Impix Develops Industrial AI Machine Vision... "Enables Easy and Fast Quality Inspection"

Impix recently announced on the 9th that it has developed an artificial intelligence (AI) machine vision product capable of anomaly detection through a CNN (Convolutional Neural Network)-based PaDim (Patch Distribution Modeling) model.

Impix's AI machine vision product, A²Lab Vision, is a deep learning-based industrial AI machine vision analytics solution. It overcomes the limitations of existing machine vision systems that could not accurately detect irregular defects or recognize complex patterns with many variations, and its differentiated learning methods and data generation capabilities enable it to solve challenging inspection and classification problems.

Existing vision solutions faced difficulties due to the use of high-resolution images. To address this issue, Impix utilized high-resolution images through tiling, which converts large images into multiple smaller images, and employed a flow-based learning algorithm to enable faster learning of more images. PaDim uses a proprietary methodology called Memory Bank and currently demonstrates the highest vision recognition accuracy among existing algorithms. Impix completed the patent application for this product on the 1st.

Impix is also participating in the Korea-UK international joint technology development manufacturing generative AI project conducted by the Korea Institute of Industrial Technology. This project involves the University of Edinburgh and Neurolabs from the UK, and UNIST (Ulsan National Institute of Science and Technology) and Impix from Korea. The project aims to overcome the limitations of existing data collection by developing a diffusion-based model that infinitely generates defect data, and to enhance image learning capabilities through diffusion-based algorithms, thereby providing a differentiated and expanded form of manufacturing AI vision solutions compared to existing products.

The diffusion-based generative model not only has superior generative capabilities compared to GANs, which are commonly used generative models, but also addresses issues such as mode collapse, making it a suitable approach for distinguishing between good and defective products. This project is expected to be the first domestic case of product development using a diffusion-based model, and Impix plans to upgrade and release the product through this project. The Korea Skills Agency is participating as a demand company for product demonstration, and the project will continue until 2026.

Lee Sang-ho, CEO of Impix, stated, “Based on 20 years of accumulated know-how specialized in the manufacturing industry, Impix plans to release an upgraded machine vision product,” and added, “Through the development of differentiated machine vision products, we hope to create an opportunity for manufacturing AI machine vision to take a leap forward.”