Infrared Dim Small Target Detection Algorithm In Complex Sea Background

Infrared image plays an important role in modern military confrontation which is dominated by scientific and technological confrontation because of its passive detection imaging and all-weather uninterrupted work.Because of the wide use of infrared image,a series of military infrared image target detection problems have become one of the hot spots for researchers.Among them,the military equipment such as distant ships,missiles and torpedoes have the characteristics of small and weak targets in the infrared imaging system because of the inherent attributes such as long distance,weak signal and small range of targets.The detection of such targets has become a long-term problem that scholars pay attention to in this field.Due to the small difference between the target and the background,the detection of small and weak targets in this scene is difficult,with high false detection rate and false alarm rate.Therefore,it is urgent to explore more innovative and robust algorithms in target detection.In order to improve the reliability and stability of the detection system and realize the accurate detection of infrared small and weak targets,this paper analyzes the imaging principle and imaging characteristics of this kind of image.After reading and learning a lot of literature algorithms,it is decided to detect the targets in the image according to the overall idea of first single frame detection and then multi frame confirmation.The main content of the article is divided into the following parts.(1)Some classical single frame image processing algorithms are studied,the noise with different characteristics is analyzed,and the classical single frame image processing algorithm with pertinence is proposed for various noises,and its background suppression effect and target detection effect are analyzed.(2)In the single frame image processing,a single frame target detection flow based on the combination of local target enhancement weighted graph,global target background separation,local target confirmation and gray level restoration is proposed.Through the combination of local information and global information,the detection success rate of single frame target is improved.(1)The construction of local target enhancement weighted graph is the process of target enhancement preprocessing.Based on the local image matrix,the weighted graph uses Harris corner detection algorithm,improves the calculation formula of local corner enhancement weight factor through the physical meaning of matrix eigenvalue,traverses the whole graph by sliding window,and obtains the same size of the original image target enhancement prior weight Figure,two images are taken as the input of the next global target background separation process.(2)The global target background separation algorithm is based on the principal component analysis algorithm(PCA)commonly used in the field of image processing of small and weak targets.The traditional image block matrix is replaced by the image block tensor to reduce the problem of image layering after matrix processing.Through the difference of sparsity between the target and the background,the small and weak targets are regarded as sparse tensor,and the background is regarded as low rank tensor to establish the algorithm model The detection of infrared dim small target in complex background is transformed into an optimization problem based on low rank tensor and sparse tensor.Then,through the tight convex relaxation of the optimization problem,the approximate equivalent problem which is easy to solve is obtained.The improved alternating direction multiplier algorithm is used for iterative operation to separate the target and background,and the separated image is obtained.(3)Local target recognition algorithm and gray level restoration algorithm are used to remove the remaining stubborn noise and confirm the final target.The algorithm mainly consists of two parts.In the confirmation part,the improved double top hat algorithm is used to filter the target morphology and eliminate the false signals that the morphology does not conform to.In order to ensure the gray value and shape of small and weak targets are not destroyed,the gray restoration algorithm uses the target location image as the decision image,the target image obtained from global separation as the gray image,and the target image is judged by the way of intersection between images,and the gray is restored to the initial gray of the target image.The final single frame target detection result is determined by eliminating the non target interference and ensuring the target gray level is not distorted.(3)In terms of multi frame confirmation algorithm,this paper proposes an accurate pipeline filtering algorithm based on the classical pipeline filtering algorithm,which is an all-round improvement algorithm for the classical pipeline filtering algorithm.Innovation has been made in many aspects,such as input target variable,pipeline global variable,pipeline length,secondary search algorithm,etc.The total gray scale of the target is used to replace the center gray scale of the target,the decimal coordinate of the gray scale weighting is used to replace the integer coordinate of the highlight,and the asymmetric pipeline is used to replace the symmetric pipeline,which makes the pipeline tracking the target more accurate in the whole process.In addition,the detection algorithm flow with feedback is also proposed in this paper.The number of opened channels in the precise multi frame pipeline filtering algorithm is compared with the number of current frame targets,and the final target recognition threshold in the single frame detection algorithm is adjusted by feedback.This paper takes the infrared image taken by the ship infrared detection system or the real infrared image taken by the artificial simulation ocean as the experimental scene,and compares the algorithm proposed in this paper with other related algorithms.Experimental results show that the algorithm in this paper achieves better results in single frame detection accuracy and multi frame confirmation accuracy.