Feature Fusion for Leaf Image Classification

Moeri Okuda, Hiroaki Ohshima

Abstract: In this paper, we tackle the problem of classifying tree species from leaf images with deep learning. First, we classify leaf images with deep learning, focusing on a single leaf feature: only the whole-leaf feature, only the leaf-shape feature, or only the leaf-vein feature. The leaf feature that contributes to the classification accuracy differs depending on the species. Second, we classify leaf images by combining the whole-leaf, the leaf-shape, and the leaf-vein feature. We define combining these features as feature fusion. After feature fusion, the classification performance improves to 92.07%. It is important for people to teach which leaf features to focus on.

External IDs:dblp:conf/bigcomp/OkudaO22