Exploring the power of advanced spectral imaging and multivariate analysis for distinguishing similar carrot cultivars

Abstract

Carrots are a widely consumed vegetable around the world. They come in several varieties, each with their distinct quality traits. This study uses hyperspectral imaging to identify and predict soluble solids content (SSC) in three carrot cultivars using machine learning-based spectral analysis. After preprocessing the raw spectra, we constructed a partial least squares regression (PLSR) model to predict SSC using the preprocessed full spectra. iPLS and competitive adaptive reweighted sampling (CARS) methods were compared for variable selection. To differentiate carrots from three orchards, we developed four classification models: CARS-LDA, iPLS-LDA, iPLS-KNN, and CARS-KNN. Results indicated that both CARS-LDA and iPLS achieved effective wavelength selection for the Fuzhou A carrot cultivar, with the CARS-LDA model demonstrating the highest predictive capability, reflected by a relative prediction deviation (RPD) value of 2.72. In terms of accuracy, specificity, sensitivity, and precision, the CARS-KNN model was the best. This study underscores the potential of hyperspectral imaging coupled with machine learning methodologies to reliably predict and distinguish between various carrot cultivars, thereby contributing significantly to improvements in food safety and quality control standards in the agricultural sector.