Milk powder particle uniformity intelligent detection method and device based on machine vision

Abstract

The invention relates to the technical field of machine vision, and particularly discloses a milk powder particle uniformity intelligent detection method and device based on machine vision, and the method comprises the steps: collecting a multi-modal image of a to-be-detected milk powder sample, extracting basic characteristic data of the to-be-detected milk powder sample, carrying out particle segmentation and adhesion analysis, and obtaining an adhesion signal of the to-be-detected milk powder sample; the method comprises the following steps: constructing a milk powder particle uniformity analysis model, determining an initial grade of milk powder particle uniformity, obtaining production parameters of a to-be-detected milk powder sample, outputting an updated grade of milk powder particle uniformity, carrying out dissolution reaction detection and thermal reaction detection on the to-be-detected milk powder sample, and outputting a milk powder particle reaction uniformity grade in combination with a BP neural network model. And obtaining a multi-dimensional uniformity evaluation report, and determining a milk powder particle regulation and control instruction. The problems that traditional detection depends on manual operation, efficiency is low, subjectivity is high, errors are large, and the requirements of modern production lines for high precision and high efficiency of milk powder quality control are difficult to meet are solved.

Description

Technical Field

[0001] The present invention relates to the technical field of machine vision, and in particular to a method and device for intelligently detecting the uniformity of milk powder particles based on machine vision. Background Art

[0002] Currently, the uniformity of milk powder particles is a key indicator affecting product quality. It not only affects the solubility and taste of milk powder, but also the uniformity of processes such as thermal reactions and dissolution reactions. With the development of machine vision technology, it can achieve automated detection through image acquisition, processing, and analysis, providing a way to overcome traditional detection challenges. Therefore, intelligent detection methods for milk powder particle uniformity based on machine vision have become a research hotspot, aiming to achieve high-precision, high-efficiency, and objective detection of milk powder particle uniformity to ensure milk powder product quality.

[0003] Nowadays, there are still some shortcomings in the research on intelligent detection of milk powder particle uniformity. Specifically, traditional detection relies on manual operations, such as screening and microscopic observation, which have problems of low efficiency, strong subjectivity and large errors. It is difficult to meet the high precision and high efficiency requirements of modern production lines for milk powder quality control. Summary of the Invention

[0004] In response to the deficiencies in the prior art, the present invention provides a method and device for intelligent detection of milk powder particle uniformity based on machine vision, which can effectively solve the problems involved in the above-mentioned background technology.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: In a first aspect, the present invention provides an intelligent detection method for uniformity of milk powder particles based on machine vision, comprising the following steps: collecting multimodal images of milk powder samples to be detected, and performing preprocessing to extract basic feature data of the milk powder samples to be detected; performing particle segmentation and adhesion analysis based on the preprocessed multimodal images of the milk powder samples to be detected to obtain adhesion signals of the milk powder samples to be detected; constructing a milk powder particle uniformity analysis model based on the basic feature data of the milk powder samples to be detected, and determining the initial grade of milk powder particle uniformity in combination with the adhesion signals of the milk powder samples to be detected; obtaining the adhesion signals of the milk powder samples to be detected; and obtaining the adhesion signals of the milk powder samples to be detected. Measure the production parameters of the milk powder samples, specifically including the production line parameters of the milk powder samples to be tested and the production environment parameters of the milk powder samples to be tested, and output the updated uniformity level of the milk powder particles; divide the milk powder samples to be tested into two parts, perform dissolution reaction detection and thermal reaction detection on the milk powder samples to be tested respectively, and obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor; based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor, combine the BP neural network model to output the milk powder particle reaction uniformity level; based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, obtain a multi-dimensional uniformity evaluation report and determine the milk powder particle control instructions.

[0006] As a further method, multimodal images of the milk powder sample to be tested are collected and preprocessed. The specific analysis process is as follows: images of the milk powder sample to be tested are taken at four orientations of the stage: 0°, 90°, 180°, and 270° to obtain static images of the milk powder sample to be tested; the milk powder sample to be tested is placed on a turntable, the turntable rotates at 10 r / min, and the industrial camera is photographed at a frame rate of 1000 fps, and a clear dynamic image of the milk powder sample to be tested is generated by a motion blur elimination algorithm; the noise frequency bands of the static image of the milk powder sample to be tested and the dynamic image of the milk powder sample to be tested are separated by wavelet transform, and then a denoising model based on Swim-UNet is called for adaptive denoising; the motion entropy value of the milk powder sample particles is extracted from the dynamic image of the milk powder sample to be tested, and the static background noise is eliminated. The area where the motion entropy value of the milk powder sample particles is lower than the threshold stored in the database is determined to be non-milk powder sample particles; and the preprocessed multimodal image of the milk powder sample to be tested is output.

[0007] As a further method, basic characteristic data of the milk powder samples to be tested are extracted, specifically including: static basic characteristic data of the milk powder samples to be tested, and dynamic basic characteristic data of the milk powder samples to be tested, wherein: the static basic characteristic data of the milk powder samples to be tested specifically include the maximum difference in roundness of the milk powder samples to be tested, the maximum difference in sphericity of the milk powder samples to be tested, and the maximum difference in diameter of the milk powder samples to be tested; the dynamic basic characteristic data of the milk powder samples to be tested specifically include the maximum difference in displacement of the milk powder samples to be tested, and the maximum difference in movement speed of the milk powder samples to be tested.

[0008] As a further method, particle segmentation and adhesion analysis are performed based on the preprocessed multimodal images of the milk powder sample to be tested to obtain the adhesion signal of the milk powder sample to be tested. The specific analysis process is: based on the preprocessed multimodal images of the milk powder sample to be tested, an adaptive threshold segmentation algorithm is used, and the threshold is dynamically adjusted based on the local grayscale gradient to preliminarily separate the independent particles of the milk powder sample in the preprocessed multimodal images of the milk powder sample to be tested to obtain suspected adhesion areas; the suspected adhesion areas are input into the trained lightweight convolutional neural network, and the cutting boundary lines of the adhesion particles are output to obtain the adhesion areas; the area ratio of the adhesion areas is calculated and recorded as the adhesion signal of the milk powder sample to be tested.

[0009] As a further method, a milk powder particle uniformity analysis model is constructed based on the basic characteristic data of the milk powder sample to be tested, and the initial level of milk powder particle uniformity is determined in combination with the adhesion signal of the milk powder sample to be tested. The specific analysis process is as follows: based on the basic characteristic data of the milk powder sample to be tested, a milk powder particle uniformity analysis model is constructed to obtain the basic characteristic signal of the milk powder sample to be tested;

[0010] Milk powder particle uniformity analysis model, the specific analysis process is as follows:

[0011] ;

[0012] in, is the maximum difference in circularity of the milk powder samples to be tested, is the maximum difference in sphericity of the milk powder sample to be tested, is the maximum diameter difference of the milk powder samples to be tested, is the maximum displacement difference of the milk powder sample to be tested, is the maximum difference in the movement speed of the milk powder sample to be tested, is the standard deviation of the circularity of the milk powder sample to be tested stored in the database, is the standard deviation of the sphericity of the milk powder sample to be tested stored in the database, is the standard deviation of the diameter of the milk powder samples to be tested stored in the database, is the displacement standard deviation of the milk powder sample to be tested stored in the database, is the standard deviation of the movement speed of the milk powder sample to be tested stored in the database, is the static basic characteristic signal of the milk powder sample to be tested, is the dynamic basic characteristic signal of the milk powder sample to be tested, is the basic characteristic signal of the milk powder sample to be tested, Stored in the database The weight factor, Stored in the database The weight factor, is the compensation constant stored in the database, which is 1, and e is a natural constant;

[0013] The basic characteristic signal of the milk powder sample to be tested and the adhesion signal of the milk powder sample to be tested are stored as the basic uniformity label of the milk powder sample to be tested, and a mapping table of the basic uniformity label of the milk powder sample to be tested and the initial uniformity level of milk powder particles stored in the database is obtained; based on the current basic uniformity label of the milk powder sample to be tested, the matching initial uniformity level of the milk powder particles is determined.

[0014] As a further method, the production parameters of the milk powder sample to be tested are obtained, specifically including the production line parameters of the milk powder sample to be tested and the production environment parameters of the milk powder sample to be tested, and the updated uniformity level of the milk powder particles is output. The specific analysis process is: obtaining the production parameters of the milk powder sample to be tested, specifically including the production line parameters of the milk powder sample to be tested and the production environment parameters of the milk powder sample to be tested, wherein: the production line parameters of the milk powder sample to be tested specifically include the spray drying inlet temperature of the milk powder sample to be tested, the atomizer speed of the milk powder sample to be tested, and the concentration temperature of the milk powder sample to be tested; the production environment parameters of the milk powder sample to be tested specifically include the production environment humidity of the milk powder sample to be tested and the production environment temperature of the milk powder sample to be tested;

[0015] Based on the production line parameters of the milk powder samples to be tested, the production line indicators of the milk powder samples to be tested are analyzed and obtained. The specific analysis process is as follows:

[0016] ;

[0017] in, The production line indicators of the milk powder samples to be tested are: The spray drying air inlet temperature of the milk powder sample to be tested is The atomizer speed is used to produce the milk powder sample to be tested. Produce concentrated temperature for the milk powder sample to be tested, The standard production spray drying inlet air temperature of the milk powder sample to be tested stored in the database, The atomizer speed of the standard production of the milk powder sample to be tested stored in the database, The standard production concentration temperature of the milk powder sample to be tested stored in the database;

[0018] Based on the production environment parameters of the milk powder samples to be tested, the production environment indicators of the milk powder samples to be tested are analyzed. The specific analysis process is as follows:

[0019] ;

[0020] in, The production environment indicators of the milk powder samples to be tested are: The humidity of the production environment of the milk powder sample to be tested, is the production environment temperature of the milk powder sample to be tested, The standard production environment humidity of the milk powder sample to be tested stored in the database, The standard production environment temperature of the milk powder sample to be tested stored in the database;

[0021] The production line indicators of the milk powder sample to be tested and the production environment indicators of the milk powder sample to be tested are stored as the production indicators of the tested milk powder sample, and the production indicators of the tested milk powder sample stored in the database - the milk powder particle uniformity impact level are obtained. Based on the current production indicators of the tested milk powder sample, the matching milk powder particle uniformity impact level is determined; the initial milk powder particle uniformity level and the milk powder particle uniformity impact level are accumulated to obtain the updated milk powder particle uniformity level.

[0022] As a further method, the milk powder sample to be tested is divided into two parts, and the dissolution reaction test and thermal reaction test are performed on the milk powder sample to be tested respectively to obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor. The specific analysis process is: performing a dissolution reaction test on the milk powder sample to be tested to obtain dissolution reaction data of the milk powder sample to be tested, specifically including the concentration field entropy value of the dissolution reaction solution of the milk powder sample to be tested and the number of residual particles in the dissolution reaction of the milk powder sample to be tested; based on the dissolution reaction data of the milk powder sample to be tested, obtaining the dissolution reaction uniformity factor;

[0023] Perform thermal reaction testing on the milk powder samples to be tested to obtain thermal reaction data of the milk powder samples to be tested, specifically including the maximum difference in thermal reaction temperature of the milk powder samples to be tested and the maximum difference in the starting time of thermal reaction caramelization of the milk powder samples to be tested; based on the thermal reaction data of the milk powder samples to be tested, obtain the thermal reaction uniformity factor.

[0024] As a further method, based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor, the BP neural network model is combined to output the reaction uniformity grade of the milk powder particles. The specific analysis process is as follows: the dissolution reaction uniformity factor and the thermal reaction uniformity factor are used as the input of the trained BP neural network model; the number of neurons in the input layer of the BP neural network model is determined to be equal to the number of input variables, and the input variables are the dissolution reaction uniformity factor and the thermal reaction uniformity factor, which is 2; the number of neurons in the output layer is equal to the number of categories of the milk powder particle reaction uniformity grade, and the number of categories of the milk powder particle reaction uniformity grade is 3, which are 1, 2, and 3, representing high, intermediate, and low, respectively; the probability distribution vector of the category of the milk powder particle reaction uniformity grade output by the output layer is obtained; and according to the probability distribution vector, the maximum membership principle is used to determine the reaction uniformity grade of the milk powder particles.

[0025] As a further method, a multi-dimensional uniformity evaluation report is obtained based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, and the milk powder particle control instructions are determined. The specific analysis process is: based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, a multi-dimensional uniformity evaluation report is output; the milk powder particle uniformity update level and the milk powder particle reaction uniformity level are recorded as adjustment indicators, and the adjustment indicator-milk powder particle control instruction mapping table stored in the database is obtained. Based on the current adjustment indicator, the matching milk powder particle control instructions are determined, specifically including the atomization pressure adjustment value and the hot air temperature adjustment value.

[0026] The second aspect of the present invention provides an intelligent detection device for uniformity of milk powder particles based on machine vision, comprising a basic feature data extraction module, a milk powder sample adhesion signal acquisition module, a uniformity initial level determination module, a uniformity level update module, a milk powder sample module reaction detection, a reaction uniformity level output module and a milk powder particle control instruction determination module, wherein: the basic feature data extraction module is used to collect multimodal images of the milk powder sample to be detected, and perform preprocessing to extract basic feature data of the milk powder sample to be detected; the milk powder sample adhesion signal acquisition module is used to perform particle segmentation and adhesion analysis based on the preprocessed multimodal images of the milk powder sample to be detected to obtain the adhesion signal of the milk powder sample to be detected; the uniformity initial level determination module is used to construct a milk powder particle uniformity analysis model based on the basic feature data of the milk powder sample to be detected, combined with the adhesion signal of the milk powder sample to be detected The signal is connected to determine the initial uniformity level of the milk powder particles; the uniformity level update module is used to obtain the production parameters of the milk powder sample to be tested, specifically including the production line parameters of the milk powder sample to be tested and the production environment parameters of the milk powder sample to be tested, and output the updated uniformity level of the milk powder particles; the milk powder sample module reaction detection is used to divide the milk powder sample to be tested into two parts, and perform dissolution reaction detection and thermal reaction detection on the milk powder sample to be tested respectively to obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor; the reaction uniformity level output module is used to output the reaction uniformity level of the milk powder particles based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor in combination with the BP neural network model; the milk powder particle control instruction determination module is used to obtain a multi-dimensional uniformity evaluation report based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, and determine the milk powder particle control instruction.

[0027] Compared with the prior art, the embodiments of the present invention have at least the following advantages or beneficial effects:

[0028] The present invention provides an intelligent detection method and device for milk powder particle uniformity based on machine vision, multimodal image acquisition and preprocessing, which can comprehensively capture the characteristics of milk powder particles, build an analysis model based on basic feature data and incorporate adhesion signals, so that the initial grade determination is more accurate; the introduction of production parameters can eliminate interference in the production process, so that the updated grade is more in line with the actual production situation; the sample is split for dissolution and thermal reaction detection, and the obtained uniformity factor can reflect the performance of the particles in actual application, and then the reaction uniformity grade is output through the BP neural network, which improves the objectivity of the evaluation; finally, the multi-dimensional grades are integrated to generate an evaluation report and determine the control instructions, which not only realizes the full-chain detection from the particles themselves to the production links to the application reactions, but also takes into account the high precision, intelligence and practicality of the detection, overcomes the disadvantages of low efficiency and strong subjectivity of traditional detection, and provides more comprehensive and reliable technical support for milk powder production quality control. BRIEF DESCRIPTION OF THE DRAWINGS

[0029] The present invention is further described with reference to the accompanying drawings. However, the embodiments in the accompanying drawings do not constitute any limitation to the present invention. A person skilled in the art can obtain other drawings based on the following drawings without creative effort.

[0030] Figure 1 Schematic diagram of the method steps of the present invention.

[0031] Figure 2 Schematic diagram of the device module connection of the present invention. DETAILED DESCRIPTION

[0032] The technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of the present invention.

[0033] Reference Figure 1 As shown, the first aspect of the present invention provides an intelligent detection method for uniformity of milk powder particles based on machine vision, comprising: collecting multimodal images of the milk powder sample to be detected, performing preprocessing, and extracting basic feature data of the milk powder sample to be detected.

[0034] The specific analysis process is as follows: images of the milk powder sample to be tested are taken at four angles of the stage: 0°, 90°, 180°, and 270°, to obtain static images of the milk powder sample to be tested; the milk powder sample to be tested is placed on a turntable, the turntable rotates at 10r / min, and the industrial camera is used to shoot at a frame rate of 1000fps, and a clear dynamic image of the milk powder sample to be tested is generated through a motion blur elimination algorithm; the noise frequency bands of the static image and the dynamic image of the milk powder sample to be tested are separated by wavelet transform, and then the denoising model based on Swim-UNet is called for adaptive denoising; the motion entropy value of the milk powder sample particles is extracted from the dynamic image of the milk powder sample to be tested, and the static background noise is eliminated. The areas where the motion entropy value of the milk powder sample particles is lower than the threshold stored in the database are determined to be non-milk powder sample particles; and the preprocessed multimodal image of the milk powder sample to be tested is output.

[0035] Extract the basic characteristic data of the milk powder samples to be tested, specifically including: static basic characteristic data of the milk powder samples to be tested, and dynamic basic characteristic data of the milk powder samples to be tested, wherein: the static basic characteristic data of the milk powder samples to be tested specifically include the maximum difference in roundness of the milk powder samples to be tested, the maximum difference in sphericity of the milk powder samples to be tested, and the maximum difference in diameter of the milk powder samples to be tested; the dynamic basic characteristic data of the milk powder samples to be tested specifically include the maximum difference in displacement of the milk powder samples to be tested, and the maximum difference in movement speed of the milk powder samples to be tested.

[0036] Multi-directional static image capture comprehensively captures static particle characteristics, while dynamic image capture combined with a motion blur removal algorithm accurately records particle motion. This combination enriches and comprehensively enhances image information. Wavelet transform noise separation combined with the Swim-UNet denoising model specifically eliminates different types of noise, while motion entropy screening precisely eliminates non-particle areas, significantly improving image quality and laying a solid foundation for subsequent analysis. Extracted static features (circularity, sphericity, and maximum diameter difference) and dynamic features (displacement and maximum velocity difference) quantify particle differences across multiple dimensions, from shape and size to motion. These features are directly linked to core uniformity indicators, ensuring that the feature data effectively reflects particle uniformity and provides precise input for subsequent grade determination, ensuring high accuracy and comprehensiveness from the very beginning.

[0037] Based on the preprocessed multimodal image of the milk powder sample to be tested, particle segmentation and adhesion analysis are performed to obtain the adhesion signal of the milk powder sample to be tested.

[0038] The specific analysis process is as follows: based on the preprocessed multimodal image of the milk powder sample to be tested, an adaptive threshold segmentation algorithm is used to dynamically adjust the threshold based on the local grayscale gradient to preliminarily separate the independent particles of the milk powder sample in the preprocessed multimodal image of the milk powder sample to be tested, and obtain the suspected adhesion area; the suspected adhesion area is input into the trained lightweight convolutional neural network, and the cutting boundary line of the adhesion particles is output to obtain the adhesion area; the area ratio of the adhesion area is calculated and recorded as the adhesion signal of the milk powder sample to be tested.

[0039] The adaptive threshold segmentation algorithm can dynamically adjust the threshold according to the local grayscale gradient, accurately separate independent particles, reduce misclassification caused by uneven lighting or grayscale differences in particles, and make the identification of suspected adhesion areas more reliable; the suspected adhesion area is input into the trained lightweight convolutional neural network, and with the help of the network's precise learning of adhesion features, it can efficiently output the cutting boundary line and accurately define the adhesion area. Compared with traditional algorithms, it is more adaptable to complex adhesion forms; by calculating the area ratio of the adhesion area, the adhesion signal is obtained, and the degree of particle adhesion is quantified. This signal directly reflects the important negative factor of particle uniformity and provides a key basis for subsequent uniformity level judgment. It not only ensures segmentation accuracy, but also takes into account processing efficiency through the lightweight network, making adhesion analysis more accurate, efficient and in line with actual detection needs.

[0040] Based on the basic characteristic data of the milk powder samples to be tested, a milk powder particle uniformity analysis model is constructed. Combined with the adhesion signal of the milk powder samples to be tested, the initial grade of milk powder particle uniformity is determined.

[0041] The specific analysis process is as follows: based on the basic characteristic data of the milk powder sample to be tested, a milk powder particle uniformity analysis model is constructed to obtain the basic characteristic signal of the milk powder sample to be tested;

[0042] Milk powder particle uniformity analysis model, the specific analysis process is as follows:

[0043] ;

[0044] in, is the maximum difference in circularity of the milk powder samples to be tested, is the maximum difference in sphericity of the milk powder sample to be tested, is the maximum diameter difference of the milk powder samples to be tested, is the maximum displacement difference of the milk powder sample to be tested, is the maximum difference in the movement speed of the milk powder sample to be tested, is the standard deviation of the circularity of the milk powder sample to be tested stored in the database, is the standard deviation of the sphericity of the milk powder sample to be tested stored in the database, is the standard deviation of the diameter of the milk powder samples to be tested stored in the database, is the displacement standard deviation of the milk powder sample to be tested stored in the database, is the standard deviation of the movement speed of the milk powder sample to be tested stored in the database, is the static basic characteristic signal of the milk powder sample to be tested, is the dynamic basic characteristic signal of the milk powder sample to be tested, is the basic characteristic signal of the milk powder sample to be tested, Stored in the database The weight factor, Stored in the database The weight factor, is the compensation constant stored in the database, which is 1, and e is a natural constant;

[0045] The basic characteristic signal of the milk powder sample to be tested and the adhesion signal of the milk powder sample to be tested are stored as the basic uniformity label of the milk powder sample to be tested, and a mapping table of the basic uniformity label of the milk powder sample to be tested and the initial uniformity level of milk powder particles stored in the database is obtained; based on the current basic uniformity label of the milk powder sample to be tested, the matching initial uniformity level of the milk powder particles is determined.

[0046] Formulas are used to correlate the static circularity, sphericity, and maximum diameter difference, and the dynamic displacement and maximum velocity difference, with the standard deviation values ​​in the database. Weighting factors and compensation constants are also introduced to comprehensively calculate basic characteristic signals from both static and dynamic dimensions. This allows for comprehensive and accurate quantification of the uniformity characteristics of milk powder particles, fully leveraging the value of basic characteristic data. This key information is then combined with adhesion signals to form a basic uniformity label. Initial grades are then quickly matched using a mapping table in the database. This not only enables a scientific and systematic analysis of particle uniformity, but also leverages existing data experience to make grade determinations efficient and well-founded. From data processing to grade determination, complex uniformity testing is standardized and intelligent, laying a solid foundation for subsequent, more accurate quality assessments and production control, making the entire testing process more scientific and practical.

[0047] Obtain the production parameters of the milk powder sample to be tested, including the production line parameters and the production environment parameters of the milk powder sample to be tested, and output the updated milk powder particle uniformity level.

[0048] The specific analysis process is as follows: obtaining the production parameters of the milk powder sample to be tested, specifically including the production line parameters of the milk powder sample to be tested and the production environment parameters of the milk powder sample to be tested, wherein: the production line parameters of the milk powder sample to be tested specifically include the spray drying air inlet temperature of the milk powder sample to be tested, the atomizer speed of the milk powder sample to be tested, and the concentration temperature of the milk powder sample to be tested; the production environment parameters of the milk powder sample to be tested specifically include the production environment humidity of the milk powder sample to be tested and the production environment temperature of the milk powder sample to be tested;

[0049] Based on the production line parameters of the milk powder samples to be tested, the production line indicators of the milk powder samples to be tested are analyzed and obtained. The specific analysis process is as follows:

[0050] ;

[0051] in, The production line indicators of the milk powder samples to be tested are: The spray drying air inlet temperature of the milk powder sample to be tested is The atomizer speed is used to produce the milk powder sample to be tested. Produce concentrated temperature for the milk powder sample to be tested, The standard production spray drying inlet air temperature of the milk powder sample to be tested stored in the database, The atomizer speed of the standard production of the milk powder sample to be tested stored in the database, The standard production concentration temperature of the milk powder sample to be tested stored in the database;

[0052] Based on the production environment parameters of the milk powder samples to be tested, the production environment indicators of the milk powder samples to be tested are analyzed. The specific analysis process is as follows:

[0053] ;

[0054] in, The production environment indicators of the milk powder samples to be tested are: The humidity of the production environment of the milk powder sample to be tested, is the production environment temperature of the milk powder sample to be tested, The standard production environment humidity of the milk powder sample to be tested stored in the database, The standard production environment temperature of the milk powder sample to be tested stored in the database;

[0055] The production line indicators of the milk powder sample to be tested and the production environment indicators of the milk powder sample to be tested are stored as the production indicators of the tested milk powder sample, and the production indicators of the tested milk powder sample stored in the database - the milk powder particle uniformity impact level are obtained. Based on the current production indicators of the tested milk powder sample, the matching milk powder particle uniformity impact level is determined; the initial milk powder particle uniformity level and the milk powder particle uniformity impact level are accumulated to obtain the updated milk powder particle uniformity level.

[0056] Key production conditions such as the spray drying air inlet temperature, atomizer speed, concentration temperature, humidity, and temperature of the production environment will affect the formation of milk powder particles. By comparing actual production parameters with database standard values ​​and calculating production line indicators and production environment indicators, we can quantify the impact of production conditions on particle uniformity. We can use production indicators to match the impact level, and finally add it to the initial level to get an updated level. Based on the original characteristics of the particles, we can also make up for the impact of production process factors. This allows uniformity assessment to no longer focus only on finished particles, but to be linked to the entire production process. It can more truly reflect the actual production situation, correct the initial level, and make the final uniformity level more consistent with production logic. This provides a more comprehensive and accurate basis for subsequent quality traceability and production regulation, helping companies better control the uniformity of milk powder particles from the production end.

[0057] It should be noted that the reason why the production line index of the milk powder sample to be tested and the production environment index of the milk powder sample to be tested in this embodiment both adopt linear combinations is as follows:

[0058] During milk powder production, the mechanisms by which production line parameters (such as spray drying temperature and atomizer speed) and environmental parameters (such as humidity and temperature) influence particle uniformity are complex, involving multiple physical processes such as heat transfer, mass transfer, and phase change. Nonlinear coupling between parameters (such as excessively high spray temperatures causing particle breakage) does exist. However, in engineering practice, linear combinations are a reasonable simplification of these complex relationships:

[0059] Quality control on production lines requires fast calculations and real-time feedback. Linear formulas (such as production line indicators and production environment indicators for milk powder samples to be tested) are simple to calculate and can be solved directly in real time using sensor data, meeting the efficiency requirements of the production line. Furthermore, nonlinear coupling relationships often require extensive experimental data to establish complex models (such as neural networks). However, the core of this embodiment is to quantify the impact of parameters by the degree to which they deviate from their standard values. Linear combinations can intuitively reflect the contribution of each parameter's deviation from the standard value, making it easier for operators to understand and adjust.

[0060] The essence of the linear combination in this example is "deviation accumulation": the deviation (the ratio of the difference from the standard value) of each parameter (such as spray temperature and atomizer speed) is considered an independent influencing factor, and the overall index is obtained by linear addition. This method assumes that the more the parameter deviates from the standard value, the greater the negative impact on particle uniformity, and the deviations of multiple parameters will have a cumulative effect. For example, too high a spray drying temperature and too low an atomizer speed may lead to overly fine and overly large particles, respectively. The negative effects of these two are independent and additive. The linear combination can directly reflect this cumulative effect without the need for in-depth analysis of the nonlinear coupling mechanism between parameters.

[0061] The production parameter indicators are ultimately used to match the "milk powder particle uniformity impact level" (achieved through a mapping table in the database). The indicator obtained by linear combination is a continuous quantitative value, which can be divided into different levels by preset thresholds to achieve the conversion from parameter deviation to impact level: if a nonlinear model is used, the mapping relationship between indicators and levels may become complex and unstable, while the monotonicity of linear combination (when the parameter deviation increases, the indicator value increases monotonically) can ensure the consistency and interpretability of the level division. For example, when or When it increases, the corresponding "impact level" will inevitably increase. The linear relationship can ensure the stability of this logic, which is consistent with the intuitive understanding in production that "the more the parameters deviate from the standard, the more significant the impact."

[0062] The milk powder sample to be tested is divided into two parts, and the dissolution reaction test and the thermal reaction test are performed on the milk powder sample to be tested respectively to obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor.

[0063] The specific analysis process is as follows: performing a dissolution reaction test on the milk powder sample to be tested, obtaining dissolution reaction data of the milk powder sample to be tested, specifically including the concentration field entropy value of the dissolution reaction solution of the milk powder sample to be tested and the number of residual particles in the dissolution reaction of the milk powder sample to be tested; and obtaining a dissolution reaction uniformity factor based on the dissolution reaction data of the milk powder sample to be tested;

[0064] Dissolution reaction uniformity factor, the specific analysis process is:

[0065] ;

[0066] in, is the dissolution reaction uniformity factor, is the concentration field entropy value of the dissolution reaction solution of the milk powder sample to be tested, is the number of residual particles in the dissolution reaction of the milk powder sample to be tested, is the concentration field entropy value of the standard solution for the dissolution reaction of the milk powder sample to be tested stored in the database, is the number of standard residual particles in the dissolution reaction of the milk powder sample to be tested stored in the database, and e is a natural constant;

[0067] Performing thermal reaction testing on the milk powder samples to be tested to obtain thermal reaction data of the milk powder samples to be tested, specifically including the maximum difference in thermal reaction temperature of the milk powder samples to be tested and the maximum difference in the caramelization start time of the thermal reaction of the milk powder samples to be tested; and obtaining a thermal reaction uniformity factor based on the thermal reaction data of the milk powder samples to be tested;

[0068] Thermal reaction uniformity factor, the specific analysis process is:

[0069] ;

[0070] in, is the dissolution reaction uniformity factor, is the maximum difference in thermal reaction temperature of the milk powder sample to be tested, The maximum difference in the caramelization starting time of the thermal reaction of the milk powder sample to be tested, is the reference difference of the thermal reaction temperature of the milk powder sample to be tested stored in the database, is the reference difference of the thermal reaction caramelization starting time of the milk powder sample to be tested stored in the database, and e is a natural constant.

[0071] By conducting dissolution and thermal reaction tests on binary samples, we can identify key links in actual milk powder application scenarios. Dissolution testing focuses on the entropy value of the concentration field and the number of residual particles, while thermal reaction testing focuses on the maximum temperature difference and the maximum difference in caramelization start time, breaking through the limitations of traditional detection based only on the characteristics of the particle itself and extending to the application performance dimension. The actual reaction data is coupled with the standard value of the database using a formula to generate dissolution and thermal reaction uniformity factors, building a bridge to link particle uniformity with application performance. The detection system covers the entire post-production application chain, accurately quantifies the impact of particle uniformity on dissolution consistency and thermal reaction synchronization, provides a key basis for milk powder quality assessment that is close to real-world usage scenarios, supports production process optimization and product quality control, enhances the practicality and integrity of the detection system, and adapts to the dairy industry's professional needs for quality traceability and performance verification.

[0072] Based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor, the BP neural network model is combined to output the reaction uniformity grade of the milk powder particles.

[0073] The specific analysis process is as follows: the dissolution reaction uniformity factor and the thermal reaction uniformity factor are used as the input of the trained BP neural network model; the number of neurons in the input layer of the BP neural network model is determined to be equal to the number of input variables, which are 2, namely the dissolution reaction uniformity factor and the thermal reaction uniformity factor; the number of neurons in the output layer is equal to the number of categories of milk powder particle reaction uniformity level, which is 3, namely 1, 2, and 3, representing high, medium, and low levels respectively; the probability distribution vector of the category of milk powder particle reaction uniformity level output by the output layer is obtained; and the maximum membership principle is used to determine the milk powder particle reaction uniformity level based on the probability distribution vector.

[0074] The advantage of using a BP neural network to process dissolution and thermal reaction uniformity factors is that these two factors accurately characterize the application reaction characteristics of milk powder and serve as network input, meeting actual quality assessment needs. The structure of the input layer (two neurons corresponding to the two types of factors) and the output layer (three neurons corresponding to high, medium, and low levels) is clearly defined, and an adaptive model framework is constructed. By using network calculations to output the probability distribution of levels and then determining the level based on the maximum membership, the powerful nonlinear fitting capabilities of the neural network are utilized to explore the complex relationship between factors and reaction uniformity levels. Furthermore, through probability and membership rules, the level determination is quantified, intelligent, and standardized, making the uniformity assessment of the reaction dimension more accurate and efficient, providing reliable application performance level conclusions for multi-dimensional control of milk powder quality, and supporting in-depth analysis of product quality and production optimization decisions.

[0075] Based on the milk powder particle uniformity update level and milk powder particle reaction uniformity level, a multi-dimensional uniformity evaluation report is obtained to determine the milk powder particle control instructions.

[0076] The specific analysis process is as follows: output a multi-dimensional uniformity evaluation report based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level; record the milk powder particle uniformity update level and the milk powder particle reaction uniformity level as adjustment indicators, obtain the adjustment indicator-milk powder particle control instruction mapping table stored in the database, and determine the matching milk powder particle control instructions based on the current adjustment indicators, specifically including the atomization pressure adjustment value and the hot air temperature adjustment value.

[0077] The multi-dimensional uniformity assessment report integrates the uniformity of the particles themselves and the uniformity of the application reaction. It can fully present the comprehensive performance of milk powder particles in the production process and actual application, providing a panoramic perspective for quality analysis. By using these two levels as adjustment indicators, the specific atomization pressure and hot air temperature adjustment values ​​are matched through the database mapping table, so that the control instructions are directly linked to the core production parameters, realizing a precise closed loop from quality assessment to production optimization. It not only ensures the comprehensiveness and systematicness of the assessment, but also makes the control based on evidence and clear targets, effectively opening up the "testing-assessment-control" link, helping the production end to quickly respond to quality needs and improve the professionalism and effectiveness of milk powder particle uniformity control.

[0078] Reference Figure 2 As shown, the second aspect of the present invention provides an intelligent detection device for uniformity of milk powder particles based on machine vision, including a basic feature data extraction module, a milk powder sample adhesion signal acquisition module, a uniformity initial level determination module, a uniformity level update module, a milk powder sample module reaction detection, a reaction uniformity level output module and a milk powder particle control instruction determination module.

[0079] The basic feature data extraction module is used to collect multimodal images of the milk powder samples to be tested, perform preprocessing, and extract basic feature data of the milk powder samples to be tested.

[0080] The milk powder sample adhesion signal acquisition module is used to perform particle segmentation and adhesion analysis based on the preprocessed multimodal image of the milk powder sample to be tested, and obtain the adhesion signal of the milk powder sample to be tested.

[0081] The module for determining the initial uniformity level is used to construct a milk powder particle uniformity analysis model based on the basic characteristic data of the milk powder sample to be tested, and to determine the initial uniformity level of the milk powder particles in combination with the adhesion signal of the milk powder sample to be tested.

[0082] The uniformity level update module is used to obtain the production parameters of the milk powder sample to be tested, including the production line parameters of the milk powder sample to be tested and the production environment parameters of the milk powder sample to be tested, and output the updated uniformity level of the milk powder particles.

[0083] The milk powder sample module reaction detection is used to divide the milk powder sample to be tested into two parts, perform dissolution reaction detection and thermal reaction detection on the milk powder sample to be tested respectively, and obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor.

[0084] The reaction uniformity level output module is used to output the reaction uniformity level of milk powder particles based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor in combination with the BP neural network model.

[0085] The milk powder particle control instruction determination module is used to obtain a multi-dimensional uniformity evaluation report based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, and determine the milk powder particle control instruction.

[0086] The above content is merely an example and explanation of the structure of the present invention. Those skilled in the art may make various modifications or additions to the described specific embodiments or replace them in a similar manner. As long as they do not deviate from the structure of the invention or exceed the scope defined by the claims, they should all fall within the scope of protection of the present invention.

Claims

1. An intelligent detection method for milk powder particle uniformity based on machine vision, characterized in that: The following steps are involved: Collect multimodal images of the milk powder samples to be tested, perform preprocessing, and extract basic feature data of the milk powder samples to be tested; Based on the pre-processed multimodal image of the milk powder sample to be tested, particle segmentation and adhesion analysis are performed to obtain the adhesion signal of the milk powder sample to be tested; Based on the basic characteristic data of the milk powder samples to be tested, a milk powder particle uniformity analysis model is constructed, and combined with the adhesion signal of the milk powder samples to be tested, the initial grade of milk powder particle uniformity is determined; Obtain the production parameters of the milk powder sample to be tested, including the production line parameters and the production environment parameters of the milk powder sample to be tested, and output the updated milk powder particle uniformity level; The milk powder sample to be tested is divided into two parts, and the dissolution reaction test and thermal reaction test are performed on the milk powder sample to be tested respectively to obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor; Based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor, the BP neural network model is combined to output the milk powder particle reaction uniformity grade; Based on the milk powder particle uniformity update level and milk powder particle reaction uniformity level, a multi-dimensional uniformity evaluation report is obtained to determine the milk powder particle control instructions.

2. The machine vision-based intelligent detection method for milk powder particle uniformity according to claim 1, characterized in that: The multimodal images of the milk powder sample to be tested are collected and preprocessed. The specific analysis process is as follows: Take images of the milk powder sample to be tested at four angles of the stage: 0°, 90°, 180°, and 270° to obtain static images of the milk powder sample to be tested; The milk powder sample to be tested is placed on a turntable that rotates at 10 r / min. The industrial camera shoots at a frame rate of 1000 fps and uses a motion blur elimination algorithm to generate a clear dynamic image of the milk powder sample to be tested. The static image and dynamic image of the milk powder sample to be tested are separated into noise frequency bands through wavelet transform, and then the denoising model based on Swim-UNet is called to perform adaptive denoising; Extracting the motion entropy value of milk powder sample particles from the dynamic image of the milk powder sample to be tested, eliminating static background noise, and determining areas where the motion entropy value of milk powder sample particles is lower than the threshold stored in the database as non-milk powder sample particles; Output the preprocessed multimodal image of the milk powder sample to be tested.

3. The machine vision-based intelligent detection method for uniformity of milk powder particles according to claim 2, characterized in that: The extracting of basic characteristic data of the milk powder sample to be tested specifically includes: Static basic characteristic data of the milk powder sample to be tested, and dynamic basic characteristic data of the milk powder sample to be tested, including: The static basic characteristic data of the milk powder samples to be tested specifically include the maximum difference in roundness of the milk powder samples to be tested, the maximum difference in sphericity of the milk powder samples to be tested, and the maximum difference in diameter of the milk powder samples to be tested; The dynamic basic characteristic data of the milk powder samples to be tested specifically include the maximum difference in displacement of the milk powder samples to be tested and the maximum difference in movement speed of the milk powder samples to be tested.

4. The machine vision-based intelligent detection method for milk powder particle uniformity according to claim 1, characterized in that: Based on the pre-processed multimodal image of the milk powder sample to be tested, particle segmentation and adhesion analysis are performed to obtain the adhesion signal of the milk powder sample to be tested. The specific analysis process is as follows: Based on the pre-processed multimodal image of the milk powder sample to be tested, an adaptive threshold segmentation algorithm is used to dynamically adjust the threshold based on the local grayscale gradient to preliminarily separate the independent particles of the milk powder sample in the pre-processed multimodal image of the milk powder sample to be tested, and obtain the suspected adhesion area; The suspected adhesion area is input into the trained lightweight convolutional neural network, which outputs the cutting boundary line of the adhesion particles to obtain the adhesion area; Calculate the percentage of the adhesion area and record it as the adhesion signal of the milk powder sample to be tested.

5. The machine vision-based intelligent detection method for milk powder particle uniformity according to claim 3, characterized in that: Based on the basic characteristic data of the milk powder sample to be tested, a milk powder particle uniformity analysis model is constructed, and the initial level of milk powder particle uniformity is determined in combination with the adhesion signal of the milk powder sample to be tested. The specific analysis process is as follows: Based on the basic characteristic data of the milk powder sample to be tested, a milk powder particle uniformity analysis model is constructed to obtain the basic characteristic signal of the milk powder sample to be tested; Milk powder particle uniformity analysis model, the specific analysis process is as follows: ； in, is the maximum difference in circularity of the milk powder samples to be tested, is the maximum difference in sphericity of the milk powder sample to be tested, is the maximum diameter difference of the milk powder samples to be tested, is the maximum displacement difference of the milk powder sample to be tested, is the maximum difference in the movement speed of the milk powder sample to be tested, is the standard deviation of the circularity of the milk powder sample to be tested stored in the database, is the standard deviation of the sphericity of the milk powder sample to be tested stored in the database, is the standard deviation of the diameter of the milk powder samples to be tested stored in the database, is the displacement standard deviation of the milk powder sample to be tested stored in the database, is the standard deviation of the movement speed of the milk powder sample to be tested stored in the database, is the static basic characteristic signal of the milk powder sample to be tested, is the dynamic basic characteristic signal of the milk powder sample to be tested, is the basic characteristic signal of the milk powder sample to be tested, Stored in the database The weight factor, Stored in the database The weight factor, is the compensation constant stored in the database, which is 1, and e is a natural constant; The basic characteristic signal of the milk powder sample to be tested and the adhesion signal of the milk powder sample to be tested are stored as the basic uniformity label of the milk powder sample to be tested, and a mapping table of the basic uniformity label of the milk powder sample to be tested and the initial uniformity grade of the milk powder particles stored in the database is obtained; Based on the uniformity basic label of the current milk powder sample to be tested, the matching initial grade of uniformity of the milk powder particles is determined.

6. The machine vision-based intelligent detection method for milk powder particle uniformity according to claim 1, characterized in that: The production parameters of the milk powder sample to be tested are obtained, specifically including the production line parameters of the milk powder sample to be tested, the production environment parameters of the milk powder sample to be tested, and the updated grade of uniformity of the milk powder particles is output. The specific analysis process is as follows: Obtain the production parameters of the milk powder sample to be tested, including the production line parameters and the production environment parameters of the milk powder sample to be tested, including: The production line parameters of the milk powder samples to be tested specifically include the spray drying air inlet temperature of the milk powder samples to be tested, the atomizer speed of the milk powder samples to be tested, and the concentration temperature of the milk powder samples to be tested; The production environment parameters of the milk powder sample to be tested specifically include the production environment humidity and the production environment temperature of the milk powder sample to be tested; Based on the production line parameters of the milk powder samples to be tested, the production line indicators of the milk powder samples to be tested are analyzed and obtained. The specific analysis process is as follows: ； in, The production line indicators of the milk powder samples to be tested are: The spray drying air inlet temperature of the milk powder sample to be tested is The atomizer speed is used to produce the milk powder sample to be tested. Produce concentrated temperature for the milk powder sample to be tested, The standard production spray drying inlet air temperature of the milk powder sample to be tested stored in the database, The atomizer speed of the standard production of the milk powder sample to be tested stored in the database, The standard production concentration temperature of the milk powder sample to be tested stored in the database; Based on the production environment parameters of the milk powder samples to be tested, the production environment indicators of the milk powder samples to be tested are analyzed. The specific analysis process is as follows: ； in, The production environment indicators of the milk powder samples to be tested are: The humidity of the production environment of the milk powder sample to be tested, is the production environment temperature of the milk powder sample to be tested, The standard production environment humidity of the milk powder sample to be tested stored in the database, The standard production environment temperature of the milk powder sample to be tested stored in the database; The production line index and the production environment index of the milk powder sample to be tested are stored as the production index of the milk powder sample to be tested, and the milk powder particle uniformity impact level stored in the database is obtained. Based on the current production index of the milk powder sample to be tested, the matching milk powder particle uniformity impact level is determined; The milk powder particle uniformity initial level and the milk powder particle uniformity impact level are accumulated to obtain the milk powder particle uniformity update level.

7. The machine vision-based intelligent detection method for milk powder particle uniformity according to claim 1, characterized in that: The milk powder sample to be tested is divided into two parts, and the dissolution reaction test and thermal reaction test are performed on the milk powder sample to be tested respectively to obtain the dissolution reaction uniformity factor and the thermal reaction uniformity factor. The specific analysis process is as follows: Performing a dissolution reaction test on the milk powder sample to be tested, and obtaining dissolution reaction data of the milk powder sample to be tested, specifically including the concentration field entropy value of the dissolution reaction solution of the milk powder sample to be tested and the number of residual particles in the dissolution reaction of the milk powder sample to be tested; Based on the dissolution reaction data of the milk powder sample to be tested, a dissolution reaction uniformity factor is obtained; Perform thermal reaction detection on the milk powder samples to be tested, and obtain thermal reaction data of the milk powder samples to be tested, specifically including the maximum difference in thermal reaction temperature of the milk powder samples to be tested and the maximum difference in thermal reaction caramelization starting time of the milk powder samples to be tested; Based on the thermal reaction data of the milk powder sample to be tested, the thermal reaction uniformity factor is obtained.

8. The machine vision-based intelligent detection method for uniformity of milk powder particles according to claim 7, characterized in that: The dissolution reaction uniformity factor and the thermal reaction uniformity factor are combined with the BP neural network model to output the milk powder particle reaction uniformity grade. The specific analysis process is as follows: The dissolution reaction uniformity factor and the thermal reaction uniformity factor are used as inputs of the trained BP neural network model; Determine that the number of neurons in the input layer of the BP neural network model is equal to the number of input variables, which are the dissolution reaction uniformity factor and the thermal reaction uniformity factor, which is 2; The number of neurons in the output layer is equal to the number of categories of the milk powder particle reaction uniformity level. The number of categories of the milk powder particle reaction uniformity level is 3, which are 1, 2, and 3, representing high, medium, and low levels respectively. Obtain the probability distribution vector of the category of the milk powder particle reaction uniformity level output by the output layer; According to the probability distribution vector, the maximum membership principle is adopted to determine the reaction uniformity level of milk powder particles.

9. The machine vision-based intelligent detection method for milk powder particle uniformity according to claim 1, characterized in that: The multi-dimensional uniformity evaluation report is obtained based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, and the milk powder particle control instructions are determined. The specific analysis process is as follows: Output a multi-dimensional uniformity assessment report based on the milk powder particle uniformity update level and milk powder particle reaction uniformity level; The milk powder particle uniformity update level and the milk powder particle reaction uniformity level are recorded as adjustment indicators, and the adjustment indicator-milk powder particle control instruction mapping table stored in the database is obtained. Based on the current adjustment indicators, the matching milk powder particle control instructions are determined, specifically including the atomization pressure adjustment value and the hot air temperature adjustment value.

10. An intelligent detection device for milk powder particle uniformity based on machine vision, applied to the intelligent detection method for milk powder particle uniformity based on machine vision according to any one of claims 1 to 9, characterized in that: It includes a basic feature data extraction module, a milk powder sample adhesion signal acquisition module, a uniformity initial level determination module, a uniformity level update module, a milk powder sample module reaction detection module, a reaction uniformity level output module and a milk powder particle control instruction determination module, wherein: The basic feature data extraction module is used to collect multimodal images of the milk powder sample to be tested, perform preprocessing, and extract the basic feature data of the milk powder sample to be tested; A milk powder sample adhesion signal acquisition module is used to perform particle segmentation and adhesion analysis based on the pre-processed multimodal image of the milk powder sample to be tested, and obtain the adhesion signal of the milk powder sample to be tested; The module for determining the initial uniformity level is used to construct a milk powder particle uniformity analysis model based on the basic characteristic data of the milk powder sample to be tested, and to determine the initial uniformity level of the milk powder particles in combination with the adhesion signal of the milk powder sample to be tested; The uniformity level update module is used to obtain the production parameters of the milk powder sample to be tested, including the production line parameters and the production environment parameters of the milk powder sample to be tested, and output the updated uniformity level of the milk powder particles; The milk powder sample module reaction detection is used to divide the milk powder sample to be tested into two parts, perform dissolution reaction detection and thermal reaction detection on the milk powder sample to be tested respectively, and obtain the dissolution reaction uniformity factor and thermal reaction uniformity factor; The reaction uniformity level output module is used to output the reaction uniformity level of milk powder particles based on the dissolution reaction uniformity factor and the thermal reaction uniformity factor in combination with the BP neural network model; The milk powder particle control instruction determination module is used to obtain a multi-dimensional uniformity evaluation report based on the milk powder particle uniformity update level and the milk powder particle reaction uniformity level, and determine the milk powder particle control instruction.

Images