Thank you for taking an interest in this problem.
- I did not display all the images because the file size would be too large. I have attached the images that I did display.
- All the images I currently have have been marked at the inflection points.
However, I will actually analyze images that do not have the marks.
si: List of images for training data
pri: List of images for testing
Predict function
cp1: List of coordinates of the lowest bending points in the training data images
cp2: List of coordinates of the second bending points from the bottom in the training data images
preas1: List of coordinates of the lowest bending points in the test data images
preas2: List of coordinates of the second bending points from the bottom in the test data images
datasetx1: List of training data in which the horizontal coordinates of the lowest bending points are associated with the training data images
datasety1: List of training data in which the vertical coordinates of the lowest bending points are associated with the training data images
datasetx2: List of training data in which the horizontal coordinates of the second bending points from the bottom are associated with the training data images
datasety2: List of training data in which the vertical coordinates of the second bending points from the bottom are associated with the training data images
corpre1: List of coordinates of the lowest bending points estimated from the test data images
corpre2: List of coordinates of the second bending points estimated from the test data images
NetChain & NetTrain
cpt: List of the bottom 6 coordinates of the bending points of the training image
Certainly, I knew I had to binarize the image before using ImageCorners, and I tried binarizing with various parameters, but unfortunately our images could not be binarized conveniently.
Some images have wide lines in the binarized image, and ImageCorners gives the positions of both the inner and outer corners.
Attachments:
