摘要
针对传统的作物病害识别方法中人为提取的分类特征,对复杂作物病害图像的形状、光照和背景比较敏感等问题,提出一种基于物联网和深度卷积神经网络(DCNN)的冬枣病害识别方法。DCNN由1个输入层、4个卷积层、3个下采样层、1个全连接层和1个输出层组成。利用该方法能够提取冬枣病害图像的有效特征,并识别病害类型,避免了传统作物病害识别方法中繁琐的特征提取过程。在4种冬枣病害果实数据库上进行了冬枣病害识别实验,识别率达到92%以上。试验结果表明,该方法适合利用物联网采集的大规模视频病害图像进行冬枣病害识别。
Focusing on the problem of traditional crop disease recognition methods that the artificially designed features are more susceptible to the crop disease image shapes,illumination and background,a recognition method of jujube disease was proposed based on the internet of things and deep convolutional neural network（ DCNN）. The network model was composed of input layer,4 convolutional layers,3 down-sampling layers,fully-connection layer and output layer. The proposed method can extract effective features of winter jujube disease image and recognize the diseases,avoiding the complicated feature extraction process of the traditional crop disease method. The proposed method was verified on the winter jujube fruit disease database,and the recognition rate was above 92%. The experimental results showed that the proposed method was suitable for winter jujube disease recognition on the large-scale disease database collected by internet of things.
出处
《浙江农业学报》
CSCD
北大核心
2017年第11期1868-1874,共7页
Acta Agriculturae Zhejiangensis
基金
国家自然科学基金项目(61473237)
陕西省自然科学基础研究计划(2016GY-141)
关键词
冬枣病害识别
冬枣病害图像
深度卷积神经网络(DCNN)
特征提取
winter rujube disease recognition
winter rujube disease image
deep convolutional neural network
fea-ture extraction
