No.1022E 2026Special Edition on the Technologies that Support Manufacturing and Manufacturing Equipment Development of AI Technology for Automated Visual Inspection

Category TECHNICAL REPORT
Author M. KANO M. MOREILLON
Data Analytics R&D Dept., Innovation Division
Y. ARAI
Process innovation Engineering Dept., Production Engineering Division
A. NAKASE
Data Analytics R&D Dept., Innovation Division
E. KONDO Y. YAMASHITA D. HIBI
Process innovation Engineering Dept., Production Engineering Division
Abstract Automated visual inspection requires accurate detection of tiny defects. However, deep learning-based image classification methods often struggle with low accuracy for small anomalies. To solve this problem, we propose a novel approach that integrates image partitioning with data cleansing based on defect locations. Our experimental results demonstrate that this method significantly improves classification accuracy, particularly for defects located near the center of an image. To facilitate practical deployment in manufacturing environments, we have developed an in-house, no-code platform for building and utilizing AI models. This report presents the key features of our platform and discusses its potential for widespread adoption in real production settings.
Keyword visual inspection, image classification, image partitioning, software KARAKURI
Reference
Search tag

Download Ranking of
ENGINEERING JOURNAL

Summary:5,2026