Carrot Quality Assessment Using Hyperspectral Imaging and Multivariate Analysis

By Samudrapom DamApr 22 2024Reviewed by Susha Cheriyedath, M.Sc.

In an article recently published in the journal Scientific Reports, researchers performed quantitative measurement of carrots' internal quality using hyperspectral imaging (HSI) and multivariate analysis involving partial least squares regression (PLSR) and least squares support vector machine (LS-SVM).

Carrot Quality and Inspection

Carrot is a root vegetable crop with high nutritional value as it contains essential micronutrients like vitamins C and A. Although carrots are usually orange, these crops also exhibit several other colors, including yellow, red, and purple, which enriches the diversity in the spectrum.

Additionally, these widely consumed crops provide substantial amounts of carotenoids, antioxidants, and provitamin A, which deliver different health benefits, including improved liver and heart health and a lower prostate cancer risk. Factors such as firmness, vitamin C levels, provitamin A content, color, and absence of bruises are the crucial quality indicators for carrots, impacting consumer satisfaction, market value, and shelf life.

Thus, the development of rapid quality control technologies providing detailed and precise information regarding nutritional content and the improvement of carrot quality inspection have become critical due to climate change effects and increasing consumption. Using this information, the storage parameters can be refined, the most suitable harvesting time can be ascertained, and the processed carrot derivatives' nutritional quality can be improved.

Quantitative Measurement of Internal Quality

In this study, researchers measured the pH and carotenoid contents of carrots using HSI. Overall, 300 images covering 472 wavebands from 400 to 1000 nm were obtained using an HSI system. Regions of interest were defined for extracting average spectra from the hyperspectral images (HIS). 300 carrot samples with comparable size and shape were procured from Putian and Fuzhou City.

Additionally, two models, including PLSR and LS-SVM, were developed using the entire spectrum to perform a quantitative analysis between the spectra and pigment amounts. The pigment contents and spectra were correlated and predicted using these models. The early warning signals (EWs)/representative wavelength selection for modeling was performed using the regression coefficients (RC) and successive projections algorithm (SPA) from PLSR models and LS-SVM.

The objective of the study was to investigate the potential of hyperspectral reflectance imaging for predicting the pH and carotenoid content of carrots to comprehensively assess the carrots' internal quality attributes. Near-infrared and visible HSI were utilized to evaluate these parameters in the xylem and cortex of carrots.

The collection of image and spectral data simultaneously from the tested sample using HSI combines digital imaging technology and conventional spectroscopy into a system. The HSI technology can provide significant information in spatial and spectral domains.

Recently, HSI has played a crucial role in detecting agricultural products' internal quality. Studies have demonstrated that near-infrared/visible HSI technology can be used for assessing interior fruit attributes, including firmness and soluble solids content, in a non-destructive manner.

Model validation is a crucial step in multivariate data analysis. In this work, the prediction model was developed using LS-SVM and PLSR, which are linear multivariate algorithms, as these algorithms are effective when a linear relationship is present between object properties and spectra.

PLSR is used extensively in chemometrics for analyzing the correlation between reference quality indicators and spectral data. The PLSR model predicted the pH and carotenoid levels using a set of latent variables that are statistically uncorrelated. Similarly, LS-SVM, an enhancement of conventional SVM, utilizes least-squares linear systems as the loss function in place of conventional convex quadratic programming.

Study Findings

The results displayed that HSI could effectively evaluate the internal attributes of the carrot cortex and xylem. The developed models could accurately predict the pH and carotenoid contents of the carrot parts. The models' predictive capabilities for pH and carotenoid attributes were improved through careful analysis and selection of EWs using SPA and RC methods. This indicates the importance of custom wavelength selection in multispectral imaging applications requiring efficient and accurate quality attribute predictions in the scientific and agricultural sectors.

The LS-SVM model effectively evaluated the carrots' internal qualities, specifically the pH level, while the PLSR model displayed a significant discrepancy between the prediction generated and the correction set. Additionally, the PLSR model did not demonstrate adequate accuracy in fitting predictions and stability regarding pH quality.

Thus, the LS-SVM model can be implemented for precisely assessing the carrots' pH level, which can enable producers and farmers to make informed decisions concerning distribution, storage, and harvesting. Similarly, the LS-SVM model showed superior performance compared to the PLSR model based on the prediction accuracy for carotenoid content, specifically for Fuzhou (Fz) carrots. This model also possessed optimal stability and fitting accuracy.

To summarize, the findings of this study demonstrated the feasibility of the proposed approach based on hyperspectral imaging and multivariate analysis for internal quality assessment of carrots.