Abstract: A portable visible and near-infrared (VIS-NIR) spectroscopy-based device with a micro-spectrometer as the core component was built for rapid and non-destructive quality detection of kabocha squash. The spectral data of kabocha squash at different development and storage periods were collected using this device and pretreated by first derivative, Savitzky-Golay (SG), multiplicative scatter correction (MSC) or their combinations. The best spectral pretreatment method was selected. The characteristic wavelengths were extracted by continuous projection algorithm, and predictive models for soluble solids content (SSC) and firmness were established by backpropagation neural network, multiple linear regression or partial least squares regression. The optimal SSC and firmness prediction models were selected and imported into the device for rapid non-destructive testing of the SSC and firmness of kabocha squash. The results showed that the optimal spectral preprocessing method for the SSC was SG + MSC, and the backpropagation neural network model was selected as the optimal prediction model. The prediction set determination coefficient (Rp2), root mean squared error of prediction (RMSEP) and residual prediction deviation (RPD) were 0.895 5, 0.874 4 °Brix and 2.809 7, respectively. The optimal spectral pretreatment method for kabocha squash firmness was SG + MSC, and the optimal prediction model was the multiple linear regression prediction model, with Rp2, RMSEP and RPD of 0.910 7, 3.029 4 kg/cm2 and 3.214 4, respectively. The above results show that the device can predict the SSC and firmness of kabocha squash well and can be used for their rapid nondestructive testing.

Key words: kabocha squash; visible and near-infrared spectroscopy; soluble solids content; firmness; non-destructive detection