Study On Rice Disease And Insect Pest Detection Based On Image Processing

As the main food crop in China,rice has penetrated into all aspects of agricultural production and people’s life.In the process of rice growth,diseases,insect pests and other objective factors have adverse effects on its high quality and high yield.With the development of intelligent agriculture,although the traditional manual detection methods are gradually replaced by the detection method based on machine learning,but due to the actual operational testing system is more complex,farmers for detecting computer knowledge is not enough,the system in the countryside can’t get a good promotion and popularization.The research area of this paper is located in Zunsan Village,Matang Town,Rudong County.A total of 1217 images of various rice diseases and insect pests were collected through field visits,and the following suggestions were put forward respectively:Based on LC significant detection algorithm and OSTU algorithm with the combination of plant diseases and insect pests image segmentation method,based on gray level co-occurrence matrix,color moment,HU invariant moments with the combination of plant diseases and insect pests image feature extraction methods,the PSO-SVM model based on RBF kernel function of plant diseases and insect pests image recognition method,and based on this design a simple and practical plant diseases and insect pests of rice detection system.The specific research contents and results are as follows:(1)Collection and pretreatment of images of rice diseases and insect pests.In this study,Huawei mate20 and Oppo findx3 smart phones were used to take a total of 1217 images of rice diseases and insect pests in zunsan Village,Matang Town,Rudong County,during the rice diseases and insect pests frequent season from early June to September 2021.Firstly,MATLAB2016 b was used to simulate the denoising experiments of different functions on the grayscale images of diseases and pests,and gaussian filtering with better denoising effect was used as the research method of image denoising in this paper.Secondly,several different image segmentation methods were used to segment the collected images of rice diseases and insect pests,and it was found that the threshold segmentation method combined with OSTU algorithm on the basis of LC significance detection algorithm was used for segmentation,and the threshold segmentation method obtained the best results.Finally,open and close operation was performed on the segmented images to retain the true shape of rice disease and insect pest images as much as possible.(2)Image detection and recognition of pests based on SVM model.This paper analyzes the SVM recognition method and proposes an improved SVM recognition method based on particle swarm optimization.The collected images were screened,and the damage images of rice blast,rice smut and leaf roller were selected.A total of 568 samples of the training set,142 samples of the test set and114 interference images were selected according to the ratio of four to one.Single or multiple image feature parameters were operated respectively by the original SVM and the improved SVM.It is concluded that the training accuracy is the highest when the PSO-SVM model recognizes the fusion features of color and shape.(3)Development of rice pest and disease detection system.In this paper,the feasibility of the system is analyzed,the main functions of the system are determined,and based on the optimal support vector machine identification model studied above,a detection system is developed which adopts the general Spring architecture and SOA design concept based on J2 EE technology system.The system mainly adopts the interaction between the front and back systems and the interaction with the peripheral systems based on HTTP protocol.The front-end uses React+Antd+Umi architecture.The background transmits image data to the server and calls the image recognition program that has been packed to the server in advance to segment and extract features from the images of rice diseases and insect pests,so as to carry out identification and detection.In this paper,we study the average diagnostic accuracy was 94.525%,the detection system of the diagnosis time is about 7 seconds,basic can achieve anticipated goal functions,compared with the traditional test model,with strong practicability,high identification and easy operation etc.,to rice farmers use,can provide them with timely,effective and nondestructive prevention advice,It can meet the needs of agricultural planting in the context of big data.