基于集成改进卷积注意力块的SAR图像目标分类算法

SAR Image Target Classification Algorithm Based on Integrated Improved Convolutional Block Attention Module

在合成孔径雷达(Synthetic Aperture Radar,SAR)图像中,目标的轮廓和细节通常比较复杂。传统的卷积神经网络(Convolutional Neural Network,CNN)只使用单一均值参数进行无差别的特征提取,不能很好地区分SAR特征之间的差异。为了解决此问题,文章提出了1种基于集成改进卷积注意力块(Improved Convolutional Block Attention Module,ICBAM)的SAR图像目标分类算法ICBAM_CNN。首先,该模块通过引入方差参数至传统CBAM模块中,设计了1种改进的CBAM注意力机制,有助于分类识别网络更好地学习SAR图像不同目标卷积层输出与通道注意力之间的差异信息,提升不同SAR目标特征的可分离性;然后,ICBAM设计了1种中心坐标注意力机制,能更好地捕捉SAR图像中目标的中心分布特征,有效抑制杂波对SAR目标分类影像的干扰;最后,为了提高效率,将改进后的ICBAM模块集成到CNN网络中,实现SAR图像目标分类。ICBAM_CNN深度融合了SAR图像目标的多层级特征,提升了SAR目标特征的可分离性,可实现SAR图像目标的高精度、高效率识别分类。通过MSTAR数据集进行实验,结果表明,相比于传统CBAM方法,改进ICBAM方法的精确率提升了2.44%,召回率提升了2.24%,F1-score提升了2.34%。

In the Synthetic Aperture Radar(SAR)images,the contours and details of targets are often complex.Traditional Convolutional Neural Network(CNN)only uses a single mean parameter for indiscriminate feature extraction,so it cannot distinguish the differences between SAR features well.To address this issue,a SAR image target classification algorithm is proposed based on the Integrated Improved Convolutional Block Attention Module(ICBAM),as ICBAM_CNN.Firstly,this module designs an improved CBAM attention mechanism by introducing variance parameters into the traditional CBAM module,which helps the classification and recognition network better learn the differential information between the convolutional layer output and channel attention of different targets in SAR images,and improves the separability of different SAR target features.In addition,ICBAM has designed a center coordinate attention mechanism to better capture the center distribution features of targets in SAR images,effectively suppressing clutter interference on SAR target classification images.Finally,in order to improve efficiency,the improved ICBAM module is integrated into the CNN network to achieve SAR image target classification.ICBAM_CNN deeply integrates multi-level features of SAR image targets and improves the separability of SAR target features,enabling high-precision and efficient recognition and classification of SAR image targets.Experiments are conducted on the MSTAR dataset,and the results showed that compared to traditional CBAM methods,the improved ICBAM method improved precision by 2.44%,recall by 2.24%,and F1-score by 2.34%.