RSS
Логотип
Баннер в шапке 1
Баннер в шапке 2

Beorg Smart Vision

Product
The name of the base system (platform): Artificial intelligence (AI, Artificial intelligence, AI)
Developers: Beorg (Biorg)
Last Release Date: 2024/10/31
Technology: EDMS - Streaming Recognition Systems

Content

Beorg Smart Vision is a solution for digitizing archives and recognizing documents. According to the developers, the program allows you to process kits (including even handwritten documents) for up to 20 minutes.

Beorg Smart Vision is a flexible and versatile solution that scales to any task and data volume. At the same time, to launch the project, the customer does not need to purchase expensive equipment and licenses, as well as involve additional specialists - scan images of documents are processed, verified and classified in a protected cloud on the Biorg side. Even "complex" data arrays (handwritten text, blurry prints, dilapidated and fuzzy records with personal data), unlike "boxed" OCR products, Beorg Smart Vision can recognize and translate into electronic form automatically. This minimizes the impact of the "human" factor on the quality of digitization, and its accuracy reaches 99.9%, which is an absolute maximum for the market.

2024: The neural network recognizes and checks the passport of the Russian Federation 4 times faster

The updated technology was presented on October 31, 2024 by Biorg. An improved neural network is 4.25 times faster and 8 times more accurate than its previous model recognizes the passport field with a machine-readable record (MFZ, MRZ). This helps to control the quality of processing of all data in the document and automatically check the passport for relevance. Speed ​ ​ and accuracy are important when scoring loan applications in banks, MFIs, when drawing up an agreement in insurance companies, etc.

The peculiarity of this neural network model is that AI does not try to localize data and recognize individual characters in a string. The network finds and reads all the information at once, just like the human eye does. The model understands that it coped with recognition well during multi-stage postprocessing - by the number of recognized characters, by the content of the line, etc. Such a check takes a split second.

File:Aquote1.png
One of the qualities of the approach is that it can be scaled to other types of documents. The model can work not only on graphics accelerators (video cards), but also on ordinary processors. This makes the recognition process more cost-effective in terms of infrastructure costs. And even more environmentally friendly, given how much electricity is absorbed by video cards and how much heat they emit in data centers. - said the general director of "Biorg" Ruslan Aligadzhiev.
File:Aquote2.png

The neural network copes well even with documents whose photos are taken in conditions of lack or excess lighting, as well as with images of documents photographed at an angle with geometric distortions. Traditionally, such images have caused processing problems. The recognition accuracy of the MFZ line is more than 99%.

2023: Application in My Shift App

Thanks to artificial intelligence (AI) technologies, users of the My Change platform - applications for quickly finding part-time jobs near home - can register an account up to 20 times faster and switch to stores of large retail chains. To do this, you need to download your passport and medical book. The documents are recognized and checked by the AI technology Beorg Smart Vision from the Russian company Biorg. Thanks to the solution, you no longer need to enter data manually, and retail chains gain confidence in the authenticity of documents of temporary executors. Beorg announced this on August 29, 2023. Read more here.

2021: Optimization of the neural network "pipeline" for automatic passport recognition

The Russian company Biorg, a Skolkovo resident, developer of the domestic platform for streaming digitization of documents Beorg Smart Vision, on July 22, 2021 announced the optimization of the neural network "conveyor" for automatic passport recognition. According to the company, the processing time of one passport spread is no more than 2 seconds while working with a flow of documents, the quality of recognition is above 99%.

The neural networks were finalized at the expense of the Strategic Computer Technologies and Software grant from the Skolkovo Foundation. The amount of the grant is five million rubles.

The estimated throughput of the updated platform has doubled. The system automatically recognizes at least 95% of fields containing information, initially this figure did not exceed 50-60%. The remaining 5% of the uncertain fields are manually verified by the data entry operators, which increases the quality of the final processing to almost one hundred percent.

File:Aquote1.png
"We use a service approach, where we focus on the final result: high quality of digital documents and the ability to save the customer's budget. The Beorg Smart Vision hybrid platform is SaaS a solution for streaming digitization of not only passports, but also almost all types of documents, including handwritten ones. The amount of information that neural networks cannot cope with is processed by our operators. Such control is necessary when digitizing archives, drawings, bank mortgage in "conveyors," etc., wherever accuracy is needed. Combining the technological approach with crowdsourcing capabilities, we digitize even low-quality scan images taken in low light, with glare, etc., which usually presents problems for automatic recognition systems, "said the founder of Biorg. Georgy Zuev
File:Aquote2.png

2020

Recognition as important and worthy for replication around the country

Beorg Smart Vision, an intelligent platform designed to recognize objects and data using artificial intelligence technologies, is recognized as important and worthy of replication around the country. The development of Biorg received this assessment from the Agency for Strategic Initiatives (ASI) in the framework of the Smart.AI & Data competition. In November 2020, in an interview with TAdviser, Biorg founder Georgy Zuev spoke about how the sphere of state information reporting becomes a platform where the most advanced technological and organizational solutions are massively introduced. Read more here.

Scaling Beorg Smart Vision Document Capture on Mail.ru Cloud Solutions

On May 12, 2020, Biorg announced that it had transferred its services for recognizing complex objects and data, including handwritten text, to the Mail.ru Cloud Solutions (MCS) cloud platform. The platform will provide privacy, speed of processing and transfer of information, as well as the ability to quickly scale Beorg Smart Vision services. Read more here.

Expanding platform functionality taking into account the demand for the remote work format

On April 14, 2020, Biorg, a Russian developer of intelligent systems based on neural networks and machine learning, announced the expansion of the functionality of its crowdsourcing platform, taking into account the change in the format of work of most commercial and state organizations. The presented opportunities will be of interest primarily for health care, education, housing and communal services, the MPSC and the social sphere.

Beorg Smart Vision

As noted in Biorg, it was in these areas that the volume of typed routine processes increased sharply (for example, processing incoming applications or requests, voice assistants, prescriptions, analyzes, sick leave, school assignments, etc.). Given the large number of requests related to solving these problems, Biorg is meeting users halfway and supporting everyone who is forced to work remotely - the company is redesigning its combined Beorg Smart Vision platform for existing needs. At the same time, crowdsourcing service, taking into account the current situation with the spread of the epidemic in the country, will be provided at cost.

According to the developer's statement, Beorg Smart Vision is an innovative business platform that allows you to process various objects and data with special accuracy thanks to the dual processing method (neural networks plus verification by operators). The core of the system includes proprietary algorithms and non-standard ensembles of specialized neural networks, the Faster-RCNN architecture of deep learning. This allows you to automatically build the most complex models that take into account the most subtle nuances of incoming information with the minimum participation of operators (there is no "narrow neck"). Platform features: high speed (from 2 seconds per document) and low cost of processing digital materials, fast scalability with recognition quality above 99%.

According to Biorg, in a pandemic situation, for new categories of Beorg Smart Vision users, the resources of the company's operator pool (more than 55,000 people), focused on specialized tasks and highly specialized topics with a well-functioning official employment system, ratings and training, come to the fore. The work will be built within the customer's SLA using special software, mandatory verification of data on directories and logical conditions, and the results will be integrated with the organization's IT system.

File:Aquote1.png
"The events of recent months have shown that the world in general and our business world in particular are experiencing serious changes. More remote work, more specialization and as little overhead as possible - this is how we see the slogan of the coming years. The work of the operators of our crowdsourcing platform becomes relevant in this regard. These are people of different specialties who know how to work well with images, do digitization, data markup, moderation and many other typed operations. They are ready to start any projects as soon as possible in 24x7 mode, "

noted' Georgy Zuev, founder of "Biorg" '
File:Aquote2.png

2019

Key about the data processing platform

  • Tasks are downloaded to the platform and processed online by neural networks and performers.
  • Initially, operators are used to train the neural network and mark up the datacet, then the processing takes place automatically. Only problematic tasks are subject to visual inspection.
  • The neural network copes with handwritten text. The natural purity of the data grows with each new array. Reprocessing and reverse reconciliation of data with the document allows you to bring the quality to the maximum.
  • Transparency for the customer: control of work in real time, online interaction with the contractor.
  • It is possible to organize work within the organization (enterprise cloud)

Processing more than 30 million documents

According to Biorg, with the help of its key development - the Beorg Smart Vision platform - and a crowdsourcing service with more than 55,000 operators, more than 30 million documents were processed in 2019.