Abstract
This project is part of the Innovation Fund Denmark project RoboWeedMaPS. The overall goal of RoboWeedMaPS is to substantially reduce the amount of herbicides in modern crop farming, which will benefit society, the environment and the farmer. To achieve this, a more efficient and precise deployment of herbicides is needed. The project will incorporate automated vision systems to assess the optimum weed treatment and thereby eliminate the need for intermediate manual decision-making and data processing.
This project seeks to apply machine learning, specifically deep learning, to automatically classify weed species and their current development stage from images. To improve the certainty of the classification, it should incorporate context relevant data such as site-specific cropping history, past weed registrations, etc. The project will also explore the potential of generating photo realistic image samples of weeds using deep generative models. These artificial samples are expected to be used for creating a more robust weed classification model. Generative models can potentially also be used to improve the quality of unfocused and blurred images.
Additional project information
The VITRACE study
Place of employment
Intensive Care Dept.
PhD author
Date and place of defense
12th July 2019, Institute of Clinical Medicine, University of Copenhagen
Supervisors
Anders Perner (main)
Kirsten Møller
Morten Hylander Møller
Links
Link to PubMed