Moving Target Detection For Multi-Spectral Imaging Of Motion Platform

Target detection and recognition of satellite remote sensing images has always been a research hotspot in the field of image processing,and it has important military significance and civilian value.The pushbroom imaging system uses the satellite platform’s own motion to acquire remote sensing images along the flight direction,which is widely used for Earth observation.The Multi-line Multi-Spectral Pushbroom Sensors(MLMS-PBSs)are composed of a plurality of Pushbroom Sensor(PBS)with different spectrum and parallel to each other.Each sensor collects remote sensing images in different spectral ranges and at different times.And multispectral images obtained provides rich spectral features for target recognition.Before the recognition,the small moving targets in remote sensing image obtain by MLMS-PBSs need to be detected.Because the multispectral image obtained by MLMS-PBSs of the visible spectrum has complex background,low signal-to-noise ratio,weak target and no shape information,the target detection of single image is difficult.But the resources on the satellitare limited,the Multi-line homomorphic sensors cannot be installed.So the work of this thesis is mainly to study the weak target detection technology in multisprectral image sequences.This thesis first studies the registration technique of multisprectral images.Because in the multisprectral images sequence image processing,the imaging positions of different spectrum are different,and the individual sensor parameters cannot be kept consistent,resulting in deviations in the spatial position between different spectrums at different times,so it is necessary to register the multispectral images.In this thesis,a local self-similarity feature point pair matching algorithm based on block Harris corners(LSSBH)is proposed.Firstly,the algorithm extract Harris corner points by dividing the image into blocks,and then uses local self-similarity to describe the corner features and obtain matching feature point pairs.The algorithm effectively solves the problem of large registration deviation of different spectrum images based on gray information.In this thesis,the simulation image is used to verify the effectiveness of the algorithm.The experimental results show that the proposed new algorithm can complete the registration processing of different spectrumimages with higher precision.Based on the registration of different spectrum images,this thesis also studies the detection algorithm of weak and small moving targets in Multisprectalimage sequences.In this thesis,a Multisprectalimage sequence weak moving target detection algorithm(MSGMA)based on Gaussian model approximation is proposed.The algorithm improves the background suppression of Multisprectalimages based on the moving target recognition algorithm(MTI).The algorithm effectively solves the problem that the background between heterogeneous images is difficult to suppress.The experimental results show that the proposed new algorithm can effectively achieve background suppression of heterogeneous images and extract weak targets of motion.