基于卷积神经网络的植物病害识别技术

Recognition Technology of Plant Disease Based on Convolution Neural Network

针对植物叶片的病害识别问题,设计一种基于卷积神经网络的识别方法,针对植物前景与背景分割问题,设计一种新的灰度变换方法,之后使用Ostu法与基于Sobel算子混合切割的方法成功将植物叶片的前景与背景分离,针对38种植物病害对Alex Net进行改进,使用19000张预处理完成后的图片对神经网络迭代30次以后,在植物病害数据集上的识别率达到98.4375%,可以为卷积神经网络在植物病害识别做一个参考。

Designs a recognition method based on convolution neural network to identify diseases of plant leaves.Aiming at the problem of plant foreground and background segmentation,designs a new method of gray scale transformation,after that,the Ostu method and Sobel based hybrid cutting method are used to separate the foreground of plant leaves from the background.The AlexNet model is improved according to the 38 plant diseases,after 19000 preprocessed images,the neural network is iterated for 30 times.The recognition rate on plant disease data sets reaches 98.4375%,this can be used as a reference for the recognition of plant disease by convolution neural network.