A deep learning approach to the study of colloidal microparticles assembly process in three-dimensional space

We present a deep learning-based approach to visualize and analyze the three-dimensional self-assembly of microparticles in a fluid using laser-plane microscopy, followed by coordinate reconstruction in three dimensions. A YOLOv8 neural network was employed for this purpose. It has been demonstrated that the proposed post-processing technique allows for the detection of characteristic features in the scattering pattern of microparticles with a mean average precision of 0.93, as well as the extraction of their coordinates in three-dimensional space with an accuracy of up to 20% of their diameter. This approach holds promise for controlling the self-assembly processes of microparticles in three dimensions and has the potential for enhancing the development of novel materials and technologies, such as three-dimensional bioprinting and micro- and nanoscale fabrication.