Non-destructive detection of local radishes’ hollow hearts by hyperspectral imaging combined with support vector machine

Abstract. Hollow hearts are the major factor which affects the edibility of local radishes in Hangzhou, P.R. China. In the local variety of the radish (Yidianhong), the rate of hollow hearts accounts for 1/3 approximately. Identifying by naked eyes after cutting open the local radish based on experience is the way to differentiate edible local radishes from inedible radishes with hollow hearts at present. This method is simple but destructive, while hyperspectral imaging combined with chemometrics could bypass this limitation and its feasibility was investigated. 155 local radishes (116 edible samples and 39 inedible samples with hollow hearts inside) were used in this work and the reflection hyperspectral image (400-1000nm) of each sample was acquired. After extracting the average spectrum and comparing various methods for spectral preprocessing, such as multiplicative scatter correction (MSC), standard normal variate (SNV) and mean-normalization, the raw spectra were applied to establish the calibration set and a SVM model was then developed. The parameter of the SVM model was classification (C-SVC) for the type, polynomial for the kernel type, 3 for degree, 0.5 for C value and 0.00195 for g. The accuracy of the calibration set and the prediction set were 100% and 96.08%, respectively. Hyperspectral imaging combined with SVM provides a rapid, non-destructive and accurate way to differentiate edible local radishes from inedible local radishes with hollow hearts.