Corn grain year identification based on Raman spectroscopy and machine learning

Background As one of the important food crops, corn’s reserve issue has attracted much attention. Given the direct impact of corn year on its quality and supply, as well as the limitations of traditional manual identification methods, this study aims to explore the feasibility of combining Raman spectroscopy with machine learning methods in corn year identification. Methods Five different years of corn were selected as experimental samples, and spectral data of corn were obtained through Raman spectroscopy. The spectral data were subjected to a combination of preprocessing methods such as convolution smoothing (SG), wavelet transform (WT), standard normal variable transform (SNV), and first derivative processing. Then, important feature information is extracted through dimensionality reduction using linear discriminant analysis (LDA) and independent component analysis (ICA). Finally, year recognition is achieved through support vector machine (SVM), random forest, and K-nearest neighbor algorithm. Results The experimental results show that the proposed method exhibits good year recognition performance. Among them, the combination of SG and SNV preprocessing and ICA feature extraction significantly shortened the running time and improved the recognition accuracy. The SVM classification model performs the best, with an accuracy of up to 96.69%, a recall rate and F1 value of 96.77% and 0.9659, respectively, and a running time of only 99.53 s, indicating that the model based on Raman spectroscopy and machine learning can efficiently and accurately identify corn years.