Detection of starch content in maize kernel based on Raman hyperspectral imaging technique

Abstract

Starch is an important component for maize to ensure the quality of maize kernels. However, maize aging has a negative effect on the quality. In this study, Raman hyperspectral imaging technique was employed to detect the starch content during maize aging. The typical Raman peaks were analyzed and two-dimensional correlation spectroscopy (2D-COS) was applied to further characterize the relationship between Raman peaks and maize aging process. The changes order of Raman peaks under the effect of maize aging were revealed. Six preprocessing methods and five modeling methods were used to establish the maize starch content prediction model. The characteristic Raman shifts extracted by variable selection methods and 2D-COS peaks were used as inputs to establish the prediction model, and further fused to explore the model. The results showed that ELM model coupled with 35 characteristic Raman shifts based on normalization preprocessing method exhibited the optimal performance with Rc, Rp, RMSEC and RMSEP of 0.8827, 0.8502, 0.0539, 0.0577, respectively. In conclusion, this study provided a feasibly and powerfully nondestructive detection method for the determination of maize starch during maize aging and laid a theoretical foundation in the cereal detection industry.