Hyperspectral imaging for trace cadmium prediction in lettuce leaves

Abstract

Cadmium (Cd) pollution presents a significant threat to the agricultural product control, and the development of detection technology for Cd content in lettuce has important application value. This study developed a nondestructive approach based on hyperspectral imaging (HSI) to detect trace Cd content in lettuce. Using heterogeneous two-dimensional correlation spectroscopy (H2D-COS) to correlate visible near-infrared and fluorescence spectra, and introducing Savitzky Golay (SG) derivative method to preprocess spectral data, the weak features related to Cd were effectively amplified through noise suppression and signal enhancement, and the optimal band (407, 432, 453, 490, 543, 576, 626, 661, 688, 754, 779, 804, 826, 871, 891, 911, 935 and 964 nm) for Cd content was determined. The predictive performance of multiple feature extraction methods (including SPA, RC, RF, 2D-COS) combined with machine learning models were also compared. The results showed that H2D-COS significantly improved the ability to extract Cd-related features by fusing the dynamic response of heterologous spectra, and the constructed support vector machine (SVM) model exhibited the best performance. The research results indicate that H2D-COS integrated with SG derivative can extract the key bands, and combined with HSI, it has potential for prediction of trace Cd content in lettuce leaves.