CN111382191A - Machine learning identification method based on deep learning - Google Patents

Abstract

The invention discloses a machine learning identification method based on deep learning, which comprises the following steps: s1, acquiring index data related to the training parameter set based on the BP neural network model according to the identification target parameter; s2, calling a corresponding data mining module and/or a Faster R-CNN model according to the acquired related index data to realize automatic detection and acquisition of the training parameter set; s3, calling a corresponding data preprocessing model according to the identification target parameters and the acquired related index parameters to realize preprocessing of the training parameter set, and acquiring training set data and test set data; and S4, inputting the training set data into the corresponding machine learning model for learning training, then updating the parameters of the neural network according to forward propagation and backward propagation until the model converges, and storing the trained model. The whole process of the invention is automatically completed depending on different neural network models, and the machine learning efficiency can be greatly improved.

Description

A machine learning recognition method based on deep learning

technical field

The invention relates to the field of machine learning, in particular to a deep learning-based machine learning identification method.

Background technique

With the advent of the era of big data, the classification and mining technology of massive data is particularly important. In massive data mining, how to use the information classified and mined from existing data to guide the classification and mining of new data has become a new research hotspot. Especially when the number of samples for some tasks is small, the use of multi-task learning can effectively reduce the time cost of classification and mining of massive data and improve the accuracy of information acquisition.

Deep learning-based methods have been proven to be effective and robust information classification methods in practice. Deep neural networks (such as deep convolutional neural networks) are the most representative machine learning methods. Deep learning models usually have dozens of learnable data processing layers and hundreds of thousands or even millions of learnable parameters. Since a large number of parameters constitute an extremely huge learning space, in order to obtain the optimal model parameters, a large amount of training data is usually required. However, in order to train a deep learning model, it is necessary to build a training dataset with a large number of samples, usually more than tens of thousands of training samples. However, constructing such a training set is very difficult in practical applications.

When deep learning methods are used for classification tasks, traditional deep learning methods require the classification model to compare the class of the samples with the same class as the production samples, that is, the model can only classify the learned classes. If there are new classes of samples that need to be classified, The machine learning model must be retrained, or some adaptive training and learning of the machine learning model must be done. This also leads to the training of machine learning models based on deep learning methods, which requires a lot of training computing resources and long training and learning time, which limits its convenience and versatility in practical applications.

SUMMARY OF THE INVENTION

In order to solve the above problems, the present invention provides a deep learning-based machine learning identification method.

To achieve the above object, the technical scheme adopted in the present invention is:

A deep learning-based machine learning identification method, comprising the following steps:

S1. According to the identification target parameters, the acquisition of relevant index data of the training parameter set is realized based on the BP neural network model;

S2. Call the corresponding data mining module and/or the Faster R-CNN model according to the obtained relevant indicator data to realize automatic detection and collection of the training parameter set;

S3, calling the corresponding data preprocessing model to realize the preprocessing of the training parameter set according to the identification target parameters and the obtained relevant index parameters, and obtain the training set data and the test set data;

S4. Input the training set data into the corresponding machine learning model for learning and training, and then update the parameters of the neural network according to forward propagation and back propagation until the model converges, and save the trained model;

S5. Use the trained model and test set data to predict the model, analyze the model according to the training log and prediction log of the model, draw the training and testing Accracy and loss curves, and judge whether the model can effectively identify target parameters.

Further, when discovering new training data, first build the relationship between the new training data and relevant index parameters, and then convert the new training data into parameters expressed by relevant index parameters, and then input the corresponding trained model. to train.

Further, the data preprocessing model in the step S3 adopts the Inception_V3 neural network model.

Further, when it is found that the data to be identified cannot be identified, first extract the characteristic data of the data to be identified based on the convolutional neural network, then construct the correlation between the characteristic parameters and the relevant index parameters, and then convert the to-be-identified data into expressions using the relevant index parameters. parameters, you can input the corresponding trained model for training.

Further, the training parameter-related index data includes at least the type of the training parameter, the feature or feature set included in the training parameter, and the field to which the training parameter belongs.

In the above scheme:

The acquisition of the training parameter set based on the relevant index data of the training parameter set can improve the pertinence of the training parameter set, reduce the number of training parameters as much as possible, and shorten the model training time.

The automatic detection and collection of the training parameter set is realized through the data mining module and/or the Faster R-CNN model, so that the automatic generation of the training parameter set can be quickly realized.

By constructing the relationship between new training data and relevant index parameters, and then converting the new training data into parameters expressed by relevant index parameters, the fine-tuning of the trained model is realized, so that it can quickly identify new parameters. Recognized function.

The whole process relies on different neural network models to be done automatically, which greatly improves the efficiency of machine learning.

Description of drawings

FIG. 1 is a flowchart of Embodiment 1 of the present invention.

FIG. 2 is a flowchart of Embodiment 2 of the present invention.

FIG. 3 is a flowchart of Embodiment 3 of the present invention.

Detailed ways

The present invention will be described in detail below with reference to specific embodiments. The following examples will help those skilled in the art to further understand the present invention, but do not limit the present invention in any form. It should be noted that, for those skilled in the art, several modifications and improvements can be made without departing from the concept of the present invention. These all belong to the protection scope of the present invention.

Example 1

A deep learning-based machine learning identification method, comprising the following steps:

S1. According to the identification target parameters, the BP neural network model is used to obtain the relevant index data of the training parameter set; the training parameter-related index data at least includes the type (picture/text) of the training parameter, the feature or feature set contained in the training parameter, and the training parameter belongs to. areas, such as finance and multimedia;

S2. Invoke the corresponding data mining module and/or the Faster R-CNN model according to the obtained relevant index data to realize automatic detection and collection of the training parameter set; that is, all the obtained training parameter sets include the relevant index data of the training parameter set. features or feature sets;

S3, call the corresponding Inception_V3 neural network model according to the identification target parameters and the obtained relevant index parameters to realize the preprocessing of the training parameter set, and obtain the training set data and the test set data;

S4. Input the training set data into the corresponding machine learning model for learning and training, and then update the parameters of the neural network according to forward propagation and back propagation until the model converges, and save the trained model;

S5. Use the trained model and test set data to predict the model, analyze the model according to the training log and prediction log of the model, draw the training and testing Accracy and loss curves, and judge whether the model can effectively identify target parameters.

Example 2

A deep learning-based machine learning identification method, comprising the following steps:

S1, realize the acquisition of training parameter set related index data based on the BP neural network model according to the identification target parameter; the training parameter related index data at least include the type of the training parameter, the feature or the feature set that the training parameter includes, and the field to which the training parameter belongs;

S2. Call the corresponding data mining module and/or the Faster R-CNN model according to the obtained relevant indicator data to realize automatic detection and collection of the training parameter set;

S3, call the corresponding Inception_V3 neural network model according to the identification target parameters and the obtained relevant index parameters to realize the preprocessing of the training parameter set, and obtain the training set data and the test set data;

S4. Input the training set data into the corresponding machine learning model for learning and training, and then update the parameters of the neural network according to forward propagation and back propagation until the model converges, and save the trained model;

S5. Use the trained model and test set data to predict the model, analyze the model according to the training log and prediction log of the model, draw the training and testing Accracy and loss curves, and judge whether the model can effectively identify target parameters.

S6. Convert the new training data into parameters expressed by relevant index parameters and input the corresponding trained model for training to obtain a new model; specifically, when new training data is found, first extract the new model based on the convolutional neural network The feature data of the training data, construct the association relationship between the feature data and the relevant indicator parameters (here refers to the features or feature set parameters contained in the training parameters), and then convert the new training data into the relevant indicator parameters (in the training parameters). The parameters expressed by the included features or feature set parameters) can be input into the corresponding trained model for training.

Example 3

A deep learning-based machine learning identification method, comprising the following steps:

S1, realize the acquisition of training parameter set related index data based on the BP neural network model according to the identification target parameter; the training parameter related index data at least include the type of the training parameter, the feature or the feature set that the training parameter includes, and the field to which the training parameter belongs;

S2. Call the corresponding data mining module and/or the Faster R-CNN model according to the obtained relevant indicator data to realize automatic detection and collection of the training parameter set;

S3, call the corresponding Inception_V3 neural network model according to the identification target parameters and the obtained relevant index parameters to realize the preprocessing of the training parameter set, and obtain the training set data and the test set data;

S4. Input the training set data into the corresponding machine learning model for learning and training, and then update the parameters of the neural network according to forward propagation and back propagation until the model converges, and save the trained model;

S5. Use the trained model and test set data to predict the model, analyze the model according to the training log and prediction log of the model, draw the training and testing Accracy and loss curves, and judge whether the model can effectively identify target parameters.

S6. When it is found that the data to be recognized cannot be recognized, convert the data to be recognized into parameters expressed by relevant index parameters and input the corresponding trained model for training to obtain a new model; specifically, first extract the data to be recognized based on the convolutional neural network. Identify the characteristic data of the data, and then construct the association relationship between the characteristic parameters and the relevant index parameters (here refers to the characteristics or characteristic set parameters contained in the training parameters), and then convert the data to be identified into the relevant index parameters (the training parameters include The parameters expressed by the features or feature set parameters) can be input to the corresponding trained model for training.

Specific embodiments of the present invention have been described above. It should be understood that the present invention is not limited to the above-mentioned specific embodiments, and those skilled in the art can make various changes or modifications within the scope of the claims, which do not affect the essential content of the present invention. The embodiments of the present application and features in the embodiments may be combined with each other arbitrarily, provided that there is no conflict.

Claims (5)

1. a machine learning identification method based on deep learning, is characterized in that: comprise the steps: S1. According to the identification target parameters, the acquisition of relevant index data of the training parameter set is realized based on the BP neural network model; S2. Call the corresponding data mining module and/or the Faster R-CNN model according to the obtained relevant indicator data to realize automatic detection and collection of the training parameter set; S3, calling the corresponding data preprocessing model to realize the preprocessing of the training parameter set according to the identification target parameters and the obtained relevant index parameters, and obtain the training set data and the test set data; S4. Input the training set data into the corresponding machine learning model for learning and training, and then update the parameters of the neural network according to forward propagation and back propagation until the model converges, and save the trained model; S5. Use the trained model and test set data to predict the model, analyze the model according to the training log and prediction log of the model, draw the training and testing Accracy and loss curves, and judge whether the model can effectively identify target parameters. 2. a kind of machine learning identification method based on deep learning as claimed in claim 1 is characterized in that: when discovering new training data, first construct the association relation between new training data and relevant index parameter, then new The training data is converted into parameters expressed by relevant index parameters, and the corresponding trained model can be input for training. 3 . The deep learning-based machine learning identification method according to claim 1 , wherein the data preprocessing model in the step S3 adopts the Inception_V3 neural network model. 4 . 4. a kind of machine learning identification method based on deep learning as claimed in claim 1 is characterized in that: when discovering that the data to be identified cannot be identified, at first extract the characteristic data of the data to be identified based on convolutional neural network, then construct the feature The relationship between the parameters and the relevant index parameters, and then the data to be identified is converted into the parameters expressed by the relevant index parameters, and the corresponding trained model can be input for training. 5. A deep learning-based machine learning identification method according to claim 1, wherein the training parameter-related index data at least includes the type of the training parameter, the feature or feature set included in the training parameter, and the training parameter belongs to field.

Images