# US 2019/0377972 A1 · Method and apparatus for training, classification model, mobile terminal, and readable storage medium

> Agents: this data is available as JSON from the patentgrep API and MCP server — see https://patentgrep.com/llms.txt

- Number: US 2019/0377972 A1
- Document ID: US20190377972A1
- Status: Published application
- Kind: A1
- Application number: 16/415,551
- Filed: 2019-05-17
- Published: 2019-12-12
- Earliest claimed: 2018-06-08
- Assignees: Guangdong Oppo Mobile Telecommunications Corp., Ltd.
- Applicants: Guangdong Oppo Mobile Telecommunications Corp., Ltd.
- Inventors: Yaoyong LIU
- Web page: https://patentgrep.com/patent/US20190377972A1

## CPC

- G06F 18/41: {Interactive pattern learning with a human teacher}
- G06F 18/217: {Validation; Performance evaluation; Active pattern learning techniques}
- G06F 18/2413: based on distances to training or reference patterns
- G06N 3/045: Combinations of networks
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/08: Learning methods
- G06N 3/0895: Weakly supervised learning, e.g. semi-supervised or self-supervised learning
- G06N 3/09: Supervised learning
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

Provided are a method and apparatus for training a classification model, a mobile terminal, and a computer-readable storage medium. The method includes that: a classification model is trained based on a preset data set until the accuracy of the classification model reaches a standard value, wherein data in the preset data set includes label information; data in the preset data set is identified based on the trained classification model to acquire category information of the data; when the category information of the data is not matched with the label information, the data is cleaned to acquire a cleaned target data set; and the classification model is trained again based on the target data set.

## Claims

### Claim 1

A method for training a classification model, comprising:
training a classification model based on a preset data set until accuracy of the classification model reaches a standard value;
identifying data in the preset data set based on the trained classification model to acquire category information of the data, the data in the preset data set comprising label information;
cleaning, in response to that the category information of the data is not matched with the label information, the data to acquire a target data set; and
training the classification model based on the target data set.

### Claim 2 (depends on 1)

The method according to claim 1, wherein cleaning, in response to that the category information of the data is not matched with the label information, the data to acquire the cleaned target data set comprises:
verifying correctness of the label information when the category information of the data is not matched with the label information;
correcting, in response to that the label information is not correct, the label information to clean the data; and
forming the target data set based on the cleaned data.

### Claim 3 (depends on 1)

The method according to claim 1, wherein cleaning, in response to that the category information of the data is not matched with the label information, the data to acquire the cleaned target data set comprises:
verifying correctness of the label information when the category information of the data is not matched with the category information;
deleting, in response to that the label information is not correct, the data or adding new data including the label information to clean the data; and
forming the target data set based on the cleaned data.

### Claim 4 (depends on 3)

The method according to claim 3, further comprising:
setting priority of verifying the label information higher than priority of verifying the category information when the category information of the data is not matched with the label information.

### Claim 5 (depends on 1)

The method according to claim 1, wherein the classification model is a neural network model; and
training the classification model based on the preset data set until the accuracy of the classification model reaches the standard value comprises:
acquiring the preset data set, and dividing the preset data set to form a training set and a test set;
inputting the training set including the label information to the neural network model, performing feature extraction through a basic network layer of the neural network model:
inputting an extracted image feature into a classification network layer and a target detection network layer;
obtaining, in the classification network layer, a first loss function reflecting a difference between a first prediction confidence of a specified image category to which a background image in the data belongs and a first true confidence;
obtaining, in the target detection network layer, a second loss function reflecting a difference between a second prediction confidence of a specified object category to which a foreground target in the data belongs and a second true confidence;
performing weighted summation on the first loss function and the second loss function to obtain a target loss function;
adjusting a parameter of the neural network model based on the target loss function; and
testing the test set based on the neural network model repeatedly adjusted, until the accuracy of the neural network model reaches the standard value.

### Claim 6 (depends on 5)

The method according to claim 5, wherein acquiring the preset data set comprises:
acquiring, based on a learning task, a first data set having a first preset quantity and comprising label information;
training the classification model on the first data set, and evaluating the accuracy of the classification model;
screening, when the accuracy does not reach a preset value, unlabeled data based on the trained classification model, and merging the screened data to the first data set to form a second data set; and
performing data cleaning in the second data set to form the preset data set, a data quantity of the second data set being greater than or equal to a data quantity of the preset data set.

### Claim 7 (depends on 1)

The method according to claim 1, further comprising:
acquiring the accuracy of the classification model trained to determine whether data cleaning is effective or not; and
cleaning, when the data cleaning is effective, the data of the preset data set until the accuracy of the classification model reaches a target value.

### Claim 8

An apparatus for training a classification model, comprising:
a memory for storing a computer program, and
a processor capable of executing the computer program to implement steps of
training a classification model based on a preset data set until accuracy of the classification model reaches a standard value;
identifying data in the preset data set based on the trained classification model to acquire category information of the data, the data in the preset data set comprising label information;
cleaning, in response to that the category information of the data is not matched with the label information, the data to acquire a target data set; and
training the classification model based on the target data set.

### Claim 9 (depends on 8)

The apparatus of claim 8, wherein the processor is capable of executing the computer program to further implement steps of:
verifying correctness of the label information when the category information of the data is not matched with the label information;
correcting, in response to that the label information is not correct, the label information to clean the data; and
forming the target data set based on the cleaned data.

### Claim 10 (depends on 8)

The apparatus of claim 8, wherein the processor is capable of executing the computer program to further implement steps of:
verifying correctness of the label information when the category information of the data is not matched with the category information;
deleting, in response to that the label information is not correct, the data or adding new data including the label information to clean the data; and
forming the target data set based on the cleaned data.

### Claim 11 (depends on 10)

The apparatus of claim 10, wherein the processor is capable of executing the computer program to further implement steps of:
setting priority of verifying the label information higher than priority of verifying the category information when the category information of the data is not matched with the label information.

### Claim 12 (depends on 8)

The apparatus of claim 8, wherein the classification model is a neural network model; and
the processor is capable of executing the computer program to further implement steps of:
acquiring the preset data set, and dividing the preset data set to form a training set and a test set;
inputting the training set including the label information to the neural network model, performing feature extraction through a basic network layer of the neural network model;
inputting an extracted image feature into a classification network layer and a target detection network layer;
obtaining, in the classification network layer, a first loss function reflecting a difference between a first prediction confidence of a specified image category to which a background image in the data belongs and a first true confidence;
obtaining, in the target detection network layer, a second loss function reflecting a difference between a second prediction confidence of a specified object category to which a foreground target in the data belongs and a second true confidence;
performing weighted summation on the first loss function and the second loss function to obtain a target loss function;
adjusting a parameter of the neural network model based on the target loss function; and
testing the test set based on the neural network model repeatedly adjusted, until the accuracy of the neural network model reaches the standard value.

### Claim 13 (depends on 12)

The apparatus of claim 12, wherein the processor is capable of executing the computer program to further implement steps of:
acquiring, based on a learning task, a first data set having a first preset quantity and comprising label information;
training the classification model on the first data set, and evaluating the accuracy of the classification model;
screening, when the accuracy does not reach a preset value, unlabeled data based on the trained classification model, and merging the screened data to the first data set to form a second data set; and
performing data cleaning in the second data set to form the preset data set, a data quantity of the second data set being greater than or equal to a data quantity of the preset data set.

### Claim 14 (depends on 1)

The apparatus of claim 1, wherein the processor is capable of executing the computer program to further implement steps of:
acquiring the accuracy of the classification model trained to determine whether data cleaning is effective or not; and
cleaning, when the data cleaning is effective, the data of the preset data set until the accuracy of the classification model reaches a target value.

### Claim 15 (depends on 8)

A mobile terminal, comprising the apparatus of claim 8.

### Claim 16 (depends on 1)

A computer-readable storage medium, having a computer program stored thereon, wherein the computer program is executable by a processor to implement the steps of the method of claim 1.