Vehicle maintenance scheme judgment method and device, storage medium and computer equipment

Abstract

The invention discloses a vehicle maintenance scheme judgment method and device, a storage medium and computer equipment, relates to the technical field of vehicle loss assessment and finance, and mainly solves the problems of insufficient fine granularity and large deviation of existing maintenance scheme judgment. Carrying out feature extraction processing on the to-be-analyzed vehicle damage picture by adopting a pre-trained feature extractor to obtain corresponding to-be-analyzed vehicle damage feature information; determining a corresponding target damage type based on the to-be-analyzed vehicle damage feature information, and obtaining a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained based on historical maintenance information and corresponding soft label training; carrying out classification processing on the to-be-analyzed vehicle damage feature information by adopting a maintenance mode classifier to obtain a prediction probability value corresponding to each maintenance mode; and obtaining a target maintenance score corresponding to the target damage type, and determining a target maintenance scheme based on the target maintenance score and the prediction probability value.

Description

Technical Field

[0001] The present invention relates to the fields of vehicle damage assessment technology and financial technology, and in particular to a method and device for determining a vehicle repair plan, a storage medium, and a computer device. Background Art

[0002] In the insurance business in the financial field, the vehicle intelligent damage assessment system needs to determine the damage details of the report pictures uploaded by the user and assist professionals to confirm the repair plan. The repair plan judgment in the existing vehicle intelligent damage assessment system is too broad. Taking sheet metal as an example, there are four repair plans: large sheet metal, medium sheet metal, small sheet metal, and replacement. However, affected by various factors such as the on-site environment and vehicle status, the actual vehicle damage varies greatly. Even if the sheet metal is large, the actual repair difficulty is not the same. That is, the current vehicle damage assessment and repair plan judgment based on the report pictures are not granular enough, which often leads to a large difference between the repair plan confirmed by the system and the actual on-site repair plan. Summary of the invention

[0003] In view of this, the present invention provides a method and device for determining a vehicle maintenance plan, a storage medium, and a computer device, the main purpose of which is to solve the problems of insufficient granularity and large deviation in the determination of maintenance plans in the existing vehicle intelligent damage assessment system.

[0004] According to one aspect of the present invention, a method for determining a vehicle maintenance plan is provided, comprising:

[0005] Acquire a vehicle damage picture to be analyzed, and use a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed;

[0006] Determine the corresponding target damage type based on the vehicle damage feature information to be analyzed, and obtain a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained based on historical maintenance information and corresponding soft label training;

[0007] Using the maintenance mode classifier to classify the vehicle damage characteristic information to be analyzed to obtain a prediction probability value corresponding to each maintenance mode;

[0008] A target maintenance score corresponding to the target damage type is acquired, and a target maintenance plan is determined based on the target maintenance score and the predicted probability value.

[0009] Furthermore, before obtaining the maintenance mode classifier corresponding to the target damage type, the method further includes:

[0010] Acquire historical maintenance information corresponding to each damage type, and divide the historical maintenance information into multiple maintenance modes;

[0011] Determining a characteristic center of each maintenance mode, and determining a soft label of each historical maintenance information based on the characteristic center;

[0012] The classifier is trained based on the historical maintenance information and the corresponding soft labels to obtain the maintenance mode classifier corresponding to each damage type.

[0013] Further, the determining of the characteristic center of each maintenance mode includes:

[0014] Acquire historical vehicle damage pictures in the historical maintenance information under each of the maintenance modes, and use the pre-trained feature extractor to perform feature extraction processing on the historical vehicle damage pictures to obtain a plurality of historical vehicle damage features corresponding to each of the maintenance modes;

[0015] Compressing each of the historical vehicle damage features into a vehicle damage feature vector;

[0016] Calculate the mean of the vehicle damage feature vector corresponding to each of the maintenance modes; and determine the mean of the vehicle damage feature vector as the feature center corresponding to each of the maintenance modes.

[0017] Further, the determining of the soft labels of each of the historical maintenance information based on the feature center includes:

[0018] Calculating the similarity between the vehicle damage feature vector and each of the feature centers;

[0019] After sorting the similarities from large to small, determine the target maintenance mode and the nearest neighbor maintenance mode to which the vehicle damage feature vector belongs based on the similarity sorting result;

[0020] Map the similarity of all vehicle damage feature vectors belonging to each maintenance mode to (0.5, 1] ​​to obtain the mapped similarity;

[0021] Obtain a target one-hot encoding of the target maintenance mode and a nearest neighbor one-hot encoding of the nearest neighbor maintenance mode for each vehicle damage feature vector; and determine a soft label for each of the historical maintenance information based on the mapped similarity, the target one-hot encoding, and the nearest neighbor one-hot encoding.

[0022] Furthermore, the training process of the classifier based on the historical maintenance information and the corresponding soft label to obtain the maintenance mode classifier corresponding to each damage type includes:

[0023] Input each vehicle damage feature vector in the historical maintenance information into a classifier to obtain a predicted label distribution; and obtain a soft label of each of the historical maintenance information to form a soft label distribution;

[0024] Calculating the KL divergence between the predicted label distribution and the soft label distribution;

[0025] The KL divergence is used as a loss function to guide the training of the classifier to obtain the maintenance mode classifier.

[0026] Further, the acquiring a target maintenance score corresponding to the target damage type, and determining a target maintenance plan based on the target maintenance score and the predicted probability value, includes:

[0027] Obtaining a preset maintenance score retrieval table, wherein the maintenance score retrieval table stores a correspondence between damage types and maintenance scores;

[0028] Acquire the target maintenance score corresponding to the target damage type from the maintenance score search table;

[0029] Calculating the sum of the products of the target maintenance score and the predicted probability value to obtain a solution score value in percentage form;

[0030] The interval position of the solution score value in the target maintenance score is determined, and the target maintenance solution is determined based on the interval position.

[0031] Furthermore, the method further comprises:

[0032] Obtaining a preset maintenance amount retrieval table, wherein the maintenance amount retrieval table stores a correspondence between damage types and maintenance amount scores;

[0033] Acquire a target repair amount score corresponding to the target damage type from the repair amount search table;

[0034] The sum of the products of the target maintenance amount score and the predicted probability value is calculated to obtain the predicted maintenance amount of the target maintenance solution.

[0035] According to another aspect of the present invention, a vehicle maintenance plan determination device is provided, comprising:

[0036] A feature extraction module is used to obtain the vehicle damage picture to be analyzed, and use a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed;

[0037] A classifier acquisition module, used to determine the corresponding target damage type based on the vehicle damage feature information to be analyzed, and acquire a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained by training based on historical maintenance information and corresponding soft labels;

[0038] A classification processing module, used to classify the vehicle damage characteristic information to be analyzed using the maintenance mode classifier to obtain a prediction probability value corresponding to each maintenance mode;

[0039] The solution determination module is used to obtain a target maintenance score corresponding to the target damage type, and determine a target maintenance solution based on the target maintenance score and the predicted probability value.

[0040] Furthermore, the device further comprises a training processing module, and the training processing module comprises:

[0041] A mode division unit, used for acquiring historical maintenance information corresponding to each damage type, and dividing the historical maintenance information into a plurality of maintenance modes;

[0042] a soft label determination unit, configured to determine a characteristic center of each maintenance mode, and determine a soft label of each historical maintenance information based on the characteristic center;

[0043] The classifier training unit is used to train the classifier based on the historical maintenance information and the corresponding soft label to obtain the maintenance mode classifier corresponding to each damage type.

[0044] Furthermore, the soft label determination unit is further used for:

[0045] Acquire historical vehicle damage pictures in the historical maintenance information under each of the maintenance modes, and use the pre-trained feature extractor to perform feature extraction processing on the historical vehicle damage pictures to obtain a plurality of historical vehicle damage features corresponding to each of the maintenance modes;

[0046] Compressing each of the historical vehicle damage features into a vehicle damage feature vector;

[0047] Calculate the mean of the vehicle damage feature vector corresponding to each of the maintenance modes; and determine the mean of the vehicle damage feature vector as the feature center corresponding to each of the maintenance modes.

[0048] Furthermore, the soft label determination unit is further used for:

[0049] Calculating the similarity between the vehicle damage feature vector and each of the feature centers;

[0050] After sorting the similarities from large to small, determine the target maintenance mode and the nearest neighbor maintenance mode to which the vehicle damage feature vector belongs based on the similarity sorting result;

[0051] Map the similarity of all vehicle damage feature vectors belonging to each maintenance mode to (0.5, 1] ​​to obtain the mapped similarity;

[0052] Obtain a target one-hot encoding of the target maintenance mode and a nearest neighbor one-hot encoding of the nearest neighbor maintenance mode for each vehicle damage feature vector; and determine a soft label for each of the historical maintenance information based on the mapped similarity, the target one-hot encoding, and the nearest neighbor one-hot encoding.

[0053] Furthermore, the classifier training unit is also used for:

[0054] Input each vehicle damage feature vector in the historical maintenance information into a classifier to obtain a predicted label distribution; and obtain a soft label of each of the historical maintenance information to form a soft label distribution;

[0055] Calculating the KL divergence between the predicted label distribution and the soft label distribution;

[0056] The KL divergence is used as a loss function to guide the training of the classifier to obtain the maintenance mode classifier.

[0057] Furthermore, the solution determination module is also used for:

[0058] Obtaining a preset maintenance score retrieval table, wherein the maintenance score retrieval table stores a correspondence between damage types and maintenance scores;

[0059] Acquire the target maintenance score corresponding to the target damage type from the maintenance score search table;

[0060] Calculating the sum of the products of the target maintenance score and the predicted probability value to obtain a solution score value in percentage form;

[0061] The interval position of the solution score value in the target maintenance score is determined, and the target maintenance solution is determined based on the interval position.

[0062] Furthermore, the device also includes a maintenance amount prediction module, which is used to:

[0063] Obtaining a preset maintenance amount retrieval table, wherein the maintenance amount retrieval table stores a correspondence between damage types and maintenance amount scores;

[0064] Acquire a target repair amount score corresponding to the target damage type from the repair amount search table;

[0065] The sum of the products of the target maintenance amount score and the predicted probability value is calculated to obtain the predicted maintenance amount of the target maintenance solution.

[0066] According to another aspect of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, wherein the executable instruction enables a processor to execute operations corresponding to the above-mentioned method for determining a vehicle maintenance plan.

[0067] According to another aspect of the present invention, there is provided a computer device, comprising a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus;

[0068] The memory is used to store at least one executable instruction, and the executable instruction enables the processor to perform operations corresponding to the above-mentioned vehicle maintenance plan determination method.

[0069] By means of the above technical solution, the technical solution provided by the embodiment of the present invention has at least the following advantages:

[0070] The present invention provides a method and device for distinguishing vehicle maintenance plans, a storage medium, and a computer device. Compared with the prior art, the present invention classifies and processes the vehicle damage feature information to be analyzed extracted from the vehicle damage image to be analyzed by a maintenance mode classifier obtained by training based on historical maintenance information and corresponding soft labels, thereby improving the accuracy and granularity of vehicle damage assessment. By obtaining a target maintenance score corresponding to the target damage type and determining a target maintenance plan based on the target maintenance score and the predicted probability value, the required maintenance operation is expressed in percentage form, further improving the granularity of the maintenance plan.

[0071] The above description is only an overview of the technical solution of the present invention. In order to more clearly understand the technical means of the present invention, it can be implemented according to the contents of the specification. In order to make the above and other purposes, features and advantages of the present invention more obvious and easy to understand, the specific implementation methods of the present invention are listed below. BRIEF DESCRIPTION OF THE DRAWINGS

[0072] Various other advantages and benefits will become apparent to those of ordinary skill in the art by reading the detailed description of the preferred embodiments below. The accompanying drawings are only for the purpose of illustrating the preferred embodiments and are not to be considered as limiting the present invention. Moreover, the same reference symbols are used throughout the accompanying drawings to represent the same components. In the accompanying drawings:

[0073] Figure 1 A schematic flow chart of a method for determining a vehicle maintenance plan provided by an embodiment of the present invention is shown;

[0074] Figure 2 A schematic flow chart of another method for determining a vehicle maintenance plan provided by an embodiment of the present invention is shown;

[0075] Figure 3 A schematic flow chart of another method for determining a vehicle maintenance plan provided by an embodiment of the present invention is shown;

[0076] Figure 4A schematic flow chart of another method for determining a vehicle maintenance plan provided by an embodiment of the present invention is shown;

[0077] Figure 5 A schematic diagram showing the structure of a vehicle maintenance plan determination device provided by an embodiment of the present invention;

[0078] Figure 6 A schematic structural diagram of a computer device provided by an embodiment of the present invention is shown. DETAILED DESCRIPTION

[0079] The exemplary embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although the exemplary embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be limited by the embodiments set forth herein. On the contrary, these embodiments are provided to enable a more thorough understanding of the present disclosure and to fully convey the scope of the present disclosure to those skilled in the art.

[0080] The embodiment of the present invention provides a method for determining a vehicle maintenance plan. The method is applicable to insurance business scenarios in the financial field, and is particularly applicable to the vehicle damage assessment link in the vehicle damage insurance business. The specific steps of the method are as follows: Figure 1 As shown, the method includes:

[0081] 101. Obtain a vehicle damage picture to be analyzed, and use a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed;

[0082] In an embodiment of the present invention, the current execution end obtains the vehicle damage picture to be analyzed submitted by the user in the vehicle intelligent damage assessment system, wherein the vehicle damage picture to be analyzed represents the picture of the vehicle damage related information, such as the vehicle paint defect picture, the door dent picture, the front windshield damage picture, etc., which is not specifically limited in the embodiment of the present invention. After obtaining the vehicle damage picture to be analyzed, the current execution end uses a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain the corresponding vehicle damage feature information to be analyzed. Among them, the pre-trained feature extractor represents a type of model or algorithm for extracting image feature information. The purpose of feature extraction is to compress high-dimensional image information into low-dimensional feature vectors for subsequent processing and analysis. YOLO v7 is preferably used to perform feature extraction processing on the vehicle damage picture to be analyzed, which is not specifically limited in the embodiment of the present invention.

[0083] 102. Determine a corresponding target damage type based on the vehicle damage feature information to be analyzed, and obtain a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained by training based on historical maintenance information and corresponding soft labels;

[0084] In an embodiment of the present invention, the current execution end determines the corresponding target damage type based on the vehicle damage feature information to be analyzed extracted in step 101. Specifically, a pre-trained classification model can be used to classify the vehicle damage feature information to be analyzed, such as a pre-trained convolutional neural network model CNN, a deep neural network model AlexNet, YOLO v7, etc., which are not specifically limited in the embodiment of the present invention. Among them, the damage type can be divided into paint damage, metal frame deformation, engine damage, oil circuit damage, etc., which are not specifically limited in the embodiment of the present invention. It should be noted that the above-mentioned damage types are all broad types, and the granularity of vehicle damage assessment is not enough in the specific vehicle damage insurance business. Therefore, further analysis and processing based on the determined target damage type is required. In an embodiment of the present invention, the current execution end obtains a maintenance mode classifier corresponding to the target damage type, wherein the maintenance mode classifier is trained based on historical maintenance information and corresponding soft labels, and is mainly used to determine a finer-grained maintenance mode from the target damage type. For example, when the target damage type is metal frame deformation, a maintenance mode classifier for distinguishing five maintenance modes, namely replacement, large sheet metal, medium sheet metal, small sheet metal, and no maintenance required, can be trained based on historical maintenance information and corresponding soft labels. The embodiment of the present invention does not make specific limitations.

[0085] 103. Using the maintenance mode classifier to classify the vehicle damage characteristic information to be analyzed, and obtaining a prediction probability value corresponding to each maintenance mode;

[0086] In the embodiment of the present invention, the current execution end uses the maintenance mode classifier obtained in step 102 to classify the vehicle damage feature information to be analyzed extracted in step 101 to obtain prediction probability values ​​corresponding to various maintenance modes. For example, when the target damage type is metal frame deformation, the vehicle damage feature information to be analyzed is classified and processed by the maintenance mode classifier, and the prediction probability values ​​corresponding to replacement, large sheet metal, medium sheet metal, small sheet metal, and no maintenance are obtained as [-0.1, 0.2, 0.5, 0.8, 0.0], etc., which are not specifically limited in the embodiment of the present invention.

[0087] 104. Obtain a target maintenance score corresponding to the target damage type, and determine a target maintenance plan based on the target maintenance score and the predicted probability value.

[0088] In an embodiment of the present invention, the current execution end obtains a target maintenance score corresponding to the target damage type, wherein the target maintenance score represents the score value corresponding to each maintenance mode under the target damage type. Taking the deformation of the metal frame in step 103 as an example, the score value corresponding to each maintenance mode under the target damage type can be set to [100, 75, 50, 25, 0], wherein 100 is the score value of the replacement maintenance mode, 75 is the score value of the sheet metal major maintenance mode, 50 is the score value of the sheet metal medium maintenance mode, 25 is the score value of the sheet metal minor maintenance mode, 0 is the score value of no maintenance required, etc., and the embodiment of the present invention does not make specific limitations. It should be noted that the larger the value of the target maintenance score, the more maintenance should be performed. After obtaining the target maintenance score, the current execution end determines the target maintenance plan based on the target maintenance score and the predicted probability value obtained in step 103.

[0089] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to improve the granularity of the maintenance plan determination, another method for determining the vehicle maintenance plan is provided, such as Figure 2 As shown, before the step of acquiring the maintenance mode classifier corresponding to the target damage type, the method further includes:

[0090] 201. Acquire historical maintenance information corresponding to each damage type, and divide the historical maintenance information into multiple maintenance modes;

[0091] In the embodiment of the present invention, the current execution end obtains historical maintenance information corresponding to each damage type, such as obtaining historical maintenance information related to damage type vehicle paint defects, obtaining historical maintenance information related to damage type metal frame deformation, etc., which is not specifically limited in the embodiment of the present invention. In addition, the current execution end also divides the historical maintenance information corresponding to each damage type into multiple maintenance modes, such as dividing the historical maintenance information related to damage type metal frame deformation into five maintenance modes: replacement, large sheet metal, medium sheet metal, small sheet metal, and no maintenance required, etc., which is not specifically limited in the embodiment of the present invention.

[0092] 202. Determine a characteristic center of each maintenance mode, and determine a soft label of each historical maintenance information based on the characteristic center;

[0093] In the embodiment of the present invention, the current execution end determines the characteristic center of each maintenance mode divided in step 201, specifically including:

[0094] (1) Obtaining historical vehicle damage pictures in the historical maintenance information under each of the maintenance modes, and using the pre-trained feature extractor to perform feature extraction processing on the historical vehicle damage pictures to obtain a plurality of historical vehicle damage features corresponding to each of the maintenance modes; wherein the pre-trained feature extractor may preferably be YOLO v7 in step 101 to perform feature extraction processing on the historical vehicle damage pictures, which is not specifically limited in the embodiment of the present invention.

[0095] (2) Compressing each of the historical vehicle damage features into a vehicle damage feature vector; wherein, when compressing the features, Average pooling can be preferably used to downsample the overall feature information by one layer to obtain a one-dimensional vehicle damage feature vector for subsequent processing and analysis.

[0096] (3) Calculate the mean of the vehicle damage feature vector corresponding to each of the maintenance modes; and determine the mean of the vehicle damage feature vector as the feature center corresponding to each of the maintenance modes. For example, when the target damage type is metal frame deformation, calculate the mean of the vehicle damage feature vector under five maintenance modes: replacement, large sheet metal, medium sheet metal, small sheet metal, and no maintenance required, respectively, to obtain five feature centers corresponding to the above five maintenance modes, etc., which are not specifically limited in the embodiments of the present invention.

[0097] In the embodiment of the present invention, after obtaining the characteristic center of each maintenance mode, the current execution end further determines the soft label of each historical maintenance information based on the characteristic center, specifically including:

[0098] (1) Calculate the similarity between the vehicle damage feature vector and each feature center. Specifically, the similarity between each vehicle damage feature vector and all feature centers can be calculated to form a similarity matrix. The similarity can be calculated using cosine similarity (its value range is [-1, 1], and the higher the score, the more similar the two feature vectors are). If there are N vehicle damage feature vectors and five feature centers, the size of the similarity matrix is ​​N*5, etc., which is not specifically limited in the embodiment of the present invention.

[0099] (2) After the similarities are sorted from large to small, the target maintenance mode and the nearest neighbor maintenance mode to which the vehicle damage feature vector belongs are determined based on the similarity sorting result; that is, the two values ​​with the highest and second highest similarities are selected from the similarity sorting result, and the maintenance modes corresponding to these two values ​​are determined as the target maintenance mode and the nearest neighbor maintenance mode to which the vehicle damage feature vector belongs. For example, when the similarities corresponding to the five maintenance modes of replacement, large sheet metal, medium sheet metal, small sheet metal, and no maintenance are [-0.1, 0.2, 0.5, 0.8, 0.0] respectively, the small sheet metal corresponding to the similarity value of 0.8 is determined as the target maintenance mode, and the medium sheet metal corresponding to the similarity value of 0.5 is determined as the nearest neighbor maintenance mode, etc., which is not specifically limited in the embodiment of the present invention.

[0100] (3) Map the similarity of all vehicle damage feature vectors belonging to each maintenance mode to (0.5, 1] ​​to obtain the mapped similarity to ensure that the softening label operation does not affect the attribution of the feature (the maximum probability score is still on the maintenance mode to which the feature originally belongs). The specific mapping formula is as follows:

[0101]

[0102] Among them, b and a refer to the similarity values ​​before and after mapping, respectively. max and b min The maximum and minimum values ​​of the target range (here should be 1 and 0.5+1e -5 ), a max and a min are the maximum and minimum values ​​of all similarity values ​​in the current maintenance mode respectively.

[0103] (4) Obtaining a target one-hot encoding of the target maintenance mode and a nearest neighbor one-hot encoding of the nearest neighbor maintenance mode corresponding to each vehicle damage feature vector; and determining a soft label of each of the historical maintenance information based on the mapped similarity, the target one-hot encoding, and the nearest neighbor one-hot encoding. The specific formula for determining the soft label is as follows:

[0104]

[0105] in, is a soft label, s is the similarity after mapping, y is the target one-hot encoding of the mode to which the vehicle damage feature vector belongs, and γ is the nearest neighbor one-hot encoding of the vehicle damage feature vector. In this way, when a vehicle damage feature vector has a high similarity with the feature center, its soft label is close to the original target one-hot encoding label; otherwise, when a vehicle damage feature vector has a low similarity with the feature center, its soft label is a bimodal probability distribution, and the peak values ​​of both peaks are close to 0.5 (the class to which it belongs is slightly greater than 0.5, and the neighbor class is slightly less than 0.5). It can also be said that the class to which it belongs and the neighbor class almost divide the original one-hot encoding label equally.

[0106] 203. Perform training processing on a classifier based on the historical maintenance information and the corresponding soft label to obtain the maintenance mode classifier corresponding to each of the damage types.

[0107] In an embodiment of the present invention, the current execution end trains the classifier based on historical maintenance information and corresponding soft labels to obtain a maintenance mode classifier corresponding to each damage type. This classifier is only used to fit and predict the maintenance mode, with Kullback-Leibler (KL) divergence as the loss. The specific classifier structure can be consistent with the classification head for predicting the target damage type in step 102. In an embodiment of the present invention, YOLO v7 can be preferably used to predict the maintenance mode. The only difference from step 102 is that the number of neurons in the last layer is changed to the number of maintenance modes, the loss function is changed to KL divergence, the optimizer is adam, and the learning rate is 1e -5 , and the epoch is 300. The maintenance mode classifier trained by the embodiment of the present invention benefits from the soft label, and the classifier can predict the maintenance mode in a more fine-grained manner.

[0108] Further, as a refinement and extension of the specific implementation of the above embodiment, in order to improve the granularity of the maintenance plan determination, another method for determining the vehicle maintenance plan is provided, such as Figure 3 As shown, the step of training a classifier based on the historical maintenance information and the corresponding soft label to obtain the maintenance mode classifier corresponding to each damage type includes:

[0109] 301. Input each vehicle damage feature vector in the historical maintenance information into a classifier to obtain a predicted label distribution; and obtain a soft label of each of the historical maintenance information to form a soft label distribution;

[0110] 302. Calculate the KL divergence between the predicted label distribution and the soft label distribution;

[0111] 303. Use the KL divergence as a loss function to guide the training of the classifier to obtain the maintenance mode classifier.

[0112] In an embodiment of the present invention, the current execution end inputs each vehicle damage feature vector in the historical maintenance information into the classifier to obtain a predicted label distribution, and obtains a soft label of each historical maintenance information to form a soft label distribution; then the KL divergence between the above-mentioned predicted label distribution and the soft label distribution is calculated, and the calculated KL divergence is used as a loss function to guide the training of the classifier. When the training reaches a preset number of iterations epoch, the training is stopped to obtain a maintenance mode classifier.

[0113] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to define the target maintenance plan by score and further reduce the deviation of the maintenance plan determination result, another method for determining the vehicle maintenance plan is provided, such as Figure 4As shown, the step obtains a target maintenance score corresponding to the target damage type, and determines a target maintenance plan based on the target maintenance score and the predicted probability value, including:

[0114] 401. Obtain a preset maintenance score retrieval table, wherein the maintenance score retrieval table stores a correspondence between damage types and maintenance scores;

[0115] 402. Acquire the target maintenance score corresponding to the target damage type from the maintenance score search table;

[0116] 403. Calculate the sum of the products of the target maintenance score and the predicted probability value to obtain a solution score value in percentage form;

[0117] 404. Determine an interval position of the solution score value in the target maintenance score, and determine the target maintenance solution based on the interval position.

[0118] In an embodiment of the present invention, the current execution end obtains a preset maintenance score retrieval table, wherein the maintenance score retrieval table stores the correspondence between each damage type and the maintenance score, such as storing the correspondence between metal frame deformation and the maintenance score [100, 75, 50, 25, 0]. The number of maintenance scores is equal to the number of maintenance modes for the corresponding damage types, that is, the above maintenance scores are 5, and the maintenance modes corresponding to the metal frame deformation are also 5. The embodiment of the present invention does not make specific limitations. After obtaining the target maintenance score, the current execution end calculates the sum of the products of the target maintenance score and the predicted probability value obtained in step 103 to obtain a solution score value in percentage form. For example, when the target damage type is metal frame deformation, the target maintenance score is [100, 75, 50, 25, 0], and the predicted probability value is [0.0, 0.1, 0.4, 0.5, 0.0], the calculated solution score value is 40 = 0.0*100 + 0.1*75 + 0.4*50 + 0.5*25 + 0.0*0, which is not specifically limited in the embodiment of the present invention. Next, determine the interval position of the solution score value in the target maintenance score, and determine the target maintenance solution based on the determined interval position. For example, the above maintenance score [100, 75, 50, 25, 0] can be divided into five intervals: [100, 87.5), [87.5, 62.5), [62.5, 47.5), [47.5, 12.5), and [12.5, 0], which respectively correspond to five maintenance modes: replacement, large sheet metal, medium sheet metal, small sheet metal, and no maintenance required. The solution score value 40 obtained by the above calculation is in the middle of the interval [47.5, 12.5), so the small sheet metal corresponding to the interval [47.5, 12.5) is determined as the target maintenance solution, etc. This embodiment of the present invention does not make any specific limitation.

[0119] It should be noted that in the embodiment of the present invention, a finer-grained division can be set. For example, the degree of sheet metal is no longer defined as large, medium, small, etc., but defined as a score, such as sheet metal 82, sheet metal 40, sheet metal 15, etc. The embodiment of the present invention does not make specific limitations.

[0120] Furthermore, as a refinement and extension of the specific implementation of the above embodiment, in order to set a fine-grained maintenance amount prediction, another method for determining a vehicle maintenance plan is provided, and the method further includes:

[0121] Obtaining a preset maintenance amount retrieval table, wherein the maintenance amount retrieval table stores a correspondence between damage types and maintenance amount scores;

[0122] Acquire a target repair amount score corresponding to the target damage type from the repair amount search table;

[0123] The sum of the products of the target maintenance amount score and the predicted probability value is calculated to obtain the predicted maintenance amount of the target maintenance solution.

[0124] In an embodiment of the present invention, the current execution end obtains a preset maintenance amount retrieval table, wherein the maintenance amount retrieval table stores the correspondence between each damage type and the maintenance amount score, such as storing the correspondence between metal frame deformation and the maintenance amount score [10000, 7500, 5000, 2500, 0]. The number of maintenance amount scores is equal to the number of maintenance modes for the corresponding damage types, that is, the number of the above maintenance amount scores is 5, and the maintenance modes corresponding to the metal frame deformation are also 5. The embodiment of the present invention does not make specific limitations. After obtaining the target maintenance amount score, the current execution end calculates the sum of the products of the target maintenance amount score and the predicted probability value obtained in step 103 to obtain the predicted maintenance amount of the target maintenance plan. For example, when the target damage type is metal frame deformation, the target maintenance amount score is [10000, 7500, 5000, 2500, 0], and the predicted probability value is [0.0, 0.1, 0.4, 0.5, 0.0], the calculated predicted maintenance amount of the target maintenance plan is 4000=0.0*10000+0.1*7500+0.4*5000+0.5*2500+0.0*0, which is not specifically limited in the embodiment of the present invention.

[0125] The embodiment of the present invention provides a method for distinguishing vehicle maintenance plans. Compared with the prior art, the present invention classifies and processes the vehicle damage feature information to be analyzed extracted from the vehicle damage image to be analyzed by a maintenance mode classifier obtained by training based on historical maintenance information and corresponding soft labels, thereby improving the accuracy and granularity of vehicle damage assessment. By obtaining a target maintenance score corresponding to the target damage type and determining a target maintenance plan based on the target maintenance score and the predicted probability value, the required maintenance operation is expressed in percentage form, further improving the granularity of the maintenance plan.

[0126] As the above Figure 1 The embodiment of the present invention provides a vehicle maintenance plan identification device, such as Figure 5 As shown, the device comprises:

[0127] The feature extraction module 51 is used to obtain the vehicle damage picture to be analyzed, and use the pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed;

[0128] A classifier acquisition module 52 is used to determine the corresponding target damage type based on the vehicle damage feature information to be analyzed, and to acquire a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained by training based on historical maintenance information and corresponding soft labels;

[0129] A classification processing module 53, used to classify the vehicle damage characteristic information to be analyzed using the maintenance mode classifier to obtain a prediction probability value corresponding to each maintenance mode;

[0130] The solution determination module 54 is used to obtain a target maintenance score corresponding to the target damage type, and determine a target maintenance solution based on the target maintenance score and the predicted probability value.

[0131] Furthermore, the device further comprises a training processing module, and the training processing module comprises:

[0132] A mode division unit, used for acquiring historical maintenance information corresponding to each damage type, and dividing the historical maintenance information into a plurality of maintenance modes;

[0133] a soft label determination unit, configured to determine a characteristic center of each maintenance mode, and determine a soft label of each historical maintenance information based on the characteristic center;

[0134] The classifier training unit is used to train the classifier based on the historical maintenance information and the corresponding soft label to obtain the maintenance mode classifier corresponding to each damage type.

[0135] Furthermore, the soft label determination unit is further used for:

[0136] Acquire historical vehicle damage pictures in the historical maintenance information under each of the maintenance modes, and use the pre-trained feature extractor to perform feature extraction processing on the historical vehicle damage pictures to obtain a plurality of historical vehicle damage features corresponding to each of the maintenance modes;

[0137] Compressing each of the historical vehicle damage features into a vehicle damage feature vector;

[0138] Calculate the mean of the vehicle damage feature vector corresponding to each of the maintenance modes; and determine the mean of the vehicle damage feature vector as the feature center corresponding to each of the maintenance modes.

[0139] Furthermore, the soft label determination unit is further used for:

[0140] Calculating the similarity between the vehicle damage feature vector and each of the feature centers;

[0141] After sorting the similarities from large to small, determine the target maintenance mode and the nearest neighbor maintenance mode to which the vehicle damage feature vector belongs based on the similarity sorting result;

[0142] Map the similarity of all vehicle damage feature vectors belonging to each maintenance mode to (0.5, 1] ​​to obtain the mapped similarity;

[0143] Obtain a target one-hot encoding of the target maintenance mode and a nearest neighbor one-hot encoding of the nearest neighbor maintenance mode for each vehicle damage feature vector; and determine a soft label for each of the historical maintenance information based on the mapped similarity, the target one-hot encoding, and the nearest neighbor one-hot encoding.

[0144] Furthermore, the classifier training unit is also used for:

[0145] Input each vehicle damage feature vector in the historical maintenance information into a classifier to obtain a predicted label distribution; and obtain a soft label of each of the historical maintenance information to form a soft label distribution;

[0146] Calculating the KL divergence between the predicted label distribution and the soft label distribution;

[0147] The KL divergence is used as a loss function to guide the training of the classifier to obtain the maintenance mode classifier.

[0148] Furthermore, the solution determination module 54 is also used for:

[0149] Obtaining a preset maintenance score retrieval table, wherein the maintenance score retrieval table stores a correspondence between damage types and maintenance scores;

[0150] Acquire the target maintenance score corresponding to the target damage type from the maintenance score search table;

[0151] Calculating the sum of the products of the target maintenance score and the predicted probability value to obtain a solution score value in percentage form;

[0152] The interval position of the solution score value in the target maintenance score is determined, and the target maintenance solution is determined based on the interval position.

[0153] Furthermore, the device also includes a maintenance amount prediction module, which is used to:

[0154] Obtaining a preset maintenance amount retrieval table, wherein the maintenance amount retrieval table stores a correspondence between damage types and maintenance amount scores;

[0155] Acquire a target repair amount score corresponding to the target damage type from the repair amount search table;

[0156] The sum of the products of the target maintenance amount score and the predicted probability value is calculated to obtain the predicted maintenance amount of the target maintenance solution.

[0157] The embodiment of the present invention provides a vehicle maintenance plan identification device. Compared with the prior art, the present invention classifies and processes the vehicle damage feature information to be analyzed extracted from the vehicle damage image to be analyzed by a maintenance mode classifier obtained by training based on historical maintenance information and corresponding soft labels, thereby improving the accuracy and granularity of vehicle damage assessment. By obtaining a target maintenance score corresponding to the target damage type and determining a target maintenance plan based on the target maintenance score and the predicted probability value, the required maintenance operation is expressed in percentage form, further improving the granularity of the maintenance plan.

[0158] According to one embodiment of the present invention, a storage medium is provided, wherein the storage medium stores at least one executable instruction, and the computer executable instruction can execute the vehicle maintenance plan determination method in any of the above method embodiments.

[0159] Figure 6 A schematic diagram of the structure of a computer device provided according to an embodiment of the present invention is shown. The specific embodiment of the present invention does not limit the specific implementation of the computer device.

[0160] like Figure 6 As shown, the computer device may include: a processor (processor) 602 , a communication interface (Communications Interface) 604 , a memory (memory) 606 , and a communication bus 608 .

[0161] The processor 602 , the communication interface 604 , and the memory 606 communicate with each other via a communication bus 608 .

[0162] The communication interface 604 is used to communicate with other devices such as clients or other servers.

[0163] The processor 602 is used to execute the program 610, and specifically can execute the relevant steps of the above-mentioned vehicle maintenance plan determination method.

[0164] Specifically, the program 610 may include program codes, which include computer operation instructions.

[0165] The processor 602 may be a central processing unit (CPU), or an application-specific integrated circuit (ASIC), or one or more integrated circuits configured to implement the embodiments of the present invention. The one or more processors included in the computer device may be processors of the same type, such as one or more CPUs; or processors of different types, such as one or more CPUs and one or more ASICs.

[0166] The memory 606 is used to store the program 610. The memory 606 may include a high-speed RAM memory, and may also include a non-volatile memory (non-volatile memory), such as at least one disk memory.

[0167] The program 610 may be specifically configured to enable the processor 602 to perform the following operations:

[0168] Acquire a vehicle damage picture to be analyzed, and use a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed;

[0169] Determine the corresponding target damage type based on the vehicle damage feature information to be analyzed, and obtain a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained based on historical maintenance information and corresponding soft label training;

[0170] Using the maintenance mode classifier to classify the vehicle damage characteristic information to be analyzed to obtain a prediction probability value corresponding to each maintenance mode;

[0171] A target maintenance score corresponding to the target damage type is acquired, and a target maintenance plan is determined based on the target maintenance score and the predicted probability value.

[0172] Obviously, those skilled in the art should understand that the above modules or steps of the present invention can be implemented by a general computing device, they can be concentrated on a single computing device, or distributed on a network composed of multiple computing devices, and optionally, they can be implemented by a program code executable by a computing device, so that they can be stored in a storage device and executed by the computing device, and in some cases, the steps shown or described can be executed in a different order than here, or they can be made into individual integrated circuit modules, or multiple modules or steps therein can be made into a single integrated circuit module for implementation. Thus, the present invention is not limited to any specific combination of hardware and software.

[0173] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. For those skilled in the art, the present invention may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention. The non-Company software tools or components appearing in the embodiments of the present invention are only examples and do not represent actual use.

Claims

1. A method for determining a vehicle maintenance plan, characterized in that: include: Acquire a vehicle damage picture to be analyzed, and use a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed; Determine the corresponding target damage type based on the vehicle damage feature information to be analyzed, and obtain a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained based on historical maintenance information and corresponding soft label training; Using the maintenance mode classifier to classify the vehicle damage characteristic information to be analyzed to obtain a prediction probability value corresponding to each maintenance mode; A target maintenance score corresponding to the target damage type is acquired, and a target maintenance plan is determined based on the target maintenance score and the predicted probability value.

2. The method according to claim 1, characterized in that Before acquiring the maintenance mode classifier corresponding to the target damage type, the method further includes: Acquire historical maintenance information corresponding to each damage type, and divide the historical maintenance information into multiple maintenance modes; Determining a characteristic center of each maintenance mode, and determining a soft label of each historical maintenance information based on the characteristic center; The classifier is trained based on the historical maintenance information and the corresponding soft labels to obtain the maintenance mode classifier corresponding to each damage type.

3. The method according to claim 2, characterized in that The determining of the characteristic center of each maintenance mode comprises: Acquire historical vehicle damage pictures in the historical maintenance information under each of the maintenance modes, and use the pre-trained feature extractor to perform feature extraction processing on the historical vehicle damage pictures to obtain a plurality of historical vehicle damage features corresponding to each of the maintenance modes; Compressing each of the historical vehicle damage features into a vehicle damage feature vector; Calculate the mean of the vehicle damage feature vector corresponding to each of the maintenance modes; and determine the mean of the vehicle damage feature vector as the feature center corresponding to each of the maintenance modes.

4. The method according to claim 2, characterized in that: The step of determining the soft labels of each of the historical maintenance information based on the feature center includes: Calculating the similarity between the vehicle damage feature vector and each of the feature centers; After sorting the similarities from large to small, determine the target maintenance mode and the nearest neighbor maintenance mode to which the vehicle damage feature vector belongs based on the similarity sorting result; Map the similarity of all vehicle damage feature vectors belonging to each maintenance mode to (0.5, 1] ​​to obtain the mapped similarity; Obtain a target one-hot encoding of the target maintenance mode and a nearest neighbor one-hot encoding of the nearest neighbor maintenance mode for each vehicle damage feature vector; and determine a soft label for each of the historical maintenance information based on the mapped similarity, the target one-hot encoding, and the nearest neighbor one-hot encoding.

5. The method according to claim 2, characterized in that: The training process of the classifier based on the historical maintenance information and the corresponding soft label to obtain the maintenance mode classifier corresponding to each damage type includes: Inputting each vehicle damage feature vector in the historical maintenance information into a classifier to obtain a predicted label distribution; and obtaining a soft label of each of the historical maintenance information to form a soft label distribution; Calculating the KL divergence between the predicted label distribution and the soft label distribution; The KL divergence is used as a loss function to guide the training of the classifier to obtain the maintenance mode classifier.

6. The method according to claim 1, characterized in that The acquiring a target maintenance score corresponding to the target damage type, and determining a target maintenance plan based on the target maintenance score and the predicted probability value, comprises: Obtaining a preset maintenance score retrieval table, wherein the maintenance score retrieval table stores a correspondence between damage types and maintenance scores; Acquire the target maintenance score corresponding to the target damage type from the maintenance score search table; Calculating the sum of the products of the target maintenance score and the predicted probability value to obtain a solution score value in percentage form; The interval position of the solution score value in the target maintenance score is determined, and the target maintenance solution is determined based on the interval position.

7. The method according to any one of claims 1 to 6, characterized in that: The method further comprises: Obtaining a preset maintenance amount retrieval table, wherein the maintenance amount retrieval table stores a correspondence between damage types and maintenance amount scores; Acquire a target repair amount score corresponding to the target damage type from the repair amount search table; The sum of the products of the target maintenance amount score and the predicted probability value is calculated to obtain the predicted maintenance amount of the target maintenance solution.

8. A vehicle maintenance plan determination device, characterized in that: include: A feature extraction module is used to obtain the vehicle damage picture to be analyzed, and use a pre-trained feature extractor to perform feature extraction processing on the vehicle damage picture to be analyzed to obtain corresponding vehicle damage feature information to be analyzed; A classifier acquisition module, used to determine the corresponding target damage type based on the vehicle damage feature information to be analyzed, and acquire a maintenance mode classifier corresponding to the target damage type; the maintenance mode classifier is obtained by training based on historical maintenance information and corresponding soft labels; A classification processing module, used to classify the vehicle damage characteristic information to be analyzed using the maintenance mode classifier to obtain a prediction probability value corresponding to each maintenance mode; The solution determination module is used to obtain a target maintenance score corresponding to the target damage type, and determine a target maintenance solution based on the target maintenance score and the predicted probability value.

9. A storage medium, wherein at least one executable instruction is stored in the storage medium, and the executable instruction executes an operation corresponding to the method for determining a vehicle maintenance plan according to any one of claims 1 to 7.

10. A computer device, comprising a processor, a memory, a communication interface and a communication bus, wherein the processor, the memory and the communication interface communicate with each other via the communication bus; The memory is used to store at least one executable instruction, and the executable instruction enables the processor to execute an operation corresponding to the vehicle maintenance plan determination method according to any one of claims 1 to 7.

Images