A novel approach to authentication of highbush and lowbush blueberry cultivars using image analysis, traditional machine learning and deep learning algorithms

Abstract The objective of this study was to classify blueberry cultivars based on image texture parameters using models built using traditional machine learning and deep learning algorithms. The blueberries belonging to highbush cultivars (‘Bluecrop’, ‘Herbert’, ‘Jersey’, and ‘Nelson’) and lowbush cultivars (‘Emil’ and ‘Putte’) were subjected to imaging using a digital camera. The texture parameters from blueberry images in color channels R , G , B , L , a , b , X , Y , Z , U , V , and S were determined. After selection image textures were used to build models for the classification of all highbush and lowbush blueberry cultivars, and highbush blueberry cultivars and lowbush blueberry cultivars, separately. In the case of distinguishing all cultivars, such as ‘Bluecrop’, ‘Herbert’, ‘Jersey’, and ‘Nelson’, ‘Emil’ and ‘Putte’, the classification accuracy reached 92.33% for a model built using a deep learning algorithm. Models built to distinguish only highbush cultivars provided an average accuracy of up to 91.25% (WiSARD). For models developed to classify two lowbush cultivars, an average accuracy reaching 96% (WiSARD) was found. The applied procedure can be used in practice to distinguish blueberry cultivars before their consumption or processing.