Non-Destructive Prediction of Loquat Sugar Content and Hardness Using Hyperspectral Imaging Combined with Weighted Band Selection

Sugar content and firmness are important indicators for measuring the quality of loquats. In order to perform non-destructive testing on the sugar content and hardness of loquat, the hyperspectral image information of 262 loquats was obtained by using a visible light/near infrared (400–1167 nm) hyperspectral imaging system. The original spectra were preprocessed using moving average smoothing (MA) and convolution smoothing (SG). On this basis, the weighted band selection algorithm (WBS), successive projection algorithm (SPA), principal component analysis (PCA) and SPA-WBS were used to reduce the dimension of hyperspectral data and extract characteristic variables. Random forest (RF) and partial least squares regression (PLSR) prediction models were established to detect the sugar content and hardness of loquat, and the optimal prediction model was established. Based on the optimal prediction model, the sugar content and firmness of loquat fruits were fitted, enabling the visualization of the internal distribution of sugar content and firmness within the fruits. The results indicate that: The RF model based on SPA-WBS has the best prediction effect on sugar content, with correlation coefficients of the training set and test set being 0.9269 and 0.9162, and root mean square errors being 0.4366 and 0.4336, respectively; The RF model based on the full spectrum band achieves the best hardness prediction effect, and the correlation coefficient of the test set can reach 0.8726. The methods proposed in this work enable non-destructive prediction and visualization research of sugar content and hardness in loquat.