I’m trying to develop a system for the visual inspection of stainless steel parts with machine learning. I want the system to detect defects on the part’s surface, such as micro welding pores or scratches.
Is there a pre-trained model (maybe a CNN) which could be adapted to the task with transfer learning? Or one that is particularly suitable for the classification of or object detection in this kind of images?
Thanks in advance.
This blog post might be helpful for you: https://blog.floydhub.com/localize-and-detect-corrosion-with-tensorflow-object-detection-api/.
@sayakpaul Thank you for the first answer!
So they are using the VGG-16 model for their work. I just thought this model is espacially optimised to be used for classifying everyday objects and might not be applicable to the kind of images I need to classify. But I guess that’s wrong, as I found some more articles about using VGG-16 for surface inspection.
Are there some alternatives which might be even more suitable? Or is this the best way to go right now?
That is exactly you would fine-tune the network on your given dataset.
Okay, I’ll do that, thanks!