摘要
针对部分遥感图像配准算法存在特征提取能力不足,存在较多误匹配点的问题,提出了一个算法。在特征提取阶段加入改进的空间注意力机制关注图像的重要区域以及显著特征,增强网络的特征表达能力。匹配阶段提出粗过滤加细过滤双层过滤的方式,提高模型的鲁棒性和配准精度。同时改进了损失函数,提高模型的泛化性和拟合能力。仿真中,与多种先进的方法相比,所提算法能获得更高的配准精度和更低的配准误差,并且针对不同类型的图像也能取得较好的配准效果。
Aiming at the problem that some remote sensing image registration algorithms have insufficient feature extraction ability and many false matching points,an algorithm is proposed.In the feature extraction stage,an improved spatial attention mechanism is added to focus on the important regions and prominent features of the image to enhance the feature expression ability of the network.In the matching stage,the method of coarse filtration plus fine filtration and double filtration is proposed to improve the robustness and registration accuracy of the model.At the same time,the loss function is improved to improve the generalization and fitting ability of the algorithm.In the simulation,compared with the four advanced methods,the proposed algorithm can achieve higher registration accuracy and lower registration error,and can also achieve better registration effects for different types of images.
作者
李先静
陈颖
石艳娇
LI Xian-jing;CHEN Ying;SHI Yan-jiao(Shanghai Institute of Technology,Shanghai 201418,China)
出处
《计算机仿真》
2024年第2期202-206,231,共6页
Computer Simulation
基金
国家自然科学基金资助项目(61976140)
国家自然科学基金青年基金项目(61806126)。
关键词
图像配准
空间注意力
双层过滤
损失函数
Image registration
Spatial attention
Double filter
Loss function
