Research On Multi-rotor Unmanned Aerial Vehicle Identification Method Based On Deep Learning

With the development of the science and the technology of Multi-rotor unmanned aerial vehicles(UAVs),which gradually expand from military field to civil field.They're extensively used in various domains such as remote sensing mapping,agriculture,searching,amusement,logistics and so on.While bringing convenience to people's lives,UAVs also bring Some hidden danger,for example,causing public order issues to airport and railway.Issues such as invasion of privacy,physical injury,and smuggling narcotics can be caused by illegally drones using.Based on the above series of incertitude problems,unmanned aerial vehicle identification and monitoring technique possesses broad application prospects.So,a UAV detection method based on deep learning is proposed.This paper mainly focuses on the identification and detection of Multi-rotor UAVs.At first,the problem of poor recognition ability and inaccurate detection of Multi-rotor UAVs by object detection method is analyzed,then,some improvement suggestions are proposed for the existing methods and a UAV identification system is built.The main works of this thesis include:(1)This paper briefly introduces the development status of object detection,and the development of traditional object detection technology based on deep neural network,the advantages and disadvantages of traditional algorithm and neural network algorithm.Moreover,the relevant theoretical knowledge of the deep neural network,which provides the theoretical basis for the follow-up research is also introduced.(2)In this paper,we study the residual learning mechanism and the structure and principle of the feature pyramid.According to these theories,we design a new convolutional neural network(MC-ResNet),and introduce the composition and training methods of the network model in detail.By comparing with the Existing algorithm,the experiment proves that the network model can fully integrate the context information and improve the object detection accuracy.(3)This paper expounds the principle and structure of SSD detection network,analyzes the advantages and disadvantages of SSD detection network,and improve SSD detection network.In addition the experiment proves that the network improves the speed and precision of object detection effectively.(4)The detection efficiency of traditional UAV detection system is poor,it has defects such as frame missing and frame dropping occur during detection.Based on the previous research results,this paper builds a multi-rotor UAV identification system based on deep convolutional neural network.Through the tests and analysis of the system,the system meets the design requirements,and realizes the high-precision and high-efficiency identification and detection of UAV.