Research On Small Infrared Target Detection Algorithm Based On Low-rank And Sparse Decomposition

Compared with visible light imaging and radar imaging,infrared imaging(IR)can work day and night.Moreover,it can work in the rain,fog and other harsh climatic conditions.With these advantages of infrared imaging,there is no doubt that infrared imaging plays an important role in infrared guidance systems,early warning systems,and monitoring systems.As one of the key techniques in these systems,small infrared target detection has attracted a lot of attention and research in recent years.It is very important to achieve accurate small target detection over long imaging distance which would provide sufficient reaction time.Long imaging distance makes small targets usually lack concrete shapes and textures,resulting in small infrared target detection becomes more difficult.Although researchers have proposed many methods,there are still areas to improve.This paper studied small target detection in single frame under various scenarios and sequential-frame small target detection under complex scenarios:Firstly,the performance of current single-frame detection methods image,the detection performance degrades rapidly in some scenarios due to that they exploit only one kind of information(e.g.,local or nonlocal)while sacrificing the other.The proposed method combine local and nonlocal information,which firstly adopt dual-window model to calculate salient map,which exploits local information.Then,method of low-rank and sparse decomposition is applied to salient map,which exploits nonlocal information.Experimental results demonstrate that the combination of these two kinds of information is more effective for small target detection in different scenarios.Secondly,aiming at the problem that current sequential small target detection algorithm ignores the time domain structure information,the proposed method combining time domain structure information and target spatiotemporal continuity information.The method firstly performs time domain expansion on the sequence image,then low-rank and sparse decomposition is applied to obtain target prediction map.Subsequently,the target is further separated from the noise by a modified pipeline filter.Tested on three sequences in complex scenarios,the proposed method has higher probability of detection and lower false alarm rate than other methods.