Systems of digital "vision" and artificial intelligence are being developed for use in agricultural machinery at Samara University, reported the press service of the university on Tuesday.

"Together with colleagues from scientific institutions in Moscow and Novocherkassk (Rostov Region), Samara scientists intend to offer the domestic agricultural producer digital 'vision' systems for agricultural machinery based on technologies that were originally developed for space. In order to do so, the university set up a center for the development of technologies for 'smart' agriculture," the statement reads.

The university expects that space technology should increase the efficiency of crop cultivation, reduce fertilizer consumption and increase crop yields by about a quarter. For example, it is possible to remotely determine soil moisture and mineral content, to detect the presence of diseases in plants and even foci of the spread of insect pests. Neural networks will automatically analyze the images obtained from the systems of agricultural machinery's "vision". Earlier, specialists from the Department of Technical Cybernetics of Samara University created a compact space hyperspectrometer for promising domestic satellites and, together with scientists from the Department of Supercomputers and General Informatics, developed methods for processing and classifying hyperspectral images of the Earth's surface obtained from orbit.

"We can equip a machine that is, for example, irrigated with hyperspectral equipment. After all, a hyperspectral image allows you to see a lot of things that cannot be seen in a normal black-and-white or color image. And the sensor will instantly determine whether a field needs to be watered or not. We plan to use less than 50 spectral channels in the wavelength range of 0.4-1.05 micrometers in order to do so. This technology saves agricultural producers, and, in fact, we create 'smart' agriculture", explained Nikolai Kazansky, professor of the Department of Technical Cybernetics at Samara University.

As part of the work, scientists will pay special attention to the technical design of sensors — it should be very simple and cheap enough for mass use in agricultural machinery. Hyperspectral sensors can be installed not only on ground vehicles, but also on drones, this will allow to immediately assess the state of big areas of agricultural land. Agreements on relevant tests have already been reached with Samara State Agrarian University.

Neural network training

In addition to creating hyperspectral sensors suitable for mass use, scientists will develop algorithms for the reconstruction and analysis of the resulting hyperspectral images using deep learning methods of neural networks.

"The scientific school of Academician of the Russian Academy of Sciences Viktor Soyfer, established and successfully operating at the university, will help us with this: the methods of recognition of hyperspectral information accumulated over decades will allow us to train the neural network so that it can analyze how much, for example, phosphorus is lacking in the soil," Kazansky added.

Source: TASS.com