- You are here:
-
-
All publications
- Deep Learning for Real-Time Hair Fescue Identification in Wild Blueberry Fields
All publications from your selection
Deep Learning for Real-Time Hair Fescue Identification in Wild Blueberry Fields
Authors: Patrick Hennessy, Travis Esau, Aitazaz Farooque, Arnold Schumann, Qamar Zaman, Kenneth Corscadden
Identifier: CSBE21522
Download file:
Published in: CSBE-SCGAB Technical Conferences » 5th CIGR and AGM Quebec City 2021 » World Congress on Computers in Agriculture and Natural Resources
Download RAW file:
DOI:
Description: Herbicides are sprayed on wild blueberry (Vaccinium angustifolium Ait.) fields in eastern Canada and Maine to manage weeds such as hair fescue (Festuca filiformis Pourr.). Commercial sprayers provide a uniform application of herbicide, while the weeds often grow in patches. Smart sprayers provide an opportunity to reduce herbicide use by only spraying areas of the field with weed cover. Research has been done using machine vision systems which rely on green colour segmentation, colour co-occurrence matrices, and deep learning convolutional neural networks (CNNs) for automatic detection of areas with weed cover. Green colour segmentation is not optimal because multiple weeds found in wild blueberry fields contain green, and each weed may require a different herbicide treatment for effective management. Colour co-occurrence matrices are limited by long processing times and are impractical for real-time detection. Two object-detection CNNs built on the Darknet framework, YOLOv3 and YOLOv3-Tiny, were trained to detect hair fescue using images of wild blueberry fields captured during application timing intervals during the 2019 season. YOLOv3 and YOLOv3-Tiny identified hair fescue with AP scores of 75.83% and 75.46%, respectively. Detections in 1280x736 resolution images were achieved at 16.2 frames per second (FPS) with YOLOv3 and 82.2 FPS with YOLOv3-Tiny using an RTX 5000 Mobile GPU in an MSI workstation laptop. The goal will be to incorporate the networks into a smart sprayer and use machine vision to control spray applications. Using CNNs to limit herbicide spray will create major cost-savings for wild blueberry producers.
Keywords:
Résumé:
Mots-clés:
Citation:
Volume:
Issue:
Pages -
Contributor:
Date: 2021-06-11
Technical field:
Conference name: 5th CIGR International Conference and CSBE-SCGAB AGM 2021, Quebec City,QC, 11-14 May 2021.
Session name: World Congress on Computers in Agriculture and Natural Resources (WCCA 2)
Other information:
Type: Text.Abstract
Format:
Publication type: Presentation
Source:
Relation:
Coverage: Canada
Language 1: en
Language 2:
Rights: Canadian Society for Bioengineering
Notes:
Other files:
DOI:
Description: Herbicides are sprayed on wild blueberry (Vaccinium angustifolium Ait.) fields in eastern Canada and Maine to manage weeds such as hair fescue (Festuca filiformis Pourr.). Commercial sprayers provide a uniform application of herbicide, while the weeds often grow in patches. Smart sprayers provide an opportunity to reduce herbicide use by only spraying areas of the field with weed cover. Research has been done using machine vision systems which rely on green colour segmentation, colour co-occurrence matrices, and deep learning convolutional neural networks (CNNs) for automatic detection of areas with weed cover. Green colour segmentation is not optimal because multiple weeds found in wild blueberry fields contain green, and each weed may require a different herbicide treatment for effective management. Colour co-occurrence matrices are limited by long processing times and are impractical for real-time detection. Two object-detection CNNs built on the Darknet framework, YOLOv3 and YOLOv3-Tiny, were trained to detect hair fescue using images of wild blueberry fields captured during application timing intervals during the 2019 season. YOLOv3 and YOLOv3-Tiny identified hair fescue with AP scores of 75.83% and 75.46%, respectively. Detections in 1280x736 resolution images were achieved at 16.2 frames per second (FPS) with YOLOv3 and 82.2 FPS with YOLOv3-Tiny using an RTX 5000 Mobile GPU in an MSI workstation laptop. The goal will be to incorporate the networks into a smart sprayer and use machine vision to control spray applications. Using CNNs to limit herbicide spray will create major cost-savings for wild blueberry producers.
Keywords:
Résumé:
Mots-clés:
Citation:
Volume:
Issue:
Pages -
Contributor:
Date: 2021-06-11
Technical field:
Conference name: 5th CIGR International Conference and CSBE-SCGAB AGM 2021, Quebec City,QC, 11-14 May 2021.
Session name: World Congress on Computers in Agriculture and Natural Resources (WCCA 2)
Other information:
Type: Text.Abstract
Format:
Publication type: Presentation
Source:
Relation:
Coverage: Canada
Language 1: en
Language 2:
Rights: Canadian Society for Bioengineering
Notes:
Other files:
©
2023
CSBE-SCGAB