Search

CN-121994660-A - Spherical particle size measurement method and device based on deep learning numerical prediction

CN121994660ACN 121994660 ACN121994660 ACN 121994660ACN-121994660-A

Abstract

The invention discloses a spherical particle size measurement method and device based on deep learning numerical prediction, and belongs to the field of particle measurement and image processing. The method comprises the steps of obtaining image data of spherical particle interference fringes with different sizes, labeling corresponding particle size labels for each image, dividing the image data into a training set, a verification set and a test set, training a pre-built neural network model by using the training set, monitoring a training process by the verification set, storing optimal model weights, testing the prediction accuracy of the model by the test set, setting the pre-built neural network model on the basis of an original UNet++ network, comprising a plurality of residual blocks and a spatial attention module, adding a fully-connected output head, realizing end-to-end mapping from the image to a size value, and finally inputting the interference fringe image of particles to be tested into the trained network, and directly outputting a particle size predicted value. The method realizes high-precision, real-time and end-to-end prediction of the spherical particle size, and is suitable for the scenes such as online monitoring of a cloud particle field.

Inventors

  • LUO SHIJIE
  • SHI XIAOXIA
  • LI YITONG
  • Sun Lincheng
  • HAN SHUZHEN
  • SUN JINLU

Assignees

  • 天津工业大学

Dates

Publication Date
20260508
Application Date
20260126

Claims (9)

  1. 1. The spherical particle size measurement method based on deep learning numerical prediction is characterized by comprising the following steps of: the method comprises the steps of preparing a data set, namely acquiring image data of spherical particle interference fringes with different sizes, labeling corresponding particle size labels for each image, and constructing a data set with labels; Dividing the data set into a training set, a verification set and a test set; The method comprises a network training step, a network training step and a network training step, wherein the training step is to train a pre-constructed neural network model by using the training set, monitor the training process through the verification set, store the optimal model weight and test the prediction precision of the model through the test set; And the step of size prediction, namely inputting interference fringe images of the particles to be detected into a trained network, and directly outputting a particle size predicted value.
  2. 2. The method according to claim 1, wherein the pre-built neural network model specifically comprises: 7 residual blocks are connected in series after the double-layer convolution structure of the encoding end and the decoding end; a spatial attention module is embedded in each residual block, and then the spatial attention weight is generated through 1 multiplied by 1 convolution and Sigmoid activation; and a fully-connected regression head is newly added at the network output end, and the high-dimensional characteristics are mapped into a single scalar through a fully-connected layer, so that the end-to-end size prediction from pixel-level representation to image-level regression indexes is realized.
  3. 3. The method of claim 1, wherein in the data set preparing step, the spherical particle interference fringe image data includes experimental acquisition data and simulation generation data; The experimental data are collected through an interference particle imaging system, the system comprises a laser, a spatial filter, a collimating lens, a cylindrical lens group, a sample cell, an imaging lens and a CCD camera, the diluted polystyrene spherical particles with different sizes are placed in the sample cell as experimental particles, and a spherical particle interference fringe pattern is obtained through sheet laser beam irradiation in the CCD camera.
  4. 4. A method according to claim 3, wherein the CCD camera resolution is 2448 x 2048 and the pixel size is 3.45 μm.
  5. 5. A method according to claim 3, wherein the experimental particles are polystyrene spherical particles of 30 μm, 45 μm, 60 μm, 90 μm, diluted and placed in a sample cell containing deionized water.
  6. 6. The method of claim 1, wherein the training process in the network training step includes selecting Adam by an optimizer, setting an initial learning rate to be 1×10 -4 , setting a weight attenuation coefficient to be 1×10 -8 , adopting a learning rate adjustment strategy, attenuating the learning rate according to a ratio of 0.1 for every 10 training rounds, and dynamically storing model parameters with minimum current loss according to a verification set loss in the training process, wherein the training round number is 150 epochs, thereby obtaining optimal network weights.
  7. 7. A spherical particle size measurement apparatus based on deep learning numerical prediction, characterized in that it is used for realizing the spherical particle size measurement method based on deep learning numerical prediction as set forth in any one of claims 1 to 6, and comprises: the data set preparation module is used for acquiring image data of spherical particle interference fringes with different sizes, labeling corresponding particle size labels for each image and constructing a data set with labels; the data set dividing module divides the data set into a training set, a verification set and a test set; The network construction and training module is used for training a pre-constructed neural network model, monitoring a training process through a verification set, storing optimal model weights and testing the prediction accuracy of the model through a test set, wherein the pre-constructed neural network model is based on an original UNet++ network and comprises a plurality of residual blocks and a spatial attention module, and a full-connection output head is added to realize end-to-end mapping from an image to a size value; and the size prediction module inputs the interference fringe image of the particle to be detected into a trained network and directly outputs a particle size predicted value.
  8. 8. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, wherein the processor implements the method for measuring spherical particle size based on deep learning numerical predictions as claimed in any one of claims 1 to 6 when executing the program.
  9. 9. A computer-readable storage medium, on which a computer program is stored, characterized in that the program, when executed by a processor, implements the spherical particle size measurement method based on deep learning numerical prediction as claimed in any one of claims 1 to 6.

Description

Spherical particle size measurement method and device based on deep learning numerical prediction Technical Field The invention relates to the technical field of particle measurement and image processing of optical imaging, in particular to a method and a device for measuring the size of spherical particles based on deep learning numerical prediction. Background The warm cloud consists of spherical water drop particles, and measuring the size of the cloud particles is not only beneficial to improving the accuracy of weather forecast and reducing the risk of weather disasters, but also has important significance in revealing the generation and evolution mechanism of the cloud, optimizing the artificial precipitation and other processes. Interference particle imaging (Interferometric PARTICLE IMAGING, IPI) is a measurement method based on particle scattering light field distribution, and the method utilizes Mie scattering theory to study the relationship between particle size and scattering light intensity distribution, and acquires particle scattering light information through an optical imaging system, so as to realize particle size measurement. In the research of using the interference particle imaging technology to realize particle size measurement, the existing patent document CN116559033a extracts the fringe frequency in the interference fringe pattern through a two-dimensional fourier transform algorithm to obtain the particle size, but cannot acquire cloud particle field information in real time. Currently, the unet++ network is applied to the field of spherical particle size measurement, and the measurement method of the prior patent CN119935850a realizes spherical particle size measurement, but the reconstructed image needs to be converted into a size value by an image reconstruction method, so that the process is complex. Disclosure of Invention The present invention has been made in view of the above problems, and has as its object to provide a spherical particle size measurement method and apparatus based on deep learning numerical prediction that overcomes or at least partially solves the above problems. In order to achieve the above purpose, the present invention adopts the following technical scheme: in a first aspect, an embodiment of the present invention provides a method for measuring a size of spherical particles based on deep learning numerical prediction, including the steps of: the method comprises the steps of preparing a data set, namely acquiring image data of spherical particle interference fringes with different sizes, labeling corresponding particle size labels for each image, and constructing a data set with labels; Dividing the data set into a training set, a verification set and a test set; The method comprises a network training step, a network training step and a network training step, wherein the training step is to train a pre-constructed neural network model by using the training set, monitor the training process through the verification set, store the optimal model weight and test the prediction precision of the model through the test set; And the step of size prediction, namely inputting interference fringe images of the particles to be detected into a trained network, and directly outputting a particle size predicted value. In one embodiment, the pre-built neural network model specifically includes: 7 residual blocks are connected in series after the double-layer convolution structure of the encoding end and the decoding end; a spatial attention module is embedded in each residual block, and then the spatial attention weight is generated through 1 multiplied by 1 convolution and Sigmoid activation; and a fully-connected regression head is newly added at the network output end, and the high-dimensional characteristics are mapped into a single scalar through a fully-connected layer, so that the end-to-end size prediction from pixel-level representation to image-level regression indexes is realized. In one embodiment, in the data set preparing step, the spherical particle interference fringe image data includes experimental acquisition data and simulation generation data; The experimental data are collected through an interference particle imaging system, the system comprises a laser, a spatial filter, a collimating lens, a cylindrical lens group, a sample cell, an imaging lens and a CCD camera, the diluted polystyrene spherical particles with different sizes are placed in the sample cell as experimental particles, and a spherical particle interference fringe pattern is obtained through sheet laser beam irradiation in the CCD camera. In one embodiment, the CCD camera resolution is 2448×2048 and the pixel size is 3.45 μm. In one example, the experimental particles were spherical polystyrene particles of 30 μm, 45 μm, 60 μm, 90 μm, which were diluted and placed in a sample cell containing deionized water. In one embodiment, the training process in the network training step