| New label noise detection feature added
| A complete overhaul of the One Class Learning feature
| Up to 5 projects can be carried out simultaneously

Sua Lab announced on the 2nd that it has released 'Sua Kit 2.3', an upgraded version of 'Sua Kit 2.2' released in November 2018.

SuaKIT is a deep learning-based machine vision defect inspection software that can detect atypical and irregular defects that could not be detected with conventional machine vision inspection through deep learning. It can be applied to product inspection in various industries, including the electrical and electronic industries such as displays, semiconductors, and solar power, as well as the automotive and food and beverage industries.

Since the first release of 'SuaKit 1.0' in June 2017, SuaLab has continuously updated its products with a focus on solving various variables and difficulties that occur in actual manufacturing vision inspection sites, such as image comparison and the Visual Debugger, a function that shows the area that serves as the basis for judgment in an image that has been judged to be defective.

In SuaKit 2.3, the 'Label Noise Detection' function has been newly added, the 'One Class Learning' function has been completely revamped, and user-friendly features have been increased.

The label noise detection feature identifies which types of images are unclear during training with the provided data and which types of additional data are needed to improve the performance of the deep learning network. This feature not only saves time during image label review but also improves detection performance.

The One-Class Learning feature has been enhanced. In cases where collecting defective images for product defect detection is difficult or time-consuming, training is now performed using only normal images, enabling the system to distinguish between normal and abnormal images. Additionally, unlike the typical learning method where labeling is done directly by humans, applying this function can reduce labeling costs by utilizing the defective location detection function.

In addition, it supports the 'Multi Project' function that allows up to 5 projects to be carried out simultaneously, and the 'Task Manager' function that automatically manages GPU resources to optimize the productivity of concurrent tasks. It has additionally equipped various functions to increase the convenience of machine vision engineers and general users who actually use the tool, such as the 'Image Tag' function that allows for detailed classification of defect detection, and the 'Memo' function that allows for easy recording of detection contents during product inspection.

“With the advent of the Fourth Industrial Revolution, one of the biggest challenges facing Korean manufacturing companies is leveraging AI technology to improve productivity,” said Song Ki-young, CEO of Suarab. “We will continue to develop products with more convenient features from the user’s perspective and showcase the best technology.”

Meanwhile, Suarlab will participate in the '2019 Busan International Smart Factory Conference & Expo' held in Busan from September 18th to 20th and hold an exhibition and demonstration of Suarlab 2.3.