Study On Obstacle Avoidance System Of Agricultural UAV Based On Deep Learning

Grape is one of the fresh fruits.It is very important to monitor its nutritional needs and pest status in time during the planting process to improve the quality of grape.With the rapid development of modern science and technology,new progress has been made in agricultural modernization and informationization.Agricultural UAVs gradually show their superiority and potential for application and development in agricultural production.The introduction and use of agricultural UAVs in grape production has improved the efficiency of grape cultivation,but at the same time it also faces the problems of safe operation of UAVs.Therefore,the real-time autonomous obstacle avoidance of agricultural UAVs has become an important research focus in the field of agricultural informatization and automation.It has been reported that most researchers use different sensors to detect the distance between UAV and obstacles in flight environment,and adopt corresponding obstacle avoidance methods.However,there are many problems such as limited detection distance and high cost.Therefore,this paper uses the latest hotspot technology in the field of artificial intelligence-depth learning method to explore the intelligent algorithm of agricultural UAV target detection,and constructs a real-time and accurate obstacle avoidance application system.Firstly,on the basis of discussing the typical convolutional neural network structure model of in-depth learning,this paper analyses the target detection algorithm based on region nomination,and develops the data set of farmland obstacles,which mainly includes seven categories: people,trees,birds,wire towers,buildings,agricultural aircraft and kites.Secondly,on the basis of in-depth analysis of target detection algorithms without region nomination represented by YOLO series,a new intelligent target detection algorithm combining region nomination target detection algorithm with region nomination-free target detection algorithm is proposed.Finally,based on the above improved intelligent algorithm and experimental research,an obstacle avoidance system for agricultural UAV based on target detection algorithm is designed.Experiments show that compared with typical obstacle avoidance systems such asultrasonic obstacle avoidance system and RTK technology,the cost of obstacle avoidance method adopted in this paper is reduced;compared with existing target detection algorithms,the improved intelligent obstacle avoidance algorithm in this study is 17.2% higher than Faster R-CNN's mAP,and the detection rate is 14 FPS faster;while guaranteeing real-time performance,mAP is higher than YOLO 2's.23.3%,6.25% higher than YOLOv3.At the same time,an obstacle detection system for agricultural UAV is designed,which is suitable for farmland environment.It provides a new method for real-time obstacle avoidance of UAV.