Evaluate The Quality Of The Fruit By Nir Spectroscopy

INTRODUCTION

NIR spectroscopy uses the interaction of near-infrared light with the fruit samples' molecular bonds to reveal details about the samples' physical characteristics and chemical makeup. Since it can potentially evaluate factors including total soluble solids content (sweetness indicator), acidity (sourness indicator), dry matter (in some case it is maturity indicator), moisture content (indicator for juiciness), lycopene content (such as in case of watermelon and tomato) and the presence of defects or contaminants without harming the sample, this approach has been explored and used extensively in the field of fruit quality assessment.

The non-destructive nature, speed of analysis, and ability to measure multiple quality indicators at once make NIR spectroscopy a favorable tool for assessing fruit quality. The quality of unknown fruit samples can be accurately estimated by creating reliable calibration models that link NIR spectra to reference data. This enables real-time monitoring, classification, and classification applications.

METHODS

From the instrument perspective, for different samples, the wavelength range and detection mode vary. When it comes to detection mode, reflectance is the most common spectral acquisition mode in NIR analysis. Light attenuation by the sample, relative to the reference, is known as reflectance (R).

Figure 1: Setup for the acquisition of reflectance spectra.

From an algorithmic perspective, the preprocessing, variable selection, and modeling parts are all important. There are different spectral preprocessing algorithms, corresponding to different functions: smoothing, minimizing noise effects, removing gradient shifts, and correcting dispersion effects. But in general, they all have the same goal of eliminating useless information in the NIR spectrum. Variable selection methods are used to make the model more efficient and effective.

Figure 2: Calibration of PLS regression by NIR systemmeasurements of soluble solid content of apple fruit.

After data pretreatment data is divided into a calibration set and a prediction set. The calibration set is the set that we use to develop the model and the prediction set is to prove the calibration set. The calibration set has absorbance X and reference value Y from which we can develop a calibration equation by regression analysis. By this process, we can check the performance of our calibration equation.

RESULTS

Figure 3: Scatter plots between actual SSC versus NIR-predicted SSC values of validation sets: (a) sapodilla, (b) banana.

The performance of the final prediction model is built on the results of the calibration model. First, the highest calibration correlation of calibration coefficient (R_c²) and lowest RMSEC. The prediction model in Figure 4, the PLS model from sapodilla and banana, shows very satisfactory performance. According to prediction model performance standards, the coefficient correlation of prediction (R_P²) in the banana and sapodilla samples, specifically 0.885 and 0.905, should be used with caution in different applications. The observed changes in the chemical composition of the samples have the potential to provide robust and reliable prediction models for several biomass species. It is difficult to obtain an acceptable fruit quality prediction model to evaluate SSC for rough and thick-skinned fruits such as dragon fruit compared to smooth and thin-skinned fruits using reflection mode due to the influence of light scattering and higher shell thickness.

BUILD SYSTEM

INTINS can provide a complete system for this application. NIRQuest+ spectrometer from Ocean Optics (Ocean Insight) features an improved optical bench design to deliver highly sensitive performance. High sensitivity allows for more accurate measurements especially in low light conditions. The long wavelength ranges from 900 - 2450 nm is a spectrometer suitable for evaluating the quantity of fruit applications, which can be used in the laboratory or on the industrial line.

Figure 4: Reflectance Measurement Setup.

The most important thing about choosing a light source for reflection is to find one with strong output over the wavelength range of interest. In this application, the light source needs to be observed in the NIR spectrum from 800 nm - 1600 nm. Our product - Tungsten halogen light source - is suitable to meet these standards. Ocean Insight's HL-2000 series offers models ranging from 380nm – 2400 nm and varies from high power models (HL-2000-HP Light Source) to long life models (HL-2000-FHSA-LL Light Source and HL -2000-LL Light Source), both meet your application requirements. Even though intensity decreases at longer wavelengths, this effect is offset by the higher sensitivity of the detectors in our NIR spectrometers at those wavelengths.

A reflection probe is great for making quick measurements and for applications where a small spot size needs to be sampled. Probes can be positioned to measure specular or diffuse reflectance and are compatible with our spectrometers and light sources (provided the probe fiber bundle matches the wavelength range of the light source). We offer reflection/backscatter probes in a variety of fiber core sizes and wavelength ranges. In this application, the R200-7-VIS-NIR reflectance probe is used, with a wavelength range of 400 nm - 2100 nm and a fiber core size of 200 um.

CONCLUSION

The application of NIR spectroscopy has demonstrated the potential for online, rapid, and non-destructive testing of maturation stages, assessing physicochemical properties (SSC, TSS, Lycopene, β-carotene, and Internal quality index) simultaneously. it can detect physical defects and contaminants in fruit. The ability to simultaneously evaluate multiple quality parameters is an advantage of the NIR technique. The composition of the fruit and its appearance is greatly influenced by the growing season, growing conditions, and environmental parameters. Therefore, large representative samples from different growing conditions need to be combined to develop accurate and robust calibration and classification models. Robust calibration models (such as PLS) can be developed using product variability and large sample sizes for accurate detection using an online system. The highest accuracy is achieved by applying appropriate spectral and spatial preprocessing methods, selecting a few important wavelengths, extracting important features, and building a classification/prediction model exactly.