CN-121994659-A - Particle classification and size measurement method and device based on interference imaging and improved UNet++

CN121994659ACN 121994659 ACN121994659 ACN 121994659ACN-121994659-A

Abstract

The invention discloses a particle classification and size measurement method and device based on interference imaging and improved UNet++, and belongs to the field of particle measurement and image processing. The method comprises the steps of obtaining a spherical particle interference fringe pattern and an irregular particle interference speckle pattern, constructing a data set containing a mask pattern and a class label, dividing the data set into a training set, a verification set and a test set, training a pre-constructed neural network model by using the training set, introducing a space and channel attention module, a frequency domain dynamic convolution unit and a multi-axis Hadamard convolution attention module based on an original UNet++ network, training the network to output a reconstructed image and the class label at the same time, inputting the particle image to be measured into the trained network, outputting the class of particles and the reconstructed image, and measuring the reconstructed image to obtain size information. The method realizes high-precision, real-time and end-to-end prediction of the particle type and size, and is suitable for mixed scenes of droplet sphericity and irregular ice crystals in the cloud.

Inventors

SUN JINLU
CHU YUNPENG
ZHANG CHENG
LI YUQIANG
QIAN LEI

Assignees

天津工业大学

Dates

Publication Date: 20260508
Application Date: 20260126

Claims (10)

1. A particle classification and size measurement method based on interference imaging and improved unet++, the method is characterized by comprising the following steps of: A data set preparation step of acquiring a spherical particle interference fringe pattern and an irregular particle interference speckle pattern and constructing a data set containing a mask pattern and a class label; 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 model identification 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 a prediction step of inputting the particle image to be detected into a trained network, directly outputting the particle type and the reconstructed image, and measuring the reconstructed image to obtain size information.
2. The method of claim 1, wherein the class labels comprise spherical particles and irregular particles, and the mask map comprises diameters of the spherical particles or widths and heights of the irregular particles.
3. The method of claim 1, wherein in the data set preparing step, both the spherical particle interference fringe pattern and the irregular particle interference speckle pattern are generated by experimental acquisition and simulation; The experimental collection is collected through an interference particle imaging system, the system comprises a laser, a spatial filter, a collimating lens, a cylindrical lens group, a sample carrier, an imaging lens and a CCD camera, the diluted polystyrene spherical particles with different sizes are used as experimental particles and are placed in a sample cell to be used as sample carriers, irregular particles are fixed on the surface of a glass slide to be used as sample carriers, the spherical particles and the irregular particles are irradiated by a sheet laser beam, and a spherical particle interference speckle pattern and an irregular particle interference speckle pattern are obtained by the CCD camera.
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. The method according to claim 3, wherein the experimental particles are polystyrene spherical particles of 30 μm, 45 μm, 60 μm, 90 μm, and the irregular particles have a width of 81.579 μm to 411.353 μm and a height of 96.098 μm to 430.711 μm.
6. The method according to claim 1, wherein the pre-built neural network model specifically comprises: The method comprises the steps that in a unet++ network coding part, a frequency domain dynamic convolution unit and a multi-axis Hadamard convolution attention module which are connected in a residual way are respectively introduced after double-layer convolution in each convolution block; In unet++) each of the networks an encoder is provided which is arranged to encode the data in the data stream, a space and channel attention module is introduced.
7. The method of claim 1, wherein the training process in the network training step comprises the steps of adopting an initial learning rate of 2×10 -5 , training a training round number of 150, an optimizer of Adam, a batch size of 2,3 data loading threads for each training, an image input size of 256×256, a single channel input and an output of single class prediction, normalizing the image in the training process, normalizing mask to [0,1], and setting an early stop strategy to inhibit overfitting.
8. An interference imaging and improved unet++ based particle classification and size measurement device, characterized in that, a base for implementing the method according to any one of claims 1-7 interferometric imaging and improved unet+++ particle classification and sizing methods, the device comprises: the data set preparation module is used for acquiring a spherical particle interference fringe pattern and an irregular particle interference speckle pattern and constructing a data set containing a mask pattern and a class label; The data set dividing module is used for dividing the data set into a training set, a verification set and a test set; The network 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, introduces a space and channel attention module, a frequency domain dynamic convolution unit and a multi-axis Hadamard convolution attention module, and trains the network to output a reconstructed image and a category label at the same time; And the prediction module is used for inputting the particle image to be detected into a trained network, directly outputting the particle type and the reconstructed image, and measuring the reconstructed image to obtain the size information.
9. A computer device comprising a memory, a processor and a computer program stored on the memory and executable on the processor, characterized in that, the processor implementing any one of claims 1 to 7 when executing the program is based on interference imaging and improved UNet +: particle classification and size measurement method of +c.
10. A computer-readable storage medium having a computer program stored thereon, characterized in that, the program when executed by a processor implements the interference imaging and improved unet++ based particle classification and sizing method of any of claims 1-7.

Description

Particle classification and size measurement method and device based on interference imaging and improved UNet++ Technical Field The invention relates to the technical field of particle measurement and image processing, in particular to a particle classification and size measurement method and device based on interference imaging and improved UNet++. Background In the case where droplets are common in the atmosphere and ice crystals coexist, they are part of the earth's water circulation system. The liquid drops and ice crystals in the cloud can be mutually transformed through melting or freezing, and play an important role in the processes of weather forecast, artificial precipitation, aircraft icing and the like. In order to reduce the icing accident of the aircraft and improve the accuracy of weather forecast, researching the morphological characteristics of cloud particles is a key task. The interference particle imaging (Interferometric PARTICLE IMAGING, IPI) technology is a measurement method based on particle scattered light distribution, and the scattered light of particles is received through an optical imaging system so as to obtain size and category information. In the research of particle classification and size measurement by using the interference particle imaging technique, the prior patent document CN106018201a uses mean filtering, edge extraction and fourier frequency analysis to calculate the average and remove the outlier under the multi-template to obtain the spherical particle size, and the patent document CN108593528a combines the image enhancement and the two-dimensional autocorrelation analysis to realize the shape and size measurement of the irregular particle. However, none of these methods can classify and measure the size of both spherical and irregular particles. At present, a convolutional neural network is applied to the field of particle measurement, and patent CN116559033A realizes classification and size measurement of spherical and irregular particles by combining interference defocused images with deep learning. This sort-before-measure procedure increases computational complexity and sort errors may lead to reduced dimensional measurement accuracy. Disclosure of Invention In view of the above-mentioned problems of the prior art, the present invention has been made to provide a method of overcoming the above problems or at least partially solving them particle classification and size measurement method based on interference imaging and improved UNet++, a method for measuring particle size and (3) a device. In order to achieve the above purpose, the present invention adopts the following technical scheme: in a first aspect of the present invention, the embodiment of the invention provides a particle classification and size measurement method based on interference imaging and improved UNet++, which comprises the following steps: A data set preparation step of acquiring a spherical particle interference fringe pattern and an irregular particle interference speckle pattern and constructing a data set containing a mask pattern and a class label; 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 model identification 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 a prediction step of inputting the particle image to be detected into a trained network, directly outputting the particle type and the reconstructed image, and measuring the reconstructed image to obtain size information. In one embodiment, the class labels include spherical particles and irregular particles, and the mask map includes diameters of the spherical particles or widths and heights of the irregular particles. In one embodiment, in the data set preparing step, the spherical particle interference fringe pattern and the irregular particle interference speckle pattern are both experimentally collected and simulated; The experimental collection is collected through an interference particle imaging system, the system comprises a laser, a spatial filter, a collimating lens, a cylindrical lens group, a sample carrier, an imaging lens and a CCD camera, the diluted polystyrene spherical particles with different sizes are used as experimental particles and are placed in a sample cell to be used as sample carriers, irregular particles are fixed on the surface of a glass slide to be used as sample carriers, the spherical particles and the irregular particles are irradiated by a sheet laser beam, and a spherical particle interference speckle pattern and an irregular particle interference speckle pattern are obtained by the CCD camera. In one embodiment, the CCD camera resolution is 2448×2048 and th