Alpine Plants Recognition with Deep Convolutional Neural Network

In this study, the ultimate goal is to build a system which identifies species of alpine plants from pictures. In this paper, in order to build such a system, its fundamental recognition part is constructed using a Deep Convolutional Neural Network (DCNN). A lot of recent studies reveal that DCNNs show excellent performance for recognition tasks by acquisition of feature representations from raw data. However, it is necessary to prepare sufficient number of data for obtaining good feature representations. Especially for alpine plants, this is rather difficult because of their habitat. In this paper, we add images of plants other than alpine ones, and examine how such supplementary data have influence on the recognition accuracy for the target domain, i.e., alpine plants. Experimental results show the effectiveness of using supplementary images for alpine plants recognition.