• 主题
高级搜索  >
历史搜索 清空历史
首页> 外文会议>International Conference on Computer Vision, Image and Deep Learning >Super Resolution Enhancement of Satellite Remote Sensing Images of Transmission Tower Based on Multi-map Residual Network and Wavelet Transform
【24h】

Super Resolution Enhancement of Satellite Remote Sensing Images of Transmission Tower Based on Multi-map Residual Network and Wavelet Transform

机译：基于多地图残差网络和小波变换的传输塔卫星遥感图像超分辨率提高

获取原文
页面导航
  • 摘要
  • 著录项
  • 相似文献
  • 相关主题

摘要

Existing satellite remote sensing images are often used to observe the fuzzy phenomenon of transmission line bodies. It is necessary to enhance super-resolution, but traditional superresolution technology is difficult to obtain rich details and edge information of transmission towers. This paper proposes a multiscale edge enhancement method combining multi-map residual convolutional neural network and wavelet transform to solve these problems. Specifically, we first use a multi-map residual convolutional neural network to directly take the low-resolution image as the initial input of the network, and then use a convolutional layer to extract features. Secondly, a multi-mapping network is established by residual learning, and a batch normalization layer is added to optimize the network to enrich the feature information needed for high-resolution image aggregation. Finally, we use deconvolution layers to complete image upsampling and output high-resolution images, so the initial image can directly complete the end-to-end mapping relationship between low-resolution images and high-resolution images without performing preprocessing. On this basis, multiscale edge enhancement is performed on the transmission tower based on wavelet transform to obtain the final super-resolution enhancement result. Experiments on different benchmark data sets show that the proposed method is superior to the existing methods in the four quantitative indicators of peak signal-to-noise ratio, structural similarity, entropy and image detail enhancement.
机译：现有的卫星遥感图像通常用于观察传输线体的模糊现象。有必要增强超级分辨率，但传统的超级化技术难以获得富有传动塔的丰富细节和边缘信息。本文提出了一种多尺度边缘增强方法，组合多映射剩余卷积神经网络和小波变换来解决这些问题。具体地，我们首先使用多映射的残余卷积神经网络直接将低分辨率图像作为网络的初始输入，然后使用卷积层提取特征。其次，通过剩余学习建立多映射网络，并添加批量归一化层以优化网络以丰富高分辨率图像聚合所需的特征信息。最后，我们使用Deconvolution层来完成图像上采样和输出高分辨率图像，因此初始图像可以直接完成低分辨率图像和高分辨率图像之间的端到端映射关系而不执行预处理。在此基础上，基于小波变换对传输塔进行多尺度边缘增强，以获得最终的超分辨率增强结果。不同基准数据集的实验表明，所提出的方法优于峰值信噪比的四个定量指标中的现有方法，结构相似性，熵和图像细节增强。

著录项

相似文献

  • 外文文献
  • 中文文献
  • 专利
获取原文

客服邮箱：kefu@zhangqiaokeyan.com

京公网安备：11010802029741号 ICP备案号：京ICP备15016152号-6 六维联合信息科技 (北京) 有限公司©版权所有

  • 客服微信

  • 服务号