一种改进的可解释SAR图像识别网络  

An Improved Interpretable SAR Image Recognition Network

作　　者：李鹏[1] 冯存前[1] 胡晓伟[1] LI Peng;FENG Cunqian;HU Xiaowei(Air Defense and Antimissile School,Air Force Engineering University,Xi’an,710051,China)

机构地区：[1]空军工程大学防空反导学院,西安710051

出　　处：《空军工程大学学报》2023年第4期49-55,共7页Journal of Air Force Engineering University

基　　金：陕西省自然科学基础研究计划(2023-JC-YB-491)。

摘　　要：SAR-BagNet模型是一种应用于合成孔径雷达(SAR)图像识别的可解释深度学习模型。为了使SAR-BagNet模型在具有可解释性的同时具有较高的识别精度,以SAR-BagNet模型为基础,在模型框架中加入了空间注意力和坐标注意力机制,并在MSTAR实测数据集上进行了实验。实验结果表明,空间注意力和坐标注意力机制增强了SAR-BagNet模型的全局信息获取能力,能够在不降低其可解释性的基础上,有效提高模型的识别精度和决策合理性。The SAR-BagNet model is an interpretable deep learning model used for Synthetic Aperture Radar(SAR)image recognition.In order to maintain the interpretability of the SAR-BagNet model while also achieving high recognition accuracy,this paper uses the SAR-BagNet model as a foundation and incorporates spatial attention and coordinate attention mechanisms into the model framework.Experimental results on the MSTAR dataset demonstrate that the spatial attention and coordinate attention mechanisms enhance the SAR-BagNet model's ability to acquire global information.This enhancement effectively improves the model's recognition accuracy and decision rationality without compromising its interpretability.

关 键 词：深度学习 SAR图像识别 模型可解释性 注意力机制 

分 类 号：TN957[电子电信—信号与信息处理]