Research On Detection Algorithm Of Infrared Dim Target In Complex Background

Compared with the visible light system,the infrared system has the advantages of great concealment,long detection distance and strong ability to penetrate the atmosphere.As a key technology in the infrared imaging system,the infrared dim target detection has important research and application value in the fields of military early warning,infrared guidance,civil security,maritime search and rescue.With the continuous improvement of infrared dim target detection technology,the detection accuracy,efficiency and robustness of infrared dim target detection methods in simple scene have been greatly developed.However,due to the influence of long imaging distance and atmospheric turbulence,infrared dim targets usually occupy only a few pixels,without obvious texture,shape,color and structural features.On the other hand,infrared dim target is surrounded by complex interference,such as various background clutter and heavy noise.Infrared dim target detection is still a difficult and challenging task.Based on the human visual system(HVS),this thesis mainly studies the infrared dim target detection technology in complex scenes,which is devoted to improve the detection accuracy,efficiency and scene adaptability.In this paper,based on the research and analysis of the development and research status of infrared dim target detection technology,the following works is carried out.1.To solve the problem of the lack of experimental data in the field of infrared dim target detection,a Multi-Scene Infrared Dim Target(MSIDT)dataset is constructed.Firstly,the data collected is pruned to eliminate the unreasonable data,and then divided into six general scenes according to different backgrounds.Secondly,all targets of the six scenes are manually labeled.Finally,the statistical and attribute characteristics of MSIDT are introduced and concluded.2.Aiming at the problem that the single feature detection methods can not suppress the complex background effectively,according to the comparison mechanism of HVS,an infrared dim target detection algorithm based on Multi Information Fusion(MIF)is proposed.Firstly,the Local Gray Rresidual Measurement(LGRM)is proposed to suppress the gently background,highlight-isolated background and homogeneous continuous edge.Secondly,a Novel Variance Difference Measure(NVDM)method is proposed to suppress the isolated continuous edge.Thirdly,the method based on the Gradient Direction Characteristics(GDC)is designed to further enhance the robustness of the algorithm.Finally,the final saliency map is obtained from the above three feature maps by multiplicative fusion.The experimental results show that the proposed method has great suppression performance for various complex background and better detection performance in challenging scene.3.In order to improve the detection accuracy,efficiency and scene adaptability of detection methods based on HVS,a new infrared dim target detection algorithm based on visual attention mechanism is proposed.From the bottom-up perspective,the multi-scale gray and variance estimation is proposed to calculate the saliency map and estimate the optimal target size fastly.Then,the ROI(Region of Interest)is extracted using FAST(Features from Accelerated Segment Test)corner detection for transforming the traditional global computing into local processing to improve the detection efficiency.From the top-down perspective,based on the theory of biological lateral inhibition and cosine similarity,a soft fuzzy adaptive resonance theory network(Soft-FART)is proposed,and a new dim target feature set is constructed to train the network.Finally,the ROI is recognized by the well-trained model.The experimental results show that the proposed method can achieve the requirements of accurate,real-time and stability.