Research And Implementation Of Crop Disease And Pest Image Retrieval Methods

As the third largest grain crop and the first largest fruit in the world,potato and citrus have caused huge economic losses every year because of the disease and insect pests.The intelligent diagnosis method of plant diseases and insect pests based on computer vision,pattern recognition and image processing technology has important practical significance for its sustainable development,and is also a research hotspot and focus of agricultural modernization and information development.This study aims to solve the problems of large environmental impacts of potato and citrus pests and diseases,low extraction speeds in disease and insect pests,and low recognition accuracy.The above issues have been studied in depth with self-built images of potato and citrus pests and diseases(a total of 35 categories and 1650 samples)as the research object.The main research contents and innovative results are as follows:(1)In view of the fact that crop pests and diseases images in the natural environment are susceptible to light,angles and scales,the preprocessing method of pest and disease image is experimentally studied.The use of color space transformation and image graying can effectively eliminate the impact of light on pest and disease recognition.Median filter can effectively retain the texture and edge details of pests and diseases,and the effect of noise reduction is good.(2)In view of the problem that the image of disease and pest is greatly influenced by the natural background,and when the feature is extracted directly on the original image,there are many problems such as large computation and severe feature redundancy.A disease and insect pest area segmentation method based on Grab-Cut algorithm is established,which can better get the disease and pest area,but the running time is 1s and above,and it needs human interaction.Based on the above problem,this paper proposes an algorithm based on key feature points for the detection of ROI.First of all,the ORB feature points are extracted after the pyramid image is down-sampled,and the SIFT feature points are extracted when the number of feature points is smaller than a given threshold.Then the coordinate values of the feature points are sorted in the horizontal and vertical directions respectively.And by calculating the mean value of the K nearest neighbor feature points,the coordinates of the pest area are determined and the ROI is extracted.The research shows that this method can automatically locate pests and diseases accurately,and the average time of ROI detection is 13 ms.(3)Aiming at the problem of high computational complexity and low recognition accuracy when extracting features from the original pest image,based on the pest and disease ROI image,the HSV color histogram and the UPLBP texture histogram are integrated as the total feature vector of the pest region,and a K nearest neighbor disease and insect pest recognition algorithm based on weighted distance is designed.Firstly,a weighted distance separator is used for pre-classification,and then the K nearest neighbor classifier is used for the second decision to obtain the final recognition result.The results showed that the average recognition accuracy of potato and citrus pests are 96.3%,93.48%,94.07% and 92.59% on the self built image library,and the average running time is up to 243 ms.In addition,a pest support classification model based on a nonlinear support vector machine based on radial basis kernel functions is used for comparative study.Experiments showed that the average recognition accuracy of potato and citrus diseases and insect pests are 95.42,92.07%,91.04% and 90.37%,and the average running time is within 113 ms.(4)Based on the open source computer vision library Open CV2.4.13 and using C/C++ programming language,an image retrieval system for crop diseases and insect pests is designed and developed on the Windows system.The system can accurately identify the diseases and insect pests of the image to be retrieved in 0.2S time and so on,and provide corresponding pest and disease control measures,and realize the automatic diagnosis of the image of potato and citrus disease and insect pests.