Research On Disease Identification Of Poplar Leaves Based On Disease Spot Enhancement And Feature Segmentation

.In recent years,the types of poplar diseases have increased,and the degree of infection has increased year by year.Accurate and rapid identification of disease types plays a vital role in preventing large-scale outbreaks of poplar diseases.In order to realize the rapid identification of common leaf disease types in poplar,a leaf leaf disease identification scheme is proposed in this thesis based on CNN and SVM,and designs and implements a processing algorithm that enhances disease characteristics to obtain disease feature maps to identify Accuracy and recognition speed.The specific research content and results are as follows:(1)According to the different background composition factors of the poplar leaf image,the image to be divided into leaf images under a single background and complex background,and different algorithms are designed to segment and extract the main image of the leaf.The improved contour extraction method based on Canny is used to extract the foreground of leaf images under a single background.The results show that this method has better segmentation effect and faster speed,and is suitable for the extraction of leaf images based on a single background.An interactive segmentation algorithm based on Grab Cut is used to extract the image of the main part of the leaf under a complex background.The results show that this method can effectively shield the effects of soil,tree trunks,branches and other complex background elements in the shooting background in the natural shooting environment,achieve a better segmentation effect,and minimize the noise caused by the natural environment.(2)In order to reduce the influence of uneven lesions and local bright spots caused by uneven lighting during shooting,the study used contrast adaptive histogram equalization to process the image.The results show that this method can effectively improve the recognition of the lesion area and avoid the problem of image discontinuity and excessive enhancement.(3)In order to reduce the learning pressure of the convolutional neural network recognition model,improve the learning efficiency and improve the recognition accuracy,the research proposes to perform Otsu processing of the adaptive threshold of the disease image,extract the disease feature area and realize the extraction of the disease feature map,and train and recognize.The results show that the training time is significantly shortened,and the verification accuracy of 96.07% is obtained,which is 7.28% higher than the original image recognition scheme,and the effect is remarkable.(4)In order to reduce hardware requirements,this thesis builds a tree support vector machine recognition model based on disease image feature extraction,and trains the color moment and SIFT input support vector machine model to improve learning efficiency.Experimental results show that compared with the CNN recognition scheme,the training speed is greatly accelerated.The same data set is used for verification,and the recognition accuracy rate is 89.9%,which effectively improves the learning efficiency.Improving the accuracy of image recognition through image preprocessing and feature enhancement is more targeted,and at the same time can avoid other adverse effects caused by the correction of NN;and the SVM recognition scheme combined with feature vector extraction can also complete disease recognition and reduce the hardware environment limit.In addition,this research combines the image processing algorithm and identification model of poplar leaf diseases to design and develop an automatic recognition system for poplar leaf diseases,which provides practical support for forest disease detection and prevention.Other disease intelligent detection schemes can also refer to this Program.