We use cookies on our website. Some are necessary for the operation of the website. You can also allow cookies for statistical purposes. You can adjust the data protection settings or agree to all cookies directly.
Collaborative project: Reliable identification of the growth center of weeds in camera-controlled mechanical weeding machines using artificial intelligence in image processing - subproject A (KIdetect)
Project
Project code: 28DK132A20
Contract period: 18.06.2021
- 17.06.2024
Budget: 364,653 Euro
Purpose of research: Experimental development
Keywords: crop production, AI Artificial Intelligence, crop production, weed, resource protection, resource efficiency, crop production, agricultural engineering
The objective of project "KIdetect" is an automatic and reliable identification of weeds in real-time using AI-methods for image processing. Based on the image information, 3D-reconstructions of partial areas of the weeds shall be generated in order to enable a target-oriented control of weeders even with strong weed infestation. The three-dimensional positions of the growth centers of individual weeds, which may not be visible directly in camera images due to occlusions by leaves of other plants, are in the center of attention. In practice, human eyes combined with human intelligence are able to identify growth centers of weeds reliably. For instance, it is possible to assign quite a lot of leaves to a specific plant even if only fragments of these leaves are visible due to dense plant covering. In addition to model based image processing algorithms, AI methods shall be used in this research project to detect plants reliably. Different architectures and configurations of CNNs (convolutional neural networks) shall be trained, evaluated and compared with the detection performance of human beings. The potential of using AI-methods for the identification of the center of growth of plants gets boosted enormously by the application of SWIR-cameras currently in the stage of research, since with the help of novel image information segments of plants can be distinguished on the basis of their spectral finger print, e.g. the water content. With regard to the detection of weeds, a multi-spectral camera is needed which covers both the visible spectrum (VIS) and the SWIR-spectrum and in this way makes a stereo image possible in real time. Goals of this research project are to study suitable technologies for cameras and to analyze the potential of novel image information with respect to highest possible detection rates.
Section overview
Subjects
- Crop Protection
Framework programme
Funding programme
Excutive institution
Funding institution
Project management agencies
Associated projects: KIdetect
- Collaborative project: Reliable identification of the growth center of weeds in camera-controlled mechanical weeding machines using artificial intelligence in image processing - subproject B
- Collaborative project: Reliable identification of the growth center of weeds in camera-controlled mechanical weeding machines using artificial intelligence in image processing - subproject C
- Collaborative project: Reliable identification of the growth center of weeds in camera-controlled mechanical weeding machines using artificial intelligence in image processing - subproject D