Accurate early detection of fertility and embryonic development during incubation is essential for enhancing hatchability efficiency in poultry production. This study employed a line-scan hyperspectral imaging (HSI) system to monitor fertility and embryonic development in white-shelled eggs. Full-transmittance hyperspectral images were acquired at the end of each of the first four incubation days from both fertile and infertile eggs. Classification based on average spectra was performed using soft independent modeling of class analogy (SIMCA), linear discriminant analysis, quadratic discriminant analysis (QDA), and artificial neural networks (ANN). Among these, ANN achieved the highest performance in embryonic stage detection (F1-score: 92.33 %), while SIMCA, ANN, QDA, and ANN provided the best fertility detection on days 1-4, respectively (F1-scores: 77.78-100 %). Pixel-wise classification, incorporating ANN, random forest, deep neural networks (DNN), and convolutional neural networks (CNN), further improved accuracy and enabled spatial visualization. The DNN achieved the highest stage discrimination (F1-score: 94.95 %), while CNN, DNN, ANN, and DNN performed best in fertility detection on days 1-3 (F1-score: 90.00 %) and day 4 (F1-score: 100.00 %), respectively. These findings demonstrate the potential of HSI combined with pixel-wise deep learning classification as a non-destructive and accurate method for real-time fertility assessment in poultry hatcheries.
Keywords: Deep learning classification; Embryo development; Fertility detection; Poultry hatchery automation; Spectral-spatial modeling.
Copyright © 2025. Published by Elsevier Inc.