High-resolution range profile target recognition online database building method based on step-by-step segmented training

Abstract

The invention discloses a high-resolution range profile target recognition online database building method based on step-by-step and segmented training. The method comprises the following steps of acquiring a radar high-resolution range profile training data set of a current task and preprocessing the radar high-resolution range profile training data set; updating the memory data set by using the frequency domain training data set; constructing an initial multilayer convolutional neural network; training the initial multi-layer convolutional neural network by using the memory data set, and sampling the importance of neurons to obtain the importance score of the current task; dividing the trained multilayer convolutional neural network according to the importance score of the current task, and training to obtain a trained multilayer convolutional neural network; and inputting test data into the trained multilayer convolutional neural network to obtain a target recognition result. According to the method, the multi-layer convolutional neural network is used, the high-dimensional features are extracted from the frequency domain information of the radar high-resolution range profile, and a problem that in the prior art, due to the fact that the feature information amount of target recognition is limited, target recognition accuracy is not high is solved.

Description

technical field [0001] The invention belongs to the technical field of radar, and in particular relates to an online database building method for high-resolution range image target recognition based on step-by-step training. Background technique [0002] Radar High-Resolution Range Profile (HRRP) contains rich radar target structure features, and has the advantages of easy acquisition, storage and processing. It is very valuable for radar target recognition and classification, and has become a radar automatic target recognition research hotspots. [0003] The premise of identifying radar targets is to establish a complete radar target HRRP database offline before training. In practical applications, due to the difficulty in obtaining non-cooperative target samples and other reasons, it is often impossible to establish a complete database in advance. Utilizing the HRRP data accumulated during the radar working process, online database building through online learning and re...

AI Technical Summary

Problems solved by technology

[0004] However, the direct tuning method has certain restrictions on parameter modification

Since specific tasks have their corresponding optimal performance parameter spaces in neural network learning, the EWC method relies on the intersection of different task parameter spaces. If the characteristics of the old and new tasks differ greatly, the intersection of parameter spaces will become smaller, and the network is learning new categories. While forgetting the old categories, the accuracy of target recognition will drop rapidly

Images