Multi-functional Radar Working Mode Recognition Based On Entropy Map And Multi-modal Prototype Enhancement

Multi-function Radar(MFR)is one of the main representatives of modern new-generation radars,which is characterized by significant intelligence and flexibility.Traditional radar mode recognition methods cannot adapt to the MFR’s varying signal waveforms and rapidly changing operating modes,which limits the analysis of MFR’s working status and reasoning of behavioral intention.Therefore,it is imperative to study new MFR mode recognition methods.This article analyzes the mechanism of MFR working modes and the temporal characteristics of pulse parameters in different working modes.Firstly,a radar state transition point detection method based on wavelet transform is proposed to detect and segment pulse sequences of multi-working mode combinations into single-working mode pulse sequence fragments.Based on this,a multifunction radar mode recognition method based on an entropy graph is proposed by combining time series analysis and data mining methods to effectively identify the MFR working modes.Furthermore,a small sample mode recognition method based on multimodal prototype enhancement under mutual information maximization(MIM)is proposed to address the problem of insufficient radar signal samples.The specific research content of this article is as follows:1.To address the limitations of existing mode recognition methods in adapting to rapidly changing MFR waveform parameters and overly idealized experimental sample data,a radar state transition point detection method based on wavelet transform is proposed to detect the state transition points of MFR pulse signal sequences and segment them into single working mode sample data.Then,from the perspective of time series entropy,a multi-function radar mode recognition method based on an entropy graph is proposed.By extracting the approximate entropy,permutation entropy,sample entropy,and fuzzy entropy of the inter-pulse parameter sequence to form the entropy graph,the entropy graph features can accurately reflect the change rules and characteristics of pulse sequence parameters in radar working modes.Finally,the MFR working mode recognition is implemented by combining the convolutional neural network model.The experimental results show that the detection accuracy of the switch points between different working modes is close to 85% when the false pulse rate or missed pulse rate is 25%;the average recognition accuracy of working modes is close to 85%when the false pulse rate is 20%,83.1% when the missed pulse rate is 20%,and close to 83.1% when the parameter error is 8%,verifying the effectiveness of the proposed method.2.Considering that it is very difficult to intercept a large number of radar signal sample data in the actual battlefield environment,and the conventional recognition methods are difficult to reflect the change rules and characteristics of MFR working mode signal waveforms under small sample conditions,a small sample mode recognition method based on multimodal prototype enhancement under MIM is proposed.Firstly,by encoding semantic knowledge of the MFR working mode pulse sequence and extracting the entropy graph,"semantic" and "visual" modal samples are obtained.Then,high-dimensional embedding features of semantic knowledge vectors and the entropy graph of two modalities are extracted,and MIM is introduced to capture the intersection of key information between two modal features and filter out irrelevant noise information.Finally,the "visual" prototype is calculated,and the semantic information is used to guide attention to obtain enhanced class prototypes.Simulation experiments show that the proposed method improves the recognition performance by an average of 5%,2% under different false pulse scenes and 6%,2.5% under different missed pulse scenes compared to the two comparative methods,verifying the effectiveness and superiority of this recognition method in the field of small sample mode recognition.