Research On Infrared Small Target Detection Algorithm Via Tensor Recovery

Infrared imaging is widely used in military,industry,agriculture,transportation and other fields because of its advantages of high concealment,long working distance,strong anti-jamming ability and long working time.With the development of science and technology,the number of infrared images acquired every day increases geometrically,only processing of these infrared images by manpower has been inadequate.Therefore,many infrared image processing algorithms came into being,and infrared small target detection algorithm is one of the research hotspots.Infrared small target detection technology can be applied to equipment injury detection,medical screening,unmanned assistance,monitoring and early warning and other aspects,can greatly reduce the cost,save manpower material and financial resources,has a very important research significance.However,due to the long distance of infrared imaging,the obtained infrared images are fuzzy,with low resolution and more noise.The target in the image is small,and there is no shape,texture and other features.The pres ence of clutter in the background makes the signal-to-noise ratio of the image low and it is difficult to detect the target.Therefore,to solve these problems,we propose a new infrared small target detection mode.The main contents are as follows:(1)Since there are few features in the infrared image,the adjacent frames of the infrared image are constructed as tensors as input in this paper to make full use of their space-time characteristics.As the target in the infrared image is sparse,background has low rank,in this paper,the infrared small target detection problem as a tensor recovery problem to solve,and use the double-nuclear norm instead of the traditional single nuclear norm as the unwinding of the rank function,solved the problem of the suboptimal solution deviation from the original solution,the better approximation rank minimization.In addition,in order to make full use of the priori knowledge of the target,in this paper joined the morphology of regularization term in the objective function,and using the weighted ring structure element instead of the traditional structure element,to make better use of the target information and its surrounding background between the local information.Finally,experiments are carried out on 5 sets of real sequences.Compared with other baseline methods,the proposed algorithm can enhance the target and suppress the background more effectively,and ensure a higher detection probability and a lower false alarm rate.(2)In recent years,due to the redundancy of the basis function of the wavelet frame,compared with the basis of the wavelet transform,the frame transform can describe the details more finely,and compared with the Fourier transform,the framelet transform can better approximate the multirank tensor,so it is often used in image restoration and denoising.Infrared small target detection can be regarded as a background image restoration,and the weak small targets under complex background is easy to be overwhelmed with background of heavy clutter.Therefore,it is particularly important to target detail depict.So this paper puts forward a kind of infrared small target detection model based on framelet transform.The definition of frame transform was introduced in detail,using the ADMM was used to solve the objective function,and the experimental results show that compared with other methods of baseline and the algorithm of weak small targets enhancement effect is better,for complex background suppression effect is also better.