现有的识别方法对新的数据需要重新进行训练,不利于模型速度和在线功能的实现。文中提出一种宽度的卷积神经网络(BCNN), 该模型由于具有“宽度”的网络结构,可以通过利用新生成的附加特征,提高BCNN模型的识别性能;此外,BCNN模型还能够利用新的训练数据进行自身的更新,从而具有增量学习能力。实验结果表明,该方法能更好地提取 数据的特征,而且比常规的CNN的识别精度提高8%以上,所提模型还可以利用新的数据进行在线更新,从而具有更强的实用性和鲁棒性。

Radar target recognition technology is a means to identify remote targets by using radar and computer, and has been widely used in modern radar field. However, the existing recognition methods need to be retrained for new data, which is not conducive to the realization of model speed and online function. Based on this, this paper proposes a broad convolution neural network (BCNN), since the model has the "width" network structure, which can use newly generated additional features, and improve the recognition performance of BCNN model, BCNN model is also able to use new training data to update themselves, thus has the incremental learning ability. Experimental results show that this method can better extract the features of the data, and the recognition accuracy is more than 8% higher than that of conventional CNN, the model can update itself to include the newly emerging radar data, so it has stronger practicability and robustness.

TN957.5

江苏省高等教育教改立项研究课题(2015JSJG360)