Infrared Dim Target Detection Based On Visual Saliency

Infrared small target detection has been widely used in infrared early warning and precision guidance,etc.It is an important means in national defense construction.In order to meet the needs of war and gain more response time for the defense system,the target must be detected as early as possible.Because the infrared weak target has no color,texture and other features,and has a long distance and a small imaging area,this makes it more difficult to carry out the detection process.Therefore,how to find infrared small and weak targets in the environment with a long distance and many interfering factors,so as to buy more time for the defense system,is the top priority of national defense technology breakthrough.This thesis summarizes the infrared imaging mechanism,analyzes the interference characteristics,target characteristics and noise characteristics around the target,introduces several common infrared image preprocessing methods,and discusses several infrared small and weak targets detection algorithms.The main contents of the discussion include the following:(1)The infrared imaging principle is elaborated from the aspects of infrared radiation theory and infrared imaging system,and the mechanism of infrared imaging is described.Through mathematical modeling,the environmental characteristics around the infrared image target are analyzed.The methods of infrared image preprocessing are discussed in detail from the aspects of frequency domain filtering and spatial domain filtering.These related mechanism studies have laid a theoretical foundation for the proposed and improved infrared small and weak target detection algorithm.(2)In view of the traditional local contrast degree method through the whole image,one by one point calculation result in low efficiency and easily affected by the noise of false alarm point problem,put forward to improve the image of the local contrast method of infrared weak small targets detection based on image local entropy of significant figure within the scope of local contrast to improve the operation,greatly saves time,improves the computational efficiency,but also greatly reduces the false alarm.(3)The attention process of the human visual system can be regarded as the process of solving the significance part of the infrared image,which can be transformed into the sparse part of the gray matrix corresponding to the infrared image.We combined the infrared images containing weak targets into corresponding matrices by columns,and conducted sparse low-rank decomposition of the obtained matrices.The constructed sparse matrix was solved to obtain the significance graph,and the threshold segmentation of the significance graph was carried out to obtain the infrared weak targets.(4)Aiming at the problem that the spectral residual algorithm is simple in calculation and high in efficiency,but has certain limitations,an improved spectral residual algorithm is proposed.The horizontal and vertical features of the infrared image were extracted,and the spectral residuals were calculated in the horizontal and vertical directions.Finally,the infrared dim targets were obtained by fuzzy c-means clustering.