You are on page 1of 11

CN114491127A Vehicle insurance buying information detection method and device, computer equipment

and storage medium

Data originating from sources other than the EPO may not be accurate, complete, or up to date.

The wording below is an initial machine translation of the original publication. To generate a version using the
latest translation technology, go to the original language text and use Patent Translate.

Detection method, device, computer equipment and storage medium for auto insurance application information

technical field

The present application relates to the field of data detection, and in particular, to a detection method, device,
computer equipment and storage medium for auto insurance application information.

Background technique

In the case of vehicle insurance claims, double insurance fraud in the field of vehicle insurance refers to the
fraudulent behavior formed by multiple claims applications for the same damage to the same accident vehicle
within the same time period without repairs. For the above fraudulent behaviors, the traditional solution mainly
relies on the professional quality and historical experience of the surveyors, and manually compares the reported
cases with the historical cases in the system, which is highly subjective and uncontrollable. If the surveyor does
not actively report the repeated insurance application, the system will not be able to detect it in time; or extract
the insurance type information from the insurance application instruction, and use the processing logic
framework to obtain the metadata to realize the comparison and confirmation of the repeated insurance
application. The above solutions have strong limitations. , The application range is not wide, resulting in low
accuracy and low efficiency of repeated insurance detection in current vehicle damage cases.

SUMMARY OF THE INVENTION

The main purpose of this application is to provide a detection method, device, computer equipment and storage
medium for auto insurance application information, which aims to solve the problems of low accuracy and low
efficiency of repeated insurance application detection in current vehicle damage cases.

In order to achieve the above purpose of the invention, the present application proposes a method for detecting
auto insurance information, including:

Receive report information of car damage cases;

Identify the report information, obtain the unique registration information of the vehicle and the image of the
vehicle damage;

Extract the damage feature of the vehicle damage image based on the deep learning network, and convert the
damage feature into a feature vector;

The feature vector is segmented and encoded to obtain feature encoding;

Obtain the historical feature code of the historical vehicle damage image of the vehicle corresponding to the
unique registration information;

The feature code is compared with the historical feature code, and if the similarity between the feature code and
the historical feature code is higher than a preset threshold, it is determined that the report information is
repeated report information.

Further, the deep rest network includes a backbone network, a region candidate network and an output network;
the extraction of damage features of the vehicle damage image based on the deep learning network includes:
Inputting the vehicle damage image to the backbone network, and performing convolution and pooling
calculations on the vehicle damage image based on the backbone network to obtain a first intermediate feature
map;

Inputting the first intermediate feature map to the region candidate network, and performing boundary
regression and region-of-interest alignment calculations on the first intermediate feature map based on the
region candidate network to obtain a second intermediate feature map;

Input the second intermediate feature map to the output network, calculate the damage frame and damage type
of the vehicle damage corresponding to the second feature map based on the output network, and determine the
vehicle damage according to the damage frame and damage type Image damage characteristics.

Further, calculating the damage frame and damage type of the vehicle damage corresponding to the second
feature map based on the output network includes:

computing bounding box coordinates of the second feature map based on the output network;

Determine the damage frame of the vehicle damage according to the bounding frame coordinates;

Calculate the degree of loss between the image in the damaged frame and the standard image;

The damage type of the vehicle damage is determined according to the damage degree.

Further, the described feature vector is segmented and encoded to obtain feature encoding, including:

obtain the length of the feature vector;

Divide the feature vector into several sub-vectors with the same length according to the length;

Randomly select vector values at several positions from the sub-vectors;

Calculate a target value according to the vector value and based on a preset mapping rule, and configure the
target value as a segmented subcoding of the subvector;

The segmented auto-encodings are combined to obtain feature encodings.

Further, if the similarity between the feature code and the historical feature code is higher than a preset
threshold, it is determined that the report information is repeated report information, including:

obtaining the segmented subcode of the feature code;

obtaining the historical segment subcoding of the historical feature coding;

When the same number of segment sub-codes as the historical segment sub-codes is greater than a preset value,
it is determined that the similarity between the feature code and the historical feature code is higher than a preset
threshold, and it is determined that the report information is repeated report information.

Further, before the time when the same number of the segment sub-codings as the historical segment sub-
codings is greater than a preset value, it also includes:

Get the detection stage of the current auto insurance coverage;

Configure the number of resources according to the detection phase;

The value of the preset value is determined according to the number of resources.

Further, after the feature vector is segmented and encoded to obtain the feature encoding, it also includes:

generating block information after associating the report information, the vehicle damage image and the feature
code;
Configuring the unique registration information of the vehicle as block header information;

A block chain is generated according to the block information and the block header information.

The application also provides a detection device for auto insurance application information, including:

The information acquisition module is used to receive report information of vehicle damage cases;

The information identification module is used to identify the report information, obtain the unique registration
information of the vehicle and the vehicle damage image;

a feature extraction module, used for extracting the damage feature of the vehicle damage image based on the
deep learning network, and converting the damage feature into a feature vector;

A vector coding module, for coding the feature vector after segmentation to obtain feature coding;

A historical case module, used to obtain the historical feature code of the historical vehicle damage image of the
vehicle corresponding to the unique registration information;

A repeat detection module, used to compare the feature code with the historical feature code, and if the
similarity between the feature code and the historical feature code is higher than a preset threshold, determine
that the report information is repeated report information .

The present application also provides a computer-readable storage medium on which a computer program is
stored, and when the computer program is executed by a processor, implements the steps of any of the foregoing
methods for detecting auto insurance application information.

This application example provides a method for feature extraction based on a car damage image and a method
for detecting whether an insured case is repeatedly insured after comparing the features after coding. To detect
the existence of repeated insurance, first receive the report information of the car damage case, identify the
report information, obtain the unique registration information of the vehicle and the car damage image, and
extract the damage characteristics of the car damage image based on the deep learning network, including the
damage part. Image features such as image gray value, image texture, image cracks, etc., based on the image
features, the corresponding damage features of the vehicle damage image can be determined, and then the
damage features are converted into feature vectors to represent different damages in the form of vectors. In
order to more easily represent different damage features, facilitate the storage of damage features and the
comparison of damage features, the feature vector is divided and encoded to obtain the feature code, and then
obtain The historical feature code of the historical vehicle damage image of the vehicle corresponding to the
unique registration information, that is, the historical feature code obtained after feature extraction and encoding
of the historical vehicle damage image in the historical report information, and the feature code is combined
with the historical feature code. The feature codes are compared, the numerical values of the corresponding bits
of the feature codes and the historical feature codes are calculated, and the similarity between the feature codes
and the historical feature codes is determined according to the difference in the numerical values. When the
similarity is higher than A preset threshold is used to determine that the report information is repeated report
information, so as to quickly and accurately detect whether the insurance application of the same vehicle
damage case has repeated insurance coverage, and improve the detection efficiency and accuracy of auto
insurance insurance information.

Description of drawings

1 is a schematic flowchart of an embodiment of a method for detecting auto insurance application information of
the application;

FIG. 2 is a schematic structural diagram of an embodiment of an apparatus for detecting auto insurance
application information of the application;

FIG. 3 is a schematic structural block diagram of an embodiment of a computer device of the present
application.
The realization, functional characteristics and advantages of the purpose of the present application will be
further described with reference to the accompanying drawings in conjunction with the embodiments.

Detailed ways

In order to make the purpose, technical solutions and advantages of the present application more clearly
understood, the present application will be described in further detail below with reference to the accompanying
drawings and embodiments. It should be understood that the specific embodiments described herein are only
used to explain the present application, but not to limit the present application.

Referring to FIG. 1 , an embodiment of the present application provides a method for detecting auto insurance
application information, including steps S10 - S60 . The steps of the method for detecting vehicle insurance
application information are described in detail as follows.

S10. Receive the report information of the vehicle damage case.

This embodiment is applied to the claim detection scenario of vehicle damage assessment. With the continuous
development of digitization, insurance companies record vehicle damage cases in a digital form. In this
embodiment, an insurance case detection platform is developed to detect vehicle damage cases. Check whether
there is duplicate insurance coverage. First, receive the report information of the car damage case, that is, when
there is a car damage case reported, upload the car damage case to the insurance case detection platform, so as
to receive the report information of the car damage case.

S20. Identify the report information, and obtain the unique registration information of the vehicle and the
vehicle damage image.

In this embodiment, after receiving the report information of the vehicle damage case, in order to accurately
determine whether there is duplicate insurance application in the insured case, the report information is first
identified, and the unique registration information of the vehicle and the vehicle damage image are obtained.
The unique registration information includes license plate number, frame number, etc. In each report
information, a unique vehicle identification mark is required, that is, the unique registration information of the
vehicle, and the vehicle damage image is the location of the vehicle damage contained in the report information.
image.

S30. Extract the damage feature of the vehicle damage image based on the deep learning network, and convert
the damage feature into a feature vector.

In this embodiment, after identifying the report information, obtaining the unique registration information of the
vehicle and the vehicle damage image, it is determined whether the insurance application is repeated insurance
through the vehicle damage image. Damage features of vehicle damage images, the deep learning network
obtains the image features of the damaged parts of the vehicle, and trains and learns the image features of the
damaged parts of the vehicle and the image features of the undamaged corresponding positions of the vehicle to
obtain the vehicle damage The image features of the parts, including the image gray value, image texture, image
cracks and other features of the damaged parts, so as to learn the image features of different vehicle damaged
parts. After receiving the vehicle damage images, the vehicle can be determined based on the image features
The damage feature corresponding to the damage image is then converted into a feature vector, and different
damage features are represented in the form of vectors, thereby reducing the data storage resource occupation of
the damage feature.

S40. The feature vector is segmented and then encoded to obtain a feature code.

In this embodiment, after extracting the damage feature of the vehicle damage image based on the deep learning
network, and converting the damage feature into a feature vector, in order to more easily represent different
damage features, it is convenient for the storage of damage features and damage features For comparison, the
feature vector is divided and encoded to obtain a feature code. Specifically, the length of the feature vector
directly converted from the damage feature is large, which is inconvenient for storage and direct comparison
between features. The vector is divided into multiple segments, and then the multiple segments are classified
and mapped to compress the information of each segment, so as to quickly compress the feature vector and
obtain the feature code.
S50. Obtain the historical feature code of the historical vehicle damage image of the vehicle corresponding to
the unique registration information.

In this embodiment, after the feature vector is segmented and encoded to obtain the feature code, since the
report information of each vehicle damage case must have the unique registration information of the vehicle,
that is, through the unique registration information of the vehicle The registration information is used to detect
whether there is duplicate insurance application in the insurance application case of the same vehicle.
Specifically, the historical insurance application information of the vehicle corresponding to the unique
registration information is obtained, and then the historical vehicle damage image is obtained from the historical
insurance application information. The historical car damage images are also extracted and encoded based on
the steps S30 and S40, so as to obtain the historical feature codes of the historical car damage images, and the
feature codes are compared, so that whether the current car damage images and the historical car damage images
are the same can be quickly compared. , so as to determine whether the current car damage case is repeated
insurance.

S60. Compare the feature code with the historical feature code, and if the similarity between the feature code
and the historical feature code is higher than a preset threshold, determine that the case report information is
repeated case report information.

In this embodiment, after obtaining the historical feature code of the historical vehicle damage image of the
vehicle corresponding to the unique registration information, the feature code is used for comparison to compare
whether the current vehicle damage image is the same as the historical vehicle damage image, so as to determine
whether the current vehicle damage image is the same as the historical vehicle damage image. Whether the car
damage case is repeated insurance, specifically, compare the feature code with the historical feature code, if the
similarity between the feature code and the historical feature code is higher than a preset threshold, specifically,
Calculate the value of the corresponding bits of the feature code and the historical feature code, determine the
similarity between the feature code and the historical feature code according to the difference in the value, and
determine the similarity when the similarity is higher than a preset threshold. The report information is repeated
report information, so as to quickly and accurately detect whether there is duplicate insurance in the insurance of
the same vehicle damage case, and improve the detection efficiency and accuracy of auto insurance insurance
information.

This embodiment provides a method for detecting whether an insurance application is repeated insured cases
after comparing the features after encoding the characteristics of a vehicle damage image. By developing an
insurance application case detection platform, it is used to detect whether the insurance application in a vehicle
damage case exists. Repeatedly apply for insurance for detection, first receive the report information of the car
damage case, identify the report information, obtain the unique registration information of the vehicle and the
car damage image, and extract the damage characteristics of the car damage image based on the deep learning
network, including the image of the damaged part Image features such as gray value, image texture, image
cracks, etc., based on the image features, the corresponding damage features of the vehicle damage image can be
determined, and then the damage features are converted into feature vectors to represent different damage
features in the form of vectors. , thereby reducing the data storage resource occupation of damage features, in
order to more easily represent different damage features, facilitate the storage of damage features and the
comparison of damage features, the feature vector is segmented and encoded to obtain the feature code, and then
obtain the The historical feature code of the historical vehicle damage image of the vehicle corresponding to the
unique registration information, that is, the historical feature code obtained after feature extraction and encoding
of the historical vehicle damage image in the historical report information, and the feature code and the
historical feature code The codes are compared, the numerical values of the corresponding bits of the
characteristic codes and the historical characteristic codes are calculated, and the similarity between the
characteristic codes and the historical characteristic codes is determined according to the difference of the
numerical values. A threshold is set to determine that the report information is duplicate report information, so
as to quickly and accurately detect whether there is duplicate insurance in the insurance application of the car
damage case of the same vehicle, and improve the detection efficiency and accuracy of auto insurance insurance
information.

In one embodiment, the deep rest network includes a backbone network, a region candidate network, and an
output network; the deep learning-based network extracts damage features of the vehicle damage image,
including:
Inputting the vehicle damage image to the backbone network, and performing convolution and pooling
calculations on the vehicle damage image based on the backbone network to obtain a first intermediate feature
map;

Inputting the first intermediate feature map to the region candidate network, and performing boundary
regression and region-of-interest alignment calculations on the first intermediate feature map based on the
region candidate network to obtain a second intermediate feature map;

Input the second intermediate feature map to the output network, calculate the damage frame and damage type
of the vehicle damage corresponding to the second feature map based on the output network, and determine the
vehicle damage according to the damage frame and damage type Image damage characteristics.

In this embodiment, the architecture of the deep rest network includes three layers of sub-networks, the sub-
network includes a backbone network, a regional candidate network and an output network, wherein the
backbone network is used to extract shallow features, for example, resnet50 is selected as the backbone
network; the region candidate network is used to extract candidate regions, for example, a standard RPN
network is used; the output network is used to fit a specific damage coordinate frame and damage classification,
specifically, the vehicle damage image is input to the The backbone network performs convolution and pooling
calculations on the vehicle damage image based on the backbone network to obtain a first intermediate feature
map, thereby extracting the shallow features of the vehicle damage image, and then inputting the first
intermediate feature map. To the region candidate network, perform boundary regression and region-of-interest
alignment calculation on the first intermediate feature map based on the region candidate network to obtain a
second intermediate feature map, thereby extracting the coordinates of the damage location of the vehicle
damage image, and then The second intermediate feature map is input to the output network, a damage frame
and a damage type corresponding to the vehicle damage of the second feature map are calculated based on the
output network, and the vehicle damage image is determined according to the damage frame and the damage
type The damage features of the damage image are extracted layer by layer based on the three-layer sub-
network, so as to accurately extract the damage features corresponding to the damage image.

In one embodiment, calculating the damage frame and damage type of the vehicle damage corresponding to the
second feature map based on the output network includes:

computing bounding box coordinates of the second feature map based on the output network;

Determine the damage frame of the vehicle damage according to the bounding frame coordinates;

Calculate the degree of loss between the image in the damaged frame and the standard image;

The damage type of the vehicle damage is determined according to the damage degree.

In this embodiment, in the process of calculating the damage frame and damage type of the second feature map
corresponding to the vehicle damage by the output network, first calculate the bounding box coordinates of the
second feature map based on the output network. , in the second feature map, the coordinates of which the
degree of difference between adjacent pixels is greater than the preset value are selected as the bounding box
coordinates, and then the damage frame of the vehicle damage is determined according to the bounding box
coordinates, that is, the damage frame is determined The position of the damage frame and the image
information contained in the position, and then calculate the loss degree of the image in the damage frame and
the standard image, calculate the loss degree of the image in the damage frame and the image in the same
position without damage, and then calculate the loss degree according to the The degree of loss determines the
damage type of the car damage. In order to better detect the fraud of auto insurance repeated insurance, it is
necessary to define the damage type more fully, and in order to define the damage type accurately, to avoid
small damage differences that are easily mistakenly detected as the same damage, This embodiment improves
the definition of damage types, and defines eight damage types: scratches, scratches, mild depressions, moderate
depressions, severe depressions, wrinkles, cracks, and deletions, thereby improving the accuracy of damage type
determination. .

In one embodiment, the encoding of the feature vector after segmentation to obtain the feature encoding
includes:

obtain the length of the feature vector;


Divide the feature vector into several sub-vectors with the same length according to the length;

Randomly select vector values at several positions from the sub-vectors;

Calculate a target value according to the vector value and based on a preset mapping rule, and configure the
target value as a segmented subcoding of the subvector;

The segmented auto-encodings are combined to obtain feature encodings.

In this embodiment, in the process of dividing the feature vector and encoding to obtain the feature code, the
length of the feature vector is first obtained, and then the feature vector is divided into a number of equal parts
according to the length. A sub-vector of length, each sub-vector contains the corresponding value at each
position, that is, each position has a corresponding vector value, and the vector values at several positions are
randomly selected from the sub-vector, according to the vector value The target value is calculated based on the
preset mapping rule, the target value is configured as the segment sub-code of the sub-vector, and the segment
auto-code is combined to obtain the feature code. Specifically, it is assumed that the length of the extracted
feature vector is 512, which is inconvenient for storage and direct comparison between features. First, the
feature vector with a length of 512 is cut into 8 segments on average, and the length of each sub-vector is 64,
that is, each sub-vector has 64-bit value, then for each sub-vector, randomly select 2 digits in 1-64, calculate the
target value according to the vector value and based on the preset mapping rule, and configure the target value
as the sub-vector For example, if the value of the sub-vector in 2 bits is less than 0.5, the sub-vector is mapped
to 0; if only one bit is less than 0.5, the sub-vector is mapped to 1; if the two bits are not less than 0.5 , the sub-
vector is mapped to 2; through this mapping process, the feature vector with the original length of 512 is finally
mapped into a segmented code with a length of 8, thereby improving the storage efficiency and comparison
efficiency of damage features, and reducing damage Features storage resources to improve resource utilization.

In one embodiment, if the similarity between the feature code and the historical feature code is higher than a
preset threshold, determining that the report information is repeated report information, including:

obtaining the segmented subcode of the feature code;

obtaining the historical segment subcoding of the historical feature coding;

When the same number of segment sub-codes as the historical segment sub-codes is greater than a preset value,
it is determined that the similarity between the feature code and the historical feature code is higher than a preset
threshold, and it is determined that the report information is repeated report information.

In this embodiment, in the process of judging that the similarity between the feature code and the historical
feature code is higher than a preset threshold, and determining that the case report information is repeated case
report information, the segmented sub-codes of the feature code are obtained, and then Obtain the historical
segment sub-codes of the historical feature code, and when the same number of the segment sub-codes as the
historical segment sub-codes is greater than a preset value, it is determined that the similarity between the
feature code and the historical feature code is high. At a preset threshold, it is determined that the reporting
information is repeated reporting information, for example, when the same number of segmented sub-codings as
the historical segmented sub-codings is greater than 5, determine the similarity between the feature coding and
the historical feature coding Above the preset threshold, it is determined that the report information is repeated
report information, thereby improving the accuracy of similarity calculation.

In one embodiment, before when the same number of segment sub-codes as the historical segment sub-codes is
greater than a preset value, the method further includes:

Get the detection stage of the current auto insurance coverage;

Configure the number of resources according to the detection phase;

The value of the preset value is determined according to the number of resources.

In this embodiment, before it is determined that the same number of segment sub-codes as the historical
segment sub-codes is greater than a preset value, different preset values need to be determined according to
different scenarios. In the detection stage, the number of resources is configured according to the detection
stage, and then the value of the preset value is determined according to the number of resources. When the same
number of the historical segment sub-codings is greater than or equal to 3, it is considered that the impairments
in the two pictures are the same impairment; when the resources are less, the similarity threshold is increased,
that is, the segment sub-coding is the same as the historical segment sub-coding. When the number is greater
than or equal to 6, the damage in the two images is considered to be the same damage, thereby improving the
efficiency of similarity determination in different scenarios.

In one embodiment, after the feature vector is segmented and encoded to obtain the feature encoding, the
method further includes:

generating block information after associating the report information, the vehicle damage image and the feature
code;

Configuring the unique registration information of the vehicle as block header information;

A block chain is generated according to the block information and the block header information.

In this embodiment, after the feature vector is split and encoded, and after the feature code is obtained, the
feature code needs to be stored confidentially to prevent the feature code from being maliciously modified. After
the car damage image and the feature code are associated, block information is generated, and then the unique
registration information of the vehicle is configured as block header information, and a block chain is generated
according to the block information and the block header information. , the report information, the car damage
image and the feature code are associated and then processed on the chain, so as to avoid malicious modification
of the car damage image and the corresponding feature code in the report information, and ensure the data
security of the car damage image. .

Referring to FIG. 2, the present application also provides a detection device for auto insurance application
information, including:

The information acquisition module 10 is used for receiving report information of vehicle damage cases;

The information identification module 20 is used to identify the report information, and obtain the unique
registration information of the vehicle and the vehicle damage image;

A feature extraction module 30, configured to extract the damage feature of the vehicle damage image based on
the deep learning network, and convert the damage feature into a feature vector;

The vector encoding module 40 is used to encode the feature vector after segmentation to obtain feature
encoding;

The historical case module 50 is used to obtain the historical feature code of the historical vehicle damage image
of the vehicle corresponding to the unique registration information;

Repeat detection module 60 is used to compare the feature code with the historical feature code, and if the
similarity between the feature code and the historical feature code is higher than a preset threshold, determine
that the report information is a repeat report information.

As described above, it can be understood that each component of the device for detecting auto insurance
application information proposed in this application can implement the functions of any one of the above
methods for detecting vehicle insurance application information.

In one embodiment, the deep rest network includes a backbone network, a region candidate network, and an
output network; the deep learning-based network extracts damage features of the vehicle damage image,
including:

Inputting the vehicle damage image to the backbone network, and performing convolution and pooling
calculations on the vehicle damage image based on the backbone network to obtain a first intermediate feature
map;
Inputting the first intermediate feature map to the region candidate network, and performing boundary
regression and region-of-interest alignment calculations on the first intermediate feature map based on the
region candidate network to obtain a second intermediate feature map;

Input the second intermediate feature map to the output network, calculate the damage frame and damage type
of the vehicle damage corresponding to the second feature map based on the output network, and determine the
vehicle damage according to the damage frame and damage type Image damage characteristics.

In one embodiment, calculating the damage frame and damage type of the vehicle damage corresponding to the
second feature map based on the output network includes:

computing bounding box coordinates of the second feature map based on the output network;

Determine the damage frame of the vehicle damage according to the bounding frame coordinates;

Calculate the degree of loss between the image in the damaged frame and the standard image;

The damage type of the vehicle damage is determined according to the damage degree.

In one embodiment, the encoding of the feature vector after segmentation to obtain the feature encoding
includes:

obtain the length of the feature vector;

Divide the feature vector into several sub-vectors with the same length according to the length;

Randomly select vector values at several positions from the sub-vectors;

Calculate a target value according to the vector value and based on a preset mapping rule, and configure the
target value as a segmented subcoding of the subvector;

The segmented auto-encodings are combined to obtain feature encodings.

In one embodiment, if the similarity between the feature code and the historical feature code is higher than a
preset threshold, determining that the report information is repeated report information, including:

obtaining the segmented subcode of the feature code;

obtaining the historical segment subcoding of the historical feature coding;

When the same number of segment sub-codes as the historical segment sub-codes is greater than a preset value,
it is determined that the similarity between the feature code and the historical feature code is higher than a preset
threshold, and it is determined that the report information is repeated report information.

In one embodiment, before when the same number of segment sub-codes as the historical segment sub-codes is
greater than a preset value, the method further includes:

Get the detection stage of the current auto insurance coverage;

Configure the number of resources according to the detection phase;

The value of the preset value is determined according to the number of resources.

In one embodiment, after the feature vector is segmented and encoded to obtain the feature encoding, the
method further includes:

generating block information after associating the report information, the vehicle damage image and the feature
code;

Configuring the unique registration information of the vehicle as block header information;
A block chain is generated according to the block information and the block header information.

Referring to FIG. 3 , an embodiment of the present application further provides a computer device. The
computer device may be a mobile terminal, and its internal structure may be as shown in FIG. 3 . The computer
equipment includes a processor, memory, network interface, and display and input devices connected by a
system bus. Wherein, the network interface of the computer equipment is used to communicate with external
terminals through network connection. The input device of the computer device is used to receive user input.
The computer is designed with a processor to provide computing and control capabilities. The memory of the
computer device includes a storage medium. The storage medium stores an operating system, a computer
program and a database. The database of the computer equipment is used to store data. When the computer
program is executed by the processor, a method for detecting auto insurance application information is realized.

The above processor executes the above method for detecting auto insurance application information, including:
receiving report information of a car damage case; identifying the report information, obtaining unique
registration information of a vehicle and a car damage image; extracting the car damage image based on a deep
learning network and convert the damage feature into a feature vector; code the feature vector after
segmentation to obtain a feature code; obtain the historical feature code of the historical vehicle damage image
of the vehicle corresponding to the unique registration information; The feature code is compared with the
historical feature code, and if the similarity between the feature code and the historical feature code is higher
than a preset threshold, it is determined that the report information is repeated report information.

The computer equipment provides a method for detecting whether an insurance application case is repeated
insurance application after comparing the characteristics after encoding the characteristics of the vehicle damage
image. By developing an insurance application case detection platform, it is used for insurance application of
the vehicle damage case. To detect whether there is duplicate insurance, first receive the report information of
the car damage case, identify the report information, obtain the unique registration information of the vehicle
and the car damage image, and extract the damage characteristics of the car damage image based on the deep
learning network, including the damage part. Based on the image features such as image gray value, image
texture, image crack, etc., the damage feature corresponding to the vehicle damage image can be determined
based on the image feature, and then the damage feature is converted into a feature vector to represent different
damage features, thereby reducing the data storage resource occupancy of damage features, in order to more
easily represent different damage features, facilitate the storage of damage features and the comparison of
damage features, the feature vector is segmented and encoded to obtain feature codes, and then Obtain the
historical feature code of the historical vehicle damage image of the vehicle corresponding to the unique
registration information, that is, the historical feature code obtained after feature extraction and encoding of the
historical vehicle damage image in the historical report information, and compare the feature code with the
Compare the historical feature codes, calculate the value of the corresponding bits of the feature code and the
historical feature code, and determine the similarity between the feature code and the historical feature code
according to the difference in the values. When the similarity is high At a preset threshold, it is determined that
the report information is repeated report information, so as to quickly and accurately detect whether the
insurance application of the same vehicle damage case has repeated insurance coverage, and improve the
detection efficiency and accuracy of auto insurance insurance information.

An embodiment of the present application further provides a computer-readable storage medium on which a
computer program is stored, and when the computer program is executed by the processor, implements a
method for detecting auto insurance application information, including the steps of: receiving a case of vehicle
damage the report information; identify the report information, obtain the unique registration information of the
vehicle and the car damage image; extract the damage feature of the car damage image based on the deep
learning network, and convert the damage feature into a feature vector; convert the feature After the vector is
segmented and encoded, the feature code is obtained; the historical feature code of the historical vehicle damage
image of the vehicle corresponding to the unique registration information is obtained; the feature code is
compared with the historical feature code, if the feature code If the similarity with the historical feature code is
higher than a preset threshold, it is determined that the report information is repeated report information.

The computer-readable storage medium provides a method for detecting whether an insured case is repeatedly
insured after comparing the features based on a vehicle damage image. By developing an insured case detection
platform, it is used to detect vehicle damage. To detect whether there is repeated insurance coverage in the case
of insurance, first receive the report information of the car damage case, identify the report information, obtain
the unique registration information of the vehicle and the car damage image, and extract the damage
characteristics of the car damage image based on the deep learning network. It includes image features such as
image gray value, image texture, and image crack of the damaged part. Based on the image features, the damage
feature corresponding to the vehicle damage image can be determined, and then the damage feature is converted
into a feature vector in the form of a vector. Characterize different damage features, thereby reducing the data
storage resource occupation of damage features. In order to more easily represent different damage features,
facilitate the storage of damage features, and compare damage features, the feature vector is segmented and
encoded to obtain features. encoding, and then obtain the historical feature encoding of the historical vehicle
damage image of the vehicle corresponding to the unique registration information, that is, the historical feature
encoding obtained by the historical vehicle damage image in the historical report information after feature
extraction and encoding, and the feature encoding Compare with the historical feature coding, calculate the
numerical value of the corresponding bit of the feature coding and the historical feature coding, determine the
similarity between the feature coding and the historical feature coding according to the difference of the
numerical value, when the When the similarity is higher than a preset threshold, it is determined that the report
information is repeated report information, so as to quickly and accurately detect whether the insurance
application of the vehicle damage case of the same vehicle has repeated insurance coverage, and improve the
detection efficiency and accuracy of auto insurance insurance information.

Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above
embodiments can be implemented by instructing relevant hardware through a computer program, and the
computer program can be stored in a non-volatile computer-readable storage In the medium, when the computer
program is executed, it may include the processes of the above-mentioned method embodiments.

Any reference to a memory, storage, database, or other medium provided herein and used in the embodiments
may include non-volatile or volatile memory.

Nonvolatile memory may include read only memory (ROM), programmable ROM (PROM), electrically
programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory.
Volatile memory may include random access memory (RAM) or external cache memory. By way of illustration
and not limitation, RAM is available in various forms such as static RAM (SRAM), dynamic RAM (DRAM),
synchronous DRAM (SDRAM), double-rate SDRAM (SSRSDRAM), enhanced SDRAM (ESDRAM),
synchronous Link (Synchlink) DRAM (SLDRAM), memory bus (Rambus) direct RAM (RDRAM), direct
memory bus dynamic RAM (DRDRAM), and memory bus dynamic RAM (RDRAM), etc.

It should be noted that, herein, the terms "comprising", "comprising" or any other variation thereof are intended
to encompass non-exclusive inclusion, such that a process, device, article or method comprising a series of
elements includes not only those elements, It also includes other elements not expressly listed or inherent to
such a process, apparatus, article or method. Without further limitation, an element qualified by the phrase
"comprising a..." does not preclude the presence of additional identical elements in the process, apparatus,
article, or method that includes the element.

The above descriptions are only preferred embodiments of the present application, and are not intended to limit
the patent scope of the present application.

All equivalent structures or equivalent process transformations made by using the contents of the description
and drawings of this application, or directly or indirectly applied in other related technical fields, are similarly
included in the scope of patent protection of this application.

You might also like