Quantitative detection of cadmium pollution in lettuce leaves under selenium influence via fluorescence hyperspectral...

Heavy metal cadmium (Cd) pollution poses a serious threat to agricultural safety, and the application of selenium (Se) to mitigate Cd stress introduces additional complexity for Cd nondestructive detection. This study investigated the feasibility of employing fluorescence hyperspectral imaging (F-HSI) for detecting Cd pollution in lettuce leaves under Se influence. The raw spectra were preprocessed using savitzky-golay filter to effectively suppress noise and enhance subtle spectral features, thereby highlighting weak fluorescence spectral information. Subsequently, a multi-modal differential fusion selector (M²DFS) was proposed to extract sensitive features closely associated with Cd by sorting out the complex fluorescence spectra-Se-Cd relationships under SeCd interaction conditions. Combined with a one-dimensional convolutional neural network (1D-CNN) based on the VGG architecture, the quantitative prediction model for Cd content in lettuce leaves under the background of SeCd interaction was developed. Compared with classical methods, M²DFS demonstrated superior feature extraction performance in both machine learning and deep learning models, significantly enhanced the predictive capability of Cd content under Se influence. Ultimately, the 1D-CNN model built using M²DFS-features achieved the highest prediction performance (R_p = 0.9173, RMSEP = 0.0272 mg/kg, RPD = 2.5108). In summary, the method system of F-HSI combined with savitzky-golay filter, M²DFS and 1D-CNN provides a reliable approach for nondestructive detection of Cd pollution in lettuce leaves under Se influence, offering a foundation for future research on nondestructive detection of heavy metal under complex agronomic environments.