基于DCResNet的SAR图像车辆目标识别

合成孔径雷达（SAR）图像自动目标识别中，特征提取和目标分类是两个重要环节。残差网络（ResNet）作为一种较新的卷积神经网络，凭借其对目标特征的自适应学习能力，在SAR图像分类领域表现突出。本文在ResNet基础上，设计出了密集连接型残差网络（DCResNet），用于SAR图像目标识别。DCResNet在残差模块中增加了跳跃性连接的密度，不仅继承了ResNet的易学习的优点，还加强了特征的传播和利用率。除此之外，DCResNet采用平均池化的方式进行下采样，抑制了SAR图像中噪声对识别精度造成的影响。关于SAR图像目标识别的实验结果证明，本文提出的DCResNet与ResNet、AlexNet相比，不仅具有更快的收敛速度和推理速度，而且目标分类的准确率更高。

SAR Image Vehicle Target Recognition Based on DCResNet

In automatic target recognition for synthetic aperture radar (SAR) image, feature extraction and target classification are two important links. As a new convolutional neural network, ResNet has outstanding performance in SAR image classification because of its adaptive learning ability to target features. On the basis of ResNet, this paper designs a dense connection residual network (DCResNet) for SAR image target recognition. DCResNet increases the density of jumping connection in the residual module, so it not only inherits the advantages of ResNet in learning ability, but also enhances the propagation and utilization of features. In addition, DCResNet adopts the average pooling method for down sampling to suppress the impact of noise in SAR image on recognition accuracy. The experimental results of SAR image target recognition show that the DCResNet proposed in this paper has not only faster convergence speed and reasoning speed, but also higher accuracy of target classification compared with ResNet and AlexNet.