# US 2021/0342999 A1 · System and method for training a damage identification model

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

- Number: US 2021/0342999 A1
- Document ID: US20210342999A1
- Status: Published application
- Kind: A1
- Application number: 17/374,352
- Filed: 2021-07-13
- Published: 2021-11-04
- Earliest claimed: 2018-08-31
- Assignees: Advanced New Technologies Co., Ltd.
- Applicants: Advanced New Technologies Co., Ltd.
- Inventors: Juan Xu
- Web page: https://patentgrep.com/patent/US20210342999A1

## CPC

- G06F 16/55: Clustering; Classification
- G06F 16/5866: {using information manually generated, e.g. tags, keywords, comments, manually generated location and time information}
- G06F 18/2113: by ranking or filtering the set of features, e.g. using a measure of variance or of feature cross-correlation
- G06F 18/214: Generating training patterns; Bootstrap methods, e.g. bagging or boosting
- G06F 18/24: Classification techniques
- G06F 18/2413: based on distances to training or reference patterns
- G06N 3/0464: Convolutional networks [CNN, ConvNet]
- G06N 3/09: Supervised learning
- G06N 5/04: Inference or reasoning models
- G06N 20/00: Machine learning
- G06T 7/0002: {Inspection of images, e.g. flaw detection}
- G06T 7/0004: {Industrial image inspection}
- G06V 10/40: Extraction of image or video features
- G06V 10/764: using classification, e.g. of video objects
- G06V 10/774: Generating sets of training patterns; Bootstrap methods, e.g. bagging or boosting
- G06V 10/82: using neural networks
- G06V 20/00: Scenes; Scene-specific elements (control of digital cameras H04N23/60)
- G06V 20/20: in augmented reality scenes
- G06N 3/045: Combinations of networks
- G06N 5/01: Dynamic search techniques; Heuristics; Dynamic trees; Branch-and-bound
- G06N 20/20: Ensemble learning
- G06T 2207/20081: Training; Learning
- G06T 2207/30156: Vehicle coating
- G06T 2207/30248: Vehicle exterior or interior
- G06V 2201/08: Detecting or categorising vehicles
- G06F: ELECTRIC DIGITAL DATA PROCESSING (computer systems based on specific computational models G06N)
- G06N: COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06T: IMAGE DATA PROCESSING OR GENERATION, IN GENERAL
- G06V: IMAGE OR VIDEO RECOGNITION OR UNDERSTANDING

## Abstract

A system is provided for identifying damages of a vehicle. During operation, the system can obtain a set of digital images associated with a set of tagged digital images as training data. Each tagged digital image in the set of tagged digital images may include at least one damage object. Based on the training data, the system can train a damage identification model by reducing noise associated with the damage object; generating a set of feature vectors for the training data; and using the set of feature vectors to optimize a set of parameters associated with the damage identification model to obtain a trained damage identification model. The system can then apply the trained damage identification model to obtain a damage category prediction result and determine a repair solution based on the damage category prediction result.

## Claims

### Claim 1

A computer-implemented method for identifying damages of a vehicle, the method comprising:
obtaining, by a computing device, a set of tagged digital images as training data, wherein each tagged digital image includes at least one damage object;
training, based on the training data, a damage identification model, wherein training the damage identification model comprises:
reducing noise associated with the damage object;
generating a set of feature vectors for the training data; and
using the set of feature vectors to optimize a set of parameters associated with the damage identification model to obtain a trained damage identification model;
applying the trained damage identification model to obtain a damage category prediction result; and
determining a repair solution based on the damage category prediction result.

### Claim 2 (depends on 1)

The method of claim 1, wherein an area comprising the damage object is tagged with a damage bounding box and a damage category tag corresponding to the damage bounding box, and wherein the damage category tag indicates a damage material and a damage mode.

### Claim 3 (depends on 2)

The method of claim 2, wherein the damage material includes one of:
a painted piece;
a plastic piece;
a metal piece;
a sanded piece; and
a glass.

### Claim 4 (depends on 2)

The method of claim 2, wherein the damage mode includes one of:
a scratch;
a deformation;
a crack; and
a fragmentation.

### Claim 5 (depends on 2)

The method of claim 2, wherein the damage category tag further indicates a damage degree, and wherein the damage degree corresponds to at least one of:
minor damage;
medium damage; and
severe damage.

### Claim 6 (depends on 1)

The method of claim 1, wherein reducing the noise comprises:
determining whether a sub-area associated with the damage object corresponds to a maximum damaged area;
in response to determining that the sub-area does not include the maximum damaged area, filtering out negative matches; and
retaining positive matches.

### Claim 7 (depends on 6)

The method of claim 6, wherein the negative matches indicate that the one or more portions in the sub-area do not include the damage object, and wherein the positive matches indicate that the one or more portions in the sub-area includes the damage object.

### Claim 8 (depends on 1)

The method of claim 1, wherein applying the trained damage identification model to obtain a damage category prediction result comprises:
obtaining an untagged digital image of a target object;
providing the digital image to the trained damage identification model;
using the trained damage identification model to:
identify a damage object from the digital image; and
determine a damage category prediction result associated with the damage object; and
determining a damage material and a damage mode of the damage object based on the damage category prediction result and a mapping table.

### Claim 9 (depends on 1)

The method of claim 1, wherein training the damage identification model further comprises:
identifying at least a damage object in the training data based on a target detection technique.

### Claim 10

An apparatus for identifying damages of a vehicle, the apparatus comprising:
one or more processors; and
a storage medium storing instructions that, when executed by the one or more processors, cause the apparatus to perform a method comprising:
obtaining, by a computing device, a set of tagged digital images as training data, wherein each tagged digital image includes at least one damage object;
training, based on the training data, a damage identification model, wherein training the damage identification model comprises:
reducing noise associated with the damage object;
generating a set of feature vectors for the training data; and
using the set of feature vectors to optimize a set of parameters associated with the damage identification model to obtain a trained damage identification model;
applying the trained damage identification model to obtain a damage category prediction result; and
determining a repair solution based on the damage category prediction result.

### Claim 11 (depends on 10)

The apparatus of claim 10, wherein an area comprising the damage object is tagged with a damage bounding box and a damage category tag corresponding to the damage bounding box, and wherein the damage category tag indicates a damage material and a damage mode.

### Claim 12 (depends on 10)

The apparatus of claim 10, wherein reducing the noise comprises:
determining whether a sub-area associated with the damage object corresponds to a maximum damaged area;
in response to determining that the sub-area does not include the maximum damaged area, filtering out negative matches; and
retaining positive matches.

### Claim 13 (depends on 12)

The apparatus of claim 12, wherein the negative matches indicate that the one or more portions in the sub-area do not include the damage object, and wherein the positive matches indicate that the one or more portions in the sub-area includes the damage object.

### Claim 14 (depends on 10)

The apparatus of claim 10, wherein applying the trained damage identification model to obtain a damage category prediction result comprises:
obtaining an untagged digital image of a target object;
providing the digital image to the trained damage identification model;
using the trained damage identification model to:
identify a damage object from the digital image; and
determine a damage category prediction result associated with the damage object; and
determine a damage material and a damage mode of the damage object based on the damage category prediction result and a mapping table.

### Claim 15 (depends on 10)

The apparatus of claim 10, wherein training the damage identification model further comprises:
identifying at least a damage object in the training data based on a target detection technique.

### Claim 16

A non-transitory computer-readable storage medium storing instructions that when executed by a computer cause the computer to perform a method for identifying damages of a vehicle, the method comprising:
obtaining, by a computing device, a set of tagged digital images as training data, wherein each tagged digital image includes at least one damage object;
training, based on the training data, a damage identification model, wherein training the damage identification model comprises:
reducing noise associated with the damage object;
generating a set of feature vectors for the training data; and
using the set of feature vectors to optimize a set of parameters associated with the damage identification model to obtain a trained damage identification model;
applying the trained damage identification model to obtain a damage category prediction result; and
determining a repair solution based on the damage category prediction result.

### Claim 17 (depends on 16)

The non-transitory computer-readable storage medium claim 16, wherein an area comprising the damage object is tagged with a damage bounding box and a damage category tag corresponding to the damage bounding box, and wherein the damage category tag indicates a damage material and a damage mode.

### Claim 18 (depends on 16)

The non-transitory computer-readable storage medium claim 16, wherein reducing the noise comprises:
determining whether a sub-area associated with the damage object corresponds to a maximum damaged area;
in response to determining that the sub-area does not include the maximum damaged area, filtering out negative matches; and
retaining positive matches.

### Claim 19 (depends on 18)

The non-transitory computer-readable storage medium claim 18, wherein the negative matches indicate that the one or more portions in the sub-area do not include the damage object, and wherein the positive matches indicate that the one or more portions in the sub-area includes the damage object.

### Claim 20 (depends on 16)

The non-transitory computer-readable storage medium claim 16, wherein applying the trained damage identification model to obtain the damage category prediction result comprises:
obtaining an untagged digital image of a target object;
providing the digital image to the trained damage identification model;
using the trained damage identification model to:
identify a damage object from the digital image; and
determine a damage category prediction result associated with the damage object; and
determine a damage material and a damage mode of the damage object based on the damage category prediction result and a mapping table.
