An ensemble deep learning method for predicting cadmium content in eggs using hyperspectral imaging

Abstract

The purpose of this research was to investigate the feasibility of applying hyperspectral imaging (HSI) technique for the determination of heavy metal cadmium (Cd) in eggs. An ensemble deep learning algorithm fusing deep learning and ensemble learning was proposed to achieve end-to-end prediction. Firstly, discrete wavelet transform (DWT) algorithm was used to decompose the raw spectra to obtain scale information. Then, one-dimensional convolutional neural network (1DCNN) sub-models were established based on the wavelet coefficients of each scale. Finally, the sub-model space was optimized using a filtering strategy to construct ensemble deep learning model. The results showed that the ensemble deep learning model had stronger stability and higher prediction accuracy than conventional hyperspectral analysis methods, with R²_p of 0.91, RMSEP of 0.0285 mg/kg and RPD of 3.40. The proposed combination of HSI with ensemble deep learning model has great potential for in-situ and non-destructive detection of heavy metals in eggs.