Abstract: A recognition method of multi-feature plant leaves based on dissimilarity-supervised locally linear embedding algorithm is proposed to improve the recognition accuracy in plant leaves recognition. The features of color, shape and texture are extracted as leaves multi-feature, and the sample dissimilarity is brought into weighted locally linear embedding to form the supervised LLE algorithm to reduce leaves multi-feature dimension; the nearest classifier is used to recognize leaves category in low dimension space. The leaves multi-feature are better than pixels to describe leaves, at the same time, dissimilarity can mine sample category information fully. The experimental results based on real plant leaf databases show that the proposed method improves the leaves recognition accuracy effectively.

Key words: recognition, leaves multi-feature, supervised locally linear embedding, weighted locally linear embedding, dimension reduction, dissimilarity

摘要： 为了提高植物叶片图像识别的准确率，提出一种基于差异性值监督局部线性嵌入（D-LLE）算法的多特征植物叶片图像识别方法。该方法提取叶片的颜色、形状和纹理作为叶片多特征，在加权局部线性嵌入（WLLE）算法中引入样本的差异性值构成差异性值监督LLE算法（D-LLE）对叶片高维特征进行降维，在低维空间采用最近邻分类器实现叶片的识别。该方法所用的叶片多特征比单一特征像素值更能描述叶片图像，同时差异性值能够充分挖掘样本的类别信息。基于实拍的叶片图像数据库的实验结果表明，该方法有效提高了叶片的识别精度。

关键词: 识别, 叶片多特征, 监督局部线性嵌入, 加权局部线性嵌入, 降维, 差异性值