“The appliance of artificial intelligence (AI) within the manufacturing industry ought to be more cautious. Unlike management and work efficiency improvement, it’s directly related to the defect or fatal lack of the product.”
Founded in 2019, Neurokkle has been specializing in the advancement and business of vision inspection software that goals for universal AI software that might be utilized by non -professionals. In consequence, the corporate has grown rapidly with a break -even point (BEP) in two years.
Roh Jun -gu, a neurokle business development manager who meets the client and listens to the voice of the positioning, emphasized that “the results of specializing in actual use.” Based on corporate feedback, it’s constantly advancing.
Neurokkle software has been introduced that there are two foremost differences.
First, I heard ‘quick application’. It emphasized that it has all of the crucial parts for site applications comparable to data labeling, learning, data synthesis, essential deep learning test model, and model performance inspection inside a single software.
The second known as ‘Auto Deep Learning Algorithm’. This routinely explores the optimal deep learning model structure and learning parameters to create a high -performance non -inspection model. In other words, it’s an algorithm that routinely creates a high -performance model, which minimizes the time and price of the corporate.
The corporate has developed model learning software, neuroti, neurox, and neuroal software for model application. Annually, the corporate ran regular updates and recently introduced version 4.3 of Neuroti and Neuroal.
Specifically, version 4.3 cited the strengthening of optical character recognition (OCR) and multiple models. This can also be the results of actively reflecting corporate feedback.
First, OCR is characterised by advanced ‘text rule’. Previously, there was a limit that it was essential to acknowledge and post -processing in certain images and documents. He explained that he would support the text rules to acknowledge only specific layouts. For instance, it’s to speed up the work speed by setting just one expiration date within the product packaging material to detect it.
It is usually equipped with a five -sword correction function to resolve the issue of decreasing accuracy when confusing similar forms (0 and O, 1 and I) or decreasing image quality.
Multiple model connections were also essential. It’s because within the manufacturing site, it is usually utilized by combining various vision inspection models comparable to ▲ classification, detection, and OCR installed within the neuronic software. But in version 4.3, the difference is that the multi -model might be applied ‘faster and fewer’ multiple models.
“We could have the most important variety of functions in Korea as a deep learning software,” he said.
“The manufacturing and market itself shouldn’t be good, however the importance of vision tests is not going to be lowered because it is directly related to the standard of the product,” he said. It can increase. ”
It also emphasized that it goals to be ‘partial automation’ somewhat than full automation for accurate vision tests.
“I’m conscious of the recent technical trends comparable to AI agent, multimodal, and orchestration,” he said. I’ll focus. ”
The concept of synthetic data also revealed. “I feel synthetic data is an indicator of assistance within the field of regard inspection,” he said. “I feel the information labeling itself may be very essential.”
In consequence of adhering to those principles and company vision, they’ve received a response from corporations. Since its inception, the number of consumers has increased yearly, but additionally provides software to 25 countries.
Noh Jun-gu, Neurokkle Business Development Manager, said, “We are going to take part in the March Smart Factory-Automation Industry Exhibition ‘Automation World’ and ‘Inter Battery 2025’ exhibition to showcase deep learning vision test solutions and inspection demo.” “We are going to upgrade the software.”
By Jang Se -min, reporter semim99@aitimes.com
