Design And Test Of Seedling Image Acquisition Robot

Phenotypic non-destructive measurement technology can provide big data support for seedling breeding,cultivation and growth monitoring.At present,the phenotypic measurement of seedlings mainly relies on manual labor,which is inefficient and destroys seedlings,and high-throughput non-destructive measurement technology is urgently needed to replace manual labor.Compared with fixed phenotypic detection methods,the use of mobile robots with image sensors is a more flexible and low-cost detection method.In this project,a seedling image acquisition robot is designed to meet the image acquisition requirements of in situ detection of seedling phenotypes.The robot consists of three parts: motion control module,image acquisition module and automatic navigation module,and the robot can complete the automatic acquisition of seedling canopy images.The main research contents and results are as follows:1.Robot mechanical body design.The overall structure of the robot was determined based on factors such as robot working scene and seedling cultivation method.and design and process its key components such as shock absorbing structure;The height of the image acquisition bracket in the greenhouse scene of seedlings was determined by experiment,and the support was designed to solve the problem of motion imbalance caused by the height of the bracket.Faced with the seedling scene with different cultivation methods,an adjustable lifting scheme of bracket is proposed.2.Robot STM32 control scheme design.The robot motion control scheme with STM32 as the control core was determined.The control scheme mainly includes three parts: robot linear speed regulation,angular speed regulation,and host computer communication;According to the functional requirements of the robot,the integrated circuit board is designed and made by Altium Designer software to improve the stability of the robot circuit,and the test circuit board runs stably,and the CPU chip has no obvious heating phenomenon,which meets the requirements.3.Image acquisition module design.According to low-cost,high-throughput design principles.Determine the equipment selection.A seedling image acquisition algorithm based on Python language was developed,and the algorithm workflow was to carry out single-point acquisition after the host computer detected the arrival signal sent by the Raspberry Pi.After the collection is completed,the completion command is issued to the Raspberry Pi to make the robot go to the next point;When running to the end point,receive the end command feedback from the Raspberry Pi to stop the job.Using Lab VIEW to build human-computer interaction software,the robot operation status can be monitored in real time and the detection information can be displayed,ensuring the safe and stable operation of the robot.4.Automatic navigation system design.According to the robot working environment,laser SLAM technology is used as the robot automatic navigation scheme,and the equipment model used for navigation is determined according to the robot design principle.The automatic navigation algorithm was developed based on the ROS operating system and deployed in the Raspberry Pi.Based on the Gmapping algorithm,the navigation map is established,and the lidar collected data is preprocessed to match it with the robot hardware to obtain the optimal mapping effect under low-cost conditions.Based on AMCL algorithm,robot self-positioning is realized;Based on Dijkstra and DWA algorithms,robot path planning is realized,and robot motion parameters are configured to make the path planning effect conform to the working scene.The operator can select the robot operation point in real time through the ROS interface or pre-set the robot working point by the algorithm to complete the robot fixed-point image acquisition operation.5.Robot test and performance analysis.The robot prototype was tested with the accuracy test of navigation target point,endurance time test and seedling image acquisition test,which proved that the robot platform design achieved the expected performance.The test shows that the minimum lateral deviation of the robot navigation and positioning accuracy is 0.03 m,the maximum lateral deviation is 0.14 m,and the average lateral deviation is 0.089 m.The minimum longitudinal deviation is 0.02 m,the maximum longitudinal deviation is 0.17 m,and the average longitudinal deviation is0.079m;when fully charged,it can run continuously for more than 3h;the platform mounts the depth camera realsense D415 and the micro-host for seedling phenotypic information collection,and can detect 1000m2 area per hour.After the test of image processing algorithm,the image quality meets the requirements of the algorithm.The robot can automatically collect seedling canopy images according to the needs of researchers,and provide data support for seedling phenotypic detection.On the basis of the existing technology,it is expected to further expand the growth status detection function of seedlings and become an intelligent detection tool for seedling factories,which has good application prospects.