Particle clustering method and device based on ART2A algorithm, equipment and storage medium

Abstract

This application relates to the field of particulate matter processing technology, and provides a particulate matter clustering method, apparatus, device, and storage medium based on the ART2A algorithm, which can reduce inter-class correlation and thus reduce the number of classifications. In this application, mass spectra of multiple particulate matter and the current round's cluster center mass spectra matrix are obtained; based on the mass spectra of the particulate matter and the current round's cluster center mass spectra matrix, each particulate matter is assigned to a cluster center with similar mass spectra, completing the current round of classification; based on the mass spectra of the particulate matter assigned to the cluster center and the similarity between the particulate matter and the cluster center, the current round's cluster center mass spectra matrix is ​​updated to obtain the next round's cluster center mass spectra matrix; wherein, the higher the similarity between the particulate matter and the assigned cluster center, the smaller the update degree of the mass spectra of that cluster center in the current round's cluster center mass spectra matrix, and the lower the similarity, the larger the update degree; based on the mass spectra of multiple particulate matter and the next round's cluster center mass spectra matrix, the multiple particulate matter is classified in the next round.

Description

Technical Field

[0001] This application relates to the field of particulate matter processing technology, and in particular to a particulate matter clustering method, apparatus, computer equipment, and storage medium based on the ART2A algorithm. Background Technology

[0002] Atmospheric particulate matter composition measurement and source apportionment are important topics in atmospheric environmental research, playing a crucial role in air pollution control. Currently, online analysis methods based on particulate matter mass spectrometry are commonly used. Clustering preprocessing of a large amount of particulate matter mass spectrometry information is an essential step, and ART2A is a commonly used method for this preprocessing.

[0003] ATR2A is a clustering method based on cluster centers, meaning it clusters particles according to the similarity between the mass spectra of the particles and the mass spectra of the cluster centers. It typically performs multiple rounds of classification on the particles, updating the mass spectra of the cluster centers at a fixed rate during each round.

[0004] However, when there are a large number of mass spectrometry data samples of particulate matter, since the update degree of the mass spectrometry of the cluster centers is fixed in each round of classification, there are periodic fluctuations in the update characteristics of the cluster centers when there are a large number of samples. This will lead to higher correlation between different categories of particulate matter, resulting in more classifications. Summary of the Invention

[0005] Therefore, it is necessary to provide a particulate clustering method, apparatus, computer equipment, and storage medium that can reduce inter-class correlation and thus reduce the number of classifications, in order to address the above-mentioned technical problems.

[0006] This application provides a particle clustering method based on the ART2A algorithm, the method comprising:

[0007] Obtain the mass spectra of multiple particles and the center-of-class mass spectra matrix for this round;

[0008] Based on the mass spectra of the particles and the mass spectra of the class centers in this round, each particle is classified into a class center with similar mass spectra, thus completing this round of classification;

[0009] Based on the mass spectra of particles assigned to class centers in this round and the similarity between particles and class centers, the class center mass spectra matrix of this round is updated to obtain the class center mass spectra matrix of the next round. Among them, the higher the similarity between particles and the assigned class centers, the smaller the update degree of the mass spectra of that class center in the class center mass spectra matrix of this round, and the lower the similarity, the greater the update degree.

[0010] Based on the mass spectra of multiple particles and the mass spectra matrix of the class centers in the next round, the multiple particles are classified in the next round.

[0011] This application also provides a particle clustering device based on the ART2A algorithm, the device comprising:

[0012] The mass spectrometry acquisition module is used to acquire the mass spectra of multiple particles and the center-of-class mass spectrometry matrix for this round.

[0013] The classification module is used to classify each particulate matter into a class center with similar mass spectra based on the mass spectra of the particulate matter and the mass spectra matrix of the class centers in this round, thus completing the classification in this round.

[0014] The matrix update module is used to update the mass spectrum matrix of the class centers in the current round based on the mass spectra of the particles assigned to the class centers and the similarity between the particles and the class centers, so as to obtain the mass spectrum matrix of the class centers in the next round. The higher the similarity between the particles and the assigned class centers, the smaller the update degree of the mass spectrum of the class center in the current round of the class center mass spectrum matrix; the lower the similarity, the larger the update degree.

[0015] The next-round classification module is used to classify multiple particles in the next round based on their mass spectra and the next-round class center mass spectra matrix.

[0016] This application also provides a computer device, which includes a memory and a processor. The memory stores a computer program, and the processor executes the computer program to perform the following steps:

[0017] Obtain the mass spectra of multiple particles and the center-of-class mass spectra matrix for this round;

[0018] Based on the mass spectra of the particles and the mass spectra of the class centers in this round, each particle is classified into a class center with similar mass spectra, thus completing this round of classification;

[0019] Based on the mass spectra of particles assigned to class centers in this round and the similarity between particles and class centers, the class center mass spectra matrix of this round is updated to obtain the class center mass spectra matrix of the next round. Among them, the higher the similarity between particles and the assigned class centers, the smaller the update degree of the mass spectra of that class center in the class center mass spectra matrix of this round, and the lower the similarity, the greater the update degree.

[0020] Based on the mass spectra of multiple particles and the mass spectra matrix of the class centers in the next round, the multiple particles are classified in the next round.

[0021] This application also provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, performs the following steps:

[0022] Obtain the mass spectra of multiple particles and the center-of-class mass spectra matrix for this round;

[0023] Based on the mass spectra of the particles and the mass spectra of the class centers in this round, each particle is classified into a class center with similar mass spectra, thus completing this round of classification;

[0024] Based on the mass spectra of particles assigned to class centers in this round and the similarity between particles and class centers, the class center mass spectra matrix of this round is updated to obtain the class center mass spectra matrix of the next round. Among them, the higher the similarity between particles and the assigned class centers, the smaller the update degree of the mass spectra of that class center in the class center mass spectra matrix of this round, and the lower the similarity, the greater the update degree.

[0025] Based on the mass spectra of multiple particles and the mass spectra matrix of the class centers in the next round, the multiple particles are classified in the next round.

[0026] The aforementioned particle clustering method, apparatus, computer equipment, and storage medium based on the ART2A algorithm, after acquiring the mass spectra and cluster center mass spectra matrices of multiple particles, classifies each particle into cluster centers with similar mass spectra and performs multiple rounds of classification. During each round of classification, the cluster center mass spectra matrix is ​​updated. The update degree of the cluster center mass spectra matrix is ​​inversely proportional to the similarity between each particle and its corresponding cluster center. Therefore, the update degree of the cluster center mass spectra varies in different rounds, which can avoid the problem of periodic fluctuations in the cluster center update characteristics, reduce inter-cluster correlation, and ultimately form fewer classifications. Attached Figure Description

[0027] Figure 1 This is a flowchart illustrating a particulate matter clustering method based on the ART2A algorithm in one embodiment.

[0028] Figure 2 This is a flowchart illustrating a particulate matter-based segmentation method in one embodiment;

[0029] Figure 3 This is a flowchart illustrating a method for classifying particulate matter based on average spectrum in one embodiment;

[0030] Figure 4 This is a structural block diagram of a particulate matter clustering device based on the ART2A algorithm in one embodiment;

[0031] Figure 5 This is an internal structural diagram of a computer device in one embodiment. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0033] In this application, the reference to "embodiment" means that a specific feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of this application. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a mutually exclusive, independent, or alternative embodiment. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described in this application can be combined with other embodiments.

[0034] In one embodiment, such as Figure 1 As shown, a particle clustering method based on the ART2A algorithm is provided, and its application to a computer device is illustrated, including the following steps:

[0035] Step S101: Obtain the mass spectra of multiple particles and the center-of-class mass spectra matrix for this round.

[0036] Step S102: Based on the mass spectra of the particles and the mass spectra of the current class centers, classify each particle into a class center with similar mass spectra, thus completing this round of classification.

[0037] Particulate matter refers to various solid or liquid particles uniformly dispersed in a gaseous medium. Mass spectrometry is a commonly used analytical method in modern science for measuring the mass-to-charge ratio of ions, and it is also frequently used to analyze the particle size and chemical composition of particulate matter. The mass spectrum of particulate matter can be represented by a vector, where each element in the vector represents a characteristic ion peak in the mass spectrum.

[0038] The atmosphere typically contains multiple particulate matter, and the mass spectra of these particles can form a sparse mass spectrum matrix. Each vector in this matrix represents the mass spectrum of a single particulate matter. For example, a sparse mass spectrum matrix can be characterized as follows:

[0039]

[0040] Where, m i =(a i1 ,…,a in ) represents the mass spectrum row vector of the i-th particulate, and n represents the position of the characteristic ion peak of the mass spectrum, i∈[1,m].

[0041] Classifying multiple particulate matter requires the use of mass spectrometry with multiple class centers. These mass spectrometers can be combined to form a class center mass spectrometry matrix. For example, a class center mass spectrometry matrix can be characterized as follows:

[0042]

[0043] Among them, w j =(b 1j ,…,b nj) T The mass spectrum column vector represents the j-th class center, and n represents the position of the characteristic ion peak in the mass spectrum, j∈[1,q].

[0044] In the above method, all mass spectra in the sparse mass spectrum matrix and the class center mass spectrum matrix need to be normalized using the Euclidean norm.

[0045] During classification, a round of classification is considered complete when all particles in a sparse mass spectrometry matrix are assigned to each class of the current class center mass spectrometry matrix. Multiple rounds of classification are typically performed to improve accuracy. In each round, the mass spectra of each class of the class center mass spectrometry matrix are updated until the class center mass spectrometry matrix meets the convergence condition, at which point the classification process stops.

[0046] Furthermore, based on the mass spectra of the particles and the current class center mass spectra matrix, multiple particles are individually assigned to class centers with similar mass spectra. This includes: randomly obtaining the mass spectrum of one particle from the multiple particles; comparing the mass spectrum of this particle with the mass spectra of each class center in the current class center mass spectra matrix to obtain the similarity between the particle and each class center; obtaining the maximum similarity among the similarities between the particle and each class center; if the maximum similarity is greater than a set threshold, then determining that the particle is mass spectrometrically similar to the class center corresponding to the maximum similarity; and assigning the particle to that class center.

[0047] Specifically, the method for comparing the mass spectrum of a particle with the mass spectra of various centers in this round involves treating the particle's mass spectrum as a row vector and the mass spectra of each center as multiple column vectors, performing dot products on each to obtain multiple similarity values, and constructing a similarity vector from these values. If the maximum similarity value in this similarity vector is greater than a set threshold, the particle is determined to be spectrally similar to the center corresponding to the maximum similarity value, and is assigned to that center. Multiple particles in the sparse mass spectrum matrix are classified using the above comparison method. Once all particles are classified, this round of classification is complete.

[0048] By classifying multiple particles by comparing them one by one, all particles in the sparse mass spectrometry matrix were compared with the class center mass spectrometry matrix, and particles that met the mass spectrometry similarity requirements were assigned to a certain class center, thus achieving the accuracy of classification in each round of classification.

[0049] Step S103: Based on the mass spectra of particles assigned to class centers in this round and the similarity between particles and class centers, update the class center mass spectra matrix for this round to obtain the class center mass spectra matrix for the next round. The higher the similarity between particles and assigned class centers, the smaller the update degree of the mass spectra of that class center in the class center mass spectra matrix in this round; the lower the similarity, the greater the update degree.

[0050] Specifically, in this round of classification, particulate matter has been assigned to various centers with similar mass spectra. At this point, the number of particles under each center can be one or more. If there is only one particle under a certain center, the mass spectrum of that particle and its similarity to that center are used to update the mass spectrum of that center. If there are multiple particles under a certain center, the mass spectrum of that center is used to update the mass spectrum of all particles under that center and the similarity of each particle to that center.

[0051] In this round of classification, the mass spectra of the class center can be updated multiple times during the classification process based on the mass spectra of individual particles and their similarity, or it can be updated once after classification based on the average mass spectra of multiple particles and their average similarity.

[0052] For example, suppose there are class centers A and B, and particles 1, 2, 3 and 4 during classification.

[0053] In one scenario, the mass spectra of the class centers are updated multiple times during classification based on the mass spectra of individual particles and their similarity. The classification process is as follows:

[0054] First, particulate matter 1 is classified by comparing its mass spectrum with that of class center A and class center B. If the mass spectrum of particulate matter 1 is similar to that of class center A, it is classified under class center A. Based on the mass spectrum of particulate matter 1 and the similarity between particulate matter 1 and class center A, the mass spectrum of class center A is updated to obtain the updated class center A.

[0055] Next, particulate matter 2 is classified by comparing its mass spectrum with the updated mass spectra of class center A and class center B. If the mass spectrum of particulate matter 2 is similar to that of class center B, it is classified under class center B. Based on the mass spectrum of particulate matter 2 and the similarity between particulate matter 2 and class center B, the mass spectrum of class center B is updated to obtain the updated class center B.

[0056] Then, particulate matter 3 is classified by comparing its mass spectrum with the updated class center A and the updated class center B. If the mass spectrum of particulate matter 3 is similar to that of the updated class center A, it is classified under the updated class center A. Based on the mass spectrum of particulate matter 3 and the similarity between particulate matter 3 and the updated class center A, the mass spectrum of the updated class center A is updated again to obtain the updated class center A.

[0057] Finally, particulate matter 4 was classified. The mass spectrum of particulate matter 3 was compared with the mass spectra of the updated class centers A and B. No specific classification results are assumed here; it is understood that the mass spectrum of the class center to which particulate matter 4 was assigned will be updated based on the classification results.

[0058] In another scenario, an update is performed after classification based on the average mass spectrometry values ​​of multiple particles and the average similarity value. The classification process is as follows:

[0059] First, particulate matter 1 is classified by comparing its mass spectrum with that of class center A and class center B. If its mass spectrum is similar to that of class center A, it is classified under class center A, while the mass spectrum of class center A remains unchanged.

[0060] Next, particulate matter 2 is classified by comparing its mass spectrum with that of class center A and class center B. If its mass spectrum is similar to that of class center B, it is classified under class center B, while the mass spectrum of class center B remains unchanged.

[0061] Then, particulate matter 3 and particulate matter 4 were classified in the same way as described above, assuming that both particulate matter 3 and particulate matter 4 were similar to class center B in terms of mass spectra.

[0062] Finally, the mass spectra of class centers A and B are updated. At this point, class center A contains particulate matter 1, and class center B contains partsiculate matter 2, 3, and 4. When updating the mass spectrum of class center A, since it contains only one particulate matter, the update is performed based on the mass spectrum of particulate matter 1 and its similarity to class center A. When updating the mass spectrum of class center B, the average mass spectra of partsic matter 2, 3, and 4, as well as the average similarity between partsic matter 2, 3, and 4 and class center B, are calculated. The mass spectrum of class center B is then updated based on these average mass spectra and the average similarity.

[0063] When updating the mass spectrometer of a certain class center, the higher the similarity between the particulate matter and the class center it is assigned to, the smaller the update degree of the mass spectrometer of that class center; the lower the similarity, the larger the update degree. That is, there is a negative correlation between the update degree and the similarity.

[0064] Step S104: Based on the mass spectra of multiple particles and the mass spectra matrix of the next class center, perform the next round of classification on multiple particles.

[0065] Specifically, after the current round of classification, the next round's class center mass spectrometry matrix is ​​obtained by updating the class center mass spectrometry matrix of the current round. The mass spectra of multiple particles are then classified particle by particle in the next round. The mass spectrum of a single particle is compared with the mass spectra of all class centers in the next round's class center mass spectrometry matrix, and the particle is assigned to the class center with the highest similarity. This particle-by-particle comparison is repeated to classify multiple particles, and the next round of classification is completed after all particles have been classified.

[0066] The particle clustering method based on the ART2A algorithm described above can, after obtaining the mass spectra and cluster center mass spectra matrices of multiple particles, classify each particle into cluster centers with similar mass spectra and perform multiple rounds of classification. During each round of classification, the cluster center mass spectra matrix is ​​updated, and the degree of update in the cluster center mass spectra matrix is ​​negatively correlated with the similarity between each particle and its respective cluster center. This method of updating the cluster center mass spectra matrix reduces inter-cluster correlation, ultimately resulting in fewer clusters and effectively suppressing the overly fine clustering problem present in the original clustering method.

[0067] In one embodiment, the initial center-of-class mass spectrometry matrix is ​​transformed from empty to non-empty. The transformation process includes the following steps: during the initial particle-by-particle partitioning, the mass spectrum of the first partitioned particle is directly added to the initial empty center-of-class mass spectrometry matrix as the center-of-class mass spectrum, resulting in the initial non-empty center-of-class mass spectrometry matrix; when partitioning non-first particles, the mass spectrum of the partitioned non-first particles is compared with the center-of-class mass spectrum of the non-empty center-of-class mass spectrometry matrix; if the mass spectrum of the partitioned non-first particles is dissimilar to the center-of-class mass spectrum of the non-empty center-of-class mass spectrometry matrix, the mass spectrum of the partitioned non-first particles is used as the new center-of-class mass spectrum and added to the non-empty center-of-class mass spectrometry matrix.

[0068] Specifically, an empty matrix is ​​created and used as the center-of-class mass spectrometry (COSMS) matrix. During the first round of classification, the mass spectrum of the first selected particle is added to this empty COSMS matrix as a center-of-class mass spectrum, transforming it into a non-empty matrix. Next, for particles other than the first particle, their mass spectra are compared with the existing center-of-class mass spectra in the non-empty matrix. If a particle's mass spectrum is dissimilar to all existing center-of-class mass spectra in the non-empty matrix, it is added as a new center-of-class mass spectrum. After comparing all remaining particles, the first round of classification is complete, yielding the final center-of-class mass spectrometry (COSMS) matrix.

[0069] In this embodiment, the initial classification of particulate matter forms a first-round class center mass spectrometry matrix. During the initial classification process, a class center mass spectrometry matrix composed of mass spectra of various classes with appropriate inter-class aggregation is established. This facilitates the reasonable classification of particulate matter mass spectra in subsequent classification rounds, and allows for the updating of the class center mass spectrometry matrix during the classification process.

[0070] After the first round of classification is completed, when a non-first round of classification is to be carried out, that is, the second and subsequent rounds of classification, the process of changing the class center mass spectrometry matrix is ​​the same as that of the first round of classification. The class center mass spectrometry matrix will continuously add new class center mass spectra based on the foundation before the start of each round of classification, so as to classify the mass spectra of all particles more accurately.

[0071] As mentioned above, during a single round of classification, the mass spectra of the class centers can be updated multiple times based on the mass spectra of individual particles and their similarity. That is, the mass spectra of the class center are updated each time a particle is assigned to a class center. In this approach, based on the particle mass spectra and the current round's class center mass spectra matrix, each particle is assigned to a class center with similar mass spectra, completing the current round of classification. Specifically, this can include: assigning multiple particles one by one to class centers with similar mass spectra based on the particle mass spectra and the current round's class center mass spectra matrix; the class center mass spectra matrix used to assign the first particle in this round is obtained from the initial mass spectra of each class center in this round; the method for obtaining the class center mass spectra matrix used to assign particles other than the first particle is: updating the mass spectra of the class centers to which the previous particle was assigned in the current round's class center mass spectra matrix based on the mass spectra of the previous particle; and using the updated class center mass spectra matrix as the class center mass spectra matrix used for assigning the current particle.

[0072] Specifically, by repeatedly comparing the mass spectra of particulate matter with the current class center mass spectra matrix, the mass spectra of multiple particulate matter are successively classified into class centers with similar mass spectra. In non-initial classifications, the mass spectra of each class center in the current round of class center mass spectra matrix used to classify the first particulate matter are the initial mass spectra of this round.

[0073] After classifying the first particle, the subsequent non-first particles are classified. When classifying non-first particles, the mass spectrum of the previous particle's cluster center is updated based on its mass spectrum. The updated cluster center mass spectrum matrix is ​​then used as the cluster center mass spectrum matrix for classifying the non-first particles, and the mass spectra of the non-first particles are compared with the updated cluster center mass spectrum matrix.

[0074] In updating the mass spectrometer of the class center, if the similarity between the previous particle and the class center to which the previous particle was assigned is higher, the update degree of the mass spectrometer of that class center is smaller; if the similarity is lower, the update degree is larger.

[0075] Based on the mass spectra of particles assigned to class centers in this round and the similarity between particles and class centers, the class center mass spectra matrix of this round is updated to obtain the class center mass spectra matrix of the next round. Specifically, this may include: after completing the classification of the last particle, updating the mass spectra of the class centers assigned to the last particle in the class center mass spectra matrix of this round based on the mass spectra of the last particle; and using the updated class center mass spectra matrix as the class center mass spectra matrix of the next round.

[0076] Specifically, after classifying the last particulate matter, the mass spectrum of the class center to which the last particulate matter was assigned is updated based on the mass spectrum of the last particulate matter. After completing this round of particulate matter classification and updating the class center mass spectrum matrix, the resulting updated class center mass spectrum matrix is ​​used as the class center mass spectrum matrix for the next round.

[0077] In this embodiment, multiple rounds of classification are performed on the mass spectra of multiple particles based on particle-by-particle mass spectrometry. In each round of classification, the class center mass spectrometry matrix is ​​updated based on the mass spectra of each particle and its similarity, and the update degree is inversely proportional to the similarity. Through the above steps, the classification of the next particle in the current round is better guided, the accuracy of classification is improved, and the correlation between class centers is reduced.

[0078] As mentioned above, during a classification round, the mass spectra of the class centers can be updated after the classification process based on the average mass spectra of multiple particles and the average similarity. That is, at the end of a classification round, the mass spectra of a class center are updated based on the multiple particles assigned to it. In this approach, based on the mass spectra of the particles and the current round's class center mass spectra matrix, each particle is assigned to a class center with similar mass spectra, completing the current round of classification. Specifically, this can include: assigning multiple particles one by one to a class center with similar mass spectra based on the particle mass spectra and the current round's class center mass spectra matrix; when this round is not the first round, the class center mass spectra matrix used to assign each particle remains unchanged during this round of classification.

[0079] Specifically, by repeatedly comparing the mass spectra of particulate matter with the current class center mass spectra matrix, the mass spectra of multiple particulate matter are successively classified into class centers similar to their mass spectra. When this is the first round, the change process of the first round class center mass spectra matrix used to classify each particulate matter is as described above. When this is not the first round, the value of the current round class center mass spectra matrix does not change during the classification process.

[0080] Based on the mass spectra of particles classified into class centers in this round and the similarity between particles and class centers, the class center mass spectrometry matrix for this round is updated to obtain the class center mass spectrometry matrix for the next round. This includes: after completing the classification of the last particle, updating the mass spectra of each class center in the class center mass spectrometry matrix for this round based on the average mass spectra of all particles under each class center and the average similarity between all particles under each class center and their respective class centers, to obtain the class center mass spectrometry matrix for the next round. Among these, the higher the average similarity between all particles under each class center and their respective class centers, the smaller the update degree of the mass spectra of each class center in the class center mass spectrometry matrix for this round; the lower the similarity, the greater the update degree.

[0081] Specifically, after classifying the last particulate matter, the average mass spectrometry value of all parts under each class center and the average similarity value between all parts under each class center and their respective class centers are calculated. The mass spectra of each class center are then updated based on these values ​​to obtain the class center mass spectrometry matrix for the next round. A higher average similarity value between all parts under each class center and their respective class center results in a smaller update of the class center mass spectra, while a lower similarity value results in a larger update.

[0082] In this embodiment, multiple rounds of mass spectra classification are performed on the mass spectra of multiple particles based on the average spectrum. After each round of classification, the mass spectra of each type of center are updated based on the average mass spectra of all particles under each type of center and the average similarity between all particles under each type of center and their respective type of center. The degree of update is inversely proportional to the average similarity. Through the above steps, the computational load during classification is reduced, the classification efficiency is improved, and the correlation between different types of centers is reduced.

[0083] In one embodiment, the degree of update of the mass spectrometry of the class center can be determined by the following steps: obtaining the similarity between the particulate matter and the assigned class center and the adaptive learning rate function; substituting the similarity between the particulate matter and the assigned class center into the adaptive learning rate function to obtain the degree of update of the mass spectrometry of the class center in the current round of the class center mass spectrometry matrix.

[0084] Specifically, when performing multiple rounds of classification of mass spectra of multiple particles based on particle-by-particle mass spectrometry, the similarity between each particle and its assigned class center is obtained. Conversely, when performing multiple rounds of classification of mass spectra of multiple particles based on average spectrum, the average similarity between all particles in each class center and their respective class centers is obtained. The obtained similarity is then substituted as an independent variable into the selected adaptive learning rate function to obtain the update level of the mass spectrum for that class center in the current round's class center mass spectrum matrix. In the above update step, the mass spectra of each class center are updated based on the obtained update level of the class center's mass spectrum. The selection of the adaptive learning rate function is not limited, as long as a negative correlation is achieved between the update level and the similarity.

[0085] For example, the formula for the adaptive learning rate function can be as follows:

[0086]

[0087] Where P is the correlation coefficient, λ0 is the initial learning rate, K is the 50% decrease (increase) rate of the adaptive learning rate function, C is the 50% decrease adaptive learning rate, and L is the minimum learning rate.

[0088] In this embodiment, the correlation coefficient P in the adaptive learning rate function is the similarity between the particulate matter and the class center, and the values ​​of λ0, K, C and L are set according to the actual situation.

[0089] According to the above formula, the larger P is, the smaller λ is; the smaller P is, the larger λ is. P and λ are negatively correlated. Introducing this into the calculation of the update degree in the above steps can achieve a negative correlation between the update degree and the similarity.

[0090] In this embodiment, by introducing an adaptive learning function, the update rate of the mass spectra of the cluster centers can be adjusted according to the similarity. Unlike the original clustering method, which updates the mass spectra of the cluster centers at a fixed learning rate, this embodiment improves upon the original clustering method by adaptively adjusting the mass spectra of the cluster centers, thereby reducing inter-cluster correlation and effectively suppressing the problem of overly refined clustering in the original clustering method.

[0091] In one embodiment, before proceeding to the next round of classification after completing the current round, it is necessary to determine whether the class center mass spectrometry matrix meets the convergence condition. The classification process ends when the class center mass spectrometry matrix converges. If the class center mass spectrometry matrix does not converge, the next round of classification continues.

[0092] When determining whether convergence has occurred, the minimum similarity among all particulate matter and its class center is obtained. If the minimum similarity is less than a set threshold, the class center mass spectrometry matrix is ​​determined to have not converged, and the multiple particulate matter is classified in the next round based on the mass spectra of multiple particulate matter and the class center mass spectrometry matrix of the next round.

[0093] In this embodiment, the minimum similarity among all particles and their respective class centers determines whether to continue the next round of classification after completing the current round. This allows for continuous updating of the class center mass spectrometry through multiple rounds of classification, even when the class center mass spectrometry matrix has not reached the convergence condition. This continuously improves the similarity between each particle and its class center, thereby continuously improving the accuracy of the classification results.

[0094] To better understand the above method, please refer to the accompanying diagram. Figure 2 and Figure 3 The following details an application example of the particle clustering method based on the ART2A algorithm proposed in this application.

[0095] In this embodiment, depending on the different update methods used when updating the center-of-class mass spectrometry matrix, there are two schemes: one based on particle-by-particle mass spectrometry and the other based on average spectrum.

[0096] Combination Figure 2 It can be seen that the scheme for updating the cluster center mass spectrum based on particle-by-particle mass spectrometry includes: Step S201, initialization of the cluster center mass spectrum matrix WM: create an empty matrix and use it as the cluster center mass spectrum matrix WM, i.e., WM = []. In the first round of classification, the transpose of the mass spectrum m1 of the first particle in the sparse mass spectrum matrix MM is... It is added to the center-of-class mass spectrometry matrix as a center-of-class mass spectrometry.

[0097] Step S202: Obtain the sparse mass spectrum matrix MM composed of the mass spectra of multiple particles, and randomly select the mass spectrum m of a particle from the sparse mass spectrum matrix MM. i The mass spectrometry of this particulate matter... i As row vectors, the mass spectra of each class center in the class-centered mass spectrometry matrix WM in this round are represented as multiple column vectors, through m i The WM dot product method is used to compare the mass spectrum of the particle with the mass spectra of each class center in the current class center mass spectrum matrix WM, and obtain a similarity vector P composed of the similarity values ​​between the particle and each class center.

[0098] Step S203: Obtain the maximum similarity value max(P) in the similarity vector P. If max(P) is greater than the set threshold VF, then determine the mass spectrometry m of the particulate matter. iThe mass spectrum w of the class center corresponding to the maximum similarity value max(P) k Since their mass spectra are similar, they are classified into this class center.

[0099] Find the mass spectrum w of the class center corresponding to the maximum similarity value max(P) from the class center mass spectrum matrix WM. k And update it, the update process is shown in the following formula:

[0100]

[0101]

[0102] In the adaptive learning rate function λ[P,K,C,L], the correlation coefficient P is the maximum similarity value max(P), and the other parameters λ0, K, C, and L are fixed values.

[0103] Specifically, the larger the maximum similarity value max(P), the smaller the value of the adaptive learning rate function λ[max(P),K,C,L], and the better the mass spectrum w of the class center. k Update level The smaller it is. Furthermore, after updating the mass spectrometer w at the class center... k After that, it needs to be normalized, which can be expressed by the formula as follows:

[0104] If the maximum similarity value max(P) is less than the set threshold VF, then the mass spectrometer of the particle is... i transpose m i T The mass spectra used as new class centers are added to the class center mass spectrum matrix WM, as expressed by the formula:

[0105]

[0106] Step S204: After classifying the mass spectrum of one particulate matter and updating the class center mass spectrum matrix WM, continue classifying the mass spectra of the remaining particulate matter. When classifying the mass spectra of particles other than the first one, the class center mass spectrum matrix WM used is the updated class center mass spectrum matrix WM obtained based on the mass spectrum of the previous particulate matter.

[0107] Step S205: In the first round of classification, the mass spectra of all particles in the sparse mass spectrometry matrix MM are compared according to steps S202, S203, and S204 above, and the class center mass spectrometry matrix WM is updated in one round according to step S203 above, finally forming the first round of class center mass spectrometry matrix WM. Before the next round of classification, it is necessary to determine whether the convergence condition is met based on the minimum similarity among all particles and their respective class centers. If the minimum similarity is less than the set threshold VF, it is determined that convergence has not been achieved, and the class center mass spectrometry matrix WM does not meet the convergence condition, and the next round of classification continues. In the next round of classification, the class center mass spectrometry matrix WM obtained in this round of classification will become the class center mass spectrometry matrix WM for the next round.

[0108] If the minimum similarity is greater than the set threshold VF, then the class center mass spectrum matrix WM satisfies the convergence condition, and the next round of classification is stopped.

[0109] Step S206: In the next round of classification, repeat the classification process described in steps S202, S203, S204, and S205. After multiple rounds of classification, continuously update the class center mass spectrometry matrix WM until the class center mass spectrometry matrix WM meets the convergence condition, then stop the next round of classification and end the classification of particulate matter.

[0110] The above steps S201 to S206 can be represented by pseudocode as follows:

[0111]

[0112] Among them, MM is in R M*N The mass spectral sparse matrix in space, WM is in R N*K The mass spectrometry matrix of the class centers in space, where M is the number of particles, N is the number of characteristic ion peaks, and K is the number of class centers; the function a = randperm[1:M] outputs a random integer sequence vector a between 1 and M with no repeated elements; the function a = min(x) outputs the minimum value a of the x vector; the function {a,b} = max(x) outputs the maximum value a and the index b of the maximum value in x; the function {a,b} = size(x) outputs the row a and column b of the x matrix; M T This is a matrix transpose operation;

[0113] Combination Figure 3 It can be seen that the schemes for updating center-like mass spectra based on average spectra include:

[0114] Step S301, Initialize the class center mass spectrometry matrix WM: Create an empty matrix and use it as the class center mass spectrometry matrix WM, i.e., WM = []. During the first round of classification, the transpose m1 of the mass spectrum m1 of the first particle in the sparse mass spectrometry matrix MM will be used. TIt is added to the center-of-class mass spectrometry matrix as a center-of-class mass spectrometry.

[0115] Step S302: Obtain the sparse mass spectrum matrix MM composed of the mass spectra of multiple particles, and randomly select the mass spectrum m of a particle from the sparse mass spectrum matrix MM. i The mass spectrometry of this particulate matter... i As row vectors, the mass spectra of each class center in the class-centered mass spectrometry matrix WM in this round are represented as multiple column vectors, through m i The WM dot product method compares the mass spectrum of the particle with the mass spectra of each class center in the current class center mass spectrum matrix WM, resulting in a similarity vector P composed of the similarity values ​​between the particle and each class center. In a single round of classification, the mass spectra of each class center in the class center mass spectrum matrix WM remain unchanged during the classification process.

[0116] Step S303: Obtain the maximum similarity value max(P) in the similarity vector P. If max(P) is greater than the set threshold VF, then determine the mass spectrometry m of the particulate matter. i The mass spectrum w of the class center corresponding to the maximum similarity value max(P) k Since their mass spectra are similar, they are classified into this class center.

[0117] If the maximum similarity value max(P) is less than the set threshold VF, then the mass spectrometer of the particle is... i transpose m i T The mass spectrometer, as the new class center, is added to the class center mass spectrometer matrix WM, as expressed by the formula:

[0118] Step S304: After classifying the mass spectra of all particles in the sparse mass spectrometry matrix MM according to steps S302 and S303 above, a round of updates is performed on the mass spectra of all class centers in the class center mass spectrometry matrix WM. The update process is shown in the following formula:

[0119]

[0120]

[0121] Wherein, the correlation coefficient P in the adaptive learning rate function λ(P,K,C,L) is the class center mass spectrum w k Average similarity of all mass spectra For center mass spectrometry w k The average value of all mass spectra is used, while the other parameters λ0, K, C, and L are fixed values.

[0122] Specifically, when the average similarity When the value is larger, the adaptive learning rate function The smaller the value, the higher the mass spectrometry value of the class center. k Update level The smaller it is.

[0123] In addition, in the updated class center of mass spectrometry w k In addition, it also needs to be normalized, which can be expressed by the formula:

[0124] Step S305 involves updating the class center mass spectrometry matrix WM to form the first-round class center mass spectrometry matrix WM. Before the next round of classification, it is necessary to determine whether the convergence condition is met based on the minimum similarity among all particulate matter and its class center. If the minimum similarity is less than a set threshold VF, convergence is deemed not to have occurred, and the class center mass spectrometry matrix WM does not meet the convergence condition, thus continuing with the next round of classification. In the next round of classification, the class center mass spectrometry matrix WM obtained in this round will become the class center mass spectrometry matrix WM for the next round.

[0125] If the minimum similarity is greater than the set threshold VF, then the class center mass spectrum matrix WM satisfies the convergence condition, and the next round of classification is stopped.

[0126] Step S306: In the next round of classification, repeat the classification process of steps S302, S303, S304, and S305. After multiple rounds of classification, continuously update the class center mass spectrometry matrix WM until the class center mass spectrometry matrix WM meets the convergence condition, then stop the next round of classification and end the classification of particulate matter.

[0127] The above steps S301 to S306 can be represented by pseudocode as follows:

[0128]

[0129] Among them, MM is in R M*N The mass spectral sparse matrix in space, WM is in R N*K The mass spectrometry matrix of the class centers in space, where M is the number of particles, N is the number of characteristic ion peaks, and K is the number of class centers; the function a = randperm[1:M] outputs a random integer sequence vector a between 1 and M with no repeated elements; the function a = min(x) outputs the minimum value a of the x vector; the function {a,b} = max(x) outputs the maximum value a and the index b of the maximum value in x; the function {a,b} = size(x) outputs the row a and column b of the x matrix; M T This is a matrix transpose operation;

[0130] In this embodiment, both the scheme of updating the cluster center mass spectrum based on particle-by-particle mass spectrometry and the scheme of updating the cluster center mass spectrum based on the average spectrum achieve accurate classification of multiple particles, reduce the inter-class correlation, and suppress the problem of overly fine clustering in the original clustering method.

[0131] To better demonstrate the beneficial effects of this application, the following explanation combines parameter settings from the algorithm pseudocode and algorithm performance data:

[0132] First, air was sampled using a single-particle aerosol mass spectrometer, obtaining 465,310 mass spectra. Classification tests were then conducted using both traditional algorithms and the algorithm proposed in this application. After calculation, classes comprising more than 97% of the total particles were selected as valid groups for algorithm performance analysis. Specific algorithm parameters are shown in Table 1.

[0133] algorithm VF λ K C L Traditional Algorithm 0.7 0.0301 - - - Algorithm in this application 0.7 0.0301 50 0.85 <![CDATA[10 -5 ]]>

[0134] Table 1

[0135] Secondly, determine the algorithm performance analysis standards: ① Correlation test: Perform a correlation test between the internal mass spectra of each category in the classification results, and count the proportion of correlation P ≥ 0.7; use the average value of all mass spectra within a category to perform a correlation test between the categories in the classification results, and count the proportion of correlation P ≤ 0.7. ② Time: Under the same algorithm parameters, count the time required for the algorithm to complete the classification. ③ Number of classes: Under the same algorithm parameters, count the number of valid classes in the algorithm's classification results / the total number of classes.

[0136] Based on the above algorithm performance analysis criteria, the statistically obtained algorithm performance parameters are shown in Table 2.

[0137]

[0138] Table 2

[0139] Finally, the classification results are analyzed:

[0140] Compared with traditional algorithms, the proportion of RN with intra-class similarity greater than 0.7 in the algorithm of this application is higher. mean The percentage decreased from 97% to 90%; while the percentage of classes with inter-class similarity less than 0.7 (RN) 1,mean The accuracy increased from 87% to 92%. The time taken for the algorithm to complete the classification decreased from 2000s to 120s. In addition, the number of valid classes in the classification results decreased from 745 to 250.

[0141] Under the same sample and algorithm parameters, compared with traditional algorithms, the algorithm in this application significantly reduces both intra-class and inter-class correlations, improves algorithm speed by 16.7 times, and reduces the number of classifications. This avoids the problem of a large number of classifications caused by high inter-class correlations in traditional algorithms, facilitating subsequent processing of mass spectrometry data in practical applications.

[0142] It should be understood that although the steps in the flowcharts of the embodiments described above are shown sequentially according to the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the embodiments described above may include multiple steps or multiple stages. These steps or stages are not necessarily completed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the steps or stages of other steps.

[0143] In one embodiment, such as Figure 4 As shown, a particulate matter clustering device based on the ART2A algorithm is provided, comprising:

[0144] The mass spectrometry acquisition module 401 is used to acquire the mass spectra of multiple particles and the center-of-class mass spectrometry matrix of the current round.

[0145] The current classification module 402 is used to classify each particulate matter into a class center with similar mass spectra based on the mass spectra of the particulate matter and the current class center mass spectra matrix, thus completing the current classification.

[0146] The matrix update module 403 is used to update the mass spectrum matrix of the class center in the current round based on the mass spectra of the particles assigned to the class center in the current round and the similarity between the particles and the class center, so as to obtain the mass spectrum matrix of the class center in the next round. Among them, the higher the similarity between the particles and the assigned class center, the smaller the update degree of the mass spectrum of the class center in the current round of the class center mass spectrum matrix, and the lower the similarity, the larger the update degree.

[0147] The next-round classification module 404 is used to perform a next-round classification of the multiple particles based on the mass spectra of the multiple particles and the next-round class center mass spectra matrix.

[0148] In one embodiment, the current classification module 402 is further configured to classify the plurality of particles into mass spectrometric similarity class centers according to the mass spectra of the particles and the current class center mass spectrometry matrix.

[0149] In this classification round, the mass spectrometry matrix of the class centers used to classify the first particle is obtained based on the initial mass spectra of each class center in this round. The method for obtaining the mass spectrometry matrix of the class centers used to classify non-first particles is as follows: based on the mass spectra of the previous particle, the mass spectra of the class centers to which the previous particle was classified in the current round of the class center mass spectrometry matrix are updated; the updated class center mass spectrometry matrix is ​​used as the class center mass spectrometry matrix used when classifying the particle. The higher the similarity between the previous particle and the class centers to which the previous particle was classified, the smaller the degree of update; the lower the similarity between the previous particle and the class centers to which the previous particle belonged, the greater the degree of update.

[0150] The matrix update module 403 is also used to update the mass spectra of the last particle to which it belongs in the current round of the class center mass spectra matrix after the classification of the last particle is completed, based on the mass spectra of the last particle; and use the updated class center mass spectra matrix as the class center mass spectra matrix for the next round.

[0151] In one embodiment, the current classification module 402 is further configured to classify the plurality of particles into mass spectrometric similarity class centers according to the mass spectra of the particles and the current class center mass spectrometry matrix.

[0152] When this round is not the first round, the class center mass spectrometry matrix used to classify each particulate matter remains unchanged during this round of classification.

[0153] The matrix update module 403 is also used to update the mass spectra of each type of center in the current round of the class center mass spectra matrix after completing the classification of the last particulate matter, based on the average mass spectra of all particulate matter under each type of center and the average similarity between all particulate matter under each type of center and its respective class center, to obtain the class center mass spectra matrix for the next round; wherein, the higher the average similarity between all particulate matter under each type of center and its respective class center, the smaller the update degree of the mass spectra of each type of center in the current round of the class center mass spectra matrix, and the lower the similarity, the greater the update degree.

[0154] In one embodiment, the current classification module 402 is further configured to: randomly obtain the mass spectrum of a particle from the plurality of particles; compare the mass spectrum of the particle with the mass spectra of each class center in the current class center mass spectrum matrix to obtain the similarity between the particle and each class center; obtain the maximum similarity among the similarities between the particle and each class center; if the maximum similarity is greater than a set threshold, determine that the particle has mass spectrum similarity with the class center corresponding to the maximum similarity; and classify the particle under that class center.

[0155] In one embodiment, the device further includes a matrix transformation module, used to directly add the mass spectrum of the first segmented particle as the center-of-class mass spectrum to the empty matrix of the center-of-class mass spectrum in the first round of particle-by-particle segmentation, thereby obtaining the non-empty matrix of the center-of-class mass spectrum in the first round; when segmenting non-first particles, the mass spectrum of the non-first particles is compared with the center-of-class mass spectrum of the non-empty matrix of the center-of-class mass spectrum; if the mass spectrum of the non-first particles is dissimilar to the center-of-class mass spectrum of the non-empty matrix, the mass spectrum of the non-first particles is added as a new center-of-class mass spectrum to the non-empty matrix of the center-of-class mass spectrum.

[0156] In one embodiment, the device further includes an update degree calculation module, used to obtain the similarity between the particulate matter and the assigned class center and the adaptive learning rate function; and to substitute the similarity between the particulate matter and the assigned class center into the adaptive learning rate function to obtain the update degree of the mass spectrum of the class center in the current round of the class center mass spectrum matrix.

[0157] In one embodiment, the device further includes a convergence determination module, which obtains the minimum similarity among all particulate matter and its class center; if the minimum similarity is less than a set threshold, it determines that the class center mass spectrometry matrix has not converged, and performs a next round of classification on the multiple particulate matter based on the mass spectra of the multiple particulate matter and the class center mass spectrometry matrix of the next round.

[0158] Specific limitations regarding the ART2A-based particle clustering device can be found in the above section on the limitations of the ART2A-based particle clustering method, and will not be repeated here. Each module in the aforementioned ART2A-based particle clustering device can be implemented entirely or partially through software, hardware, or a combination thereof. These modules can be embedded in or independent of the processor in a computer device, or stored in the computer device's memory as software, so that the processor can call and execute the corresponding operations of each module.

[0159] In one embodiment, a computer device is provided, the internal structure of which can be shown as follows: Figure 5As shown, the computer device includes a processor, memory, and network interface connected via a system bus. The processor provides computational and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system, computer programs, and a database. The internal memory provides the environment for the operation of the operating system and computer programs in the non-volatile storage media. The database stores particle clustering data. The network interface communicates with external terminals via a network connection. The computer device also includes input / output interfaces (I / O interfaces), which are connection circuits between the processor and external devices for exchanging information; they are connected to the processor via a bus. When the computer program is executed by the processor, it implements a particle clustering method based on the ART2A algorithm.

[0160] Those skilled in the art will understand that Figure 5 The structure shown is merely a block diagram of a portion of the structure related to the present application and does not constitute a limitation on the computer device to which the present application is applied. Specific computer devices may include more or fewer components than those shown in the figure, or combine certain components, or have different component arrangements.

[0161] In one embodiment, a computer device is provided, including a memory and a processor, the memory storing a computer program, and the processor executing the computer program to implement the steps in the various method embodiments described above.

[0162] In one embodiment, a computer-readable storage medium is provided having a computer program stored thereon, which, when executed by a processor, implements the steps in the various method embodiments described above.

[0163] In one embodiment, a computer program product is provided having a computer program stored thereon, the computer program being executed by a processor of the steps described in the various method embodiments above.

[0164] Those skilled in the art will understand that all or part of the processes in the methods of the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments of the above methods. Any references to memory, storage, databases, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, or optical storage, etc. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc.

[0165] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0166] The above embodiments are merely illustrative of several implementation methods of this application, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this patent application should be determined by the appended claims.

Claims

1. A particulate matter clustering method based on the ART2A algorithm, characterized in that, The method includes: Obtain the mass spectra of multiple particles and the center-of-class mass spectra matrix for this round; Based on the mass spectra of the particles and the mass spectra of the class centers in this round, each particle is classified into a class center with similar mass spectra, thus completing this round of classification; Based on the mass spectra of particles classified under the current class centers and the similarity between particles and class centers, the mass spectra of the class centers corresponding to the maximum similarity values ​​are found from the current class center mass spectra matrix. If the maximum similarity value is greater than or equal to a set threshold, the mass spectra of the class centers corresponding to the maximum similarity value are updated to obtain the class center mass spectra matrix for the next round. The update process of the mass spectra of the class centers corresponding to the maximum similarity value can be expressed as follows: ； ； Among them, the The adaptive learning rate function is used to realize the negative correlation between the update degree of the mass spectrometry of the class center and the similarity; max(P) is the maximum similarity value; m i The mass spectrum of a randomly selected particle from a sparse mass spectrum matrix, wherein the sparse mass spectrum matrix consists of mass spectra of multiple particles; the w k The mass spectra of the class centers corresponding to the maximum similarity values; K, C, and L are fixed values; For normalization processing; the larger the maximum similarity value max(P), the larger the adaptive learning rate function. The smaller the value, the better the mass spectrum w of the said class center. k Update level The smaller; If the maximum similarity value is less than the set threshold, the mass spectrum of the particulate matter is added to the center mass spectrum matrix as a new center mass spectrum, the number of center mass spectra in the center mass spectrum matrix is ​​updated, and the column number of the mass spectrum of the particulate matter in the center mass spectrum matrix is ​​updated. Among them, the higher the similarity between particulate matter and the assigned class center, the smaller the update degree of the mass spectrum of that class center in the current round of class center mass spectrometry matrix; the lower the similarity, the greater the update degree. Based on the mass spectra of the multiple particles and the mass spectra matrix of the class centers in the next round, the multiple particles are classified in the next round.

2. The method according to claim 1, characterized in that, The process of classifying particles into classes with similar mass spectra based on their mass spectra and the current class center mass spectra matrix, thus completing this round of classification, includes: Based on the mass spectra of the particles and the mass spectra of the current class centers, the multiple particles are classified one by one into class centers with similar mass spectra. In this round of classification, the mass spectrometry matrix of the class center used to classify the first particulate matter was obtained based on the initial mass spectra of each class center in this round. The method for obtaining the class center mass spectrometry matrix used to classify non-first particles is as follows: Based on the mass spectrum of the previous particle, the mass spectrum of the class center to which the previous particle was assigned in the current round of class center mass spectrum matrix is ​​updated; the updated class center mass spectrum matrix is ​​used as the class center mass spectrum matrix used when classifying the particle; the higher the similarity between the previous particle and the class center to which the previous particle was assigned, the smaller the degree of update; the lower the similarity between the previous particle and the class center to which the previous particle belonged, the greater the degree of update. The process involves updating the class center mass spectrum matrix for the current round based on the mass spectra of particles assigned to class centers and the similarity between particles and class centers, to obtain the class center mass spectrum matrix for the next round. This includes: After classifying the last particulate matter, the mass spectra of the last particulate matter are used to update the mass spectra of the class centers in the current round of mass spectra matrix to which the last particulate matter is classified; the updated class center mass spectra matrix is ​​then used as the class center mass spectra matrix for the next round.

3. The method according to claim 1, characterized in that, The process of classifying particles into classes with similar mass spectra based on their mass spectra and the current class center mass spectra matrix, thus completing this round of classification, includes: Based on the mass spectra of the particles and the mass spectra of the current class centers, the multiple particles are classified one by one into class centers with similar mass spectra. When this round is not the first round, the class center mass spectrometry matrix used to classify each particulate matter remains unchanged during this round of classification. The process involves updating the class center mass spectrum matrix for the current round based on the mass spectra of particles assigned to class centers and the similarity between particles and class centers, to obtain the class center mass spectrum matrix for the next round. This includes: After completing the classification of the last particulate matter, the mass spectra of each type of center in the current round of the center-class mass spectra matrix are updated based on the average mass spectra of all particulate matter under each type of center and the average similarity between all particulate matter under each type of center and its respective center, so as to obtain the center-class mass spectra matrix for the next round. Among them, the higher the average similarity between all particles under each type of center and their respective type of center, the smaller the update degree of the mass spectrometry of each type of center in this round of center mass spectrometry matrix, the lower the similarity, and the greater the update degree.

4. The method according to claim 2 or 3, characterized in that, Based on the mass spectra of the particles and the mass spectra of the current class centers, the multiple particles are individually classified into class centers with similar mass spectra, including: The mass spectrum of one particle is randomly obtained from the plurality of particles; The mass spectrum of the particulate matter was compared with the mass spectra of each type of center in the current center mass spectrum matrix to obtain the similarity between the particulate matter and each type of center. Obtain the maximum similarity among the similarities between the particulate matter and various centers; If the maximum similarity is greater than a set threshold, then the particulate matter is determined to be mass spectrometrically similar to the class center corresponding to the maximum similarity. This particulate matter is classified under this category center.

5. The method according to claim 1, characterized in that, The first round of center-like mass spectrometry matrix transformation involves changing from empty to non-empty, and the transformation process includes: In the first round of particle-by-particle segmentation, the mass spectrum of the first segmented particle is directly added to the empty matrix of the first round of center mass spectra as the center mass spectrum, thus obtaining the non-empty matrix of the first round of center mass spectra. When classifying non-first particles, the mass spectra of the non-first particles are compared with the mass spectra of the non-empty matrix of the class center mass spectra; If the mass spectrum of the non-first particle being classified is dissimilar to the mass spectrum of the class center in the non-empty matrix of the class center mass spectrum, then the mass spectrum of the non-first particle being classified is taken as a new class center mass spectrum and added to the non-empty matrix of the class center mass spectrum.

6. The method according to claim 1, characterized in that, Based on the similarity between particulate matter and its assigned class center, the update level of the mass spectrum for that class center is determined in this round of the class center mass spectrometry matrix, including: Obtain the similarity between particles and their assigned class centers, as well as the adaptive learning rate function; Substituting the similarity between the particulate matter and the assigned class center into the adaptive learning rate function, we obtain the update level of the mass spectrum of the class center in the current round of the class center mass spectrum matrix.

7. The method according to claim 1, characterized in that, Before performing the next round of classification on the multiple particulate matter based on the mass spectra of the multiple particles and the mass spectra of the class centers in the next round, the method further includes: Find the minimum similarity among all particulate matter and its class center; Based on the mass spectra of multiple particles and the mass spectra matrix of the next class center, the multiple particles are classified in the next round, including: If the minimum similarity is less than the set threshold, it is determined that the class center mass spectrometry matrix has not converged, and the multiple particles are classified in the next round based on the mass spectra of multiple particles and the class center mass spectrometry matrix of the next round.

8. A particulate matter clustering device based on the ART2A algorithm, characterized in that, The device includes: The mass spectrometry acquisition module is used to acquire the mass spectra of multiple particles and the center-of-class mass spectrometry matrix for this round. The classification module is used to classify each particulate matter into a class center with similar mass spectra based on the mass spectra of the particulate matter and the mass spectra matrix of the class centers in this round, thus completing the classification in this round. The matrix update module is used to find the mass spectrum of the class center corresponding to the maximum similarity value from the class center mass spectrum matrix of the current round based on the mass spectrum of the particles classified under the class center in the current round and the similarity between the particles and the class center. If the maximum similarity value is greater than or equal to a set threshold, the mass spectrum of the class center corresponding to the maximum similarity value is updated to obtain the class center mass spectrum matrix of the next round. The update process of the mass spectrometry of the class center corresponding to the maximum similarity value can be represented as follows: ； ； Among them, the The adaptive learning rate function is used to realize the negative correlation between the update degree of the mass spectrometry of the class center and the similarity; max(P) is the maximum similarity value; m i The mass spectrum of a randomly selected particle from a sparse mass spectrum matrix, wherein the sparse mass spectrum matrix consists of mass spectra of multiple particles; the w k The mass spectra of the class centers corresponding to the maximum similarity values; K, C, and L are fixed values; For normalization processing; the larger the maximum similarity value max(P), the larger the adaptive learning rate function. The smaller the value, the better the mass spectrum w of the said class center. k Update level The smaller; If the maximum similarity value is less than the set threshold, the mass spectrum of the particulate matter is added to the center mass spectrum matrix as a new center mass spectrum, the number of center mass spectra in the center mass spectrum matrix is ​​updated, and the column number of the mass spectrum of the particulate matter in the center mass spectrum matrix is ​​updated. Among them, the higher the similarity between particulate matter and the assigned class center, the smaller the update degree of the mass spectrum of that class center in the current round of class center mass spectrometry matrix; the lower the similarity, the greater the update degree. The next-round classification module is used to classify the multiple particles in the next round based on their mass spectra and the next-round class center mass spectra matrix.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the method of any one of claims 1 to 7.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it implements the method of any one of claims 1 to 7.

Images