SK Planet Combines Hyperspectral Technology with AI to Commercialize Defect Detection and Quality Control in Manufacturing Processes

SK Planet is pursuing defect detection and object identification businesses across various fields such as manufacturing, food, and environment by combining hyperspectral technology with its own developed AI. Hyperspectral cameras subdivide light from visible light to infrared and ultraviolet, generating three-dimensional information in the form of a Data Cube, through which the unique spectral characteristics of objects are analyzed.

AI-Based Data Processing and Analysis

Hyperspectral data is complex to analyze due to atmospheric distortion, high correlation, and the Curse of Dimensionality. SK Planet utilizes deep learning models based on Stacked Auto-Encoder(SAE), Deep Belief Network(DBN), and CNN to address this. In particular, 3D-CNN receives Raw data as input without preprocessing and simultaneously learns spatial and spectral information, and models applying PCA or Gabor filters as preprocessing also contribute to improved classification performance.

Application Cases in Manufacturing and Food Industries

• Secondary Batteries: Reduced finished product sample testing, which previously took more than a day, to full inspection within about 10 minutes, and detects the content and purity of raw materials in real time.
• Food Production: Capable of detecting even atypical foreign substances in flakes or powder, and checks for foreign substances inside instant food by penetrating vinyl packaging.
• Plastic Sorting: Quickly and accurately sorts collected plastics by type, going beyond the limits of manual work.

Implications and Expected Effects

The combination of hyperspectral technology and AI enhances the accuracy and predictive capability of manufacturing processes by precisely grasping multidimensional information. Real-time monitoring enables immediate detection and response to process anomalies, contributing to improved productivity and reduced labor.