Comparative evaluation of computer vision and electronic nose technologies for quality classification of cardamom powder using machine learning approaches

This study evaluated and compared the potential of electronic nose (E-nose) technology and computer vision (CV) coupled with machine learning algorithms for quality grading of cardamom powder. Nine distinct classes of powder were formulated by combining seeds from three cardamom categories (black, brown, and yellow). Images and E-nose signals of cardamom powder samples were acquired for quality analysis. From each image, 52 color and texture descriptors were extracted, while 54 features were derived from aroma sensor responses. To reduce dimensionality, two feature selection strategies—Correlation-based Feature Selection (CFS) and Principal Component Analysis (PCA)—were applied, yielding 14 optimal variables for CV data and 17 for E-nose data. The refined datasets were classified using four algorithms: Multilayer Perceptron Neural Network (MLPNN), Support Vector Machines (SVM), Random Forest (RF), and Bayesian Networks (BN). For CV data, the CFS-driven MLPNN with eight hidden neurons achieved the best performance, with a test classification accuracy of 97.78 %, a precision of 0.981, a recall of 0.978, and an F1-score of 0.978. For E-nose data, the CFS-based RF classifier was superior, attaining 93.33 % accuracy and a precision of 0.944 during the test stage. These findings demonstrate that both visual and olfactory feature extraction combined with artificial intelligence enable the development of accurate, rapid, and non-destructive grading frameworks for cardamom powder quality assessment, with CV showing particular promise for online applications.