Bi-level classification using naive bayes classifier and gaussian line-seeker method on rice leaf images

In this study, supervised and contextual classification are combined for a bi-level model of categorizing rice leaf pixels according to three main clusters: vein, mesophyll cells and unknown (chlorophyll concentration and other parts of the leaf). Gaussian Line-Seeker Method (GLSM) performs first-level classification based on neighboring pixels. Higher- intensity pixels are identified as candidates for further classification. These candidate pixels are subjected to Naive Bayes Classifier. Hue, saturation and intensity are the independent predictors used to construct the posterior probability of class membership. This bi-level model is developed to facilitate an automated screening process for the C4 Rice Project. It is a pro ject of the International Rice Research Institute (IRRI) requiring the screening of more than 100,000 leaf samples of rice varieties and cultivars. The veins were classified with an average precision of 75.07%, 71.99% for mesophyll cells and 98.45% for the unknown cluster.