李衡霞,龙陈锋,曾蒙,申佳.一种基于深度卷积神经网络的油菜虫害检测方法[J].湖南农业大学学报:自然科学版,2019,45(5):.
一种基于深度卷积神经网络的油菜虫害检测方法
  
DOI:
中文关键词:  油菜害虫  检测  深度卷积神经网络  VGG16网络
英文关键词:rape pest  identification  deep convolutional neural network  VGG16 network
基金项目:国家自然科学基金项目(61772031);湖南省长株潭国家自主创新示范区专项(2017XK2054);湖南省教育厅优秀青年项目(12B061);湖南农业大学双一流建设项目(SYL201802002)
作者单位
李衡霞,龙陈锋,曾蒙,申佳 1.湖南农业大学信息科学技术学院湖南 长沙 4101282.湖南省农村农业信息化工程技术研究中心湖南 长沙 410128 
摘要点击次数: 65
全文下载次数: 25
中文摘要:
      针对目前油菜虫害识别在背景、角度、姿态、光照等方面的鲁棒性问题, 提出一种基于深度卷积神经网络的油菜虫害检测方法:首先在卷积神经网络和区域候选网络的基础上,构建油菜虫害检测模型,再在深度学习tensorflow框架上实现模型的检测,最后对比分析结果。油菜虫害检测模型利用VGG16网络提取油菜虫害图像的特征, 区域候选网络生成油菜害虫的初步位置候选框,Fast R–CNN实现候选框的分类和定位。结果表明, 该方法可实现对蚜虫、菜青虫(幼虫)、菜蝽、跳甲、猿叶甲5种油菜害虫的快速准确检测, 平均准确率达94.12%,与RCNN、Fast R–CNN、多特征融合方法、颜色特征提取方法相比,准确率分别提高了28%、23%、12%、2%。
英文摘要:
      Aiming at the robustness of rape pest identification methods in terms of background, angle, posture and illumination, a method was proposed based on deep convolutional neural network to detect rape pests. Firstly, the detection model of rape pest was constructed on the basis of convolutional neural network and region proposal network. Secondly, the model was tested on the deep learning tensor flow framework. Finally, the experimental results were compared and analyzed. Based on convolutional neural networks and regional candidate networks, a rape pest detection model was constructed by using the VGG16 network to extract the features of the rape pest image, the region proposal network to generate the preliminary position candidate box of the rape pest, and the Fast R-CNN to realize the classification and localization of the candidate box. The results showed that the method can quickly and accurately detect five species of rape pests such as aphids, caterpillars (larvae), dish, jumping and leaf, with the average accuracy rate of 94.12%. Compared with the RCNN, Fast R-CNN, multi-feature fusion method and color feature extraction method, the accuracy rate of new method improved 28%, 23%, 12%, and 2%, respectively.
查看全文  查看/发表评论  下载PDF阅读器
关闭
Copyright © 2008 《湖南农业大学学报》自然科学版 地址:湖南 长沙市 芙蓉区 
邮编:410128 电话:0731-84638380 0731-84618035 0731-84617599 0731-84673730 E-mail: zkb4618035@hunau.net  京ICP备09084417
分享按钮