| As an important part of precision agriculture technology,the automatic navigation control system is of great significance to solve the problem of labor shortage and improve the level of intelligent agricultural machinery and agricultural production efficiency.In recent years,with the continuous optimization of China's industrial structure and the vigorous implementation of innovation driven development strategy,many new kinetic energy has been injected into the automatic navigation control technology of agricultural vehicles,which has gradually become a research hot spot in the field of precision agriculture.China is the country with the largest fruit planting area and the highest yield in the world.At the same time,the problem of aging population is serious.Therefore,in order to meet the requirements of green and sustainable development of modern orchard operations,alleviate the shortage of agricultural labor,it is imperative to develop and design advanced orchard operating equipment using automatic navigation control technology.In this paper,the standardized orchard apple orchard is used as the research environment,and the agricultural electric vehicle is modified as a research platform.Considering the problem of agricultural production costs,2D laser is selected as the main sensing element.The three key technologies in the automatic navigation system of orchard intelligent vehicle are studied,which are environment perception positioning,navigation path tracking and obstacle detection.The main work and conclusions are as follows:(1)Through consulting a large number of references at home and abroad,investigating the existing agricultural machinery products on the spot,the main research methods and key technologies adopted are understood.At the same time,the main problems existing in the automatic navigation system of orchard vehicles are analyzed through comparison.Finally,the research content,research scheme and technical route of this paper are determined in view of the above problems.(2)On the basis of SH-GL agricultural electric vehicle,the hardware platform of orchard intelligent navigation vehicle is built,including the installation of 2D laser sensor,front wheel angle detection system,the modification of its starting system and automatic steering system.The performance of laser sensor,front wheel angle detection system and automatic steering system are tested through different experiments.The test results showthat the designed hardware systems can complete the corresponding work well,has good reliability and stability,which lays the necessary foundation for the follow-up study.(3)In order to solve the problem that the laser data is easily affected by the factors such as cross line fruit trees or small obstacles between lines,which results in poor positioning accuracy,a data pretreatment method combining the distance threshold method and the statistical filtering method is proposed In order to reduce the computation of the navigation system,a method of obtaining the feature center points of tree trunks is proposed,which using the three points collinear principle,including the shortest ? value point,laser emission point and tree trunk center point after Euclidean clustering.The test results show that,in the X axis direction,the maximum deviation is 8.0cm,the average deviation is 3.2cm,the standard deviation is 2.7cm.In the Y axis direction,the maximum deviation is 9.9cm,the average deviation is 6.5cm,the standard deviation is 2.3cm.Its accuracy can meet the actual needs.At the same time,the least square method is used to complete the fitting of navigation straight line path,and the course deviation and lateral deviation are taken as navigation control parameters to obtain.Compared with the actual value,the maximum absolute error of course deviation is 0.95 °,the average absolute error is 0.682 °,the root mean square error is 0.237 °;the maximum absolute error of lateral deviation is 4.66 cm,the average absolute error is 2.119 cm,the root mean square error is1.010 cm.The results show that the accuracy of navigation parameters can meet the requirements of independent operation in orchard.(4)According to the characteristics of orchard vehicle path tracking,based on the two wheeled vehicle model,a pure tracking control model under the condition of straight-line tracking is constructed.In view of the problem that the forward-looking distance of pure tracking control model can not be adjusted in real time,a fuzzy controller is designed to adjust the forward-looking distance in real time with the input of lateral deviation and heading deviation.A query table of forward-looking distance fuzzy control is made by Matlab software.In the Matlab / Simulink environment,the overall simulation model is built,and the simulation test is carried out for two paths of straight line and curve under the condition of constant speed.The experimental results show that the system has the advantages of fast response,high tracking accuracy and good stability.At the same time,the field test was carried out in the artificial simulated orchard.The test results show that the maximum tracking deviation of vehicles is 0.13 M,the average tracking deviation is0.048 m,and the lack of fruit trees has no significant impact on the designed control system.The control accuracy basically meets the actual operation requirements of the orchard.(5)In order to better realize the vehicle path navigation planning and improve the driving safety,aiming at the existing problems of obstacle detection in orchard,an obstacle target detection method using 2D laser to obtain the distance information of obstacles and monocular CCD camera to obtain the category information of obstacles is proposed.An obstacle detection model of SSD deep learning target is built and trained,and the accuracy performance evaluation is completed through the single obstacle recognition experiment.The installation platform for orchard test is made,and the distance and category combination detection experiment for multiple obstacles in orchard row under three different positions is carried out.The experimental results show that the method proposed in this paper is reasonable and effective,and the confidence of class detection is above90%. |
PDF Full Text Request