Research On Flying Target Detection Method Based On Push-broom Multispectral Images

With the development of aviation industry,aircraft represented by civil aircraft has gradually become a common thing.Therefore,the monitoring and management of aircraft has gained extensive attention.The traditional aircraft monitoring.method uses the Air Traffic Control Radar Beacon System as the signal source,and the radar base station is easily restricted by the terrain.Later,the automatic dependent surveillance-broadcast is used for aircraft monitoring.In recent years,with the continuous advance of sensor manufacturing technology,optical remote sensing technology has got gradually mature.As one of the most common methods to observe earth,multispectral remote sensing technology has been widely applied in many important fields,such as urban planning,camouflage recognition and target detection.At present,target detection in multispectral images is mainly focused on stationary targets,and there are relatively a few researches on moving targets detection.Moving object detection based on multispectral image faces the main problem that time and displacement information of moving target is missing.To the above problems,this paper studied the multi-spectral imaging principle,and use parallax characteristic of the push-broom sensor,aimed to make the most of the spatial-temporal and spectral information of the object for target detection,the main research contents are as follows:Firstly,with the beginning of the imaging principle of the multispectral sensor,the imaging process of the push-broom sensor represented by the Sentinel-2 multispectral sensor is studied,and the parallax characteristics of the push-broom sensor are obtained.By analyzing,it is found that the aircraft and other man-made objects in flight have strong reflection characteristics.Based on this,the characteristics of spectral distortion,rectilinear motion and displacement proportional to time are studied,and the feasibility of the flying target detection is analyzed,which lays a foundation for the following detection.Then,according to the spectral distortion characteristics of the flying target,the flying target detection based on push-broom multispectral image is studied.Superpixel segmentation is used to cluster the images to obtain the purified background.And,the fractional Fourier transform is used to enhance the target and expand the distinction between the background and the target.Finally,combined with the spatial information,a collaborative representation Reed-Xiaoli(RX)anomaly detection algorithm based on significance weighting is proposed.By comparing with the traditional anomaly detection algorithm,the good performance of the proposed algorithm is proved.Finally,with the spatial and temporal information of the object,the moving detection algorithm based on the background estimation and the difference between bands is used to realize the object detection.Combined with the result of anomaly detection and the direction and time shift characteristics of the target,the false alarm is removed,the accurate extraction and motion estimation of the target are realized.By comparing the results of motion estimation with the reference data,the validity of the proposed method are verified.