Research On Target Detection Method Of Low-altitude Slow Infrared Unmanned Aerial Vehicle In Complex Background

Small Unmanned Aerial Vehicle(SUAV)is emerging non-manned small flying vehicle.The rapid development of SUAV has brought great convenience as well as great security risks.In recent years,the frequent "black fly" of drones has caused great threat to public social security.In the military field,SUAV has quickly become a powerful combat force.Therefore,the demand for anti-SUAV monitoring and counter-systems is unprecedentedly high,and the detection of SUAV targets is the basis and focus of the system development.In this paper,the infrared detection method is used to capture the the SUAV target,and the infrared SUAV target detection method under complex background is developed for the anti-SUAV monitoring system requirements for fixed area security.In order to study the infrared characteristics of the SUAV target and improve the speed and accuracy of the SUAV detection algorithm,the main work of this paper is organized as follows:(1)In order to make up for the vacancy of the current infrared SUAV target data set,meet the requirement of target characteristic analysis,and the training requirements of the deep learning algorithm,the SUAV data set for target detection experiment are carried out.(2)In order to analyze the infrared characteristics of the infrared SUAV target image,firstly,the infrared characteristics of the SUAV target and its background are modeled according to the Planck’s law and the Fresnel formula,and the imaging principle and complexity are analyzed.Then,The measured data obtained by the SUAV detection experiment is used to analyze the infrared radiation characteristics,motion characteristics and visual saliency of the SUAV target in the infrared image,and provide theoretical guidance for the subsequent algorithm design.(3)In order to balance the speed and accuracy of infrared SUAV target detection algorithm and improve the recall of small targets,this paper proposes a real-time detection method of infrared SUAV based on multi-scale feature map fusion.This method combines a multi-scale feature map fusion module based on dense horizontal connection and a sliding window search module based on target statistical features.The single-stage detection framework is uesd to achieves high real-time performance under the condition of ensuring detection accuracy.(4)In order to solve the problem of accurate detection of infrared SUAV in complex background,this paper proposes an accurate detection and segmentation method based on the rotationally unconstrained region proposals.The model combines a rotationally unconstrained region proposal network,a rotating region of interest region-align layer,and adaptive non-maximum suppression approach.Aiming at the problem of high background complexity in the target neighborhood,the idea of rotating area is proposed,and the multi-task joint learning method is combined to realize the parallel output of the target category,position bounding box and segmentation mask,which provide the prerequisites of advanced identification such as model identification,attitude estimation,and behavior analysis provides prerequisites.The research and related works in this paper provides theoretical and technical support for the development of anti-SUAV infrared video surveillance system for fixed area protection.It has a great significance to promote the development of anti-SUAV technology and infrared target detection technology.