methods:

• Acquisition of full-transmittance hyperspectral images from white-shelled eggs on the first four incubation days.
• Classification using average spectra with methods including SIMCA, LDA, QDA, and ANN.
• Pixel-wise classification using ANN, Random Forest, DNN, and CNN for enhanced accuracy and spatial visualization.

Main Results:

• Artificial Neural Networks (ANN) showed high performance in embryonic stage detection (F1-score: 92.33%).
• Deep Neural Networks (DNN) achieved the highest stage discrimination (F1-score: 94.95%) in pixel-wise classification.
• HSI combined with deep learning models demonstrated excellent accuracy for fertility detection across incubation days (up to 100% F1-score).

Conclusions:

• Hyperspectral imaging offers a promising non-destructive approach for early fertility assessment in poultry.
• Pixel-wise deep learning classification significantly enhances the accuracy and spatial resolution of fertility and embryonic development monitoring.
• This integrated HSI and deep learning methodology can improve hatchability efficiency in commercial poultry hatcheries.