B. Rajabifar
3M,
United States
Keywords: AFM, Deep learning, CBIR
Summary:
Atomic force microscopy (AFM) is widely used to characterize industrial polymeric samples at high resolution. Acquired images provide various types of information, such as surface topography, morphology, and mechanical properties. The amount of data in each AFM image and the rate at which images are generated limit the applicability of traditional data processing approaches. Therefore, introducing tools that are better suited to process “big datasets” is essential to the efficient processing of large sets of historical AFM data. A content-based image retrieval (CBIR) tool was developed that can look for images with topographical similarity to a height map of interest across an entire corpus of acquired AFM height maps. The tool helps microscopists to better understand a sample of interest within a broader chronological and similarity-aware context that spans multiple research labs. The designed unsupervised deep learning CBIR model is equipped with a user interface to facilitate selecting the image of interest and the display of search results.