Research on egg yolk color detection based on near infrared spectroscopy and machine vision.

Yolk color is a key indicator of egg quality, as customers prefer eggs with intensely yellow yolks, which also signal nutrient richness. At present, the commonly used method for yolk color detection is to open the eggs and evaluate the yolk color using the Roche yolk color fan (RYCF), so developing a non-destructive method for discrimination of yolk color is of great significance. In order to overcome the human subjectivity associated with RYCF based on yolk color scoring, a machine vision method was built to classify the yolk color grades more objectively and precisely. In this work, a total of 150 egg samples with yolk color scores from 5 to 11 were collected and the near-infrared (NIR) spectral data of intact eggs and egg yolks were gathered independently, while the true scores of yolk color grades were acquired using the machine vision system as the target set for modeling. Finally, different regression prediction models for egg yolk color grades were constructed using chemometric Partial Least Squares (PLS) and machine learning techniques, such as Temporal Convolutional Network - Gated Recurrent Unit-Attention (TCN-GRU-Attention), Least Squares Support Vector Machines (LSSVM) and Convolutional Neural Network-Bidirectional Long Short Term Memory-Adaptive Boosting (CNN-BiLSTM-Adaboost). For the intact egg and separated yolk spectral data, the results show that the PLS model achieved the best prediction accuracy in the test set, with R² values of 0.9035 and 0.9274, and the root mean square errors (RMSE) were 0.3665 and 0.2933, respectively, which accomplished the non-destructive quantitative detection of egg yolk color scores.