Study on bruising degree classification of apples using hyperspectral imaging and GS-SVM

Abstract Bruising degree can affect the end use and sales price of apples directly. In order to identify the bruising degrees of apples quickly and accurately, using hyperspectral imaging technology, this study proposed a method combining successive projections algorithm (SPA) with support vector machine based on grid search parameter optimization (GS-SVM) to classify and identify apple samples with different degrees of bruising. In the process of research, firstly, random forest method was used to extract spectral data of bruised areas of apples with a high accuracy, then Kennard-Stone algorithm was performed to partition sample set reasonably to improve performance of the model. After comparing different pretreatment methods, standard normal variate (SNV) transformation method with the best performance was selected to process the spectral data. Finally, in order to reduce the time required to build the model, the GS-SVM models were set up based on the characteristic variables selected by SPA and competitive adaptive reweighted sampling (CARS), and the classification results were compared with the results of the model constructed with the full spectra. The experimental results showed that the SNV-SPA-GS-SVM model had the best prediction effect, and the prediction accuracy of apples with four kinds of bruising degrees was 95%.