基于卷积神经网络的景象匹配区适配性筛选方法  

A CNN-based Scene Matching Adaptability Screening Method

作　　者：李邦杰 武健 张大巧 朱宏伟 胡超帝 白双豪 LI Bangjie;WU Jian;ZHANG Daqiao;ZHU Hongwei;HU Chaodi;BAI Shuanghao(Rocket Force University of Engineering,Xi’an 710025,Shaanxi;Troops No.96728,Golmud 816099,Qinghai)

机构地区：[1]火箭军工程大学,陕西西安710025 [2]96728部队,青海格尔木816099

出　　处：《火箭军工程大学学报》2024年第3期45-50,59,共7页Journal of Rocket Force University of Engineering

摘　　要：针对基于图像特征指标的景象匹配区适配性筛选方法效率低、景象匹配区规划周期长的问题,提出一种基于卷积神经网络(Convolutional Neural Network,CNN)的适配性筛选算法。首先以典型的城市、荒漠、农田等6类景象为对象,构建包括灰度均方差、高程方差、独立像元数和最大峰值比等的图像特征指标,形成相关数据集;设计基于CNN的景象匹配区适配性筛选模型,并分析了迭代次数、训练样本数量、像元数等对模型计算精度的影响;最后选取合适的迭代次数和全连接层神经元数,利用数据集验证了本文算法。结果表明:所提方法可以有效提升大规模典型景象的匹配适配性筛选效率。In order to solve problems of low efficiency and long planning period of image characteristic index-based scene matching area adaptability screening method,an adaptability screening algorithm based on convolutional neural network(CNN)was proposed.By taking six types of typical scenes,such as cities,deserts,and farmlands as objects,image characteristic indexes including gray mean square deviation,elevation variance,number of independent pixels,and maximum peak ratio were constructed to form relevant datasets.Then,a scene matching area adaptability screening model of classified CNN was designed to analyze the influence of iteration times,training sample number and pixel number on the accuracy of model computation.Accordingly,by selecting proper iteration times and numbers of fully connected layer neurons,the proposed algorithm was verified on datasets.Results showed that the proposed method can effectively improve the efficiency of matching and adaptability screening of typical large-scale scenes.

关 键 词：景象匹配 卷积神经网络 图像特征 适配性筛选 

分 类 号：TJ410[兵器科学与技术—火炮、自动武器与弹药工程]