Design And Implementation Of UAV Obstacle Avoidance Based On Synergetics And Transfer Learning

Because of Unmanned Aerial Vehicle(UAV)are widely used in various fields,such as aerial aerial photography,inspection,reconnaissance,etc,the field of UAV has become a research direction which is a major topic of debate in recent years,and the technology of UAV has been booming.In the development of UAV technology,how to avoid obstacle automatically is one of the emphases for research.It mainly detects the obstacles appearing in the flight path of the UAV through various sensors work alone or together,and adopts appropriate obstacle avoidance strategies ensuring the UAV fly safely.Today,deep learning is one of the hot directions in the field of science,various of emerging theory based on deep learning can be seen oftenly.The research topic of this thesis is realizing automatic obstacle avoidance by using synergetics and transfer learning.The constructed deep synergetic neural network has improved performance significantly compared with the traditional synergetic neural network,and expands its function of target detection,and finally applies it on the UAV obstacle avoidance system.First,this thesis introduces the related concepts and basic ideas of synergetics,and the synergetic neural network based on the idea of synergetics,and the mathematical model and operation process of the synergetic neural network are expounded.Then introduces the basic concepts and implementation methods of transfer learning,and expounds using model migration to realize migration learning.This method take advantage of characteristic of deep convolutional neural network,which extract features from the sample,the general features are stored in low layers and the high layers store classification features.The target detection algorithm SSD based on Inception pre-training model is used to achieve target detection,and the fully-connected layer of the pre-training model is redefined,so that the pre-training model can complete the detection and recognition function in the target domain and tasks.Then,completing the design of UAV obstacle avoidance algorithm based on synergetic neural network and algorithm SSD.Combined with the boundary regression idea of the target detection algorithm SSD,this thesis design an obstacle avoidance algorithm based on the deep synergetic neural network.Firstly,the obstacle is detected and its position is determined.Then completing obstacle recognition by using deep synergetic neural network.And the prototype pattern is generated by using generative adversarial network to participate in the evolution of the network.On the basis of the synergetic neural network,increasing the number of network layers combined with the idea of deep learning to achieve better recognition results,it provides theoretical support for the subsequent detection and recognition algorithms.Finally,completing the overall design and implementation of the UAV obstacle avoidance system.According to the relative position of the obstacle in the field of view,a corresponding obstacle avoidance strategy is formulated to enable the UAV can complete filght task safely.In addition,a simple upper computer is designed to control the drone and airborne motion camera.In the end,the actual flight test of the UAV is carried out,verifying the feasibility and practicability of the system.