基于稀疏表示字典学习的植物分类方法

Plant classification method based on dictionary learning with sparse representation

基于叶片图像的植物分类方法研究是植物分类学的一个重要研究方向。由于叶片图像的复杂性和对季节、光照等条件比较敏感,使得现有的植物分类方法的分类效果不佳。该文提出了一种基于稀疏表示字典学习的植物物种识别方法,该方法将植物分类问题转化为求解待分类叶片图像对于训练样本植物叶片图像的稀疏表示问题;再利用面向植物叶片图像类别的字典学习,寻求一个较小的、并经过优化的超完备字典来计算待识别叶片图像的稀疏表示。与已有植物分类方法比较,该方法的创新点为直接对原始叶片图像进行处理,不需要从每幅叶片图像中提取颜色、纹理和形状等分类特征,从而极大降低了植物分类方法的复杂度,提高了分类方法的实时性和鲁棒性。在公开的植物叶片图像数据库中对50类植物叶片图像进行了分类实验,识别率高达92%以上。

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%.