Design And Implementation Of UAV Obstacle Avoidance Based On Deep Synergetic Neural Network

Nowadays,rotary-wing Unmanned Aerial Vehicle(UAV)are widely used in various aspects of human life and production,and play an important role in aerial photography,rescue,and search.Obstacle avoidance is a research focus ofUAV flight.Its main purpose is to detect obstacles in its flight path and adopt certain evasive strategies so that it can avoid obstacles and allow UAVs to fly autonomously.Deep learning is one of the most popular topics in the field of science and technology.Research based on deep learning theory is also emerging.This thesis mainly studies deep synergetic neural networks built on the principle of synergetics and has greatly improved the performance of traditional synergetic neural networks.Compared with the currently popular deep neural network,the network model is smaller and the recognition speed is faster.The method of detection and identification applicable to the network is proposed and applied to the design of the obstacle avoidance system of the UAV to realize the recognition of the surrounding environment by the UAV,avoidance of obstacles,and the safety of UAV flight.First,this thesis introduces the basic ideas and key concepts of synergetic theory.Then the neural network model based on synergetics theory is introduced: Synergetic neural network.The mathematical model,structural model and operation flow of the synergetic neural network are described,and the characteristics of synergetics and synergetic neural networks are described in detail.Based on synergetic principle and the model structure of traditional deep neural network and PCA algorithm,a deep synergetic neural network is constructed,and the model structure,running process and algorithm steps of the deep synergetic neural network are described in detail.It provides sufficient theoretical support for the follow-up study of detection and recognition algorithms applied to deep synergetic neural networks.Secondly,this thesis analyzes the traditional sliding window detection method and the development course of regional nomination detection method based on convolutional neural network represented by Faster-RCNN.It also introduces the no-nomination end-to-end detection and identification algorithm YOLO,SSD,and through analyzing and summarizing the advantages of the above algorithms,proposes a detection and identification method suitable for deep synergetic neural networks.It alsocreates a sample database of obstacles and uses the sample database to analyze and verify the performance of the algorithm.Finally,based on the above theories and experiments,a UAV avoidance system based on deep synergetic neural network was designed.The obstacle avoidance scheme under different conditions was given,and the obstacle avoidance test was carried out for the actual flight of the UAV.Besides,the practicality and effectiveness of the system was verified.