Research On Small Target Detection Of Staring Sequence Images

In recent years,with the rapid development of science and technology,the safety problems of China’s far-sea coastal defense and maritime traffic have become increasingly prominent.Therefore,there is an urgent need to improve China’s maritime surveillance capability.In response,China has added the geostationary orbit highresolution staring satellite(GF4)to the high-score major project.However,there are still many problems in the maritime monitoring,especially in the monitoring of ships in the far sea.On the one hand,the detection results are poor due to the small proportion of the target,the indistinct maritime background and the interference of cloud and fog.On the other hand,there is currently no relevant method to apply the high temporal resolution and imaging characteristics of GF4 staring image to target detection.Aiming at the above problems,this paper mainly studies the detection method of small and small targets at sea from the following three aspects:(1)GF4 data exists same problems of excessive redundant information irrelevant to ship target detection and low target local signal-to-clutter ratio.On the basis of water extraction,a multi-band cloud detection method based on spectral difference and cloud motion characteristics was studied based on GF4 waveband imaging characteristics,which effectively eliminated redundant information irrelevant to detection in the image.Further,aiming at the problem of low local signal-to-clutter ratio,a fog removal method is studied,which effectively improves the contrast between target and background,and also provides refined data for subsequent target detection.(2)There are some problems in GF4 data,such as the missing of target space information and the serious influence of high noise on detection results.Aiming at these problems,on the one hand,based on the low-rank characteristics of GF4 sea images,the slice tensor small and weak target detection algorithm based on local weighting is studied.On the other hand,the spectral reflectance difference of different bands and the time difference between bands are used to eliminate the false targets and random noise which have great difference with the real target spectral reflectance,which can effectively reduce the false alarm rate of a single frame image target detection.(3)Aiming at the problems of large time difference between frames and poor target correlation in GF4 data,a dynamic small and small target detection algorithm based on frame difference method combined with single frame detection was studied,and static false targets such as islands were preliminarily eliminated,taking the inter-frame correlation of staring sequence images as the starting point.Furthermore,an optimization algorithm based on kernel correlation filter tracking and motion smoothing constraint is proposed based on the motion characteristics of Marine targets,which can effectively eliminate false targets while missing the target correction.