Classification Of Main Pests In Cruciferous Vegetables Based On Image Recognition

The prediction and forecast of pests is the premise of reasonable pest control.At present,the method of measuring and reporting the main pests of cruciferous vegetables is: trapping and killing pests through pest report lamps.After a period of time,the trapped pests are manually retrieved and then identified and counted by the forecasters in the laboratory.This method takes time and effort,and the accuracy of the forecast is not high.In order to reduce the workload of the forecasters and improve the accuracy of the forecast,this paper presents a classification study on the main pests of cruciferous vegetables based on image recognition.The main work includes:(1)Samples of major pests of cruciferous vegetables were collected through different channels.After manual classification,according to the image collection standards set in this article,the image of the pest samples was collected to obtain 4218 images of pests,and the database of main pests of cruciferous vegetables was established;(2)The characteristics of insect taxonomy were studied,by comparing the taxonomic characteristics of the three parts of the head,chest and abdomen,as well as the pest body data measured in the laboratory.Six new types of shape features are proposed: head wing area ratio,head wing length ratio,head wing average width ratio,single wing double wing diagonal length ratio,longest axis to corresponding short axis ratio,perimeter median axis ratio;(3)In order to extract 6 new shape features more quickly and accurately,this paper designs a new segmentation method to automatically segment pests,and designs a GUI interface for the extraction of six shape features;(4)The six shape features of the pest image and the four traditional features of Gabor feature,LBP feature,HOG feature and color feature are extracted respectively,and the five major cruciferous vegetable pests are classified by classifier,and the six shape features are compared the accuracy of the four traditional features.The results show,The six shape features proposed in this paper have high accuracy and can better classify and identify the main pests of five cruciferous vegetables;(5)In order to improve the recognition rate,this paper integrates 6 kinds of shape features and 4 kinds of traditional features,and obtains 5 kinds of fusion features.The SVM classifier is used to classify the five major pests of cruciferous vegetables.The results show,it is not that the more fusion features,the higher the recognition accuracy.The combination of six shape features and color features is more suitable for the classification and identification of major pests of cruciferous vegetables;(6)By adding incomplete images to the original database,the recognition rate of major pests of cruciferous vegetables has dropped significantly,with an average accuracy rate of 55.2%.By fusing four traditional features,the recognition accuracy is improved,and the average recognition accuracy of the 6 shape features fused with LBP features is improved by 41.49%,reaching 96.69%.This article establishes the collection standard of the main pest images of cruciferous vegetables,and establishes an image database,which contains most of the pest gestures.This provides a good basis for future research on the intelligent classification and identification of major pests of cruciferous vegetables.The six shape features proposed in this paper have high recognition accuracy,which provides a good research idea for the subsequent research on automatic classification and recognition of major pests of cruciferous vegetables.