The Augmented Agronomist Pipeline and Time Series Forecasting

EasyChair Preprint 3087

Abstract

We propose a new pipeline to facilitate deep learning at scale for agriculture and food robotics, and exemplify it using strawberry tabletop. We use this multimodal, autonomously self-collected, distributed dataset for predicting strawberry tabletop yield, aiming at informing both agronomists and creating a robotic attention system. We call this system the augmented agronomist, which is designed for agronomy forecasting, and support, maximizing the human time and awareness to areas most critical. This project seeks to be relatively protective of both its neural networks, and its data, to prevent things such as adversarial attacks, or sensitive method leaks from damaging the future growers livelihoods. Toward this end this project shall take advantage of, and further our existing distributed-deep-learning framework Nemesyst. The augmented agronomist will take advantage of our existing strawberry tabletop in our Riseholme campus, and will use the generalized robotics platform Thorvald for the autonomous data collection.

Keyphrases: Agriculture, database, deep learning, nemesyst, strawberries, thorvald

@booklet{EasyChair:3087,
  author    = {George Onoufriou and Marc Hanheide and Georgios Leontidis},
  title     = {The Augmented Agronomist Pipeline and Time Series Forecasting},
  howpublished = {EasyChair Preprint 3087},
  year      = {EasyChair, 2020}}