LabVision: An Intelligent Mobile System for Laboratory Equipment Detection and Safety Management Using On-Device Deep Learning — A Developing Country Perspective

Abstract

LabVision is a smart computer-vision-based laboratory equipment management system. The system integrates a fine-tuned MobileNetV2 Single Shot Detector (SSD) TensorFlow Lite model for on-device equipment identification, a 47-rule forward-chaining expert system for context-aware safety validation, and a Supabase cloud backend for real-time session logging and user management, delivered within a React-based mobile application packaged for Android via a Flutter wrapper. A custom dataset of 1,200 annotated images across eight equipment classes was assembled and used for model training. Post-training INT8 quantisation reduced the model size by 72.8% — from 22.4 MB to 6.1 MB — with no measurable accuracy regression. System evaluation over 450 test frames achieved 92.0% detection accuracy ([email protected]), 296 ms mean end-to-end latency, 91.7% expert system rule correctness, and 96.0% inventory accuracy, surpassing all benchmark systems compared. LabVision demonstrates a cost-effective and practically deployable intelligent laboratory management solution for resource-constrained Nigerian academic institutions.