[Effectively predicting soluble solids content in apple based on hyperspectral imaging].

Area of Science:

• Agricultural Science
• Spectroscopy
• Data Analysis

Background:

• Accurate assessment of internal fruit quality, such as soluble solids content (SSC), is crucial for agricultural produce.
• Hyperspectral imaging offers a non-destructive method for evaluating fruit quality, but requires effective wavelength selection.

Purpose of the Study:

• To identify optimal wavelengths for predicting Yantai "Fuji" apple SSC using hyperspectral imaging.
• To compare different algorithms (Genetic Algorithm - GA, Successive Projections Algorithm - SPA, GA-SPA) for wavelength selection.
• To evaluate various modeling techniques (Partial Least Squares - PLS, Least Squared Support Vector Machine - LS-SVM, Multiple Linear Regression - MLR) for SSC prediction.

Main Methods:

• Hyperspectral images (400-1000 nm) of 160 apple samples were acquired.
• Effective wavelengths were extracted using GA, SPA, and a combined GA-SPA algorithm.
• Predictive models for SSC were built using PLS, LS-SVM, and MLR based on selected wavelengths.

Main Results:

• The Successive Projections Algorithm (SPA) combined with Multiple Linear Regression (MLR) yielded the best prediction performance.
• The SPA-MLR model achieved a coefficient of determination (Rp(2)) of 0.9501, a Root Mean Square Error of Prediction (RMSEP) of 0.3087, and a Ratio of Performance to Deviation (RPD) of 4.4766.
• SPA proved effective in selecting optimal wavelengths from hyperspectral data.

Conclusions:

• The SPA-MLR model demonstrates high accuracy and efficiency for predicting apple SSC.
• The selected wavelengths and the MLR model offer potential for developing online detection instruments and portable devices for apple quality assessment.