Hyperspectral Unmixing of Agricultural Images Taken from UAV Using Adapted U-Net Architecture

The hyperspectral unmixing method is an algorithm that extracts material (usually referred to as endmembers) data from hyperspectral data, along with their corresponding abundances.Due to the lower spatial resolution of hyperspectral sensor data compared to conventional cameras, each pixel is more likely to contain mixed information from multiple endmembers.In turn, mixed hyperspectral data is less valuable for use in research or predictive models.One of the problems in hyperspectral unmixing is the lack of openly available, field-collected datasets, particularly those from agricultural and other UAV-gathered sources.In turn, hyperspectral unmixing algorithms are rarely tested on this data type.This paper proposes a hyperspectral unmixing algorithm based on the U-Net network architecture to achieve more accurate unmixing results on existing and newly created hyperspectral unmixing datasets.The proposed model is fully unsupervised and is not limited by data shape and size.We also developed and shared a hyperspectral unmixing dataset derived from blueberry field data collected using a hyperspectral camera mounted on a Unmanned Aerial Vehicle (UAV).Compared to the state-of-the-art transformer-based unmixing model, our proposed algorithm achieved approximately 20% lower endmember RMSE and more than 50% lower reconstruction error values.