Estimation Method of Soluble Solid Content in Peach Based on Deep Features of Hyperspectral Imagery

Abstract

Soluble solids content (SSC) is one of the important components for evaluating fruit quality. The rapid development of hyperspectral imagery provides an efficient method for non-destructive detection of SSC. Previous studies have shown that the internal quality evaluation of fruits based on spectral information features achieves better results. However, the lack of comprehensive features limits the accurate estimation of fruit quality. Therefore, the deep learning theory is applied to the estimation of the soluble solid content of peaches, a method for estimating the SSC of fresh peaches based on the deep features of the hyperspectral image fusion information is proposed, and the estimation models of different neural network structures are designed based on the stack autoencoder–random forest (SAE-RF). The results show that the accuracy of the model based on the deep features of the fusion information of hyperspectral imagery is higher than that of the model based on spectral features or image features alone. In addition, the SAE-RF model based on the 1237-650-310-130 network structure has the best prediction effect (R2 = 0.9184, RMSE = 0.6693). Our research shows that the proposed method can improve the estimation accuracy of the soluble solid content of fresh peaches, which provides a theoretical basis for the non-destructive detection of other components of fresh peaches.