CN116189078A

CN116189078A - Intelligent preparation method and system of water-based propylene ink

Info

Publication number: CN116189078A
Application number: CN202211696811.0A
Authority: CN
Inventors: 吴健; 田仁富; 方德建
Original assignee: Wenzhou Jinma Stationery Manufacturing Co ltd
Current assignee: Wenzhou Jinma Stationery Manufacturing Co ltd
Priority date: 2022-12-28
Filing date: 2022-12-28
Publication date: 2023-05-30

Abstract

The application discloses an intelligent preparation method and a system thereof for water-based propylene ink, wherein the intelligent preparation method comprises the steps of obtaining the adding speed values of all components of the water-based propylene mark pen ink at a plurality of preset time points in a preset time period, the rotating speed values of a stirrer at the preset time points and a stirring monitoring video of a mixture in the preset time period, adopting a deep neural network model based on deep learning, and adjusting parameters of the deep neural network model through a gradient descent back propagation algorithm to simulate complex nonlinear association among things, so as to establish a complex mapping relation among the adding speeds of all components of the water-based propylene mark pen ink, the stirring speeds of the stirrer and the mixing state change of the mixture, and enable the self-adaptive control of the stirring speeds to be carried out based on the real-time change of the added quality and the mixing state of actual raw materials. In this way, the mixing effect can be improved, and the quality of the produced ink can be further improved.

Description

Intelligent preparation method and system of water-based propylene ink

Technical Field

The application relates to the technical field of intelligent preparation, and more particularly relates to an intelligent preparation method and system of water-based propylene ink.

Background

The acrylic pigment is popular with many painters and lovers because of quick drying and bright and full color, but is inconvenient for many painting beginners because the pigment needs to be prepared and used quickly on site. In recent years, various pigment acrylic mark pens have appeared, and pigment acrylic ink is filled into the pen, so that the pen is convenient to carry and use. At present, some international brands produce acrylic mark pens with various colors, but due to the advantages of quick drying, better pigment solubility and better glossiness of the organic solvent applied in the ink, the volatile organic solvent and the organic micromolecular material are used in the ink to different degrees at present, and long-term contact easily has health influence on users.

Aiming at the problems, chinese patent application CN111117354A proposes a water-based acrylic mark pen ink and a preparation method thereof. The titanium dioxide and pigment are used, and the environment-friendly and nontoxic polymer is used as a functional interphase regulator to be matched with the film-forming emulsion, so that the ink with stable performance, smooth water outlet, uniform distribution and high color fastness is obtained.

However, the water-based acrylic marker ink prepared in the actual treatment process of the scheme is found to have poor quality and slower efficiency because the raw materials are added each time, and in the scheme, the quality of the produced ink is improved by controlling the mass percent of each component to be within a certain range and controlling the stirring speed to be within a fixed range, and the inherent nonlinear correlation between the mass percent of the raw materials and the rotating speed of the stirrer is not considered.

Thus, an optimized intelligent preparation scheme for water-based propylene inks is desired.

Disclosure of Invention

The present application has been made in order to solve the above technical problems. The embodiment of the application provides an intelligent preparation method and system of water-based propylene ink, wherein the intelligent preparation method and system acquire the adding speed values of each component of the water-based propylene mark pen ink at a plurality of preset time points in a preset time period, the rotating speed values of a stirrer at the preset time points and a stirring monitoring video of a mixture in the preset time period, a deep neural network model based on deep learning is adopted, and parameters of the deep neural network model are adjusted through a gradient descent back propagation algorithm to simulate complex nonlinear association among things, so that a complex mapping relation among the adding speed of each component of the water-based propylene mark pen ink, the stirring speed of the stirrer and the mixing state change of the mixture is established, and the stirring speed is adaptively controlled based on real-time change of the actual raw material adding quality and the mixing state. In this way, the mixing effect can be improved, and the quality of the produced ink can be further improved.

According to one aspect of the present application, there is provided an intelligent preparation method of a water-based propylene ink, comprising:

acquiring the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in a preset time period, the rotation speed values of a stirrer at the preset time points and a stirring monitoring video of a mixture in the preset time period;

arranging the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an addition speed input matrix according to the time dimension and the sample dimension, and then obtaining a component addition characteristic vector through a first convolution neural network model serving as a filter;

the rotational speed values of the mixers at a plurality of preset time points are arranged into rotational speed input vectors according to time dimensions, and then the rotational speed input vectors are processed through a multi-scale neighborhood feature extraction module to obtain rotational speed feature vectors;

stirring monitoring video of the mixture in the preset time period is processed through a second convolution neural network model using a three-dimensional convolution kernel to obtain a stirring state monitoring feature vector;

adding a feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to the components based on the Gaussian density map for feature data enhancement so as to obtain first to third Gaussian density maps;

Fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map;

performing Gaussian discretization on the posterior Gaussian density map to obtain a classification feature matrix;

performing feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix; and

and passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the stirrer at the current time point is increased or decreased.

In the above-mentioned intelligent preparation method of water-based propylene ink, the arranging the adding speed values of each component of the water-based propylene mark pen ink at a plurality of predetermined time points in the predetermined time period into an adding speed input matrix according to the time dimension and the sample dimension, and then obtaining a component adding feature vector through a first convolutional neural network model as a filter, including:

each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer:

carrying out convolution processing on the input data to obtain a convolution characteristic diagram;

carrying out mean pooling treatment based on a feature matrix on the convolution feature map to obtain a pooled feature map; and

Non-linear activation is carried out on the pooled feature map so as to obtain an activated feature map;

and the output of the last layer of the first convolutional neural network model serving as the filter adds a feature vector to the component, and the input of the first layer of the first convolutional neural network model serving as the filter is the addition speed input matrix.

In the above intelligent preparation method of water-based propylene ink, the multi-scale neighborhood feature extraction module comprises a first convolution layer and a second convolution layer which are parallel, and a multi-scale fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer respectively use one-dimensional convolution kernels with different scales.

In the above-mentioned intelligent preparation method of water-based propylene ink, the steps of arranging the rotational speed values of the stirrer at a plurality of predetermined time points into rotational speed input vectors according to a time dimension, and obtaining rotational speed feature vectors by a multi-scale neighborhood feature extraction module include:

inputting the rotating speed input vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale rotating speed feature vector, wherein the first convolution layer is provided with a first one-dimensional convolution kernel with a first length;

Inputting the rotation speed input vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale rotation speed feature vector, wherein the second convolution layer is provided with a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and

and cascading the first scale rotational speed feature vector and the second scale rotational speed feature vector by using a multi-scale fusion layer of the multi-scale neighborhood feature extraction module to obtain the rotational speed feature vector.

In the above-mentioned intelligent preparation method of water-based propylene ink,

the inputting the rotation speed input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale rotation speed feature vector, further comprises: performing one-dimensional convolution encoding on the rotating speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first-scale rotating speed feature vector;

wherein, the formula is:

wherein a is the width of the first one-dimensional convolution kernel in the X direction, F (a) is a parameter vector of the first one-dimensional convolution kernel, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the first one-dimensional convolution kernel, and X represents the rotational speed input vector;

The inputting the rotation speed input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale rotation speed feature vector, further comprising: performing one-dimensional convolution encoding on the rotating speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale rotating speed feature vector;

wherein, the formula is:

wherein b is the width of the second one-dimensional convolution kernel in the X direction, F (b) is a parameter vector of the second one-dimensional convolution kernel, G (X-b) is a local vector matrix calculated by a convolution kernel function, m is the size of the second one-dimensional convolution kernel, and X represents the rotational speed input vector.

In the above-mentioned intelligent preparation method of water-based propylene ink, the mixing monitoring video of the mixture of the predetermined time period is obtained by using a second convolution neural network model of a three-dimensional convolution kernel to obtain a mixing state monitoring feature vector, which comprises:

extracting a plurality of image key frames from the stirring monitoring video of the mixture in the preset time period, and arranging the image key frames according to the time dimension to obtain a three-dimensional input tensor;

performing three-dimensional convolution processing, mean pooling processing and nonlinear activation processing based on the three-dimensional convolution kernel on input data in forward transfer of layers through each layer of the second convolution neural network model using the three-dimensional convolution kernel to monitor a feature map by output of a last layer of the second convolution neural network model using the three-dimensional convolution kernel, wherein input of a first layer of the second convolution neural network model using the three-dimensional convolution kernel is the three-dimensional input tensor: and

And carrying out global average pooling on each feature matrix of the stirring state monitoring feature map to obtain the stirring state monitoring feature vector.

In the above method for intelligently preparing water-based propylene ink, the fusing the first to third gaussian density maps to obtain a posterior gaussian density map using a bayesian probability model includes:

fusing the first gaussian density map, the second gaussian density map, and the third gaussian density map using a bayesian probability model in the following formula to obtain the posterior gaussian density map;

wherein, the formula is:

wherein ,

representing the posterior Gaussian density map, < >>

Representing said first Gaussian density map, ">

Representing the second Gaussian density map, +.>

Representing the third gaussian density map.

In the above method for intelligently preparing water-based propylene ink, the performing feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix includes:

carrying out feature distribution correction on the classification feature matrix by using the following formula to obtain a corrected classification feature matrix;

wherein, the formula is:

wherein M_c And M is the classification characteristic matrix and the corrected classification characteristic matrix, respectively, and ReLU (·) represents a ReLU activation function,

The method is characterized in that the method comprises the steps of multiplying a matrix, dividing the matrix eigenvalues by the division between a numerator matrix and a denominator matrix, wherein exp (·) represents the exponential operation of the matrix, and the exponential operation of the matrix represents the calculation of a natural exponential function value with the eigenvalues of all positions in the matrix as powers.

In the above-mentioned intelligent preparation method of water-based propylene ink, the step of passing the corrected classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the rotational speed value of the mixer at the current time point should be increased or decreased, includes:

expanding the corrected classification characteristic matrix into classification characteristic vectors according to row vectors or column vectors;

performing full-connection coding on the classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and

and the coding classification feature vector is passed through a Softmax classification function of the classifier to obtain the classification result.

According to another aspect of the present application, there is provided an intelligent preparation system for water-based propylene ink, comprising:

the data acquisition module is used for acquiring the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in a preset time period, the rotation speed values of the stirrer at the preset time points and the stirring monitoring video of the mixture in the preset time period;

The first convolution module is used for arranging the adding speed values of the components of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an adding speed input matrix according to the time dimension and the sample dimension, and then obtaining component adding feature vectors through a first convolution neural network model serving as a filter;

the multi-scale feature extraction module is used for arranging the rotating speed values of the stirring machines at a plurality of preset time points into rotating speed input vectors according to the time dimension and then obtaining rotating speed feature vectors through the multi-scale neighborhood feature extraction module;

the second convolution module is used for obtaining a stirring state monitoring feature vector through a second convolution neural network model using a three-dimensional convolution kernel according to the stirring monitoring video of the mixture in the preset time period;

the characteristic data enhancement module is used for adding a characteristic vector, the rotating speed characteristic vector and the stirring state monitoring characteristic vector to the component based on the Gaussian density map, and carrying out characteristic data enhancement to obtain first to third Gaussian density maps;

a fusion module for fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map;

The Gaussian discretization module is used for carrying out Gaussian discretization on the posterior Gaussian density map to obtain a classification characteristic matrix;

the characteristic distribution correction module is used for carrying out characteristic distribution correction on the classified characteristic matrix to obtain a corrected classified characteristic matrix; and

and the classification result generation module is used for passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the stirrer at the current time point should be increased or decreased.

Compared with the prior art, the intelligent preparation method and the system of the water-based propylene ink provided by the application acquire the adding speed values of each component of the water-based propylene mark pen ink at a plurality of preset time points in a preset time period, the rotating speed values of the stirrer at the plurality of preset time points and the stirring monitoring video of the mixture in the preset time period, adopt a deep neural network model based on deep learning, adjust the parameters of the deep neural network model through a gradient descent back propagation algorithm to simulate complex nonlinear association among things, and thus establish a complex mapping relation among the adding speeds of each component of the water-based propylene mark pen ink, the stirring speeds of the stirrer and the mixing state change of the mixture, so that the self-adaptive control of the stirring speeds is performed based on the real-time change of the quality and the mixing state of actual raw material addition. In this way, the mixing effect can be improved, and the quality of the produced ink can be further improved.

Drawings

The foregoing and other objects, features and advantages of the present application will become more apparent from the following more particular description of embodiments of the present application, as illustrated in the accompanying drawings. The accompanying drawings are included to provide a further understanding of embodiments of the application and are incorporated in and constitute a part of this specification, illustrate the application and not constitute a limitation to the application. In the drawings, like reference numerals generally refer to like parts or steps.

Fig. 1 is a schematic view of a scenario of an intelligent preparation method of a water-based propylene ink according to an embodiment of the present application.

FIG. 2 is a flow chart of a method for intelligent preparation of water-based propylene ink in accordance with an embodiment of the present application.

Fig. 3 is a schematic architecture diagram of an intelligent preparation method of water-based propylene ink according to an embodiment of the present application.

Fig. 4 is a flow chart of sub-steps of step S130 in the intelligent preparation method of the water-based propylene ink according to the embodiment of the present application.

Fig. 5 is a flow chart of sub-steps of step S140 in the intelligent preparation method of the water-based propylene ink according to the embodiment of the present application.

Fig. 6 is a flow chart of sub-steps of step S190 in the intelligent preparation method of the water-based propylene ink according to the embodiment of the present application.

FIG. 7 is a block diagram of an intelligent preparation system for water-based propylene ink in accordance with an embodiment of the present application.

Detailed Description

Hereinafter, example embodiments according to the present application will be described in detail with reference to the accompanying drawings. It should be apparent that the described embodiments are only some of the embodiments of the present application and not all of the embodiments of the present application, and it should be understood that the present application is not limited by the example embodiments described herein.

Scene overview

As described above, in the process of actually using the water-based acrylic mark pen ink and the method for preparing the same, it was found that the quality of the water-based acrylic mark pen ink prepared is poor and the efficiency is slow, because each time raw materials are added, in this scheme, only by controlling the mass percentage of each component to be added in a certain range and controlling the stirring speed to be in a fixed range, the inherent nonlinear correlation between the two is not considered, so that the mass percentage of raw materials to be added and the rotational speed of the stirrer need to be dynamically monitored and controlled in real time to improve the quality of the produced ink. Thus, an optimized intelligent preparation scheme for water-based propylene inks is desired.

Accordingly, in consideration of the fact that in the preparation process of the water-based acrylic mark pen ink, the stirring speed value of the stirrer needs to be controlled in real time based on the addition mass percentage of each component of the actual water-based acrylic mark pen ink and the mixing state change of the mixture, so that the stirring uniformity and the mixing effect of the mixture are improved, and the quality of the produced ink is improved. Further, since it is difficult to measure the mass percentages of the components of the water-based acrylic mark pen ink, the determination of the additive mass can be performed by selecting the addition rate value of the components of the water-based acrylic mark pen ink. In this process, it is difficult to establish a mapping relationship between the addition speed of each component of the water-based acrylic mark pen ink, the stirring speed of the stirrer, and the mixing state change of the mixture, so as to perform adaptive control of the stirring speed based on the actual quality of raw material addition and the real-time change of the mixing state, thereby achieving the purpose of improving the mixing effect.

In recent years, deep learning and neural networks have been widely used in the fields of computer vision, natural language processing, text signal processing, and the like. In addition, deep learning and neural networks have also shown levels approaching and even exceeding humans in the fields of image classification, object detection, semantic segmentation, text translation, and the like.

Deep learning and development of neural networks provide new solutions and solutions for mining complex mappings between the addition rate of each component of the water-based acrylic mark pen ink, the stirring rate of the stirrer, and the mixing state variation of the mixture. Those of ordinary skill in the art will appreciate that the deep-learning based deep neural network model may be tuned by appropriate training strategies, such as by a gradient-descent back-propagation algorithm, to enable it to simulate complex nonlinear correlations between things, which is obviously suitable for simulating and establishing complex mappings between the addition rates of the individual components of the water-based acrylic Mark pen ink, the stirring rate of the stirrer, and the mixing state variation of the mixture.

Specifically, in the technical scheme of the application, firstly, adding speed values of each component of the water-based propylene mark pen ink at a plurality of preset time points in a preset time period, rotating speed values of a stirrer at the preset time points and stirring monitoring videos of a mixture in the preset time period are obtained. Next, for the addition speed values of the respective components of the water-based acrylic mark pen ink at the respective predetermined time points, time-series implicit correlation feature mining of the addition speed values of the respective components is performed using a convolutional neural network model having excellent performance in implicit correlation feature extraction in consideration of feature distribution information having correlation in both the time dimension and the sample dimension. Specifically, the adding speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period are arranged into an adding speed input matrix according to the time dimension and the sample dimension, and then feature extraction is carried out in a first convolution neural network model serving as a filter, so that relevance feature distribution information of the adding speed of each component in the time dimension and the sample dimension is extracted, and a component adding feature vector is obtained.

Then, regarding the rotational speed values of the stirrer at the plurality of preset time points, considering that the rotational speed values of the stirrer have fluctuation in time dimension and have different mode dynamic change characteristics under different time period spans in the preset time period, in order to accurately extract the dynamic change implicit characteristics of the rotational speed of the stirrer, so as to extract implicit mapping related characteristics of the addition speed of each component and the change of the stirring mixing state.

Then, for the agitation monitoring video of the mixture for the predetermined period of time, since the agitation monitoring video is composed of a plurality of agitation monitoring frames at a plurality of time points, it has not only an implicit characteristic of the agitation state of the mixture in each monitoring frame but also a dynamic variation characteristic in the time dimension. Therefore, in the technical scheme of the application, feature mining is performed on the stirring monitoring video of the mixture in the preset time period in a second convolution neural network model by using a three-dimensional convolution kernel, so that dynamic change feature distribution information of the state feature of the mixture in the stirring monitoring video of the mixture in the time dimension is extracted, and therefore the stirring state monitoring feature vector is obtained. In particular, the convolution kernel of the second convolutional neural network model is a three-dimensional convolution kernel, which has W (width), H (height) and C (channel dimension), and in the technical solution of the present application, the channel dimension of the three-dimensional convolution kernel corresponds to the time dimension of the stirring monitoring video of the mixture, so that when performing three-dimensional convolutional encoding, dynamic change feature information of the state feature of the mixture in the time dimension can be extracted.

Further, in order to improve the accuracy of the real-time control of the stirring speed, it is necessary to perform the line data enhancement in the high-dimensional feature space for the respective addition speed-related feature of the respective components, the rotational speed dynamic feature of the mixer, and the mixing state change feature of the mixture, considering that not only the addition speed value of the respective components, the stirring speed value of the mixer, but also the mixing state change of the mixture have fluctuation and uncertainty in the time dimension at the time of actually performing the ink manufacturing.

It should be understood that, as a learning target of the neural network model, the gaussian density map may represent a joint distribution in the case where a plurality of feature values constitute an overall distribution due to probability density thereof, that is, a feature distribution is taken as a priori distribution, and probability density under the influence of correlation with other a priori distribution positions at each a priori distribution position is obtained as a posterior distribution, thereby describing the feature distribution more accurately in a higher dimension. Therefore, in the technical solution of the present application, the data enhancement can be performed on the addition speed association characteristic of each component, the rotational speed dynamic characteristic of the mixer, and the mixing state change characteristic of the mixture by using the addition speed of each component, the rotational speed of the mixer, and the mixing state of the mixture as prior distributions, i.e., gaussian distributions, respectively. Specifically, the gaussian density maps of the component addition feature vector, the rotational speed feature vector, and the stirring state monitoring feature vector are respectively constructed to obtain first to third gaussian density maps.

Next, in order to be able to perform the adaptive adjustment of the rotational speed of the mixer based on the actual added mass of each component of the water-based acrylic mark pen ink and the real-time state change of the mixture, it is necessary to extract the correlation between the three components, and to generate the result of the adaptive control of the rotational speed of the mixer using the correlation feature. It should be understood that, considering that the second gaussian density map corresponding to the rotational speed feature vector is used as a prior probability vector, the technical solution of the present application aims to update the prior probability to obtain the posterior probability when new evidence, that is, when the adding speed of each component changes. Then, according to a bayesian formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability, so in the technical scheme of the application, a bayesian probability model is used for fusing the corresponding first to third gaussian density maps respectively corresponding to the component addition feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to obtain a posterior gaussian density map. Specifically, the second gaussian density map corresponding to the rotational speed feature vector is used as a priori probability, the first gaussian density map corresponding to the component addition feature vector is used as evidence probability, the third gaussian density map corresponding to the stirring state monitoring feature vector is used as event probability, and the posterior probability of real-time change of the rotational speed of the stirrer when the addition speed of each component is changed is calculated, so that the posterior gaussian density map is obtained. And then, carrying out Gaussian discretization processing on the posterior Gaussian density map so as not to generate information loss when the data features are amplified, thereby obtaining a classification feature matrix, and classifying the posterior Gaussian density map so as to improve the accuracy of subsequent classification.

That is, after the classification feature matrix is obtained, it is subjected to classification processing in a classifier to obtain a classification result indicating whether the rotational speed value of the stirrer at the current time point should be increased or decreased. That is, in the technical solution of the present application, the label of the classifier includes that the rotational speed value of the mixer at the current time point should be increased or should be decreased, wherein the classifier determines to which classification label the classification feature matrix belongs through a soft maximum function. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control policy label of the rotation speed value of the mixer at the current time point, so after the classification result is obtained, the rotation speed value of the mixer at the current time point can be adaptively adjusted based on the classification result, so as to perform adaptive control of the mixing speed based on the real-time change of the quality and the mixing state of the actual raw material addition, so as to achieve the purpose of improving the mixing effect.

Particularly, in the technical scheme of the application, when feature data enhancement is performed on the component adding feature vector, the rotating speed feature vector and the stirring state feature vector based on the Gaussian density map respectively to obtain the first to third Gaussian density maps and when Gaussian discretization is performed on the posterior Gaussian density map to obtain the classification feature matrix, due to the random characteristics of the Gaussian density map and the Gaussian discretization, negative correlation values relative to global feature distribution are introduced into local feature distribution of the classification feature matrix, so that classification accuracy of the classification feature matrix is affected.

Therefore, the applicant of the present application corrects the classification feature matrix in a non-linear re-weighting manner of full orthographic projection, expressed as:

mc and M are the classification feature matrices after and before correction, respectively, and the division between the numerator matrix and the denominator matrix is the division by position of the matrix feature values.

Here, the full orthographic projection nonlinear re-weighting ensures full orthographic projection through a ReLU function to avoid aggregation of negatively related information, and simultaneously introduces a nonlinear re-weighting mechanism to aggregate the eigenvalue distribution of the classification feature matrix, so that the internal structure of the classification feature matrix after correction can penalize long-distance connection to strengthen local coupling. Therefore, the synergistic effect of the spatial feature transformation corresponding to the full orthographic projection re-weighting of the classification feature matrix in the high-dimensional feature space is realized, so that the negative correlation value relative to the global feature distribution in the local feature distribution of the classification feature matrix is eliminated, and the classification accuracy of the classification feature matrix is improved. Therefore, the self-adaptive control of the stirring speed can be accurately performed in real time based on actual conditions, so that the mixing efficiency and effect of the mixture are improved, and the preparation quality of the water-based propylene ink is improved.

Based on this, the application provides an intelligent preparation method of water-based propylene ink, which comprises the following steps: acquiring the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in a preset time period, the rotation speed values of a stirrer at the preset time points and a stirring monitoring video of a mixture in the preset time period; arranging the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an addition speed input matrix according to the time dimension and the sample dimension, and then obtaining a component addition characteristic vector through a first convolution neural network model serving as a filter; the rotational speed values of the mixers at a plurality of preset time points are arranged into rotational speed input vectors according to time dimensions, and then the rotational speed input vectors are processed through a multi-scale neighborhood feature extraction module to obtain rotational speed feature vectors; stirring monitoring video of the mixture in the preset time period is processed through a second convolution neural network model using a three-dimensional convolution kernel to obtain a stirring state monitoring feature vector; adding a feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to the components based on the Gaussian density map for feature data enhancement so as to obtain first to third Gaussian density maps; fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map; performing Gaussian discretization on the posterior Gaussian density map to obtain a classification feature matrix; performing feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix; and passing the corrected classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the stirrer at the current time point is increased or decreased.

Fig. 1 is an application scenario diagram of an intelligent preparation method of a water-based propylene ink according to an embodiment of the present application. As shown in fig. 1, in this application scenario, the addition speed values of the respective components of the water-based propylene marker ink at a plurality of predetermined time points within a predetermined period of time (e.g., D1 as illustrated in fig. 1), the rotation speed values of the agitators at the plurality of predetermined time points (e.g., D2 as illustrated in fig. 1), and the agitation monitoring video of the mixture for the predetermined period of time (e.g., D3 as illustrated in fig. 1) are acquired, and then the addition speed values of the respective components of the water-based propylene marker ink at a plurality of predetermined time points within the predetermined period of time, the rotation speed values of the agitators at the plurality of predetermined time points, and the agitation monitoring video of the mixture for the predetermined period of time are entered into a server (e.g., S as illustrated in fig. 1) that is provided with an intelligent preparation algorithm of the water-based propylene ink, wherein the server is capable of generating a classification result indicating that the rotation speed value of the agitators at the current time point should be increased or decreased based on the intelligent preparation algorithm of the water-based propylene ink.

Having described the basic principles of the present application, various non-limiting embodiments of the present application will now be described in detail with reference to the accompanying drawings.

Exemplary method

FIG. 2 is a flow chart of a method for intelligent preparation of water-based propylene ink in accordance with an embodiment of the present application. As shown in fig. 2, the intelligent preparation method of the water-based propylene ink according to the embodiment of the application comprises the following steps: s110, obtaining the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in a preset time period, the rotation speed values of a stirrer at the preset time points and a stirring monitoring video of a mixture in the preset time period; s120, arranging the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an addition speed input matrix according to a time dimension and a sample dimension, and then obtaining component addition characteristic vectors through a first convolution neural network model serving as a filter; s130, arranging the rotation speed values of the stirring machines at a plurality of preset time points into rotation speed input vectors according to a time dimension, and then obtaining rotation speed feature vectors through a multi-scale neighborhood feature extraction module; s140, obtaining a stirring state monitoring feature vector by using a second convolution neural network model of the three-dimensional convolution kernel through stirring monitoring video of the mixture in the preset time period; s150, adding a feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to the components based on the Gaussian density map, and carrying out feature data enhancement to obtain first to third Gaussian density maps; s160, fusing the first Gaussian density map to the third Gaussian density map by using a Bayesian probability model to obtain a posterior Gaussian density map; s170, performing Gaussian discretization on the posterior Gaussian density map to obtain a classification feature matrix; s180, carrying out feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix; and S190, passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotation speed value of the stirrer at the current time point is increased or decreased.

Fig. 3 is a schematic architecture diagram of an intelligent preparation method of water-based propylene ink according to an embodiment of the present application. As shown in fig. 3, in the network architecture, first, the addition speed values of the respective components of the water-based propylene mark pen ink at a plurality of predetermined time points within a predetermined period of time, the rotation speed values of the mixer at the plurality of predetermined time points, and the mixing monitoring video of the mixture of the predetermined period of time are acquired; then, arranging the adding speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an adding speed input matrix according to the time dimension and the sample dimension, and then obtaining component adding characteristic vectors through a first convolution neural network model serving as a filter; then, arranging the rotation speed values of the stirring machines at a plurality of preset time points into rotation speed input vectors according to a time dimension, and obtaining rotation speed feature vectors through a multi-scale neighborhood feature extraction module; then, the stirring monitoring video of the mixture in the preset time period is processed through a second convolution neural network model using a three-dimensional convolution kernel to obtain a stirring state monitoring feature vector; then, adding a feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to the components based on the Gaussian density map, and carrying out feature data enhancement to obtain first to third Gaussian density maps; then, fusing the first to third Gaussian density maps by using a Bayesian probability model to obtain a posterior Gaussian density map; then, carrying out Gaussian discretization on the posterior Gaussian density map to obtain a classification feature matrix; then, carrying out feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix; and finally, the corrected classification characteristic matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the stirrer at the current time point is increased or decreased.

More specifically, in step S110, the addition speed values of the respective components of the water-based propylene mark pen ink at a plurality of predetermined time points within a predetermined period of time, the rotation speed values of the mixer at the plurality of predetermined time points, and the mixing monitor video of the mixture of the predetermined period of time are acquired. In the actual preparation process of the water-based acrylic mark pen ink, the stirring speed value of the stirrer needs to be controlled in real time based on the addition mass percent of each component of the actual water-based acrylic mark pen ink and the mixing state change of the mixture, so that the stirring uniformity and the mixing effect of the mixture are improved, and the quality of the produced ink is further improved.

More specifically, in step S120, the addition speed values of the respective components of the water-based acrylic mark pen ink at a plurality of predetermined time points within the predetermined period are arranged as an addition speed input matrix in terms of a time dimension and a sample dimension, and then the addition speed input matrix is passed through a first convolutional neural network model as a filter to obtain component addition feature vectors. Regarding the addition speed value of each component of the water-based acrylic mark pen ink at each predetermined time point, in consideration of the feature distribution information having relevance in both the time dimension and the sample dimension, the time-series implicit relevance feature mining of the addition speed value of each component is performed using a convolutional neural network model having excellent performance in the implicit relevance feature extraction. Specifically, the adding speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period are arranged into an adding speed input matrix according to the time dimension and the sample dimension, and then feature extraction is carried out in a first convolution neural network model serving as a filter, so that relevance feature distribution information of the adding speed of each component in the time dimension and the sample dimension is extracted, and a component adding feature vector is obtained.

Accordingly, in one specific example, the arranging the adding speed values of the components of the water-based acrylic mark pen ink at a plurality of predetermined time points within the predetermined time period according to the time dimension and the sample dimension into the adding speed input matrix, and then obtaining component adding feature vectors through a first convolution neural network model serving as a filter, includes: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer: carrying out convolution processing on the input data to obtain a convolution characteristic diagram; carrying out mean pooling treatment based on a feature matrix on the convolution feature map to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the first convolutional neural network model serving as the filter adds a feature vector to the component, and the input of the first layer of the first convolutional neural network model serving as the filter is the addition speed input matrix.

More specifically, in step S130, the rotational speed values of the mixers at the plurality of predetermined time points are arranged according to a time dimension to form a rotational speed input vector, and then the rotational speed input vector is passed through a multi-scale neighborhood feature extraction module to obtain a rotational speed feature vector. Considering that the rotational speed values of the stirrer at the plurality of preset time points have fluctuation in time dimension, the rotational speed values of the stirrer have different mode dynamic change characteristics under different time period spans in the preset time period, so that in order to accurately extract the dynamic change implicit characteristics of the rotational speed of the stirrer, the implicit mapping related characteristics of the addition speeds of the components and the change of the stirring mixing state are extracted, and in the technical scheme of the application, the rotational speed values of the stirrer at the plurality of preset time points are further arranged into rotational speed input vectors according to the time dimension and then are subjected to encoding processing in a multi-scale neighborhood characteristic extraction module, so that the dynamic multi-scale neighborhood related characteristics of the rotational speed values of the stirrer under different time spans in the preset time period are extracted, and the rotational speed characteristic vectors are obtained.

Accordingly, in one specific example, the multi-scale neighborhood feature extraction module includes a first convolution layer and a second convolution layer in parallel, and a multi-scale fusion layer connected to the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer each use one-dimensional convolution kernels having different scales.

Accordingly, in a specific example, as shown in fig. 4, the arranging the rotational speed values of the stirring machines at the plurality of predetermined time points into the rotational speed input vector according to the time dimension, and then obtaining the rotational speed feature vector through the multi-scale neighborhood feature extraction module, includes: s131, inputting the rotating speed input vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale rotating speed feature vector, wherein the first convolution layer is provided with a first one-dimensional convolution kernel with a first length; s132, inputting the rotating speed input vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale rotating speed feature vector, wherein the second convolution layer is provided with a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and S133, cascading the first scale rotational speed feature vector and the second scale rotational speed feature vector by using a multi-scale fusion layer of the multi-scale neighborhood feature extraction module to obtain the rotational speed feature vector.

Accordingly, in one specific example, the inputting the rotational speed input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale rotational speed feature vector further includes: performing one-dimensional convolution encoding on the rotating speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first-scale rotating speed feature vector;

wherein, the formula is:

wherein a is the width of the first one-dimensional convolution kernel in the X direction, F (a) is a parameter vector of the first one-dimensional convolution kernel, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the first one-dimensional convolution kernel, and X represents the rotational speed input vector.

Accordingly, in a specific example, the inputting the rotational speed input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale rotational speed feature vector further includes: performing one-dimensional convolution encoding on the rotating speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale rotating speed feature vector;

wherein, the formula is:

More specifically, in step S140, the agitation monitoring video of the mixture for the predetermined period of time is passed through a second convolutional neural network model using a three-dimensional convolutional kernel to obtain an agitation state monitoring feature vector. For the stirring monitoring video of the mixture in the preset time period, since the stirring monitoring video is composed of a plurality of stirring monitoring frames at a plurality of time points, the stirring monitoring video has not only implicit characteristics of the stirring state of the mixture in each monitoring frame, but also dynamic change characteristics in the time dimension. Therefore, in the technical scheme of the application, feature mining is performed on the stirring monitoring video of the mixture in the preset time period in a second convolution neural network model by using a three-dimensional convolution kernel, so that dynamic change feature distribution information of the state feature of the mixture in the stirring monitoring video of the mixture in the time dimension is extracted, and therefore the stirring state monitoring feature vector is obtained. In particular, the convolution kernel of the second convolutional neural network model is a three-dimensional convolution kernel, which has W (width), H (height) and C (channel dimension), and in the technical solution of the present application, the channel dimension of the three-dimensional convolution kernel corresponds to the time dimension of the stirring monitoring video of the mixture, so that when performing three-dimensional convolutional encoding, dynamic change feature information of the state feature of the mixture in the time dimension can be extracted.

Accordingly, in one specific example, as shown in fig. 5, the mixing monitoring video of the mixture of the predetermined period of time is obtained by using a second convolution neural network model of a three-dimensional convolution kernel to obtain a mixing state monitoring feature vector, including: s141, extracting a plurality of image key frames from the stirring monitoring video of the mixture in the preset time period, and arranging the image key frames according to the time dimension to obtain a three-dimensional input tensor; s142, respectively performing three-dimensional convolution processing, mean pooling processing and nonlinear activation processing based on the three-dimensional convolution kernel on input data in forward transfer of layers through each layer of the second convolution neural network model using the three-dimensional convolution kernel to output the stirring state monitoring feature map by the last layer of the second convolution neural network model using the three-dimensional convolution kernel, wherein the input of the first layer of the second convolution neural network model using the three-dimensional convolution kernel is the three-dimensional input tensor; and S143, carrying out global averaging on each feature matrix of the stirring state monitoring feature map to obtain the stirring state monitoring feature vector.

In view of the fact that, when the ink production is actually performed, not only the addition speed value of each component, the stirring speed value of the stirrer, but also the mixing state change of the mixture have fluctuation and uncertainty in the time dimension, in order to improve the accuracy of real-time control of the stirring speed, it is necessary to perform data enhancement in a high-dimensional feature space for each of the addition speed-related feature of each component, the rotational speed dynamic feature of the stirrer, and the mixing state change feature of the mixture. It should be understood that, as a learning target of the neural network model, the gaussian density map may represent a joint distribution in the case where a plurality of feature values constitute an overall distribution due to probability density thereof, that is, a feature distribution is taken as a priori distribution, and probability density under the influence of correlation with other a priori distribution positions at each a priori distribution position is obtained as a posterior distribution, thereby describing the feature distribution more accurately in a higher dimension. Therefore, in the technical solution of the present application, the data enhancement can be performed on the addition speed association characteristic of each component, the rotational speed dynamic characteristic of the mixer, and the mixing state change characteristic of the mixture by using the addition speed of each component, the rotational speed of the mixer, and the mixing state of the mixture as prior distributions, i.e., gaussian distributions, respectively. Specifically, the gaussian density maps of the component addition feature vector, the rotational speed feature vector, and the stirring state monitoring feature vector are respectively constructed to obtain first to third gaussian density maps. And the component adding feature vector, the rotating speed feature vector and the stirring state monitoring feature vector correspond to the first Gaussian density map to the third Gaussian density map respectively.

More specifically, in step S150, feature level data enhancement is performed on the component addition feature vector, the rotational speed feature vector, and the agitation state monitoring feature vector based on the gaussian density map to obtain first to third gaussian density maps.

More specifically, in step S160, the first to third gaussian density maps are fused using a bayesian probability model to obtain a posterior gaussian density map. In order to perform the adaptive adjustment of the rotational speed of the mixer based on the actual added mass of each component of the water-based acrylic mark pen ink and the real-time state change of the mixture, it is necessary to extract the correlation relationship between the three components, and to generate the result of the adaptive control of the rotational speed of the mixer by using the correlation feature. It should be understood that, considering that the second gaussian density map corresponding to the rotational speed feature vector is used as a prior probability vector, the technical solution of the present application aims to update the prior probability to obtain the posterior probability when new evidence, that is, when the adding speed of each component changes. Then, according to a bayesian formula, the posterior probability is the prior probability multiplied by the event probability divided by the evidence probability, so in the technical scheme of the application, a bayesian probability model is used for fusing the corresponding first to third gaussian density maps respectively corresponding to the component addition feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to obtain a posterior gaussian density map. Specifically, the second gaussian density map corresponding to the rotational speed feature vector is used as a priori probability, the first gaussian density map corresponding to the component addition feature vector is used as evidence probability, the third gaussian density map corresponding to the stirring state monitoring feature vector is used as event probability, and the posterior probability of real-time change of the rotational speed of the stirrer when the addition speed of each component is changed is calculated, so that the posterior gaussian density map is obtained.

Accordingly, in one specific example, the fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map includes: fusing the first gaussian density map, the second gaussian density map, and the third gaussian density map using a bayesian probability model in the following formula to obtain the posterior gaussian density map; wherein, the formula is:

wherein ,

representing the posterior Gaussian density map, < >>

Representing said first Gaussian density map, ">

Representing the second Gaussian density map, +.>

Representing the third gaussian density map.

More specifically, in step S170, the posterior gaussian density map is subjected to gaussian discretization to obtain a classification feature matrix. And carrying out Gaussian discretization processing on the posterior Gaussian density map so as not to generate information loss when the data characteristics are amplified, thereby obtaining a classification characteristic matrix, and classifying the posterior Gaussian density map so as to improve the accuracy of subsequent classification.

More specifically, in step S180, feature distribution correction is performed on the classification feature matrix to obtain a corrected classification feature matrix.

Particularly, in the technical scheme of the application, when feature data enhancement is performed on the component adding feature vector, the rotating speed feature vector and the stirring state feature vector based on the Gaussian density map respectively to obtain the first to third Gaussian density maps and when Gaussian discretization is performed on the posterior Gaussian density map to obtain the classification feature matrix, due to the random characteristics of the Gaussian density map and the Gaussian discretization, negative correlation values relative to global feature distribution are introduced into local feature distribution of the classification feature matrix, so that classification accuracy of the classification feature matrix is affected. Therefore, the applicant of the present application corrects the classification feature matrix in a non-linear re-weighting manner by full orthographic projection.

Accordingly, in one specific example, the performing feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix includes: carrying out feature distribution correction on the classification feature matrix by using the following formula to obtain a corrected classification feature matrix; wherein, the formula is:

wherein Mc and M are the classification feature matrix and the corrected classification feature matrix, respectively, and ReLU (·) represents a ReLU activation function,

More specifically, in step S190, the corrected classification feature matrix is passed through a classifier to obtain a classification result indicating whether the rotational speed value of the mixer at the current point in time should be increased or decreased. In the technical scheme of the application, the label of the classifier comprises that the rotation speed value of the stirrer at the current time point should be increased or decreased, wherein the classifier determines which classification label the classification feature matrix belongs to through a soft maximum function. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control policy label of the rotation speed value of the mixer at the current time point, so after the classification result is obtained, the rotation speed value of the mixer at the current time point can be adaptively adjusted based on the classification result, so as to perform adaptive control of the mixing speed based on the real-time change of the quality and the mixing state of the actual raw material addition, so as to achieve the purpose of improving the mixing effect.

Accordingly, in a specific example, as shown in fig. 6, the passing the corrected classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the rotational speed value of the mixer at the current time point should be increased or decreased, includes: s191, expanding the corrected classification feature matrix into classification feature vectors according to row vectors or column vectors; s192, performing full-connection coding on the classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and S193, passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

In summary, according to the method and the system for intelligently preparing the water-based propylene ink according to the embodiments of the present application, the adding speed values of the components of the water-based propylene mark pen ink at a plurality of predetermined time points in a predetermined time period, the rotating speed values of the stirrer at the plurality of predetermined time points and the stirring monitoring video of the mixture in the predetermined time period are obtained, the deep neural network model based on deep learning is adopted, and the parameters of the deep neural network model are adjusted through the backward propagation algorithm based on gradient descent so as to simulate complex nonlinear association between things, thereby establishing a complex mapping relation between the adding speed of the components of the water-based propylene mark pen ink, the stirring speed of the stirrer and the mixing state change of the mixture, so that the self-adaptive control of the stirring speed is performed based on the real-time change of the quality and the mixing state of the actual raw material addition. In this way, the mixing effect can be improved, and the quality of the produced ink can be further improved.

Exemplary System

Fig. 7 is a block diagram of an intelligent preparation system 100 for water-based propylene ink in accordance with an embodiment of the present application. As shown in fig. 7, the intelligent preparation system 100 of the water-based propylene ink according to the embodiment of the present application includes: a data acquisition module 110, configured to acquire an addition speed value of each component of the water-based acrylic mark pen ink at a plurality of predetermined time points in a predetermined time period, a rotation speed value of the mixer at the plurality of predetermined time points, and a mixing monitoring video of the mixture in the predetermined time period; a first convolution module 120, configured to arrange addition velocity values of each component of the water-based acrylic mark pen ink at a plurality of predetermined time points within the predetermined time period into an addition velocity input matrix according to a time dimension and a sample dimension, and then obtain component addition feature vectors through a first convolution neural network model serving as a filter; the multi-scale feature extraction module 130 is configured to arrange the rotational speed values of the stirring machine at the plurality of predetermined time points into rotational speed input vectors according to a time dimension, and then obtain rotational speed feature vectors through the multi-scale neighborhood feature extraction module; a second convolution module 140, configured to obtain a stirring state monitoring feature vector by using a second convolution neural network model of a three-dimensional convolution kernel through a stirring monitoring video of the mixture in the predetermined period; a feature level data enhancement module 150, configured to add a feature vector to the component, the rotational speed feature vector, and the stirring state monitoring feature vector based on a gaussian density map, and perform feature level data enhancement to obtain first to third gaussian density maps; a fusion module 160 for fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map; the gaussian discretization module 170 is configured to perform gaussian discretization on the posterior gaussian density map to obtain a classification feature matrix; the feature distribution correction module 180 is configured to perform feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix; and a classification result generating module 190, configured to pass the corrected classification feature matrix through a classifier to obtain a classification result, where the classification result is used to indicate that the rotational speed value of the mixer at the current time point should be increased or decreased.

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the first convolution module 120 is further configured to: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer: carrying out convolution processing on the input data to obtain a convolution characteristic diagram; carrying out mean pooling treatment based on a feature matrix on the convolution feature map to obtain a pooled feature map; performing nonlinear activation on the pooled feature map to obtain an activated feature map; and the output of the last layer of the first convolutional neural network model serving as the filter adds a feature vector to the component, and the input of the first layer of the first convolutional neural network model serving as the filter is the addition speed input matrix.

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the multi-scale neighborhood feature extraction module includes a first convolution layer and a second convolution layer in parallel, and a multi-scale fusion layer connected to the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer each use one-dimensional convolution kernels having different scales.

In one example, in the intelligent production system 100 of water-based propylene ink described above, the multi-scale feature extraction module 130 comprises: a first scale domain feature extraction unit, configured to input the rotation speed input vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first scale rotation speed feature vector, where the first convolution layer has a first one-dimensional convolution kernel with a first length; a second scale domain feature extraction unit, configured to input the rotation speed input vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale rotation speed feature vector, where the second convolution layer has a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and the multi-scale fusion unit is used for cascading the first-scale rotating speed feature vector and the second-scale rotating speed feature vector by using a multi-scale fusion layer of the multi-scale neighborhood feature extraction module so as to obtain the rotating speed feature vector.

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the first scale domain feature extraction unit is further configured to: performing one-dimensional convolution encoding on the rotating speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first-scale rotating speed feature vector; wherein, the formula is:

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the second scale domain feature extraction unit is further configured to: performing one-dimensional convolution encoding on the rotating speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale rotating speed feature vector; wherein, the formula is:

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the second convolution module 140 is further configured to: extracting a plurality of image key frames from the stirring monitoring video of the mixture in the preset time period, and arranging the image key frames according to the time dimension to obtain a three-dimensional input tensor; respectively performing three-dimensional convolution processing, mean pooling processing and nonlinear activation processing based on the three-dimensional convolution kernel on input data in forward transfer of layers through each layer of the second convolution neural network model using the three-dimensional convolution kernel to monitor a characteristic diagram by the output of the last layer of the second convolution neural network model using the three-dimensional convolution kernel, wherein the input of the first layer of the second convolution neural network model using the three-dimensional convolution kernel is the three-dimensional input tensor; and carrying out global average pooling on each feature matrix of the stirring state monitoring feature map to obtain the stirring state monitoring feature vector.

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the fusion module 160 is further configured to: fusing the first gaussian density map, the second gaussian density map, and the third gaussian density map using a bayesian probability model in the following formula to obtain the posterior gaussian density map; wherein, the formula is:

wherein ,

representing the posterior Gaussian density map, < >>

Representing said first Gaussian density map, ">

Representing the second Gaussian density map, +.>

Representing the third gaussian density map.

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the feature distribution correction module 180 is further configured to: carrying out feature distribution correction on the classification feature matrix by using the following formula to obtain a corrected classification feature matrix; wherein, the formula is:

wherein Mc and M are the classification bits, respectivelyThe sign matrix and the corrected classification feature matrix, the ReLU (·) represents a ReLU activation function,

In one example, in the above-described intelligent preparation system 100 for water-based propylene ink, the classification result generation module is further configured to: expanding the corrected classification characteristic matrix into classification characteristic vectors according to row vectors or column vectors; performing full-connection coding on the classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

Here, it will be understood by those skilled in the art that the specific functions and operations of the respective units and modules in the above-described intelligent preparation system 100 for water-based propylene ink have been described in detail in the above description of the intelligent preparation method for water-based propylene ink with reference to fig. 1 to 6, and thus, repetitive descriptions thereof will be omitted.

As described above, the intelligent preparation system 100 of the water-based propylene ink according to the embodiment of the present application may be implemented in various wireless terminals, such as a server or the like having an intelligent preparation algorithm of the water-based propylene ink. In one example, the intelligent preparation system 100 of water-based propylene ink in accordance with embodiments of the present application may be integrated into a wireless terminal as one software module and/or hardware module. For example, the intelligent preparation system 100 of the water-based propylene ink may be a software module in the operating system of the wireless terminal, or may be an application developed for the wireless terminal; of course, the intelligent preparation system 100 of the water-based propylene ink may also be one of the many hardware modules of the wireless terminal.

Alternatively, in another example, the intelligent preparation system 100 of the water-based propylene ink and the wireless terminal may also be separate devices, and the intelligent preparation system 100 of the water-based propylene ink may be connected to the wireless terminal through a wired and/or wireless network and transmit interactive information in accordance with a agreed data format.

The basic principles of the present application have been described above in connection with specific embodiments, however, it should be noted that the advantages, benefits, effects, etc. mentioned in the present application are merely examples and not limiting, and these advantages, benefits, effects, etc. are not to be considered as necessarily possessed by the various embodiments of the present application. Furthermore, the specific details disclosed herein are for purposes of illustration and understanding only, and are not intended to be limiting, as the application is not intended to be limited to the details disclosed herein as such.

The block diagrams of the devices, apparatuses, devices, systems referred to in this application are only illustrative examples and are not intended to require or imply that the connections, arrangements, configurations must be made in the manner shown in the block diagrams. As will be appreciated by one of skill in the art, the devices, apparatuses, devices, systems may be connected, arranged, configured in any manner. Words such as "including," "comprising," "having," and the like are words of openness and mean "including but not limited to," and are used interchangeably therewith. The terms "or" and "as used herein refer to and are used interchangeably with the term" and/or "unless the context clearly indicates otherwise. The term "such as" as used herein refers to, and is used interchangeably with, the phrase "such as, but not limited to.

It is also noted that in the apparatus, devices and methods of the present application, the components or steps may be disassembled and/or assembled. Such decomposition and/or recombination should be considered as equivalent to the present application.

The previous description of the disclosed aspects is provided to enable any person skilled in the art to make or use the present application. Various modifications to these aspects will be readily apparent to those skilled in the art, and the generic principles defined herein may be applied to other aspects without departing from the scope of the application. Thus, the present application is not intended to be limited to the aspects shown herein but is to be accorded the widest scope consistent with the principles and novel features disclosed herein.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit the embodiments of the application to the form disclosed herein. Although a number of example aspects and embodiments have been discussed above, a person of ordinary skill in the art will recognize certain variations, modifications, alterations, additions, and subcombinations thereof.

Claims

1. An intelligent preparation method of water-based propylene ink is characterized by comprising the following steps: acquiring the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in a preset time period, the rotation speed values of a stirrer at the preset time points and a stirring monitoring video of a mixture in the preset time period; arranging the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an addition speed input matrix according to the time dimension and the sample dimension, and then obtaining a component addition characteristic vector through a first convolution neural network model serving as a filter; the rotational speed values of the mixers at a plurality of preset time points are arranged into rotational speed input vectors according to time dimensions, and then the rotational speed input vectors are processed through a multi-scale neighborhood feature extraction module to obtain rotational speed feature vectors; stirring monitoring video of the mixture in the preset time period is processed through a second convolution neural network model using a three-dimensional convolution kernel to obtain a stirring state monitoring feature vector; adding a feature vector, the rotating speed feature vector and the stirring state monitoring feature vector to the components based on the Gaussian density map for feature data enhancement so as to obtain first to third Gaussian density maps; fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map; performing Gaussian discretization on the posterior Gaussian density map to obtain a classification feature matrix; performing feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix; and passing the corrected classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the stirrer at the current time point is increased or decreased.

2. The intelligent preparation method of the water-based propylene ink according to claim 1, wherein the arranging the adding speed values of the components of the water-based propylene mark pen ink at a plurality of preset time points in the preset time period according to the time dimension and the sample dimension into an adding speed input matrix, and then obtaining component adding feature vectors through a first convolutional neural network model serving as a filter comprises the following steps: each layer of the first convolutional neural network model used as the filter performs the following steps on input data in forward transfer of the layer: carrying out convolution processing on the input data to obtain a convolution characteristic diagram; carrying out mean pooling treatment based on a feature matrix on the convolution feature map to obtain a pooled feature map; non-linear activation is carried out on the pooled feature map so as to obtain an activated feature map; and the output of the last layer of the first convolutional neural network model serving as the filter adds a feature vector to the component, and the input of the first layer of the first convolutional neural network model serving as the filter is the addition speed input matrix.

3. The intelligent preparation method of the water-based propylene ink according to claim 2, wherein the multi-scale neighborhood feature extraction module comprises a first convolution layer and a second convolution layer which are parallel, and a multi-scale fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer respectively use one-dimensional convolution kernels with different scales.

4. The intelligent preparation method of the water-based propylene ink according to claim 3, wherein the steps of arranging the rotational speed values of the stirrer at the plurality of preset time points into rotational speed input vectors according to a time dimension, and obtaining rotational speed feature vectors through a multi-scale neighborhood feature extraction module comprise the following steps: inputting the rotating speed input vector into a first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale rotating speed feature vector, wherein the first convolution layer is provided with a first one-dimensional convolution kernel with a first length; inputting the rotation speed input vector into a second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second-scale rotation speed feature vector, wherein the second convolution layer is provided with a second one-dimensional convolution kernel with a second length, and the first length is different from the second length; and cascading the first scale rotational speed feature vector and the second scale rotational speed feature vector by using a multi-scale fusion layer of the multi-scale neighborhood feature extraction module to obtain the rotational speed feature vector.

5. The method for intelligent preparation of water-based propylene ink as defined in claim 4, wherein inputting the rotational speed input vector into the first convolution layer of the multi-scale neighborhood feature extraction module to obtain a first-scale rotational speed feature vector, further comprises: performing one-dimensional convolution encoding on the rotating speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a first-scale rotating speed feature vector; wherein, the formula is:

Wherein a is the width of the first one-dimensional convolution kernel in the X direction, F (a) is a parameter vector of the first one-dimensional convolution kernel, G (X-a) is a local vector matrix calculated by a convolution kernel function, w is the size of the first one-dimensional convolution kernel, and X represents the rotational speed input vector; the inputting the rotation speed input vector into the second convolution layer of the multi-scale neighborhood feature extraction module to obtain a second scale rotation speed feature vector, further comprising: performing one-dimensional convolution encoding on the rotating speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following formula to obtain a second-scale rotating speed feature vector;

wherein, the formula is:

6. The intelligent preparation method of the water-based propylene ink according to claim 5, wherein the mixing monitoring video of the mixture of the predetermined time period is obtained by using a second convolution neural network model of a three-dimensional convolution kernel to obtain a mixing state monitoring feature vector, comprising: extracting a plurality of image key frames from the stirring monitoring video of the mixture in the preset time period, and arranging the image key frames according to the time dimension to obtain a three-dimensional input tensor; respectively performing three-dimensional convolution processing, mean pooling processing and nonlinear activation processing based on the three-dimensional convolution kernel on input data in forward transfer of layers through each layer of the second convolution neural network model using the three-dimensional convolution kernel to monitor a characteristic diagram by the output of the last layer of the second convolution neural network model using the three-dimensional convolution kernel, wherein the input of the first layer of the second convolution neural network model using the three-dimensional convolution kernel is the three-dimensional input tensor; and carrying out global average pooling on each feature matrix of the stirring state monitoring feature map to obtain the stirring state monitoring feature vector.

7. The method of intelligent preparation of water-based propylene ink according to claim 6, wherein the fusing the first to third gaussian density maps using bayesian probability models to obtain a posterior gaussian density map comprises: fusing the first gaussian density map, the second gaussian density map, and the third gaussian density map using a bayesian probability model in the following formula to obtain the posterior gaussian density map; wherein, the formula is:

wherein ,

representing the posterior Gaussian density map, < >>

Representing said first Gaussian density map, ">

Representing the second Gaussian density map, +.>

Representing the third gaussian density map.

8. The method for intelligently preparing the water-based propylene ink according to claim 7, wherein the step of performing feature distribution correction on the classification feature matrix to obtain a corrected classification feature matrix comprises the steps of:

wherein, the formula is:

Representing the multiplication of the matrix,and the division between the numerator matrix and the denominator matrix is the division by position of the matrix eigenvalue, exp (·) represents the exponential operation of the matrix, which represents the calculation of the natural exponential function value raised to the power of the eigenvalue at each position in the matrix.

9. The intelligent preparation method of the water-based propylene ink according to claim 8, wherein the step of passing the corrected classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotation speed value of the stirrer at the current time point should be increased or decreased comprises the following steps: expanding the corrected classification characteristic matrix into classification characteristic vectors according to row vectors or column vectors; performing full-connection coding on the classification feature vectors by using a plurality of full-connection layers of the classifier to obtain coded classification feature vectors; and passing the coding classification feature vector through a Softmax classification function of the classifier to obtain the classification result.

10. An intelligent preparation system of water-based propylene ink, which is characterized by comprising: the data acquisition module is used for acquiring the addition speed values of each component of the water-based acrylic mark pen ink at a plurality of preset time points in a preset time period, the rotation speed values of the stirrer at the preset time points and the stirring monitoring video of the mixture in the preset time period; the first convolution module is used for arranging the adding speed values of the components of the water-based acrylic mark pen ink at a plurality of preset time points in the preset time period into an adding speed input matrix according to the time dimension and the sample dimension, and then obtaining component adding feature vectors through a first convolution neural network model serving as a filter; the multi-scale feature extraction module is used for arranging the rotating speed values of the stirring machines at a plurality of preset time points into rotating speed input vectors according to the time dimension and then obtaining rotating speed feature vectors through the multi-scale neighborhood feature extraction module; the second convolution module is used for obtaining a stirring state monitoring feature vector through a second convolution neural network model using a three-dimensional convolution kernel according to the stirring monitoring video of the mixture in the preset time period; the characteristic data enhancement module is used for adding a characteristic vector, the rotating speed characteristic vector and the stirring state monitoring characteristic vector to the component based on the Gaussian density map, and carrying out characteristic data enhancement to obtain first to third Gaussian density maps; a fusion module for fusing the first to third gaussian density maps using a bayesian probability model to obtain a posterior gaussian density map; the Gaussian discretization module is used for carrying out Gaussian discretization on the posterior Gaussian density map to obtain a classification characteristic matrix; the characteristic distribution correction module is used for carrying out characteristic distribution correction on the classified characteristic matrix to obtain a corrected classified characteristic matrix; and the classification result generation module is used for passing the corrected classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the rotating speed value of the stirrer at the current time point should be increased or decreased.