Research On Visual Aid-navigation Method For UAV Rice And Rape Seed Farming Operations

The popularization and application of hybrid rice and hybrid rape are important measures to ensure the national grain and oil safety.Pollination and plant protection are important parts of the seed production of hybrid rice and hybrid rape.In the process of seed production,the track of plant protection unmanned aerial vehicle(UAV)is inconsistent with the target crop,and the centerline of the target crop can not be tracked in real-time,which results in damage and unsatisfactory seed production effect of parents’ plant.Addressing the above issues,taking hybrid rice and hybrid rape seed production as an example,a method based on convolution neural network and GPS combination navigation was proposed to construct a row recognition model of rice and rape target crops based on convolution neural network to extract the centerline of crop rows.Based on absolute position and relative position information,the GPS position information of the target into the centerline of target crops and the navigation line was obtained.The purpose of this project is to construct a set of auxiliary operation system of flight correction control device,to associate the flight guidance of plant protection UAV with the actual situation of farmland,so as to realize the correction of field operation route of plant protection UAV,to increase and improve the breeding of rice and rape.This article mainly studies as follows:1)The extraction method of the central line of rice and rape target crops was studied.According to the characteristics and agronomic requirements of rice and rape parents,the image of low altitude hybrid rice and rape under different working conditions was collected,and the recognition model of rice and rape target line based on convolution neural network was constructed.By the recognition model,the sequence image of the image was obtained,the morphological method was used to denoising,and the parent line is obtained by smoothing the edge processing of the feature region by the least square method,and the feature points are extracted by the horizontal segmentation feature region.The parent line centerline is fitted by the least square method and Hough transformation.And experiments were designed to verify the accuracy and stability of image verification methods in different environments,the results show that the least square method was superior to the Hough transformation in accuracy and real-time,and the average accuracy of the proposed method was 95%,the average time per picture was about 1204.48 ms,which basically meets the field navigation requirements.Compared with the traditional machine vision method,this method has great advantages.2)The method of coordinate information implantation in the central line of low-altitude rows was studied.The method based on absolute position and relative position information was proposed to carry out coordinate implantation.The camera was calibrated to eliminate the influence of camera distortion,carry out perspective transformation and calibration test on the image,and the relationship between the pixel and physical size was established.Finally,the relative coordinate system and absolute coordinate system with the camera shooting point as the origin are established to calculate the geodetic coordinates of each pixel,and the navigation line can be obtained by inserting GPS coordinates into the centerline of the target crop line.The maximum error of abscissa and ordinate was 0.17 m,the minimum error was0.07 m,the average error was 0.11 m;the maximum distance error was 0.17 m,the minimum error was0.07 m,the average error was 0.12 m,which can meet the positioning requirements of UAV.3)The method of tracking guidance of plant protection unmanned aerial vehicle was studied.The flight principle and attitude calculation principle of plant protection unmanned aircraft were analyzed.A method of route tracking based on the midpoint method was proposed,which was calculated the guiding point according to the current position,velocity direction,and navigation line information of the plant protection unmanned aircraft vehicle.It realized to control the plant protection UAV along the navigation datum.The plant protection UAV at different course angle(angle between flight direction and navigation line)of-30 °,0 °,10 °,20 °,30 ° under 5m/s conditions of plant protection UAV single-step flight were simulated.4)Experimental verification.A set of experimental platform was built to verify the performance of the method proposed in the study.A mean value of the distance deviation between the centerline of the target crop line and the centerline of the target row artificially fitted was 4.23 cm.The source of this error was mainly due to the human measurement error and the pixel pick-up error of the marker on the picture,and the different course angle of-30°,0°,10°,20°,30°under 5 m/s speed of plant protection UAV were tested.When the horizontal distance from the navigation line was 1.6 m,when the operating speed was 5 m/s,the average adjustment time was 0.73 s,and the longitudinal adjustment distance was about 7 m,basically meet the plant protection UAV field operation requirements.