Scientists of Samara University have created the unique system consisting of a hyper-spectrometer and a special neural network for plant recognition, capable of detecting weeds. According to experts, the development will make it possible to create a sensor that, in tandem with the sprayer and the nozzle control system, will not only find, but also immediately destroy weeds, distinguishing them from “useful” plants. Results of the study are published in the collection IOP Conference Series: Earth and Environmental Science.

According to the University scientists, weeds develop very quickly, considerably outpacing development of cultivated plants. They take away light, water, and nutrients, creating unfavorable conditions for other crops growth. Moreover, young weeds can spread plant diseases, and curly weeds can make it difficult to harvest, which leads to large crop losses.
Researchers from Samara University have proposed a new way to combat weeds. They have developed the system consisting of a hyper-spectrometer and a special neural network for recognizing plants of different species, which is capable of distinguishing both weeds from cultivated plants and different types of cultivated plants.

“We have obtained the system to solve a number of applied tasks, primarily in framework of precision agriculture. The hyperspectral camera developed at our University forms 150 spectral channels. It is not always possible to analyze this data array by applying simple methods, so the key result of the work was creation of the neural network that can recognize three types of plants. In future, as the technology develops, the list of recognized crops will expand”, told us Roman Skidanov, Professor of the Department of Engineering Cybernetics of Samara University.

He noted that the developed system will make it possible to create a weed sensor to targetedly destroy weeds, without affecting cultivated plants. To do this, the device must be used together with the automatic sprayer to be integrated into the nozzle control system.

Besides, according to the scientists, the neural network combined with the hyper-spectrometer can be used as the basis of vision systems for unmanned agricultural machines. It can also be applied on UAVs as the system for plant health diagnostics and general monitoring or control of crops.
As Romar Skidanov noted, the development will help save and reduce the cost of some processes in agriculture.

“The economic effect may amount to several hundred rubles for each cultivated hectare. It consists of saving expensive chemicals in course of targetedly processing plants, and increasing yields”, he stressed.
Samara University is a participant of the Russian Government University Support Program “Priority 2030” of the National Project “Science and Universities”.