Have you ever tried to identify cooling holes using YOLO? I have, and it works reasonably well. But the real challenge comes when trying to ensure that hole number 1 is indeed hole number 1 in any dataset. The problem is that these holes can deform and split into smaller holes before fully blocking. I’ve tried using K-means clustering, but I’m only getting around 20% accuracy. Has anyone else faced this issue?
I’m working with a series of xy matrices, with each hole being a single matrix. I’m eager to hear any advice on how to improve my hole numbering method. Should I try a different clustering algorithm or perhaps use a more advanced neural network approach?
Let me know your thoughts!