Varietal Classification of Rice seeds using ML Models

Abstract

The work primarily addresses classifying a diverse set of 90 rice seed varieties. The classi fication process involves leveraging spatial and color data obtained from RGB images and extracting spectral information from hyperspectral images. This study integrated spatial, color, and spectral information, and subsequently, Linear Discriminant Analysis (LDA) was applied to reduce its dimensionality. For the classification task, three machine learning mod els, namely Support Vector Machine (SVM), K-Nearest Neighbour (KNN), and Random For est classifier (RFC) were employed. An investigation was conducted to compare the results obtained from the original raw data with those obtained from the preprocessed data. Prepro cessing techniques, specifically Standard NormalVariate(SNV)andSavitzky-GolaySmooth ing (SG-Smoothing), employed to preprocess spectral information extracted from the hyper spectral images (HSI). The SVM classifier demonstrated the highest F1-score of 92.84% when employing the fusion of multiple features from the raw data, including spectral, spatial, and color information. Subsequently, the models were evaluated using preprocessed data, where only the spectral features underwent preprocessing. The SVM classifier, combined with SG smoothing preprocessing, outperformedothermodelswithanimpressiveF1-scoreof92.96% and testing accuracy of 92.99%. Interestingly, the findings indicate that preprocessing the spectral information had minimal influence on enhancing the model’s performance.