Industrial Vision Inspection System (Concept)
Computer vision prototype for automated defect detection in manufacturing environments.
Context / Problem
Manual visual inspection in manufacturing is slow and inconsistent. The concept explores how a camera-based inspection system could detect surface defects in real time.
Approach
Designed a dataset strategy using labeled defect images, trained a baseline CNN model, and evaluated performance with precision/recall metrics. Emphasis was placed on dataset balance and realistic augmentation strategies.
Deployment Considerations
Focused on inference latency and integration feasibility. The system was structured for exportable models (ONNX-compatible) and edge deployment on industrial hardware.
Outcome
Demonstrated feasibility of automated defect classification under controlled conditions. Identified key bottlenecks: lighting variation, dataset quality, and edge compute constraints.