En
期刊文章
文章精选
当期目录
过刊浏览
高被引
高下载
特邀文章
专刊合集
待刊文章
Volume 111
您当前的位置：首页 > 期刊文章 > 过刊浏览 > Volumes 108-119 (2025) > Volume 111
Multi-parameter extraction of droplets with inclusions based on enhanced MobileViT and rainbow scattering
Liwei Yan, Can Li *, Linbin Huang, Yang Kang, Xiaolong Huang, Xudong Fan, Ning Li
National Key Laboratory of Transient Physics, Nanjing University of Science and Technology, Nanjing, 210094, China
10.1016/j.partic.2026.02.004
Volume 111, April 2026, Pages 175-185
Received 20 November 2025, Revised 26 January 2026, Accepted 3 February 2026, Available online 11 February 2026, Version of Record 27 February 2026.
E-mail: lican@njust.edu.cn
Highlights

• Droplet multi-parameters extraction algorithm based on improved MobileViT model.

• Acquire dataset of rainbow images of droplets with inclusions through experiments.

• Fast and accurate extraction of droplet size and inclusion volume concentration.

• Algorithm shows good robustness under Gaussian and salt pepper noise environment.


Abstract

In this study, a multi-parameter extraction method of droplets with inclusions is proposed based on an enhanced MobileViT model and the rainbow scattering technique, which allows for the fast and accurate extraction of droplet size and inclusion volume concentration from extinct rainbow patterns. The rainbow scattering optical measurement system equipped with a monodisperse droplet generation system are then demonstrated, along with typical measurement signals of droplets with inclusions. Under different host droplet sizes (120–140 μm) and inclusion concentrations (0%∼0.3%), the results are analyzed and compared with previous extinction rainbow technique. The average relative error of droplet size is ±0.2%, and the maximum absolute error of the inclusion concentration not exceed 0.003%. Anti-noise tests are then performed with the peak signal-to-noise ratio, ranging from −5 dB to 15 dB. The accuracy of the method in extracting droplet size in Gaussian noise and Pepper noise below 5 dB is comparable to its noiseless performance, and the accuracy of the concentration extraction in certain conditions is superior to that of the extinction rainbow technique in no noise case. The proposed method, based on the enhanced MobileViT, takes only 5 ms to process a single image, enabling rapid dynamic monitoring of suspension sprays.

Graphical abstract
Keywords
Deep learning; Rainbow scattering; Droplets with inclusions; Size; Inclusion concentration; Measurement
文章导读