| 基于Spark的支持向量机在小麦病害图像识别中的应用 |
| |
| 引用本文： | 林中琦,牟少敏,时爱菊,孙肖肖,李磊.基于Spark的支持向量机在小麦病害图像识别中的应用[J].河南农业科学,2017,46(7). |
| |
| 作者姓名： | 林中琦 牟少敏 时爱菊 孙肖肖 李磊 |
| |
| 作者单位： | 1. 山东农业大学信息科学与工程学院,山东泰安,271018;2. 山东农业大学化学与材料科学学院,山东泰安,271018 |
| |
| 基金项目： | 山东省自然科学基金项目，2013年山东省农业重大应用技术创新课题 |
| |
| 摘 要： | 为了提高小麦病害图像分类的效率,提出了一种基于Spark的并行式支持向量机算法。首先对小麦病害图像进行滤波去噪、灰度压缩等处理,利用灰度共生矩阵、不变矩阵等从颜色、纹理和形状3个方面提取49个特征向量;然后通过数据集的切分和并行框架的支持,将大数据并行处理技术Spark与支持向量机结合,运用Scala语言实现了串行支持向量机算法的并行化,并将其应用于小麦病害图像识别。针对小麦锈病和白粉病的图像分类测试结果表明,当测试图像分别是2 600、3 900、5 120张时,该算法对锈病的分类精度依次是76.03%、81.18%、77.82%,对白粉病的分类精度依次是83.27%、85.91%、83.14%,比串行支持向量机分类精度有所提升。分类时间依次是13 928.0、18 506.1、24 897.2 ms,明显低于串行支持向量机的分类时间。改进的算法实现了小麦病害分类精度的小幅度提升,明显提高了处理速度,具有较快的学习收敛速率。
|
| 关 键 词： | 小麦病害 图像分类 Spark 支持向量机 大数据 并行计算 图像特征提取 |
Classification of Wheat Disease Images Using Parallelized Support Vector Machine Based on Spark |
| |
| LIN Zhongqi,MU Shaomin,SHI Aiju,SUN Xiaoxiao,LI Lei.Classification of Wheat Disease Images Using Parallelized Support Vector Machine Based on Spark[J].Journal of Henan Agricultural Sciences,2017,46(7). |
| |
| Authors: | LIN Zhongqi MU Shaomin SHI Aiju SUN Xiaoxiao LI Lei |
| |
| Abstract: | In order to improve the efficiency of image classification for wheat diseases,a parallelized support vector machine algorithm based on Spark was proposed.First of all,the wheat disease images were denoised by filtering and compressed at gray-scale.Gray level co-occurrence matrix and invariant matrix and others were used to extract 49 feature vectors from color,texture and shape.Secondly,we combined the Spark with support vector machine through the support of segmentation of data sets and parallel framework.Finally,Scala language was used to realize the parallel processing of single support vector machine,and it was applied in the recognition of wheat disease images.The experimental results on the image classification of wheat diseases showed that the classification accuracies of wheat leaf rust were 76.03％,81.18％,77.82％,and the classification accuracyies of powdery mildew were 83.27％,85.91％,83.14％,when the numbers of test images were 2 600,3 900 and 5 120,respectively.The classification accuracy had been improved compared with the single support vector machine.The classification times were 13 928.0 ms,18 506.1 ms,24 897.2 ms respectively,which were obviously lower than that of the single support vector machine.The improved algorithm could make the classification accuracy of wheat diseases get a small increasing while the processing speed get a obvious ascension.The proposed algorithm owns a faster convergence rate. |
| |
| Keywords: | wheat diseases image classification Spark support vector machine big data parallel computing image characteristics extraction |
| 本文献已被 CNKI 万方数据 等数据库收录！ |
|
