基于S-WLLE算法和SVR的植物叶片图像识别方法  

An Exploration of Recognition Method of Plant Leaves Based on S-WLLE Algorithm and SVR

作　　者：杨利[1] 叶明全[1] 

机构地区：[1]皖南医学院计算机教研室,安徽芜湖241002

出　　处：《宿州学院学报》2014年第11期69-74,共6页Journal of Suzhou University

基　　金：国家自然科学基金面上项目"网络环境下数字图像可信性度量理论与方法研究"(61272540);安徽省高校省级自然科学研究重点项目"面向肿瘤基因表达数据的特征选择与集成分类研究"(KJ2014A266)

摘　　要：针对加权局部线性嵌入(Weighted Locally Linear Embedding,WLLE)算法不能充分挖掘样本类别信息以及传统流形学习算法中利用已有训练样本流形邻域关系近似得到测试样本低维嵌入的低精确性,提出了基于监督加权局部线性嵌入(Supervised Weighted Locally Linear Embedding,S-WLLE)算法和支持向量机回归(Support Vector Regression,SVR)的植物叶片图像识别方法。首先利用叶片样本监督距离代替WLLE算法中的欧式距离,对训练样本进行降维;然后学习训练样本已有数据得到SVR模型,预测测试样本的低维嵌入;最后利用最近邻分类器分别实现正负类样本以及负负类样本之间的识别。实验表明,该算法不仅提高了正负类叶片的识别精度,而且能够有效实现负负类叶片的识别。Weighted locally linear embedding algorithm can＇t mine adeqtely sample category information,and the low accuracy of traditional manifold learning algorithm uses existed manifold neighborhood of training sample to obtain test sample wity low embedding accuracy, this paper proposed a recognition method of plant leaves based on supervised weighted locally linear embedding algorithm and support vector regression. First, it used super- vised distance to replace sample Euclidean distance in WLLE to reduce training sample dimensions. Then, it learned the training sample data to form SVR model to predict test sample＇s low dimension embedding. Finally, it recognized leaves between positive and negative and between negative and negative by the nearest classifier. The experimental results show that the proposed method not only improves the leaves classification accuracy be- tween positive and negative sort of the leaves but also distinguishes two negative sorts of the leaves effectively.

关 键 词：叶片识别 监督距离 加权局部线性嵌入 降维 支持向量机回归 

分 类 号：TP391.41[自动化与计算机技术—计算机应用技术]