Overcoming deep learning data dependency in proximal and remote sensing

On the 15th of March 2023, Maria Culman Forero will publicly defend her PhD entitled “Overcoming deep learning data dependency in proximal and remote sensing: applications in agriculture”. She will take you on a fascinating journey along innovative deep learning and remote sensing science, looking for exotic palm trees and Belgian conference pear trees. This, to tell you her story for the hunt for solutions to overcome some major barriers that deep learning researchers and practitioners face when working with proximal and remote sensing data in an agricultural setting. Strategies to reduce the dependency on large-scale and high- quality datasets will be demonstrated.

She will formulate answers to the following questions:

  • How to exploit noisy labels in deep learning?
  • How to compensate for spatial coarse imagery in deep learning?
  • How to reduce the manual annotation effort in deep learning?

The findings will be demonstrated through her research on palm tree and individual pear detection. For both applications, only noisy and/ or limited labelled datasets were available. The proposed strategies, namely, transfer learning from noisy labels, multimodal data integration, and automatic image annotation, enabled to either mitigate or circumvent the data dependency of deep learning in tree and fruit detection applications.

Fruit detections and segmentations from image sample

Related blog posts by Maria Culman Forero:

Maria, who is born in Bucaramanga, Colombia, decided to follow her passion for science and agriculture all the way to Belgium. She started at VITO Remote Sensing in 2019 and thereby choose to dive into the magic world of remote sensing and deep learning to solve societal relevant agricultural issues. Initially she started as an early-stage researcher at a Marie Sklodowska-Curie Innovative Training Network, TRuStEE. In 2020, she made the decision to start a Ph.D. under the joint supervision of Dr. Stephanie Delalieux (VITO Remote Sensing) and Prof. Ben Somers (KU Leuven, Bioscience Engineering). During her PhD she was financed by the Belgian Science Policy Office (Belspo) in the framework of the PalmWatch project.