An Agricultural Precision Sprayer Deposit Identification System

Harry Rogers, Beatriz De La Iglesia, Tahmina Zebin, Grzegorz Cielniak, Ben Magri

Research output: Chapter in Book/Report/Conference proceedingConference contribution

5 Downloads (Pure)

Abstract

Data-driven Artificial Intelligence systems are playing an increasingly significant role in the advancement of precision agriculture. Currently, precision sprayers lack fully automated methods to evaluate effectiveness of their operation, e.g. whether spray has landed on target weeds. In this paper, using an agricultural spot spraying system images were collected from an RGB camera to locate spray deposits on weeds or lettuces. We present an interpretable deep learning pipeline to identify spray deposits on lettuces and weeds without using existing methods such as tracers or water sensitive papers. We implement a novel stratification and sampling methodology to improve results from a baseline. Using a binary classification head after transfer learning networks, spray deposits are identified with over 90% Area Under the Receiver Operating Characteristic (AUROC). This work offers a data-driven approach for an automated evaluation methodology for the effectiveness of precision sprayers.

Original languageEnglish
Title of host publication2023 IEEE 19th International Conference on Automation Science and Engineering (CASE)
ISBN (Electronic)9798350320695
DOIs
Publication statusPublished - 26 Aug 2023
Event2023 IEEE 19th International Conference on Automation Science and Engineering (CASE) - Auckland, New Zealand
Duration: 26 Aug 202330 Aug 2023

Publication series

NameIEEE International Conference on Automation Science and Engineering
Volume2023-August
ISSN (Print)2161-8070
ISSN (Electronic)2161-8089

Conference

Conference2023 IEEE 19th International Conference on Automation Science and Engineering (CASE)
Country/TerritoryNew Zealand
CityAuckland
Period26/08/2330/08/23

Cite this