Disclosure of Invention
The embodiment of the application provides a commodity purchase event prediction method and device, which are used for solving the problem that in the prior art, the matching degree between a commodity to be recommended and the actual demand of a user is low due to single photographing and identification function.
The embodiment of the invention provides a commodity purchase event prediction method, which comprises the following steps:
acquiring commodity images, identifying the commodity images, and determining commodity categories in the commodity images;
acquiring historical information of a commodity purchased by a user, predicting the purchase event by using an artificial intelligence method based on the historical information and the commodity category, and predicting the purchase probability of the commodity category purchased by the user, wherein the historical information comprises the time, region, price and frequency of the commodity purchased by the user;
and if the purchase probability is larger than a preset threshold value, recommending the commodity category to the user, so that the user refers to and purchases the commodity category.
Optionally, the acquiring the historical information of the user purchasing the commodity, and predicting the purchasing event by using an artificial intelligence method based on the historical information and the commodity category includes:
acquiring historical information of the commodity purchased by the user by utilizing a multidimensional Hawkes process;
the historical information is associated with the commodity category based on a graph convolution neural network, and an association matrix is obtained;
inputting the incidence matrix into a cyclic neural network model, and predicting the purchase event through the cyclic neural network model.
Optionally, the historical information of the user purchasing goods is obtained by using a multidimensional Hawkes process; the historical information is associated with the commodity category based on a graph convolution neural network, and an association matrix is obtained; inputting the incidence matrix into a cyclic neural network model, and predicting the purchase event through the cyclic neural network model, wherein the method comprises the following steps:
according to the dimension of the time sequence, the user and the historical purchased goods form an association sequence;
establishing a correlation model, inputting the correlation sequence into the correlation model, and outputting feature vectors of the user and the historical purchased goods;
inputting the feature vector into the graph convolution neural network to acquire graph adjacent information;
inputting the graph adjacent information into a GCN network, and outputting a space-time correlation matrix of historical purchased goods and purchase time by using the GCN network;
and inputting the time-space correlation matrix of the historical purchased goods and the purchase time into a cyclic neural network model to output the purchase probability of the goods category.
Optionally, before inputting the feature vector into the graph convolution neural network to obtain graph adjacency information, the method further includes:
constructing the graph roll-up neural network, comprising:
a graph embedding stage, wherein a double-layer parallel graph convolutional neural network is used for capturing graph space characteristics;
in the fourier domain, constructing a filter for capturing spatial features of nodes in the graph, traversing the filter through each node in the graph;
and after the spatial characteristics of each node are acquired, constructing a graph convolution neural network model by stacking a plurality of group convolution layers.
Optionally, identifying the commodity image, and determining the commodity category in the commodity image includes:
performing geometric analysis and cluster analysis on the commodity image, identifying fitting data of the commodity outline, and determining the position of the commodity outline on the image;
extracting the outline of the commodity, and identifying the type of the extracted commodity;
the geometric analysis and cluster analysis processing are carried out on the commodity image, and the fitting data of the commodity outline are identified, which comprises the following steps:
performing binarization processing on the commodity image;
carrying out connected region scanning on the binarized image;
performing contour screening on the scanned communication area;
splitting the screened profile to obtain fitting data of a profile boundary;
performing least square fitting on the fitting data to determine the position of the commodity outline in the commodity image;
the method for extracting the contour of the commodity and identifying the type of the extracted commodity comprises the following steps:
establishing a depth residual error network model;
inputting the commodity image into the depth residual error network model end, sequentially carrying out feature extraction, characterization enhancement, target detection and gesture measurement on the commodity image through the depth residual error network, outputting and identifying a target area of the commodity, and determining the commodity category.
Optionally, before the characterizing enhancement of the commodity image by the depth residual network, the method further comprises:
and sequentially carrying out brightness extraction, global tone mapping and Laplace domain fusion on the commodity image, and recovering the color of the commodity image.
Optionally, the sequentially performing brightness extraction, global tone mapping, and laplace domain fusion on the commodity image, and recovering the color of the commodity image, including:
converting the commodity image from RGB space to CIEXYZ space, and separating the brightness component of the image;
global tone mapping is carried out on the commodity image by using a Reinhard global tone mapping algorithm, and three compressed brightness images are obtained;
correcting the overexposed area and the overexposed area in the three brightness images by adopting a gamma correction algorithm;
fusing the three brightness images by adopting a double-pyramid frequency domain fusion algorithm, and outputting a final brightness image;
and carrying out image color recovery on the final brightness image.
Optionally, the fusing the three brightness images by using a double pyramid frequency domain fusion algorithm, and outputting a final brightness image, including:
respectively calculating the contrast ratio, the exposure well and the edge information of the three brightness images to serve as three different quality measure factors;
multiplying the three quality measure factors to obtain a weight image;
carrying out normalization processing on the weight image, and then carrying out Gaussian pyramid decomposition on the weight image;
carrying out Laplacian pyramid decomposition on the three input images;
multiplying the images after the two pyramid decomposition according to pixel distribution to obtain a new pyramid;
and carrying out Laplacian pyramid reconstruction on the new pyramid to obtain a final brightness output image.
Optionally, the commodity category is teaching aid or child game commodity.
The embodiment of the invention also provides a device which is characterized by comprising a memory and a processor, wherein the memory stores computer executable instructions, and the processor realizes the method when running the computer executable instructions on the memory.
According to the method provided by the embodiment of the invention, based on the historical information and the commodity category, the purchasing event is predicted by utilizing an artificial intelligence method, specifically, the purchasing event can be predicted by adopting a multidimensional Hoxwell process and a neural network, so that the commodity category is better determined to be recommended or not recommended, personalized commodity recommendation after combining a commodity image recognition function is realized, and the commodity recommendation success rate is increased.
Detailed Description
The following description of the embodiments of the present application will be made clearly and fully with reference to the accompanying drawings, in which it is evident that the embodiments described are some, but not all, of the embodiments of the present application. All other embodiments, which can be made by one of ordinary skill in the art without undue burden from the present disclosure, are within the scope of the present disclosure.
It should be understood that the terms "comprises" and "comprising," when used in this specification and the appended claims, specify the presence of stated features, integers, steps, operations, elements, and/or components, but do not preclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and/or groups thereof.
It is also to be understood that the terminology used in the description of the present application is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in this specification and the appended claims, the singular forms "a," "an," and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise.
It should be further understood that the term "and/or" as used in this specification and the appended claims refers to any and all possible combinations of one or more of the associated listed items, and includes such combinations.
As used in this specification and the appended claims, the term "if" may be interpreted as "when..once" or "in response to a determination" or "in response to detection" depending on the context. Similarly, the phrase "if a determination" or "if a [ described condition or event ] is detected" may be interpreted in the context of meaning "upon determination" or "in response to determination" or "upon detection of a [ described condition or event ]" or "in response to detection of a [ described condition or event ]".
FIG. 1 is a flow chart of one of the product purchase event predictions of an embodiment of the present invention, as shown in FIG. 1, the method comprising:
s101, acquiring commodity images, identifying the commodity images, and determining commodity categories in the commodity images;
in the embodiment of the invention, the commodity image can be acquired through special image acquisition equipment, such as a camera, an intelligent terminal, an internet of things device, an industrial high-definition camera device and the like, and the commodity image can be acquired by shooting the commodity outdoors in daytime or at night or shooting a certain product indoors. For example, the method can take images of commodities such as electric bicycles and shoes outdoors, upload the taken images to a cloud server at the rear end, and perform subsequent image processing and commodity identification through the cloud server.
After acquiring the commodity image, the cloud server identifies the commodity image, determines the commodity category in the commodity image, specifically,
the cloud server needs to perform geometric analysis and cluster analysis processing on the commodity image, identifies fitting data of the commodity outline and determines the position of the commodity outline on the image; and after the contour is identified, contour extraction is carried out on the commodity, and type identification is carried out on the extracted commodity.
The geometric analysis and cluster analysis processing are carried out on the commodity image, and the fitting data of the commodity outline are identified, which comprises the following steps:
s1011, performing binarization processing on the commodity image;
the image binarization algorithm mainly comprises the following three types, and any one of the three types can be adopted in the embodiment of the invention:
the conventional edge detection binarization operator, a common edge detection operator is Roberts, sobel, prewitt, and the characteristic points with larger gray value change in the image are calculated through derivative and differentiation. The method has the characteristics of high speed, simple operation and easy understanding. However, the threshold needs to be manually input to realize edge detection, the quality of the result cannot be accurately measured, and the same selected parameter is not suitable for other image data and lacks universality.
The method selects all gray values in all pixels of an image as objects, takes an average value as a binarized threshold value, determines pixels larger than the threshold value as characteristic points, and the rest are non-characteristic points. And a unified threshold value is selected for the whole image to carry out edge detection, so that the accuracy is poor and partial characteristic information is lost while manual parameter input is avoided.
The method is similar to plane convolution, and the threshold value on the pixel is determined according to the 8 neighborhood block pixel gray value distribution average value of the pixel, wherein the partial self-adaptive threshold value is obtained by calculating different areas of the image by using different threshold values. In the embodiment of the invention, an improved partial self-adaptive threshold edge detection binarization algorithm is adopted, a square window is changed into an annular neighborhood window, the gray value of the pixel point is calculated again according to the area occupied by each pixel point covered by the annular window, the distance weighting is comprehensively considered, and the threshold of the central pixel point is calculated by using the combination of annular windows with different radiuses.
S1012, carrying out connected region scanning on the binarized image;
and after the connected area is scanned, the data such as the contour lines of the area, the perimeter and the like of the area can be obtained.
S1013, carrying out contour screening on the scanned communication area;
and (3) carrying out contour screening on the scanned communication area, namely carrying out quick geometric analysis on the contour extracted in the last step, and only retaining the contour similar to a polygon so as to reduce the processing time of the subsequent step. Specifically, the contour line is smoothed once by using a local average method, then tangent vectors of each point on the contour line are calculated by using a neighbor difference method, and finally all the tangent vector coordinates are analyzed by using a hierarchical clustering method, and if the formed significant clusters are exactly n, the contour is considered to be similar to an n-sided polygon (n is a positive integer).
S1014, splitting the screened outline to obtain fitting data of an outline boundary;
based on the cluster analysis result of the last step, extracting the coordinates of the contour points corresponding to the n significant clusters into n sets, wherein the coordinates correspond to fitting data of n sides of the n-sided polygon.
S1015, performing least square fitting on the fitting data, and determining the position of the commodity outline in the commodity image.
In addition, the contour extraction is performed on the commodity, and the type identification is performed on the extracted commodity, which may specifically be:
s1016, establishing a depth residual error network model;
s1017, inputting the commodity image into the depth residual error network model end, sequentially carrying out feature extraction, characterization enhancement, target detection and attitude measurement on the commodity image through the depth residual error network, outputting and identifying a target area of the commodity, and determining the commodity category.
Specifically, the depth residual error network anchors characteristic points of the commodity image, acquires a plurality of characteristic points through a residual error deep learning algorithm, carries out straight line fitting, acquires characteristic information of the commodity in the image, sequentially carries out characterization enhancement, specific target detection and specific gesture detection, and finally outputs a target area and related parameters of the commodity, thereby carrying out commodity category identification, for example, the commodity category can be a teaching aid or a child game commodity. The method and the device for detecting the object and measuring the attitude of the object belong to the prior art, and the embodiment of the invention is not tired.
In addition, in the field of commodity image recognition, there is often a problem that the image capturing picture causes a false commodity recognition, for example, the commodity recognition often occurs in a shooting scene that is too dark or too exposed outdoors. Therefore, if the image is taken outdoors, in order to improve a better commodity identification success rate, before the commodity image is characterized and enhanced through the depth residual error network, the embodiment of the invention also needs to perform tone mapping adjustment on the commodity image so as to better output a high-quality image and increase the identification success rate. The tone mapping adjustment method specifically comprises the following steps: and sequentially carrying out brightness extraction, global tone mapping and Laplace domain fusion on the commodity image, and recovering the color of the commodity image. The specific process is as follows:
converting the commodity image from RGB space to CIEXYZ space, and separating the brightness component of the image;
the HDR image is first converted from RGB space to CIEXYZ space because the correlation of the three components X, Y, Z in this space is very small so that the chromaticity of the image is not greatly affected when the dynamic range of luminance is compressed. Wherein X, Z represents chromaticity of the image, luminance component is related to Y only, and the conversion formula is as follows:
global tone mapping is carried out on the commodity image by using a Reinhard global tone mapping algorithm, and three compressed brightness images are obtained;
optionally, the prior art is divided into two algorithms, namely local tone mapping and global tone mapping, wherein the local tone mapping can obviously improve the image contrast, but the time consumption is obviously increased, and the local mapping result also faces the problem of excessively dark local brightness.
Specifically, compared to the global tone mapping algorithm in the prior art, the embodiment of the present invention adopts the global tone mapping method to compress the dynamic range of the high exposure part, for example:
L(mask)=(settedMaxL-thL)*(L(mask)-thL)/(maxL-thL)+thL (2)
the mean value is further inversely transformed (indexed) by means of a mean function to equalize the brightness. Such as
meanLw=exp(mean(LL(:))) (3)
In addition, a gamma correction algorithm can be adopted to correct the overexposure area and the overdrising area in the three brightness images; wherein the gamma correction algorithm comprises: a gamma correction algorithm with a gamma greater than 1 and a gamma correction algorithm with a gamma less than 1. The gamma correction algorithm with the gamma larger than 1 restores details of the image overexposed region, and the gamma correction algorithm with the gamma smaller than 1 improves brightness information and contrast information of the dark region of the image.
Finally, a double pyramid frequency domain fusion algorithm can be adopted to fuse the three brightness images and output a final brightness image; specifically, respectively calculating the contrast ratio, the exposure well and the edge information of the three brightness images as three different quality measure factors; multiplying the three quality measure factors to obtain a weight image; carrying out normalization processing on the weight image, and then carrying out Gaussian pyramid decomposition on the weight image; carrying out Laplacian pyramid decomposition on the three input images; multiplying the images after the two pyramid decomposition according to pixel distribution to obtain a new pyramid; and carrying out Laplacian pyramid reconstruction on the new pyramid to obtain a final brightness output image.
And carrying out image color recovery on the final brightness image.
Fig. 2a is an image of a commodity photographed in the embodiment of the present invention, which is a learning machine, but the photographed image is too dark due to the problem of the amount of light incident upon photographing, and fig. 2b is too bright due to the higher amount of light incident upon photographing, and the normal image of fig. 2c can be obtained after the above-mentioned double pyramid processing.
S102, acquiring historical information of a commodity purchased by a user, and predicting the purchase event by using an artificial intelligence method based on the historical information and the commodity category to predict the purchase probability of the commodity category purchased by the user, wherein the historical information comprises the time, region, price and frequency of commodity purchase by the user;
specifically, S102 may employ multidimensional hopks (hawkes) and neural networks to make predictions of the process, specifically including:
acquiring historical information of the commodity purchased by the user by utilizing a multidimensional Hawkes process;
the historical information is associated with the commodity category based on a graph convolution neural network, and an association matrix is obtained;
inputting the incidence matrix into a pre-constructed cyclic neural network model, and predicting the purchase event through the cyclic neural network model.
The above process is specifically described as follows:
s1021, forming a correlation sequence by the user and the historical purchased goods according to the dimension of a time sequence (time series);
for example, the commodity purchased by the user 1 in the period T1 is a, B, and the commodity purchased in the period T2 is C, d i,j =[Item i,j ,t i,j ]Item represents a commodity identification, t represents a period, i, j is a positive integer, and 0<i,j<N。
S1022, establishing a correlation model, inputting the correlation sequence into the correlation model, and outputting feature vectors of the user and the historical purchased goods;
in one embodiment, the correlation model formula is as follows:
where F (Item, t) is the correlation model function, and H (Item) and G (t) represent the commodity intensity function and the time intensity function, respectively.
The feature vector includes, but is not limited to, a correlation coefficient/fitness between the user and the commodity, a purchase frequency, and the like, wherein the correlation coefficient/fitness can represent the correlation relationship between the user and the commodity, and the higher the numerical value is, the higher the correlation degree between the user and the commodity is proved, and the higher the fitness is, the higher the success rate of recommending the commodity is.
S1023, inputting the feature vector into the graph convolution neural network to acquire graph adjacent information;
the graph adjacency information comprises an adjacency matrix of the user node and a user node feature matrix acquired by SVD.
In addition, the construction phase of the graph convolution neural network comprises:
a graph embedding stage, wherein a double-layer parallel graph convolutional neural network is used for capturing graph space characteristics;
in the fourier domain, constructing a filter for capturing spatial features of nodes in the graph, traversing the filter through each node in the graph;
and after the spatial characteristics of each node are acquired, constructing a graph convolution neural network model by stacking a plurality of group convolution layers.
S1024, inputting the graph adjacent information into a graph convolution nerve GCN network, and outputting a space-time correlation matrix of historical purchased goods and purchase time by using the GCN network;
the graph convolutional neural network (Graph Convolutional Network, GCN) is a method that enables deep learning of graph data. The algorithm principle is as follows:
the first step: each node transmits (send) the characteristic information of itself to the neighbor node after transformation. This step is to extract and transform the characteristic information of the node.
And a second step of: each node receives (receives) aggregates the characteristic information of the neighboring nodes. This step is to fuse the local structure information of the nodes.
And a third step of: the transformation (transform) gathers the previous information and then performs nonlinear transformation to increase the expressive power of the model.
The graph convolution neural network has the following properties of a convolution neural network:
1. local parameter sharing, operator is applied to each node (circle represents operator), and is shared everywhere.
2. The receptive field is proportional to the number of layers, each node contains the information of the direct neighbors at the beginning, and the information of the neighbors can be contained when the second layer is calculated, so that the information of the neighbors participating in the operation is more and more sufficient. The more the number of layers, the wider the receptive field and the more information is involved in the operation.
FIG. 3 is a network architecture diagram of a GCN, wherein in one embodiment, graph adjacency information is input into the GCN network, a spatiotemporal correlation matrix of historical purchase items and purchase times is output, and the steps of:
the image adjacency information is input into an encoding layer Encoder, after encoding of the Encoder is completed, the encoded image adjacency information is input into a decoding layer Decoder for decoding, wherein the Encoder and the Decoder both comprise a multi-diffusion convolution recursion layer (Diffusion Convolutional Recurrent Layer) and an activation function ReLU, and the image adjacency information comprises a historical purchase commodity adjacency matrix A and a purchase time adjacency matrix B acquired by SVD.
In a spectrum-based graph neural network, the graph is assumed to be an undirected graph, one robust mathematical representation of which is a regularized graph laplace matrix, i.e
Wherein A is the adjacency matrix of the graph, D is the diagonal matrix and
D ii =Σ j (A i,j ) (6)
regularized graph the laplace matrix has the property of real symmetric semi-positive definite. Regularized Laplace matrix can be decomposed into
Wherein
U is a matrix of eigenvectors of L, Λ is a diagonal matrix, and the value on the diagonal is the eigenvalue of L. The eigenvectors of the regularized laplacian matrix form a set of orthogonal basis.
In the course of processing the image signals, the signal of one image
x∈R N (9)
Is a feature vector, x, formed by each node of the graph i Representing the ith node.
The fourier transform of the graph X is thus defined as
The Fourier inverse transformation is
Wherein isFourier transformed results.
Fourier transformation projects the input image signal into an orthogonal space whose basis is composed of regularized image laplacian eigenvectors.
The elements of the converted signal are the coordinates of the map signal in the new space, so the original input signal can be expressed as
This is the result of the inverse fourier transform.
Thus defining a graph convolution operation on the input signal X:
wherein g defines a filter; as indicated by the letter Hadamard product.
In addition, the space-time correlation matrix of the historical purchase commodity and the purchase time comprises a space dimension besides a time dimension, and specifically comprises correlation factors such as the season, the region and the like of the purchase.
S1025, inputting the time-space correlation matrix of the historical purchased goods and the purchase time into a cyclic neural network model so as to output the purchase probability of the goods category.
The recurrent neural network (Recurrent Neural Network, RNN) is a type of recurrent neural network (recursive neural network) that takes sequence data as input, performs recursion (recovery) in the evolution direction of the sequence, and all nodes (circulation units) are chained. RNNs are therefore able to process sequence data efficiently, thanks to their special structure. The traditional NN model has a simple structure: input layer- > hidden layer- > output layer; the greatest difference between RNNs and traditional NN models is that: the input of the hidden layer includes not only the output of the input layer but also the output of the hidden layer at the previous time.
The predictive mechanism of the recurrent neural network model is prior art and is skipped here.
And S103, recommending the commodity category to the user if the purchase probability is larger than a preset threshold value, so that the user refers to and purchases the commodity category.
The preset threshold is set manually, for example 80%.
According to the method provided by the embodiment of the invention, based on the historical information and the commodity category, the purchasing event is predicted by utilizing an artificial intelligence method, specifically, the purchasing event can be predicted by adopting a multidimensional Hoxwell process and a neural network, so that the commodity category is better determined to be recommended or not recommended, personalized commodity recommendation after combining a commodity image recognition function is realized, and the commodity recommendation success rate is increased.
The embodiment of the invention also comprises a device which is characterized by comprising a memory and a processor, wherein the memory stores computer executable instructions, and the processor realizes the method when running the computer executable instructions on the memory.
The embodiment of the present invention also provides a computer-readable storage medium having stored thereon computer-executable instructions for performing the method of the above-described embodiment.
The embodiment of the invention also provides a device which comprises a memory and a processor, wherein the memory stores computer executable instructions, and the processor realizes the method when running the computer executable instructions on the memory.
FIG. 4 is a schematic diagram of the hardware components of the device in one embodiment. It will be appreciated that figure 4 only shows a simplified design of the device. In practical applications, the apparatus may further include other necessary elements, including but not limited to any number of input/output systems, processors, controllers, memories, etc., and all apparatuses capable of implementing the big data management method of the embodiments of the present application are within the scope of protection of the present application.
The memory includes, but is not limited to, random access memory (random access memory, RAM), read-only memory (ROM), erasable programmable read-only memory (erasable programmable read only memory, EPROM), or portable read-only memory (compact disc read to only memory, CD to ROM) for the associated instructions and data.
The input system is used for inputting data and/or signals, and the output system is used for outputting data and/or signals. The output system and the input system may be separate devices or may be a single device.
A processor may include one or more processors, including for example one or more central processing units (central processing unit, CPU), which in the case of a CPU may be a single core CPU or a multi-core CPU. The processor may also include one or more special purpose processors, which may include GPUs, FPGAs, etc., for acceleration processing.
The memory is used to store program codes and data for the network device.
The processor is used to call the program code and data in the memory to perform the steps of the method embodiments described above. Reference may be made specifically to the description of the method embodiments, and no further description is given here.
In the several embodiments provided in this application, it should be understood that the disclosed systems and methods may be implemented in other ways. For example, the division of the unit is merely a logic function division, and there may be another division manner when actually implemented, for example, a plurality of units or components may be combined or may be integrated into another system, or some features may be omitted or not performed. The coupling or direct coupling or communication connection shown or discussed with each other may be through some interface, system or unit indirect coupling or communication connection, which may be in electrical, mechanical or other form.
The units described as separate units may or may not be physically separate, and units shown as units may or may not be physical units, may be located in one place, or may be distributed over a plurality of network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.
In the above embodiments, it may be implemented in whole or in part by software, hardware, firmware, or any combination thereof. When implemented in software, may be implemented in whole or in part in the form of a computer program product. The computer program product includes one or more computer instructions. When loaded and executed on a computer, produces a flow or function in accordance with embodiments of the present application, in whole or in part. The computer may be a general purpose computer, a special purpose computer, a network of computers, or other programmable system. The computer instructions may be stored in or transmitted across a computer-readable storage medium. The computer instructions may be transmitted from one website, computer, server, or data center to another website, computer, server, or data center by a wired (e.g., coaxial cable, fiber optic, digital subscriber line (digital subscriber line, DSL)) or wireless (e.g., infrared, wireless, microwave, etc.). The computer readable storage medium may be any available medium that can be accessed by a computer or a data storage device such as a server, data center, etc. that contains an integration of one or more available media. The usable medium may be a read-only memory (ROM), or a random-access memory (random access memory, RAM), or a magnetic medium such as a floppy disk, a hard disk, a magnetic tape, a magnetic disk, or an optical medium such as a digital versatile disk (digital versatile disc, DVD), or a semiconductor medium such as a Solid State Disk (SSD), or the like.
The foregoing is merely a specific embodiment of the present application, but the protection scope of the present application is not limited thereto, and any equivalent modifications or substitutions will be apparent to those skilled in the art within the scope of the present application, and these modifications or substitutions should be covered in the scope of the present application. Therefore, the protection scope of the present application shall be subject to the protection scope of the claims.