一种基于多任务学习的微多普勒目标识别方法

近年来为了更好地适应不同场景下的智能化探测,人们对雷达目标识别任务的需求越来越高,因此本文提出了一种基于多任务学习方法的微多普勒目标识别方法。该方法基于BLSTM框架,针对两个不同的微多普勒目标识别任务使用特征提取器共享的方法训练模型。本方法既减小了识别系统整体的计算量,同时由于共享了不同任务之间的特征表示,又提高了模型的鲁棒性和准确率。

Multi-Task Learning for Target Classification Using Micro-Doppler Signatures

In recent years,there is a growing demand for radar target classification in order to achieve intelligent detection in different scenarios.In this paper,a target classification method using micro-Doppler signatures is proposed based on a multi-task learning approach.The proposed method utilizes the BLSTM framework and a shared feature extractor to train the models for two different target classification tasks.The proposed method reduces the overall computational load of the classification system and improves the robustness and accuracy of the models due to the shared feature representation between different tasks.