Intelligent Analysis of Optical Microscopy Images for Microfluidic Synthesis Results
With the active development of microfluidic synthesis technology, automation and intellectualization of the processes associated with conducting experiments both in scientific research and in production is becoming an increasingly urgent task. Frequent problems during the experiments include the sel...
Saved in:
Published in | 2023 XXVI International Conference on Soft Computing and Measurements (SCM) pp. 297 - 300 |
---|---|
Main Authors | , , , , , |
Format | Conference Proceeding |
Language | English |
Published |
IEEE
24.05.2023
|
Subjects | |
Online Access | Get full text |
Cover
Loading…
Summary: | With the active development of microfluidic synthesis technology, automation and intellectualization of the processes associated with conducting experiments both in scientific research and in production is becoming an increasingly urgent task. Frequent problems during the experiments include the selection of parameters, control over the course of the synthesis, and evaluation of the synthesis results. Approaches and methods designed to solve these problems will significantly reduce the number of iterations and reduce the burden on the researcher, as there will be no need for many routine operations and calculations. This fact will also have a positive impact on the efficiency of research and production. One of the important processes subject to intellectualization is the process of continuous screening of morphological and dynamic characteristics of monodisperse droplets. The data obtained during the screening allows not only real-time evaluation of the experiment progress but also the automated selection of synthesis parameters to obtain the better result. This paper proposes an approach for the rapid detection of inclusions in monodisperse droplets based on deep learning methods. A synthetic data generation method is applied to achieve the required detection quality. A procedure for searching and processing the most informative droplets in the image is also proposed. The generated dataset was used to train the object detection algorithm Yolo v7. A number of tests were performed both on simple samples and in real time. |
---|---|
DOI: | 10.1109/SCM58628.2023.10159049 |