基于级联卷积神经网络的作物病害叶片分割

针对传统卷积神经网络在作物病害叶片图像中分割精度低的问题，提出一种基于级联卷积神经网络（Cascade Convolutional Neural Network，CCNN）的作物病害叶片图像分割方法。该网络由区域病斑检测网络和区域病斑分割网络组成。基于传统VGG16模型构建区域病斑检测网络（Regional Detection Network，RD-net），利用全局池化层代替全连接层，由此减少模型参数，实现叶片病斑区域精确定位。基于Encoder-Decoder模型结构建立区域分割网络（Regional Segmentation Network，RS-net），并利用多尺度卷积核提高原始卷积核的局部感受野，对病斑区域精确分割。在不同环境下的病害叶片图像上进行分割实验，分割精度为87.04%、召回率为78.31%、综合评价指标值为88.22%、单幅图像分割速度为0.23?s。实验结果表明该方法能够满足不同环境下的作物病害叶片图像分割需求，可为进一步的作物病害识别方法研究提供参考。

Crop Diseases Leaf Segmentation Method Based on Cascade Convolutional Neural Network

In order to solve the problem of low segmentation accuracy of traditional convolution neural network in crop disease leaf image, a method of crop disease leaf image segmentation based on cascade convolution neural network is proposed.The network is composed of regional disease spot detection network and regional disease spot segmentation network. The regional disease detection network is constructed based on the traditional VGG16 model, and the global pooling layer is used to replace the fully connection layer, so as to reduce the model parameters and realize the accurate location of leaf disease spot. The region segmentation network is established based on the structure of encoder-decoder model. The local receptive field of the original convolution nucleus is improved by using the multi-scale convolution kernel, and the disease region is segmented accurately.The segmentation experiments are carried out on the diseased leaf images in different environments, the segmentation accuracy is 87.04%, the recall rate is 78.31%, the comprehensive evaluation index value is 88.22%, and the segmentation speed of single image is 0.23 s.The experimental results show that this method can meet the needs of leaf image segmentation of crop diseases in different environments, and can provide reference for further research on crop disease identification methods.