Two ways to do that: 1. Separate classes for each type 2. Object detection, then classification. This is commonly used for street sign identification