This is my final year university project.
It's a deep learning project used to classify metal defects on shaft surfaces. Images were collected for this project from the NEU database containing six defect types. Below, the result is shown, where the first name is the prediction from the model and the second name is in the bracket true category of respective images. You can download the data set from here.
This dataset contains a single folder with 1800 data. If you want to implement this project the way I did, you first need to separate these datasets into six folders with 300 images of each.
For this model, I split the dataset into three parts (train, test, valid) with proposition of 60/20/20.
Then you are done just run the Notebook into your data directory you will get the same results. I hope you can also improve the model by integrating it with any transfer learning model.