Study On Spectral Characteristics Of UAV Target In Low Altitude Background And Application Of Target Detection

In recent years,with the rapid development of unmanned aerial vehicle（UAV）technology and the gradual opening of low-altitude airspace,the application of small UAVs has spread throughout all walks of life.However,the “black flight” and“indiscriminate flight” of consumer UAVs pose a serious threat to the safety of low-altitude airspace.The existing radar detection,radio frequency detection,photoelectric detection and other methods have the problems of missing detection and false detection for non-metallic material UAVs with transmission characteristics and small volume under complex background and complex weather.In view of the above problems,this paper introduces multispectral imaging technology,taking the spectral characteristics of target materials as the breakthrough point,and studies the spectral characteristics and target detection application of UAV targets in low altitude background.Through short-wave infrared multispectral images,the spectral characteristics difference between UAV targets and common targets in low altitude background and the imaging characteristic bands of UAV targets in low altitude background are studied and analyzed.Combined with deep learning method,the efficient detection application of UAV targets is realized,which provides new technical support for UAV target detection in low altitude background.The main research contents of this paper are as follows:（1）Study on the spectral characteristics of UAV materials.At present,the popular manufacturing materials of consumer UAVs in the market are mainly engineering plastics.Through the research and analysis of the infrared spectrum analysis principle and the infrared spectrum characteristics of engineering plastics,it provides a theoretical basis for the study of UAV targets in the short-wave infrared band,and also verifies the feasibility of using multi-spectral imaging technology to realize the detection of small UAV targets in the low-altitude background.（2）Aiming at the problem that there is no public UAV multispectral image data at present,the principle of multispectral imaging and the principle of multispectral imaging based on LCTF are studied and analyzed.The multispectral imaging system based on LCTF is independently built in the laboratory.After imaging test and comparison,the imaging method of LCTF placed before the combined short-wave infrared camera is finally adopted,and it is also verified that the multispectral imaging system can carry out effective data acquisition.（3）Aiming at the problem of low altitude target reflectance spectral library construction and target recognition based on multispectral images,the data characteristics of short-wave infrared multispectral images of UAV targets and other common targets and various materials under static background and the necessity of selecting regions of interest are studied and analyzed.The reflectance spectra of various targets are calculated by reflectance spectral reconstruction model,and a small reflectance spectral database is established.A typical spectral matching method based on spectral angle,minimum distance similarity and spectral information divergence is proposed to identify the target category and material of UAV targets.The experimental results further verify the effectiveness of the target reflectivity spectrum calculated based on multispectral images and the feasibility of using spectral characteristics to realize target detection and recognition of small UAVs.（4）Aiming at the selection of imaging characteristic band of UAV target in low altitude background,the characteristics of short-wave infrared multispectral image data of dynamic UAV target in low altitude background are studied and analyzed.According to the different performance of target signal strength and background signal strength in different bands and the selection criteria of target imaging characteristic band,the selection method of characteristic band based on image information content,band index method and image brightness characteristics is proposed.The experimental results show that the proposed method is effective.The bands of 1000 nm,1020 nm,1080-1110 nm,1200 nm and 1260-1340 nm can be selected as the imaging characteristic bands of UAV targets in low altitude background.（5）Aiming at the problem of further application of feature band image in UAV target detection,a UAV target detection method based on feature band image and deep learning is proposed.Firstly,the low altitude UAV target detection method based on YOLOv3 is studied.The experimental results show that the average accuracy of 90.43%in the short-wave infrared feature band is higher than that of 87.43% in the visible band.It also verifies the advantages of feature band images in UAV detection compared with visible band images.Aiming at the problem of low detection accuracy of UAV small target in complex background,a UAV target detection algorithm based on feature band image and improved YOLOv3 is proposed.By improving the multi-scale feature fusion network and a priori frame clustering method,the receptive field is increased and the convergence speed of the model is accelerated.The experimental results show that the average accuracy of the proposed algorithm is 93.20%,which is 4.96% higher than that of the original YOLOv3 algorithm,and further improves the detection accuracy of UAV targets.