Site-specific weeding robot using high electrical voltage actuator combined with UAV hyperspectral imagery for predictive management and post-assessment. – WeedElec2017

The project combines two complementary but independent detection approaches: (i) - an aerial detection approach focusing on the spatial location of «live« or «destroyed« adventices (post weeding) also usable for optimizing solutions using alternative methods of destruction, - (ii) - a ground-based approach using a robot with self-usable reconnaissance and precision destruction capabilities. The identification of species is based on the expertise acquired by Irstea in the processing of hyper-spectral images in the field, and by INRIA and CIRAD in the framework of the project Pl@ntNet ‘Artificial intelligence and deep-learning’.
The modeling of characterization processes and their integration into decision support tools, as well as the design of global weeding strategies, will be other scientific issues for which the consortium’s expertise will be required.
The destruction of high-voltage electrical adventices was previously the subject of research at IRSTEA in 1997-2000. It is an original way of destroying roots and aerial parts as glyphosate could. The proposed device helps to control the weed control and improve the efficiency of the destruction adapting the signal to the electrical impedance signature of plants (INRA) subjected to high voltage.
AGRIAL brings its expertise and means of real field experimentation.

Early results have helped develop the first solutions for detecting and discriminating weeds between them and seedlings to be preserved.
The first results reinforce our approach around the use of artificial intelligence and deep-learning techniques with the ambition of coupling with the hyper-spectral approach. The first tests of high voltage destruction are positive, although not yet using power and signal modulation.
Research on the signing of the adventitious impedance has defined the range of power and frequency that improves upon destruction.

WeedElec aims to lift, in particular, the following major scientific locks:
· Detection/identification of adventitious by hyper-spectral and color imaging, and associated deep-learning and chemometry techniques.
· The coupling of aerial detection and robot onboard detection to decide on the elimination of adventices.
· The HT electrical discharge coupling and complex adventitious impedance to determine the best high voltage action characteristics required.
We'll be interested in putting in place an integrated strategy of weeding on a typical robot.

No publications produced in 2019.
Demonstrations during the dryrun and ECPA 2019 and ANRT days

WEEDELEC 2017

We propose in this project an alternative solution to global chemical weeding, which combines aerial means for weed detection coupled with a robotized ground weeding system based on high voltage electrical energy.

The project will rely on commercial solutions concerning the aerial and ground vehicles (respectively UAV and robot). It will more particularly focus on major technical and scientific issues for the development of a future integrated weeding solution, i.e.:
- weed detection and identification, using hyperspectral imagery and deep learning techniques
- weed behavior when they are exposed to electrical stress, especially by investigating the relationship between the kind of electrical shock to be applied and the weed electrical impedance and phenology.

Questions related to aerial and robot-embedded weed detection systems will also be addressed, as well as possible environmental and safety effects of electrical shock usage on weeds, in order to design a safe integrated weeding strategy.

This project draws on previous results obtained in the FP4 European project Patchwork (electrical weeding 1995), in the FP7 European project RHEA (Integrated weeding solution, 2012) and on the Plant@Net project, devoted to automatic plant identification by deep learning. It also relies on the expertise of plant phenology scientists and weed scientists.

The WeedElec project will be an opportunity to enrich Plant@net image databases, to produce a database of electrical signatures of main weed species, to develop and test new robust algorithms for weed detection and identification, and to validate an innovative weeding solution with no chemicals.

The experimentations will be led in field crop and market gardening plots, in order to cover variate crop and weed typologies