Automated Visual Inspection Library for Computer Vision Toolbox

Identify anomalies or defects in images to assist and improve quality assurance processes.

The Computer Vision Toolbox™ Automated Visual Inspection Library offers functions that enable you to train, evaluate, and deploy anomaly detection and object detection networks for use in visual inspection and defect detection systems.

Anomaly Detection

Train state of the art anomaly detectors including PatchCore, FCDD, and FastFlow. All detectors support standalone deployment with MATLAB Coder, GPU Coder, and MATLAB Compiler.

Compute anomaly thresholds automatically using anomalyThreshold. Use criterias including maximum precision/recall, desired false positive and false negative rates for choosing thresholds.

Create visualizations of anomaly map inference data and manage details of transparency and different styles of blending with anomalyMapOverlay.

Validate anomaly detectors using the viewAnomalyDetectionResults UI. The UI aids in the visual exploration of how a trained anomaly detector is performing on a sample set of input images.

Object Detection

The popular real-time, single stage object detector YOLOX is included as a network object and associated trainer. YOLOX is a powerful object detector and works well in low-latency/real-time systems. The version of YOLOX provided in this library supports deployment via MATLAB Coder, GPU Coder and MATLAB Compiler.

For more information, please visit the automated visual inspection documentation page which shows you how to get started with anomaly detection using deep learning.

The documentation also features dedicated examples such as the Pill QC example which uses anomaly detection features in the support package in an end-to-end detection workflow. The Detect Defects on Printed Circuit Boards Using YOLOX Network example similarly demonstrates object detection features in the support package in the context of a visual inspection workflow.