CN116704412A

CN116704412A - Rose petal processing method and system

Info

Publication number: CN116704412A
Application number: CN202310679605.7A
Authority: CN
Inventors: 董卫强; 徐继雄; 王建清
Original assignee: Liangdang Qinxiangyi Rose Biotechnology Co ltd
Current assignee: Liangdang Qinxiangyi Rose Biotechnology Co ltd
Priority date: 2023-06-09
Filing date: 2023-06-09
Publication date: 2023-09-05

Abstract

The processing method and system of the rose petals acquire grinding state monitoring videos of a preset time period acquired by a camera, and grinding speed values of a plurality of preset time points in the preset time period; by adopting an artificial intelligence technology based on deep learning, a mapping relation between the time sequence change of the grinding state of the rose petals and the time sequence change of the grinding speed value is established, so that the grinding speed value is adaptively adjusted based on the characteristic of the time sequence change of the grinding state of the rose petals, thereby realizing intelligent processing control, optimizing the grinding effect and efficiency of the rose petals, and improving the production quality of the rose petals.

Description

Rose petal processing method and system

Technical Field

The application relates to the technical field of intelligent processing, in particular to a processing method and system for rose petals.

Background

The rose petals are petals of rose of Rosa of Rosaceae, are mild in rose property, sweet in fragrance, capable of expelling toxin, beautifying, promoting bile secretion, helping digestion and regulating mechanism, and can be made into rose petal tea, rose petal porridge, rose cake and the like after being dried in the sun.

During daily processing, the rose petals need to be crushed and ground. In general, a conventional polishing apparatus for processing rose petals is used by simply putting the rose petals into a polishing apparatus and polishing them for a fixed polishing time and polishing rate, and there is no concern about suitability of the polishing state and polishing rate of the rose petals. That is, when the addition amount or quality of the rose petals is different, the grinding state is changed differently in different time periods, and at this time, different grinding speeds are required to optimize the grinding efficiency and effect, so as to improve the processing quality of the rose petals.

Thus, an optimized rose petal processing regimen is desired.

Disclosure of Invention

The present application has been made to solve the above-mentioned technical problems. The embodiment of the application provides a rose petal processing method and a rose petal processing system, which are used for acquiring a grinding state monitoring video of a preset time period acquired by a camera and grinding speed values of a plurality of preset time points in the preset time period; by adopting an artificial intelligence technology based on deep learning, a mapping relation between the time sequence change of the grinding state of the rose petals and the time sequence change of the grinding speed value is established, so that the grinding speed value is adaptively adjusted based on the characteristic of the time sequence change of the grinding state of the rose petals, thereby realizing intelligent processing control, optimizing the grinding effect and efficiency of the rose petals, and improving the production quality of the rose petals.

In a first aspect, a method for processing rose petals is provided, which comprises:

acquiring a grinding state monitoring video of a preset time period acquired by a camera, and grinding speed values of a plurality of preset time points in the preset time period;

extracting image frames corresponding to the plurality of preset time points from the grinding state monitoring video to serve as a plurality of grinding state monitoring key frames;

the grinding state monitoring key frames are respectively passed through a convolutional neural network model serving as a filter to obtain a plurality of grinding granularity state feature vectors;

calculating Euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors to obtain a particle size change time sequence feature vector composed of the Euclidean distance values;

arranging the grinding speed values of the plurality of preset time points into grinding speed input vectors according to time dimensions, and then obtaining grinding speed time sequence feature vectors through a multi-scale neighborhood feature extraction module;

calculating the response estimation of the granularity change time sequence feature vector relative to the grinding speed time sequence feature vector to obtain a classification feature matrix; and

and passing the classification feature matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the grinding speed value of the current time point is increased or decreased.

In the above processing method for rose petals, the step of passing the plurality of grinding state monitoring key frames through a convolutional neural network model as a filter to obtain a plurality of grinding particle size state feature vectors includes: each layer of the 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; the output of the last layer of the convolutional neural network model serving as the filter is the plurality of grinding granularity state characteristic vectors, and the input of the first layer of the convolutional neural network model serving as the filter is the plurality of grinding state monitoring key frames.

In the above rose petal processing method, calculating euclidean distance values between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors to obtain a particle size change time sequence feature vector composed of a plurality of euclidean distance values, includes: calculating Euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors by using the following distance calculation formula to obtain a plurality of Euclidean distance values; wherein, the distance formula is:

Wherein V is _i And V _j Representing every adjacent two of the plurality of particle size status feature vectors,and->Representing the grinding particle size state characteristic vector V _i And V _j Characteristic value of each position, d (V _i ，V _j ) Representing a Euclidean distance between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors; and arranging the Euclidean distance values to obtain the granularity change time sequence feature vector.

In the above rose petal processing method, the multi-scale neighborhood feature extraction module includes: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use one-dimensional convolution kernels with different scales.

In the above rose petal processing method, the step of arranging the polishing speed values at the plurality of predetermined time points into a polishing speed input vector according to a time dimension, and then obtaining a polishing speed time sequence feature vector by a multi-scale neighborhood feature extraction module includes: performing one-dimensional convolution coding on the grinding speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula to obtain a first-scale grinding speed feature vector; wherein the first convolution formula is:

Wherein a is the width of the first one-dimensional convolution kernel in the X direction, F (a) is a first one-dimensional convolution kernel parameter vector, G (X-a) is a local vector matrix calculated by a first one-dimensional convolution kernel function, w is the size of the first one-dimensional convolution kernel, X represents the grinding speed input vector, and Cov (X) represents one-dimensional convolution encoding of the grinding speed input vector; performing one-dimensional convolution coding on the grinding speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula to obtain a second-scale grinding speed feature vector; wherein the second convolution 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 function of the second one-dimensional convolution kernel, m is the size of the second one-dimensional convolution kernel, X represents the grinding speed input vector, and Cov (X) represents one-dimensional convolution encoding of the grinding speed input vector; and cascading the first-scale grinding speed feature vector and the second-scale grinding speed feature vector by using a fusion layer of the multi-scale neighborhood feature extraction module to obtain the grinding speed time sequence feature vector.

In the above rose petal processing method, calculating a responsiveness estimate of the granularity variation time series feature vector with respect to the grinding speed time series feature vector to obtain a classification feature matrix includes: constructing a granularity change Gaussian density map of the granularity change time sequence feature vector according to the following first Gaussian formula; wherein, the first gaussian formula is:

wherein mu ₁ Representing the granularity variation timing feature vector, and sigma ₁ Representing the variance between the eigenvalues of the corresponding two locations in the granularity variation time sequence eigenvector; constructing a grinding speed Gaussian density chart of the grinding speed time sequence characteristic vector according to the following second Gaussian formula; wherein, the second gaussian formula is:

wherein mu ₂ Representing the polishing rate time sequence feature vector, and sigma ₂ The value of each position of the polishing rate time sequence feature vector represents the variance between the feature values of the corresponding two positions; calculating a responsiveness estimate of the particle size change gaussian density map relative to the grinding speed gaussian density map with a responsiveness formula to obtain a responsiveness gaussian density map; wherein, the responsiveness formula is:

wherein mu ₃ Mean vector, sigma, representing the responsive gaussian density map ₃ Covariance matrix representing the response Gaussian density map, +.The vector point-multiply, +.1 indicates the reciprocal of the value at each position of the vector, andrepresenting a matrix multiplication; carrying out Gaussian discretization processing on the Gaussian distribution of each position in the responsive Gaussian density map so as to reduce the Gaussian distribution of each position in the responsive Gaussian density map into a one-dimensional feature vector; and, two-dimensionally arranging the one-dimensional feature vectors of the respective positions to obtain the classification bitsA sign matrix.

The rose petal processing method further comprises training the convolutional neural network model serving as a filter, the multi-scale neighborhood feature extraction module and the classifier; wherein training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier comprises: acquiring training data, wherein the training data comprises training grinding state monitoring videos of a preset time period, training grinding speed values of a plurality of preset time points in the preset time period, and a true value of the grinding speed value of the current time point to be increased or decreased; extracting training image frames corresponding to the preset time points from the training grinding state monitoring video to serve as a plurality of training grinding state monitoring key frames; respectively passing the training grinding state monitoring key frames through the convolutional neural network model serving as a filter to obtain a plurality of training grinding granularity state feature vectors; calculating Euclidean distance values between every two adjacent training granularity state feature vectors in the training granularity state feature vectors to obtain training granularity change time sequence feature vectors composed of the Euclidean distance values; the training grinding speed values of the plurality of preset time points are arranged into training grinding speed input vectors according to time dimensions, and then the training grinding speed input vectors are passed through the multi-scale neighborhood feature extraction module to obtain training grinding speed time sequence feature vectors; calculating the response estimation of the training granularity change time sequence feature vector relative to the training grinding speed time sequence feature vector to obtain a training classification feature matrix; performing class Fourier scale domain probability correction on the training classification feature matrix to obtain an optimized training classification feature matrix; the optimized training classification characteristic matrix passes through the classifier to obtain a classification loss function value; and training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier with the classification loss function value as a loss function value and by back propagation of gradient descent.

In the above processing method for rose petals, performing fourier-like scale domain probability correction on the training classification feature matrix to obtain an optimized training classification feature matrix, including: carrying out class Fourier scale domain probability correction on the training classification feature matrix by using the following optimization formula to obtain the optimized training classification feature matrix; wherein, the optimization formula is:

wherein m is _i，j Is the eigenvalue of the (i, j) th position of the training classification feature matrix, W and H are the height and width of the training classification feature matrix respectively, and alpha and beta are the hyper-parameters for scale adjustment, exp (·) represents the natural exponential function value calculated to be the power of the value, m' _i，j Is the eigenvalue of the (i, j) th position of the optimization training classification eigenvalue matrix.

In the above processing method for rose petals, the step of passing the optimized training classification feature matrix through the classifier to obtain a classification loss function value includes: the classifier processes the optimized training classification feature matrix with the following classification formula to generate a training classification result, wherein the classification formula is: softmax { (W) _n ，B _n )：...：(W ₁ ，B ₁ ) X, where X represents the optimized training classification feature matrix, W ₁ To W _n Is a weight matrix, B ₁ To B _n Representing a bias matrix; and calculating a cross entropy value between the training classification result and a true value as the classification loss function value.

In a second aspect, there is provided a rose petal processing system comprising:

the data acquisition module is used for acquiring grinding state monitoring videos of a preset time period acquired by the camera and grinding speed values of a plurality of preset time points in the preset time period;

the key frame extraction module is used for extracting image frames corresponding to the preset time points from the grinding state monitoring video to serve as a plurality of grinding state monitoring key frames;

the characteristic extraction module is used for respectively passing the grinding state monitoring key frames through a convolutional neural network model serving as a filter to obtain a plurality of grinding particle size state characteristic vectors;

the Euclidean distance calculation module is used for calculating Euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors so as to obtain a particle size change time sequence feature vector composed of the Euclidean distance values;

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

The responsiveness estimation calculation module is used for calculating responsiveness estimation of the granularity change time sequence feature vector relative to the grinding speed time sequence feature vector so as to obtain a classification feature matrix; and

and the grinding speed value control module is used for passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the grinding speed value of the current time point should be increased or decreased.

Compared with the prior art, the rose petal processing method and system provided by the application have the advantages that the grinding state monitoring video of the preset time period acquired by the camera is acquired, and the grinding speed values of a plurality of preset time points in the preset time period are obtained; by adopting an artificial intelligence technology based on deep learning, a mapping relation between the time sequence change of the grinding state of the rose petals and the time sequence change of the grinding speed value is established, so that the grinding speed value is adaptively adjusted based on the characteristic of the time sequence change of the grinding state of the rose petals, thereby realizing intelligent processing control, optimizing the grinding effect and efficiency of the rose petals, and improving the production quality of the rose petals.

Drawings

In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings used in the description of the embodiments or the prior art will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present application, and other drawings may be obtained according to these drawings without inventive effort for a person skilled in the art.

Fig. 1 is a schematic view of a processing method of rose petals according to an embodiment of the present application.

Fig. 2 is a flowchart of a rose petal processing method according to an embodiment of the present application.

Fig. 3 is a schematic diagram of a processing method of rose petals according to an embodiment of the present application.

Fig. 4 is a flowchart of the sub-steps of step 180 of the rose petal processing method according to an embodiment of the present application.

Fig. 5 is a block diagram of a rose petal processing system according to an embodiment of the present application.

Detailed Description

The following description of the technical solutions according to the embodiments of the present application will be given with reference to the accompanying drawings in the embodiments of the present application, and it is apparent that the described embodiments are some embodiments of the present application, but not all embodiments. All other embodiments, which can be made by those skilled in the art based on the embodiments of the application without making any inventive effort, are intended to be within the scope of the application.

Unless defined otherwise, all technical and scientific terms used in the embodiments of the application have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used in the present application is for the purpose of describing particular embodiments only and is not intended to limit the scope of the present application.

In describing embodiments of the present application, unless otherwise indicated and limited thereto, the term "connected" should be construed broadly, for example, it may be an electrical connection, or may be a communication between two elements, or may be a direct connection, or may be an indirect connection via an intermediate medium, and it will be understood by those skilled in the art that the specific meaning of the term may be interpreted according to circumstances.

It should be noted that, the term "first\second\third" related to the embodiment of the present application is merely to distinguish similar objects, and does not represent a specific order for the objects, it is to be understood that "first\second\third" may interchange a specific order or sequence where allowed. It is to be understood that the "first\second\third" distinguishing objects may be interchanged where appropriate such that embodiments of the application described herein may be practiced in sequences other than those illustrated or described herein.

As described above, in the conventional polishing apparatus for processing rose petals, the polishing process is usually performed by merely placing the rose petals in the polishing apparatus at the time of use and by fixing the polishing time and the polishing rate, and there is no concern about the suitability of the polishing state and the polishing rate of the rose petals. That is, when the addition amount or quality of the rose petals is different, the grinding state is changed differently in different time periods, and at this time, different grinding speeds are required to optimize the grinding efficiency and effect, so as to improve the processing quality of the rose petals. Thus, an optimized rose petal processing regimen is desired.

Specifically, in the technical scheme of the application, a processing method for rose petals is provided, which comprises the following steps: cleaning and airing fresh rose petals to obtain pretreated rose petals; placing the pretreated rose petals into a grinder, and crushing the pretreated rose petals by using a fine grinding head to obtain rose petal powder, wherein in the process, the grinding time and speed can be adjusted according to the needs to ensure that the rose petal powder reaches the required powder particle size; and (3) placing the rose petal powder into a dry container for sealing, and storing the container in a cool, dry and ventilated place to avoid being affected with damp.

Accordingly, in the actual processing process of rose petals, the self-adaptive control of the grinding speed is important in the grinding process, that is, if the grinding speed is too high, uneven abrasion of the surfaces of the rose petals may be caused, and the situation of scratch or insufficient abrasion occurs; however, too low a polishing rate can prolong the processing time and reduce the polishing efficiency. Therefore, in order to enable the rose petal powder to reach the required powder granularity, the control of the grinding speed is adapted to the grinding state change condition of the rose petals, namely, the grinding speed value is adaptively adjusted based on the grinding state time sequence change characteristics of the rose petals, so that intelligent processing control is realized, the grinding effect and efficiency of the rose petals are optimized, and the production quality of the rose petals is improved. However, the grinding state characteristics of the rose petals are difficult to capture and describe in the actual monitoring process, and the grinding state characteristics and the grinding speed values have time sequence dynamic cooperative association relation in the time dimension. Therefore, in this process, it is difficult to establish a mapping relationship between the time sequence change of the grinding state of the rose petals and the time sequence change of the grinding speed value, so as to adaptively adjust the grinding speed value in real time and accurately based on the change condition of the grinding state, thereby optimizing the grinding effect and efficiency of the rose petals and improving the production quality of the rose petals.

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. The development of deep learning and neural networks provides new solutions and schemes for mining complex mapping relationships between the polishing state time sequence changes of the rose petals and the polishing speed values. Those of ordinary skill in the art will appreciate that a deep learning based deep neural network model may adjust parameters of the deep neural network model by appropriate training strategies, such as by a gradient descent back-propagation algorithm, to enable modeling of complex nonlinear correlations between things, which is obviously suitable for modeling and establishing complex mappings between the grinding state timing changes of the rose petals and the grinding speed values.

Specifically, in the technical scheme of the application, firstly, a grinding state monitoring video of a preset time period is acquired through a camera, and grinding speed values of a plurality of preset time points in the preset time period are acquired. Then, considering that the information amount contained in the grinding state monitoring video is large due to the fact that the grinding state hidden characteristic information about the rose petals at each moment exists in the grinding state monitoring video, overfitting is easily caused in the process of carrying out grinding state characteristic mining subsequently, and classification accuracy is low. Therefore, in the technical solution of the present application, in order to facilitate the subsequent association between the polishing state time-series change feature and the polishing speed value time-series change feature, image frames corresponding to the plurality of predetermined time points are further extracted from the polishing state monitoring video as a plurality of polishing state monitoring key frames, so as to facilitate the subsequent extraction of implicit association feature information about the polishing state at the plurality of predetermined time points.

Then, feature mining of the plurality of grinding state monitoring key frames is performed using a convolutional neural network model as a filter having excellent performance in terms of implicit feature extraction of images to extract grinding state high-dimensional implicit feature information about rose petals in the respective grinding state monitoring key frames, respectively, thereby obtaining a plurality of grinding particle size state feature vectors. It should be noted that, here, the convolutional neural network model as a filter is a depth residual network model.

Further, it is considered that since the high-dimensional implicit features of the grinding state of the rose petals in the respective grinding state monitoring key frames have a time-series correlation relationship in the time dimension, that is, the grinding state of the rose petals is constantly changing in the time dimension, and such time-series change feature information is fine feature information of a small scale, capturing and extraction are difficult, resulting in weaker expressive ability of the time-series dynamic change features of the grinding state features of the rose petals, and influence on control of the grinding speed. Therefore, in the technical scheme of the application, the Euclidean distance value between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors is further calculated, so that the time sequence dynamic change feature information of the high-dimensional implicit feature about the grinding particle size in every two adjacent preset time points in the preset time points is calculated, and the particle size change time sequence feature vector consisting of the Euclidean distance values is obtained.

For the polishing rate values at the plurality of predetermined time points, it is considered that the polishing rate values also have a dynamic change rule in a time dimension, and the polishing rate values have different dynamic change characteristic information at different time period spans within the predetermined time period. Therefore, in the technical scheme of the application, the grinding speed values of the plurality of preset time points are further arranged into the grinding speed input vector according to the time dimension, and then feature mining is carried out in the multi-scale neighborhood feature extraction module, so that dynamic multi-scale neighborhood associated features of the grinding speed values under different time spans are extracted, and a grinding speed time sequence feature vector is obtained.

Then, it is also considered that since the polishing state time-series variation information and the time-series dynamic variation information of the polishing rate value are not obvious in the actual monitoring process, it is desirable to perform the feature expression enhancement on the particle size variation time-series feature vector and the polishing rate time-series feature vector after they are obtained. 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 granularity change time series feature vector and the polishing speed time series feature vector by using the prior distribution of the polishing state information and the polishing speed value, that is, the gaussian distribution, that is, the feature expression enhancement is performed on the granularity change time series feature vector and the polishing speed time series feature vector based on the gaussian density map. Specifically, the Gaussian density maps of the granularity change time sequence feature vector and the grinding speed time sequence feature vector are respectively constructed to obtain a granularity change Gaussian density map and a grinding speed Gaussian density map.

And then calculating the response estimation of the particle size change Gaussian density diagram relative to the grinding speed Gaussian density diagram so as to represent the correlation characteristic distribution information between the grinding state time sequence dynamic change characteristic of the rose petals and the time sequence multi-scale correlation characteristic of the grinding speed, thereby obtaining a response Gaussian density diagram. And then, carrying out Gaussian discretization processing on the responsive Gaussian density map so as not to generate information loss when the data features are amplified, thereby obtaining a classification feature matrix.

Further, the classification feature matrix is subjected to classification processing in a classifier to obtain a classification result which is used for indicating whether the grinding speed value of 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 grinding speed value of the current time point should be increased (first label) and that the grinding speed value of the current time point should be decreased (second label), wherein the classifier determines to which classification label the classification feature vector belongs through a soft maximum function. It should be noted that the first tag p1 and the second tag p2 do not include the concept of artificial setting, and in fact, during the training process, the computer model does not have the concept of "the grinding speed value of the current time point should be increased or should be decreased", which is only two kinds of classification tags, and the probability that the output feature is under the two kinds of classification tags, that is, the sum of p1 and p2 is one. Therefore, the classification result that the polishing rate value should be increased or decreased is actually a class probability distribution converted from classifying the tag into a class classification conforming to the natural law, and the physical meaning of the natural probability distribution of the tag is essentially used instead of the language text meaning that the polishing rate value at the current time point should be increased or decreased. It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control policy label that the polishing speed value of the current time point should be increased or decreased, so that after the classification result is obtained, the polishing speed value of the current time point can be adaptively adjusted to be increased or decreased based on the classification result, so as to optimize the polishing effect and efficiency of the rose petals and improve the production quality of the rose petals.

In particular, in the technical solution of the present application, herein, when the classification feature matrix is obtained by calculating the responsiveness estimation of the granularity variation time series feature vector with respect to the grinding speed time series feature vector based on a gaussian density chart, the randomness introduced in the gaussian discretization process is considered so that there is a distribution disturbance in the probability density distribution direction of the classification feature matrix perpendicular to the time series distribution directions of the granularity variation time series feature vector and the grinding speed time series feature vector. Therefore, if the feature learning association degree of the Gaussian density map in the feature space represented by the time sequence-probability density cross dimension in the calculation response estimation process can be improved, the feature expression effect of the obtained classification feature matrix can be obviously improved, and the accuracy of the classification result of the classification feature matrix obtained by the classifier can be improved.

Based on the above, in the training process, the classification feature matrix M is subjected to fourier-like scale domain probability correction, which is specifically expressed as:

wherein m is _i，j E M is the eigenvalue of the (i, j) th position of the classification feature matrix M, W and H are the height and width of the classification feature matrix M, respectively, and α and β are the hyper-parameters for scale adjustment.

Here, the fourier-scale-domain-like probability correction considers the homology of the high-dimensional feature distribution and the scale domain where the high-dimensional feature distribution is located, and can capture the potential distribution association under the homologous space based on the low-rank constraint of the scale space through the fourier-sparse low-rank transformation of the scale space, so that in the training process of the gaussian density graph, the joint feature learning of the spatial scale coherence with the cross dimension constraint of the feature ensemble is realized while the high-dimensional feature representation in the feature space on the time sequence-probability density cross dimension of the feature value is obtained, and the feature expression effect of the obtained classification feature matrix is improved by improving the learning association degree of the gaussian density graph under the overall time sequence-probability density dimension, so that the accuracy of the classification result obtained by the classifier of the classification feature matrix is improved. Therefore, the grinding speed value can be adaptively adjusted in real time and accurately based on the change condition of the grinding state, so that the grinding effect and efficiency of the rose petals are optimized, and the production quality of the rose petals is improved.

Fig. 1 is a schematic view of a processing method of rose petals according to an embodiment of the present application. As shown in fig. 1, in the application scenario, first, a grinding state monitoring video (e.g., C1 as illustrated in fig. 1) of a predetermined period of time acquired by a camera is acquired, and grinding speed values (e.g., C2 as illustrated in fig. 1) at a plurality of predetermined points of time within the predetermined period of time; then, the acquired grinding state monitoring video and grinding speed value are input into a server (e.g., S as illustrated in fig. 1) deployed with a rose petal processing algorithm, wherein the server is capable of processing the grinding state monitoring video and the grinding speed value based on the rose petal processing algorithm to generate a classification result indicating that the grinding speed value at the current point in time should be increased or decreased.

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.

In one embodiment of the present application, fig. 2 is a flow chart of a rose petal processing method according to an embodiment of the present application. As shown in fig. 2, the rose petal processing method 100 according to the embodiment of the present application includes: 110, acquiring a grinding state monitoring video of a preset time period acquired by a camera, and grinding speed values of a plurality of preset time points in the preset time period; 120, extracting image frames corresponding to the plurality of preset time points from the grinding state monitoring video as a plurality of grinding state monitoring key frames; 130, passing the plurality of grinding state monitoring key frames through a convolutional neural network model serving as a filter to obtain a plurality of grinding granularity state feature vectors; 140, calculating euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors to obtain a particle size change time sequence feature vector composed of a plurality of euclidean distance values; 150, arranging the grinding speed values of the plurality of preset time points into a grinding speed input vector according to a time dimension, and then obtaining a grinding speed time sequence feature vector through a multi-scale neighborhood feature extraction module; 160, calculating a response estimate of the granularity variation time sequence feature vector relative to the grinding speed time sequence feature vector to obtain a classification feature matrix; and 170, passing the classification characteristic matrix through a classifier to obtain a classification result, wherein the classification result is used for indicating that the grinding speed value of the current time point is increased or decreased.

Fig. 3 is a schematic diagram of a processing method of rose petals according to an embodiment of the present application. As shown in fig. 3, in the network architecture, firstly, a grinding state monitoring video of a predetermined time period acquired by a camera is acquired, and grinding speed values of a plurality of predetermined time points in the predetermined time period are acquired; then, extracting image frames corresponding to the plurality of preset time points from the grinding state monitoring video as a plurality of grinding state monitoring key frames; then, the grinding state monitoring key frames are respectively passed through a convolutional neural network model serving as a filter to obtain a plurality of grinding particle size state feature vectors; then, calculating Euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors to obtain a particle size change time sequence feature vector composed of the Euclidean distance values; then, arranging the grinding speed values of the plurality of preset time points into a grinding speed input vector according to a time dimension, and then obtaining a grinding speed time sequence feature vector through a multi-scale neighborhood feature extraction module; then, calculating the response estimation of the granularity change time sequence feature vector relative to the grinding speed time sequence feature vector to obtain a classification feature matrix; and finally, the classification feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the grinding speed value of the current time point should be increased or decreased.

Specifically, in step 110, a grinding state monitoring video of a predetermined time period acquired by a camera is acquired, and grinding speed values of a plurality of predetermined time points within the predetermined time period are acquired. As described above, in the conventional polishing apparatus for processing rose petals, the polishing process is usually performed by merely placing the rose petals in the polishing apparatus at the time of use and by fixing the polishing time and the polishing rate, and there is no concern about the suitability of the polishing state and the polishing rate of the rose petals. That is, when the addition amount or quality of the rose petals is different, the grinding state is changed differently in different time periods, and at this time, different grinding speeds are required to optimize the grinding efficiency and effect, so as to improve the processing quality of the rose petals. Thus, an optimized rose petal processing regimen is desired.

Specifically, in the technical scheme of the application, firstly, a grinding state monitoring video of a preset time period is acquired through a camera, and grinding speed values of a plurality of preset time points in the preset time period are acquired.

Specifically, in step 120, image frames corresponding to the plurality of predetermined time points are extracted from the grinding state monitoring video as a plurality of grinding state monitoring key frames. Then, considering that the information amount contained in the grinding state monitoring video is large due to the fact that the grinding state hidden characteristic information about the rose petals at each moment exists in the grinding state monitoring video, overfitting is easily caused in the process of carrying out grinding state characteristic mining subsequently, and classification accuracy is low.

Therefore, in the technical solution of the present application, in order to facilitate the subsequent association between the polishing state time-series change feature and the polishing speed value time-series change feature, image frames corresponding to the plurality of predetermined time points are further extracted from the polishing state monitoring video as a plurality of polishing state monitoring key frames, so as to facilitate the subsequent extraction of implicit association feature information about the polishing state at the plurality of predetermined time points.

Specifically, in step 130, the plurality of grinding status monitoring key frames are respectively passed through a convolutional neural network model as a filter to obtain a plurality of grinding granularity status feature vectors. Then, feature mining of the plurality of grinding state monitoring key frames is performed using a convolutional neural network model as a filter having excellent performance in terms of implicit feature extraction of images to extract grinding state high-dimensional implicit feature information about rose petals in the respective grinding state monitoring key frames, respectively, thereby obtaining a plurality of grinding particle size state feature vectors. It should be noted that, here, the convolutional neural network model as a filter is a depth residual network model.

The step of passing the plurality of grinding state monitoring key frames through a convolutional neural network model serving as a filter to obtain a plurality of grinding granularity state feature vectors comprises the following steps: each layer of the 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; the output of the last layer of the convolutional neural network model serving as the filter is the plurality of grinding granularity state characteristic vectors, and the input of the first layer of the convolutional neural network model serving as the filter is the plurality of grinding state monitoring key frames.

The convolutional neural network (Convolutional Neural Network, CNN) is an artificial neural network and has wide application in the fields of image recognition and the like. The convolutional neural network may include an input layer, a hidden layer, and an output layer, where the hidden layer may include a convolutional layer, a pooling layer, an activation layer, a full connection layer, etc., where the previous layer performs a corresponding operation according to input data, outputs an operation result to the next layer, and obtains a final result after the input initial data is subjected to a multi-layer operation.

The convolutional neural network model has excellent performance in the aspect of image local feature extraction by taking a convolutional kernel as a feature filtering factor, and has stronger feature extraction generalization capability and fitting capability compared with the traditional image feature extraction algorithm based on statistics or feature engineering.

Specifically, in step 140, a euclidean distance value between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors is calculated to obtain a granularity variation timing feature vector composed of a plurality of euclidean distance values. Further, it is considered that since the high-dimensional implicit features of the grinding state of the rose petals in the respective grinding state monitoring key frames have a time-series correlation relationship in the time dimension, that is, the grinding state of the rose petals is constantly changing in the time dimension, and such time-series change feature information is fine feature information of a small scale, capturing and extraction are difficult, resulting in weaker expressive ability of the time-series dynamic change features of the grinding state features of the rose petals, and influence on control of the grinding speed.

Therefore, in the technical scheme of the application, the Euclidean distance value between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors is further calculated, so that the time sequence dynamic change feature information of the high-dimensional implicit feature about the grinding particle size in every two adjacent preset time points in the preset time points is calculated, and the particle size change time sequence feature vector consisting of the Euclidean distance values is obtained.

The calculating the euclidean distance value between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors to obtain a particle size change time sequence feature vector composed of a plurality of euclidean distance values comprises the following steps: calculating Euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors by using the following distance calculation formula to obtain a plurality of Euclidean distance values; wherein, the distance formula is:

Specifically, in step 150, the polishing speed values at the plurality of predetermined time points are arranged according to a time dimension to form a polishing speed input vector, and then the polishing speed input vector is passed through a multi-scale neighborhood feature extraction module to obtain a polishing speed time sequence feature vector. For the polishing rate values at the plurality of predetermined time points, it is considered that the polishing rate values also have a dynamic change rule in a time dimension, and the polishing rate values have different dynamic change characteristic information at different time period spans within the predetermined time period.

Therefore, in the technical scheme of the application, the grinding speed values of the plurality of preset time points are further arranged into the grinding speed input vector according to the time dimension, and then feature mining is carried out in the multi-scale neighborhood feature extraction module, so that dynamic multi-scale neighborhood associated features of the grinding speed values under different time spans are extracted, and a grinding speed time sequence feature vector is obtained.

The multi-scale neighborhood feature extraction module comprises: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use one-dimensional convolution kernels with different scales.

Further, the polishing speed values at the plurality of predetermined time points are arranged into a polishing speed input vector according to a time dimension, and then the polishing speed input vector is passed through a multi-scale neighborhood feature extraction module to obtain a polishing speed time sequence feature vector, which comprises the following steps: performing one-dimensional convolution coding on the grinding speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula to obtain a first-scale grinding speed feature vector; wherein the first convolution formula is:

It should be noted that the multi-scale neighborhood feature extraction module is essentially a deep neural network model based on deep learning, which is capable of fitting any function by a predetermined training strategy and has a higher feature extraction generalization capability compared to the conventional feature engineering.

The multi-scale neighborhood feature extraction module comprises a plurality of parallel one-dimensional convolution layers, wherein in the process of feature extraction by the multi-scale neighborhood feature extraction module, the plurality of parallel one-dimensional convolution layers respectively carry out one-dimensional convolution coding on input data by one-dimensional convolution check with different scales so as to capture local implicit features of a sequence.

Specifically, in step 160, a responsiveness estimate of the granularity variation timing feature vector relative to the polishing rate timing feature vector is calculated to obtain a classification feature matrix. Then, it is also considered that since the polishing state time-series variation information and the time-series dynamic variation information of the polishing rate value are not obvious in the actual monitoring process, it is desirable to perform the feature expression enhancement on the particle size variation time-series feature vector and the polishing rate time-series feature vector after they are obtained.

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 granularity change time series feature vector and the polishing speed time series feature vector by using the prior distribution of the polishing state information and the polishing speed value, that is, the gaussian distribution, that is, the feature expression enhancement is performed on the granularity change time series feature vector and the polishing speed time series feature vector based on the gaussian density map. Specifically, the Gaussian density maps of the granularity change time sequence feature vector and the grinding speed time sequence feature vector are respectively constructed to obtain a granularity change Gaussian density map and a grinding speed Gaussian density map.

Wherein calculating a responsiveness estimate of the granularity variation timing feature vector relative to the polishing rate timing feature vector to obtain a classification feature matrix comprises: constructing a granularity change Gaussian density map of the granularity change time sequence feature vector according to the following first Gaussian formula; wherein, the first gaussian formula is:

wherein mu ₃ Mean vector, sigma, representing the responsive gaussian density map ₃ Covariance matrix representing the response Gaussian density map, +.The vector point-multiply, +.1 indicates the reciprocal of the value at each position of the vector, andrepresenting a matrix multiplication; carrying out Gaussian discretization processing on the Gaussian distribution of each position in the responsive Gaussian density map so as to reduce the Gaussian distribution of each position in the responsive Gaussian density map into a one-dimensional feature vector; and two-dimensionally arranging the one-dimensional feature vectors of the positions to obtain the classification feature matrix.

Specifically, in step 170, the classification feature matrix is passed through a classifier to obtain a classification result, which is used to indicate whether the polishing rate value at the current time point should be increased or decreased. Further, the classification feature matrix is subjected to classification processing in a classifier to obtain a classification result which is used for indicating whether the grinding speed value of 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 grinding speed value of the current time point should be increased (first label) and that the grinding speed value of the current time point should be decreased (second label), wherein the classifier determines to which classification label the classification feature vector belongs through a soft maximum function. It should be noted that the first tag p1 and the second tag p2 do not include the concept of artificial setting, and in fact, during the training process, the computer model does not have the concept of "the grinding speed value of the current time point should be increased or should be decreased", which is only two kinds of classification tags, and the probability that the output feature is under the two kinds of classification tags, that is, the sum of p1 and p2 is one. Therefore, the classification result that the polishing rate value should be increased or decreased is actually a class probability distribution converted from classifying the tag into a class classification conforming to the natural law, and the physical meaning of the natural probability distribution of the tag is essentially used instead of the language text meaning that the polishing rate value at the current time point should be increased or decreased.

It should be understood that, in the technical solution of the present application, the classification label of the classifier is a control policy label that the polishing speed value of the current time point should be increased or decreased, so that after the classification result is obtained, the polishing speed value of the current time point can be adaptively adjusted to be increased or decreased based on the classification result, so as to optimize the polishing effect and efficiency of the rose petals and improve the production quality of the rose petals.

Further, the classification feature matrix is passed through a classifier to obtain a classification result, wherein the classification result is used for indicating that the grinding speed value of the current time point should be increased or decreased, and the classification result comprises the following steps: expanding the classification feature matrix into classification feature 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.

In one embodiment of the present application, the processing method of rose petals further includes training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier; fig. 4 is a flowchart of a sub-step of step 180 in the processing method for processing rose petals according to an embodiment of the present application, where training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier, as shown in fig. 4, includes: 181, acquiring training data, wherein the training data comprises training grinding state monitoring videos of a preset time period, training grinding speed values of a plurality of preset time points in the preset time period, and a true value of the grinding speed value of the current time point to be increased or decreased; 182, extracting training image frames corresponding to the predetermined time points from the training grinding state monitoring video as a plurality of training grinding state monitoring key frames; 183, respectively passing the training grinding state monitoring key frames through the convolutional neural network model serving as a filter to obtain a plurality of training grinding granularity state feature vectors; 184, calculating euclidean distance values between every two adjacent training granularity state feature vectors in the plurality of training granularity state feature vectors to obtain a training granularity change time sequence feature vector composed of a plurality of euclidean distance values; 185, arranging the training grinding speed values of the plurality of preset time points into training grinding speed input vectors according to a time dimension, and then obtaining training grinding speed time sequence feature vectors through the multi-scale neighborhood feature extraction module; 186, calculating a response estimate of the training granularity variation time sequence feature vector relative to the training grinding speed time sequence feature vector to obtain a training classification feature matrix; 187, performing class fourier scale domain probability correction on the training classification feature matrix to obtain an optimized training classification feature matrix; 188, passing the optimized training classification characteristic matrix through the classifier to obtain a classification loss function value; and, 189 training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier with the classification loss function value as a loss function value and by back propagation of gradient descent.

Based on the above, in the training process, the classification feature matrix M is subjected to fourier-like scale domain probability correction, which is specifically expressed as: carrying out class Fourier scale domain probability correction on the training classification feature matrix by using the following optimization formula to obtain the optimized training classification feature matrix; wherein, the optimization formula is:

Further, passing the optimized training classification feature matrix through the classifier to obtain a classification loss function value, including: the classifier processes the optimized training classification feature matrix with the following classification formula to generate a training classification result, wherein the classification formula is: softmax { (W) _n ，B _n )：...：(W ₁ ，B ₁ ) X, where X represents the optimized training classification feature matrix, W ₁ To W _n Is a weight matrix, B ₁ To B _n Representing a bias matrix; and calculating a cross entropy value between the training classification result and a true value as the classification loss function value.

In summary, the rose petal processing method 100 according to the embodiment of the present application is explained, which acquires a grinding state monitoring video of a predetermined period of time acquired by a camera, and grinding speed values at a plurality of predetermined time points within the predetermined period of time; by adopting an artificial intelligence technology based on deep learning, a mapping relation between the time sequence change of the grinding state of the rose petals and the time sequence change of the grinding speed value is established, so that the grinding speed value is adaptively adjusted based on the characteristic of the time sequence change of the grinding state of the rose petals, thereby realizing intelligent processing control, optimizing the grinding effect and efficiency of the rose petals, and improving the production quality of the rose petals.

In one embodiment of the present application, fig. 5 is a block diagram of a rose petal processing system according to an embodiment of the present application. As shown in fig. 5, the rose petal processing system 200 according to an embodiment of the present application includes: a data acquisition module 210, configured to acquire a grinding status monitoring video of a predetermined time period acquired by a camera, and grinding speed values of a plurality of predetermined time points in the predetermined time period; a key frame extracting module 220, configured to extract, from the grinding state monitoring video, image frames corresponding to the plurality of predetermined time points as a plurality of grinding state monitoring key frames; the feature extraction module 230 is configured to pass the plurality of grinding state monitoring key frames through a convolutional neural network model serving as a filter to obtain a plurality of grinding granularity state feature vectors; the euclidean distance calculating module 240 is configured to calculate euclidean distance values between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors to obtain a granularity variation time sequence feature vector composed of a plurality of euclidean distance values; the multi-scale feature extraction module 250 is configured to arrange the polishing speed values of the plurality of predetermined time points into a polishing speed input vector according to a time dimension, and then obtain a polishing speed time sequence feature vector through the multi-scale neighborhood feature extraction module; a responsiveness estimation calculation module 260, configured to calculate a responsiveness estimation of the granularity variation time sequence feature vector relative to the polishing speed time sequence feature vector to obtain a classification feature matrix; and the grinding speed value control module is used for enabling the classification feature matrix to pass through a classifier to obtain a classification result, wherein the classification result is used for indicating whether the grinding speed value of the current time point should be increased or decreased.

In a specific example, in the above rose petal processing system, the feature extraction module is configured to: each layer of the 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; the output of the last layer of the convolutional neural network model serving as the filter is the plurality of grinding granularity state characteristic vectors, and the input of the first layer of the convolutional neural network model serving as the filter is the plurality of grinding state monitoring key frames.

In a specific example, in the above rose petal processing system, the euclidean distance calculating module may further include: a distance calculating unit, configured to calculate euclidean distance values between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors according to the following distance calculation formula to obtain a plurality of euclidean distance values; wherein, the distance formula is:

Wherein V is _i And V _j Representing every adjacent two of the plurality of particle size status feature vectors,and->Representing the grinding particle size state characteristic vector V _i And V _j Characteristic value of each position, d (V _i ，V _j ) Representing a Euclidean distance between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors; and a vector arrangement unit configured to arrange the plurality of euclidean distance values to obtain the granularity variation timing feature vector.

In a specific example, in the above rose petal processing system, the multi-scale neighborhood feature extraction module includes: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use one-dimensional convolution kernels with different scales.

In a specific example, in the above rose petal processing system, the multi-scale feature extraction module includes: the first scale unit is used for carrying out one-dimensional convolution coding on the grinding speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula so as to obtain a first scale grinding speed feature vector; wherein the first convolution formula is:

Wherein a is the width of the first one-dimensional convolution kernel in the X direction, F (a) is a first one-dimensional convolution kernel parameter vector, G (X-a) is a local vector matrix calculated by a first one-dimensional convolution kernel function, w is the size of the first one-dimensional convolution kernel, X represents the grinding speed input vector, and Cov (X) represents one-dimensional convolution encoding of the grinding speed input vector; the second scale unit is used for carrying out one-dimensional convolution coding on the grinding speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula so as to obtain a second scale grinding speed feature vector; wherein the second convolution 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 function of the second one-dimensional convolution kernel, m is the size of the second one-dimensional convolution kernel, X represents the grinding speed input vector, and Cov (X) represents one-dimensional convolution encoding of the grinding speed input vector; and the cascading unit is used for cascading the first-scale grinding speed feature vector and the second-scale grinding speed feature vector by using the fusion layer of the multi-scale neighborhood feature extraction module so as to obtain the grinding speed time sequence feature vector.

In a specific example, in the above rose petal processing system, the responsiveness estimation calculation module includes: a gaussian construction unit for constructing a granularity-changing gaussian density map of the granularity-changing time-series feature vector with a first gaussian formula as follows; wherein, the first gaussian formula is:

wherein mu ₂ Representing the polishing rate time sequence feature vector, and sigma ₂ The value of each position of the polishing rate time sequence feature vector represents the variance between the feature values of the corresponding two positions; a responsiveness unit for calculating a responsiveness estimate of the particle size change gaussian density map relative to the grinding speed gaussian density map with a responsiveness formula to obtain a responsiveness gaussian density map; wherein, the responsiveness formula is:

wherein mu ₃ Mean vector, sigma, representing the responsive gaussian density map ₃ Covariance matrix representing the response Gaussian density map, +.The vector point multiplication, +.1 represents each bit of the vectorThe value of the setting is inverted andrepresenting a matrix multiplication; the Gaussian discrete unit is used for carrying out Gaussian discrete processing on the Gaussian distribution of each position in the responsive Gaussian density map so as to reduce the Gaussian distribution of each position in the responsive Gaussian density map into a one-dimensional feature vector; and a two-dimensional arrangement unit, configured to two-dimensionally arrange the one-dimensional feature vectors of the respective positions to obtain the classification feature matrix.

In a specific example, in the rose petal processing system, the processing system further includes a training module for training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module and the classifier; wherein, training module includes: the training data acquisition unit is used for acquiring training data, wherein the training data comprises training grinding state monitoring videos of a preset time period, training grinding speed values of a plurality of preset time points in the preset time period and a true value of which the grinding speed value of the current time point is increased or reduced; the training key frame extraction unit is used for extracting training image frames corresponding to the preset time points from the training grinding state monitoring video to serve as a plurality of training grinding state monitoring key frames; the training feature extraction unit is used for enabling the training grinding state monitoring key frames to respectively pass through the convolutional neural network model serving as a filter so as to obtain a plurality of training grinding granularity state feature vectors; the training Euclidean distance calculation unit is used for calculating Euclidean distance values between every two adjacent training abrasive grain state characteristic vectors in the plurality of training abrasive grain state characteristic vectors so as to obtain training granularity change time sequence characteristic vectors composed of a plurality of Euclidean distance values; the training multi-scale feature extraction unit is used for arranging training grinding speed values of the plurality of preset time points into training grinding speed input vectors according to time dimension, and then obtaining training grinding speed time sequence feature vectors through the multi-scale neighborhood feature extraction module; the training response estimation calculation unit is used for calculating the response estimation of the training granularity change time sequence feature vector relative to the training grinding speed time sequence feature vector so as to obtain a training classification feature matrix; the training optimization unit is used for carrying out class Fourier scale domain probability correction on the training classification characteristic matrix to obtain an optimized training classification characteristic matrix; the classification loss function value calculation unit is used for enabling the optimization training classification feature matrix to pass through the classifier to obtain a classification loss function value; and a training unit for training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module and the classifier by using the classification loss function value as a loss function value and by back propagation of gradient descent.

In a specific example, in the above rose petal processing system, the training optimizing unit is configured to: carrying out class Fourier scale domain probability correction on the training classification feature matrix by using the following optimization formula to obtain the optimized training classification feature matrix; wherein, the optimization formula is:

wherein m is _i，j Is the eigenvalue of the (i, j) th position of the training classification feature matrix, W and H are the height and width of the training classification feature matrix, respectively, and alpha and beta are hyper-parameters for scale adjustment, exp (magnitude) represents calculating a natural exponential function value exponentiated by the value, m' _i，j Is the eigenvalue of the (i, j) th position of the optimization training classification eigenvalue matrix.

In a specific example, in the above rose petal processing system, the classification loss function value calculation unit includes: the classifying subunit is configured to process the optimized training classifying feature matrix by using the following classifying formula to generate a training classifying result, where the classifying formula is: softmax { (W) _n ，B _n )：...：(W ₁ ，B ₁ ) X, where X represents the optimized training classification feature matrix, W ₁ To W _n Is a weight matrix, B ₁ To B _n Representing a bias matrix; and a calculation subunit for calculating a cross entropy value between the training classification result and a true value as the classification loss function value.

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 rose petal processing system have been described in detail in the above description of the rose petal processing method with reference to fig. 1 to 4, and thus, repetitive descriptions thereof will be omitted.

As described above, the rose petal processing system 200 according to the embodiment of the present application can be implemented in various terminal devices, such as a server or the like for rose petal processing. In one example, the rose petal processing system 200 according to an embodiment of the present application may be integrated into the terminal device as a software module and/or hardware module. For example, the rose petal processing system 200 may be a software module in the operating system of the terminal device or may be an application developed for the terminal device; of course, the rose petal processing system 200 could equally be one of a number of hardware modules of the terminal device.

Alternatively, in another example, the rose petal processing system 200 and the terminal device may be separate devices, and the rose petal processing system 200 may be connected to the terminal device via a wired and/or wireless network and communicate the interaction information in accordance with a agreed data format.

The present application also provides a computer program product comprising instructions which, when executed, cause an apparatus to perform operations corresponding to the above-described method.

In one embodiment of the present application, there is also provided a computer-readable storage medium storing a computer program for executing the above-described method.

It should be appreciated that embodiments of the present application may be provided as a method, system, or computer program product. Accordingly, the forms of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects may be utilized. Furthermore, the computer program product may take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) having computer-usable program code embodied therein.

Methods, systems, and computer program products of embodiments of the present application are described in the flow diagrams and/or block diagrams. It will be understood that each flow and/or block of the flowchart illustrations and/or block diagrams, and combinations of flows and/or blocks in the flowchart illustrations and/or block diagrams, can be implemented by computer program instructions. These computer program instructions may be provided to a processor of a general purpose computer, special purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create means for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be stored in a computer-readable memory that can direct a computer or other programmable data processing apparatus to function in a particular manner, such that the instructions stored in the computer-readable memory produce an article of manufacture including instruction means which implement the function specified in the flowchart flow or flows and/or block diagram block or blocks.

These computer program instructions may also be loaded onto a computer or other programmable data processing apparatus to cause a series of operational steps to be performed on the computer or other programmable apparatus to produce a computer implemented process such that the instructions which execute on the computer or other programmable apparatus provide steps for implementing the functions specified in the flowchart flow or flows and/or block diagram block or blocks.

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 intended to be limiting, and these advantages, benefits, effects, etc. are not to be considered as essential to 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 necessarily limited to practice with the above described specific details.

The block diagrams of the devices, apparatuses, devices, systems referred to in the present 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 aspects of 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.

Finally, it is further noted that relational terms such as first and second, and the like are used solely to distinguish one entity or action from another entity or action without necessarily requiring or implying any actual such relationship or order between such entities or actions. Moreover, the terms "comprises," "comprising," or any other variation thereof, are intended to cover a non-exclusive inclusion, such that a process, method, article, or terminal that comprises a list of elements does not include only those elements but may include other elements not expressly listed or inherent to such process, method, article, or terminal. Without further limitation, an element defined by the phrase "comprising one … …" does not exclude the presence of other like elements in a process, method, article or terminal device comprising the element.

The foregoing description has been presented for purposes of illustration and description. Furthermore, this description is not intended to limit 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. The processing method of the rose petals is characterized by comprising the following steps:

2. The processing method of rose petals according to claim 1, wherein passing the plurality of grinding state monitoring key frames through a convolutional neural network model as a filter to obtain a plurality of grinding particle size state feature vectors, respectively, comprises: each layer of the 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;

the output of the last layer of the convolutional neural network model serving as the filter is the plurality of grinding granularity state characteristic vectors, and the input of the first layer of the convolutional neural network model serving as the filter is the plurality of grinding state monitoring key frames.

3. The method according to claim 2, wherein calculating euclidean distance values between every adjacent two of the plurality of grinding particle size state feature vectors to obtain a particle size variation timing feature vector composed of a plurality of euclidean distance values, comprises:

Calculating Euclidean distance values between every two adjacent grinding particle size state feature vectors in the grinding particle size state feature vectors by using the following distance calculation formula to obtain a plurality of Euclidean distance values;

wherein, the distance formula is:

wherein V is _i And V _j Representing every adjacent two of the plurality of particle size status feature vectors,and->Representing the grinding particle size state characteristic vector V _i And V _j Characteristic value of each position, d (V _i ，V _j ) Representing a Euclidean distance between every two adjacent grinding particle size state feature vectors in the plurality of grinding particle size state feature vectors; and

and arranging the Euclidean distance values to obtain the granularity change time sequence feature vector.

4. The rose petal processing method according to claim 3, wherein the multi-scale neighborhood feature extraction module comprises: and a fusion layer connected with the first convolution layer and the second convolution layer, wherein the first convolution layer and the second convolution layer use one-dimensional convolution kernels with different scales.

5. The method according to claim 4, wherein the step of arranging the polishing rate values at the predetermined time points in the time dimension into the polishing rate input vector and then passing the polishing rate input vector through the multi-scale neighborhood feature extraction module to obtain the polishing rate time sequence feature vector comprises:

Performing one-dimensional convolution coding on the grinding speed input vector by using a first convolution layer of the multi-scale neighborhood feature extraction module according to the following first convolution formula to obtain a first-scale grinding speed feature vector;

wherein the first convolution formula is:

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

performing one-dimensional convolution coding on the grinding speed input vector by using a second convolution layer of the multi-scale neighborhood feature extraction module according to the following second convolution formula to obtain a second-scale grinding speed feature vector;

wherein the second convolution 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 function of the second one-dimensional convolution kernel, m is the size of the second one-dimensional convolution kernel, the grinding speed input vector is represented, and Cov (X) represents one-dimensional convolution encoding of the grinding speed input vector; and

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

6. The method of claim 5, wherein calculating a responsiveness estimate of the time series feature vector for the change in particle size relative to the time series feature vector for the grinding speed to obtain a classification feature matrix comprises:

constructing a granularity change Gaussian density map of the granularity change time sequence feature vector according to the following first Gaussian formula;

wherein, the first gaussian formula is:

wherein mu ₁ Representing the granularity variation timing feature vector, and sigma ₁ Representing the variance between the eigenvalues of the corresponding two locations in the granularity variation time sequence eigenvector;

constructing a grinding speed Gaussian density chart of the grinding speed time sequence characteristic vector according to the following second Gaussian formula;

wherein, the second gaussian formula is:

wherein mu ₂ Representing the polishing rate time sequence feature vector, and sigma ₂ The value of each position of the polishing rate time sequence feature vector represents the variance between the feature values of the corresponding two positions;

Calculating a responsiveness estimate of the particle size change gaussian density map relative to the grinding speed gaussian density map with a responsiveness formula to obtain a responsiveness gaussian density map;

wherein, the responsiveness formula is:

wherein mu ₃ Mean vector, sigma, representing the responsive gaussian density map ₃ Covariance matrix representing the response Gaussian density map, +.The vector point-multiply, +.1 indicates the reciprocal of the value at each position of the vector, andrepresenting a matrix multiplication;

carrying out Gaussian discretization processing on the Gaussian distribution of each position in the responsive Gaussian density map so as to reduce the Gaussian distribution of each position in the responsive Gaussian density map into a one-dimensional feature vector; and

and carrying out two-dimensional arrangement on the one-dimensional feature vectors of each position to obtain the classification feature matrix.

7. The rose petal processing method according to claim 6, further comprising training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier;

wherein training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier comprises:

Acquiring training data, wherein the training data comprises training grinding state monitoring videos of a preset time period, training grinding speed values of a plurality of preset time points in the preset time period, and a true value of the grinding speed value of the current time point to be increased or decreased;

extracting training image frames corresponding to the preset time points from the training grinding state monitoring video to serve as a plurality of training grinding state monitoring key frames;

respectively passing the training grinding state monitoring key frames through the convolutional neural network model serving as a filter to obtain a plurality of training grinding granularity state feature vectors;

calculating Euclidean distance values between every two adjacent training granularity state feature vectors in the training granularity state feature vectors to obtain training granularity change time sequence feature vectors composed of the Euclidean distance values;

the training grinding speed values of the plurality of preset time points are arranged into training grinding speed input vectors according to time dimensions, and then the training grinding speed input vectors are passed through the multi-scale neighborhood feature extraction module to obtain training grinding speed time sequence feature vectors;

calculating the response estimation of the training granularity change time sequence feature vector relative to the training grinding speed time sequence feature vector to obtain a training classification feature matrix;

Performing class Fourier scale domain probability correction on the training classification feature matrix to obtain an optimized training classification feature matrix;

the optimized training classification characteristic matrix passes through the classifier to obtain a classification loss function value; and

training the convolutional neural network model as a filter, the multi-scale neighborhood feature extraction module, and the classifier with the classification loss function value as a loss function value and by back propagation of gradient descent.

8. The method of claim 7, wherein performing fourier-like scale domain probability correction on the training classification feature matrix to obtain an optimized training classification feature matrix comprises:

carrying out class Fourier scale domain probability correction on the training classification feature matrix by using the following optimization formula to obtain the optimized training classification feature matrix;

wherein, the optimization formula is:

9. The method of claim 8, wherein passing the optimized training classification feature matrix through the classifier to obtain a classification loss function value comprises:

the classifier processes the optimized training classification feature matrix with the following classification formula to generate a training classification result, wherein the classification formula is: softmax { (W) _n ，B _n )：...：(W ₁ ，B ₁ ) X, where X represents the optimized training classification feature matrix, W ₁ To W _n Is a weight matrix, B ₁ To B _n Representing a bias matrix; and

and calculating a cross entropy value between the training classification result and a true value as the classification loss function value.

10. A rose petal processing system, comprising: