Abstract:
In the field of precision agriculture, integration of multi-modal data sources such as soil Nitrogen-
Phosphorus-Potassium (NPK) sensors with high-resolution RGB and infrared imagery is essential
for improving crop monitoring. Aligning these multi-modal data, which vary widely in
resolution and format, to a uniform spatial resolution is a challenge. This research proposes a
novel approach to integrate data collected from multiple sources with varying spatial resolutions
and unify them to the same spatial resolution. Point-wise data is collected using NPK sensors to
map the basic macronutrients (Nitrogen, Phosphorus, and Potassium) using state-of-the-art technology
known as FarmBot. After data collection, heat maps are generated to represent nutrient
distribution as the initial layer.
Image up-sampling is performed using a deep learning algorithm called Enhanced Super-Resolution
Generative Adversarial Network (ESRGAN) on these low-resolution heat maps to enhance their
resolution without losing critical features. Additionally high-resolution drone images are collected,
encompassing Red-edge and Near-Infrared (NIR) bands in addition to RGB bands, for
a broader health analysis of the crop. Multi-source images are integrated into a single multichannel
image that includes both soil nutrient data and multispectral information. Hyperspectral
imaging is used to measure soil water content through reflectance, providing additional depth to
the reflectance-based soil analysis. By deploying these methods, farmers and agronomists are
equipped to make more informed decisions, leading to better agricultural outcomes.
