Microsoft Azure AI Fundamentals (AI-900)Describe features of computer vision workloads on AzureHard
A smart factory is implementing a system to detect anomalies in product packaging. The system needs to identify if a package is damaged, torn, or improperly sealed. These defects can appear anywhere on the package and vary in shape and size. Which computer vision capability would be most effective for this task?
- AOptical Character Recognition (OCR)
- BSemantic Segmentation
- CFace Detection
- DImage Classification
Show answer & explanationAnswer & explanation
Correct answer: B. Semantic Segmentation
Semantic Segmentation is most effective. It can precisely delineate the damaged areas at a pixel level, allowing for detailed analysis of the defect's shape, size, and location, crucial for quality control where varying and subtle defects need to be identified.
Why the other options are wrong
- A. OCR extracts text and is not applicable for detecting physical damage.
- C. Face Detection is specialized for human faces and irrelevant to packaging defects.
- D. Image Classification would only label the entire package as 'damaged' or 'undamaged', not specify the location or nature of the defect.
Semantic Segmentation for Anomaly Detection
Using semantic segmentation to precisely identify and delineate anomalous regions within an image.
- Allows for granular analysis of defect size, shape, and location.
- Useful when defects are irregular or subtle.
- Provides more detail than bounding boxes for quality control.
Memory trick: Segment the pixels to find defect anomalies.