Height Detection System For Agricultural UAV Based On Multi-sensor Fusion

Agricultural UAV is one of the important representatives of agricultural aviation industry in China,which has developed rapidly in recent years.The research of many scientific researchers has shown that the operational height of agricultural UAV has an important relationship with the operational quality.In the field of agricultural unmanned aerial vehicle monitoring,research on height monitoring at home and abroad is relatively mature,but there is little research on the height monitoring of agricultural unmanned aerial vehicles from the crop canopy.Aiming at the characteristics of weak monitoring reliability and stability of a single height monitoring sensor,a multi-sensor fusion height monitoring method based on the maximum value algorithm combined with BP neural network was proposed,which was suitable for agricultural UAV planted in pieces and the canopy Realtime operational height monitoring on sparse crops(such as sugar cane,rice,etc.).Theoretical research and experimental analysis were focused on the characteristics analysis and selection of height monitoring sensors,the design of multi-sensor fusion algorithm,the construction of height system monitoring system,and artificial neural network modeling.The thesis mainly completed the following work and reached corresponding conclusions:(1)In the feature analysis and selection application of altitude monitoring sensors,the measurement accuracy and terrain following performance of three commercial altitude monitoring sensors were studied.The experiment found that the millimeter wave radar and lidar had good terrain following performance and high measurement accuracy after correction.Ultrasonic radar monitoring under simulated high-speed horizontal movement of the UAV would have obvious data errors,and the sound wave monitoring principle made it less time-effective,and it was difficult to meet the requirements of high stability and realtime monitoring under high-speed operation of agricultural drones.When the movement direction and the detection direction were in the same straight line,the measurement accuracy was high,and it could be used in the hovering and lifting of the drone.At the same time,the terrain following performance detection test verified the weak reliability and stability of the monitoring data of a single altitude monitoring sensor.(2)The monitoring method of the ultra-low altitude real working height of agricultural drones was studied,and the height monitoring method based on the maximum value algorithm combined with multi-sensor fusion technology was proposed.The system was built to realize the real working height monitoring.The information and fusion solution took a total of 15.7ms,and the data update frequency could reach 50 Hz,meeting real-time requirements.Compared with a single height monitoring sensor,the monitoring reliability and stability were significantly improved.The sample standard deviation of the effective collection information of each test group was less than 11 cm,with a decrease of more than70%.Taking the calibration value as the reference value,the mean absolute error decreased by a multiple of 16.2cm at most.The difference of extreme value decreased from more than200 cm to about 50 cm.The maximum relative error was 11.8%.(3)Aiming at the shortcomings of the height monitoring system,an error correction method based on back propagation neural network was studied.The corresponding experiment was designed to select the error compensation parameters,and the sample data was collected to train the neural network model.The results showed that the model was stable and reliable,the correlation coefficient reaches 0.994,and the performance index was 0.959.After inspection,the model could identify the outliers of collected information and made corrections.The effectively collected samples of each test group were input into the model.The standard deviation of the output samples does not exceed 3.51,the extreme value difference was about 20 cm,The mean absolute error was up to 5.3cm,and the maximum relative error was only 4.3%.All the stability data index parameters have increased by multiples,which met the requirements of high monitoring accuracy.The model could verify the feasibility of combining the maximum value algorithm with the BP neural network algorithm.(4)The Beidou navigation Satellite system was designed and assembled by myself.The results of the design test analysis showed that the relative height error was detected in the centimeter range,and the static maximum relative error was 2.67%;the dynamic monitoring stability was strong,and the test extreme value difference was 8cm.The path was basically the same and there was no deviation from the trajectory.The Beidou navigation Satellite system had high altitude and route monitoring accuracy,and was indispensable in the stability control of agricultural drones and route planning.It is one of the important technologies for achieving precision agricultural aviation.