DANS: Deep Attention Network for Single Image Super-Resolution

Talreja, Jagrati; Aramvith, Supavadee; Onoye, Takao

doi:10.1109/access.2023.3302692

Cited by 1 publication

(1 citation statement)

References 66 publications

Supporting

Mentioning

Contrasting

Order By: Relevance

“…These methods rely on extensive training data to establish a direct relationship between low-resolution and high-resolution images. However, their effectiveness can be limited and it may lead to the loss of high-frequency information [24][25][26], particularly when dealing with the non-linear mapping relationship between the low-resolution images and high-resolution images. This highlights the need for further advancements in the field of downscaling methods to address these complex non-linear correlations.…”

Section: Introductionmentioning

confidence: 99%

A New Spatial Downscaling Method for Long-Term AVHRR NDVI by Multiscale Residual Convolutional Neural Network

Sun,

Zhao,

Zhao

et al. 2024

IEEE J. Sel. Top. Appl. Earth Observations Remote Sensing

View full text Add to dashboard Cite

Monitoring vegetation dynamics is essential for ecological processes, environmental changes, and natural resource protection. Fine-scale representation of vegetation indices is necessary for regions with complex topography and high species diversity. However, the advanced very-high-resolution radiometer (AVHRR), which covers a large time range with high temporal resolution, does not provide normalized difference vegetation index (NDVI) data with sufficient spatial resolutions for a detailed analysis of vegetation changes. While the Moderate Resolution Imaging Spectroradiometer (MODIS), which has a higher temporal and spatial resolution, is only limited to the last few decades. In order to deal with these isues, we propose an approach called Multi-scale Residual Convolutional Neural Net-work (MRCNN) that utilizes a multi-scale structure together with and residual convolutional neural network to combine MODIS NDVI and AVHRR NDVI data. The MRCNN algorithm improved Mean Absolute Error (MAE), and Root Mean Squared Error (RMSE) by 0.026 and 0.032, respectively, resulting in a 64.38% improvement for MAE and 62.79% improvement for RMSE compared to AVHRR NDVI. It also increased the Peak-Signal-to-Noise Ratio (PSNR) by 28.5% and the Structural Similarity index (SSIM) by 16.2% . The MRCNN method accurately captures the actual state of MODIS NDVI and consistently tracks changing trends in the vegetation index. It is exact in complex terrain and diverse vegetation areas. This method enhances the spatial resolution of AVHRR NDVI and significantly improves the accuracy of monitoring nationwide vegetation index changes over 30 years. The findings establish a solid scientific foundation for implementing ecological conservation measures and promoting sustainable vegetation growth.

show abstract

Section: Introductionmentioning

confidence: 99%