Holographic projection with deep learning for microparticles detection from water samples

Abstract

This thesis proposes a complete holographic system to be applied in scientific research and monitoring, which is able to detect microparticles from the holographic projection of water samples, using a deep learning approach. The system proposed in this thesis uses digital holography techniques to acquire holograms from these particles (a device was built for this purpose), reconstruct them numerically by obtaining phase and intensity information, and classify them using machine learning models. In addition, we have developed an application on the web capable of performing all stages of the hologram reconstruction and the classification process using trained models, which is also available. The need for studies on particles that are invisible to the naked eye and that can be dangerous to the health of living beings is an increasingly important research topic and there are many concerns about it. An example is the various types of microplastics found on a large scale in different parts of the planet, even within the human body. Another particle that can help identify microplastics and that can be used to calculate bioindicators of water quality are diatoms. The detection of microplastics and diatoms is subject to difficult studies due to their size, in the order of the micrometer.