# US 12,182,984 B2 · Repair method and repair apparatus for flexible panel, device and storage medium

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

- Number: US 12,182,984 B2
- Document ID: US12182984B2
- Status: Granted
- Kind: B2
- Application number: 17/478,711
- Filed: 2021-09-17
- Published: 2022-09-29
- Granted: 2024-12-31
- Earliest claimed: 2021-03-29
- Assignees: Beijing Xiaomi Mobile Software Co., Ltd.
- Applicants: Beijing Xiaomi Mobile Software Co., Ltd.
- Inventors: Xiao Xiang; Jian Li
- Web page: https://patentgrep.com/patent/US12182984B2

## CPC

- G06T 7/0004: {Industrial image inspection}
- G06T 7/001: {using an image reference approach}
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/2431: Multiple classes
- G06N 20/00: Machine learning
- G06T 2207/20081: Training; Learning
- G06T 2207/20084: Artificial neural networks [ANN]
- G06T 2207/30108: Industrial image inspection
- G06T 2207/30121: CRT, LCD or plasma display
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS

## Abstract

A repair method for a flexible panel. The method includes: obtaining first image information of a flexible panel to be repaired; determining a first classification tag of the flexible panel to be repaired based on a deep learning model and the first image information; and determining an operating parameter of a repair device according to the first classification tag, and repairing the flexible panel to be repaired based on the operating parameter.

## Claims

### Claim 1

A method for repairing a flexible panel, comprising:
obtaining first image information of a flexible panel to be repaired;
determining a first classification tag of the flexible panel to be repaired based on a deep learning model and the first image information; and
determining an operating parameter of a repair device according to the first classification tag, and repairing the flexible panel to be repaired based on the operating parameter;
wherein the method further comprises:
determining the deep learning model based on a preset number of flexible panel samples to be repaired,
wherein determining the deep learning model based on a preset number of flexible panel samples to be repaired comprises:
dividing the preset number of flexible panel samples to be repaired into training flexible panel samples and verification flexible panel samples according to a preset ratio;
training with the training flexible panel samples to obtain a first model;
inputting the verification flexible panel samples into the first model to obtain a first prediction accuracy rate of the first model; and
determining the deep learning model based on the first prediction accuracy rate;
wherein training with the training flexible panel samples to obtain the first model comprises:
obtaining second image information of each training flexible panel sample;
for each training flexible panel sample, obtaining third image information of a repaired flexible panel sample after the training flexible panel sample is repaired based on the adjusted candidate operating parameter;
determining whether the third image information meets a repair standard for each repaired flexible panel sample;
in response to the third image information of the repaired flexible panel sample not meeting the repair standard, adjusting the candidate operating parameter and repairing the training flexible panel sample based on the adjusted candidate operating parameter, until obtained image information of the repaired flexible panel sample meets the repair standard;
determining a candidate operating parameter corresponding to the obtained image information meeting the repair standard as a second classification tag of the training flexible panel sample;
obtaining the second image information and a second classification tag of each training flexible panel sample, to form a second image information collection and a second classification tag collection;
determining the second image information collection and the second classification tag collection as training sample data; and
obtaining the first model by training with the training sample data.

### Claim 2 (depends on 1)

The method of claim 1, wherein training with the training flexible panel samples to obtain the first model comprises:
in response to the third image information of the repaired flexible panel sample not meeting the repair standard, adjusting the candidate operating parameter, repairing the training flexible panel sample based on the adjusted candidate operating parameter, and determining whether the obtained image information of the repaired flexible panel sample meets the repair standard after the candidate operating parameter is adjusted for m times;
in response to the image information of the repaired flexible panel sample not meeting the repair standard after the candidate operating parameter is adjusted for m times, determining a preset classification tag as the second classification tag of the training flexible panel sample, where m is a preset number.

### Claim 3 (depends on 1)

The method of claim 1, wherein different training flexible panels samples have different defect characteristics, the defect characteristics comprise at least one of: a defect classification and defect area.

### Claim 4 (depends on 1)

The method of claim 1, wherein determining the deep learning model based on the first prediction accuracy rate comprises:
in response to the first prediction accuracy rate being smaller than a preset accuracy rate, adjusting a model parameter of the first model until the first prediction accuracy rate of the first model is greater than or equal to the preset accuracy rate; and
determining the first model with the first prediction accuracy rate greater than or equal to the preset accuracy rate as the deep learning model.

### Claim 5 (depends on 1)

The method of claim 1, wherein determining the deep learning model based on the first prediction accuracy rate comprises:
in response to the first prediction accuracy rate being smaller than a preset accuracy rate, determining a second prediction accuracy rate of an alternative second model; and
determining a second model with a second prediction accuracy rate greater than or equal to the preset accuracy rate as the deep learning model.

### Claim 6 (depends on 1)

The method of claim 1, wherein inputting the verification flexible panel samples into the first model to obtain the first prediction accuracy rate of the first model comprises:
obtaining a third classification tag collection of the verification flexible panel samples;
inputting the verification flexible panel samples into the first model to obtain a prediction result collection of the verification flexible panel samples; and
determining the first prediction accuracy rate of the first model based on the prediction result collection and the third classification tag collection.

### Claim 7 (depends on 1)

The method of claim 1, wherein repairing the flexible panel to be repaired based on the operating parameter comprises:
in response to the repair device being multiple, repairing the flexible panel to be repaired by delivering the flexible panel to be repaired to a repair device having a first target operating parameter corresponding to the first classification tag; or
in response to the repair device being one, adjusting the operating parameter of the repair device to a second target operating parameter corresponding to the first classification tag, and repairing the flexible panel to be repaired based on the second target operating parameter.

### Claim 8

A device for repairing a flexible panel, comprising:
a processor; and
a memory capable to store computer programs executable by the processor;
wherein the processor implements a repair method for a flexible panel when executing the computer programs, the method comprising:
obtaining first image information of a flexible panel to be repaired;
determining a first classification tag of the flexible panel to be repaired based on a deep learning model and the first image information; and
determining an operating parameter of a repair device according to the first classification tag, and repairing the flexible panel to be repaired based on the operating parameter;
wherein the method comprises:
determining the deep learning model based on a preset number of flexible panel samples to be repaired,
wherein determining the deep learning model based on a preset number of flexible panel samples to be repaired comprises:
dividing the preset number of flexible panel samples to be repaired into training flexible panel samples and verification flexible panel samples according to a preset ratio;
training with the training flexible panel samples to obtain a first model;
inputting the verification flexible panel samples into the first model to obtain a first prediction accuracy rate of the first model; and
determining the deep learning model based on the first prediction accuracy rate;
wherein training with the training flexible panel samples to obtain the first model comprises:
obtaining second image information of each training flexible panel sample;
for each training flexible panel sample, obtaining third image information of a repaired flexible panel sample after the training flexible panel sample is repaired based on the adjusted candidate operating parameter;
determining whether the third image information meets a repair standard for each repaired flexible panel sample;
in response to the third image information of the repaired flexible panel sample not meeting the repair standard, adjusting the candidate operating parameter and repairing the training flexible panel sample based on the adjusted candidate operating parameter, until obtained image information of the repaired flexible panel sample meets the repair standard;
determining a candidate operating parameter corresponding to the obtained image information meeting the repair standard as a second classification tag of the training flexible panel sample;
obtaining the second image information and a second classification tag of each training flexible panel sample, to form a second image information collection and a second classification tag collection;
determining the second image information collection and the second classification tag collection as training sample data; and
obtaining the first model by training with the training sample data.

### Claim 9 (depends on 8)

The device of claim 8, wherein training with the training flexible panel samples to obtain the first model comprises:
in response to the third image information of the repaired flexible panel sample not meeting the repair standard, adjusting the candidate operating parameter, repairing the training flexible panel sample based on the adjusted candidate operating parameter, and determining whether the obtained image information of a the repaired flexible panel sample meets the repair standard after the candidate operating parameter is adjusted for m times;
in response to the image information of the repaired flexible panel sample not meeting the repair standard after the candidate operating parameter is adjusted for m times, determining a preset classification tag as the second classification tag of the training flexible panel samples, where m is a preset number.

### Claim 10 (depends on 8)

The device of claim 8, wherein different training flexible panels samples have different defect characteristics, the defect characteristics comprise at least one of: a defect classification and defect area.

### Claim 11 (depends on 8)

The device of claim 8, wherein determining the deep learning model based on the first prediction accuracy rate comprises:
in response to the first prediction accuracy rate being smaller than a preset accuracy rate, adjusting a model parameter of the first model until the first prediction accuracy rate of the first model is greater than or equal to the preset accuracy rate; and
determining the first model with the first prediction accuracy rate greater than or equal to the preset accuracy rate as the deep learning model.

### Claim 12 (depends on 8)

The device of claim 8, wherein determining the deep learning model based on the first prediction accuracy rate comprises:
in response to the first prediction accuracy rate being smaller than a preset accuracy rate, determining a second prediction accuracy rate of an alternative second model; and
determining a second model with a second prediction accuracy rate greater than or equal to the preset accuracy rate as the deep learning model.

### Claim 13 (depends on 8)

The device of claim 8, wherein inputting the verification flexible panel samples into the first model to obtain the first prediction accuracy rate of the first model comprises:
obtaining a third classification tag collection of the verification flexible panel samples;
inputting the verification flexible panel samples into the first model to obtain a prediction result collection of the verification flexible panel samples; and
determining the first prediction accuracy rate of the first model based on the prediction result collection and the third classification tag collection.

### Claim 14 (depends on 8)

The device of claim 8, wherein repairing the flexible panel to be repaired based on the operating parameter comprises:
in response to the repair device being multiple, repairing the flexible panel to be repaired by delivering the flexible panel to be repaired to a repair device having a first target operating parameter corresponding to the first classification tag; or
in response to the repair device being one, adjusting the operating parameter of the repair device to a second target operating parameter corresponding to the first classification tag, and repairing the flexible panel to be repaired based on the second target operating parameter.

### Claim 15

A non-transitory computer-readable storage medium having stored therein instructions that, when executed by a processor, causes the processor to perform a repair method for a flexible panel, the method comprising:
obtaining first image information of a flexible panel to be repaired;
determining a first classification tag of the flexible panel to be repaired based on a deep learning model and the first image information; and
determining an operating parameter of a repair device according to the first classification tag, and repairing the flexible panel to be repaired based on the operating parameter;
wherein the method comprises:
determining the deep learning model based on a preset number of flexible panel samples to be repaired,
wherein, determining the deep learning model based on a preset number of flexible panel samples to be repaired comprises:
dividing the preset number of flexible panel samples to be repaired into training flexible panel samples and verification flexible panel samples according to a preset ratio;
training with the training flexible panel samples to obtain a first model;
inputting the verification flexible panel samples into the first model to obtain a first prediction accuracy rate of the first model; and
determining the deep learning model based on the first prediction accuracy rate;
wherein training with the training flexible panel samples to obtain the first model comprises:
obtaining second image information of each training flexible panel sample;
for each training flexible panel sample, obtaining third image information of a repaired flexible panel sample after the training flexible panel sample is repaired based on the adjusted candidate operating parameter;
determining whether the third image information meets a repair standard for each repaired flexible panel sample;
in response to the third image information of the repaired flexible panel sample not meeting the repair standard, adjusting the candidate operating parameter and repairing the training flexible panel sample based on the adjusted candidate operating parameter, until obtained image information of the repaired flexible panel sample meets the repair standard;
determining a candidate operating parameter corresponding to the obtained image information meeting the repair standard as a second classification tag of the training flexible panel sample;
obtaining the second image information and a second classification tag of each training flexible panel sample, to form a second image information collection and a second classification tag collection;
determining the second image information collection and the second classification tag collection as training sample data; and
obtaining the first model by training with the training sample data.