基于深度学习的宽基线弱纹理影像匹配研究  

Research on Wide Baseline Weak Texture Image Matching Based on Deep Learning

作　　者：傅水清 Fu Shuiqing(Xiamen Zhongtu Geographic Information Co.,Ltd.,Xiamen,China)

机构地区：[1]厦门众图地理信息有限公司,福建厦门

出　　处：《科学技术创新》2024年第14期47-50,共4页Scientific and Technological Innovation

摘　　要：为了解决宽基线弱纹理影像匹配中的挑战性问题,本文深入研究了基于深度学习的匹配方法。首先,设计了宽基线弱纹理影像匹配的完整流程,包括几何纠正、粗等级匹配预测和最终匹配等关键步骤。在几何纠正方面,本文采用了一种有效的影像配准方法,以纠正由于摄影角度和地形变化引起的几何畸变。接着,通过引入深度学习模型,实现了对宽基线弱纹理影像的粗等级匹配预测,为后续精细匹配提供了可靠的初值。在最终匹配阶段,本文提出了一种结合传统特征和深度学习特征的综合匹配算法。研究结果表明,应用本文方法进行宽基线弱纹理影像匹配时,匹配正确率均在95.00%以上,显示了该方法在处理具有挑战性的影像数据时的优越性能。In order to solve the challenging problem of wide baseline weak texture image matching,this paper delves into deep learning based matching methods.Firstly,a complete process for matching wide baseline weak texture images was designed,including key steps such as geometric correction,coarse level matching prediction,and final matching.In terms of geometric correction,this article adopts an effective image registration method to correct geometric distortions caused by photography angles and terrain changes.Subsequently,by introducing deep learning models,coarse level matching prediction for wide baseline weak texture images was achieved,providing reliable initial values for subsequent fine matching.In the final matching stage,this paper proposes a comprehensive matching algorithm that combines traditional features and deep learning features.The research results show that when using the method proposed in this paper for wide baseline weak texture image matching,the matching accuracy is above 95.00%,demonstrating the superior performance of this method in processing challenging image data.

关 键 词：弱纹理特征 深度学习 宽基线 影像 自动 拼接 

分 类 号：TP391[自动化与计算机技术—计算机应用技术]