A team must automate inspection of drone photos to find corroded areas (multiple areas per photo possible). Corrosion appears in only about 0.1% of photos. Engineers categorize each corroded area as urgent repair, scheduled maintenance, or no action. Which combination of steps should the team take to build this solution? (Choose three.)

An object detection model is necessary because the task requires locating multiple corroded areas within each photo, not just classifying the entire image. Amazon Rekognition label detection is too general and may not be precise enough for specific corrosion types or locations. An XGBoost classifier is suitable for predicting the severity (urgent repair, scheduled maintenance, no action) of the detected corrosion areas, as it's a robust tabular data classifier. K-means clustering is an unsupervised technique for grouping data, not for predicting severity categories. Image augmentation on images containing corrosion is crucial because corrosion is a rare class (0.1% of photos), and augmentation helps increase the number of training examples for this critical class, improving model performance. Augmenting images without corrosion would worsen the class imbalance.