Abstract: Plant classification based on leaf image is an important research area in plant taxonomy. Because the leaf image is complex and is sensitive to the season and illumination, the classification results of the existing plant classification methods are not robust. Based on the dictionary learning with sparse representation, a plant classification method was proposed in this paper. The plant classification problem was transformed to solve the sparse representation problem of the test sample to the training samples. A small optimal over-complete dictionary was designed to calculate the sparse representation of the leaf image by using the class-specific dictionary learning. Comparing to the other methods, the proposed method didn't need to extract the features of color, texture and shape of the leaf image. So the computing complexity was reduced and the robustness and the real-time performance of the automatic identification of plant were improved. The experimental results on the real-world database of 50 kinds of the plant leaf images showed the feasibleness of the proposed algorithm. The recognition rate was more than 92%.

Key words: plant classification, leaf image, sparse representation, dictionary learning