Research On Pepper Target Recognition And Positioning Technology Based On Machine Vision

It takes a lot of labor to pick pepper,and the labor force in rural areas is not only in short supply,but also the cost increases year by year,which makes the production cost of pepper increased sharply,becoming the main factor restricting its large-scale production.Therefore,the study of high-efficiency agricultural picking technology is the development requirement of modern agriculture.In this subject,the automatic identification and location of pepper targets in natural environment are studied by using computer vision,digital image processing and pattern recognition,in order to enhance the accuracy of robot field operation and improve work efficiency.Specifically,the following work has been carried out:(1)The collection and color characteristics of pepper images were studied.On the basis of studying the actual growth status of pepper,the characteristics of pepper fruit in common color space model were emphatically expounded.Firstly,multiple color components in the color space model were used to process the pepper image,and compared the color characteristics of pepper fruits in each component;Secondly,the grayscale values of fruits,branches,leaves and backgrounds in each component map were analyzed and counted.(2)The method of segmentation of pepper images in the natural environment was studied.Based on R and H component graph,and combined with the Great Law(OTSU)to divide the pepper image interest area,and compared with the segmentation results of R-G chromatic aberration,ITS method and K-means clustering algorithm,which show that the OTSU division method based on R and H component can better divide the pepper interest area,and more realistic close to the actual pepper area.(3)The positioning of pepper picking points in the natural environment was studied.Firstly,morphological treatment and canny algorithm were used extract the contour region of pepper string,and then the pixel coordinates of pepper picking points are obtained according to the center-heart deviation method and the inertial spindle method,and compared with the pixels of the best picking area in the image.Finally,the pixel error of thepositioning of 100 pepper string picking points under different light was analyzed,of which83 images were between the picking point and the best picking point pixel error between 3and 10,and the number of images within 10 pixels reached 83%,indicating that the method could basically achieve the accurate positioning of pepper picking points.(4)In order to obtain the three-dimensional spatial coordinates of pepper fruit,the two-dimensional information of the pepper string image is analyzed,the principle of visual matching of binoculars is studied,and the three-dimensional coordinates of pepper string are designed and realized.Using the two-eye visual center matching method to carry out the experiment,and the obtained target pepper three-dimensional coordinate value error analysis,the test results show that: within the range of 350~1100mm from the baseline,In the range of 350~1100mm from the baseline,the error range of the concentric matching of pepper strings on X,Y and Z coordinates is-1~1.2mm,-2.4~2.7mm,and-11.3~11.5mm,which can basically meet the picking requirements of the picking robot.This topic completes the study of the target identification and positioning technology based on machine vision,and the effective analysis of the identification and positioning methods,corresponding algorithms and algorithms proposed in this paper lays the foundation for the accurate identification and positioning of peppers in the natural environment,and plays an important role in improving the efficiency of pepper picking on the spot.