Applying Texture Feature Based on Local Ternary Pattern for Binary Particle Swarm Optimization plus Support Vector Machine-based Classification of Rice Varieties

Abstract

In this study, a proposed method for classification of seventeen rice varieties planted in Vietnam was presented. Features of rice grains were extracted based on an improved local ternary pattern (ILTP). To enhance the classification accuracy and decrease the number of used features, a classifier was built by combining binary particle swarm optimization (BPSO) with support vector machine (SVM). The experiment of classification for the seventeen rice varieties achieved the overall accuracy of 95.06% for BPSO+SVM method. The result showed that BPSO+SVM method could enhance the classification accuracy to 3.12%, and decrease the number of used features by 67.75%, compared to SVM alone. In addition, the extended ILTP features gave the classification accuracy higher than feature set of our previous research. This result could be developed for applications of automatic rice varieties classification.