Publicación:

Desarrollo de un modelo de red neuronal profunda para imágenes obtenidas mediante teledetección utilizando la técnica de superresolución multiimagen.

Resumen

This presents an outstanding chance for the field of remote detecting to acquire extra information and data from caught information. Nonetheless, most of as of late distributed endeavors have focused on the Single-image Super-resolution issue. In this paper the examination introduced here proposes a Residual Attention Multi-image Super-resolution network (RAMS) that effectively finishes the Multi-image Super-resolution challenge by at the same time interfacing different pictures in light of spatial and temporal connections. To beat the restrictions of the neighborhood district of convolutional activities and accomplish cognizant information combination and data extraction from various low-resolution pictures, we present the instrument of visual component consideration utilizing 3D convolutions. In addition to using established residual associations to stream superfluous low-recurrence signals and focus the calculation on more significant high-recurrence parts, our depiction learning network uses various contributions with the same scene. The suggested substantial learning-based technique is leading-edge for distant identification applications, as shown by extensive testing and comparisons with existing open strategies for single or multi-image super resolution. Although CNN models outperform other multi-image super-resolution models quantitatively, our test results demonstrate that while RAMS models appear to perform well, they fall short of other models when it comes to measuring picture quality. © 2023 IEEE.