AiVibeImage Classification
Image classification is the computer vision task of assigning one or more labels to an entire image, such as good or defective, or the type of component shown, without locating where in the image the object or defect is.
A classifier, typically a convolutional neural network or vision transformer, outputs a probability for each class. Single-label classification chooses one class, while multi-label classification can assign several. Models are usually created by fine-tuning a network pre-trained on a large dataset such as ImageNet, whose annual recognition challenge drove much of the progress in deep learning for vision.
In manufacturing, classification sorts products, verifies part types, grades surface quality and makes pass or fail decisions when the image is tightly framed around one item. It is also used for recognising document types and sorting waste or recyclables.
Because it gives no location, classification is unsuitable when several objects appear or when the position or size of a defect matters; detection or segmentation is then needed. Class imbalance is common in inspection, where defects are rare, so per-class precision and recall matter more than overall accuracy. Confidence thresholds should be set using validation data.
Key points
- Assigns labels to a whole image
- Usually built by fine-tuning a pre-trained network
- Suited to tightly framed pass or fail inspection decisions
- Use detection or segmentation when location or size matters
Where AiVibe comes in
AiVibe delivers AI and machine learning services, chatbots and virtual assistants with RAG, MCP tools and voice, AI quality management including bias detection and model validation, and the AIMURUGA AI agent, and builds Intel-based edge AI devices using the Intel Distribution of OpenVINO toolkit.