摘要
目的:本文通过实验探究一种在多种场景下识别伪装目标的识别模型,用于解决在复杂场景下难以准确发现受伤人员的问题,以提高复杂场景识别伪装目标的准确率。方法:本实验从反射率、光谱指数、纹理和空间频率四个方面提取了适用于地面伪装目标识别的多域特征,设计了荒漠、林地和城市典型场景实验,构建SVM分类模型筛选出特定场景下的最优特征子集,筛选出可应用于多种场景的最优特征集,增强了识别方法的鲁棒性。结果:实验结果表明基于SVM分类模型的分类精准度在荒漠场景下可达99.65%,在林地场景下可达98.93%,在城市场景下可达99.74%。结论:该分类模型在三个低对比度的典型场景下分别准确地识别出了这三种迷彩服的伪装目标,为复杂多变环境中对伪装状态下的目标识别奠定实验基础。
Objective:This paper explores a model for identifying the simulated target in multiple scenes through experiments to solve the problem that it is difficult to accurately find the injured personnel in the complex scenes,so as to improve the accuracy of identifying the simulated target in the complex scenes.Methods:This experiment extracts multi-domain characteristics applied to identify the ground simulated target from four aspects:reflectivity,spectral index,texture and spatial frequency,and designs the desert,forest land and city typical scenario experiments,constructs the SVM classification model to screen the optimal feature subset in specific scenarios and the optimal selection applied in the complex scenes,which enhances the robustness of the identification method.Results:The experiment results show that,the classification accuracy based on the SVM classification model can reach 99.65%in the desert scene,98.93%in the forest land scene,and 99.74%in the city scene.Conclusion:The classification model accurately identifies the simulated target of three kinds of camouflage uniforms in three typical low contrast scenes,which lay the experimental foundation for the target identification under the camouflage condition in the complex and changeable environment.
作者
夏娟娟
祁富贵
李钊
雷涛
路国华
Xia Juan-juan;Qi Fu-gui;Li Zhao;Lei Tao;Lu Guo-hua(School of Biomedical Engineering,Air Force Medical University,Xi’an 710032,Shaanxi Province,China)
出处
《科学与信息化》
2023年第14期34-37,共4页
Technology and Information
基金
2021年度陕西省科学技术厅,陕西省重点产业创新链,野外人员搜救无人机系统研究,项目名称:项目编号:2021ZDLGY09-07。
关键词
无人机
多光谱检测
跨场景
SVM分类模型
人体目标检测
UAV
multi-spectral detection
cross-scene
SVM classification model
human target detection
