Study On The Extraction Of Saliency Feature Of Infrared Small Target

Today,the progress of military science and technology has had a great influence on the pattern and form of the modern war.Infrared small target detection is a key technology for infrared guidance,infrared tracking search and early warning system.However,Infrared imaging systems are affected by many factors such as weather,clouds and filed view,which result in more clutter in infrared images.Therefore,infrared small target detection technology under complex background has always been a challenging and difficult research hotspot.Because the small target has a significant singularity,the target detection probability can be effectively enhanced by extracting the significant features of the infrared small target.Therefore,starting from the visual attention mechanism,this paper deeply studies the significant features of infrared small target in terms of fractal,grayscale,direction,and entropy,which is used for infrared small target detection.The main work and research results are as follows:1.Fractal theory can describe the self-similarity of an area in an image,which is a more reasonable and effective approximation model.The fractal dimension can describe the roughness of the image surface,and the target and background can be effectively distinguished by calculating the fractal dimension.This chapter constructs the fractal features of infrared small target by calculating the improved Hurst index feature,and applies it to infrared small target detection.An infrared small target detection algorithm based on fractal feature is proposed.The algorithm uses the discrete fractional Brownian random field model to describe the natural background in the infrared image,and the improved Hurst index to the fractal characteristics of the infrared small target.At the same time,the background noise is further removed according to the idea of OTSU and the saliency of the target is enhanced.Finally,experiments were performed in different backgrounds and compared with different algorithms,the superiority of the algorithm can be proved by analysis of performance parameters.2.Infrared images have only one channel of grayscale,which provides relatively little information.Therefore,it is very important to study the grayscale features of infrared small targets.Since the gradient of the small target area varies greatly in each direction in the infrared image,and the contrast value between the small target and the neighboring area is relatively large based on the humane vision contrast mechanism,therefore,this chapter studies the gradient and local contrast,constructs new infrared target gradient feature and local contrast feature which are used for target detection.This chapter proposes an infrared small target detection algorithm based on gray features.The algorithm is mainly studied from the aspects of gradient feature and local contrast feature.First,a multi-directional gradient feature is proposed to extract the gradient feature of small target from multiple directions.Then,local contrast is used to calculate the local contrast feature of the small target.Finally,experimental verification and data analysis show that the algorithm has good robustness and detection performance.3.In the infrared small target detection,when the algorithm is based on a single feature,the robustness of the algorithm is poor.Therefore,in order to use more information of small target to increase the reliability of test results,this chapter extracts the features of different aspects of small targets,and proposes a new infrared small target detection algorithm based on the idea of feature fusion.Firstly,the algorithm uses an improved high boost filter for preprocessing.Then,extract the directional feature,spectral feature and entropy feature of the infrared small target,respectively.The algorithm adopts a normalized feature fusion method to enhance target saliency and suppress background.Finally,experimental verification shows that the algorithm can effectively achieve infrared target detection and improve detection probability.