Research On Technologies Of Cotton Image Segmentation And Positioning Based On Machine Vision

Over the past few years,with the development of the cotton industry,how to reduce the cost of cotton-planting and increase the efficiency and quality of cotton-picking are the research priorities of cotton researchers.The traditional mechanical cotton picking method has high impurity rate,low quality level,and high operating cost.Therefore,the intelligent cotton picker is the future direction of cotton picking machinery.This paper proposes an intelligent picking scheme for the application environment of the cotton picker.It uses the binocular vision system as the “eye” of the cotton picker to identify and locate the cotton target,and transmits cotton position information to the robot to achieve intelligent picking.In cotton target recognition,this paper first analyzes the main segmentation problems in the cotton environment,and then proposes three adaptive segmentation algorithms: adaptive threshold image segmentation based on Otsu algorithm,improved BP neural network algorithm and extreme learning machine..Through the image segmentation experiments of the three algorithms,the following rules are analyzed and summarized:(1)In complex natural environment,there are many background regions whose gray information is similar to that of cotton.Due to the occlusion,some cotton regions have low luminance information.Using threshold segmentation methods,it is difficult to completely separate cotton from the background region.(2)In the training process of BP neural network and extreme learning machine,the eigenvalues of RGB and OHTA color space have different advantages and disadvantages for different regions.(3)Compared with BP neural network,extreme learning machine has better real-time performance when the segmentation performance is similar.Therefore,this paper proposes an image segmentation algorithm based on extreme learning machine.The feature components in RGB and OHTA color space and the average brightness of the image are used as the input of the network model to realize the accurate segmentation of the cotton area in the natural environment.In the visual positioning system,a parallel binocular stereo vision system was selected according to the needs of the project.The theoretical model of the binocular vision system was introduced,and the calculation principle of camera calibration was analyzed and expounded.The camera calibration method based on plane calibration plate was used to complete stereo calibration of binocular camera.In stereo matching,the advantages of commonly used matching methods are analyzed and summarized.Combined with the picking situation in this paper,a region matching method based on cotton segmentation information is proposed.This method utilizes the segmentation information of images to achieve a fast matching of cotton picking points.Good real-time performance and low false-match rate.Finally,in order to improve the accuracy of the positioning system,this paper analyzes and summarizes the main influencing factors of measurement errors under the parallel binocular stereo vision system,and establishes the error model of the measurement results,proposes the idea of error compensation,and uses BP neural network model to predict the positioning system.Measurement error,and by learning the sample data to obtain a prediction model of measurement error,error compensation of the measurement data,analysis and comparison of the accuracy of the original measurement data and the modified measurement results,verifying the feasibility of the error compensation method in improving the positioning system accuracy.