# US 2024/0203081 A1 · Processing method for image recognition model and related product

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

- Number: US 2024/0203081 A1
- Document ID: US20240203081A1
- Status: Published application
- Kind: A1
- Application number: 18/587,550
- Filed: 2024-02-26
- Published: 2024-06-20
- Earliest claimed: 2022-08-01
- Assignees: Mashang Consumer Finance Co., Ltd.
- Applicants: MaShang Consumer Finance Co., Ltd.
- Inventors: Yongchun LV; Hui ZHU; Xunyi ZHOU; Ning JIANG; Haiying WU
- Web page: https://patentgrep.com/patent/US20240203081A1

## CPC

- G06V 10/22: by selection of a specific region containing or referencing a pattern; Locating or processing of specific regions to guide the detection or recognition
- G06V 10/25: Determination of region of interest [ROI] or a volume of interest [VOI]
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/771: Feature selection, e.g. selecting representative features from a multi-dimensional feature space
- G06V 10/7715: {Feature extraction, e.g. by transforming the feature space, e.g. multi-dimensional scaling [MDS]; Mappings, e.g. subspace methods}
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 10/82: using neural networks
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

Embodiments of the present application provide a processing method for an image recognition model and a related product. The processing method for the image recognition model includes: obtaining an image sample; determining a target object positioning box of the image sample, where the target object positioning box covers an area where an effective feature of the image sample is located; adjusting a pixel value of at least one pixel within the target object positioning box to obtain a preprocessed image sample; and training a network model based on the preprocessed image sample to obtain the image recognition model. The present application improves the concentration of the effective feature in the sample image and thus improves the regularization effect and reduces overfitting of the network model.

## Claims

### Claim 1

A processing method for an image recognition model, comprising:
obtaining an image sample;
determining a target object positioning box of the image sample, wherein the target object positioning box covers an area where an effective feature of the image sample is located;
adjusting a pixel value of at least one pixel within the target object positioning box to obtain a preprocessed image sample; and
training a network model based on the preprocessed image sample to obtain the image recognition model.

### Claim 2 (depends on 1)

The method according to claim 1, further comprising:
obtaining an image sample set, wherein the image sample set comprises at least one of the image sample.

### Claim 3 (depends on 2)

The method according to claim 2, wherein the adjusting the pixel value of the at least one pixel within the target object positioning box to obtain the preprocessed image sample comprises:
adjusting the pixel value of the at least one pixel within the target object positioning box of each image sample, and forming the at least one preprocessed image sample into a preprocessed image sample set.

### Claim 4 (depends on 1)

The method according to claim 1, wherein the determining the target object positioning box of the image sample comprises:
inputting the image sample into a self-supervised learning model to obtain feature map information, wherein the self-supervised learning model is used to extract the effective feature of the image sample; and
determining the target object positioning box of the image sample according to the feature map information.

### Claim 5 (depends on 4)

The method according to claim 4, wherein an amount of the feature map information is N, and N is a natural number greater than or equal to 1, and the determining the target object positioning box of the image sample according to the feature map information comprises:
for each piece of the feature map information, performing a normalization processing on the feature map information to obtain a heat map, determining a target point with a heat value greater than a preset threshold from the heat map, and determining an initial target object positioning box based on the target point, wherein the heat map contains at least one pixel, a magnitude of a heat value of the pixel represents a probability that the pixel contains an effective feature, and the heat value of the at least one pixel is within a preset interval; and
determining an average value of the N initial target object positioning boxes to obtain the target object positioning box.

### Claim 6 (depends on 5)

The method according to claim 5, wherein the initial target object positioning box is a polygonal box, and the determining the average value of the N initial target object positioning boxes to obtain the target object positioning box comprises:
for each vertex of any one of the N initial target object positioning boxes, determining an average value of coordinates corresponding to each vertex, wherein a number of coordinates of each vertex is N; and
obtaining the target object positioning box according to the average value of the coordinates corresponding to each vertex.

### Claim 7 (depends on 3)

The method according to claim 3, wherein the adjusting the pixel value of the at least one pixel within the target object positioning box of each image sample, and forming the at least one preprocessed image sample into the preprocessed image sample set comprises:
for each image sample, generating an associated image corresponding to the image sample according to the target object positioning box of the image sample, wherein a size of the associated image is identical to a size of the image sample; and
adjusting the pixel value of the at least one pixel within the target object positioning box of each image sample according to the associated image corresponding to each image sample to obtain the preprocessed image sample set.

### Claim 8 (depends on 7)

The method according to claim 7, wherein the generating the associated image corresponding to the image sample according to the target object positioning box of the image sample comprises:
generating an initial associated image with a same size as the image sample;
selecting M center points from the initial associated image, and determining M target areas based on the M center points, wherein M is a natural number greater than or equal to 1, and the M center points are pixels within the target object positioning box; and
setting a pixel value of at least one pixel in the initial associated image to generate the associated image corresponding to the image sample, wherein the pixel in the initial associated image contains a pixel corresponding to the M target areas and a pixel corresponding to an area other than the M target areas, and a pixel value of the pixel corresponding to the M target areas is set in a different manner than a pixel value of the pixel corresponding to the area other than the M target areas.

### Claim 9 (depends on 8)

The method according to claim 8, wherein the setting the pixel value of the at least one pixel in the initial associated image to generate the associated image corresponding to the image sample comprises:
randomly setting the pixel value of the pixel corresponding to the M target areas;
setting the pixel value of the pixel corresponding to the area other than the M target areas in the initial associated image to 1 to obtain the associated image corresponding to the image sample;
correspondingly, the adjusting the pixel value of the at least one pixel within the target object positioning box of each image sample according to the associated image corresponding to each image sample to obtain the preprocessed image sample set comprises:
for each image sample, multiplying a pixel value of each pixel of the image sample by a pixel value of a corresponding pixel of the associated image to obtain the preprocessed image sample set.

### Claim 10 (depends on 1)

The method according to claim 1, further comprising:
obtaining a to-be-recognized image, wherein the to-be-recognized image contains a recognition object; and
inputting the to-be-recognized image into the image recognition model to obtain a recognition result.

### Claim 11

An electronic device, comprising: a processor, and a memory communicatively connected to the processor;
wherein the memory stores computer-executed instructions;
the processor, when executing the computer-executed instructions stored in the memory, is configured to:
obtain an image sample;
determine a target object positioning box of the image sample, wherein the target object positioning box covers an area where an effective feature of the image sample is located;
adjust a pixel value of at least one pixel within the target object positioning box to obtain a preprocessed image sample; and
train a network model based on the preprocessed image sample to obtain the image recognition model.

### Claim 12 (depends on 11)

The electronic device according to claim 11, wherein the processor is configured to:
obtain an image sample set, wherein the image sample set comprises at least one of the image sample.

### Claim 13 (depends on 12)

The electronic device according to claim 12, wherein the processor is configured to:
adjust the pixel value of the at least one pixel within the target object positioning box of each image sample, and form the at least one preprocessed image sample into a preprocessed image sample set.

### Claim 14 (depends on 11)

The electronic device according to claim 11, wherein the processor is configured to:
input the image sample into a self-supervised learning model to obtain feature map information, wherein the self-supervised learning model is used to extract the effective feature of the image sample; and
determine the target object positioning box of the image sample according to the feature map information.

### Claim 15 (depends on 14)

The electronic device according to claim 14, wherein an amount of the feature map information is N, and Nis a natural number greater than or equal to 1, and the processor is configured to:
for each piece of the feature map information, perform a normalization processing on the feature map information to obtain a heat map, determine a target point with a heat value greater than a preset threshold from the heat map, and determine an initial target object positioning box based on the target point, wherein the heat map contains at least one pixel, a magnitude of a heat value of the pixel represents a probability that the pixel contains an effective feature, and the heat value of the at least one pixel is within a preset interval; and
determine an average value of the N initial target object positioning boxes to obtain the target object positioning box.

### Claim 16 (depends on 15)

The electronic device according to claim 15, wherein the initial target object positioning box is a polygonal box, and the processor is configured to:
for each vertex of any one of the N initial target object positioning boxes, determine an average value of coordinates corresponding to each vertex, wherein a number of coordinates of each vertex is N; and
obtain the target object positioning box according to the average value of the coordinates corresponding to each vertex.

### Claim 17 (depends on 13)

The electronic device according to claim 13, wherein the processor is configured to:
for each image sample, generate an associated image corresponding to the image sample according to the target object positioning box of the image sample, wherein a size of the associated image is identical to a size of the image sample; and
adjust the pixel value of the at least one pixel within the target object positioning box of each image sample according to the associated image corresponding to each image sample to obtain the preprocessed image sample set.

### Claim 18 (depends on 17)

The electronic device according to claim 17, wherein the processor is configured to:
generate an initial associated image with a same size as the image sample;
select M center points from the initial associated image, and determine M target areas based on the M center points, wherein M is a natural number greater than or equal to 1, and the M center points are pixels within the target object positioning box; and
set a pixel value of at least one pixel in the initial associated image to generate the associated image corresponding to the image sample, wherein the pixel in the initial associated image contains a pixel corresponding to the M target areas and a pixel corresponding to an area other than the M target areas, and a pixel value of the pixel corresponding to the M target areas is set in a different manner than a pixel value of the pixel corresponding to the area other than the M target areas.

### Claim 19 (depends on 18)

The electronic device according to claim 18, wherein the processor is configured to:
randomly set the pixel value of the pixel corresponding to the M target areas;
set the pixel value of the pixel corresponding to the area other than the M target areas in the initial associated image to 1 to obtain the associated image corresponding to the image sample; and
for each image sample, multiply a pixel value of each pixel of the image sample by a pixel value of a corresponding pixel of the associated image to obtain the preprocessed image sample set.

### Claim 20

A non-transitory computer-readable storage medium with computer-executed instructions stored therein, wherein a processor, when executing the computer-executed instructions, is configured to:
obtain an image sample;
determine a target object positioning box of the image sample, wherein the target object positioning box covers an area where an effective feature of the image sample is located;
adjust a pixel value of at least one pixel within the target object positioning box to obtain a preprocessed image sample; and
train a network model based on the preprocessed image sample to obtain the image recognition model.
