Dissertation > Excellent graduate degree dissertation topics show

Research on the Automatic Identification of Rice Planthopper

Author: YangLingLing
Tutor: DingWeiMin;LiuDeYing
School: Nanjing Agricultural College
Course: Agricultural Mechanization Engineering
Keywords: rice planthopper digital image color texture BP neural network automatic recognition
CLC: TP391.41
Type: Master's thesis
Year: 2008
Downloads: 132
Quote: 1
Read: Download Dissertation

Abstract

Rice is one of the most important crops,and living creature disaster is a key factor to influence stable and high-yield production of rice.Rice planthopper is a major pest to rice, and different kinds of rice planthopper have a different effect on rice.So identifying the species of rice planthopper exactly and recording the number of accidents are effective measures to prevent the pest disaster.There are some abuses in the traditional recognition method.For example,it is not stable and it has a heavy reliance on experts.The automatic identification technology on the species of rice planthopper has a bright future,when it can solve the contradiction between the increasing of requirement of professional species identification and the decreasing of examiners.Using image processing technique and BP neural network technique,proper image processing and identification technology were researched,mainly at the static image of sogatella and laodelphax.The results proved that this method was feasible to do the species identification about rice planthopper.The main contents and results of the research:(1) The determination of research projectWe distinguish rice planthopper depending on the color and texture of its back.The size of insects can be used as a subsidiary basis.So the back of rice planthopper was used as the target of research.Using 43 samples of female sogatella,67 samples of male sogatella,67 samples of female laodelphax and 72 samples of male laodelphax,took some microscopic digital images of their backs in order to use them in automatic recognition.(2) Image preprocessingBy comparing the mean filter and the median filter,conclusion that median filter can protect image details was gotten.Through the experimental analysis,satisfying results using 3×3 square window was gotten.Global threshold method was used at the stage of image segmentation.By analyzing and comparing the distribution of gray value among each channel of R,G and B,the conclusion that the differences of gray value between the aim and background in channel B were the most obvious.Because of this,image was segmentated in the channel of B.When confirming the best segmentation threshold, morphologic processing needed to be combined,and remarked the effect of segmentation by area ratio named P.When threshold was 98,|P| was the minimum by comparing.That was to say, the target area approached the insect body and the segmentation was the best.(3) Image feature extractionAccording to the main and stable characteristics of distinguishing rice planthopper, effective features that can reflect the differences between different species of rice planthopper were extracted.The color and texture features of rice planthopper were described mainly in this research because of characterizing images using color feature only can cause decrease of expression accuracy,and the method of percent histogram was used to extract the percent.The chrominance is between 20 and 40.It would not be influenced by the size and the shape of objects,and background noise.It could be used as an important input feature of image pattern recognition.In addition,color moment of images was extracted.In the texture feature extraction,using texture features of mesonotum area to represent the texture features of the whole rice planthopper was proposed.The principal axis direction must be kept identical with y axis direction when extracting sub-image.The sub-image could not contain background noise,or the statistical meaning would be lost.(4) Image classification and recognitionThe training improvement methods of three common BP Neural network were compared and analyzed.Conclusion that L-M method had faster convergence rate was drawn by comparing training error,convergence rate and testing rate,lower network error and higher testing rate.The experimental results proved that the network error was the lowest when input vector included 18 eigenvalues ex aera feature,and that the trained network had a better recognition rate to identifying samples.This result showed that the neural network classifier was proper.(5) Design of the recognition system of rice planthopper.The function was extracting feature vector after a series of processing,then judging types of individual according to rice planthopper images information commited by users.

Related Dissertations

CLC: > Industrial Technology > Automation technology,computer technology > Computing technology,computer technology > Computer applications > Information processing (information processing) > Pattern Recognition and devices > Image recognition device
© 2012 www.DissertationTopic.Net Mobile