Momahmoses/ng-counterfeit-drug-detection

||Multi-modal counterfeit drug detection for Nigeria using NIR spectroscopy, pill vision, PaddleOCR packaging verification and NAFDAC registry lookup for offline Android deployment. # Counterfeit Drug Detection, AI for Nigerian Medicine Markets > Multi-modal ML system combining NIR spectroscopy, pill image analysis, and packaging OCR to identify counterfeit and substandard drugs in Nigerian markets, deployable as an offline Android app with a clip-on NIR spectrometer. Targeting the 42% of Nigerian drugs estimated to be falsified or substandard (WHO). --- ## The Problem The WHO estimates **42% of antimalarials and antibiotics** in Nigerian markets are substandard or falsified. A community health worker has no way to verify drug authenticity in the field without a lab. Counterfeit drugs kill directly through treatment failure and contribute to antimicrobial resistance. The entire Nigerian supply chain, from port to patent medicine vendor, lacks a real-time authentication layer. --- ## Solution: Three-Modality Detection Pipeline | Modality | Technology | What It Detects | |---|---|---| | **NIR Spectroscopy** | PLS-DA + CNN-1D | Active pharmaceutical ingredient (API) presence and concentration | | **Pill Vision** | EfficientNet-B0 | Shape irregularity, colour deviation, imprint mismatch, surface defects | | **Packaging OCR** | PaddleOCR + NAFDAC API | Fake NAFDAC numbers, misspellings, format violations, expired batch logic | **Fusion layer**: weighted ensemble → `AUTHENTIC / SUSPICIOUS / LIKELY COUNTERFEIT` --- ## Priority Drug Coverage - Artemether-Lumefantrine (ACT), most counterfeited antimalarial - Amoxicillin, most counterfeited antibiotic - Oxytocin, supply chain integrity critical for maternal survival - ARV medications, HIV treatment efficacy - Metformin / Glibenclamide, diabetes management --- ## System Architecture ``` [Field worker scans drug] ↓ [MODALITY 1: Clip-on NIR spectrometer (SCiO/Tellspec)] → SNV preprocessing → PLS-DA + CNN-1D → API present? Correct concentration? [MODALITY 2: Camera photo of pill] → EfficientNet-B0 multi-label visual classifier → Shape / colour / imprint / surface texture score [MODALITY …