基于多特征降维的植物叶片识别方法

发布时间：2018-01-26 12:20

本文关键词： 叶片识别多特征降维主成分分析线性评判分析支持向量机　出处：《农业机械学报》2017年03期 　论文类型：期刊论文

【摘要】：植物种类识别方法主要是根据叶片低维特征进行自动化鉴定。针对低维特征不能全面描述叶片信息,识别准确率低的问题,提出一种基于多特征降维的植物叶片识别方法。首先通过数字图像处理技术对植物叶片彩色样本图像进行预处理,获得去除颜色、虫洞、叶柄和背景的叶片二值图像、灰度图像和纹理图像。然后对二值图像提取几何特征和结构特征,对灰度图像提取Hu不变矩特征、灰度共生矩阵特征、局部二值模式特征和Gabor特征,对纹理图像提取分形维数,共得到2 183维特征参数。再采用主成分分析与线性评判分析相结合的方法对叶片多特征进行特征降维,将叶片高维特征数据降到低维空间。降维后的训练样本特征数据使用支持向量机分类器进行训练。试验结果表明:使用训练后的支持向量机分类器对Flavia数据库和ICL数据库的测试叶片样本进行分类识别,平均正确识别率分别为92.52%、89.97%,有效提高了植物叶片识别的正确率。
[Abstract]:The method of plant species recognition is mainly based on the low-dimensional characteristics of leaves for automatic identification, aiming at the low-dimensional features can not fully describe the leaf information, the recognition accuracy is low. A method of plant leaf recognition based on multi-feature dimensionality reduction is proposed. Firstly, the color of plant leaves is preprocessed by digital image processing technology to get the removal of color and wormhole. The binary image of leaf blade, gray image and texture image of petiole and background. Then, the geometric and structural features of binary image, Hu invariant moment feature and co-occurrence matrix feature of gray image are extracted. Local binary pattern feature and Gabor feature are used to extract fractal dimension from texture image. Then the principal component analysis (PCA) combined with linear evaluation analysis was used to reduce the dimension of multi-feature of blade. The high dimensional feature data of blade is reduced to low dimensional space. The training sample feature data after dimensionality reduction is trained by support vector machine classifier. The experimental results show that:. The trained SVM classifier is used to classify and identify the test blade samples from Flavia database and ICL database. The average correct recognition rate was 92.52 and 89.97 respectively, which effectively improved the correct rate of plant leaf recognition.
【作者单位】：北京林业大学工学院;中国科学院电工研究所;
【基金】：中央高校基本科研业务费专项资金项目(2015ZCQ-GX-04) 北京市科技计划项目(Z161100000916012)
【分类号】：Q949;TP391.41
【正文快照】： 引言植物的种类识别是植物学研究和农业生产的基础性工作,传统方法主要是根据植物形态进行人工鉴定,识别结果存在主观性且效率较低。植物叶片含有大量种属信息,且大部分叶片外形为扁平状,因此可对植物叶片进行图像采集和处理,实现植物种属快速识别。目前,大多数研究采用少量特

【相似文献】