CN110210513B - Data classification method and device and terminal equipment - Google Patents

Data classification method and device and terminal equipment Download PDF

Info

Publication number
CN110210513B
CN110210513B CN201910328170.5A CN201910328170A CN110210513B CN 110210513 B CN110210513 B CN 110210513B CN 201910328170 A CN201910328170 A CN 201910328170A CN 110210513 B CN110210513 B CN 110210513B
Authority
CN
China
Prior art keywords
data
result
network
detected
prediction
Prior art date
Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
Active
Application number
CN201910328170.5A
Other languages
Chinese (zh)
Other versions
CN110210513A (en
Inventor
孟凡阳
柳伟
梁永生
杨火祥
黄玉成
王昌伟
Current Assignee (The listed assignees may be inaccurate. Google has not performed a legal analysis and makes no representation or warranty as to the accuracy of the list.)
Shenzhen Institute of Information Technology
Original Assignee
Shenzhen Institute of Information Technology
Priority date (The priority date is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the date listed.)
Filing date
Publication date
Application filed by Shenzhen Institute of Information Technology filed Critical Shenzhen Institute of Information Technology
Priority to CN201910328170.5A priority Critical patent/CN110210513B/en
Publication of CN110210513A publication Critical patent/CN110210513A/en
Application granted granted Critical
Publication of CN110210513B publication Critical patent/CN110210513B/en
Active legal-status Critical Current
Anticipated expiration legal-status Critical

Links

Images

Classifications

    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/24—Classification techniques
    • G06F18/241—Classification techniques relating to the classification model, e.g. parametric or non-parametric approaches
    • G—PHYSICS
    • G06—COMPUTING; CALCULATING OR COUNTING
    • G06F—ELECTRIC DIGITAL DATA PROCESSING
    • G06F18/00—Pattern recognition
    • G06F18/20—Analysing
    • G06F18/25—Fusion techniques

Landscapes

  • Engineering & Computer Science (AREA)
  • Data Mining & Analysis (AREA)
  • Theoretical Computer Science (AREA)
  • Computer Vision & Pattern Recognition (AREA)
  • Bioinformatics & Cheminformatics (AREA)
  • Bioinformatics & Computational Biology (AREA)
  • Artificial Intelligence (AREA)
  • Evolutionary Biology (AREA)
  • Evolutionary Computation (AREA)
  • Physics & Mathematics (AREA)
  • General Engineering & Computer Science (AREA)
  • General Physics & Mathematics (AREA)
  • Life Sciences & Earth Sciences (AREA)
  • Image Analysis (AREA)

Abstract

The invention is suitable for the technical field of deep learning, and provides a data classification method, a data classification device and terminal equipment, wherein the data classification method comprises the following steps: inputting first data to be measured into a data augmentation-based neural network, wherein the data augmentation-based neural network comprises a transformation network, a prediction network and a decision layer, and the transformation network and the prediction network are both neural networks with deep learning capability; according to the first data to be detected, second data to be detected is obtained through the transformation network; according to the first to-be-detected data and the second to-be-detected data, respectively obtaining a first prediction result corresponding to the first to-be-detected data and a second prediction result corresponding to the second to-be-detected data through the prediction network; and obtaining a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result. The embodiment of the invention can improve the accuracy of data classification.

Description

Data classification method and device and terminal equipment
Technical Field
The invention belongs to the technical field of deep learning, and particularly relates to a data classification method and device and terminal equipment.
Background
With the development of deep learning technology, deep neural networks are widely used in various data processing such as image recognition, image segmentation, speech recognition, and gesture recognition. The key of the deep neural network in processing the data is detection and classification of the object to be detected, so that the data can be generally classified by the deep neural network regardless of recognition of data such as images, voice and the like and image segmentation.
The deep neural network is used for classifying data, a large amount of sample data is required to be collected and labeled in advance, however, the acquisition of the large amount of sample data is time-consuming and labor-consuming, and the sample data is difficult to acquire, such as medical data and abnormal detection data, so that the problem of insufficient data classification accuracy caused by insufficient sample data exists when the deep neural network is used for classifying the data.
Disclosure of Invention
In view of this, embodiments of the present invention provide a data classification method, an apparatus, and a terminal device, so as to solve the problem of how to improve accuracy of data classification in the prior art.
A first aspect of an embodiment of the present invention provides a data classification method, including:
inputting first data to be measured into a data augmentation-based neural network, wherein the data augmentation-based neural network comprises a transformation network, a prediction network and a decision layer, and the transformation network and the prediction network are both neural networks with deep learning capability;
according to the first data to be detected, second data to be detected is obtained through the transformation network;
according to the first to-be-detected data and the second to-be-detected data, respectively obtaining a first prediction result corresponding to the first to-be-detected data and a second prediction result corresponding to the second to-be-detected data through the prediction network;
and obtaining a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result.
A second aspect of an embodiment of the present invention provides a data classification apparatus, including:
the device comprises an input unit, a data amplification-based neural network and a decision layer, wherein the input unit is used for inputting first data to be tested into the data amplification-based neural network, the data amplification-based neural network comprises a transformation network, a prediction network and the decision layer, and the transformation network and the prediction network are both neural networks with deep learning capability;
the transformation unit is used for obtaining second data to be tested through the transformation network according to the first data to be tested;
the prediction unit is used for respectively obtaining a first prediction result corresponding to the first data to be tested and a second prediction result corresponding to the second data to be tested through the prediction network according to the first data to be tested and the second data to be tested;
and the classification result determining unit is used for obtaining a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result.
A third aspect of the embodiments of the present invention provides a terminal device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, where the processor implements the steps of the data classification method when executing the computer program.
A fourth aspect of embodiments of the present invention provides a computer-readable storage medium having stored thereon a computer program which, when executed by a processor, performs the steps of the data classification method as described.
Compared with the prior art, the embodiment of the invention has the following beneficial effects: in the embodiment of the invention, in the process of classifying the first data to be detected, the second data to be detected is generated through the transformation network, and then the prediction results corresponding to the first data to be detected and the second data to be detected are fused to obtain the final classification result.
Drawings
In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the drawings needed for the embodiments or the prior art descriptions will be briefly described below, and it is obvious that the drawings in the following description are only some embodiments of the present invention, and it is obvious for those skilled in the art to obtain other drawings without creative efforts.
Fig. 1 is a schematic flow chart of an implementation of a first data classification method provided in an embodiment of the present invention;
fig. 2 is a schematic structural diagram of a data-augmentation-based neural network according to an embodiment of the present invention;
fig. 3 is a schematic structural diagram of a method for transforming a network according to an embodiment of the present invention;
fig. 4 is a schematic flow chart of an implementation of a second data classification method provided in the embodiment of the present invention;
FIG. 5 is a schematic diagram of data transmission during training of a neural network based on data augmentation according to an embodiment of the present invention;
FIG. 6 is a flowchart illustrating an implementation of a third data classification method according to an embodiment of the present invention;
FIG. 7 is a schematic diagram of a data sorting apparatus according to an embodiment of the present invention;
fig. 8 is a schematic diagram of a terminal device according to an embodiment of the present invention.
Detailed Description
In the following description, for purposes of explanation and not limitation, specific details are set forth, such as particular system structures, techniques, etc. in order to provide a thorough understanding of the embodiments of the invention. It will be apparent, however, to one skilled in the art that the present invention may be practiced in other embodiments that depart from these specific details. In other instances, detailed descriptions of well-known systems, devices, circuits, and methods are omitted so as not to obscure the description of the present invention with unnecessary detail.
In order to explain the technical means of the present invention, the following description will be given by way of specific examples.
It will be understood that the terms "comprises" and/or "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 herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application. As used in the specification of the present application 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 and includes any and all possible combinations of one or more of the associated listed items.
As used in this specification and the appended claims, the term "if" may be interpreted contextually as "when", "upon" or "in response to a determination" or "in response to a detection". Similarly, the phrase "if it is determined" or "if a [ described condition or event ] is detected" may be interpreted contextually to mean "upon determining" or "in response to determining" or "upon detecting [ described condition or event ]" or "in response to detecting [ described condition or event ]".
In addition, in the description of the present application, the terms "first," "second," "third," and the like are used solely to distinguish one from another and are not to be construed as indicating or implying relative importance.
The first embodiment is as follows:
fig. 1 shows a schematic flow chart of a first data classification method provided in an embodiment of the present application, which is detailed as follows:
in S101, a first to-be-measured data is input into a data-augmentation-based neural network, where the data-augmentation-based neural network includes a transformation network, a prediction network, and a decision layer, and both the transformation network and the prediction network are neural networks with deep learning capabilities.
The first data x to be measured refers to data to be classified, and the first data to be measured may be image data, voice data, text data, or the like to be classified. A data-based augmentation neural network, specifically an end-to-end deep neural network with sample data augmentation function, is composed of a transformation network, a prediction network, and a decision layer, as shown in fig. 2. The transformation network and the prediction network are both neural networks with deep learning capability, namely the transformation network can automatically transform the first to-be-detected data after sample training and automatic learning of adjustment parameters; after the prediction network automatically learns and adjusts parameters through sample training, the result prediction can be automatically carried out on input data. The data-augmentation-based neural network in the embodiment of the invention is specifically a neural network trained in advance. Before the data based on the data augmentation neural network is used for data classification, a preset amount of sample data is input to train the data based on the data augmentation neural network, and during training, because the data augmentation based neural network has a data augmentation function, more generated sample data are generated on the basis of original sample data to perform model training, so that the finally trained data based on the data augmentation neural network has stronger generalization capability and judgment capability.
And inputting the first data to be detected into a pre-trained neural network based on data augmentation, and starting an identification and classification processing flow. Optionally, the first data to be measured may be preprocessed, for example, de-noising, data cleaning, and the like, before being input into the data-augmented neural network.
In S102, according to the first data to be measured, second data to be measured is obtained through the transformation network.
Inputting the first data x to be tested into a transformation network, learning the characteristics of the first data x to be tested through the transformation network, and generating and outputting data similar to the first data x to be tested in turn, namely the transformation network can learn and reconstruct the first data x to be tested to obtain the second data x to be tested
Figure BDA0002036857580000051
Optionally, the transform network is an encoding-decoding network structure as shown in fig. 3, the first data x to be measured is input into the encoder for encoding and input into the hidden layer, feature extraction, learning and reconstruction are performed through the hidden layer, and then the generated data corresponding to the first data x to be measured, that is, the second data x to be measured, is obtained through decoding and output by the decoder
Figure BDA0002036857580000052
Specifically, the transformation network may be an Auto-encoder (AE) network structure or a Variational Auto-encoder (VAE) network structure.
Optionally, if the first data to be measured is one-dimensional data, the transformation network is a long-short term memory (LSTM) network;
and if the first data to be detected is two-dimensional data, the transformation network is a convolutional neural network.
If the first data to be measured is data in a one-dimensional data, i.e. a sequence class, such as voice sequence data and text sequence data, the structure of the transform network is specifically a Long Short-Term Memory (Long Short-Term Memory) network structure. If the first data to be measured is two-dimensional data, for example image data, the structure of the transformation network is in particular a convolutional neural network structure suitable for processing image type data. Optionally, the transformation network includes both an LSTM network and a convolutional neural network, and if it is detected that the first data to be measured is one-dimensional data, the first data to be measured is input into the LSTM network in the transformation network to be transformed to obtain second data to be measured; and if the first data to be detected is detected to be two-dimensional data, inputting the first data to be detected into a convolutional neural network in a transformation network for transformation processing.
In S103, according to the first to-be-measured data and the second to-be-measured data, a first prediction result corresponding to the first to-be-measured data and a second prediction result corresponding to the second to-be-measured data are obtained through the prediction network, respectively.
The first data x to be tested and the second data to be tested generated according to the first data to be tested
Figure BDA0002036857580000061
Respectively inputting the prediction networks in the neural network based on data augmentation to respectively obtain first prediction results corresponding to the first data x to be measured
Figure BDA0002036857580000062
And second data to be tested
Figure BDA0002036857580000063
Corresponding second prediction result
Figure BDA0002036857580000064
In S104, a final classification result of the first to-be-detected data is obtained through the decision layer according to the first prediction result and the second prediction result.
The first prediction result is obtained
Figure BDA0002036857580000065
And a second predicted result
Figure BDA0002036857580000066
And inputting the data into a decision layer together, and performing weighted fusion calculation to obtain a final classification result y of the first to-be-detected data x. For example, taking image recognition as an example, a recognition result (i.e., a final classification result of the first to-be-detected data) of the first to-be-detected image is obtained through the decision layer according to a first recognition result (i.e., a first prediction result) corresponding to the first to-be-detected image (i.e., the first to-be-detected data) and a second recognition result (i.e., a second prediction result) corresponding to a generated image (i.e., the second to-be-detected data) transformed and reconstructed according to the first to-be-detected image. The identification result may be that the first image to be detected is identified to contain or not contain the target detection object, and the final classification result is one of a yes classification result and a no classification result; or the recognition result can be one of four classification results that the first image to be detected contains the first target detection object, the second target detection object, the third target detection object or does not contain any target detection object; by analogy, the total number of the classified categories can be set according to actual needs, and the final classification result of the first to-be-detected data is one of the total categories.
Optionally, obtaining a final classification result of the first to-be-measured data through the decision layer according to the first prediction result and the second prediction result, including:
obtaining a decision layer parameter lambda according to a first prediction result
Figure BDA0002036857580000067
And a second predicted result
Figure BDA0002036857580000068
Obtaining a final classification result y of the first to-be-detected data through a decision calculation formula, wherein the decision calculation formula specifically comprises:
Figure BDA0002036857580000071
wherein, lambda belongs to [0,1 ].
Obtaining a parameter lambda of a decision layer as a parameter for carrying out weighted fusion calculation on the decision layer, and according to an input first prediction result
Figure BDA0002036857580000072
And a second predicted result
Figure BDA0002036857580000073
By means of decision-making calculation formulas
Figure BDA0002036857580000074
Calculating to obtain a final classification result y, wherein the lambda belongs to [0,1]]. In general, λ is set to any real number between 0 and 1 (excluding 0 and 1), i.e. λ ∈ (0,1), so that the final classification result y combines with the first prediction result corresponding to the original data (i.e. the first data to be detected)
Figure BDA0002036857580000075
And a second prediction result corresponding to the generated data (i.e. the second data to be measured)
Figure BDA0002036857580000076
Therefore, more data characteristics can be integrated to enable the classification result to be more accurate.
In a special case where the data classification processing time or the computing power of the system is not sufficient, λ may be made 0, and when λ is detected to be 0, step S102 and step S103 may be skipped, and step S104 may be directly performed, that is, the first prediction result is obtained by predicting only the original first data to be measured
Figure BDA0002036857580000077
And taking the first prediction result as a final classification result y. When lambda is 0, the prediction result of generated data does not need to be combined, namely, a part of a transformation network is omitted, and a part of generated data input into a prediction network for calculation is also omitted, so that the calculation complexity is reduced, the calculation time and the system resources are saved, and the mode is a flexible choice for reducing the calculation complexity by sacrificing certain accuracy under special conditions. Alternatively, λ may be 1, and a second prediction result corresponding to the second data to be measured, which is the data, may be generated at this time
Figure BDA0002036857580000078
The method can be used for the condition that the performance of the transformation network is better as the final classification result of the first data to be detected, the data characteristics of the second data to be detected generated by the transformation network can include all the data characteristics of the first data to be detected and more data characteristics hidden by the first data to be detected are added, and only the second prediction result is used for the time
Figure BDA0002036857580000079
A final classification result with higher accuracy can be obtained.
Preferably, λ is 0.5, in which case the first prediction can be maximally combined at the same time
Figure BDA00020368575800000710
And a second predicted result
Figure BDA00020368575800000711
The final classification result y of the first data to be detected is obtained, that is, the data features of the original data (the first data to be detected) and the generated data (the second data to be detected) are maximally combined to obtain the final classification result, so that the accuracy of data classification can be improved by combining more data features.
Optionally, before the step S101, the method further includes: and receiving a decision-making layer parameter setting instruction, and setting the value of a decision-making layer parameter lambda according to the setting instruction. Optionally, when the set λ value does not meet the condition λ ∈ [0,1], a prompt message of setting error is issued to instruct the user to re-input the correct λ value.
In the embodiment of the invention, in the process of classifying the first data to be detected, the second data to be detected is generated through the transformation network, and then the prediction results corresponding to the first data to be detected and the second data to be detected are fused to obtain the final classification result.
Example two:
fig. 4 shows a schematic flow chart of a second data classification method provided in an embodiment of the present application, which is detailed as follows:
in S401, original sample data is obtained, where the original sample includes first original sample data carrying a first result tag and second original sample data carrying a second result tag.
The method comprises the steps of obtaining original sample data, for example, when first data to be classified are image data, acquiring a preset number of target images as the original sample data in an image acquisition mode, and enabling the original sample data to carry corresponding result labels by receiving result label marking instructions. Or, acquiring a preset number of original sample data carrying result tags in a manner of downloading and reading an existing target image database. Wherein, the first result label y is carriediFirst original sample x ofiAnd carrying a second result label yjSecond original sample xjAny two samples of data in the original sample.
In S402, according to the first original sample data, first generation sample data carrying a first result tag is obtained through a transform network.
According to carryingFirst result tag yiFirst original sample data x ofiAnd carrying out sample data augmentation through a transformation network to obtain generated sample data which is the same as the result label of the first original sample data but has different sample data characteristics, so as to obtain first generated sample data carrying the first result label. When the neural network based on data augmentation is used for network model training, the transformation network plays a role in generating more sample data, the diversity of the sample data is ensured, the generalization capability of the neural network based on data augmentation is improved, and therefore the accuracy of data classification by using the neural network based on data augmentation is improved.
In S403, sample fusion is performed according to the first generated sample data carrying the first result tag and the second original sample data carrying the second result tag, so as to obtain fusion sample data carrying a fusion result tag.
Will carry a first result tag yiFirst generation sample data of
Figure BDA0002036857580000091
And carries a second result tag yjSecond original sample data xjCarrying out weighting operation to realize sample fusion and obtain a label carrying a fusion result
Figure BDA0002036857580000092
Of (2) fusion sample data
Figure BDA0002036857580000093
Optionally, the step S403 specifically includes:
according to the label y carrying the first resultiFirst generation sample data of
Figure BDA0002036857580000094
Carrying a second result label yjSecond original sample data xjAnd fusion coefficient beta, and obtaining a label carrying a fusion result through a fusion sample calculation formula
Figure BDA0002036857580000095
Of (2) fusion sample data
Figure BDA0002036857580000096
The fused sample calculation formula specifically comprises:
Figure BDA0002036857580000097
Figure BDA0002036857580000098
wherein, beta belongs to [0,1 ].
In the training process, the fusion coefficient beta is any real number which is greater than or equal to 0 and less than or equal to 1, and fusion sample data calculated by a fusion sample calculation formula according to the fusion coefficient
Figure BDA0002036857580000099
Randomly fusing the first generation sample data
Figure BDA00020368575800000910
And data characteristics of the second original sample data, and a fusion result label corresponding to the fusion sample data
Figure BDA00020368575800000911
Also randomly fusing the first result label yiAnd a second result label yjThe two results can generate diversified sample data to train the prediction network, so that the accuracy of data classification by adopting the data-based augmented neural network is improved.
Optionally, the fusion coefficient β in the fused sample calculation formula satisfies the following constraint condition:
β∈Beta(a,a)
Figure BDA00020368575800000912
beta (a, a) represents Beta distribution with two parameters being a, N and N respectively represent current iteration times and total target iteration times in the training process, and m is a real number which is greater than 0 and less than or equal to 0.1.
Beta (a, a) refers to Beta Distribution (Beta Distribution) in which the first parameter and the second parameter are both a, and the fusion parameter obeys the Beta (a, a) Distribution. Wherein the parameter a is taken
Figure BDA0002036857580000101
m, the larger of the two values. In that
Figure BDA0002036857580000102
In the training process, n represents the current iteration times in the training process, namely represents that the current training is the first round of training; n represents the total number of target iterations, i.e. the total number of rounds of training, which can be set before training, and the value of N is usually large, and the magnitude is 10^3 or more, and after setting, the model will perform N rounds of training together. m is a real number close to 0, and specifically, the value of m is constrained to satisfy the condition 0<m ≦ 0.1, and the value of m may be set before training, e.g., set m to 0.01. Specifically, the value of m may be set according to the order of N in the total training round, for example, if N is 10^3, m may be set to a real number less than or equal to 1/(10^3), for example, m is set to 1/(10^3) or m is set to 1/(10^3 × 10), and the like.
The N, m value is set so that at the beginning of training, when the n value is relatively small,
Figure BDA0002036857580000103
close to 1, i.e.
Figure BDA0002036857580000104
When Beta (a, a) ═ Beta (1,1) ═ U (0,1), where U (0,1) refers to a Uniform Distribution between 0 and 1 (Uniform Distribution), that is, when Beta Distribution is equivalent to a Uniform Distribution between 0 and 1, that is, the fusion coefficient β obeys U (0, 1); meltWhen the fusion coefficient β is subject to U (0,1), the probability that β is 0 or β is 1 is small, that is, in this case, β is usually a number between 0 and 1, and the tag carrying the fusion result is calculated according to the fusion calculation formula
Figure BDA0002036857580000105
Of (2) fusion sample data
Figure BDA0002036857580000106
And combining the data characteristics of the original sample and the learned production sample, namely training the model parameters of the transformation network in the initial training stage. In the later period of training, the value of n is larger,
Figure BDA0002036857580000107
is less than the value of m, i.e. when a is equal to m, e.g. when m is equal to 0.001, Beta (a, a) is equal to Beta (0.001 ); in the Beta distribution, when the two parameters are both 0, namely Beta (0,0) is equivalent to the bernoulli distribution, and because Beta (0,0) may have unstable conditions, a parameter m close to 0 is taken as the parameter of the Beta distribution at the moment, so that the fusion coefficient Beta at the moment approximately follows the bernoulli distribution; when the fusion coefficient beta obeys Bernoulli distribution, the value of the beta is 0 or 1, namely the label carrying the fusion result is obtained by calculation according to a fusion calculation formula
Figure BDA0002036857580000108
Of (2) fusion sample data
Figure BDA0002036857580000109
Basically, the prediction method includes the steps of obtaining sample data obtained by training a converged transformation network at a later training stage by using an original sample (specifically, a first original sample) or a generated sample (specifically, a first generated sample), and sending the sample data into a prediction network for training.
By constraining the fusion coefficient beta, the transformation network can be trained in the early stage of training, the learning parameters of the transformation network are adjusted, and meanwhile, various fusion sample data are obtained for training of the prediction network; and generating pure original sample data and pure generated sample data for the training of the prediction network at the later training stage, so that the prediction network can accurately process the fusion sample data and can accurately process the original sample data and generate the sample data.
In S404, deep learning training is performed on the data-based augmented neural network according to the original sample data and the fusion sample data.
And (4) inputting the original sample data and the fusion sample data obtained in the step (S403) into a prediction network in the data-augmented neural network for model training, and automatically learning and adjusting parameters of the prediction network according to the sample data. Alternatively, when the fusion calculation formula in S403 is used to calculate the fusion sample data, and the fusion parameter β satisfies the constraint condition: beta e Beta (a, a) and
Figure BDA0002036857580000111
in the meantime, the fusion sample data of the data-based augmented neural network fusion sample data at the later stage of training already contains the original sample data, so step S404 is specifically: and performing deep learning training on the data augmentation-based neural network according to the fusion sample data.
And repeatedly executing the steps S402 to S404, and performing multi-round combined training on the transformation network and the prediction network in the data-augmentation-based neural network, so as to realize end-to-end training and obtain the trained data-augmentation-based neural network for later data classification application. The processing flow of each round of training data is shown in fig. 5.
In S405, the first data to be measured is input to the data-augmented neural network.
In S406, according to the first data to be measured, second data to be measured is obtained through the transformation network.
In S407, according to the first to-be-measured data and the second to-be-measured data, a first prediction result corresponding to the first to-be-measured data and a second prediction result corresponding to the second to-be-measured data are obtained through the prediction network, respectively.
In S408, according to the first prediction result and the second prediction result, a final classification result of the first data to be measured is obtained through the decision layer.
Step S405 to step S408 are processes of classifying data by using a trained neural network based on data augmentation, and step S405 to step S408 in this embodiment are respectively the same as step S101 to step S104 in the first embodiment, and refer to the description of step S101 to step S104 in the first embodiment, which is not repeated herein.
In the embodiment of the invention, as more sample data are generated through the transformation network with deep learning capability according to the original sample data, the performance of the neural network based on data augmentation obtained by final training can be better, and the accuracy of data classification can be improved when the neural network based on data augmentation is used for data classification application; meanwhile, the transformation network in the embodiment of the invention is a neural network with deep learning capability, so that compared with the existing data augmentation method without learning parameters, the neural network based on data augmentation in the embodiment of the invention can jointly train the transformation network and the prediction network during training, so that end-to-end joint training of the neural network is realized, and training can be automatically completed only by inputting an original sample into the neural network based on data augmentation, so that the model training is more efficient.
Example three:
fig. 6 is a schematic flow chart of a third data classification method according to an embodiment of the present invention, where the data classification method in the embodiment of the present invention is specifically a human skeleton behavior identification method, the first to-be-detected data is specifically first to-be-detected human image data, the second to-be-detected data is specifically second to-be-detected human image data, and the classification result is specifically a human skeleton behavior identification result, which is detailed as follows:
in S601, inputting first human body image data to be detected into a data-augmentation-based neural network, where the data-augmentation-based neural network includes a transformation network, a prediction network, and a decision layer, and both the transformation network and the prediction network are neural networks with deep learning capability.
And acquiring first human body image data to be detected, wherein the first human body image data to be detected can be obtained by tracking shooting of a target human body by a camera or by downloading an existing human body behavior recognition image data set. Preferably, the first human body image data to be detected is acquired through a depth camera, so that the target human body skeleton information can be better displayed in the acquired first human body image data to be detected. Optionally, the first human body image data to be detected is subjected to denoising processing and thresholding processing in advance, so that the processed human body image data to be detected can more clearly display the target human body skeleton information. The neural network based on data augmentation in the embodiment of the present invention is consistent with the network structure of the neural network based on data augmentation described in S101 in the first embodiment, and please refer to the related description of the neural network based on data augmentation in S101 in the first embodiment. Inputting the first human body image data to be detected into a data-augmentation-based neural network, and starting to execute a human body skeleton behavior identification process.
In S602, according to the first human body image data to be detected, second human body image data to be detected is obtained through the transformation network.
And obtaining second human body image data to be detected through a transformation network, wherein the second human body image data to be detected is subjected to feature learning and reconstruction according to the first human body image data to be detected. Because the data to be detected in the embodiment of the invention is image data, the transformation network is specifically a convolutional neural network with deep learning capability. In the embodiment of the present invention, a specific process of obtaining the second human body image data by the transform network is similar to the description of step S102 in the first embodiment, and refer to the related description of step S102 in the first embodiment.
In S603, according to the first human body image data to be detected and the second human body image data to be detected, a first prediction result corresponding to the first human body image data to be detected and a second prediction result corresponding to the second human body image data to be detected are obtained through the prediction network, respectively.
Inputting the first human body image data to be detected into a prediction network to obtain a first prediction result, and inputting the second human body image data to be detected into a second prediction result of the prediction network. The prediction result may be a result of predicting whether the first human body image data or the second human body image data to be detected includes the target behavior.
In S604, a final human skeleton behavior recognition result of the first human image data to be detected is obtained through the decision layer according to the first prediction result and the second prediction result.
And carrying out decision making through a decision-making layer according to the first prediction result and the second prediction result to obtain a final human skeleton behavior recognition result, wherein the decision-making layer comprises a decision-making layer parameter lambda. Optionally, the human skeleton behavior recognition result is a result indicating whether the first to-be-detected human image data includes the target behavior, and the human skeleton behavior recognition result may be prompted in a text prompt mode, a voice prompt mode or an image display mode.
In order to verify the recognition accuracy of the human skeleton behavior recognition method based on the data augmentation neural network in the embodiment of the invention, three existing and disclosed human behavior recognition data sets are respectively adopted: the accuracy test of the Human skeletal behavior Recognition method was performed using a Nanyang technology University RGB-D motion Recognition Dataset (NTU RGB-D), a North-West University Los Angeles Multiview Action3D Dataset (North western University of California, Los Angeles Multiview Action3D Dataset, NUCLA), a Texas multimodality Human Action Dataset (University of Texas at Dallas modified Man Action Dataset, UTD-MHAD), where the NTU-D Dataset includes a Cross-View Dataset NTU-CV (NTU Cross-View) and a tested Cross-CV (NTU Cross-Subject). The accuracy test results are shown in table 1, wherein the identification methods a to F are control group methods, including: identification method A without data augmentation, identification method B with L2 regularization, identification method C with Dropout layer, identification method D with Zoneout layer, identification method E with data augmentation through rotation of original image, and baseThe identification method F is added to the existing Mixup data; the identification method of steps S601 to S604 of the embodiment of the invention is represented by a code S, wherein S0Method for identifying when a decision-making layer parameter λ is equal to 0, S0.5A recognition method when the decision layer parameter λ is 0.5; the data reported in the table are percent identification accuracy.
Table 1:
Figure BDA0002036857580000141
as can be seen from the test results in Table 1, the accuracy of the human skeleton behavior recognition method is superior to that of the existing neural network recognition method without data augmentation and other data augmentation network recognition methods to different degrees no matter what the value of the decision layer parameter lambda is.
Table 2 also shows the accuracy comparison results of the recognition tests performed by using the human skeleton behavior recognition method of the present invention and other existing classical human skeleton behavior recognition methods, respectively, with NTU-CV and NTU-CS as data sets, and the data recorded in the table is the recognition accuracy percentage. The classic human skeleton behavior recognition method shown in the table comprises the following steps: a Global content aware Attention long-short term memory Network (GCA-LSTM), a spatial Attention Temporal Attention LSTM, a Zoneout layer-based Network (hereinafter referred to as Zoneout), an independent Recurrent Neural Network (indRNN), a residual-Temporal Convolutional Network (Res-TCN), an Enhanced Visualization Convolutional Neural Network (EVCNN), a space-Temporal Convolutional Network (ST-GCN); the identification method of steps S601 to S604 using the embodiment of the present invention is denoted by symbol S.
Table 2:
Figure BDA0002036857580000151
as can be seen from table 2, the test accuracy of the human skeleton behavior recognition method of the present invention is higher than that of the existing classical human skeleton behavior recognition method. In addition, although the performances of the identification methods of ST-GCN, EVCNN and indRNN are similar to those of the identification method, the complexity of the networks of ST-GCN, EVCNN, indRNN and the like is higher than that of the neural network based on data augmentation, so that the human skeleton behavior identification method can improve the identification accuracy and control the calculation complexity, thereby improving the identification efficiency.
In the embodiment of the invention, the data classification method is used for solving the problem of human skeleton behavior identification, and as the final human skeleton behavior identification result is based on the data characteristics of the original human body image data to be detected and the data characteristics of the second human body image data to be detected obtained through a transformation network, the accuracy of human skeleton behavior identification can be improved through the complementarity between the original data and the generated data, namely, more data characteristics are combined; experiments prove that the accuracy of the human skeleton behavior identification method in the embodiment of the invention is superior to that of the existing other data augmentation network identification methods and the existing classical human skeleton behavior identification method.
It should be understood that, the sequence numbers of the steps in the foregoing embodiments do not imply an execution sequence, and the execution sequence of each process should be determined by its function and inherent logic, and should not constitute any limitation to the implementation process of the embodiments of the present invention.
Example four:
fig. 7 is a schematic structural diagram of a data classification apparatus provided in an embodiment of the present application, and for convenience of description, only parts related to the embodiment of the present application are shown:
the data classification apparatus includes: an input unit 71, a transformation unit 72, a prediction unit 73, and a classification result determination unit 74. Wherein:
the input unit 71 is configured to input the first data to be measured into a data-augmentation-based neural network, where the data-augmentation-based neural network includes a transformation network, a prediction network, and a decision layer, and the transformation network and the prediction network are both neural networks with deep learning capabilities.
The first data x to be measured refers to data to be classified, and the first data to be measured may be image data, voice data, text data, or the like to be classified. A data-based augmentation neural network, specifically an end-to-end deep neural network with sample data augmentation function, is composed of a transformation network, a prediction network, and a decision layer, as shown in fig. 2. The transformation network and the prediction network are both neural networks with deep learning capability, namely the transformation network can automatically transform the first to-be-detected data after sample training and automatic learning of adjustment parameters; after the prediction network automatically learns and adjusts parameters through sample training, the result prediction can be automatically carried out on input data. The data-augmentation-based neural network in the embodiment of the invention is specifically a neural network trained in advance. Before the data based on the data augmentation neural network is used for data classification, a preset amount of sample data is input to train the data based on the data augmentation neural network, and during training, because the data augmentation based neural network has a data augmentation function, more generated sample data are generated on the basis of original sample data to perform model training, so that the finally trained data based on the data augmentation neural network has stronger generalization capability and judgment capability.
And inputting the first data to be detected into a pre-trained neural network based on data augmentation, and starting an identification and classification processing flow. Optionally, the first data to be measured may be preprocessed, for example, de-noising, data cleaning, and the like, before being input into the data-augmented neural network.
And the transformation unit 72 is configured to obtain second data to be tested through the transformation network according to the first data to be tested.
Inputting the first data x to be tested into a transformation network, learning the characteristics of the first data x to be tested through the transformation network, and generating and outputting data similar to the first data x to be tested in turn, namely the transformation network can input the first data x to be tested into the transformation networkPerforming learning and reconstruction to obtain second data to be tested
Figure BDA0002036857580000171
Optionally, the transform network is an encoding-decoding network structure as shown in fig. 3, the first data x to be measured is input into the encoder for encoding and input into the hidden layer, feature extraction, learning and reconstruction are performed through the hidden layer, and then the generated data corresponding to the first data x to be measured, that is, the second data x to be measured, is obtained through decoding and output by the decoder
Figure BDA0002036857580000172
Specifically, the transformation network may be an Auto-encoder (AE) network structure or a Variational Auto-encoder (VAE) network structure.
Optionally, the transformation unit 72 comprises a first transformation module and/or a second transformation module:
and the first conversion module is used for converting the network into a long-term and short-term memory (LSTM) network if the first data to be measured is one-dimensional data.
And the second transformation module is used for determining that the transformation network is a convolutional neural network if the first data to be tested is two-dimensional data.
And the prediction unit 73 is configured to obtain a first prediction result corresponding to the first data to be detected and a second prediction result corresponding to the second data to be detected, respectively, through the prediction network according to the first data to be detected and the second data to be detected.
The first data x to be tested and the second data to be tested generated according to the first data to be tested
Figure BDA0002036857580000181
Respectively inputting the prediction networks in the neural network based on data augmentation to respectively obtain first prediction results corresponding to the first data x to be measured
Figure BDA0002036857580000182
And second data to be tested
Figure BDA0002036857580000183
Corresponding second prediction result
Figure BDA0002036857580000184
A classification result determining unit 74, configured to obtain a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result.
The first prediction result is obtained
Figure BDA0002036857580000185
And a second predicted result
Figure BDA0002036857580000186
And inputting the data into a decision layer together, and performing weighted fusion calculation to obtain a final classification result y of the first to-be-detected data x. For example, taking image recognition as an example, a recognition result (i.e., a final classification result of the first to-be-detected data) of the first to-be-detected image is obtained through the decision layer according to a first recognition result (i.e., a first prediction result) corresponding to the first to-be-detected image (i.e., the first to-be-detected data) and a second recognition result (i.e., a second prediction result) corresponding to a generated image (i.e., the second to-be-detected data) transformed and reconstructed according to the first to-be-detected image. The identification result may be that the first image to be detected is identified to contain or not contain the target detection object, and the final classification result is one of a yes classification result and a no classification result; or the recognition result can be one of four classification results that the first image to be detected contains the first target detection object, the second target detection object, the third target detection object or does not contain any target detection object; by analogy, the total number of the classified categories can be set according to actual needs, and the final classification result of the first to-be-detected data is one of the total categories.
Optionally, the classification result determining unit 74 includes:
a decision calculation module for obtaining a decision layer parameter lambda according to the first prediction result
Figure BDA0002036857580000187
And a second predicted result
Figure BDA0002036857580000188
Obtaining a final classification result y of the first to-be-detected data through a decision calculation formula, wherein the decision calculation formula specifically comprises:
Figure BDA0002036857580000189
wherein, lambda belongs to [0,1 ].
Optionally, the data classification apparatus further includes:
the system comprises an original sample data acquisition unit, a first result tag acquisition unit and a second result tag acquisition unit, wherein the original sample data comprises first original sample data carrying the first result tag and second original sample data carrying the second result tag;
a sample data generating unit, configured to obtain, according to the first original sample data, first generation sample data carrying a first result tag through a transform network;
the sample fusion unit is used for carrying out sample fusion according to the first generation sample data carrying the first result label and the second original sample data carrying the second result label to obtain fusion sample data carrying a fusion result label;
and the training unit is used for carrying out deep learning training on the data augmentation-based neural network according to the original sample data and the fusion sample data.
Optionally, the sample fusion unit comprises:
a sample fusion calculation module for calculating the fusion of the sample according to the label y carrying the first resultiFirst generation sample data of
Figure BDA0002036857580000191
Carrying a second result label yjSecond original sample data xjAnd a fusion coefficient beta by fusing the sample meterCalculating formula to obtain label carrying fusion result
Figure BDA0002036857580000192
Of (2) fusion sample data
Figure BDA0002036857580000193
The fused sample calculation formula specifically comprises:
Figure BDA0002036857580000194
Figure BDA0002036857580000195
wherein, beta belongs to [0,1 ].
Optionally, the sample fusion calculation module includes:
a fusion coefficient constraint module for carrying the first result label yiFirst generation sample data of
Figure BDA0002036857580000196
Carrying a second result label yjSecond original sample data xjAnd fusion coefficient beta, and obtaining a label carrying a fusion result through a fusion sample calculation formula
Figure BDA0002036857580000197
Of (2) fusion sample data
Figure BDA0002036857580000198
The fused sample calculation formula specifically comprises:
Figure BDA0002036857580000201
Figure BDA0002036857580000202
wherein, beta belongs to [0,1 ].
Optionally, the data classification device is specifically configured to recognize a skeleton behavior of a human body, and the processing data in the data classification device appearing in each unit specifically includes: the first data to be detected is human body image data to be detected, the second data to be detected is second human body image data to be detected obtained through a transformation network according to the human body image data to be detected, and the classification result is a human skeleton behavior recognition result.
In the embodiment of the invention, in the process of classifying the first data to be detected, the second data to be detected is generated through the transformation network, and then the prediction results corresponding to the first data to be detected and the second data to be detected are fused to obtain the final classification result.
Example five:
fig. 8 is a schematic diagram of a terminal device according to an embodiment of the present invention. As shown in fig. 8, the terminal device 8 of this embodiment includes: a processor 80, a memory 81 and a computer program 82, such as a data classification program, stored in said memory 81 and operable on said processor 80. The processor 80, when executing the computer program 82, implements the steps in the various data classification method embodiments described above, such as the steps S101 to S104 shown in fig. 1. Alternatively, the processor 80, when executing the computer program 82, implements the functions of the modules/units in the above-described device embodiments, such as the functions of the units 71 to 74 shown in fig. 8.
Illustratively, the computer program 82 may be partitioned into one or more modules/units that are stored in the memory 81 and executed by the processor 80 to implement the present invention. The one or more modules/units may be a series of computer program instruction segments capable of performing specific functions, which are used to describe the execution of the computer program 82 in the terminal device 8. For example, the computer program 82 may be divided into an input unit, a transformation unit, a prediction unit, and a classification result determination unit, and each unit has the following specific functions:
the device comprises an input unit, a prediction unit and a decision layer, wherein the input unit is used for inputting first data to be tested into a data-augmentation-based neural network, the data-augmentation-based neural network comprises a transformation network, a prediction network and a decision layer, and the transformation network and the prediction network are both neural networks with deep learning capability.
And the transformation unit is used for obtaining second data to be tested through the transformation network according to the first data to be tested.
And the prediction unit is used for respectively obtaining a first prediction result corresponding to the first data to be tested and a second prediction result corresponding to the second data to be tested through the prediction network according to the first data to be tested and the second data to be tested.
And the classification result determining unit is used for obtaining a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result.
The terminal device 8 may be a desktop computer, a notebook, a palm computer, a cloud server, or other computing devices. The terminal device may include, but is not limited to, a processor 80, a memory 81. Those skilled in the art will appreciate that fig. 8 is merely an example of a terminal device 8 and does not constitute a limitation of terminal device 8 and may include more or fewer components than shown, or some components may be combined, or different components, e.g., the terminal device may also include input-output devices, network access devices, buses, etc.
The Processor 80 may be a Central Processing Unit (CPU), other general purpose Processor, a Digital Signal Processor (DSP), an Application Specific Integrated Circuit (ASIC), a Field Programmable Gate Array (FPGA) or other Programmable logic device, discrete Gate or transistor logic, discrete hardware components, etc. A general purpose processor may be a microprocessor or the processor may be any conventional processor or the like.
The storage 81 may be an internal storage unit of the terminal device 8, such as a hard disk or a memory of the terminal device 8. The memory 81 may also be an external storage device of the terminal device 8, such as a plug-in hard disk, a Smart Media Card (SMC), a Secure Digital (SD) Card, a Flash memory Card (Flash Card), and the like, which are provided on the terminal device 8. Further, the memory 81 may also include both an internal storage unit and an external storage device of the terminal device 8. The memory 81 is used for storing the computer program and other programs and data required by the terminal device. The memory 81 may also be used to temporarily store data that has been output or is to be output.
It will be apparent to those skilled in the art that, for convenience and brevity of description, only the above-mentioned division of the functional units and modules is illustrated, and in practical applications, the above-mentioned function distribution may be performed by different functional units and modules according to needs, that is, the internal structure of the apparatus is divided into different functional units or modules to perform all or part of the above-mentioned functions. Each functional unit and module in the embodiments may be integrated in one processing unit, or each unit may exist alone physically, or two or more units are integrated in one unit, and the integrated unit may be implemented in a form of hardware, or in a form of software functional unit. In addition, specific names of the functional units and modules are only for convenience of distinguishing from each other, and are not used for limiting the protection scope of the present application. The specific working processes of the units and modules in the system may refer to the corresponding processes in the foregoing method embodiments, and are not described herein again.
In the above embodiments, the descriptions of the respective embodiments have respective emphasis, and reference may be made to the related descriptions of other embodiments for parts that are not described or illustrated in a certain embodiment.
Those of ordinary skill in the art will appreciate that the various illustrative elements and algorithm steps described in connection with the embodiments disclosed herein may be implemented as electronic hardware or combinations of computer software and electronic hardware. Whether such functionality is implemented as hardware or software depends upon the particular application and design constraints imposed on the implementation. Skilled artisans may implement the described functionality in varying ways for each particular application, but such implementation decisions should not be interpreted as causing a departure from the scope of the present invention.
In the embodiments provided in the present invention, it should be understood that the disclosed apparatus/terminal device and method may be implemented in other ways. For example, the above-described embodiments of the apparatus/terminal device are merely illustrative, and for example, the division of the modules or units is only one logical division, and there may be other divisions when actually implemented, for example, a plurality of units or components may be combined or integrated into another system, or some features may be omitted, or not executed. In addition, the shown or discussed mutual coupling or direct coupling or communication connection may be an indirect coupling or communication connection through some interfaces, devices or units, and may be in an electrical, mechanical or other form.
The units described as separate parts may or may not be physically separate, and parts displayed as units may or may not be physical units, may be located in one place, or may be distributed on a plurality of network units. Some or all of the units can be selected according to actual needs to achieve the purpose of the solution of the embodiment.
In addition, functional units in the embodiments of the present invention may be integrated into one processing unit, or each unit may exist alone physically, or two or more units are integrated into one unit. The integrated unit can be realized in a form of hardware, and can also be realized in a form of a software functional unit.
The integrated modules/units, if implemented in the form of software functional units and sold or used as separate products, may be stored in a computer readable storage medium. Based on such understanding, all or part of the flow of the method according to the embodiments of the present invention may also be implemented by a computer program, which may be stored in a computer-readable storage medium, and when the computer program is executed by a processor, the steps of the method embodiments may be implemented. Wherein the computer program comprises computer program code, which may be in the form of source code, object code, an executable file or some intermediate form, etc. The computer-readable medium may include: any entity or device capable of carrying the computer program code, recording medium, usb disk, removable hard disk, magnetic disk, optical disk, computer Memory, Read-Only Memory (ROM), Random Access Memory (RAM), electrical carrier wave signals, telecommunications signals, software distribution medium, and the like. It should be noted that the computer readable medium may contain content that is subject to appropriate increase or decrease as required by legislation and patent practice in jurisdictions, for example, in some jurisdictions, computer readable media does not include electrical carrier signals and telecommunications signals as is required by legislation and patent practice.
The above-mentioned embodiments are only used for illustrating the technical solutions of the present invention, and not for limiting the same; although the present invention has been described in detail with reference to the foregoing embodiments, it will be understood by those of ordinary skill in the art that: the technical solutions described in the foregoing embodiments may still be modified, or some technical features may be equivalently replaced; such modifications and substitutions do not substantially depart from the spirit and scope of the embodiments of the present invention, and are intended to be included within the scope of the present invention.

Claims (9)

1. A method of data classification, comprising:
acquiring original sample data, wherein the original sample data comprises first original sample data carrying a first result label and second original sample data carrying a second result label;
obtaining first generation sample data carrying a first result label through a transformation network according to the first original sample data;
performing weighted operation on the first generated sample data carrying the first result label and the second original sample data carrying the second result label to realize sample fusion and obtain fusion sample data carrying a fusion result label;
according to the original sample data and the fusion sample data, deep learning training is carried out on a data augmentation-based neural network;
inputting first data to be measured into the data augmentation-based neural network, wherein the data augmentation-based neural network comprises a transformation network, a prediction network and a decision layer, and the transformation network and the prediction network are both neural networks with deep learning capability;
according to the first data to be detected, second data to be detected is obtained through the transformation network;
according to the first to-be-detected data and the second to-be-detected data, respectively obtaining a first prediction result corresponding to the first to-be-detected data and a second prediction result corresponding to the second to-be-detected data through the prediction network;
and obtaining a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result.
2. The data classification method of claim 1,
if the first data to be measured is one-dimensional data, the transformation network is a long-short term memory (LSTM) network;
and if the first data to be detected is two-dimensional data, the transformation network is a convolutional neural network.
3. The data classification method according to claim 1, wherein the performing a weighted operation on the first generated sample data carrying the first result tag and the second original sample data carrying the second result tag to realize sample fusion and obtain fused sample data carrying a fused result tag includes:
according to the label y carrying the first resultiFirst generation sample data of
Figure FDA0002752298670000021
Carrying a second result label yjSecond original sample data xjAnd fusion coefficient beta, and obtaining a label carrying a fusion result through a fusion sample calculation formula
Figure FDA0002752298670000022
Of (2) fusion sample data
Figure FDA0002752298670000023
The fused sample calculation formula specifically comprises:
Figure FDA0002752298670000024
Figure FDA0002752298670000025
wherein, beta belongs to [0,1 ].
4. The data classification method according to claim 3, characterized in that the fusion coefficient β in the fused sample calculation formula satisfies the following constraint:
β∈Beta(a,a)
Figure FDA0002752298670000026
beta (a, a) represents Beta distribution with two parameters being a, N and N respectively represent current iteration times and total target iteration times in the training process, and m is a real number which is greater than 0 and less than or equal to 0.1.
5. The data classification method of claim 1, wherein the obtaining, by the decision layer, a final classification result of the first to-be-detected data according to the first prediction result and the second prediction result comprises:
obtaining a decision layer parameter lambda according to a first prediction result
Figure FDA0002752298670000027
And a second predicted result
Figure FDA0002752298670000028
Obtaining a final classification result y of the first to-be-detected data through a decision calculation formula, wherein the decision calculation formula specifically comprises:
Figure FDA0002752298670000031
wherein, lambda belongs to [0,1 ].
6. The data classification method according to any one of claims 1 to 5, characterized in that the data classification method is specifically a human skeleton behavior recognition method, the first data to be detected is first human body image data to be detected, the second data to be detected is second human body image data to be detected obtained through a transformation network according to the first human body image data to be detected, and the classification result is a human skeleton behavior recognition result.
7. A data sorting apparatus, comprising:
the system comprises an original sample data acquisition unit, a first result tag acquisition unit and a second result tag acquisition unit, wherein the original sample data comprises first original sample data carrying the first result tag and second original sample data carrying the second result tag;
a sample data generating unit, configured to obtain, according to the first original sample data, first generation sample data carrying a first result tag through a transform network;
the sample fusion unit is used for performing weighted operation on the first generation sample data carrying the first result label and the second original sample data carrying the second result label to realize sample fusion and obtain fusion sample data carrying a fusion result label;
the training unit is used for carrying out deep learning training on the neural network based on data augmentation according to the original sample data and the fusion sample data;
the input unit is used for inputting first data to be tested into the data augmentation-based neural network, wherein the data augmentation-based neural network comprises a transformation network, a prediction network and a decision layer, and the transformation network and the prediction network are both neural networks with deep learning capability;
the transformation unit is used for obtaining second data to be tested through the transformation network according to the first data to be tested;
the prediction unit is used for respectively obtaining a first prediction result corresponding to the first data to be tested and a second prediction result corresponding to the second data to be tested through the prediction network according to the first data to be tested and the second data to be tested;
and the classification result determining unit is used for obtaining a final classification result of the first to-be-detected data through the decision layer according to the first prediction result and the second prediction result.
8. A terminal device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that the processor implements the steps of the method according to any of claims 1 to 6 when executing the computer program.
9. A computer-readable storage medium, in which a computer program is stored which, when being executed by a processor, carries out the steps of the method according to any one of claims 1 to 6.
CN201910328170.5A 2019-04-23 2019-04-23 Data classification method and device and terminal equipment Active CN110210513B (en)

Priority Applications (1)

Application Number Priority Date Filing Date Title
CN201910328170.5A CN110210513B (en) 2019-04-23 2019-04-23 Data classification method and device and terminal equipment

Applications Claiming Priority (1)

Application Number Priority Date Filing Date Title
CN201910328170.5A CN110210513B (en) 2019-04-23 2019-04-23 Data classification method and device and terminal equipment

Publications (2)

Publication Number Publication Date
CN110210513A CN110210513A (en) 2019-09-06
CN110210513B true CN110210513B (en) 2021-04-09

Family

ID=67786170

Family Applications (1)

Application Number Title Priority Date Filing Date
CN201910328170.5A Active CN110210513B (en) 2019-04-23 2019-04-23 Data classification method and device and terminal equipment

Country Status (1)

Country Link
CN (1) CN110210513B (en)

Families Citing this family (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN111310892B (en) * 2020-01-20 2022-08-02 南京邮电大学 Language model modeling method based on independent cyclic neural network
CN111291667A (en) * 2020-01-22 2020-06-16 上海交通大学 Method for detecting abnormality in cell visual field map and storage medium
CN111401141B (en) * 2020-02-25 2022-07-15 浙江大学 3D gesture estimation method based on skeleton
CN111382271B (en) * 2020-03-09 2023-05-23 支付宝(杭州)信息技术有限公司 Training method and device of text classification model, text classification method and device
CN111523649B (en) * 2020-05-09 2022-06-10 支付宝(杭州)信息技术有限公司 Method and device for preprocessing data aiming at business model
CN111985645A (en) * 2020-08-28 2020-11-24 北京市商汤科技开发有限公司 Neural network training method and device, electronic equipment and storage medium
CN112232349B (en) * 2020-09-23 2023-11-03 成都佳华物链云科技有限公司 Model training method, image segmentation method and device
CN113160138B (en) * 2021-03-24 2022-07-19 山西大学 Brain nuclear magnetic resonance image segmentation method and system
CN114882333A (en) * 2021-05-31 2022-08-09 北京百度网讯科技有限公司 Training method and device of data processing model, electronic equipment and storage medium
CN113837794A (en) * 2021-08-26 2021-12-24 润联软件系统(深圳)有限公司 Chain retail store sales prediction method based on space-time graph convolutional network

Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108108766A (en) * 2017-12-28 2018-06-01 东南大学 Driving behavior recognition methods and system based on Fusion
CN109544512A (en) * 2018-10-26 2019-03-29 浙江大学 It is a kind of based on multi-modal embryo's pregnancy outcome prediction meanss

Family Cites Families (10)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108734719A (en) * 2017-04-14 2018-11-02 浙江工商大学 Background automatic division method before a kind of lepidopterous insects image based on full convolutional neural networks
CN107909021A (en) * 2017-11-07 2018-04-13 浙江师范大学 A kind of guideboard detection method based on single deep layer convolutional neural networks
CN107748900B (en) * 2017-11-08 2020-09-25 山东财经大学 Mammary gland tumor classification device and storage medium based on discriminative convolutional neural network
CN107945875A (en) * 2017-11-17 2018-04-20 合肥工业大学 Pulmonary nodule detection method and system based on data enhancing
CN108230339B (en) * 2018-01-31 2021-08-03 浙江大学 Stomach cancer pathological section labeling completion method based on pseudo label iterative labeling
CN108564587A (en) * 2018-03-07 2018-09-21 浙江大学 A kind of a wide range of remote sensing image semantic segmentation method based on full convolutional neural networks
CN108460764B (en) * 2018-03-31 2022-02-15 华南理工大学 Ultrasonic image intelligent segmentation method based on automatic context and data enhancement
CN108764082A (en) * 2018-05-17 2018-11-06 淘然视界(杭州)科技有限公司 A kind of Aircraft Targets detection method, electronic equipment, storage medium and system
CN109409198B (en) * 2018-08-31 2023-09-05 平安科技（深圳）有限公司 AU detection method, AU detection device, AU detection equipment and AU detection medium
CN109614979B (en) * 2018-10-11 2023-05-02 北京大学 Data augmentation method and image classification method based on selection and generation

Patent Citations (2)

* Cited by examiner, † Cited by third party
Publication number Priority date Publication date Assignee Title
CN108108766A (en) * 2017-12-28 2018-06-01 东南大学 Driving behavior recognition methods and system based on Fusion
CN109544512A (en) * 2018-10-26 2019-03-29 浙江大学 It is a kind of based on multi-modal embryo's pregnancy outcome prediction meanss

Also Published As

Publication number Publication date
CN110210513A (en) 2019-09-06

Similar Documents

Publication Publication Date Title
CN110210513B (en) Data classification method and device and terminal equipment
CN111860573B (en) Model training method, image category detection method and device and electronic equipment
CN109086811B (en) Multi-label image classification method and device and electronic equipment
CN111027576B (en) Cooperative significance detection method based on cooperative significance generation type countermeasure network
CN112418292B (en) Image quality evaluation method, device, computer equipment and storage medium
CN114067119B (en) Training method of panorama segmentation model, panorama segmentation method and device
CN110942090A (en) Model training method, image processing method, device, electronic equipment and storage medium
CN107729885B (en) Face enhancement method based on multiple residual error learning
Sun et al. Two-stage deep regression enhanced depth estimation from a single RGB image
CN113393385B (en) Multi-scale fusion-based unsupervised rain removing method, system, device and medium
CN114818864A (en) Gesture recognition method based on small samples
CN113158904B (en) Twin network target tracking method and device based on double-mask template updating
CN116861262B (en) Perception model training method and device, electronic equipment and storage medium
CN110768864B (en) Method and device for generating images in batches through network traffic
CN110210314B (en) Face detection method, device, computer equipment and storage medium
CN117033657A (en) Information retrieval method and device
CN111582449A (en) Training method, device, equipment and storage medium for target domain detection network
CN111161238A (en) Image quality evaluation method and device, electronic device, and storage medium
CN115862119A (en) Human face age estimation method and device based on attention mechanism
CN116246161A (en) Method and device for identifying target fine type of remote sensing image under guidance of domain knowledge
CN109165587A (en) intelligent image information extraction method
CN116091763A (en) Apple leaf disease image semantic segmentation system, segmentation method, device and medium
CN114511702A (en) Remote sensing image segmentation method and system based on multi-scale weighted attention
CN113837062A (en) Classification method and device, storage medium and electronic equipment
CN110705611A (en) Fundus image sample expansion method, device, medium, and electronic apparatus

Legal Events

Date Code Title Description
PB01 Publication
PB01 Publication
SE01 Entry into force of request for substantive examination
SE01 Entry into force of request for substantive examination
GR01 Patent grant
GR01 Patent grant