Challenge

Automatic detection and localisation of anomalies in nano-fibrous materials help to reduce the cost of the production process and the time of the post-production visual inspection process. The basis for a corresponding task provides Scanning Electron Microscope (SEM) imaging of the material parts.

Solution

We designed a solution based on a deep variational Auto-Encoder which can be trained in a purely unsupervised manner and does not require labelled data. By computing the reconstruction error from a fixed window around each pixel we can precisely identify the location of the anomal parts. The images on the right illustrate a few examples. The detected anomalies are marked green.

Impact

The resulting solution automatically detects anomalies in the present material and has an advantage of relying on an unsupervised method which does not require labelled data for training.