基于R-VGG和多分支注意力的无人机图像配准模型

UAV Image Registration Model Based on R-VGG and Multi-Branch Attention

无人机图像通常分辨率较大且含有大面积的弱纹理区域,导致在配准时图像特征提取不足和误匹配增加.针对这些问题,提出一种基于R-VGG特征提取和多分支注意力的无监督配准模型.首先,利用两个具有共享权重参数的特征提取网络来提取运动图像和参考图像的低、高层融合特征;然后,在初步特征匹配之后,加入以残差块为单位的多分支注意力(Multi-Branch Attention,MBA)以滤除错误特征匹配;最后,采用卷积神经网络进行单应性估计,使用空间变换网络(Spatial Transform Network,STN)将运动图像根据单应性矩阵扭曲得到配准结果图像.通过实验将其与另外4种图像配准方法进行了比较,并根据结构相似性(SSIM)、互信息量(MI)和平均绝对误差(MAE)三种评价指标进行了衡量.结果表明,所提方法具有很好的性能表现,能够准确、稳定地完成无人机图像的配准任务.

UAV images usually have a large resolution and contain large areas of low texture,which leads to insufficient feature extraction and increased mismatches during registration.Aim at these problems,an unsupervised registration model based on R-VGG feature extraction and multi-branch attention was proposed.Firstly,two deep feature extraction networks with shared parameters were used to extract the low-level and high-level fusion features of the moving image and the fixed image;then,after the preliminary feature matching,it was proposed to add multi-branch attention in units of residual blocks to filter out mismatches;finally,following by using the convolutional neural network to estimate the homography,the spatial transform network(STN)was used to warp the moving image according to the homography matrix to obtain the registration result image of moved image.Compared with the other four registration methods and measured according to three evaluation indexes of SSIM,MI and MAE,the results demonstrate that the proposed method has good performance and can complete the UAV image registration task accurately and stably.