Research On Image Acquisition And Diagnosis System Of Rice Diseases And Pests Based On Embedded Mobile Terminal

At present,the diagnosis and control of rice diseases and pests in China is mainly based on manual observation,and the intellectualization of disease and pest diagnosis is relatively backward.There are some problems such as labor consumption,low diagnosis efficiency and high misjudgment rate,which can not meet the needs of the agricultural market.In order to solve the problems of high requirements for rice image acquisition and processing equipment and large sample size required to determine the types of rice diseases and pests,The acquisition and preprocessing of rice diseases and pests in agricultural field based on embedded technology and image processing were realized in this dissertation.The main research contents and results of this paper are as follows.(1)A rice image acquisition and processing system based on embedded mobile terminal is designed and implemented.The itop4412 development board based on cortex-A9 processor,external handheld USB camera and LCD screen are selected as the mobile terminal platform.The GUI function program is designed in the embedded QT integrated development environment to realize the graphical human-computer interaction function based on the embedded mobile terminal.The image acquisition application program uses v4l2 video acquisition specification and converts the image format for real-time dynamic display,acquisition and storage of images.The image processing application calls the opencv library related functions to extract the disease spots,binarize the disease spots and the leaves of the read images,so as to provide the basis for users to make the preliminary judgment of rice disease in the agricultural field.(2)The far-end image diagnosis algorithm of rice diseases and insect pests is studied and implemented.Taking healthy rice,brown spot disease,iron beetle disease and rice blast as the research objects,a method of weighted fusion of HOG and LBP features based on the accuracy of classification and recognition is proposed in this dissertation,and uses support vector machine to classify rice images.In the case of small sample size,the accuracy of classification and recognition of rice images is 94.2%,which is better than the classification results of single feature,series fusion feature and equal weight fusion feature.On the basis of this method,the target detection of three kinds of disease spots is further combined with multi-scale detection.The detection rate of disease spots is 90.5%,which is higher than the detection results of YOLO v3 and Fast R-CNN.The far-end realizes the accurate recognition of four types of rice images and the efficient detection of three kinds of disease spots.The system studied in this dissertation can meet the acquisition and processing of rice pest images in agricultural field,as well as its diagnosis in far-end.It has obvious advantages in small sample size and processing efficiency,and has broad application prospects in the field of agricultural informatization in the future.