Combining hyperspectral imaging technology and visible-near infrared spectroscopy with a data fusion strategy for the detection of soluble solids content in apples

Soluble solids content (SSC) is an important indicator for evaluating apple quality. This study aimed to assess two spectral techniques (hyperspectral imaging (HSI) and visible-near infrared (Vis-NIR)) combined with low-level and mid-level fusion strategies (LLF and MLF) for the detection of SSC in apples. Firstly, baseline correction (BC), detrending (DT), multiplicative scatter correction (MSC), savitzky-golay (SG), and standard normal variables (SNV) were used for the preprocessing of spectral data. Secondly, genetic algorithm (GA), competitive adaptive reweighted sampling (CARS), and their combinations were used for feature variable extraction. Finally, models were developed using partial least squares regression (PLSR) for spectral data. The results showed that Vis-NIR has an advantage over HSI in predicting apple SSC if only a single spectral technique was considered. The data fusion strategy showed better performance in predicting SSC metrics compared to individual spectral data. Among them, the LLF strategy showed the best performance in predicting SSC, with an $R_p^2$ of 0.927, an RMSEP of 0.529 °Brix, and an Akike information criterion (AIC) of 332.96. In addition, an optimal model based on HSI data was used to achieve visual maps of SSC. It was demonstrated that the fusion of HSI and Vis-NIR data provided a promising method for detecting the SSC in apples.