Advances in imaging for the design and investigation of materials have been remarkable: the growth of x-ray brilliance was 18 orders of magnitude in 5 decades, and extremely quick snapshots have enabled description of dynamic systems at the atomic scale. From industry to national laboratories, shape and structural properties of new compounds imaged through x-rays are used to measure the function and resilience of new materials. What drastically changed is the frequency in which this data modality is collected and used as a key scientific record, which is unprecedented. One of the main challenges is how to couple increasing data rate experiments to new Data Science methods in support of more automated analytical tasks for scientific discovery. Recent efforts in machine learning applied to data representation and structural fingerprints have streamlined sample sorting and ranking, including the identification of special materials configurations from million-sized image databases. Methods such as convolutional neural networks have allowed automated characterization of abstract pictures, such as scattering patterns, based on prototypes stipulated by experts, or simulated at leadership computing facilities. Such characterizations or signatures show accelerated image similarity search with real-time feedback in million-size image collections. The ability to survey samples more broadly allied to computational algorithms to compare millions of samples offers unique opportunities for deeper scientific interpretation of experiments, but also impose hurdles such as availability of storage, data transfers, large memory footprint and intensive computation. This talk discuss some strategies and software that tackles detection, segmentation and classification of materials such as carbon fibers, concrete, CMC and more.