密集结构改进双通道神经网络的遥感图像配准

REMOTE SENSING IMAGE REGISTRATION OF DUAL-CHANNEL CONVOLUTIONAL NEURAL NETWORK WITH DENSE STRUCTURE IMPORVEMENT

针对部分传统算法对于遥感图像配准精度较低的问题,提出一种密集结构改进双通道卷积神经网络的遥感图像配准方法。对输入的图像采用密集结构改进的双通道卷积神经网络模型进行特征提取;用粒子群算法改进的随机一致性点漂移算法进行特征匹配得到仿射变换系数;使待配准图像能够根据该系数实现变换,达到配准目的。实验表明,改进算法比传统算法的配准精度平均提高了15%以上,对具有显著地貌差异的遥感图像对的配准精度可以有效地提高。

In order to solve the problem that some traditional algorithms have low accuracy in remote sensing image registration,a remote sensing image registration method of dual-channel convolutional neural network with dense structure improvement is proposed.The input image was extracted by using the dual-channel convolutional neural network model with improved dense structure,and the affine transformation coefficient was obtained by using the improved random consistency point drift algorithm of particle swarm optimization.The image to be registered could be transformed according to the coefficient to achieve the purpose of registration.The experimental results show that the registration accuracy of the improved algorithm is more than 15%higher than that of the traditional algorithm,and the registration accuracy of remote sensing images with significant geomorphologic differences can be effectively improved.