WeedElec: Localised weeding robot using high voltage electrical process.
The WeedElec project offers an alternative to global chemical weed control. It combines an aerial means of weed detection by drone coupled to an ECOROBOTIX delta arm robot equipped with a high voltage electrical weeding tool. While agro-ecology has become the daily life of many farmers, WeedElec provides an innovative, efficient and eco-responsible solution thanks to precision farming.
Substitution of chemical weeding by high voltage weeding
WeedElec’s objective is to remove the following major scientific obstacles in particular:
- Weed detection/identification, using hyperspectral and colour imaging, and associated chemometric and deep learning techniques.
- The coupling of aerial detection and onboard detection by the robot to decide on the elimination of weeds.
- A detailed study of the behaviour of weeds subjected to high voltage electrical discharge by working on the coupling of electrical discharge and complex impedance of weeds with their phenology.
We will focus on the environmental and safety effects associated with the use of electrical discharges on weeds, in order to implement an integrated and secure weed control strategy.
Weed detection is a key point of WeedElec. The available resources can be classified into three categories according to the type of information used:
- Spatial position of seedlings: the aim is to detect the seedlings by using the fixed distance that separates them during planting and then to improve their detection around their pre-positioning using algorithms that discriminate between the soil and the vegetation.
Most often, colour or near-infrared image segmentation techniques are used.
Once the seedling is located, everything that will be detected around it is then classified as a weed. This method is very effective in centimetric planting (laser guidance) and at a very early stage of vegetation. However, this technique, which does not allow weeds to be individualized on the row, will only be used as a first approach in WeedElec.
- Taxonomic identification by deep learning: The performance of image recognition algorithms is able to identify one plant among thousands of others with high recognition rates. The latest results in this field show recognition rates above 85%. (http://www.lifeclef.org/), and access to these techniques has become widely available in recent years, particularly through the Pl@ntNet mobile application (www.plantnet-project.org ; Joly & al., 2016). This approach, which requires large amounts of data, requires computing power that is difficult to mobilize on an embedded electronic solution. Knowledge distillation (Hinton & all, 2015) and transfer learning (Carranza-Rojas & all, 2017) or domain adaptation (Tzeng & all, 2017) techniques will be used to remove this lock.
- Spectral information: The hyperspectral imaging technique (one to several hundred bands, combined with chemometric treatments such as PLS-DA, has shown its ability to effectively discriminate (Hadoux & all, 2014) between weeds and crop plants. WeedElec will mainly use this technique to detect weeds by drone and proxy detection.
The principle of electric weeding was studied by Diprose (1984). This showed that the energy required to destroy a plant is proportional to its size and more important for outdoor plants than for those under glass. Irstea during the PATCHWORK and D3E projects (1995-2000) had already developed a specific technique for localized high-voltage electrical destruction of weeds.
However the characteristics of complex electrical impedance vary according to the species. It leads to different behaviours when they are subjected to high voltage. WeedElec aims to study these characteristics to generate the optimal high voltage signal from a lethal efficiency and energy consumption point of view.
The purpose of WeedElec
In the end, WeedElec’s overall objective is to assemble all these solutions on a robot to obtain a viable prototype solution that can be used in whole or in independent modules for industry.
The role of TERIA
TERIA is involved in several eco-responsible projects over a wide range of applications. Respect and improvement of the environment are an integral part of the company’s DNA. Our team of experts work daily with its partners to develop efficient and environmentally friendly solutions.
WeedElec, which uses TERIA geo-positioning, is one of the robotic agricultural applications where TERIA is a pioneer.
Source : www.challenge-rose.fr