Intelligent Grading of Green Cardamom Using Data Fusion of Electronic Nose and Computer Vision Methods

Abstract

In this research, the intelligent quality grading of green cardamom was carried out using electronic nose (e-nose) and computer vision (CV) methods along with machine learning (ML) approaches. Cardamom samples were analyzed in three grades including Grade 1 (healthy and green), Grade 2 (healthy with yellow color), and Grade 3 (immature and shriveled) for capsules and Grade 1 (Black), Grade 2 (Brown), and Grade 3 (Yellow and red) for seeds. Three ML algorithms including Decision Tree (DT), Bayesian Network (BN), and Support Vector Machine (SVM) were used to classify the quality grades. Results showed that the correlation-based feature selection (CFS) algorithm decreased the number of input features and increased the classification performance. For classifying cardamom capsule samples based on the visual features, the CFS-BN model was the best classifier, with the root mean squared error (RMSE) and accuracy of 0.1408 and 96.67%, respectively. The RMSE and accuracy of this model for classifying cardamom seeds based on image features were 0.1220 and 96.67%, respectively. In classifying cardamom seeds using e-nose data, the CFS-DT model was the best classifier with RMSE and accuracy of 0.2093 and 93.33%, respectively. The CFS-BN model was the best for classifying cardamom capsules with an RMSE of 0.1126 and an accuracy of 96.67%. The fusion of e-nose and CV data increased the model performance compared to the separate use of e-nose and CV datasets. The accuracy of the CFS-BN model using the combination of CV and e-nose data was 100% during both the calibration and evaluation stages. It can be concluded that data fusion of e-nose and CV methods can be effectively used to develop an intelligent, accurate, reliable, fast, and non-destructive system for quality grading of cardamom capsules and seeds.