VIS-NIR Spectroscopy for Ripeness Detection in Table Grapes

01 Research Background

Fruit ripeness is a critical quality indicator for fruits, directly affecting yield, processing methods, and storage. Grapes, with their rich vitamin content and diverse applications (fresh consumption, juice, wine), hold significant agricultural and economic value.

 

Phenolic compounds, primarily found in grape skins and seeds, serve as key maturity markers. Traditional chemical assays for phenolics are time-consuming, destructive, and costly. Visible-near infrared (Vis-NIR) spectroscopy offers a rapid, non-destructive alternative by analyzing spectral features correlated with phenolic content.

 

Advantages of Vis-NIR Spectroscopy:

• Non-destructive analysis

• High-speed data acquisition

• Suitable for portable/online detection systems

 

02 Experimental Procedure

Sample Collection:

 

  

• Sampling Method: Five-point sampling from grape clusters (upper, middle, lower, sunlit, shaded sides).

• Sampling Period: From veraison to full maturity, every 3–5 days.

• Total Samples: Collecting 280 to 320 grains each time, 24 batches (178 groups, 40 grapes/group).

• Storage: Grapes frozen at −40°C for subsequent spectral and chemical analysis.

 

Spectral Acquisition:

 

 

As show in the figure, in order to minimize the stray light interference, spectral acquisition was conducted in a dark room using a whiteboard as background. The system consists of halogen light, fiber optical probe, ATP3030 high-resolution miniature spectrometer of Optosky (spectral range 200-1100nm, 0.5nm resolution, integration time 1ms), standard diffuse reflectance whiteboard, and samples. The handheld spectral acquisition module integrates the halogen light source, dark room, and probe, connected via fiber and spectrometer.

 

The spectral acquisition procedure follows three key steps：

• Calibration: Capture reference spectra by aligning the spectral acquisition module with the standard diffuse reflectance whiteboard.

• Sample Measurement: Place individual grape samples on the whiteboard within the handheld dark room and acquire reflectance spectra using dedicated software.

• Data Processing: Average three measurements taken at 120°intervals along the sample equator as final spectral data.

 

 

Fig. 1 Reflectance curve of grapes during mature period

 

Spectral Analysis

To eliminate spectrometer artifacts and external interference, spectral data were cropped to 400-1100nm for subsequent modeling. Key observations:

• 500-720nm: Spectral divergence correlated with grape maturity (skin color variations affecting light absorption).

• ~720nm: Prominent absorption peak, likely linked to carotene and chlorophyll.

• 850-920nm: Absorption band potentially tied to phenolics, anthocyanins, cellulose, and sucrose.

• 1000-1050nm: Absorption features suggesting contributions from water and vitamins.

Substance determination:

Following spectral measurements, grape samples were frozen at -40°C. Subsequent processing included separation of peel and seeds, followed by grinding into powder, ultrasonic treatment, centrifugation, and supernatant collection for quantitative analysis of total phenolics and tannins.

 

Fig. 2 Total Phenolic Calibration Curve

 

Total phenolics, the dominant phenolic components in grape peel/seed extracts, were measured using a modified Folin-Ciocalteu method. Quantification employed gallic acid calibration curves, expressed as mg/g gallic acid equivalents (GAE). Tannins were determined via methyl cellulose precipitable (MCP) assay, with concentrations calculated against catechin calibration standards and reported as mg/g catechin equivalents (CE).

 

 Fig. 3 Tannins Calibration Curve

03 Model Development

Data Preprocessing & Modeling:

• Preprocessing: Savitzky-Golay smoothing, multiplicative scatter correction (MSC).

• Feature Selection: Competitive Adaptive Reweighted Sampling (CARS) identified critical wavelengths.

• CNN Architecture: Deep learning model for phenolic content prediction.

 

 

Fig. 9 CARS-Selected Feature Wavelengths

 

Through systematic evaluation of spectral preprocessing techniques and full-spectrum modeling, we developed feature wavelength-based predictive models. This methodology enables efficient and precise quantification of total phenolics and tannins in grape peels/seeds using spectral data.

 

 

Fig. 10 Optimal modeling of total skin-seed phenols and tannins in visible/near infrared spectra of grape samples

 

Results:

• High-accuracy models for skin/seed phenolics and tannins (R² > 0.90).

• Optimal performance in 400–1100 nm range.

04 Conclusions

It was shown that by using visible near infrared spectroscopy, collecting grape reflectance spectra and establishing a prediction model, it is possible to realize the detection of total phenol and tannin content of grape skin-seed, and then realize the detection of grape ripeness and quality.

Spectroscopic technology has the characteristics of fast spectral formation time and no damage to the sample, and through model optimization, it can achieve highly accurate detection, which has great potential for the detection of ripeness of grapes or other fruits.

 

 

References

1. Liu W., et al. "Vis-NIR Spectroscopy for Ripeness Evaluation in Table Grapes," Journal of Agricultural Machinery, 2024, 02. DOI:10.6041/j.issn.1000-1298.2024.02.037.
2. Gao S., et al. "Portable Vis/NIR Detector for Grape Quality Assessment," Journal of Agricultural Machinery, 2021, 02. DOI:10.6041/j.issn.1000-1298.2021.02.030.