Unknown individual identification method and device for radar signals

Abstract

The invention discloses an unknown individual recognition method and device for radar signals. The method comprises the steps: constructing and storing a known category sample set N and an unknown category sample set UN; inputting each to-be-identified sample in the known category sample set N and each to-be-identified sample in the unknown category sample set UN into a coding network, extractingfeature vectors of the to-be-identified samples, and classifying the to-be-identified samples; generating an attention probability distribution vector by using a DDPG algorithm according to the feature vector of the to-be-identified sample; generating a conditional feature vector according to the feature vector and the attention probability distribution vector of the to-be-identified sample, and inputting the conditional feature vector into a decoding network to judge the category of the to-be-identified sample; performing network training in a coding network, a decoding network and a DDPG algorithm. The method has the advantages that unknown signals are prevented from being mistakenly recognized as certain known signals, and the known signals and the unknown signals are accurately separated.

Description

technical field [0001] The invention relates to the field of electronic reconnaissance, and more specifically relates to a method and device for identifying unknown individuals of radar signals. Background technique [0002] Based on the classification network of deep learning, the output type of the model is usually fixed, so when the model is trained, the category of the test data is also known. In real application scenarios, unknown categories that do not exist during training usually appear, and traditional classification networks cannot correctly classify unknown categories that do not exist during training. This defect will cause the recognition accuracy of the classification model to be greatly reduced in the real application environment, so solving the recognition problem of unknown categories is a key factor to improve the recognition accuracy of the classification network. There are two main solutions to the problem of unknown class recognition for deep learning. ...

AI Technical Summary

Problems solved by technology

The first method is to distinguish known categories from unknown categories by adding unknown sample classes to the training set. It is highly dependent on unknown samples, and the sample set is difficult to collect. It is difficult to include all potential unknown types in unknown classes. When the model encounters When it comes to the type that does not exist in the training set, it still cannot be correctly classified

The second method is to extract the middle layer features of the classification network, and then combine the commonly used machine learning methods for cluster analysis to distinguish the unknown classes. It can only have a better effect when the similarity between the unknown class and the known class is not large. , and the individual recognition of radar signals is distinguished by the subtle changes in the signals generated by the differences in radar hardware. The similarity of different radar individual signals is very large. individuals are completely separated

So to sum up, the existing solutions to the unknown class recognition problem of deep learning have the problem of not being able to classify correctly. In-depth research on cutting-edge technologies in the field of deep learning, and explore better unknown class recognition methods to improve the accuracy of radar signal individual recognition of great significance

Images