Challenge

During different steps of an industrial production process  several machines touch flat steel by the time it’s ready to ship. As a result this can damage the quality in various ways. The task here is to localize and classify surface defects on a steel sheet by using images from high frequency cameras.

Solution

We cast the problem as the task of semantic segmentation and trained a DNN for detecting different types of predefined defects. The resulting network provides very accurate results with respect to the localisation and identification of defects.