Hyperspectral image information fusion-based detection of soluble solids content in red globe grapes

Abstract

• Red globe grape are widely loved for their full grain, firm flesh, sweetness and high nutrient content. SSC is one of the important internal qualities of the fruit and is an important indicator of fruit ripeness. • The traditional method for testing SSC of fruits is to conduct destructive detecting, which is tedious and time-consuming, and only samples can be detected, and the experimental samples are completely damaged after detecting and cannot be sold and consumed. • In this paper, the hyperspectral images and spectral information of the samples were collected by hyperspectral imaging technology, and a PLSR model of SSC of red globe grape based on the fusion of spectral information, image information and hyperspectral image information was built. • After comparative analysis, an optimal detection model of SSC of red globe grape was determined, which provides a new method for the non-destructive detection of SSC of red globe grape. The Soluble Solids Content (SSC) of red globe grapes is an important indicator of internal quality. In this paper, 360 red globe grapes in the growing stage were collected as samples and the spectral information and images of the samples were extracted. The Raw spectral (RAW) information was extracted using the one-time dimensionality reduction algorithm (GA, CARS, SPA, UVE) and the combined dimensionality reduction algorithm (CARS-SPA, UVE-SPA) to build the PLSR model of the spectral information. The grey-scale co-occurrence matrix of the image was extracted as the texture feature information of the image and combined with the color information of the image (R, G, B, H, S, V, L, a, b) to form 19 image features to build the PLSR model of the image information. Thus, the PLSR model based on the fusion of hyperspectral image information was built by fusing the spectra extracted with the successive projection algorithm (SPA) feature wavelength and the image information after dimensionality reduction by the principal component analysis algorithm (PCA). The results showed that if only the spectral information was used for modelling, the SPA algorithm effectively extracted the characteristic wavelengths of the red globe grapes of SSC spectral information and improved the prediction performance of the model. If only image information was used for modelling, the PCA algorithm effectively improved the detection performance of the model by dimensionality reduction, but the improved performance was limited. The correlation coefficients of the calibration set and prediction set of the PLSR model were 0.9775 and 0.9762, and the detection effect and stability of the model were greatly improved compared with those built unilaterally based on spectral information or image information, and a new non-destructive detection method was found for the detection about SSC of red globe grapes.