基于多特征降维的植物叶片识别方法

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。

Method of Leaf Identification Based on Multi-feature Dimension Reduction

The identification of plant species is an essential part of botanical study and agricultural production. However, low dimension features cannot describe the leaf information. Thus, it cannot differentiate varieties of plants, and the accuracy is low. A method of plant species identification was proposed based on multi-feature dimension reduction. Firstly, color images of plant leaves were preprocessed by the digital image processing technology. The binary image, gray scale image and texture image without the petiole, wormhole and background were obtained after the preprocessing. Secondly, geometric characteristics and structural characteristic were extracted from the binary image. Hu moment invariants features, gray level co-occurrence matrix features, LBP features and Gabor features were extracted from the gray scale image. The fractal dimension was extracted from the texture images and 2183 features were extracted to describe leaf samples in number. Thirdly, the method of combining principal component analysis (PCA) and linear discriminant analysis (LDA) was adopted to reduce the feature dimension. Then the feature data of training samples was adopted to train the support vector machine classifier. Finally, the support vector machine classifier was used to classify the feature data of test samples. The experiments were carried out on Flavia database and ICL database. The average accuracy was 92.52% and 89.97%, respectively. The experiments showed that the average accuracy of the proposed method was better than that of the compared researches.