Research Of Key Algorithms Of Crop Disease Recognition Based On Image

The diagnosis and identification of crop disease is of great significance for improving crop quality.The application of image processing and machine vision technology to crop disease identification is superior to the traditional artificial diagnosis and recognition method,and improves the ability of crop disease monitoring and early warning.Through comprehensive analysis of the complex factors(such as leaf shade,leaf twist and deformation,different climatic conditions,different light conditions,shadow,other plants and soil),the growth and development of agricultural crops are analyzed and excavated in the field.The image of crop disease in complex environment has the characteristics of huge background information,changeable disease characteristics,high disease association degree and high disease complexity.Based on the physiological characteristics of crop disease,image processing technology and bioinformatics are combined to study the image preprocessing of crop disease under complex background.A series of key algorithms,such as disease segmentation,disease spot feature extraction and disease spot type identification,are used to establish the dynamic prediction and analysis model of crop disease identification,and monitor the crop growth process in real time intelligently.The specific work is as follows:Firstly,according to the image of crop disease collected under the simple background and complex background of the field,the preprocessing operations such as the occlusion,image fusion,edge detection and the image enhancement based on the equalization are carried out,and the image quality of the preprocessed images is evaluated.A new algorithm of disease image dis-occlusion based on thermograph is proposed.The experimental results of the dis-occlusion show that the disease blade area from the original disease image has not only removed the occlusion information,but also removes most of the background area,which makes the disease image smaller and ensures the integrity of the disease blade.Then,in the phase of the disease image segmentation under the complex background,the disease image segmentation method based on the significant graph is proposed in view of the disease images with complex background collected in the field environment.The salient map is obtained by using the significant graph detection strategy.The significant image is made as the mask image and the disease image is recovered.The results show that the disease image segmentation method,which is combined with Grab Cut,has strong anti-interference ability to the image noise,uneven illumination and inhomogeneous color of disease spots.It can completely separate most of the disease leaves,and can retain the details of the disease spots.The experimental results show that compared with the CNN-based segmentation algorithm,the Grab Cut segmentation algorithm based on saliency map is more suitable for image segmentation on fine-grained category datasets.Physiological characteristics of Chinese wolfberry leaf diseases and rice leaf disease were analyzed in the stage of feature extraction and optimization.The multidimensional features of the disease spot image are extracted,including color features,shape features and texture features.The main component analysis PCA method is used to optimize the features and reduce the PCA dimension.The feature vectors are then transformed by LDA projection to get the best classification features,which provides a reliable guarantee for the next classification and recognition.Aiming at the problem of insufficiently labeled disease image recognition,the crop disease recognition algorithm based on the discriminative deep confidence network is studied.In the experiments of classification and recognition on small-scale and medium-scale datasets,the algorithm shows stable,accurate and superior classification performance when there are not enough annotated images or only a few annotated images.And the classification accuracy,classification error rate,confusion matrix and ROC curve are used to evaluate the classification performance of the disease image recognition classifier proposed in this paper.Finally,in view of the problem of the large area crop image detection in the field,an automatic detection method of disease image based on Faster R-CNN is proposed which realizes the rapid positioning of diseased leaves.Based on the strategy of "first detection,then identification",a crop disease identification system is developed.It can be applied to disease identification and growth environment monitoring of Chinese wolfberry.The dynamic monitoring of the growth process and the automatic diagnosis of disease information were realized.In this paper,the key algorithm of crop disease image recognition based on complex background is studied,and a feasible solution for automatic identification and detection of crop diseases is proposed.Good experimental results are achieved on small-scale data sets of Chinese wolfberry disease images.This study also provides a new idea and method for intelligent detection and recognition of crop diseases,provides necessary information for farmers to prevent and treat diseases in time,and provides technical support for real-time and accurate diagnosis,identification and prediction of diseases.