Visual inspection with object detection 'positive' data as well?


I am trying to train a model for object detection. I want it to be able to detect some defects on metallic surfaces.
First I am not sure if it is possible to detect different kinds of defects as the same object called ‘defect’ or if it is better to train the model with multiple objects for the different defects like ‘crazing’, ‘inclusion’, ‘scratches’, …?
Additionally I am wondering if it is reasonable to mark some ‘positive’ areas as well because the surface is inhomogenous and I do not want the detection model to detect defects where there are none?

Thanks in advance!