Identification/Classification using BOW and SVM

Question

I have been working on a project which would identify diseases from a leaf. I did search and worked out a few things. However some confusions remains.

I believe following should be the flow (suggestions required)

1. Crop diseased area (Manually) from leafs for building Vocabulary.
2. Use SIFT to get keypoints and descriptors
3. Create Bag of Words Vocabulary and Cluster (K means)
4. Train SVM from descriptors obtained above
5. To Evaluate/Classify Take input image of entire leaf and crop it to extract diseased area using HarCascade
6. Use SIFT to get keypoints and Descriptors and then use SVM to Predict.

Questions are

1. Is above workflow reasonable ? or i am missing something?
2. I am confused about how does SVM learns the Name of object or disease for example where does SVM get Name of object it learned or detected?
3. How does SVM outputs the Name of object it identified ?