Russians no longer hide from drones: Armed Forces are equipped with the latest machine vision system

Machine vision reduces the dependence of drones on the operator, the geolocation of GPS and the counteraction to the reb - the drone is looking for the target itself. In the Russian-Ukrainian war, both sides compete in the implementation of this technology. At the end of 2023, Talan Systems presented its own machine vision system for MLP Talan drones, which uses machine training and is available for installing almost all drones now being implemented in the Armed Forces.

The developers spoke about the system dev. ua. The universal platform of machine learning is based on artificial intelligence technology for object recognition. MLP Talan allows you to identify images of drone cameras in radio electronic fighting. "When the connection is lost, the difference between Machine Vision, ordinary trekking and the detection of objects, especially in priorities. The system is taught to recognize objects hidden by others.

In hostilities, people and equipment are often hidden behind obstacles where they need to be found. Given all the details, one specific model studies during the month, which allows to obtain 60-70% of the accuracy of recognition. The platform is suitable for intelligence: it monitors the movement of objects in real time and "leads" their camera. Also, in real time, MLP Talan displays in the interface information about the target: its type, status and other data.

In the battle, drones with MLP Talan can automatically track the whole objects without operator intervention and in the conditions of HR. You can prioritize the choice of specific types of goals in advance. After determining, the system can automatically control the drone for further observation or damage. The system can be installed on almost any air or land drone. These can be aircraft, terrestrial self -propelled drones, armed turels, including reconnaissance and impact UAVs.

The platform uses machine learning to assign classes to objects, recognize the types of enemy equipment and other objects that can be the purpose of drone. The developers argue that the software solution should be used taking into account the hardware and in contact with drones manufacturers to integrate the program into specific models of drones. Of course, in the integration process, hardware restrictions of drones will be manifested, but compatibility is very wide.

The developers are counting on close cooperation with the military to improve their own product, taking into account real combat conditions. There is a process of implementing the development of the Armed Forces, but the details and terms are not called for safety reasons. According to the developers, the solution is industrial and systemic. They are based on modern neural models of production of Neurotechnology (Lithuania), which is a partner of Talan Systems.

Tens of thousands of photos are used to teach identifying the goals of the neural network. Each image has a marking to identify objects on it. About a fifth of images can be generated or fictional. In the training finals, the system knows what the face looks, specific samples of technology and other items. "We can also work with houses, food, appliances, can be taught to distinguish between Ketchup, sauce, as well as tanks, distinguish BMP from cars and planes.