Research on apple identity recognition method based on hyperspectral technology and chemometrics method

Abstract

This study focuses on using hyperspectral technology and chemometric methods to identify apples from four production areas and ten varieties to improve the accuracy of apple identification. Physical and chemical data and hyperspectral images of apple samples were collected for preprocessing and feature selection of spectral data. In the next step, K-nearest neighbour algorithm (KNN) and the support vector machine-radial basis kernel function (SVM-RBF) models were established. Regarding the identification of apple varieties based on origin information, research was conducted for both the same and different origins. For the same origin modeling, the best results were obtained from the KNN model, with accuracy, F1-score, and recall all reaching 100% for both the modeling and prediction sets. The PLS-DA prediction results were slightly inferior to KNN, and SVM modeling produced the worst results. The PLS-DA model produced the best results for predicting different apple origins, with accuracy, F1-score, and recall all achieving over 98% for both the modeling and prediction sets. Regarding the identification of apple origin based on variety information, research was conducted for both the same and different varieties. The best model for identifying the origin of apples with the same variety was SD-PLS-DA, with accuracy, F1-score, and recall of 97.38%, 99.59%, and 99.60% for the modeling set, and 99.61%, 97.35%, and 97.84% for the prediction set. For different varieties, both the KNN and PLS-DA models were effective, with the KNN model outperforming the PLS-DA model, and all indicators reaching 100% for both the modeling and prediction sets. The results indicated that the use of variety and origin identification models in this study was effective for apple identification, and can accurately identify apples.