Research On Anti-jamming Recognition Algorithm Of Infrared Target

In a complex confrontation environment,the target and interference imaging process is gradually separated from the aliased point target state to showing appearance characteristics.The effective recognition of infrared targets is an important factor in achieving precision strikes.Recognizing the number and location of targets and interference as early as possible from the aliasing state can provide earlier information for the identification of infrared targets and improve military strike capabilities.This thesis studies the anti-artificial-jamming algorithm of infrared targets,the main work is as follows:Aiming at the problem that the target and the interference are aliased in the image plane at the initial stage of the countermeasure,the number and position information cannot be distinguished,this thesis proposes a multi-frame based coarse and precise estimation combined with a super-resolution algorithm of Closed-Spaced-Objects.In this algorithm,the coarse estimation module uses the frequency domain analysis method to estimate the interframe offset of the target block,aligns and superimposes the high-resolution target blocks obtained by interpolation,and estimates the number and position of the target based on the superimposed energy peaks.The fine estimation module constructs a super-resolution function based on the imaging model and uses the result of the rough estimation as a constraint to estimate the precise position using a quantum particle swarm optimization algorithm.Experimental results prove that the algorithm has excellent super-resolution performance and provides early information for anti-jamming tasks.Aiming at the problem that the current anti-jamming algorithms are insufficient for mining motion patterns,this thesis proposes an anti-jamming algorithm based on Detection Cascade-Motion Feature Analysis-Appearance Timing Analysis network(DC-1DCNNCLSTM).The algorithm extracts the trajectory pipeline through the DC network,and expresses the difference between the target and the interference movement pattern by using the temporal changes of dynamic characteristics and appearance characteristics,and thus designs two network branches.The one-dimensional convolutional neural network(1DCNN)branch network analyzes dynamic characteristics,and the CLSTM branch network analyzes the temporal changes of appearance characteristics,and merges the classification probabilities of the two branch networks as the final classification result.Experimental results prove that the algorithm can achieve 93.3% classification accuracy in anti-jamming tasks.Aiming at the problem that single-frame candidate target extraction and interframe cascading performance in the DC network seriously restrict the anti-jamming ability,this thesis proposes an end-to-end anti-jamming algorithm based on Spatio-TemporalProgressive learning.The algorithm starts from the coarse-scale proposals,adopts the idea of gradual learning,uses the spatial refinement module to complete the detection and cascade tasks,and uses the temporal expansion module to extract more temporal information and improve the accuracy of anti-jamming classification.In this thesis,the attention mechanism is used to highlight the temporal and spatial characteristics and movement characteristics of the target and interference,so as to further improve the performance of anti-jamming.Experimental results prove that the algorithm can achieve a classification accuracy of 96.3% and has excellent anti-jamming performance.