Towards Automated Nanoenergetic Reaction Characterization with Computational Vision.

In order to utilize aluminum-based nanoenergetic materials effectively, the mechanisms by which reactive aluminum fuel escapes a passivating aluminum shell must be better understood. These reactions can be classified based upon pre- and post-reaction imagery taken of the material. To aide in quantitatively understanding these reactions we investigate creating an improved reaction classifier through leveraging larger, state-of-the-art deep learning models. Additionally, we experiment with various pre-training methods and image analogs to enable effective training of large models even with a small dataset. Image features used from the past experimentation may be used in conjunction with reaction imagery to aid in reaction classification.