Research On Image Segmentation And Its Application Of Plant Disease Leaves Under Complex Environment

Plant diseases are one of the important factors affecting the yield and quality of agricultural products.It has been found that many diseases of plants mainly occur on the leaves,and the types of diseases can be directly judged by observing the leaves of plants.Traditional plant disease identification is mainly based on artificial judgment.This method has subjectivity,blindness and inefficiency.With the continuous development of computer vision technology,the use of computer vision to intelligently identify plant diseases has become an effective means.In the process of using computer vision to identify diseases,accurate segmentation of diseased areas in plant leaves is a key technology for plant disease identification.In recent years,deep convolutional neural networks have achieved outstanding results in the field of computer vision.Among them,image segmentation methods based on Convolutional Neural Networks(CNNs)are widely used to solve multiple computer vision problems such as medical image segmentation,autopilot and remote sensing image classification.For different research fields,many researchers have proposed multiple image segmentation models such as FCN,SegNet,U-Net and DeepLabV3.However,when the existing image segmentation model is applied to the image segmentation of plant disease leaves,there are problems such as long model training time,poor segmentation effect and high model complexity.In order to effectively solve the image segmentation problem of plant diseases,this study proposes Multi-Scale Residual Convolutional Neural Networks(MSR-CNNs)and multiscale multi-column convolutional neural networks(MSMC-CNNs)based on existing image segmentation networks.MSMC-CNNs are used to solve the image segmentation of plant diseases and the image segmentation of plant diseases in different environments.The MSRCNNs mainly includes two parts: an Encoder Network and a Decoder Network.The encoder network includes structures such as Convolutional Layers,Pooling Layers,and Activation Layers.The encoder network introduces Deep Residual Structure,which is used to enhance the feature extraction capability of the network model.The MSMC-CNNs consists of a Multi-Column Encoder Network structure and a multi-scale decoder network.In the multi-column coding network structure,convolutional layers of different scales are cascaded using Skip Connection operations,which can effectively enhance the transmission of feature information in the convolutional layer.In the multi-scale decoding network structure,a Nine-Point Linear Interpolation algorithm is used as Deconvolution Layers to restore the feature map resolution of the input image.In the course of the experiment,this study proposed Multi-Scale Residual Convolutional Neural Networks(MSR-CNNs)and Multi-Scale Multi-Column Convolutional Neural Network(MSMC-CNNs),the two image segmentation models for plant disease leaves can effectively complete the segmentation of lesions in plant disease leaf images.Among them,MSR-CNNs are mainly used for leaf segmentation of plant diseases in complex environments and MSMC-CNNs are used to segment leaves of different degrees of plant diseases.Moreover,the experimental results show that the proposed two network models are superior to existing convolutional neural network based image segmentation methods in segmentation accuracy and model training time.