Research On Key Technology Of Detection Of Jujube Tree Pest Based On Image Recognition

Jujube tree,as ancitent chinese fruit tree,ranks the top in dried fruit production.In recent years,threat of pests is increased as planting area of jujube tree expanding.As long as damaged by pests,the jujube tree will be affected its harvest in the next many years,which always brings great financial losses to jujube farmers.When disaster of jujube tree pest occurs,the most important thing is that it shall recognise which pest it is,then it can suit the remedy to the case.However,facing to problems of numerous varities of jujube tree pests,and large planting jujube trees,as well as time-wasting and error-prone only relying on expert and technician as disaster of jujube tree pest occurs,it can't meet jujube tree farmers' requirements at all.With the development of image recognition technology,it also explores image recognition technology instead of manual intervention in argricultural field so as to reduce response time of disaster.Based on previous research,this paper explores key technology of image recognition applying on aspect of detection of jujube tree pest.Due to poor image environment in jujube tree orchard achived,this paper analysis the method of improving image quality applied by image preprocessing technology.As well,this paper explores and comes up with proper image proprocessing technology method,which is more suitable for jujube tree pest,respectively according to study image preprocessing technologies algorithm principle such as the technology of image gray level change,histogram equalization,and image filtering,as well as study its corresponding image.In order to reduce the image recognition information,the image segmentation technology is applied.This paper mainly discusses the algorithm principle of threshold segmentation technology,region based segmentation method and edge detection based segmentation method,and proposes the specific application methods of Sobel operator and Canny operator in image segmentation of jujube tree pest detection.In the aspect of image pattern recognition,the methods of detecting jujube tree pests using by machine learning and neural network are discussed respectively.In the part of machine learning,HOG algorithm is proposed for image feature extraction and SVM algorithm for image classification.In the part of neural network,the hierarchical scheme and algorithm principle of classical convolutional neural network are introduced,and the steps and methods of using LeNet-5 model to detect jujube pests are put forward.In this paper,some jujube orchards of group 8 and group 16 in Alar City were selected to take pest images.More than 400 images of red spider,moth larva,geometrid and gall midge were taken.The experimental results show that the image preprocessing method can reduce the amount of data processing by about two-thirds,and the image segmentation technology can quickly extract the pest images to be identified.In the part of pattern recognition,the recognition rate of red spider,moth larva,geometrid and gall midge were 80.00%,81.82%,75.00% and 77.78%,respectively.The accuracy rate of using lenet-5model to identify red spider was 76.92%,the recognition rate of E.spinosa larvae was 81.82%,the recognition rate of jujube geometrid was 87.5%,and the recognition rate of jujube gall midge was 76.92%.