Research On Detection And Recognition Of Point Targets In Infrared Imaging Systems

The missile defense system plays an important role in maintaining national and regional security.The rapid detection and accurate recognition of missile targets is the basis and premise of missile interception,and it is also the great technical challenge faced by modern missile defense systems.This thesis focuses on two issues of missile target detection and warhead target recognition based on infrared imaging system.The contents of this thesis include:(1)Aiming at the problem of infrared point target detection under the complex background in space-based early warning system,a target detection framework based on satellite-ground combination is proposed.Under this framework,the onboard information processing system needs to complete two tasks of point target detection and complex background region extraction based on single-frame image.The ground information processing system needs to perform multi-frame association and moving point target detection.The experimental results show that the newly proposed detection framework can significantly reduce the target false alarm probability while maintaining the target detection probability under the complex background.(2)Aiming at the problem of infrared point target detection based on single-frame image in the onboard information processing system,an infrared background modeling method based on eight-direction linear prediction is proposed.The method includes two steps for the background prediction of each pixel.Firstly,the background gray value of each pixel is linearly predicted from eight directions to obtain eight predicted values,and then the Gaussian weighted sum of the eight values is obtained as the final prediction result.The experimental results show that the proposed algorithm can effectively suppress the background clutter and improve the target signal-to-noise ratio on the basis of ensuring the real-time performance.(3)Aiming at the problem of complex background region extraction based on singleframe image in the onboard information processing system,a complex region detection method based on background suppression and adaptive clustering is proposed.The method firstly detects the target candidate points in the single-frame image through background suppression,then adaptively clusters the candidate points by using the region growing method,and finally makes the complexity judgment of the image regions formed by the candidate points in a same cluster.The experimental results show that the method can detect complex background regions of any size quickly and effectively.(4)Aiming at the problem of infrared moving point target detection based on sequence images in the ground information processing system,an infrared moving point target detection algorithm based on spatial-temporal local contrast is proposed firstly,and the target detection performance of the algorithm is verified by experiments.Then,based on this,an enhanced spatial-temporal local contrast algorithm is proposed.The algorithm uses eight direction filtering to obtain the spatial local contrast.In the time domain,three frame images with equal intervals are used to make two differences,and then multiplying the two difference images to obtain the temporal local contrast.Finally,an enhanced threshold segmentation method is proposed to reduce the false alarm probability.The experimental results show that the proposed algorithm can effectively suppress the image background,significantly improve the target signal-to-noise ratio and improve the detection probability of the target under the complex background.(5)Aiming at the problem of point target recognition,the infrared radiation intensity sequence model of the spatial point target is established firstly.This simulation model takes a comprehensive consideration of the physical characteristics of the target,the motion characteristics of the target and the imaging effect of the detector.It can provides data support for the research of the point target recognition.Then a time series classification algorithm based on sparse representation is proposed.K singular value decomposition is applied to each class of training samples to obtain a corresponding dictionary.For a test sample,it is sparsely reconstructed by each dictionary respectively,and assign it to the class with the smallest reconstruction error.The experimental results show that the algorithm has good adaptability to the length of time series,has high tolerance to time series data loss,and has strong anti-noise ability.Finally,combined with the idea of deep learning and sparse representation,a time series classification algorithm based on sparse modulation convolutional neural network is proposed.The input signal of the algorithm includes both the original and the sparsely modulated time series,which makes the input information is more abundant.In addition,the convolutional neural network can mine and extract the local structural features of the time series,forming distinguished deep features layer by layer independently.The experimental results demonstrate the superior performance of the algorithm in time series classification.