Veröffentlichung von "Detection of Rumex spp. in Grassland using Multispectral Images and Deep Learning" / Publication of "Detection of Rumex spp. in Grassland using Multispectral Images and Deep Learning"   [07.11.24]

ENGLISH VERSION BELOW.

Ein weiteres Paper ist aus der Arbeit des KI-Fachgebiets erschienen.

Hier kommen die Hard-Facts zur Veröffentlichung.

Titel:
Detection of Rumex spp. in Grassland using Multispectral Images and Deep Learning

Autoren:
J. Heil and A. Stein

Venue:
81. Internationale Tagung LAND.TECHNIK AgEng 2024

Abstract:
Rumex spp. (Rumex) is an important weed in grassland fodder production. Most livestock avoid Rumex due to its oxalic acids. Furthermore, Rumex is competing with other plants for space and nutrients in grasslands. The roots of established Rumex are robust and the early development and lifespan of seeds assure a widespread occurrence [1]. Therefore, adequate control of Rumex is necessary. Conventional farms can resort to the application of herbicides, either manually and targeted to individual plants with a backpack sprayer or by utilizing mechanisation, i.e., larger plant protection sprayers. In organic farming, Rumex is predominantly controlled manually [2]. To decrease the usage of herbicides and the manual labour to treat Rumex, an automated individual plant control constitutes a viable solution path. A first step is the automatic detection and localisation of Rumex based on visual sensors. This spatially registered information can be later used to generate application maps. In a second step, using real-time detection with high-resolution proximal sensing data, as considered in this study, enables the automated treatment of Rumex. To this end, the state-of-the-art deep learning models YOLOv8n, YOLOv8l, YOLOv9t and YOLOv9c have been compared for their ability to detect Rumex on proximal sensing image data comprising both multispectral and RGB channels. The combination of certain RGB channels with a multispectral channel revealed partially improved performance for real-time capable YOLO-variants.

Unter folgendem Link kommen Sie zur gesamten Übersicht der bisherig veröffentlichten Forschungsarbeiten des Fachgebiets: ki-agrartechnik.uni-hohenheim.de/veroeffentlichungen

--

Another paper has been published with the participation of the AI department.

Here are the hard facts about the publication.

Title:
Detection of Rumex spp. in Grassland using Multispectral Images and Deep Learning

Authors:
J. Heil and A. Stein

Venue:
81st International Conference LAND.TECHNIK AgEng 2024

Summary:
Rumex spp. (Rumex) is an important weed in grassland fodder production. Most livestock avoid Rumex due to its oxalic acids. Furthermore, Rumex is competing with other plants for space and nutrients in grasslands. The roots of established Rumex are robust and the early development and lifespan of seeds assure a widespread occurrence [1]. Therefore, adequate control of Rumex is necessary. Conventional farms can resort to the application of herbicides, either manually and targeted to individual plants with a backpack sprayer or by utilizing mechanisation, i.e., larger plant protection sprayers. In organic farming, Rumex is predominantly controlled manually [2]. To decrease the usage of herbicides and the manual labour to treat Rumex, an automated individual plant control constitutes a viable solution path. A first step is the automatic detection and localisation of Rumex based on visual sensors. This spatially registered information can be later used to generate application maps. In a second step, using real-time detection with high-resolution proximal sensing data, as considered in this study, enables the automated treatment of Rumex. To this end, the state-of-the-art deep learning models YOLOv8n, YOLOv8l, YOLOv9t and YOLOv9c have been compared for their ability to detect Rumex on proximal sensing image data comprising both multispectral and RGB channels. The combination of certain RGB channels with a multispectral channel revealed partially improved performance for real-time capable YOLO-variants.

The following link will take you to the complete overview of the research work published to date by the department: ki-agrartechnik.uni-hohenheim.de/veroeffentlichungen

Zurück zu Aktuelle Themen