基于深度学习的马铃薯病害智能识别

Intelligent Identification of Potato Diseases Based on Deep Learning

针对现有通过叶片病斑识别马铃薯病害的方法中,只对简单背景下的单片叶片进行病害识别处理,难以应用于真实复杂环境的问题,提出了一种基于深度学习的现场环境下马铃薯病害智能识别的方法。首先采用Deeplab v3+语义分割网络在生长背景中分割出马铃薯叶片,然后使用自适应对比度增强和颜色空间转换的方法提取马铃薯病斑,最后结合病斑的纹理特征和VGG16网络提取的特征,通过搭建一维卷积神经网络识别出病害。实验结果表明,上述方法能准确有效地识别复杂背景下的马铃薯病害。

In view of the existing methods of identifying potato diseases by leaf spot,it is difficult to apply to a real and complex environment,and only a single leaf under simple background is identified and processed.A method of intelligent identification of potato diseases in a field environment based on deep learning is proposed.First,the Deeplab v3+semantic segmentation network was used to segment the potato leaves in the growing background,and then the potato lesions were extracted using the method of adaptive contrast enhancement and color space conversion.Finally,combining the texture features of the lesions and the features extracted by the VGG16 network,the disease was identified by building a one-dimensional convolutional neural network.Experimental results show that this method can accurately and effectively identify potato diseases in a complex background.