Non-Destructive Measurement of Sugar Content in Litchis Using Visible and Near-Infrared Spectroscopy and Fuzzy...

Area of Science:

• Agricultural Science
• Spectroscopy
• Machine Learning

Background:

• Non-destructive sugar content measurement is crucial for efficient litchi sorting.
• Existing models often suffer from poor fitting and lengthy training times.

Purpose of the Study:

• To develop an efficient predictive model for litchi sugar content using visible-near-infrared (Vis-NIR) spectral data.
• To overcome limitations of current models by proposing a novel Fuzzy Stochastic Configuration Network (FSCN).

Main Methods:

• Developed an FSCN model integrating fuzzy neural networks and stochastic configuration networks.
• Evaluated 12 spectral preprocessing methods and 4 feature extraction techniques.
• Utilized 656 spectral data samples from litchis for model training and validation.

Main Results:

• The optimal preprocessing combination was multiplicative scatter correction (MSC) and successive projections algorithm (SPA).
• The FSCN model achieved high accuracy with R² of 0.9676, RMSE of 0.5487, and MAE of 0.4441.
• FSCN outperformed PLSR, ANN, and SVR models in predictive performance.

Conclusions:

• The FSCN model demonstrates superior capability for accurate sugar content detection in litchis.
• This model is highly suitable for practical application in automated litchi sorting systems.
• The study highlights the potential of intelligent algorithms in agricultural product quality assessment.