Adacotech AdaInspector Good-Sample Learning AI

Adacotech is a Tokyo company established in 2012 to take over an AIST-certified venture founded in 2006, commercializing anomaly detection technology from Japan's National Institute of Advanced Industrial Science and Technology. It builds AI for visual inspection in manufacturing, aimed at defects that are hard to define with rules and at factories without AI specialists.

AdaInspector uses HLAC (Higher-order Local Auto-Correlation) features and learns mainly from 100-300 images of good products plus a few defect images. Models are built without code in AdaInspector Cloud, then downloaded to run offline on ordinary CPU-based PCs or existing equipment. The product line also includes Shiwaketter for defect image classification, POKAMIRU for visual inspection support in high-mix production and Q-Smart for quality data analysis.

• Good-sample learning: builds models from about 100-300 good-product images and a few defect images.
• No-code cloud training: AdaInspector Cloud includes automatic parameter search and threshold tuning.
• CPU-only inference: runs offline on general-purpose PCs or existing inspection equipment without GPUs.
• Shiwaketter: classifies defect images for grading and reduces over-detection by existing detectors.
• POKAMIRU: assists visual checks of missing, misoriented or mixed parts in high-mix, low-volume lines.
• Q-Smart: analyses quality data to find root causes and improve yield.

A coil or motor component plant could inspect winding defects that rule-based vision misses. A nonwoven or filter producer could detect foreign matter in patterned material. A resin molding supplier to Japanese electronics brands could flag silver streaks and dents on existing inspection PCs.

Models need mainly good-product images, which suits lines where defects are rare and hard to collect. No GPU is required, and a 5-megapixel image takes about 200-300 ms on a Core i7 PC, so the AI can be added to existing rule-based systems. Software-only projects can be delivered within two weeks, and initial verification is free.

Many Japanese-invested factories in Vietnam produce coils, molded resin, filters and precision parts where defects are rare and varied, making defect-image collection for deep learning difficult. Lightweight AI that learns from good samples and runs on existing PCs lowers the barrier for small and mid-sized suppliers adopting automated inspection.

Adacotech Incorporated is headquartered in Nihonbashi, Tokyo, and led by CEO Kei Nagai. It raised a 400 million JPY Series A led by University of Tokyo Edge Capital Partners and DNX, and a 1.5 billion JPY Series B led by Real Tech Fund and Spiral Capital. Customers include Olympus, Denka, Nitto and Toyota Boshoku.