Anomaly and Defect Detection
R2026bSince R2026b
Train deep learning networks using normal images to establish an expected appearance, then identify images or regions that deviate from this learned representation. Visual Inspection Toolbox™ supports these anomaly detection capabilities:
Training — Train with normal data alone, or optionally incorporate anomalous examples when available.
Scoring and localization — Generate anomaly scores for image-level classification and pixel-level spatial anomaly maps that highlight defect regions.
Threshold optimization — Determine classification thresholds using receiver operating characteristic (ROC) curves or precision-recall metrics.
Evaluation — Assess detector performance with confusion matrices and accuracy, precision, recall, and F1 scores.
Visualization — Overlay spatial anomaly maps on images as heat maps and interactively review results. Normalize anomaly maps using percentile statistics of maps from normal examples.
To interactively detect and visualize anomalies, as well as train and evaluate anomaly detection networks, use the Visual Anomaly Detector app.
To get started with anomaly detection, see Get Started with Anomaly Detection Using Deep Learning.
Apps
| Visual Anomaly Detector | Train, evaluate, and compare deep learning anomaly detectors |
Functions
Topics
- Get Started with Anomaly Detection Using Deep Learning
Learn about the basics of anomaly detection.
- Select Detection Approach for Visual Inspection
Choose among anomaly detection, shape matching, object detection, and object counting for your visual inspection application.
- Train and Evaluate Anomaly Detection Model Using Visual Anomaly Detector
This example shows how to train and evaluate an anomaly detection model using the Visual Anomaly Detector app.
- Detect Anomalies Using Visual Anomaly Detector
This example shows how to detect and localize industrial production defects in images using a trained anomaly detector in the Visual Anomaly Detector app.
- Detect PCB Defects in Live Image Stream Using Visual Anomaly Detector
This example shows how to detect and localize defects on printed circuit board (PCB) images in a live image stream using a trained anomaly detector in the Visual Anomaly Detector app.
Featured Examples
Localize Screen Manufacturing Defects Using Student-Teacher Anomaly Detector
Train a Student-Teacher anomaly detector on normal screen images, then localize manufacturing defects by comparing student and teacher feature outputs.
Localize Industrial Defects Using PatchCore Anomaly Detector
Perform localization of anomalous defects in printed circuit boards (PCBs) using anomaly heat maps generated with the PatchCore anomaly detector.
Detect Defects Using Tiled Training of Student-Teacher Anomaly Detector
Detect and localize industrial production defects on chewing gum images by training a Student-Teacher anomaly detection network on tiled images.
Detect Image Anomalies Using Explainable FCDD Network
Use an anomaly detector to distinguish between normal pills and pills with anomalous chips or contamination.
Detect Anomalies in Pills During Live Image Acquisition
Detect anomalies in pills during live image acquisition.
Classify Defects on Wafer Maps Using Deep Learning
Classify eight types of manufacturing defects on wafer maps using a simple convolutional neural network (CNN).
Detect Industrial Defects Using Zero-Shot AnomalyCLIP
Detect and localize industrial production defects in pill images using an AnomalyCLIP anomaly detection network.
Detect Image Anomalies Using PatchCore Anomaly Detector
Detect and localize crack anomalies in concrete images using a PatchCore anomaly detection network.
Identify Defects in Air Compressors Using Spectrogram Images
Detect and localize defects in acoustic recordings of air compressors using Mel spectrogram images and a Student-Teacher anomaly detector.
Detect Irregularities in 3-D Volumetric Lung CT Data Using PatchCore Anomaly Detector
Detect irregularities in 3-D volumetric computed tomography (CT) data of lungs by using a PatchCore anomaly detector that uses a pretrained ResNet-101 model as the backbone.
Detect Anomaly in Dense 3-D Point Cloud Data Using FPFH Feature Embeddings
Detect anomalies in 3-D dense point cloud data using fast point feature histograms (FPFH) feature embeddings extracted for a single class.
