| 基于SVM和DS证据理论融合多特征的玉米病害识别研究 |
| |
| 引用本文： | 毛彦栋,宫鹤.基于SVM和DS证据理论融合多特征的玉米病害识别研究[J].中国农机化学报,2020(4):152-157. |
| |
| 作者姓名： | 毛彦栋 宫鹤 |
| |
| 作者单位： | 吉林农业大学信息技术学院;吉林省智能环境工程研究中心;吉林省农业物联网科技协同创新中心 |
| |
| 基金项目： | 吉林省教育厅项目(20170204038NY);吉林省发改委项目(2014Y108);长春市科技局项目(12SF31)。 |
| |
| 摘 要： | 针对玉米叶部病害图像的颜色、纹理、形状特征对病害影响的差异性,提出一种结合单特征下的SVM识别准确率和识别结果的融合多特征玉米病害识别方法。首先对预处理后的玉米病害图片提取颜色、纹理、形状3种特征,对应每一种特征构建一个SVM分类器,结合3个SVM分类器的平均准确率和识别结果作为证据理论的3个证据,构建D-S证据理论的基本概率分配函数(BPA),最后根据D-S证据理论决策规则进行决策级融合,依据决策条件输出最终识别结果。结果表明,结合SVM识别准确率和识别结果来对玉米的灰斑病、弯孢菌叶斑病、锈病三种病害进行识别,准确率分别为95%,85%,100%,平均准确率为93.33%,该方法对玉米叶部病害的识别更准确和稳定。
|
| 关 键 词： | 多特征 病害识别 D-S证据理论 支持向量机 基本概率函数 |
Corn disease identification study based on SVM and DS evidence theory fusion multi-features |
| |
| Mao Yandong,Gong He.Corn disease identification study based on SVM and DS evidence theory fusion multi-features[J].Chinese Agricultural Mechanization,2020(4):152-157. |
| |
| Authors: | Mao Yandong Gong He |
| |
| Institution: | (College of Information Technology,Jilin Agricultural University,Changchun,130118,China;Jilin Province Intelligent Environmental Engineering Research Center,Changchun,130118,China;Jilin Province Agricultural Internet of Things Science and Technology Collaborative Innovation Center,Changchun,130118,China) |
| |
| Abstract: | Aiming at the difference of color,texture and shape characteristics of corn leaf disease images,a fusion multi-feature corn disease identification method combining SVM recognition accuracy and recognition results under single feature was proposed.Firstly,three kinds of features of color,texture and shape were extracted from the pre-treated corn disease images,and an SVM classifier was constructed for each feature.The average accuracy and recognition results of the three SVM classifiers were used as evidence for evidence theory.The basic probability distribution function(BPA)of DS evidence theory is constructed.Finally,the decision-level fusion is carried out according to the DS evidence theory decision rule,and the final recognition result is output according to the decision condition.The results showed that the SVM identification accuracy and recognition results were used to identify the gray spot,Curvularia leaf spot and rust diseases of maize,and the accuracy rates were 95%,85%and 100%,the average accuracy rate is 93.33%.respectively.The identification of leaf diseases is more accurate and stable. |
| |
| Keywords: | multiple features disease identification D-S evidence theory SVM BPA |
| 本文献已被 CNKI 维普 等数据库收录！ |
|
