Tree Species Classification Using Leaf and Tree Trunk Images

摘要

The classification of tree species is an important task in many fields such as ecological monitoring, natural conservation and tourism. The classification with the use of digital cameras would be useful since we do not have to check in the plant books and it is not time-consuming. Many efforts have been made for the species classification with images with hand-crafted features or machine learning based methods. Those methods succeed in the classification with a high accuracy. However, the previous studies used a single still image although tree species can be estimated accurately when considering the leaf and tree trunk simultaneously. Therefore, we classified the tree species with the image of leaf and tree trunk image. The number of species classified in this study was 20 and over 4000 images were collected by the authors. The features were extracted with a pre-trained network and the extracted features of leaf and tree trunk images were concatenated, then the features were fed into support vector machine. Our result showed that the classification accuracy with the multi input was significantly higher than that only with leaf or tree trunk image. The method considering multi images would be useful in other tasks in remote sensing.