Particle image processing measurement model training sample image synthesis method and measurement model construction method

Abstract

The invention discloses a particle image processing measurement model training sample image synthesis method. The method comprises the following steps: setting a light source and an image shooting device facing the light of the light source; at least a background plane, a first plane and a second plane are sequentially arranged between the light source and the image shooting equipment; light rays of the light source pass through the background plane, first particles located between the background plane and the first plane are projected to the first plane, and a first image is generated; a second particle located between the first plane and a second plane, the second particle and the first image being projected to the second plane to generate a second image; the second image projection is received by an image shooting device and serves as a training sample image of the particle image processing measurement model.

Description

Technical Field

[0001] The present invention belongs to the technical field of particle measurement, and in particular relates to a method for generating a synthetic image of a particle group with defocus and overlapping imaging, and a method for measuring the particle size and concentration of a particle group based on a convolutional neural network model trained with the synthetic image. Background Art

[0002] In recent years, particle measurement methods based on image processing technology have developed rapidly due to their advantages of visualization, non-contact, rapidity, and global measurement. Image measurement systems generally use backlight imaging technology, which is mainly used to measure the particle size and shape of particles. They are widely used in the measurement of particle flow or two-phase flow, such as droplets sprayed from sprays or bubbles in gas-liquid two-phase flow. Compared with laser diffraction technology and phase Doppler technology, the advantage of image measurement technology is that it can measure non-spherical particles because the size and shape of the measured object are visualized. At the same time, in some occasions, such as when it is necessary to measure the color or appearance of particles, particle luminescence imaging methods such as reflected light illumination, excited fluorescence or phosphorescence imaging, and self-luminescence imaging are also used. Summary of the invention

[0003] In one aspect of the present invention, a particle image processing measurement model training sample image synthesis method is provided.

[0004] Setting a first plane and a second plane;

[0005] Arrange the first particles and the second particles so that the first particles and the second particles are located on both sides of the first plane respectively, and the second particles are located between the first plane and the second plane;

[0006] A first image is generated by projecting the first particle on the first plane, and the first image and the second particle are projected on the second plane to generate a second image;

[0007] The second image projection is received by an image capturing device as a sample image for training the particle image processing measurement model.

[0008] In one aspect of the present invention, a particle image processing measurement model training sample image synthesis method comprises the steps of:

[0009] Setting a light source and an image capturing device facing the light of the light source;

[0010] A background plane, a first plane, and a second plane are sequentially arranged between the light source and the image capturing device;

[0011] The light from the light source passes through the background plane, and projects the first particles between the background plane and the first plane onto the first plane, thereby generating a first image;

[0012] A second particle located between the first plane and the second plane is projected onto the second plane with the first image to generate a second image;

[0013] The second image projection is received by an image capturing device as a sample image for training the particle image processing measurement model.

[0014] In one aspect of the present disclosure, a method for generating a synthetic image of a particle group with defocus and overlapping imaging phenomena is provided. The method is based on the two-dimensional image projection principle of optical imaging and generates a synthetic image of all particles in the field of view of an imaging system from far to near.

[0015] When imaging each particle, both the particle's shielding effect on the background and the particle's luminescence imaging effect (including self-luminescence, excitation light or reflected light) are considered;

[0016] When considering the particle occlusion defocus effect, the background image I is generated in sequence bg , the particle occludes the background image I c , the particle occluded area background defocus image I cd And the particle occlusion effect defocus image I b ;

[0017] Considering the particle self-luminescence or reflected light, the particle luminescence effect defocusing diagram I f Add together to obtain the defocus image I of the particle;

[0018] The defocused image I is used as the background image of the next composite image of particles closer to the lens until the occlusion and luminous effects of all particles in the field of view are fully considered, that is, the composite image of the particle group is obtained.

[0019] The particle group composite image generation method is applicable to backlight imaging, reflected light or self-luminous imaging, and the case where both of the above lighting conditions exist.

[0020] Another aspect of the present disclosure is a convolutional neural network model trained based on the synthetic image to measure the particle size and concentration of the particle group. The model includes training the convolutional neural network model using the synthetic image of the particle group taking into account defocus and overlapping imaging, and predicting the experimental or synthetic defocused particle image, and finally obtaining the particle size distribution and concentration data of the particle group;

[0021] The training model includes two steps: binary classification of particle images (within or outside the measurement depth range) and particle size recognition of particle images within the measurement depth range.

[0022] In response to the defocus and overlap problems of particle images in the optical direct imaging method, the present invention proposes a method for generating a synthetic image of a particle group with defocus and overlap imaging, as well as a method for measuring the particle size and concentration of a particle group based on a convolutional neural network model trained with synthetic images. The method utilizes the principle of optical projection imaging to measure the particle size and shape of particles in the field of multiphase flow testing, thereby solving the existing measurement problems of particle flow or two-phase flow.

[0023] The beneficial effects of the present disclosure are:

[0024] A method for generating synthetic images of particle groups that takes into account defocus and overlapping imaging phenomena is proposed. This method can obtain synthetic images that are very close to real particle group images, provide a large amount of training data with known true values ​​for convolutional neural networks, and provide verification data for other defocus image measurement methods.

[0025] And a particle group size and concentration measurement method based on a synthetic image training convolutional neural network model, which uses a binary classification training strategy to identify particles within the measurement depth range to achieve the purpose of concentration measurement, which is of great significance in the field of online particle measurement. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] The above and other objects, features and advantages of the exemplary embodiments of the present invention will become readily understood by reading the detailed description below with reference to the accompanying drawings. In the accompanying drawings, several embodiments of the present invention are shown in an exemplary and non-limiting manner, in which:

[0027] Figure 1 Schematic diagram of the synthesis principle of backlight imaging overlapping particle defocused images according to one embodiment of the present invention.

[0028] Figure 2 A flow chart of overlapping particle defocused image synthesis according to one embodiment of the present invention.

[0029] Figure 3 An example of a backlit composite image of a spherical particle group of different concentrations according to one embodiment of the present invention.

[0030] Figure 4 Examples of non-backlit composite images of irregular particle groups of different concentrations according to one embodiment of the present invention.

[0031] Figure 5 A convolutional neural network flow chart according to one embodiment of the present invention.

[0032] Figure 6 Schematic diagram of a particle image processing and measurement system according to one embodiment of the present invention.

[0033] 1——Backlight source,

[0034] 2 – initial background image,

[0035] 3 - the first particle,

[0036] 4——First image (single out-of-focus particle composite image),

[0037] 5 - the second particle,

[0038] 6 - the second image (a composite image of two overlapping out-of-focus particle images),

[0039] 7——Camera. DETAILED DESCRIPTION

[0040] When using image processing technology to measure particles, for two-phase flow or online particle measurement, there are inevitably blurred images caused by limited depth of field and overlapping images of two or more particles at high concentrations, which are the two main reasons affecting the measurement accuracy of image methods. In this case, it is difficult to achieve effective segmentation of particle images and accurate identification of particle size and depth position using traditional image processing methods.

[0041] The convolutional neural network model based on data training is one of the effective and advanced ways to deal with the above problems. However, the convolutional neural network model requires a large amount of training data with true values, which is often difficult to obtain in experiments; how to obtain a large amount of reliable training data is still an urgent problem to be solved in the online particle size measurement technology based on the convolutional neural network model. At present, there are no reports on the accurate image measurement of particle group concentration using the convolutional neural network model.

[0042] After research, the present disclosure proposes that one of the ways to solve the above-mentioned problem of lack of training data is to use synthetic images. However, in the existing literature, there is no reliable method for generating synthetic images of defocused and overlapping particle groups, or the generated images are quite different from the images actually taken. The difficulty lies in that, in actual situations, the different depth positions of each particle in the particle group result in different degrees of blurring of each particle image, especially when the particle concentration is high, the particle images are prone to overlap, and the imaging process cannot be obtained by simply adding the blurred images of multiple particles, or the result of simple addition is quite different from the characteristics of the actual imaging picture. This situation was discovered by the present disclosure in the research.

[0043] According to one or more embodiments, a method for generating a synthetic image is proposed for a particle group imaging process in which defocused and overlapping imaging phenomena exist in particle image measurement.

[0044] Taking the process of two particles being out of focus and overlapping as an example, the schematic diagram of the imaging principle is as follows: Figure 1As shown. That is, the light emitted by the backlight source 1 first forms an initial background image 2, and then passes through two particles 3 and 5 at different depths but partially overlapping imaging positions, and is projected onto the camera 7 to obtain an image 6. The particles may or may not reflect or emit light at the same time. The two particles may not be in focus, and defocus blur will occur during imaging. When generating a composite image, the influence of particles farther from the lens is considered first, that is, a composite image 4 of particles farther from the lens is generated first. This composite image is used as the background light intensity for imaging closer particles to take into account the influence of overlap. This method can be used to synthesize images of three or more particles that are out of focus and overlapping. Figure 1 In the figure, the second image 6, the first image 4, and the initial background image 2 are actually on the axis formed by the background light source 1 and the camera 7. For the convenience of explanation, these three images are moved out of the axis position for identification. In addition, two particle composite images are described here. For multiple particles, the same applies. At the same time, for the self-luminous particle group, the backlight source here is not necessary. Even if there is no backlight source, the required training image set can be synthesized according to the above scheme.

[0045] When considering each particle, the particle's occlusion effect on the background and the particle's luminous imaging effect (including self-luminous or reflected light) are considered at the same time; when considering the particle occlusion defocus effect, the background image is generated in sequence (i is the particle number), the background image of the area blocked by the particle Background defocus map of the particle occluded area And the defocus map of the particle occlusion effect Considering the particle self-luminescence or reflected light, the particle luminescence effect defocused diagram Add together to obtain the defocus image I of the particle i ; The defocus image I i It is used as the background image of the next composite image of particles closer to the lens until the occlusion and luminescence effects of all particles in the field of view are fully considered, that is, the composite image of the particle group is obtained 6. A large number of composite images containing particle size and depth position information are input into the convolutional neural network model for training, so that it learns the characteristics of the particles and finally realizes the measurement of particle size and concentration.

[0046] Here, the occlusion effect of particles on the background means that when particles are within the field of view of the imaging system, they block the background light, thus forming shadows or dark areas in the imaging. This effect is conducive to improving the contrast between the background image and the particle image, that is, making the particle image easy to identify, so this lighting and imaging method is often used in the particle size measurement process. In the imaging of particle groups, especially when the concentration is high, image overlap is difficult to avoid. Only by accurately identifying the edge of each particle image can the particle size distribution information be accurately obtained, and the background image and particle space distribution can be accurately reconstructed.

[0047] The luminous effect of particles includes the light emitted by the particles themselves (self-luminescence) or the light reflected from the surface of the particles (reflected light). Particles may emit light due to their material properties or external light sources. For example, when it is necessary to obtain information about the apparent morphology of the particles, such as color, a light source on the same side of the lens is used for illumination instead of backlighting. Whether it is self-luminescence or reflected light, particles will produce additional brightness in the imaging, which needs to be considered in image synthesis or analysis.

[0048] The particle defocus effect refers to the fact that during the imaging process, particles often do not lie exactly on the focal plane of the imaging system, whether they block background light or have a luminous effect. This causes the image to be out of focus and blurred. Therefore, the defocus map refers to the image with different degrees of blur or out-of-focus effects generated during the imaging process due to the different distances of the particles relative to the focal plane of the imaging system. The defocus map reflects the visual effect of particles during the imaging process due to their position deviating from the focal plane.

[0049] Therefore, the following steps can be included here:

[0050] S301, first create a background image without particle occlusion

[0051] S302, considering the occlusion area, for a certain particle, generate a background image that only includes the occlusion area of ​​the particle

[0052] S303, considering the defocus effect, generating a background defocus map of the particle occlusion area

[0053] S304, generating an occlusion effect defocus map, combining the occlusion and defocus effects to generate

[0054] S305, considering the particle luminescence effect, for the self-luminescence or reflected light of the particles, a particle luminescence effect defocus map is generated

[0055] S306, synthesizing the defocus map, adding the shading effect defocus map of the particle and the luminous effect defocus map to obtain the defocus map I of the particle i ;

[0056] S306, synthesizing the image, taking the defocused image of each particle as the background of the next particle synthesized image, repeating the above process until all particles are taken into account, and obtaining a synthesized image of the entire particle group.

[0057] After obtaining a synthetic image sample set of a particle group, a convolutional neural network is trained using the synthetic image sample set containing particle size and depth position information so that the model learns the characteristics of the particles, and the trained model is used to measure the particle size and concentration.

[0058] According to one or more embodiments, a method for generating a particle group synthetic image of defocused and overlapped imaging and a particle group size and concentration measurement method based on a convolutional neural network model trained with the synthetic image is provided. Figure 2 As shown, the specific steps include:

[0059] 1) Given the particle group information and imaging system related parameters for generating a synthetic image, including the particle size distribution of the particle group (particle size in pixels), the volume concentration of the particle group, the depth range of the particle group, the illumination intensity imaging grayscale, the field of view size of the imaging system (in pixels), and the defocus blur operator at different depths of the imaging system.

[0060] 2) Based on the above information and parameters, a three-dimensional spatial distribution of a particle group at a certain moment is randomly generated for subsequent generation of a synthetic image of the particle group;

[0061] 3) First, generate an image representing the background lighting conditions to simulate the situation where there are no particles in the system; generally, it is an image with uniform grayscale, and the grayscale value is selected according to actual needs; if there is no backlight source, the grayscale value is 0; if the backlight source is uneven, an uneven grayscale image can also be generated according to the actual lighting conditions.

[0062] 4) Arrange the particles in the three-dimensional space mentioned in step 2) in order from far to near from the camera, and generate a composite image of the particle group in this order. When processing the first particle, that is, generating a composite image of the particle farthest from the camera, the image generated in step 3) is used as the background image.

[0063] 5) In the background image Cut out the particle projection position area ( Figure 2 The red dotted circle in the middle is the occluded area, forming the background image of the particle occluded area According to the depth position of the particle, the defocus blur operator of the actual imaging system is used (for example, the general imaging system can use the Gaussian blur operator, and the characteristic parameter of the Gaussian blur kernel is σ) to perform convolution to obtain the background defocus map of the particle occluded area. According to the principle of occlusion effect, minus This is the defocus map of the particle occlusion effect At this time, the grayscale value of the self-luminous effect defocused image is 0; for the particle luminous effect of particle self-luminescence or reflected light, the particle luminous effect defocused image is generated by convolving the particle luminous clear image with the defocus blur operator at the corresponding position. this

[0064] When , the gray value of the defocused image due to occlusion effect is 0; and Add together to get the defocus image I of the particle A ;

[0065] 6) The defocus image I in step 5) A As the background image for the next particle synthesis image generation process According to the order mentioned in step 4), step 5) is repeated for all particles in the field of view, and finally a composite image of the particle group with instantaneous three-dimensional spatial distribution is obtained.

[0066] 7) Repeat steps 2)-6) to generate multiple composite images of the particle group under the given parameters in step 1).

[0067] 8) By changing the relevant information or parameters in step 1), a large number of composite images of particle groups with different particle size distributions, concentrations, background image grayscales, particle defocus imaging blur, etc. can be generated; for example, under certain conditions where other parameters are the same, three composite images of particle groups with different volume concentrations (0.001%-0.1%) with backlight or non-backlight and spherical or irregular shapes can be generated. Figure 3 and Figure 4 shown.

[0068] 9) The relevant true value corresponding to each particle image in the particle group defocused image can be obtained through the parameters in steps 1) and 2), including the clear image boundary, particle size, and depth position.

[0069] 10) Construct a convolutional neural network model with two steps, namely, firstly classify the particle images into two categories (within or outside the measurement depth range), and then perform particle size recognition on the particle images within the measurement depth range. Figure 5 shown.

[0070] 11) The image obtained in step 8) and the true value data obtained in step 9) are input into step 10) for convolution

[0071] Training of neural network models;

[0072] 12) Using the convolutional neural network model trained in step 11), the actual image is detected; including identifying particles within the measurement depth range and counting their particle sizes to obtain particle size distribution; counting the particle size and number, and calculating the volume concentration based on the measurement depth range and field size.

[0073] According to one or more embodiments, a particle image processing measurement model training sample image synthesis method, when considering the particle occlusion defocus effect, that is, when backlight is illuminated, generates a background image in sequence (i=A, B, C... is the particle number), the background image of the area blocked by the particle Background defocus map of the particle occluded area And the defocus map of the particle occlusion effect The specific implementation method includes the following steps:

[0074] S101, given the particle group information and imaging system related parameters for generating a synthetic image, including the particle group size distribution (particle size in pixels), the particle group volume concentration, the depth range of the particle group, the illumination light intensity imaging grayscale, the imaging system field of view size (in pixels) and the defocus blur operator at different depths of the imaging system.

[0075] S102, randomly generating a three-dimensional spatial distribution of a particle group at a certain moment according to the above information and parameters, for subsequent generation of a composite image of the particle group.

[0076] S103, first generate an image representing the background lighting conditions to simulate the situation where there are no particles in the system; generally, it is an image with uniform grayscale, and the grayscale value is selected according to actual needs; if the backlight source is uneven, an uneven grayscale image can also be generated according to the actual lighting conditions.

[0077] S104, the particles in the three-dimensional space mentioned in step S102 are arranged in order from far to near from the camera, and a composite image of the particle group is generated in this order. When processing the first particle, i.e., generating a composite image of the particle farthest from the camera, the image generated in step S101 is used as the background image I bg .

[0078] S105, in the background image Cut out the particle projection position area ( Figure 2 The red dotted circle in the middle is the occluded area, forming the background image of the particle occluded area According to the depth position of the particle, the defocus blur operator of the actual imaging system is used (for example, the general imaging system can use the Gaussian blur operator, and the characteristic parameter of the Gaussian blur kernel is σ, which corresponds to the distance between the particle and the focal plane of the imaging system) to perform convolution to obtain the background defocus map of the particle occluded area. According to the principle of occlusion effect, minus This is the defocus map of the particle occlusion effect At this time, the self-luminous effect defocuses the image The gray value is 0, that is, the image Equal to I A ; S106, the defocus image I in step S105 A As the background image for the next particle synthesis image generation process Step S105 is repeated for all particles in the field of view in the order mentioned in step S104, and finally a composite image of the particle group distributed in the instantaneous three-dimensional space is obtained.

[0079] S107, repeating steps S102-S106, and generating a plurality of particle group composite images under the given parameters in step S101.

[0080] S108, by changing the relevant information or parameters in step S101, a large number of spherical or irregular particle group composite images with different particle size distribution, concentration, background image grayscale, particle defocus imaging blur, etc. can be generated; for example, under certain conditions where other parameters are the same, three different volume concentrations (0.001%-0.1%) of spherical particle group backlighting composite images are as follows: Figure 3 shown.

[0081] S109, the relevant true value corresponding to each particle image in the particle group defocused image can be obtained through the parameters in steps S101 and S102, including the clear image boundary, particle size, and depth position.

[0082] According to one or more embodiments, Figure 2 As shown, a method for synthesizing overlapping particle defocused images specifically comprises the following steps:

[0083] S201, obtaining the overall background image of particle A In this figure, the occluded area of ​​particle A is cut out to obtain the background image of the occluded area

[0084] S202, for the background image of the occluded area By σ A Defocus blur, get the background defocus image of the occluded area

[0085] S203, for the overall background image of particle A Defocused image of the background and the occluded area Obtain the defocused image of the occlusion effect by subtracting the image of the backlight occlusion effect

[0086] S204, occlusion effect defocused image Defocused image with self-illumination effect Add together to obtain the defocused image I of particle A A ,

[0087] S205, particle B overall background image Crop the occluded area of ​​particle B to obtain the background image of the occluded area

[0088] S206, for the background image of the occluded area By σ B Defocus blur, get the background defocus image of the occluded area

[0089] S207, particle B overall background image Defocused image of the background and the occluded area Obtain the defocused image of the occlusion effect by subtracting the pictures of the occlusion effect

[0090] S208, occlusion effect defocused image Defocused image with self-illumination effect Add together to obtain the defocused image I with particle A as the background and particle B B .

[0091] In the above method, if only backlight imaging is performed, the grayscale value of the defocused image due to the self-luminous effect is 0; if there is no backlight illumination, the grayscale value of the defocused image due to the occlusion effect is 0; if there is both backlight and luminous effects, the corresponding grayscale image is formed according to the above steps based on the actual situation.

[0092] like Figure 5 As shown, the steps of using a convolutional neural network (CNN) to measure particle size and concentration include the following steps:

[0093] S501, starting the particle image synthesis process to synthesize a particle group image;

[0094] S502, obtaining a training sample set of single particle images, when ( where d p is the particle size) is less than or equal to a certain value (such as 0.35), the image is set to category 1, Images larger than a certain value are set to category 2;

[0095] S503, training using an instance segmentation deep learning model Mask-RCNN to identify and segment particles in the image;

[0096] S504, using the trained Mask-RCNN model to predict the image that has been captured and input, so as to identify and segment particles in the image and identify category 1 or category 2;

[0097] S505, for the particles in the category 2 image, no statistical analysis is performed; for the particle image in category 1, a segmented image of the category 1 image is obtained, the effective depth of field of the composite image is calculated according to the optical constants of the measurement system, and the diameter and concentration values ​​of the particles are further obtained.

[0098] S506, according to the above The value determines a specific measurement depth range, counts the particle sizes and numbers of all particles in the corresponding measurement volume, and obtains the particle size distribution and concentration results of the particle group.

[0099] According to one or more embodiments, a method for synthesizing overlapping particle defocused images, when considering the particle self-luminous effect, that is, when not backlit, the particle defocus nature is consistent with the backlit illumination, and the difference is that the foreground (greater grayscale value) and background (smaller grayscale value) of non-backlit illumination are opposite to the backlit illumination. In order to realize the particle image synthesis function of backlit or non-backlit illumination in the same program, it can be realized by changing the foreground grayscale value and the background grayscale value. The specific implementation method includes the following steps:

[0100] a) Given the particle group information and imaging system related parameters for generating a synthetic image, including the particle group size distribution (particle size in pixels), the particle group volume concentration, the depth range of the particle group, the illumination intensity imaging grayscale, the imaging system field of view size (in pixels) and the defocus blur operator characteristic parameters at different depths of the imaging system.

[0101] b) randomly generating a three-dimensional spatial distribution of a particle group at a certain moment according to the above information and parameters, for subsequent generation of a synthetic image of the particle group;

[0102] c) first generate an image representing background lighting conditions, simulating the situation where there are no particles in the system;

[0103] It is usually an image with uniform grayscale, and the grayscale value is selected according to actual needs;

[0104] d) Arrange the particles in the three-dimensional space mentioned in step b) in order from far to near from the camera, and generate a composite image of the particle group in this order. When processing the first particle, i.e., generating a composite image of the particle farthest from the camera, the image generated in step a is used as the background image I bg .

[0105] e) In the background image Cut out the particle projection position area ( Figure 2 The red dotted circle in the middle is the occluded area, forming the background image of the particle occluded area According to the depth position of the particle, the defocus blur operator of the actual imaging system is used (for example, the general imaging system can use the Gaussian blur operator, and the characteristic parameter of the Gaussian blur kernel is σ) to perform convolution to obtain the background defocus map of the particle occluded area. at this time

[0106] equal Out-of-focus image due to occlusion effects The gray value is 0, that is, the image is equal to I;

[0107] f) using the defocus image I in step e) as the background image for the next particle composite image generation process Repeat step e) for all particles in the field of view in the order mentioned in step d), and finally obtain a composite image of the instantaneous three-dimensional spatial distribution of the particle group.

[0108] g) Repeating steps b)-f) can generate multiple composite images of the particle group under the given parameters in step a).

[0109] h) By changing the relevant information or parameters in step a), a large number of spherical or irregular particle group synthetic images with different particle size distribution, concentration, background image grayscale, particle defocus imaging blur, etc. can be generated; for example, under certain conditions where other parameters are the same, three different volume concentrations (0.001%-0.1%) of irregular particle group backlighting synthetic images are shown in FIG. Figure 4 shown.

[0110] i) The relevant true value corresponding to each particle image in the particle group defocused image can be obtained through the parameters in steps a) and b), including the clear image boundary, particle size, and depth position.

[0111] j) Construct a convolutional neural network model with two steps, namely, firstly classify the particle images into two categories (within or outside the measurement depth range), and then identify the particle size of the particle images within the measurement depth range. Figure 5 shown.

[0112] k) inputting the image obtained in step h) and the true value data obtained in step i) into step j) to train the convolutional neural network model;

[0113] l) Using the convolutional neural network model trained in step k), the actual image is detected; including identifying particles within the measurement depth range and counting their particle sizes to obtain particle size distribution; counting the particle size and number, and calculating the volume concentration according to the measurement depth range and field size.

[0114] The beneficial effects of the present disclosure are that a method for generating a synthetic image of a particle group taking into account the defocus and overlapping imaging phenomena is proposed, which can obtain a synthetic image that is very close to the real particle group image, provide a large amount of training data with known true values ​​for the convolutional neural network, and provide verification data for other defocus image measurement methods; and a particle group size and concentration measurement method based on a synthetic image training convolutional neural network model, which adopts a binary classification training strategy to identify particles within the measurement depth range and achieve the purpose of concentration measurement, which is of great significance in the field of online particle measurement.

[0115] It should be understood that in the embodiments of the present invention, the term "and / or" is only a description of the association relationship of the associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone. In addition, the character " / " in this article generally indicates that the associated objects before and after are in an "or" relationship.

[0116] Those of ordinary skill in the art will appreciate that the units and algorithm steps of each example described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, computer software, or a combination of the two. In order to clearly illustrate the interchangeability of hardware and software, the composition and steps of each example have been generally described in terms of function in the above description. Whether these functions are performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention.

[0117] In the several embodiments provided in the present application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are only schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. In addition, the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interfaces, devices or units, or it can be an electrical, mechanical or other form of connection.

[0118] In addition, each functional unit in each embodiment of the present invention may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0119] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention is essentially or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk and other media that can store program codes.

[0120] The above is only a specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any technician familiar with the technical field can easily think of various equivalent modifications or replacements within the technical scope disclosed by the present invention, and these modifications or replacements should be included in the protection scope of the present invention. Therefore, the protection scope of the present invention shall be based on the protection scope of the claims.

Claims

1. A particle image processing measurement model training sample image synthesis method, characterized in that: Setting a first plane and a second plane; Arrange the first particles and the second particles so that the first particles and the second particles are located on both sides of the first plane respectively, and the second particles are located between the first plane and the second plane; A first image is generated by projecting the first particle on the first plane, and the first image and the second particle are projected on the second plane to generate a second image; The second image projection is received by an image capturing device as a sample image for training the particle image processing measurement model.

2. The method according to claim 1, characterized in that A plurality of planes including a first plane and a second plane are sequentially arranged, and at least one particle is arranged between each plane. According to the method of claim 1, the particle image processing measurement model training sample image is synthesized by analogy.

3. A particle image processing measurement model training sample image synthesis method, characterized in that: The method comprises the steps of: Setting a light source and an image capturing device facing the light of the light source; A background plane, a first plane, and a second plane are sequentially arranged between the light source and the image capturing device; The light from the light source passes through the background plane, and projects the first particles between the background plane and the first plane onto the first plane, thereby generating a first image; A second particle located between the first plane and the second plane is projected onto the second plane with the first image to generate a second image; The second image projection is received by an image capturing device as a sample image for training the particle image processing measurement model.

4. The method according to claim 1, characterized in that: A plurality of planes including a background plane, a first plane, and a second plane are sequentially arranged between the light source and the image capturing device, and at least one particle is arranged between each plane. According to the method of claim 3, the particle image processing measurement model training sample image is synthesized by analogy.

5. A particle image processing measurement model training sample image synthesis method, characterized in that: S301, set the backlight source to create a background image without particle occlusion S302, considering the occlusion area, for a certain particle, generate a background image that only includes the occlusion area of ​​the particle S303, considering the defocus effect, generating a background defocus map of the particle occlusion area S304, generating an occlusion effect defocus map, combining the occlusion and defocus effects to generate S305, considering the particle luminescence effect, for the self-luminescence or reflected light of the particles, a particle luminescence effect defocus map is generated S306, synthesizing the defocus map, adding the shading effect defocus map of the particle and the luminous effect defocus map to obtain the defocus map I of the particle; S306, synthesizing the image, taking the particle's defocused image as the background of the next particle synthesized image, repeating the aforementioned steps until all particles are considered, and obtaining a set of synthesized images of the entire particle group, which is used as a training sample for the particle image processing measurement model.

6. A method for constructing a particle image processing measurement model, characterized in that: Based on the training sample image synthesis method described in claim 2 or 3, a training set is obtained. The convolutional neural network model is trained using the training set to obtain the particle image processing measurement model.

7. A particle image processing measurement method, characterized in that: An image acquisition device is set up to obtain particle flow or two-phase flow or multi-phase flow images in real time, and the particle image processing measurement model as described in claim 6 is input to obtain particle size and concentration measurement data.

8. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein: The processor runs the computer program to implement the method according to any one of claims 2 to 7.

9. A storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 2 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that The computer program is executed by a processor to implement the method according to any one of claims 2 to 7.

Images