The research is presented in two recent studies. In the first project, scientists focused on sweet potatoes and created a large dataset called Agro-HSR. This resource includes more than one thousand reconstructed image pairs from hundreds of samples.
The goal was to allow users to study chemical attributes such as sugar content, moisture and firmness without damaging the crop. The researchers tested several reconstruction models and found that two approaches performed the best, showing strong alignment with laboratory measurements. The dataset is freely available so others can develop or improve similar tools for different crops.
In the second study, the team explored how RGB images could be used to monitor maize growth by estimating chlorophyll content, which indicates plant health. They worked in greenhouse and field locations with different soil conditions and levels of stress. From this research, they developed a new model called the Window Adaptive Spatial Spectral Attention Transformer.
This method helps separate crops from soil and other objects, producing accurate ten band reconstructed images. The team also created a handheld device that allows users to take an RGB photo and receive multispectral information directly in the field. Future plans include adding a prediction feature so farmers can view results instantly.
These advancements offer a cost-effective way to observe crop quality without relying on expensive instruments. The work supports more precise crop management, helps detect stress early and provides practical tools for both industry and research.
The studies were completed through collaboration within the university and supported by national agricultural funding programs.
Photo Credit: University of Illinois Urbana-Champaign