Research On Infrared Small Target Detection Algorithm In Heterogeneous Complex Background

Dim and small targets detection in infrared(IR)images with complex heterogeneous background is an important research direction in the field of target recognition and detection,which is widely used in military and civil fields.The heterogeneous background,such as complex clouds,sea waves etc.,usually exhibit unique characteristics,including dramtic gray change of local region,no specific shape features,low signal-to-noise ratio,and heavy clutter,which make the infrared targets detection under complex heterogenous background still difficult problems.Unfortunately,there is no such specific research work in the open literature.When processing heterogeneous background IR image,the current mainstream detection algorithms usually exhihit many drawbacks,including high false alarm rate,low detection rate and poor stability.It is difficult to meet the requriements of practical applications.In this dissertation,local features of gradient vector field,local features of facet-kernel based directional derivative,and inter-region edge features of heterogeneous background are analyzed,and a series of infrared small target detection algorithms for heterogeneous background are proposed.In order to solve the serious false alarm problem caused by gray-statistics-based local contrast methods,the features of targets region,local background region and clutter region in IR gradient vector field are analyzed,and an infrared small target detection method based on local gradient field feature contrast measure are proposed in this dissertation.First,the modulus features and direction features of local gradient field are extracted,and then a local contrast measure based on gradient field is proposed,finally,the real targets are confirmed by calculating the flux of each candidate target.The experimental results indicate that the detection performance of the proposed method is superior to that of LCM-based method.Heterogeneous background such as complex cloud background and sea wave background usually exhibit typical directional features,to eliminate this features,a small target detection method based on facet kernel and directional derivative features is proposed in this dissertation.First,the directional derivative images are obtained by using a fast facet-kernel-based directional derivative algorithm.Then,a LCM-based multi directional derivative saliency maps are proposed.Finally,the fused saliency map is segmented,and the targets are confirmed and marked.The experimental results indicate that the proposed method can reduce the false alarm rate and enhance detection rate significantly,which can be applied to most of the practical application scenarios.In order to reduce the false alarm rate of the current state-of-the-art method when processing the heterogeneous background IR image,an infrared small target detection algorithm based on adaptive multi local contrast measure is proposed in this dissertation.Most of human-visual-system-based IR small target detection algorithms only use the uniform contrast measure to replace the current pixel(the local features of heterogeneous background is not considered),which leads to the appearance of false alarm.In order to solve this problem,the local features of region,where the target located at the edge of different heterogeneous background,are investigated.The surrounding region of the target exhibits some unique features,including dramatical grayscale variation,small signal-to-noise ratio,and complex background edge.On the basis of these local background features,an infrared small target detection algorithm based on adaptive multi local contrast measure is proposed.For different local background,the proposed algorithm can generate local contrast measure adaptively.The experimental results indicate that the proposed algorithm can effectively reduce the false alarm rate,and has high practical value.The infrared small target detection algorithms based on the low-rank and sparse matrix(tensor)decomposition are current research focus.The core of these algorithms is how to obtain the input data model which satisfies the low-rank and sparse hypothesis.The original image is directly selected as input data in these algorithms,which can not meet the assumption,and the detection performance needs to be further improved.An optimal infrared patch-image model for small target detection in complex heterogeneous background is proposed in this dissertation.The model can effectively retain the original data and exhibit a strong low-rank and sparse characteristic.In addition,an adaptive inexact augmented Lagrange multiplier algorithm is proposed to solve the convex optimization problem of RPCA.The algorithm exhibit strong robustness and can significantly improve the detection rate,which can meet the needs of small IR target detection with various backgrounds.