Browsing by Author "Dias, Ana Margarida Mendes"
Now showing 1 - 1 of 1
Results Per Page
Sort Options
- Super-resolution Satellite Imagery for Crop Health MonitoringPublication . Dias, Ana Margarida Mendes; Neves, João Carlos RaposoThe exponential growth of global population and the increasing concerns regarding sustainability have fostered the interest in precision agriculture. This dissertation focuses on developing a super-resolution method that integrates remote sensing and multispectral data to generate high-resolution images for effective monitoring of soil and plant health in crops. The use of drone technology has been exploited during the last years in crop monitoring allowing to determine different metrics from plant and soil health. However, their use requires human intervention increasing their usage cost, being unaffordable for small producers. Also, their limited autonomy restricts the monitoring to large-sized areas. The use of satellite imagery addresses these problems, but, in turn, the resolution of the images when compared to drone-captured data represents a challenge. This work aims to address the problem of low-resolution satellite imagery using innovative image processing techniques and deep learning, specifically in the field of super-resolution. The methodology involves leveraging satellite imagery paired with UAV-acquired data, to train a model capable of approximating the high resolution of UAV images. In particular, this work proposed a novel clustering-based strategy for improving the attention mechanism of vision transformers. The results obtained on a proprietary and publicly available remote sensing dataset suggest that the proposed strategy is capable of achieving competitive performance with state-of-the-art approaches. More importantly, the results obtained suggest that the use of proposed approach can be used to determine plant health using satellite imagery since the estimated NDVI from the super-resolved data differs by a maximum of 13% from the values derived from UAV-acquired data, where at least half of the values present a difference of less than or equal to 3.6%.