Influence of near-infrared on-line detection device parameters on the applicability of apple soluble solid content model

Objective  The performance of the apple soluble solid content model was influenced by the detection speed and integration time of the near-infrared (NIR) spectroscopy on-line detection device. The aim of this study was to analyze the influence of detection speed and integration time on the applicability of the model, and improve the accuracy of on-line detection.

Method  The detection speed and integration time of the on-line detection device of NIR spectroscopy were setted as 0.3 m/s and 100 ms, 0.5 m/s and 70 ms, 0.5 m/s and 100 ms, 0.5 m/s and 120 ms, 0.5 m/s and 150 ms respectively. A total of 180 apple samples were used for the experiment, the NIR spectra of five experimental groups of apples were collected at 350~1 150 nm, and the partial least squares (PLS) method was applied to establish the prediction model of apple soluble solid content (SSC).

Result  There was a threshold for the integration time. When the integration time was shorter than 70 ms, the prediction performance of the model was poor. The prediction performance of the prediction model with the same detection speed and integration time for modeling set and prediction set was superior to that of the prediction model with different dection speed and integration time. Detection speed and integration time would affect the accuracy of on-line detection. The different transmission routes of light inside the apple at different detection speeds and integration time could lead to differences in the internal information obtained by the fiber optic probe, making the prediction performance worse. The 135 representative sample spectra were selected from four groups of 0.3 m/s and 100 ms, 0.5 m/s and 100 ms, 0.5 m/s and 120 ms, 0.5 m/s and 150 ms using the Kennard-Stone algorithm, the prediction models of the mixed detection speed and integration time were established, with the correlation coefficients (R_P) of the prediction set all above 0.85, the root mean square errors (RMSE_P) all below 0.65.

Conclusion  The established prediction model of mixed detection speed and integration time can better predict apple sugar content and meet the requirements of apple sugar content on-line detection under different parameters of detection devices.