The quality inspection of industrial products is becoming increasingly complex as the manufacturing industry advances in its comprehensive intelligent development. Steel surface defects contain a variety of intricate properties, and those brought on by various production lines typically differ from one another. As a result, the generalization performance of the detection algorithms for steel surface defects should be good. Currently, there is no available dataset with geometrical shapes containing steel defects. Therefore, this work introduces the creation of a new dataset, which is different from the NEU dataset in terms of the shape of the surface. We created a dataset of six different classes of geometrical shapes using the NEU dataset with the goal of identifying defects in the surface of steel. This research mainly concentrated on the concave-convex geometric shape for the creation of the dataset and introduced a new method to classify where the defects are located in those samples.