马铃薯典型病害图像自适应特征融合与快速识别

发布时间：2018-03-01 18:02

本文关键词： 马铃薯典型病害 Hough变换主成分分析加权融合支持向量机　出处：《农业机械学报》2017年12期 　论文类型：期刊论文

【摘要】：针对自然条件下马铃薯典型病害区域定位和识别难的问题,提出了一种马铃薯典型病害图像的自适应特征融合与快速识别方法。该方法利用K-means、Hough变换与超像素算法定位叶片,结合二维Otsu与形态学法分割病斑区域,通过病斑图像颜色、形状、纹理的自适应主成分分析(PCA)特征加权融合,进行支持向量机(SVM)病害识别。对3类马铃薯典型病害图像进行识别试验,结果表明:SVM识别模型下,自适应特征融合方法相比PCA降维、特征排序选择等传统自适应方法,平均识别率至少提高了1.8个百分点;13个自适应融合特征下,识别方法平均识别率为95.2%,比人工神经网络、贝叶斯分类器提高了3.8个百分点和8.5个百分点,运行时间为0.600 s,比人工神经网络缩短3 s,可有效保证识别精度,大大加快了识别速度。
[Abstract]:An adaptive feature fusion and fast recognition method based on K-means-Hough transform and super-pixel algorithm is proposed to locate the leaves of potato typical diseases. Combining two-dimensional Otsu and morphological method to segment the disease spot region, the adaptive principal component analysis (PCA) method of image color, shape and texture is used for weighted fusion. Three kinds of typical potato disease images are identified by using support vector machine (SVM). The results show that the adaptive feature fusion method is better than the traditional adaptive methods such as PCA dimension reduction, feature ranking selection and so on. The average recognition rate is at least 1.8 percentage points higher than that of the artificial neural network, and the average recognition rate of 13 adaptive fusion features is 95.2 percentage points, which is 3.8 percentage points and 8.5 percentage points higher than that of the artificial neural network and Bayesian classifier. The operating time is 0.600 s, which is 3 s shorter than that of artificial neural network, which can effectively guarantee the recognition accuracy and greatly accelerate the recognition speed.
【作者单位】：内蒙古工业大学电力学院;
【基金】：国家自然科学基金项目(61661042) 内蒙古自治区自然科学基金项目(2015MS0617)
【分类号】：S435.32;TP391.41

【相似文献】