Surface-enhanced Raman spectroscopy with gold nanorods modified by sodium citrate and liquid-liquid interface self-extraction for detection of deoxynivalenol in Fusarium head blight-infected wheat kernels coupled with a fully convolution network

Surface-enhanced Raman spectroscopy (SERS) and deep learning network were adopted to develop a detection method for deoxynivalenol (DON) residues in Fusarium head blight (FHB)-infected wheat kernels. First, the liquid-liquid interface self-extraction was conducted for the rapid separation of DON in samples. Then, the gold nanorods modified with sodium citrate (Cit-AuNRs) were prepared as substrate for a gigantic enhancement of SERS signal. Results showed that the spectral characteristic peaks for DON residues of 99.5-0.5 mg/L were discernible with the relative standard deviation of 4.2%, with the limit of detection of 0.11 mg/L. Meanwhile, the fully convolutional network for the spectra of matrix input form was developed and obtained the optimal quantitative performance, with a root-mean-square error of prediction of 4.41 mg/L and coefficient of determination of prediction of 0.9827. Thus, the proposed method provides a simple, sensitive, and intelligent detection for DON in FHB-infected wheat kernels.

Automated summary

The study developed a detection method for DON residues in wheat using SERS and deep learning network, achieving rapid and accurate detection of DON at different concentrations through liquid-liquid interface self-extraction and SERS signal enhancement by gold nanorods. The fully convolutional network showed optimal quantitative performance, making the detection of DON in FHB-infected wheat kernels simple, sensitive, and intelligent.