Non-destructive detection of egg qualities based on hyperspectral imaging

Egg quality detection is important to food processing and people consumption. The aim of this study is to detect egg freshness, scattered yolk and eggshell cracks by applying hyperspectral imaging (HSI), multivariate analysis and image process. The transmission visible-near infrared hyperspectral images of egg samples were acquired in the wavelength range of 401-1002 nm. Standard normal variate (SNV) was applied to normalize the spectral data, and iteratively retains informative variable (IRIV) was used to optimize wavelength selection. Based on the feature wavelengths, egg freshness quantitative model was established by using Extreme Gradient Boosting (XGBoost), with coefficient of determination for prediction (R2p) of 0.91 and root mean square error for prediction (RMSEP) of 4.64. An algorithm including image contrast enhancement, denoising and threshold segmentation was proposed to extract the morphological features of yolk. Based on the morphological feature ratio, the recognition accuracy of scattered yolk eggs reached 97.33%. In addition, a method including crack enhancement, double threshold segmentation was developed to extract the geometric features of cracks. The cracked eggs could be discriminated by XGBoost classification model with identification accuracy of 93.33%. The results indicate that HSI can be useful for the non-destructive detection of egg qualities.

Automated summary

This study successfully detects egg freshness, scattered yolk, and eggshell cracks using hyperspectral imaging, multivariate analysis, and image processing. The study also proposes models and algorithms for analyzing and extracting the features of eggs, achieving quantitative assessment and identification of egg quality.