Neurocle Releases Neuro-T & Neuro-R v4.0 for AI Visual Inspection

Neurocle has launched version 4.0 of ‘Neuro-T’ and ‘Neuro-R’. Neuro-T is an AI deep learning vision trainer that creates AI models for vision inspection. Neuro-R is a runtime API that applies the inspection model created by Neuro-T to the manufacturing process line in real time. This new version solves the problem of insufficient defect data and labeling resources in the manufacturing field.

The newly added models and functions are as follows.

AI Model for Virtual Defect Creation: GAN Model · Generation Center

To create an AI model that detects defects, both normal and defective image data is needed. And the richer the amount of data, the more likely it is to create a high-performance model. However, in certain industries, it is difficult to collect defect data. This makes it difficult for users to advance the performance of the model. GAN Model · Generation Center are designed to solve these cases. With this function, users can create virtual defects similar to the real ones based on a small number of defect images. Virtual defect created through GAN Model · Generation Center can be inserted into a normal image using various ‘defect creation tools. Once virtual defect images are made, they can be utilized to create model that decides normal / defective.

Unsupervised Learning Based Models: Anomaly Classification, Anomaly Segmentation

In this update, 'Unsupervised Learning', which can identify normal and abnormal by learning only from normal images, was also added. This function is also good news for users who lack defect data. Users can choose between Anomaly Classification and Anomaly Segmentation depending on the judgment criteria and the method of result confirmation.

AI-assisted Labeling Tools: Auto-Selector, Keyword Labeler

One of the most cumbersome and resource-intensive processes in deep learning projects is ‘image labeling.’ This is because how precisely the defective areas are labeled affects the performance of the model. In this update, Auto-Selector and Keyword Labeler were added to help users significantly reduce labeling resources.

‘Auto-Selector’ is a function in which a deep learning algorithm automatically labels areas with similar characteristics based on objects. This function is recommended when the area to be labeled has a complex shape and requires precise labeling. Meanwhile, ‘Keyword Labeler’ is a function that, when you enter a specific keyword, the part of the image corresponding to that keyword is labeled in the form of a box. This function is recommended when uniform labeling of objects in an image is required because the object characteristics are distinct.

In addition, new models and functions such as △Multi-Model Export and △Transfer Learning model have been added to support users’ successful AI vision inspection.