Recognition model for Costa Rican plant species identification

Abstract

In the last decade, research in Computer Vision has developed algorithms to help botanists and non-experts classify plants based on images of their leaves. Nevertheless, very few efficient tools have resulted from that research and have actually been used in the field. The most popular system to date is LeafSnap. It is considered a state-of-the art leaf recognition mobile application. It uses a multi scale curvature model of the leaf margin to classify leaf images into species. LeafSnap was applied to 184 tree species from Northeast US and achieved high levels of accuracy for that group of trees. In this document, we extend the research that led to the development of LeafSnap along several lines. First, LeafSnap’s underlying algorithms are applied to a set of species from Costa Rica. Then, texture is used as an additional criteria in order to improve the level of accuracy of LeafSnap’s original algorithms. Thus, the main goal of this research is to measure the level of improvement in automatic Costa Rican tree species identification achieved when texture analysis is added to the curvature model of margins of leaves. Our results confirm our hypothesis since the level of improvement reaches a 0.168 for the Costa Rican clean subset, and 0.431 for the Costa Rican noisy subset. In both cases, our results show this increment as statistically significant.