Russian specialists from the Faculty of Information Technologies of the Moscow Technical University of Communications and Informatics, under the strict guidance of Mikhail Gorodnichev, the dean of the faculty, were able to independently create a unique neural network of its kind. The latest system was developed specifically to automatically recognize various brands of vehicles, simplifying the work of a person in certain areas of his activity. This was announced today, on January 31, with reference to representatives of the university, according to the information publication CNews.
University specialists said that in the modern world there are many different tasks that can be automated using the so-called artificial intelligence, thereby increasing the efficiency of work in many areas. But before, the list of actions that can be automated through neural networks and machine learning consisted solely of tasks that do not require creative thinking, because the machine, which is quite natural, does not have such qualities. But thanks to the active development in this direction in recent years, the list of tasks that AI can solve has noticeably expanded.
And for one of these tasks, the university staff decided to create their own ultra-precise neural network (or CNN), which is, of course, quite complex technically. The fact is that the task of the system developed in Russia is to receive various images that are supplied to artificial intelligence in the format of input information. Next, the system should analyze the received data and, based on the results of its work, issue the exact name of the object class shown in the image. And in order for the machine to understand these very classes and be able to distribute objects between them, the neural network was previously trained using the mechanics of the robust loss function.
The authors of the project said that when developing their neural network, they used data obtained from the Auto.ru service, as well as content from outdoor video surveillance cameras. Ultimately, the project database contains more than 90,000 copies, which were then repeatedly placed and processed by specialists. Now the neural network is able to identify various cars and, importantly, their brands even for individual body elements, and the creators of this technology claim that the accuracy of the analysis is extremely high. Accordingly, the development can be used for video surveillance cameras on the roads, systems responsible for automatically issuing fines and many other tasks.
Source: Trash Box
I’m Meagan Diaz, a news writer and author at World Stock Market. My main focus is on technology and stock market trends, and I’m passionate about helping readers stay informed on the ever-changing landscape. I bring extensive knowledge of the industry to my work as well as a knack for storytelling that makes my articles both accessible and engaging.
