MLA-C01 Practice Question 85: Semantic Segmentation Pixels
AWS Certified Machine Learning Engineer – Associate (MLA-C01) · Test 9 · question 5 of 10
An agriculture company flies drones over fields and wants to estimate what percentage of each image is covered by weeds, crops, and bare soil, so it can plan targeted herbicide spraying. Weeds grow in irregular patches that blend into the crop rows. The company has labeled training images in which every pixel is marked as weed, crop, or soil, and it wants to use a SageMaker AI built-in algorithm.
Which algorithm fits this task?
Show answer and explanation
Answer:
The Semantic Segmentation algorithm tags every pixel in an image with a class label from a predefined set of classes. A per-pixel mask is exactly what an area estimate needs: count the pixels in each class and divide by the total. It also traces irregular patches that follow no neat shape, and the company's per-pixel labels are the training data this algorithm expects (option D).
- A is wrong because image classification analyzes only whole images and assigns each image a category. It cannot say how much of the image is weed. - B is wrong because bounding boxes are rectangles around objects. Irregular weed patches mixed into crop rows would overlap and overcount, so boxes cannot give an accurate area. - C is wrong because k-means is unsupervised and ignores the labels. Weeds and crops are often similar shades of green, so color clusters would not line up reliably with the three classes.
Reference: Semantic Segmentation Algorithm