Research On Target Visual Recognition Technology Of Greenhouse Green Pepper Under Near Color Background

China is a big producer of green pepper,the picking work of green pepper is intensive.Due to the shortage of agricultural labor force,in order to ensure the picking of green pepper,it is necessary to improve the efficiency of green pepper picking,so the automatic picking of green pepper is an urgent problem to be solved.Since the color between green peppers and its background is approximate,it is an important task for the pepper picking robot to accurately recognize the green pepper.This paper integrated research hot spots of machine vision in agriculture,and green pepper image under greenhouse were used as the research objective.The research mainly focuses on key technologies of image acquisition and preprocessing,image segmentation,image feature extraction and target recognition,realizing the target recognition of green pepper image in the color approximate background,finally verifying the reliability and feasibility of the method by the experiment,the main research content involves the following several aspects:(1)On the basis of acquisition and pretreatment of green pepper image information,Green pepper images were collected by the Uni Fly M088 Camera on the green pepper picking robot in greenhouse,and the image pixel was 640×480.There were green peppers,leaves,lands,and parts of the sky in the picture.Histogram was used to visualize the characteristics of the near color,and the main color space was introduced.Finally,the bilateral filtering operation was performed on collected images,and noises were removed while the edge of green pepper was preserved.(2)To solve the problem of image segmentation and post-processing of green pepper,the K-Means segmentation method was used to split the green pepper image first.Because of the similarity between the color green pepper and its background,the segmentation effect is not ideal.Therefore,the threshold segmentation method was used to the image of green pepper once more;in the later stage,morphological method was used to remove noises of image,and the area method was applied to preserve the effective area of the target and get the whole green pepper segmentation image.Finally,in order to extract features of the image,images of green pepper weretransformed into color images,and the details of the image were kept as intact as possible.(3)The problem of feature extraction of green pepper image.Consider the color approximation between green pepper and background,the feature of the shape and texture was obvious,therefore,the seven shape feature which can describe the shape of green pepper image was extracted by Hu invariant moment with invariant characteristics of translation,scaling and rotation.The vectors of roughness,contrast,and direction were extracted by Tamura texture feature method to express texture features of green pepper.In order to reduce the complexity of data calculation and improve the efficiency,the extracted shape feature vector and texture feature vector were normalized.(4)Target recognition of green pepper using the improved particle swarm optimization least squares support vector machine(IPSO-LS-SVM).The extracted features were divided into training samples and testing samples,the mutation strategy was introduced to maintain particle activity in particle swarm.Optimal parameters of The Least Squares Support Vector Machine were obtained.The identification model of green pepper training was improved,and was tested with test samples.Experimental analysis showed the accuracy rate can reach 95.89%.It indicated that the method adopted in this paper was feasible in this paper was feasible for the identification of green pepper in the greenhouse.