Weedy Rice Classification Using Image Processing and a Machine Learning Approach

Weedy rice infestation has become a major problem in all rice-growing countries, especially Malaysia. Challenges remain finding rapid technique to identify the weedy seeds that tend pose similar taxonomic and physiological features as cultivated seeds. This study presents image processing machine learning techniques classify seed variants A vision unit was set up for acquisition using an area scan camera Red, Green Blue (RGB) monochrome images of five varieties variant. Sixty-seven from RGB kernels were extracted three primary parameters, namely morphology, colour texture, used input learning. Seven classifiers used, classification performance evaluated. Analyses best model based on overall measures, such sensitivity, specificity, accuracy average correct described unbalanced dataset. Results showed optimum developed by logistic regression (LR) achieved 85.3% 99.5% 97.9% 92.4% utilising 67 features. In conclusion, this proved have higher sensitivity identifying than with selected colour, morphological textural