| One of the important bases for the comprehensive prevention and control of pests is to obtain the data of the dynamic changes of pests.Because the pests in the greenhouse have the characteristics of small insect body,migration and masking(such as obstruction of foliage,accustomed to the back of leaves),artificial sensories are used to determine the species of pests by means of magnifying glasses.This method is labor intensive,inefficient and the accuracy is influenced heavily by subjective.The method of pest detection and recognition based on image processing machine vision is time-saving,labor-saving,continuous and intelligent.It has become a research hotspot in the field of modern agricultural pest monitoring.Based on the common whitefly and thrips adults in Cucumber Greenhouse,the method of detection and recognition of greenhouse crop pests based on image processing is studied.The main research contents are as follows:(1)To obtain the image of whitefly and thrips in cucumber greenhouse environment.Adult whiteflies and thrips are small,the differences between them are not obvious.The main difference is color,and they are taxis to yellow.A pest automatic monitoring device developed and designed by our group is used to obtain insect pest images.It is mainly whitefly and thrips in the image,and there are also a small number of other pests,such as larger flies which is removed by the setting threshold during image processing.(2)The methods of image preprocessing,image segmentation and insect target extraction algorithm are described in detail.The contrast between the target and the background is improved by transforming the original image from RGB color space to HSI and L*a*b* color space.In this paper,the Prewitt is used to detect the edge of a single pest in I component of HIS binary image and the Canny operator is used to segment the single head pests on the b component of L*a*b* binary image.Then morphological treatment,empty filling processing,the final fusion of these two binary images to complete single-headed pest area extraction.(3)The target features of pests are extracted,the feature parameters are normalized and the feature vectors are formed.This paper selects the 9 color characteristics and five morphological characteristics as the characteristic parameters of cucumber pests.Two kinds of pest prediction models,BP and SVM,are constructed.The average recognition accuracy was 93.5%,and the identification rate of whitefly and thrips was 96.0% and 91.0% respectively.The recognition results show that the recognition effect of SVM is better than that of BP,and the color feature vector is the main component of the pest identification.(4)In order to get better recognition results,the parameters that affect the quality of image acquisition are discussed,and the best parameters of automatic monitoring device and image are determined.Illumination change has a great impact on the image of pests,which not only affects the process of image preprocessing,but also affects the recognition effect.The paper chooses to optimize the hardware parameters and image processing algorithms to reduce the influence of illumination.Aiming at the applicability of pest identification system,the standard of parameters and image parameters of automatic monitoring device is made in this paper.(5)The system of identifying and counting agricultural pests includes three parts: model building,pest identification and monitoring.The purpose of model building is to build a pest identification classification model.The pest identification module is used to identify the collected insect images and the pest monitoring module monitors the insect images returned by the automatic monitoring device in real time.From the results,the image segmentation and recognition algorithm can automatically and effectively segment,count and identify pests.It can provide support for monitoring and early warning of pests,and provide important basis for timely prevention and control. |
PDF Full Text Request