Thanks for visiting Imaging and Machine Vision Europe.

You're trying to access an editorial feature that is only available to logged in, registered users of Imaging and Machine Vision Europe. Registering is completely free, so why not sign up with us?

By registering, as well as being able to browse all content on the site without further interruption, you'll also have the option to receive our magazine (multiple times a year) and our email newsletters.

SuaKit v2.3

Share this on social media:

SUALAB, a deep learning machine vison solution provider, released new version 2.3 of SuaKIT with an enhancement of performance and functions – on September 2nd.

SuaKIT is deep learning machine vision software that can detect atypical and irregular defects which are not easily captured by a traditional rule-based machine vision system. It is applicable to the product surface inspections in various industries such as automobile, electric and electronics, semiconductor, display, solar panel, and packaging as well as food & beverage.

SuaKIT's deep learning algorithms used for the training can automatically learn and analyze the defect criteria and differences from the normal. Only with few clicks within the SuaKIT's straightforward GUI, users can easily load images, label NOK images and defect area, train them, and create the inspecting models. The created models can be exported as API using C# or C++ and be combined with the existing inspecting algorithms.

Since launching SuaKIT 1.0 version in June 2017, SUALAB has continued developing user-friendly functions and training architectures, such as ‘Image Comparison’ or ‘Visual Debugger,’ to solve the difficulties and variables occurring at manufacturing sites.

‘Image Comparison’ is a deep learning architecture that trains based on sets of two different images and detects significant differences found in them. It is useful for finding out the defects when optical conditions or images background vary, which is very common at manufacturing sites. By focusing on meaningful differences during the inspection, the detection accuracy can be maximized.

‘Visual Debugger’ offers the users a heatmap of an image that shows parts of the image the deep learning algorithms has focused, allowing the users to fix the cause of the unsatisfying result of their inspection. This function has been developed to solve the problem that the debugging is not possible in the deep learning methodology.

In the newly released SuaKIT 2.3 version, this September, the existing functions of 'Label Noise Detection' and 'One Class Learning' have been enhanced, and the GUI has been updated in a more user-friendly way.

'Label Noise Detection' tells you which images have been mislabeled by comparing to the other trained images, and what types of additional data would be needed to improve the training performance. This function helps to reduce time and to increase accuracy in labeling large amounts of images. In the actual mass production lines, without using this function, the detection performance wasn’t improved proportionally to the image quantity increase. Having removed incorrect and ambiguous labels thanks to 'Label Noise Detection,' the detection performance was markedly enhanced.

The existing function of 'One Class Learning' has also been enhanced. It is a kind of semi-supervised learning that does not require defect image training to find out the defects. It enables the detection of NOK image and the areas with defects by training only OK images. This is particularly useful in the manufacturing processes where NOK images are insufficiently generated. Also, defect location can be found automatically, which enables the reduction of labeling cost, comparing to the general training method.

In addition, SuaKIT 2.3 version includes various functions for better usability: ‘Multi Project’, which enables users to manage up to five projects at a SuaKIT GUI window, ‘Task Manager’, which allocates GPU resources automatically for productivity optimization, ‘Image Tag’, which marks the images in one defect class into more detailed classes. Users also can leave a memo in each image, model, and project for better management of project history.

Song Ki-young, CEO of SUALAB, says “In the era of 4th industrial Revolution, it is one of the biggest challenges to increase the productivity of Korean manufacturing companies by using disruptive technologies such as artificial intelligence. With our deep learning solution, we will continue developing better functions and methodologies for users, and contribute to the new era of industry.

SUALAB will demonstrate the new features of SuaKIT 2.3 in international industry exhibitions this year – ‘Busan Smart Factory Conference & Expo 2019’ in South Korea, ‘Packaging Machinery and Manufacturing Show(PPMA) 2019’ in Birmingham, the U.K., Collaborative Robots, Advanced Vision & A.I. (CRAV.ai) in San Jose, the U.S. and etc.

Tags: 
Company: 
Other tags: