【作者】 林涛；

【导师】 刘梅；

【作者基本信息】 哈尔滨工业大学 ， 电子信息（专业学位）， 2022， 硕士

【摘要】 随着现代科技的进步,防空系统对空中目标进行识别时,使用分布在不同位置,拥有不同观测角度的多种不同参数传感器成为可能。同时防空系统识别并打击有威胁空中目标后,也需要对打击命中率进行评估。本文从空中目标的一维距离像仿真出发,采用了深度学习方法,在多个参数不完全相同的雷达,即多传感器;每个传感器位于不同的位置拥有不同的观测角度,即多角度观测下基础上,进行了防空系统目标识别、打击后命中率评估两个阶段的研究。论文研究了四种空中目标:MG-15战斗机、F-22战斗机、J-20战斗机和S92直升机的电磁特性,提出并实践了空中目标一维距离像的仿真方案。首先是建立空中目标的3D模型,将其导入CADFEKO软件,仿真步进频率下空中目标全方位角的雷达散射特性,进而建立空中目标的静态一维距离像数据库。通过预设空中目标飞行轨迹从而推算雷达观测角,进而建立空中目标飞行过程中多传感器观测的一维距离像数据集。论文研究了空中目标一维距离像的时频分析方法,在基于空中目标的时频特征基础上,对比了多种Res Net模型在单传感器和多传感器融合识别中的性能,使用Res Net18网络进行了基于注意力机制的融合识别研究,有效提升了基于卷积神经网络的多传感器多角度融合识别性能。深入研究了胶囊网络的特点,在CIFAR-10数据集上研究了胶囊网络的改进方案,提出了Res Caps网络。该网络结合了Res Net和Caps Net的优点,在多传感器多角度融合识别任务中获得了明显优于典型卷积神经网络——Res Net18的优异性能。最后,利用前面的研究方法作为打击命中率评估系统的基础,进行了防空系统打击后命中率评估阶段的研究。分析了对空中目标进行打击命中率评估的理论基础,建立了空中目标的易损性模型,并进行了建模仿真,并将Res Caps应用于打击命中率评估中。经过验证,该打击命中率评估系统不仅能给出空中目标在打击后的命中概率,还可以准确的评估打击命中的毁伤等级。更多还原

【Abstract】 With the advancement of modern technology,when the air defense system identifies air targets,it is possible to use a variety of sensors with different parameters distributed in different locations and with different observation angles.At the same time,after the air defense system identifies and strikes a threatening air target,it also needs to evaluate the strike rate.Starting from the HRRP of aerial targets,this paper adopts the deep learning method.In multiple radars with different parameters,that is,multi-sensors;each sensor is located at different positions and has different observation angles,that is,under multiangle observation.On this basis,two stages of research on air defense system target recognition and post-strike hit rate evaluation are carried out.The thesis studies the electromagnetic characteristics of four kinds of air targets:MG-15 fighter jet,F-22 fighter jet,J-20 fighter jet and S92 helicopter,and proposes and practices the simulation scheme of HRRP of air targets.The first is to establish a 3D model of the air target,import it into CADFEKO software,simulate the radar scattering characteristics of the air target in all directions under the step frequency,and then establish a static HRRP database of the air target.By presetting the flight trajectory of the air target,the radar observation angle is calculated,and then the one-dimensional range image data set of multi-sensor observation during the flight of the air target is established.The paper studies the joint time-frequency analysis method of HRRP of aerial targets.Based on the time-frequency characteristics of aerial targets,the performance of various Res Net models in single-sensor and multi-sensor fusion recognition is compared.The fusion recognition research of attention mechanism has effectively improved the performance of multi-sensor and multi-angle fusion recognition based on CNN.The characteristics of the capsule network are deeply studied,and the improvement scheme of the capsule network is studied on the CIFAR-10 data set,and the Res Caps network is proposed.The network combines the advantages of Res Net and Caps Net,and achieves significantly better performance than the typical convolutional neural networkRes Net18 in the multi-sensor and multi-angle fusion recognition task.The paper studies the time-frequency analysis method of HRRP of aerial targets,and uses Res Net18 network to carry out identification and classification research based on the time-frequency characteristics of aerial targets.The attention mechanism is introduced,which effectively improves the recognition performance of multi-sensor and multi-angle fusion based on convolutional neural network.The characteristics of the capsule network are deeply studied,and the improvement scheme of the capsule network is studied on the CIFAR-10 data set,and the Res Caps network is proposed.The network combines the advantages of Res Net and Caps Net,and achieves significantly better performance than the typical convolutional neural networkRes Net18 in the multi-sensor and multi-angle fusion recognition task.Finally,using the previous research method as the basis of the strike hit rate evaluation system,the research on the hit rate evaluation stage of the air defense system after the strike is carried out.The theoretical basis of the air target’s hit rate evaluation is analyzed,the vulnerability model of the air target is established,and the modeling and simulation are carried out,and Res Caps is applied to the hit rate evaluation.After verification,the hit rate evaluation system can not only give the hit probability of the air target after the hit,but also accurately evaluate the damage level of the hit.更多还原