Laser scattering pollen particle size spectrum distribution detection method and system based on artificial intelligence
By constructing and training intelligent detection models and analyzing the optical signals of laser scattered pollen, the problem of inaccurate detection of pollen particle size distribution characteristics in the prior art is solved, and high-precision pollen particle size distribution detection is achieved.
Patent Information
- Application Number
- CN202411287961.5
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- Filing Date
- 2024-09-14
- Publication Date
- 2024-11-29
- Estimated Expiration
- 2044-09-14
AI Technical Summary
The existing pollen particle size distribution detection technology cannot ensure accurate and sufficient detection of pollen particle size distribution characteristics while utilizing artificial intelligence technology.
By building the initial intelligent model, training the initial intelligent model to obtain an intelligent detection model. The intelligent detection model analyzes and processes the scattered light signal generated by laser irradiation of the pollen to be measured, and obtains particle size distribution data, thereby obtaining the pollen particle size spectrum distribution.
It realizes that while ensuring the detection accuracy of pollen particle size, it accurately detects the characteristics of pollen particle size distribution, improving the accuracy and robustness of the detection.
Smart Images

Figure CN118794848B_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the technical field of pollen particle size distribution detection, and in particular to a laser scattering pollen particle size spectrum distribution detection method and system based on artificial intelligence. Background Art
[0002] Pollen particle size distribution detection technology is a scientific means to measure the size and distribution of pollen particles through specific methods and equipment, mainly including laser light scattering, microscopy, screening, etc.; this technology is not only used in biological research, but also widely used in environmental monitoring, food production, agricultural breeding and other fields. For example, in agricultural breeding, the pollen particle size detection can predict the pollination potential of crops, thereby guiding breeding practices and increasing crop yields.
[0003] Traditional methods for detecting pollen particle size include screening, microscopy, and electrical induction. Although these traditional methods are simple in principle and structure, most of them are difficult to use for real-time measurement, and the detection results have large errors. The light scattering method has the characteristics of high measurement accuracy, good real-time performance, and easy integration, and has also been widely used. However, calculating the pollen particle size through the light signal scattered by pollen requires special algorithms for solution and calculation, which can be mainly divided into non-independent mode algorithms and independent mode algorithms. However, most of the current algorithms are based on the non-negative and smooth characteristics of particle size distribution. Through non-negative constraints and regularization terms such as the first norm and the second norm, the optimal pollen particle size is obtained. It is an optimal solution, but it cannot fully describe the characteristics of pollen particle size distribution, which may lead to a large difference between the calculated result and the actual pollen particle size distribution. The existing technology also uses image recognition technology to identify pollen particles. For example, in the patent application with application publication number CN112465753A, a pollen particle detection method, device and electronic equipment are disclosed. This technology is mainly used to better distinguish pollen particles from background images, and cannot be used for pollen particle size detection. Therefore, the existing pollen particle size distribution detection technology cannot ensure accurate and sufficient detection of pollen particle size distribution characteristics while using artificial intelligence technology. Summary of the invention
[0004] The present invention aims to solve at least one of the technical problems in the prior art to a certain extent, by constructing and creating an initial intelligent model, training the initial intelligent model, and obtaining an intelligent detection model. The intelligent detection model analyzes and processes the scattered light signal generated by laser irradiation of the pollen to be tested, obtains particle size distribution data, and then obtains the pollen particle size spectrum distribution; so as to solve the problem that the existing pollen particle size distribution detection technology cannot use artificial intelligence technology combined with laser light scattering method to ensure the accuracy of pollen particle size detection while ensuring the accurate and sufficient detection of pollen particle size distribution characteristics.
[0005] To achieve the above objectives, in a first aspect, the present application provides a method for detecting the particle size spectrum distribution of pollen by laser scattering based on artificial intelligence, comprising the following steps:
[0006] Create an initial training data set, and process the initial training data set to obtain a model training data set;
[0007] Create an initial intelligent model, and use the model training data set to train the initial intelligent model to obtain an intelligent detection model;
[0008] A laser generating device is used to generate laser light to illuminate the pollen to be tested, and scattered light signals are collected to obtain light intensity information of the pollen to be tested;
[0009] Based on the light intensity information of the pollen to be tested, the particle size distribution data is obtained by using the intelligent detection model for analysis and processing;
[0010] The pollen particle size spectrum distribution is obtained based on the particle size distribution data.
[0011] Furthermore, creating an initial training data set includes the following sub-steps:
[0012] Common pollen particle size information is obtained based on the database, and computer simulation is performed according to the common pollen particle size information and pollen types. According to the Mie scattering theory, the intensity of the scattered light signal is observed and recorded at different angles in the first angle range with a first angle step, and corresponding marks are made according to the pollen type and the pollen particle size information to obtain N groups of data, which are classified according to the pollen type and marked as the initial training data set. The first angle step is b, and the first angle range is B.
[0013] Furthermore, it is characterized in that the initial training data set is processed to obtain the model training data set, which includes the following sub-steps:
[0014] Data noise processing: The data of each pollen type in the initial training data set are divided into 4 equal parts, and 3 equal parts are randomly selected from the 4 equal parts to add additive white noise with amplitudes of a1, a2, and a3 respectively. After the processing is completed, the noisy training data set is obtained;
[0015] Data normalization: The data of each pollen type in the noise training set are normalized according to the data type, and the data size is scaled to [0, 1]. After the processing is completed, the model training data set is obtained.
[0016] Furthermore, creating the initial intelligent model includes the following sub-steps:
[0017] Based on the Xception model, the convolutional layer and pooling layer in the Xception model structure are set to one-dimensional convolutional layer and one-dimensional pooling layer, and a fully connected layer is added. The softmax function is set as the activation function of the output layer of the Xception model, and the Relu function is set as the activation function of other layers except the output layer. A BN layer is added to each convolutional layer. After completion, the initial intelligent model is obtained.
[0018] Furthermore, the initial intelligent model is trained using the model training data set to obtain the intelligent detection model, which includes the following sub-steps:
[0019] Based on Mie scattering theory, the particle size information output by the initial intelligent model is used to calculate the intensity of the corresponding scattered light, which is marked as the model scattered light intensity. The particle size information output by the initial intelligent model is marked as the model particle size distribution. The difference between the model scattered light intensity and the actual scattered light intensity collected is calculated, which is marked as the scattered light difference. The difference between the model particle size distribution and the actual particle size distribution is calculated, which is marked as the particle size distribution difference. The scattered light difference and the particle size distribution difference are combined to construct a loss function.
[0020] Select the Adam algorithm as the optimization algorithm for the initial intelligent model, set the initial intelligent model training hyperparameters, including the number of training rounds, batch size, and learning rate, set the output layer, and train the intelligent detection model based on the model training data set. After completion, the intelligent detection model is obtained.
[0021] Further, using a laser generating device to generate laser light to irradiate the pollen to be tested includes the following sub-steps:
[0022] Obtain the pollen to be tested, use a dispersant to evenly disperse the pollen to be tested to prepare a test sample, use a laser generator, set the laser wavelength to λ, generate a laser with a wavelength of λ, marked as λ laser, adjust the level of the laser generator, let the λ laser pass vertically through the center of the spatial filter and the beam expander lens, generate a parallel monochromatic light beam, and let the parallel monochromatic light beam irradiate the test sample.
[0023] Furthermore, the scattered light signal is collected to obtain the light intensity information of the pollen to be measured; the following sub-steps are included:
[0024] A parabolic reflector is used to reflect the scattered light generated by the parallel monochromatic light beam irradiating the test sample into parallel light rays, which are marked as first parallel light rays. A first convex lens is used to converge the first parallel light rays at the focus of the convex lens, and a circular aperture is placed at the focus of the convex lens. The light rays diverged after the first parallel light rays pass through the focus of the first convex lens and the circular aperture are marked as first scattered light rays. A second convex lens is used to convert the first scattered light rays into parallel light rays, which are marked as second parallel light rays. The second parallel light rays are projected onto an image sensor to obtain light intensity information of the pollen to be tested.
[0025] Furthermore, based on the light intensity information of the pollen to be measured, the intelligent detection model is used for analysis and processing to obtain the particle size distribution data, which includes the following sub-steps:
[0026] The light intensity information of the pollen to be tested is normalized to obtain normalized light intensity information, which is then input into the intelligent detection model. The intelligent detection model reduces the dimension of the normalized light intensity information through a convolutional layer, and then passes through the Inception structure and the residual module. By using convolution kernels of different sizes and maximum pooling operations, features at different scales in the normalized light intensity information after dimensionality reduction are extracted. Finally, the extracted features at different scales are merged through a fully connected layer, and the particle size distribution data is output through the output layer.
[0027] Further, obtaining the pollen particle size spectrum distribution based on the particle size distribution data includes the following sub-steps:
[0028] Based on the particle size distribution data, the particle size distribution curve is drawn in a rectangular coordinate system with the pollen particle size as the X-axis and the particle size number frequency as the Y-axis to obtain the pollen particle size spectrum.
[0029] In a second aspect, the present application provides a laser scattering pollen particle size spectrum distribution detection system based on artificial intelligence, including a data management module, a model building module, a scattering measurement module, an intelligent processing module and a particle size spectrum module;
[0030] The data management module includes a data construction unit and a data processing unit, the data construction unit is used to create an initial training data set, and the data processing unit is used to process the initial training data set to obtain a model training data set;
[0031] The model building module includes a model building unit and a model training unit. The model building unit is used to create an initial intelligent model. The model training unit uses a model training data set to train the initial intelligent model to obtain an intelligent detection model.
[0032] The scattering measurement module includes a laser generating unit and a signal collecting unit. The laser generating unit generates laser light using a laser generating device to irradiate the pollen to be measured, and the signal collecting unit is used to collect scattered light signals to obtain light intensity information of the pollen to be measured.
[0033] The intelligent processing module includes an intelligent detection model, which uses the intelligent detection model to analyze and process the light intensity information of the pollen to be tested to obtain particle size distribution data;
[0034] The particle size spectrum module obtains the pollen particle size spectrum distribution based on the particle size distribution data.
[0035] The beneficial effects of the present invention are as follows: the present invention first creates an initial training data set, and processes the initial training data set to obtain a model training data set; then creates an initial intelligent model, and uses the model training data set to train the initial intelligent model to obtain an intelligent detection model; then uses a laser generating device to generate a laser to irradiate the pollen to be tested, and collects scattered light signals to obtain light intensity information of the pollen to be tested; then uses an intelligent detection model to analyze and process the light intensity information of the pollen to be tested to obtain particle size distribution data; finally, obtains the pollen particle size spectrum distribution based on the particle size distribution data; uses artificial intelligence technology in combination with laser light scattering method to ensure the accuracy of pollen particle size detection while ensuring the accurate and sufficient detection of pollen particle size distribution characteristics.
[0036] The present invention adds additive white noise when constructing a model training data set, so that the model can learn the characteristics of both noisy data and noise-free data, thereby enhancing the robustness of the model and the accuracy when used in actual scenarios; an intelligent detection model is constructed based on the Xception model, which uses deep separable convolution to replace traditional convolution operations, splits the convolution kernel into a single channel form, and performs independent convolution processing on each channel, which not only reduces the number of parameters, reduces the computational complexity, and improves the detection efficiency and accuracy, but also speeds up the training speed and makes the training process more stable; an angular scattering method is used to obtain scattered light signals, because the angular scattering method relies on Mie scattering theory, is easy to operate, can accurately measure the distribution of pollen particle size, and the measurable particle size range is also wider. BRIEF DESCRIPTION OF THE DRAWINGS
[0037] Figure 1 It is a functional block diagram of the system of the present invention.
[0038] Figure 2 The figure is a flow chart of the steps of the method of the present invention.
[0039] Figure 3 This is a schematic diagram of the Mie scattering principle of the present invention.
[0040] Figure 4 This is a schematic diagram of the light scattering signal collection principle of the present invention. DETAILED DESCRIPTION
[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.
[0042] Example 1, please refer to Figure 1As shown, in the first aspect, the present application provides a laser scattering pollen particle size spectrum distribution detection system based on artificial intelligence, including a data management module, a model building module, a scattering measurement module, an intelligent processing module and a particle size spectrum module;
[0043] The data management module includes a data construction unit and a data processing unit. The data construction unit is used to create an initial training data set, and the data processing unit is used to process the initial training data set to obtain a model training data set.
[0044] The data construction unit is configured with a data construction strategy, which includes: obtaining common pollen particle size information based on a database, the database stores common pollen particle size information, performing computer simulation according to pollen types based on the common pollen particle size information, observing and recording the intensity of scattered light signals at different angles within a first angle range with a first angle step according to Mie scattering theory. When light passes through an inhomogeneous medium, the propagation direction changes and scatters around, which is called light scattering. Pollen particles of different sizes are placed in a uniform medium, and incident light is scattered when it is irradiated on the pollen particles. The intensity of the scattered light generated is closely related to the particle size distribution of the pollen particles. Scientist Mie obtained a strict mathematical solution to the boundary conditions of the Maxwell equations for a single spherical particle of isotropic uniformity in a uniform medium under the irradiation of a beam of parallel light by studying the scattering process of the particles, which is called the Mie scattering theory. Please refer to Figure 3 As shown in the scattering process of a single uniform spherical particle based on Mie scattering theory, the wavelength of the electromagnetic wave incident along the z-axis is λ0, the intensity is I0, and it passes through the uniform spherical particle located at the origin of the coordinate. At point P, the scattered light intensity I of the spherical particle to the incident electromagnetic wave can be obtained. s , and decompose it into scattered light component I parallel to the scattering surface l and the scattered light component I perpendicular to the scattering surface r The expressions of the three scattered light intensities are: Where θ is the scattering angle of the incident light after passing through the particle, λ0 is the wavelength of the incident light, r is the distance from point P to the particle, i1(θ) is the function representing the intensity of the scattered light in the horizontal direction, and i2(θ) is the function representing the intensity of the scattered light in the vertical direction. is the azimuth angle;
[0045] And according to the pollen type and pollen particle size information, corresponding marking is performed to obtain N groups of data, which are classified according to the pollen type and marked as the initial training data set. The first angle step is b, and the first angle range is B. In this embodiment, the first angle step b is 1°, and the first angle range B is [40°, 130°];
[0046] The data processing unit is equipped with a data processing strategy, which includes: data noise processing: the data of each pollen type in the initial training data set are evenly divided into 4 equal parts, and 3 equal parts are randomly selected from the 4 equal parts to add additive white noise with amplitudes of a1, a2 and a3, and the noisy training data set is obtained after the processing is completed; in this embodiment, a1 is 1%, a2 is 3%, and a3 is 5%; the additive white noise is added to the data to simulate the common practice in actual applications, so that the model can learn the characteristics of noisy data in the subsequent model training process, thereby enhancing the robustness of the model;
[0047] Data normalization: The data of each pollen type in the noise training set are normalized according to the data type, and the data size is scaled to [0, 1]. After the processing is completed, the model training data set is obtained; the intensity of the scattered light signal is normalized by summation. The summation normalization formula is as follows: , where I g is the intensity of the scattered light signal after the sum normalization process, I is the intensity of the original scattered light signal, and n represents the number of first angles when collecting the intensity of the scattered light signal. For example, in this embodiment, the first angle step b is 1°, and the first angle range B is [40°, 130°], then n=[(130-40) / 1]+1=91,
[0048] The maximum value normalization processing is used for the pollen particle size information. The maximum value normalization processing formula is as follows; , where X g is the pollen grain size after normalization of the maximum value, X is the original pollen grain size, max(X i ) represents the maximum pollen particle size of the same pollen type;
[0049] In the specific implementation process, the particle size information of pollen refers to the equivalent particle size. The equivalent particle size means that when a certain physical property of a particle is the same or similar to that of a homogeneous spherical particle, we use the diameter of the spherical particle to represent the diameter of the actual particle. Since the shape of the actual pollen particles is usually non-spherical, it is difficult to directly use the particle size value to represent its size, and the diameter is the simplest quantity to describe the size of a geometric body, so the concept of equivalent particle size is adopted; that is, the particle size mentioned in the particle size distribution spectrum detection process is not the real diameter of the pollen particle, but a virtual "equivalent diameter"; the detection methods of pollen particle size include screening method, microscopy method, electric induction method, ultrasonic method and light scattering method, among which the light scattering method includes light scattering method, which is further divided into total light scattering method, diffraction method, polarization ratio method and angle scattering method, and the angle scattering method is used in this embodiment. The basic principle of the angular scattering method is that after the incident light passes through the particles to be measured, scattered light will be generated in all directions. The particle size distribution of the particles is calculated by collecting the scattered light signals in all directions. The angular scattering method relies on the Mie scattering theory and can accurately measure the particle size distribution of the particles. The measurable particle size range is wider than other methods.
[0050] The model building module includes a model building unit and a model training unit. The model building unit is used to create an initial intelligent model. The model training unit uses a model training data set to train the initial intelligent model to obtain an intelligent detection model.
[0051] The model building unit is configured with a model building strategy, which includes: based on the Xception model, the convolution layer and the pooling layer in the Xception model structure are set to a one-dimensional convolution layer and a one-dimensional pooling layer, and a fully connected layer is added, and the softmax function is set as the activation function of the output layer of the Xception model, because the particle size distribution finally output by the model is a probability distribution, and it is necessary to ensure that the non-negative and sum-to-one conditions are met; the Relu function is set as the activation function of other layers except the output layer, because the Xception model has many layers and is a deep neural network. The Relu function has the advantages of fast and deep propagation, and the input is the intensity of the scattered light signal, which is a non-negative value after normalization, so the activation function will not inactivate the neuron; a BN layer is added to each convolution layer to prevent the gradient from disappearing and exploding during training; after completion, the initial intelligent model is obtained;
[0052] The model training unit is configured with a model training strategy, which includes: based on Mie scattering theory, using the particle size information output by the initial intelligent model to calculate the intensity of the corresponding scattered light, marked as the model scattered light intensity, and marking the particle size information output by the initial intelligent model as the model particle size distribution; calculating the difference between the model scattered light intensity and the actual scattered light intensity collected, marked as the scattered light difference; calculating the difference between the model particle size distribution and the actual particle size distribution, marked as the particle size distribution difference, combining the scattered light difference and the particle size distribution difference to construct a loss function, the loss function formula is as follows,
[0053] ,in, is the pollen particle size distribution output by the model, The scattered light intensity corresponding to the pollen particle size distribution output by the model, F is the real pollen particle size distribution, P is the real scattered light intensity corresponding to the real pollen particle size distribution, m represents the number of classifications of the particle size distribution, for example, from 10 nm to 3650 nm, with a classification step of 40 nm, the number of classifications is [(3650-10) / 40]=96, n represents the number of the first angle when collecting the intensity of the scattered light signal, α represents the weight of the scattered light intensity term in the loss function, and the α value in this embodiment is set to 1;
[0054] The Adam algorithm is selected as the optimization algorithm for the initial intelligent model. The Adam algorithm is an optimization algorithm with adaptive learning rate. It combines the advantages of the momentum method and RMSProp optimization methods. Compared with other optimization algorithms, the Adam algorithm has a faster convergence speed, strong adaptability, and high computational efficiency.
[0055] Set the initial intelligent model training hyperparameters, which include the number of training rounds, batch size, and learning rate. In this embodiment, the number of training rounds is set to 200 rounds, the batch size is set to 256, and the learning rate is set to 0.001; set the output layer, train the intelligent detection model based on the model training data set, and obtain the intelligent detection model after completion;
[0056] In the specific implementation process, the Xception model uses depthwise separable convolution to replace the traditional convolution operation. Since the Xception model was proposed in the image field and belongs to a two-dimensional convolutional neural network, the Xception model needs to be modified to meet the needs of pollen particle size distribution detection. The convolution layer and pooling layer of the Xception model are set to one-dimensional convolution layer and one-dimensional pooling layer to adapt to the dimension of the training data set of this model. In addition, since the input of the convolution layer needs to increase the dimension of a channel, a dimension is added to the input data, for example, from (n, 90) to (n, 90, 1), where n represents the number of data in a batch. At the same time, the sigmoid function is selected as the activation function of the output layer to limit the median and standard deviation of the network output to between [0, 1].
[0057] The scattering measurement module includes a laser generating unit and a signal collecting unit. The laser generating unit generates laser light using a laser generating device to irradiate the pollen to be measured, and the signal collecting unit is used to collect scattered light signals to obtain light intensity information of the pollen to be measured.
[0058] The laser generating unit is equipped with a laser generating strategy, which includes: obtaining pollen to be tested, using a dispersant to evenly disperse the pollen to be tested to prepare a test sample, in this embodiment, the dispersant is a hyperdispersant, using a laser generating device, setting the laser wavelength to λ, generating a laser with a wavelength of λ, marked as λ laser, adjusting the level of the laser generating device, allowing the λ laser to vertically pass through the center of the spatial filter and the beam expander lens, generating a parallel monochromatic light beam, and allowing the parallel monochromatic light beam to irradiate the test sample;
[0059] The signal collection unit has a signal collection strategy, which includes: Figure 4 As shown, a parabolic reflector is used to reflect the scattered light generated by irradiating a parallel monochromatic light beam to a test sample as parallel light rays, which are marked as first parallel light rays. A first convex lens is used to converge the first parallel light rays at the focus of the convex lens, and a circular aperture is placed at the focus of the convex lens. The light rays diverging after the first parallel light rays pass through the focus of the first convex lens and the circular aperture are marked as first scattered light rays. A second convex lens is used to convert the first scattered light rays into parallel light rays, which are marked as second parallel light rays. The second parallel light rays are projected onto an image sensor, and a final scattered light image can be obtained. In the scattered light image, different pixel points represent the scattered light signal intensities in different directions, and the light intensity information of the pollen to be tested is obtained.
[0060] In the specific implementation process, the pollen grains of the test sample often exist in the form of a particle system, that is, there are pollen grains of various sizes. When the particle system undergoes light scattering, if the two conditions of uncorrelated scattering and single scattering are met, the collected scattered light signal can be approximately regarded as the sum of the scattered light signals of all single particles at that point, wherein uncorrelated scattering means that the light scattering process of a single particle is not affected by the surrounding particles. In fact, the scattered light of a certain particle and the surrounding particles in the same direction has a certain coherence, but when the distance between the particles is greater than 3 times the particle diameter, the influence of this coherence can be ignored and it can be regarded as uncorrelated scattering; single scattering means that the scattered light signal collected at a certain angle is obtained by superposition of single scattered light signals of some particles. In fact, some particles will scatter the scattered light of other particles again, but the measurement accuracy of the sensor for measuring the scattered light signal is generally very small, so it can also be regarded as single scattering.
[0061] The intelligent processing module includes an intelligent detection model, which uses the intelligent detection model to analyze and process the light intensity information of the pollen to be tested to obtain the particle size distribution data;
[0062] The intelligent processing module is configured with an intelligent processing strategy, which includes: normalizing the light intensity information of the pollen to be measured, and using sum normalization in this embodiment; obtaining normalized light intensity information, inputting the normalized light intensity information into the intelligent detection model, and the intelligent detection model reduces the dimension of the normalized light intensity information through a convolution layer, and in this embodiment, a 1*1 convolution kernel is used, and then through an Inception structure and a residual module, the residual module can avoid the gradient disappearance; by using convolution kernels of different sizes and maximum pooling operations to extract features of different scales in the normalized light intensity information after dimensionality reduction, and finally merging the extracted features of different scales through a fully connected layer, and in this embodiment, the merging method is linear addition, and the particle size distribution data is output through the output layer;
[0063] In the specific implementation process, the core idea of the Inception module is to combine different convolutional layers in parallel, and splice the result matrices processed by different convolutional layers in the depth dimension to form a deeper matrix. The Inception module can adjust the size and number of convolution kernels according to actual needs and flexibly adapt to different application scenarios. It can efficiently expand the depth and width of the model, improve the accuracy of the deep learning network and prevent overfitting. The residual module usually includes one or more convolutional layers, a batch normalization layer and a nonlinear activation function. In this embodiment, the residual module consists of two convolutional layers, each of which is followed by a batch normalization layer and a ReLU activation function. The input data is transformed through these layers, and the original input and the transformed output are added through jump connections to form the final module output.
[0064] The particle size spectrum module obtains the pollen particle size spectrum distribution based on the particle size distribution data;
[0065] The particle size spectrum module is configured with a particle size spectrum generation strategy. The particle size spectrum processing generation includes: based on the particle size distribution data, the pollen particle size is used as the X-axis, and the particle size number frequency is used as the Y-axis to depict the particle size distribution curve in the rectangular coordinate system to obtain the pollen particle size spectrum;
[0066] In the specific implementation process, computer software, such as MATLAB, can be used to plot and analyze the particle size distribution data, obtain the pollen particle size spectrum, and generate a pollen particle size test report.
[0067] Example 2, please refer to Figure 2 As shown, in the second aspect, the present application provides a method for detecting the particle size spectrum distribution of pollen by laser scattering based on artificial intelligence, comprising the following steps:
[0068] Step S1, creating an initial training data set, and performing data processing on the initial training data set to obtain a model training data set; Step S1 includes the following sub-steps:
[0069] Step S101, obtaining common pollen particle size information based on a database;
[0070] Step S102, performing computer simulation according to pollen types based on common pollen particle size information;
[0071] Step S103, observing and recording the intensity of the scattered light signal at different angles in the first angle range at a first angle step according to Mie scattering theory, where the first angle step is b and the first angle range is B;
[0072] Step S104, correspondingly marking the pollen types and the pollen particle size information to obtain N groups of data, classifying them by pollen type and marking them as initial training data sets;
[0073] Step S105, data noise processing: the data of each pollen type in the initial training data set are evenly divided into 4 equal parts, and 3 equal parts are randomly selected from the 4 equal parts to add additive white noise with amplitudes of a1, a2 and a3 respectively, and the noised training data set is obtained after the processing is completed;
[0074] Step S106, data normalization processing: normalize the data of each pollen type in the noise training set according to the data type, and scale the data size to [0, 1]; after the processing is completed, the model training data set is obtained.
[0075] Step S2, creating an initial intelligent model, and training the initial intelligent model using a model training data set to obtain an intelligent detection model; Step S2 includes the following sub-steps:
[0076] Step S201, based on the Xception model, the convolution layer and the pooling layer in the Xception model structure are set to a one-dimensional convolution layer and a one-dimensional pooling layer, and a fully connected layer is added;
[0077] Step S202, setting the softmax function as the activation function of the output layer of the Xception model;
[0078] Step S203, setting the Relu function as the activation function of the layers other than the output layer, adding a BN layer to each convolutional layer, and obtaining the initial intelligent model after completion;
[0079] Step S204, based on Mie scattering theory, using the particle size information output by the initial intelligent model, calculate the intensity of the corresponding scattered light, mark it as model scattered light intensity, and mark the particle size information output by the initial intelligent model as model particle size distribution;
[0080] Step S205, calculating the difference between the model scattered light intensity and the collected real scattered light intensity, and marking it as scattered light difference;
[0081] Step S206, calculating the difference between the model particle size distribution and the actual particle size distribution, and marking it as particle size distribution difference;
[0082] Step S207, combining the scattered light difference and the particle size distribution difference to construct a loss function;
[0083] Step S208, select the Adam algorithm as the optimization algorithm of the initial intelligent model, set the initial intelligent model training hyperparameters, the training hyperparameters include the number of training rounds, batch size and learning rate, and set the output layer;
[0084] Step S209, training the intelligent detection model with the model training data set, and obtaining the intelligent detection model after completion.
[0085] Step S3, using a laser generating device to generate laser to irradiate the pollen to be tested, and collecting scattered light signals to obtain light intensity information of the pollen to be tested; Step S3 includes the following sub-steps:
[0086] Step S301, obtaining pollen to be tested, and using a dispersant to evenly disperse the pollen to be tested to prepare a test sample;
[0087] Step S302, using a laser generating device, setting the laser wavelength to λ, generating laser with a wavelength of λ, marked as λ laser;
[0088] Step S303, adjusting the level of the laser generating device to allow the λ laser to vertically pass through the center of the spatial filter and the beam expander lens to generate a parallel monochromatic light beam, and allowing the parallel monochromatic light beam to irradiate the detection sample;
[0089] Step S304, using a parabolic reflector to reflect the scattered light generated by the parallel monochromatic light beam irradiating the test sample into parallel light rays, marked as first parallel light rays;
[0090] Step S305, using a first convex lens to converge the first parallel light at the focus of the convex lens;
[0091] Step S306, placing a circular aperture at the focus of the convex lens, marking the light diverging from the first parallel light after passing through the focus of the first convex lens and the circular aperture as the first scattered light;
[0092] Step S307, using a second convex lens to transform the first scattered light into parallel light, marked as second parallel light;
[0093] Step S308, projecting the second parallel light onto the image sensor to obtain the light intensity information of the pollen to be measured.
[0094] Step S4, based on the light intensity information of the pollen to be measured, the intelligent detection model is used for analysis and processing to obtain the particle size distribution data; Step S4 includes the following sub-steps:
[0095] Step S401, normalizing the light intensity information of the pollen to be measured to obtain normalized light intensity information;
[0096] Step S402, inputting the normalized light intensity information into the intelligent detection model, and the intelligent detection model performs dimension reduction on the normalized light intensity information through a convolutional layer;
[0097] Step S403, after passing through the Inception structure and the residual module, the features of different scales in the normalized light intensity information after dimensionality reduction are extracted by using convolution kernels of different sizes and maximum pooling operations;
[0098] Step S404: Finally, the features extracted at different scales are merged through the fully connected layer, and the particle size distribution data is output through the output layer.
[0099] Step S5, obtaining the pollen particle size spectrum distribution based on the particle size distribution data; Step S5 includes the following sub-steps:
[0100] Step S501, based on the particle size distribution data, a particle size distribution curve is drawn in a rectangular coordinate system with the pollen particle size as the X-axis and the particle size number frequency as the Y-axis to obtain a pollen particle size spectrum.
[0101] Through the description of the above implementation methods, the embodiments of the present invention can be provided as methods, systems or computer program products. Based on such an understanding, the above technical solutions can be essentially or partly contributed to the prior art in the form of software products, which can be stored in a computer-readable storage medium, such as ROM / RAM, a magnetic disk, an optical disk, etc., and include several instructions for enabling a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.
[0102] In the embodiments provided in the present application, it should be understood that the disclosed system or method can be implemented in other ways. The embodiments described above are merely schematic. For example, the division of modules or units is only a logical function division. There may be other division methods in actual implementation. For example, multiple modules or units can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of systems, modules and units can be electrical, mechanical or other forms.
[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present application, rather than to limit it. Although the present application has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present application.
Claims
1. A laser scattering pollen particle size spectrum distribution detection method based on artificial intelligence, characterized in that: The steps include: Create an initial training data set, and process the initial training data set to obtain a model training data set; Create an initial intelligent model, and use the model training data set to train the initial intelligent model to obtain an intelligent detection model; A laser generating device is used to generate laser light to illuminate the pollen to be tested, and scattered light signals are collected to obtain light intensity information of the pollen to be tested; Based on the light intensity information of the pollen to be tested, the particle size distribution data is obtained by using the intelligent detection model for analysis and processing; The pollen particle size spectrum distribution is obtained based on the particle size distribution data; Creating the initial training dataset consists of the following sub-steps: Common pollen particle size information is obtained based on the database, and computer simulation is performed according to the common pollen particle size information and pollen types. According to the Mie scattering theory, the intensity of the scattered light signal is observed and recorded at different angles in the first angle range with a first angle step, and corresponding marks are made according to the pollen types and the pollen particle size information to obtain N groups of data, which are classified according to the pollen types and marked as the initial training data set, with the first angle step being b and the first angle range being B; Processing the initial training data set to obtain the model training data set includes the following sub-steps: Data noise processing: The data of each pollen type in the initial training data set are divided into 4 equal parts, and 3 equal parts are randomly selected from the 4 equal parts to add additive white noise with amplitudes of a1, a2, and a3 respectively. After the processing is completed, the noisy training data set is obtained; Data normalization: Normalize the data of each pollen type in the noise training set according to the data type and scale the data size to [0, 1]. After the processing is completed, the model training data set is obtained; Creating an initial intelligent model includes the following sub-steps: Based on the Xception model, the convolution layer and pooling layer in the Xception model structure are set to one-dimensional convolution layer and one-dimensional pooling layer, and a fully connected layer is added. The softmax function is set as the activation function of the output layer of the Xception model, and the Relu function is set as the activation function of other layers except the output layer. A BN layer is added to each convolution layer. After completion, the initial intelligent model is obtained; Using the model training data set to train the initial intelligent model to obtain the intelligent detection model includes the following sub-steps: Based on Mie scattering theory, the particle size information output by the initial intelligent model is used to calculate the intensity of the corresponding scattered light, which is marked as the model scattered light intensity. The particle size information output by the initial intelligent model is marked as the model particle size distribution. The difference between the model scattered light intensity and the actual scattered light intensity collected is calculated and marked as the scattered light difference. Calculate the difference between the model particle size distribution and the actual particle size distribution, marked as the particle size distribution difference, and combine the scattered light difference and the particle size distribution difference to construct a loss function; Select the Adam algorithm as the optimization algorithm for the initial intelligent model, set the initial intelligent model training hyperparameters, including the number of training rounds, batch size, and learning rate, set the output layer, and train the intelligent detection model based on the model training data set. After completion, the intelligent detection model is obtained.
2. The method for detecting the laser scattering pollen particle size spectrum distribution based on artificial intelligence according to claim 1, characterized in that: The laser generating device is used to generate laser light to irradiate the pollen to be tested, which includes the following sub-steps: Obtain the pollen to be tested, use a dispersant to evenly disperse the pollen to be tested to prepare a test sample, use a laser generator, set the laser wavelength to λ, generate a laser with a wavelength of λ, marked as λ laser, adjust the level of the laser generator, let the λ laser pass vertically through the center of the spatial filter and the beam expander lens, generate a parallel monochromatic light beam, and let the parallel monochromatic light beam irradiate the test sample.
3. The method for detecting the laser scattering pollen particle size spectrum distribution based on artificial intelligence according to claim 2, characterized in that: Collecting scattered light signals to obtain light intensity information of the pollen to be tested includes the following sub-steps: A parabolic reflector is used to reflect the scattered light generated by the parallel monochromatic light beam irradiating the test sample into parallel light rays, which are marked as first parallel light rays. A first convex lens is used to converge the first parallel light rays at the focus of the convex lens, and a circular aperture is placed at the focus of the convex lens. The light rays diverged after the first parallel light rays pass through the focus of the first convex lens and the circular aperture are marked as first scattered light rays. A second convex lens is used to convert the first scattered light rays into parallel light rays, which are marked as second parallel light rays. The second parallel light rays are projected onto an image sensor to obtain light intensity information of the pollen to be tested.
4. The method for detecting the laser scattering pollen particle size spectrum distribution based on artificial intelligence according to claim 3 is characterized in that: Based on the light intensity information of the pollen to be tested, the intelligent detection model is used to analyze and process the particle size distribution data, which includes the following sub-steps: The light intensity information of the pollen to be tested is normalized to obtain normalized light intensity information, which is then input into the intelligent detection model. The intelligent detection model reduces the dimension of the normalized light intensity information through a convolutional layer, and then passes through the Inception structure and the residual module. By using convolution kernels of different sizes and maximum pooling operations, features at different scales in the normalized light intensity information after dimensionality reduction are extracted. Finally, the extracted features at different scales are merged through a fully connected layer, and the particle size distribution data is output through the output layer.
5. The method for detecting the laser scattering pollen particle size spectrum distribution based on artificial intelligence according to claim 4, characterized in that: Obtaining the pollen particle size spectrum distribution based on the particle size distribution data includes the following sub-steps: Based on the particle size distribution data, the particle size distribution curve is drawn in a rectangular coordinate system with the pollen particle size as the X-axis and the particle size number frequency as the Y-axis to obtain the pollen particle size spectrum.
6. A laser scattered pollen particle size spectrum distribution detection system based on artificial intelligence, applicable to the laser scattered pollen particle size spectrum distribution detection method based on artificial intelligence according to any one of claims 1 to 5, characterized in that: It includes data management module, model building module, scattering measurement module, intelligent processing module and particle size spectrum module; The data management module includes a data construction unit and a data processing unit, the data construction unit is used to create an initial training data set, and the data processing unit is used to process the initial training data set to obtain a model training data set; The model building module includes a model building unit and a model training unit. The model building unit is used to create an initial intelligent model. The model training unit uses a model training data set to train the initial intelligent model to obtain an intelligent detection model. The scattering measurement module includes a laser generating unit and a signal collecting unit. The laser generating unit generates laser light using a laser generating device to irradiate the pollen to be measured, and the signal collecting unit is used to collect scattered light signals to obtain light intensity information of the pollen to be measured. The intelligent processing module includes an intelligent detection model, which uses the intelligent detection model to analyze and process the light intensity information of the pollen to be tested to obtain particle size distribution data; The particle size spectrum module obtains the pollen particle size spectrum distribution based on the particle size distribution data.
Citation Information
Patent Citations
Pollen particle detection method and device and electronic equipment
CN112465753A
Method for measuring particle size distribution of particles, and apparatus thereof
CN103983546A
Particle size distribution measurement method and system based on light scattering field
CN110553955A
