| 摘 要: 针对传统的导弹型号识别主要依赖人工经验和专家知识对导弹外观特征进行分析和比对的方法存在准确率和实时性难以兼顾的问题,提出了一种基于ResNet18卷积神经网络的导弹型号识别模型用于解决这一问题。首先,根据作战任务背景分析,提取导弹的主要特征指标,构建几种常见的导弹模型;其次,以各个角度的导弹照片作为输入,输出图片的均值和方差;再次,通过对特征值进行量化,生成训练样本数据;最后,利用基于ResNet18卷积神经网络模型对导弹型号进行识别训练。实验结果表明,该模型在保证实时性和满足实战要求的前提下,具有较高的准确率,对实验中选择的6种导弹型号的识别准确率达到了99.80%。 |
| 关键词: 导弹型号识别;特征分析;ResNet18 |
|
中图分类号: TP3-05
文献标识码: A
|
|
| Research on Missile Model Recognition Based on Convolutional Neural Network |
|
QIN Haokun1, WEI Na2, QU Yubo3, ZHANG Qingchao1, CHEN Bingyu4
|
(1.Department of Inf ormation Security, Naval University of Engineering, Wuhan 430033, China;
2.School of Electronic Engineering, Naval University of Engineering, Wuhan 430033, China;
3.College of Ordnance Engineering, Naval University of Engineering, Wuhan 430033, China;
4.School of Power Engineering, Naval University of Engineering, Wuhan 430033, China)
3196582500@qq.com; weina1223@126.com; 2901093261@qq.com; 790152218@qq.com; 1246650112@qq.com
|
| Abstract: Traditional missile model recognition mainly relies on the analysis and comparison of missile appearance features based on manual experience and expert knowledge, which faces the challenge of balancing accuracy and real-time performance. Aiming at the problems, this paper proposes a missile model recognition model based on ResNet18 convolutional neural network. Firstly, based on the analysis of operational mission background, the main feature indicators of missiles are extracted, and several common missile models are constructed. Secondly, missile photos from various angles are taken as input to calculate the mean and variance of the images. Next, by quantifying the feature values, training sample data is generated. Finally, the missile model recognition is trained using the ResNet18 convolutional neural network model. Experimental results show that the model achieves a high accuracy rate of 99.80% in recognizing the six selected missile models, while ensuring real-time performance and meeting the requirements of actual combat scenarios. |
| Keywords: missile model recognition; feature analysis; ResNet18 |
