多级可逆网络在SAR图像识别中的应用

Application of Multi-level Reversible Network in SAR Image Recognition

针对传统卷积神经网络目标识别方法在处理合成孔径雷达图像时存在的信息丢失等问题,我们引入了数据增强技术和多级可逆网络:采用数据增强方法,增加训练样本数量;通过多级可逆网络无损地提取合成孔径雷达图像特征;由交叉熵损失函数和二进制交叉熵损失函数联合训练网络,使用运动和静止目标获取与识别数据集,准确率达到99.75%,与现有的方法相比,准确率较高.

To solve the problem of information loss in the processing of SAR images by the traditional convolutional neural networks object recognition method,we introduced the data augmentation technology and Multi-level reversible network:the data augmentation method was used to increase the number of training samples;the features of SAR images were extracted losslessly by a Multi-level reversible network;the network was trained by the cross entropy loss function and the binary cross entropy loss function so that the data set of moving and stationary objects obtained and recognized reached an accuracy rare of 99.75%,which is higher than that achieved by using the existing methods.