Neural networks based leaf identification using shape and structural decomposition

Plants are sine-qua-non for existence of human life. The benefits provided by plants are manifold. Plant identification is challenging but it is extremely useful for making accurate decisions regarding livestock systems, conservation and ecology. Though most plants may look similar, they might not be the same. Hence, it becomes essential to develop a system which will identify plants by studying the uniqueness they exhibit at a granular level. Leaves serve as unique identifiers for plants. Hence, physical characteristics demonstrated by leaves act as fingerprints for their identification. This paper explores and proposes leaf identification method based on color, shape, structural and textural characteristics of leaves. These features are then subjected to three different well-trained Neural Network architectures which categorically classify the leaves. A performance comparison is made between the architectures by assessing their test set classification accuracies on the leaf image database considered in this study.