Teaching Reform through Automated Grading for Machine-Vision Defect Inspection Labs

Abstract

This paper proposes an engineering-oriented teaching reform for a vocational AI Machine Vision course, targeting industrial-style print defect inspection (stain and scratch) on a reproducible synthetic defect dataset. Guided by outcome-based education (OBE), we design a project-based task chain covering data specification, synthetic defect generation, preprocessing, classical machine vision baselines (thresholding, morphology, connected components, and line-structure detection), and engineering delivery of a command-line inspection tool with standardized outputs. To support formative assessment at scale and reduce subjective grading, we implement an automated assessment pipeline that verifies input/output (I/O) contract compliance, evaluates functional correctness via unit tests (pytest), measures code quality via static analysis (pylint/flake8), and validates robustness using a perturbation suite and boundary-case tests quantified by performance degradation ΔF1. Demonstration results using representative submissions show that the proposed pipeline provides consistent, transparent grading and promotes engineering deliverables and robust behavior, providing a reusable blueprint for machine vision education aligned with industrial inspection requirements.