Hyperspectral imaging with multivariate analysis for detection of exterior flaws for quality evaluation of apples and pears

Abstract

Hyperspectral imaging (HSI) has emerged as a highly effective, non-destructive technique for detecting external and subsurface defects in agricultural products, offering rapid and accurate quality assessment. This study employed an advanced HSI system operating in the 400–2500 nm range, encompassing visible near-infrared (VNIR) and short-wave infrared (SWIR) wavelengths, to identify multiple defects in apples and pears. Specifically, this research focused on three key defect types, bruises, scars, and diseases, while addressing challenges related to defect size, shape, and severity variations. To enhance the spectral variability and improve detection accuracy, apples and pears were manually bruised in varying sizes, simulating both the early and advanced stages of bruising. These samples were analyzed at regular intervals over two days, capturing the progression of defect characteristics. This process yielded over 8300 spectral data points from carefully selected regions of interest (ROIs), providing a comprehensive dataset for analysis. Three multivariate models, linear discriminant analysis (LDA), support vector machines (SVM), and partial least squares discriminant analysis (PLS-DA), were employed to classify normal and defective fruit. Among these, the PLS-DA model demonstrated the highest performance, achieving validation classification accuracies of 97.5 % for apples and 100 % for pears in the VNIR range and 98 % for apples and 99.9 % for pears in the SWIR range. Moreover, successive projection algorithms (SPA) were employed for optimal band selection, and the obtained SPA beta coefficients were derived from the high-performing models and generated chemical visualization maps, enabling precise localization and detailed characterization of defects. By combining rapid and accurate defect identification with comprehensive flaw characterization, this study provides a practical and efficient framework for industrial post-harvest quality control, ultimately enhancing high-value agricultural products' marketability and storage stability.