A Hierarchical Approach for Multi-class Classification of Leaf Images

Plants are an essential factor in biodiversity. For biodiversity conservation and a wholesome ecosystem, plant recognition is the necessary move to aware people. Various studies in literature proved it as a primary step. In this article, we are proposing a hierarchical framework for plant recognition using Support Vector Machine and Similarity clustering. The idea not only reduces the computational cost of classifying any new sample but also provide significant accuracy. The similarity clustering method provides significant clusters that maximize the classification area between each group and best suited to the maximum margin notion of Support Vector Machine. It also overcomes the problem of imbalanced class learning. For comparative analysis, a hierarchical approach is compared with traditional Support vector Machine. Experiments are performed using the Flavia Database with different subsets of varying training and testing ratio. It is observed that the proposed approach is efficient and robust.