AOI-Based IC Counterfeit Detection System

• 

  home page

• 

  home page

🚀 Inspiration

Counterfeit ICs have become one of the biggest threats to the electronics supply chain. Industries such as defense, aerospace, and consumer electronics face huge reliability and financial risks due to fake chips entering production. Traditional manual inspection is slow, inconsistent, and cannot reliably detect subtle font changes, texture mismatches, or counterfeit markings. We wanted to build a fast, accurate, and automated solution that industries like BEL can use to verify IC authenticity in real time.

🧠 What It Does

Our system performs Automated Optical Inspection (AOI) of Integrated Circuits using a hybrid combination of Computer Vision + AI + OCR to detect whether an IC is genuine or fake.

It has two modes:

1️⃣ Training Mode

A genuine IC is scanned, and the system extracts key features such as:

Logo patterns IC text and serial numbers (OCR) QR / DataMatrix code Surface texture (LBP, GLCM) Geometry, angle, edges Color profile (CIEDE2000 ΔE)

All extracted features are stored as the “golden reference”.

2️⃣ Testing Mode

A new IC undergoes the same preprocessing. The system compares it with the reference using threshold values:

Template Match ≥ 0.8

OCR accuracy ≥ 85%

ΔE < 5 (color difference)

SSIM ≥ 0.8 (texture & defects)

The outcome: ✔ Genuine IC ✖ Fake IC

The system processes each IC in <0.2 seconds, achieving 95%+ accuracy and supporting 12,000–17,000 ICs/hour, making it suitable for industrial real-time inspection.

🛠️ How We Built It

We followed a modular pipeline:

🔍 Image Acquisition

Used industrial camera setup (5–8 MP) inside a controlled lighting environment to avoid shadows and glare.

🧹 Preprocessing Layers

Noise removal, thresholding Canny edge detection ROI extraction Hough Transform for angle detection LAB & HSV color conversion

🧠 Hybrid AI Model

Our AI agent uses a BLIP-based Vision-Language Model, hosted on Hugging Face Spaces. It helps in enhanced reasoning, textual matching, and final decision support.

📘 Feature Extraction

Using:

OpenCV for template matching Tesseract OCR for text recognition Pyzbar for QR/DMC codes Scikit-Image for texture analysis NumPy for vector comparisons

📊 Verification Engine

All features are compared with the golden IC vector. If a mismatch exceeds tolerance → chip is flagged as counterfeit.

🧩 Challenges We Ran Into

Handling varying fonts, colors, and surface textures across different manufacturers Removing noise caused by reflections on IC surfaces Achieving consistent OCR accuracy on small engraved text Creating a balanced dataset of genuine vs counterfeit IC markings Integrating a Vision-Language Model into the verification workflow

🎓 What We Learned

Advanced OCR techniques significantly improve accuracy on industrial IC text Lighting plays a crucial role in computer vision inspection Hybrid models (CV + AI agent) outperform standalone approaches Real-time industrial AOI requires optimized preprocessing pipelines Deployment on Hugging Face Spaces with fast inference is highly effective

🌍 Impact Economic Prevents losses from counterfeit ICs (estimated $75B globally per year). Environmental Reduces scrap, rework, and e-waste by detecting fake ICs early. Social Increases trust and reliability in supply chains, especially in defense and manufacturing sectors.

🏁 Conclusion

Our AOI-based IC Marking Detection system offers a robust, fast, and highly accurate solution to detect counterfeit chips using Computer Vision, OCR, and AI-driven reasoning. It is scalable, industry-ready, and capable of modernizing IC verification workflows.

Built With

• blip
• computer
• face
• gradio
• hugging
• image
• learning
• machine
• model
• numpy
• ocr
• opencv
• python
• pyzbar
• scikit-image
• spaces
• tesseract
• vision
• vision-language