Aapresid Congress 2023 – Algorithms that learn to optimize herbicide application

One of the topics that drew the most attention at the Aapresid 2023 Congress was the use of technologies to optimize the application of herbicides.

Speaking to AgroPages, expert Guy Coleman, from the University of Sydney, assured that ″the revolution in the targeted application″ of pesticides is directly linked to ″Deep Learning Algorithms″.

He spoke on the panel titled ″From Nature to Algorithms: Detecting Weeds with Machine Learning″.

Coleman recalled that targeted pesticide application has been around for decades. Initially, it started with machines able to distinguish mainly the colors of plants.

In this way, explains the specialist, direct applications were always made in fallow conditions.

However, now instead of inserting rules, a large number of images, and manually marking the weeds in the first step, the algorithm searches for all the data of each layer that composes that image and analyzes them.

Through "Deep Learning" standards are established and with them, it is possible to "predict" which plant is a weed, even in difficult places within the crop.

He also highlighted that this innovation is led by growers around the world and can be adapted to work across cultures.

Therefore, the collection of regionalized data that reflect the wide variety of weeds present in the area, and even in their different phenological stages, is a fundamental part of the process.

The panel on new technologies applied to weed management also included Marcos Mammarella, founder of the startup DeepAgro.

For her part, Mammarella pointed out that the area occupied by weeds in Argentina is only 30%, which is why these selective control tools are strategic.

This inefficiency, added to the increase in resistant weeds in constant growth, leads to the search for new solutions where selective applications, now intelligent, stand out.

This technology, which can now be used year-round, offers benefits related to less use of agrochemicals, reduced phytotoxicity in the crop, reduced water use, and greater autonomy, including less use of drums, thus contributing to the sustainability of the system.

In turn, this economy in the use of herbicides allows access to other more expensive and previously unused active principles, diversifying the mechanisms of action and delaying the advance of resistance.

According to Coleman, for the next years, the possibilities are immense in the management of weeds.

The next challenge, according to him, will be the differential control, based on the identification and size of the weed, taking into account the context of the crop, such as, for example, differentiating grasses from cereal crops.

For this, a multimodal approach is essential, with a lot of information at the same time, such as batch history, crop biomass, and impact on yield.

(Editing by Leonardo Gottems, reporter for AgroPages)