Simple Algorithm for detecting defective ball bearing images using this real life industrial dataset
1- reading the image files from the dataset (ok or def)
2- turning the images into black and white
3 - vectorizing each image using the Histogram of Oriented Gradients(HOG)
4- using K-Means clustering on the dataset
5- calculating the mean vector for each cluster and saving it
6- doing the steps 1-5 for the other dataset
7- comparing the vectorized version of the test images to each of the meaned vectors calculating their cosine similarity.
8- choosing the mean vector to which the test image vector is most similar to.
| metric | value |
|---|---|
| Accuracy | 0.78 |
| Precision | 0.83 |
| recall | 0.66 |
| f1-score | 0.732 |