星载SAR系统灵敏度对舰船分类影响的研究

Study on the influence of spaceborne SAR system sensitivity on ship classification

针对不同星载SAR系统灵敏度下的SAR图像舰船分类准确率不同的问题,文中提出了一种系统灵敏度对舰船分类影响的研究方法。采用所提出的基于背景像素填充的目标旋转数据扩充方法,并结合微调卷积神经网络模型对舰船进行分类。同时降低图像的信噪比以等效获得不同系统灵敏度的数据集,再分析其灵敏度对舰船分类的影响。仿真结果表明,随着系统灵敏度的降低,舰船分类准确率的下降趋势逐渐变缓,且当最差系统灵敏度降为-13.58 dB时,准确率可达到75%。因此,所提方法可应用于舰船分类对星载SAR系统灵敏度的需求分析,而仿真结果也为低系统灵敏度的星载SAR舰船分类提供了参考。

In view of the problem that the accuracy of ship classification in Syntactic Aperture Radar(SAR) images under different spaceborne SAR system sensitivities is different,a method is proposed to study the impact of system sensitivity for the ship classification. A data augmentation method of target rotation based on background pixel filling is proposed and combined with fine-tuning convolutional neural network model to classify ships. The datasets with different system sensitivities are equivalently obtained by reducing the signal-to-noise ratio of the images to analyze the influence on the system sensitivity on the ship classification. The simulation results show that the downward trend of the accuracy for ship classification gradually get slowly with the decrease of system sensitivity and the accuracy can reach 75%when the worst system sensitivity drops to-13.58 dB. The method proposed in this paper can be applied to study the requirements of the spaceborne SAR systems sensitivity for ship classification and the simulation results provide reference for ship classification of low system sensitivity spaceborne SAR.