Research On Infrared Small Target Detection Algorithm Based On Visual Saliency Mechanism

Infrared imaging detection technology has an important role in infrared search and tracking,precise navigation,early warning,etc.It is used in military strategy,national defense,industrial production and other fields widely.Infrared small targets generally occupy fewer pixels,while being affected by various factors of noise.Infrared small target detection technology has become a very significant role in the technology,and is a research problem.Many researchers have studied and concerned about this.The application of human visual function to the infrared small target detection process,and the use of attention selection mechanism of the human vison system can achieve better detection results.Because the target in the infrared image is the significant region in the image,it can cause the human visual priority to be selected and extracted,and the non-significant region is suppressed,and the information such as the infrared small target position is obtained.Based on the study of previous research algorithms,this paper analyzes the characteristics of infrared images and the target characteristics and background characteristics in infrared small target images,and proposes the following infrared small target detection based on visual saliency algorithm.The main contents are summarized as follows:(1)Based on the improved morphological reconstruction,the method based on the visual contrast mechanism and the method based on the image transformation domain significance information are proposed.Using the improved morphological reconstruction algorithm,it can suppress most of the background in the image.In the local contrast algorithm based on the visual contrast mechanism,the region of interest is extracted quickly to reduce the computation and the probability of error in the subsequent local contrast metric.The local contrast ratio algorithm is defined,and the difference between the gray value of the target pixel and the pixel in the surrounding is made to form a significant graph,and the target is extracted from the background.Based on the algorithm of saliency information of image transformation domain,the saliency information of the image in the transformation domain is extracted by the gray level residual,the gradient residual,and the phase spectrum information,and the information such as the infrared small target position is obtained.The experimental results illustrate that the two methods have satisfied results and can effectively extract the target component.(2)A target detection method based on two-dimensional empirical mode decomposition is proposed.The two-dimensional modal decomposition algorithm is used to decompose the image to obtain the detail information.According to the obtained detail component diagram,the feature extraction is carried out.The weighted Gaussian difference kernel function and the high-pass gradient kernel function are used to analyze the different detail components.In order to obtain the saliency information of the image,which is mainly to get the salient target,and suppress the background noise,the saliency feature map of each detail component is assigned the weight value.The saliency information contained in each component is different to fuse.According to the degree of display of the usefulness information of each detail component,the contribution of the significant graph of each component to the saliency graph of the final infrared small object is determined to achieve the suppression of background noise and highlight the target intensity.The simulation results illustrate that the method has preferable detection effect on the infrared small target image with complex background,which can enhance the signal of target and suppress the noise.(3)A target detection method based on weighted multi-directional gradient information is proposed.The gradient information of the image in different directions is obtained to achieve the purpose of analyzing the significance information in all directions of the small target.According to the gradient information of each direction,the derivative information of different directions is obtained by deriving the gradient information in the corresponding direction based on the Facet model,and the detailed information of the image is further displayed.The improved local information entropy is used to construct the weight of the image significance graph.Since the improved local information entropy can reflect the degree of gray information distribution in the image and highlight the bright gray information,the saliency of the image is expressed from local information entropy.to guide the construction of the gradient direction of the derivative map.It can achieve a better highlight of the infrared small target salient characteristics of the effect.From the experimental results,the method can effectively suppress the background noise of the infrared small target image,and enhance the significance information of the target,and can separate the target from the background,and can achieve satisfactorily detection in the analysis of results.