基于卷积神经网络迁移学习的飞机目标识别

飞机目标识别是地面情报系统的一项重要关键技术。近年来火热的深度学习方法,如卷积神经网络,展现出对于图像识别任务的优越性能。但是,训练卷积神经网络需要大量的带标签样本以估计规模庞大的模型参数,因而限制了其在雷达目标识别领域中的应用。针对飞机目标识别中的小样本问题,文中引入适用于有限数据场景的迁移学习技术,预先在其他大样本高分辨距离像数据上训练一个初始卷积神经网络模型,再结合当前飞机目标识别任务调优模型参数。在实测数据上的实验结果显示,与仅使用卷积神经网络的方法相比,所提方法可显著提升识别准确率,验证了方法的有效性。

Aircraft Target Recognition Based on Convolutional Neural Network with Transfer Learning

Aircraft target recognition is an important key technique of ground information systems. Recent hot deep learning methods, e. g. convolutional neural network, have shown superior performance for image recognition tasks. However, training a convolutional neural network requires a number of annotated samples to estimate numerous model parameters, which restricts its application to radar target recognition. Aiming at the small sample issue in aircraft target recognition, the transfer learning technique that suits the limited data is used in this study, combined with the advantages of convolutional neural network, an initial convolutional neural network has been trained in advance on big samples of radar high resolution range profiles, and then the model parameters are fine-tuned based on the current aircraft target recognition task. Experimental results of measured data show that compared to the results of merely applying convolutional neural networks, the proposed method can obviously increase the recognition accuracy rate, which validate the effectiveness of the proposed method.