Almond kernel variety identification and classification using decision tree

Nowadays, an identification system is needed in agriculture and food processing industries to boost the efficiency of production to meet international standards. The manual approach is used in product grading and quality control. Unfortunately, it leads to uneven products, higher time expenses, and fatigue by human operators. So, quality assessment is one of the significant factors for products and a great impact on the final prices. In this study, we have proposed an image processing and computational intelligence-based method for identifying and classifying almond varieties as Nonpareil (NP), Mission (MI), Carmel (CR), and California (C). The scanner is used to obtain a kernel image for 2000 samples of almond. The proposed system involves four stages, they are pre-processing (used a median filter to eliminate noise and Otsu threshold algorithm used for segmentation), feature extraction (total 66 features- 4 for geometric; 10 for shape; 37 for colour, and 15 for texture), feature selection and reduction (principal component analysis-PCA and modified sequential floating forward selection-MSFFS), and classification (decision tree). We used three strategies in classification, they are strategy-1 (considered whole features set without feature selection and reduction applied to DT), strategy-2 (considered whole features set with PCA), and strategy-3 (considered whole features set with MSFFS). Overall accuracy is obtained from DT as 80.8% for strategy-1, 90.8% for strategy-2, and 97.13% for strategy-3. Among all, strategy-3 (DT with MSFFS) is outperformed for the classification of almonds kernel variety. The developed method can be easily extended to online sorting machines.