Search for: All records

A fuzzy clustering algorithm with the ability to learn unsupervised can be used to detect objects of interest in semi-structured data. An online application of a fuzzy clustering algorithm with merging was implemented in both software and hardware to test anomaly detection in atomic force microscopy (AFM). The requisite components of the algorithm were all estimated, measured, and verified to meet real time constraints for incoming data. After clusters have been formed, representing the background of an image, any new cluster is an abnormality to the surface which is of interest to the user. This real-time detection of anomalies is important for identifying regions of interest for faster and higher resolution scanning. Results show this application is successfully capable of detecting anomalies in AFM topographic images. The approach taken in this paper is generic and can be applied to other applications with a continuous data stream.