Doctoral Student Researcher, Machine Learning and Sensing Lab, University of Florida, Gainesville, FL
Tree Species Classification using Hyperspectral Imagery Aug. 2021 - Present
Developed a semi-supervised classification approach based on pseudo-labeling to utilize a significant amount of unlabeled data (from NEON) containing hyperspectral images of tree crowns
created using Deepforest.
Leveraged statistical analysis techniques, such as pixel average and covariance calculations, to derive valuable insights into the differences in hyperspectral signatures across different tree species. The analysis included data from various North American forest sites starting from the year 2018.
Employed an ecology-informed hierarchical spectral-spatial attention network to enable hyperspectral image classification of tree species, leading to a marked improvement in classification
accuracy.
Signals in the Soil Analysis using Hyperspectral Imagery Aug. 2021 - Present
Compiled a comprehensive dataset of hyperspectral images of plant roots, spanning multiple species, within minirhizotron boxes, for rigorous plant physiology research.
Developed a two-stage regression model using ResNet architecture to predict water potential/content, a key plant physiology trait.
Implemented an efficient semantic segmentation pipeline for root and soil segmentation in various plant species. Incorporated active learning methods to augment the dataset and enhance model performance. Utilized a pre-trained Unet model trained on the PRMI dataset) and also evaluated conventional hyperspectral unmixing algorithms to segment roots from soil using spectral information.
