摘要
针对叶缘叶裂明显的植物叶片识别问题,提出一种基于叶片形状特征的识别方法。该方法首先使用阈值分割与形态学操作对叶片进行二值化处理;然后从二值图像上提取了8种描述叶片形状的特征,经过对8维特征的皮尔森相关系数分析与主成分分析,确定对分类贡献最大的5个主成分,考虑到形态特征与分类结果间存在非线性相关关系;最终的分类器使用了BP神经网络。采用UCI数据集的实验表明,该方法对15种植物171张叶片的综合识别率为89.1%,其中,对叶缘叶裂较为明显的6种植物70张叶片识别率达95%以上。
A recognition method based on shape features was proposed for the recognition of plant leaves whose leaf margin and leaf lobes were more obvious. In this method, threshold segmentation and morphological operation were used to perform a binarization process to leaves. Then, 8 kinds of features that describe the shape of leaves were extracted from binary images. After Pearson correlation coefficient analysis and principal components analysis of the 8 features, the 5 principal components of the largest contribution to the classification were determined. The final classifier applied the BP （ Back Propagation） neural network, given the non-linear correlations between morphological characters and classification results. Experiments on UCI data set show that the comprehensive recognition rate with 171 leaves of 15 plants of this method is 89. 1%; among them, the recognition rate of 70 leaves of 6 species with more obvious leaf margin and leaf lobes is above 95%.
出处
《计算机应用》
CSCD
北大核心
2016年第A02期200-202,226,共4页
journal of Computer Applications
基金
国家自然科学基金青年基金资助项目(61502060)
关键词
叶片图像
特征提取
叶片识别
主成分分析
BP神经网络
leaf image
feature extraction
teaf recognition
principal component analysis
BP neural network
