Published on: 2026-07-23
Source: Novosibirsk State University –
An important disclaimer is at the bottom of this article.
Employees Research Center in the field of Artificial Intelligence at Novosibirsk State University (NSU AI Center) They have begun field trials of a system that allows detecting and accurately localizing infestations of potato fields by the Colorado potato beetle using a drone. The development is being conducted at the request of the large agricultural company “Dary Ordynska” and is intended to help transition from widespread chemical treatment of fields to targeted, more environmentally friendly and economical pest control.
Today, agronomists often simply walk across the field diagonally and inspect plants along selected routes. This allows them to assess the overall situation but does not guarantee that breeding sites of the beetle will be noticed at an early stage and precisely localized.
— As a result, the treatment is carried out across the entire field, although the actual infected areas may occupy limited sections. If they can be accurately identified, it is possible to treat them locally and avoid flooding the entire field with chemicals, — explained the engineer of NSU AI Center Maxim Magerov.
The “Colorado-Detector” project is an independent development, but it relies on the previously acquired competencies of the center’s staff in the field of onboard data processing from unmanned aerial vehicles.
In particular, the Center’s team worked on a system for locating people using UAVs, where neural networks analyzed the video stream directly onboard, and processed information was sent to the ground. In the new project, this approach has been adapted to the tasks of the agricultural sector: based on an existing quadcopter, a special module with four cameras arranged in a fan-shaped array was developed to capture the widest possible strip of potato crops from a low altitude in a single pass.
A drone is flying at an altitude of about 3-4 meters above the bushes; currently, the team is selecting the optimal distance. On board is a high-performance single-board computer that receives images from cameras, analyzes them using machine vision algorithms, and detects adult Colorado potato beetles.
The feature of the system is that it does not focus on damaged leaves but searches directly for the insects themselves. At early stages, the beetles are already present on the plants, but there are no visible signs of feeding yet — according to the developer, relying on defoliation would result in lost time.
— It is more difficult to see larvae from the air because they are often under leaves, and the current version of the algorithm is not trained on them. However, adult beetles sitting on the upper parts of plants can be detected individually, — Maxim Magerov clarified.
After processing the images onboard, the results are transmitted via a digital channel to the ground station. The agronomist opens a web application on a laptop with a field map marked with points where beetles were detected. This allows for decisions regarding localized treatment of specific areas, reducing the use of insecticides and lessening the environmental impact. This scheme fits into the global trend of precision agriculture, where drones and neural networks are used to monitor pests, diseases, and crop conditions.
The system is currently in the active field testing stage: developers regularly go out to actual fields, adjust shooting parameters, collect new data, and test algorithms in real conditions.
The key task for the near future is to expand the training dataset. Until now, the neural network has been trained mainly on synthetic images of the beetle, with few real shots available.
— Field tests are currently underway, where we not only select optimal operating parameters, such as flight altitude, but also accumulate images obtained under real conditions, which will allow us to further train the neural network, — the developer said.
This is necessary for the model to reliably distinguish the beetle under various conditions: different lighting, at different growth stages of the potato, with various backgrounds and angles. Global studies show that the use of synthetic data can significantly improve the quality of detectors, but full practical effectiveness requires a combination of artificial and real samples.
According to the team’s plans, the current season will be dedicated to debugging the hardware complex, collecting data, and improving the accuracy of the models. In autumn and winter, the developers expect to carry out technology refinement, legal registration of rights to it, and preparation for scaling the solution.
In the future, the “Colorado Detector” is planned to be offered to other agricultural producers as well. The first customer will remain “Dary Ordynska,” but the developers hope that other farms seeking to reduce chemical protection costs and transition to more precise pest control will show interest in this technology.
In a broader context, the project demonstrates how competencies in artificial intelligence and onboard data processing on drones can transition from tasks such as saving people and area monitoring to applied projects for agriculture and managing elements of urban infrastructure, on which other developer teams of the NSU AI Center are working.
Please note; This information is raw content obtained directly from the information source. It is an accurate report of what the source claims and does not necessarily reflect the position of MIL-OSI or its clients.