AI-Driven Edge Quality Inspection for Logistics 4.0: Reproducible Code and Experimental Framework
Summary
No summary yet. VIRAT can read this paper and write a plain-language summary of its aim, methods, key findings and limitations.
Summarise with VIRAT Free to try with your Science Catalogs account.Abstract
This repository provides the official implementation of the AI-driven quality inspection framework for Logistics 4.0, as described in the associated manuscript. The framework integrates:- Edge-based deep learning perception using a multi-scale CNN (FPN-UNet)- Uncertainty-aware decision routing for cost-sensitive quality control- Evaluation under industrial anomaly detection settings using the MVTec AD dataset- Operational performance analysis including latency, throughput, and robustness- Automatic generation of figures and tables for reproducibility The implementation supports:- One-class anomaly detection (training on defect-free samples only)- Pixel-level anomaly localization- Image-level anomaly scoring- Cost-sensitive threshold calibration and uncertainty-based routing The code includes:- Training pipeline with synthetic anomaly augmentation- Evaluation metrics (Image AUROC, Pixel AUROC, mAP, IoU, F1-score)- Robustness analysis under noise, blur, and compression- Edge deployment benchmarking (FP32 vs FP16 latency and throughput)- Ablation study configurations Dataset:The experiments follow the official MVTec Anomaly Detection (MVTec AD) benchmark protocol:https://www.mvtec.com/company/research/datasets/mvtec-ad This repository ensures full reproducibility of the experimental results reported in the manuscript and provides a deployable reference implementation for Logistics 4.0 inspection systems.
Fields of Study & Subjects
Industrial Vision Systems and Defect Detection
Anomaly Detection Techniques and Applications
Non-Destructive Testing Techniques
Publication details
- Journal / source
- Zenodo (CERN European Organization for Nuclear Research)
- Publication date
- 30 March 2026
- Language
- EN