Vehicle defect detection method and system based on neural network

Abstract

The application discloses a vehicle defect detection method and system based on a neural network, and the method comprises the following steps: acquiring a plurality of vehicle data of a vehicle to be identified; determining a defect detection result corresponding to each vehicle data according to a neural network corresponding to the data type of each vehicle data; predicting a future predicted defect of the vehicle to be identified according to all the vehicle data and the corresponding defect detection result; and determining a vehicle defect of the vehicle to be identified based on a defect correlation judgment rule according to the defect detection result and the future predicted defect. It can be seen that the application can improve the accuracy and efficiency of vehicle defect identification based on the prediction results of multiple neural networks and joint judgment, provide accurate data reference for vehicle production or maintenance, and improve the safety of the vehicle.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a vehicle defect detection method and system based on neural networks. Background Technology

[0002] With the development of intelligent identification algorithms, more and more vehicle manufacturers and repair shops are adopting more intelligent and automated defect detection technologies. Given the diverse range of components and corresponding defect types within vehicles, accurately identifying vehicle defects has become a crucial technical challenge. Current technologies for vehicle defect identification largely rely on single data points and pre-defined algorithm models, neglecting the joint prediction and judgment of multiple neural network algorithms. This results in deficiencies in both accuracy and efficiency, compromising vehicle safety. Therefore, existing technologies have shortcomings that urgently need to be addressed. Summary of the Invention

[0003] The technical problem to be solved by the present invention is to provide a vehicle defect detection method and system based on neural networks, which can improve the accuracy and efficiency of vehicle defect identification based on the prediction results and joint judgment of multiple neural networks, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0004] To address the aforementioned technical problems, the first aspect of this invention discloses a vehicle defect detection method based on a neural network, the method comprising:

[0005] Acquire data from multiple vehicles to be identified;

[0006] Based on the neural network corresponding to the data type of each vehicle data, determine the defect detection result corresponding to each vehicle data.

[0007] Based on all the vehicle data and the corresponding defect detection results, predict the future defects of the vehicle to be identified.

[0008] Based on the defect detection results and the predicted future defects, the vehicle defects of the vehicle to be identified are determined according to the defect association judgment rules.

[0009] As an optional implementation, in the first aspect of the present invention, the data type of the vehicle data is in-vehicle video recording data, historical driving operation data, historical sensor recording data, and real-time vehicle test data; the historical sensor recording data is obtained by sensors installed on the vehicle to be identified operating during historical time periods.

[0010] As an optional implementation, in the first aspect of the present invention, determining the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data includes:

[0011] For each type of vehicle data, a predictive neural network corresponding to the data type is determined from a preset algorithm model library; the predictive neural network is trained using a training dataset that includes multiple training vehicle data corresponding to the data type of the vehicle data and corresponding defect annotations.

[0012] The vehicle data is input into the predictive neural network to obtain the defect detection result corresponding to the vehicle data; the defect detection result includes the defect location, defect type and defect occurrence probability.

[0013] As an optional implementation, in the first aspect of the present invention, when the data type of the vehicle data is in-vehicle video recording data, historical sensor recording data, or real-time vehicle test data, the network type of the predictive neural network is an RNN neural network; when the data type of the vehicle data is historical driving operation data, the network type of the predictive neural network is a composite network; the composite network includes an operation filtering network and an operation prediction network; the operation filtering network includes multiple classifier algorithm models, an LSTM neural network, and an output judgment model, the classifier algorithm models are used to predict the hazard level of each driving operation in the historical driving operation data, the LSTM neural network is used to predict the next operation of any two consecutive driving operations with a hazard level greater than a first hazard threshold, and compare the operation similarity between the predicted next operation and the actual next driving operation; the output judgment model is used to determine all corresponding driving operations with a hazard level greater than a second hazard threshold or an operation similarity greater than a first similarity threshold as dangerous driving operations; the operation prediction network is used to predict the corresponding defect detection result based on the dangerous driving operation.

[0014] As an optional implementation, in the first aspect of the invention, the RNN neural network or the operation prediction network includes a feature extraction network and a classification network; each of the RNN neural network or the operation prediction network is jointly trained through the following steps:

[0015] Multiple training datasets containing training vehicle data of different data types are used as a unified dataset. The unified feature extraction network and the unified classification network are trained on the unified dataset until convergence, resulting in the trained unified feature extraction network and unified classification network.

[0016] For each data type, obtain multiple historical data corresponding to that data type; the historical data is the training vehicle data or the vehicle data obtained at historical time points;

[0017] Based on the complexity algorithm, calculate the data complexity corresponding to each of the historical data;

[0018] Calculate the average of the data complexity corresponding to all the historical data to obtain the type complexity corresponding to the data type;

[0019] Based on the pre-defined correspondence between complexity and distillation parameters, the distillation parameters corresponding to the type complexity of the data type are determined; the distillation parameters include at least one of distillation step size, distillation dataset size, distillation loss function parameters, and distillation model architecture parameters.

[0020] Based on the distillation parameters, the trained unified feature extraction network and unified classification network are distilled to obtain the neural network basic model corresponding to this data type.

[0021] The neural network base model obtained by distillation is trained in a targeted manner based on the training dataset corresponding to the data type until convergence, so as to obtain the trained RNN neural network or the operation prediction network corresponding to the data type.

[0022] As an optional implementation, in the first aspect of the invention, predicting future defects of the vehicle to be identified based on all the vehicle data and the corresponding defect detection results includes:

[0023] Randomly sample all the vehicle data to obtain multiple sampling sets;

[0024] For each of the sampling sets, all the vehicle data in the sampling set are sorted from earliest to latest based on the acquisition time to obtain the vehicle data sequence corresponding to the sampling set;

[0025] The vehicle data sequence and the defect detection result corresponding to each vehicle data are determined as the data sequence to be predicted;

[0026] The data sequence to be predicted is input into a trained LSTM neural network to obtain the future predicted defects corresponding to the sampling set; the LSTM neural network is trained using a training dataset that includes multiple training vehicle sensor data sequences and corresponding historical defect detection result annotations and future defect detection result annotations; the future predicted defects include defect location, defect type and defect occurrence probability.

[0027] As an optional implementation, in the first aspect of the present invention, determining the vehicle defect of the vehicle to be identified based on the defect detection result and the predicted future defect, according to a defect association judgment rule, includes:

[0028] Based on association rules, the defect association corresponding to each of the predicted future defects is determined.

[0029] Filter out all the predicted future defects whose correlation with the defect is greater than the correlation threshold to obtain at least one high-probability future defect;

[0030] The objective function is set to maximize the number of defect results included in the defect set; the defect results are the defect detection results or the high-probability future defects.

[0031] The specified limitations include:

[0032] The defect set includes at least one of the high-probability future defects;

[0033] The distance between the defect locations corresponding to any two defect results in the defect set is less than a distance threshold.

[0034] The area of ​​the minimum envelope shape corresponding to the location of the defect location for all the aforementioned defect results in the defect set is less than the area threshold.

[0035] The average probability of occurrence of all the aforementioned defects in the defect set is greater than the probability threshold.

[0036] Based on the dynamic programming algorithm, the optimal defect set is obtained by iterative calculation according to the objective function and the limiting conditions, and the set of defects of concern corresponding to the vehicle to be identified is obtained.

[0037] The set of defects of concern is determined as the vehicle defects of the vehicle to be identified.

[0038] As an optional implementation, in the first aspect of the invention, determining the defect correlation corresponding to each of the future predicted defects based on association rules includes:

[0039] For each of the predicted future defects, calculate the first occurrence count of the defect type of the predicted future defect in all the defect detection results;

[0040] Calculate the second occurrence number of the defect location of the predicted future defect in all the defect detection results;

[0041] Calculate a first weight that is proportional to the first occurrence frequency;

[0042] Calculate the second weight that is proportional to the second occurrence frequency;

[0043] The defect correlation corresponding to the predicted defect is obtained by calculating the product of the first weight, the second weight, and the probability of occurrence of the predicted defect.

[0044] A second aspect of this invention discloses a vehicle defect detection system based on a neural network, the system comprising:

[0045] The acquisition module is used to acquire multiple vehicle data of the vehicle to be identified;

[0046] The first prediction module is used to determine the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data.

[0047] The second prediction module is used to predict the future defects of the vehicle to be identified based on all the vehicle data and the corresponding defect detection results.

[0048] The determination module is used to determine the vehicle defects of the vehicle to be identified based on the defect detection results and the predicted future defects, and according to the defect association judgment rules.

[0049] As an optional implementation, in a second aspect of the present invention, the data type of the vehicle data is in-vehicle video recording data, historical driving operation data, historical sensor recording data, and real-time vehicle test data; the historical sensor recording data is obtained by sensors installed on the vehicle to be identified operating during historical time periods.

[0050] As an optional implementation, in a second aspect of the invention, the first prediction module determines the specific method by which it determines the defect detection result corresponding to each type of vehicle data based on a neural network corresponding to the data type of each vehicle data, including:

[0051] For each type of vehicle data, a predictive neural network corresponding to the data type is determined from a preset algorithm model library; the predictive neural network is trained using a training dataset that includes multiple training vehicle data corresponding to the data type of the vehicle data and corresponding defect annotations.

[0052] The vehicle data is input into the predictive neural network to obtain the defect detection result corresponding to the vehicle data; the defect detection result includes the defect location, defect type and defect occurrence probability.

[0053] As an optional implementation, in the second aspect of the present invention, when the data type of the vehicle data is in-vehicle video recording data, historical sensor recording data, or real-time vehicle test data, the network type of the predictive neural network is an RNN neural network; when the data type of the vehicle data is historical driving operation data, the network type of the predictive neural network is a composite network; the composite network includes an operation filtering network and an operation prediction network; the operation filtering network includes multiple classifier algorithm models, an LSTM neural network, and an output judgment model, the classifier algorithm models are used to predict the hazard level of each driving operation in the historical driving operation data, the LSTM neural network is used to predict the next operation of any two consecutive driving operations with a hazard level greater than a first hazard threshold, and compare the operation similarity between the predicted next operation and the actual next driving operation; the output judgment model is used to determine all corresponding driving operations with a hazard level greater than a second hazard threshold or an operation similarity greater than a first similarity threshold as dangerous driving operations; the operation prediction network is used to predict the corresponding defect detection result based on the dangerous driving operation.

[0054] As an optional implementation, in a second aspect of the invention, the RNN neural network or the operation prediction network includes a feature extraction network and a classification network; each of the RNN neural network or the operation prediction network is jointly trained through the following steps:

[0055] Multiple training datasets containing training vehicle data of different data types are used as a unified dataset. The unified feature extraction network and the unified classification network are trained on the unified dataset until convergence, resulting in the trained unified feature extraction network and unified classification network.

[0056] For each data type, obtain multiple historical data corresponding to that data type; the historical data is the training vehicle data or the vehicle data obtained at historical time points;

[0057] Based on the complexity algorithm, calculate the data complexity corresponding to each of the historical data;

[0058] Calculate the average of the data complexity corresponding to all the historical data to obtain the type complexity corresponding to the data type;

[0059] Based on the pre-defined correspondence between complexity and distillation parameters, the distillation parameters corresponding to the type complexity of the data type are determined; the distillation parameters include at least one of distillation step size, distillation dataset size, distillation loss function parameters, and distillation model architecture parameters.

[0060] Based on the distillation parameters, the trained unified feature extraction network and unified classification network are distilled to obtain the neural network basic model corresponding to this data type.

[0061] The neural network base model obtained by distillation is trained in a targeted manner based on the training dataset corresponding to the data type until convergence, so as to obtain the trained RNN neural network or the operation prediction network corresponding to the data type.

[0062] As an optional implementation, in a second aspect of the invention, the second prediction module predicts future defects of the vehicle to be identified based on all the vehicle data and the corresponding defect detection results, including:

[0063] Randomly sample all the vehicle data to obtain multiple sampling sets;

[0064] For each of the sampling sets, all the vehicle data in the sampling set are sorted from earliest to latest based on the acquisition time to obtain the vehicle data sequence corresponding to the sampling set;

[0065] The vehicle data sequence and the defect detection result corresponding to each vehicle data are determined as the data sequence to be predicted;

[0066] The data sequence to be predicted is input into a trained LSTM neural network to obtain the future predicted defects corresponding to the sampling set; the LSTM neural network is trained using a training dataset that includes multiple training vehicle sensor data sequences and corresponding historical defect detection result annotations and future defect detection result annotations; the future predicted defects include defect location, defect type and defect occurrence probability.

[0067] As an optional implementation, in a second aspect of the invention, the determining module determines the specific method by which it determines the vehicle defect of the vehicle to be identified based on the defect detection result and the predicted future defect, according to a defect association judgment rule, including:

[0068] Based on association rules, the defect association corresponding to each of the predicted future defects is determined.

[0069] Filter out all the predicted future defects whose correlation with the defect is greater than the correlation threshold to obtain at least one high-probability future defect;

[0070] The objective function is set to maximize the number of defect results included in the defect set; the defect results are the defect detection results or the high-probability future defects.

[0071] The specified conditions include:

[0072] The defect set includes at least one of the high-probability future defects;

[0073] The distance between the defect locations corresponding to any two defect results in the defect set is less than a distance threshold.

[0074] The area of ​​the minimum envelope shape corresponding to the location of the defect location for all the aforementioned defect results in the defect set is less than the area threshold.

[0075] The average probability of occurrence of all the aforementioned defects in the defect set is greater than the probability threshold.

[0076] Based on the dynamic programming algorithm, the optimal defect set is obtained by iterative calculation according to the objective function and the limiting conditions, and the set of defects of concern corresponding to the vehicle to be identified is obtained.

[0077] The set of defects of concern is determined as the vehicle defects of the vehicle to be identified.

[0078] As an optional implementation, in a second aspect of the invention, the determining module determines the specific method by which it determines the defect correlation corresponding to each of the future predicted defects based on association rules, including:

[0079] For each of the predicted future defects, calculate the first occurrence count of the defect type of the predicted future defect in all the defect detection results;

[0080] Calculate the second occurrence number of the defect location of the predicted future defect in all the defect detection results;

[0081] Calculate a first weight that is proportional to the first occurrence frequency;

[0082] Calculate the second weight that is proportional to the second occurrence frequency;

[0083] The defect correlation corresponding to the predicted defect is obtained by calculating the product of the first weight, the second weight, and the probability of occurrence of the predicted defect.

[0084] A third aspect of this invention discloses another vehicle defect detection system based on a neural network, the system comprising:

[0085] Memory containing executable program code;

[0086] A processor coupled to the memory;

[0087] The processor calls the executable program code stored in the memory to execute some or all of the steps in the neural network-based vehicle defect detection method disclosed in the first aspect of the present invention.

[0088] The fourth aspect of the present invention discloses a computer storage medium storing computer instructions, which, when invoked, are used to execute some or all of the steps in the neural network-based vehicle defect detection method disclosed in the first aspect of the present invention.

[0089] Compared with the prior art, the embodiments of the present invention have the following beneficial effects:

[0090] This invention can determine the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data, and then predict the future defects of the vehicle based on the vehicle data and the corresponding defect detection result. Based on the defect association judgment rule, the vehicle defects are comprehensively determined. Thus, the accuracy and efficiency of vehicle defect identification can be improved based on the prediction results and joint judgment of multiple neural networks, providing accurate data reference for vehicle production or maintenance and improving vehicle safety. Attached Figure Description

[0091] To more clearly illustrate the technical solutions in the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the accompanying drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0092] Figure 1 This is a schematic flowchart of a vehicle defect detection method based on a neural network disclosed in an embodiment of the present invention.

[0093] Figure 2 This is a schematic diagram of a vehicle defect detection system based on a neural network disclosed in an embodiment of the present invention.

[0094] Figure 3 This is a schematic diagram of another vehicle defect detection system based on a neural network disclosed in an embodiment of the present invention. Detailed Implementation

[0095] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0096] The terms "first," "second," etc., used in the specification, claims, and accompanying drawings of this invention are used to distinguish different objects, not to describe a specific order. Furthermore, the terms "comprising" and "having," and any variations thereof, are intended to cover non-exclusive inclusion. For example, a process, method, apparatus, product, or device that includes a series of steps or units is not limited to the listed steps or units, but may optionally include steps or units not listed, or may optionally include other steps or units inherent to these processes, methods, products, or devices.

[0097] In this document, the term "embodiment" means that a particular feature, structure, or characteristic described in connection with an embodiment may be included in at least one embodiment of the invention. The appearance of this phrase in various places throughout the specification does not necessarily refer to the same embodiment, nor is it a separate or alternative embodiment mutually exclusive with other embodiments. It will be explicitly and implicitly understood by those skilled in the art that the embodiments described herein can be combined with other embodiments.

[0098] This invention discloses a vehicle defect detection method and system based on neural networks. It can determine the defect detection result corresponding to each vehicle data type based on the corresponding neural network, and then predict future defects of the vehicle based on the vehicle data and the corresponding defect detection results. Finally, it comprehensively determines the vehicle defects based on defect association judgment rules. This improves the accuracy and efficiency of vehicle defect identification by combining the prediction results and joint judgments of multiple neural networks, providing accurate data references for vehicle production or repair, and improving vehicle safety. Detailed descriptions follow.

[0099] Example 1

[0100] Please see Figure 1 , Figure 1 This is a schematic flowchart of a vehicle defect detection method based on a neural network disclosed in an embodiment of the present invention. Wherein, Figure 1 The described neural network-based vehicle defect detection method can be applied to data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 1 As shown, the neural network-based vehicle defect detection method may include the following operations:

[0101] 101. Obtain multiple vehicle data for the vehicle to be identified.

[0102] 102. Determine the defect detection result for each vehicle data based on the neural network corresponding to the data type of each vehicle.

[0103] 103. Based on all vehicle data and the corresponding defect detection results, predict the future defects of the vehicle to be identified.

[0104] 104. Based on the defect detection results and future defect predictions, determine the vehicle defects of the vehicle to be identified according to the defect association judgment rules.

[0105] As can be seen, the above-described embodiments of the invention can determine the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data, and then predict the future defects of the vehicle based on the vehicle data and the corresponding defect detection result. Based on the defect association judgment rule, the vehicle defects are comprehensively determined. Thus, the accuracy and efficiency of vehicle defect identification can be improved based on the prediction results and joint judgment of multiple neural networks, providing accurate data reference for vehicle production or maintenance and improving vehicle safety.

[0106] As an optional embodiment, in the above steps, the data types of vehicle data are in-vehicle video recording data, historical driving operation data, historical sensor recording data, and real-time vehicle test data; the historical sensor recording data is obtained by the sensors set on the vehicle to be identified working during historical time periods.

[0107] As can be seen, the specific content of the vehicle data is defined through the above optional embodiments, which can fully characterize the features of the vehicle in its historical driving operations. This data can then be used to accurately predict vehicle defects, assist in realizing prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0108] As an optional embodiment, the step above, determining the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data, includes:

[0109] For each vehicle data, a predictive neural network corresponding to the data type is determined from a pre-defined algorithm model library; optionally, the predictive neural network is trained using a training dataset that includes training vehicle data corresponding to multiple data types of the vehicle data and corresponding defect annotations.

[0110] The vehicle data is input into a predictive neural network to obtain the defect detection results corresponding to the vehicle data; the defect detection results include the defect location, defect type, and defect occurrence probability.

[0111] As can be seen, through the above optional embodiments, the defect detection results corresponding to the vehicle data can be predicted by the predictive neural network corresponding to the data type of the vehicle data, so as to accurately predict vehicle defects, assist in realizing the prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0112] As an optional embodiment, in the above steps, when the data type of the vehicle data is in-vehicle video data, historical sensor record data, or real-time vehicle test data, the network type of the predictive neural network is an RNN neural network; when the data type of the vehicle data is historical driving operation data, the network type of the predictive neural network is a composite network; the composite network includes an operation filtering network and an operation prediction network; the operation filtering network includes multiple classifier algorithm models, an LSTM neural network, and an output judgment model. The classifier algorithm models are used to predict the hazard level of each driving operation in the historical driving operation data. The LSTM neural network is used to predict the next operation for any two consecutive driving operations with a hazard level greater than a first hazard threshold, and compare the operation similarity between the predicted next operation and the actual next driving operation; the output judgment model is used to determine all driving operations with a hazard level greater than a second hazard threshold or an operation similarity greater than a first similarity threshold as dangerous driving operations; the operation prediction network is used to predict the corresponding defect detection results based on the dangerous driving operations.

[0113] As can be seen, the above optional embodiments define the network type and architecture function of the predictive neural network, enabling accurate prediction of vehicle defects represented by vehicle data of different types. This allows for precise prediction of vehicle defects, assists in improving the accuracy and efficiency of vehicle defect identification by combining prediction results and joint judgments based on multiple neural networks, provides accurate data references for vehicle production or maintenance, and enhances vehicle safety.

[0114] As an optional embodiment, the RNN neural network or operation prediction network in the above steps includes a feature extraction network and a classification network; each RNN neural network or operation prediction network is jointly trained through the following steps:

[0115] Multiple training datasets containing training vehicle data of different data types are used as a unified dataset. The unified feature extraction network and the unified classification network are trained on the unified dataset until convergence, resulting in the trained unified feature extraction network and unified classification network.

[0116] For each data type, obtain multiple historical data corresponding to that data type; optionally, the historical data can be training vehicle data or vehicle data obtained at historical time points.

[0117] Based on the complexity algorithm, calculate the data complexity corresponding to each historical data point;

[0118] Calculate the average of the data complexity for all historical data to obtain the type complexity for that data type;

[0119] Based on the predefined correspondence between complexity and distillation parameters, determine the distillation parameters corresponding to the type complexity of the data type; optionally, the distillation parameters include at least one of distillation step size, distillation dataset size, distillation loss function parameters, and distillation model architecture parameters.

[0120] Based on the distillation parameters, the trained unified feature extraction network and unified classification network are distilled to obtain the basic neural network model corresponding to this data type.

[0121] Based on the training dataset corresponding to this data type, the basic neural network model obtained by distillation is trained in a targeted manner until convergence, resulting in a trained RNN neural network or operational prediction network corresponding to this data type.

[0122] As can be seen, through the above optional embodiments, it is possible to achieve unified training of basic networks for multiple data types based on a unified dataset to obtain a unified basic network with more relevant prediction results. Then, based on the calculation of the complexity corresponding to the data type, the corresponding network model distillation is achieved to obtain a more accurate and reasonable prediction network for each data type, so as to accurately predict vehicle defects. This helps to improve the accuracy and efficiency of vehicle defect identification by using prediction results and joint judgment based on multiple neural networks, providing accurate data references for vehicle production or maintenance, and improving vehicle safety.

[0123] As an optional embodiment, the step above, predicting future defects of the vehicle to be identified based on all vehicle data and corresponding defect detection results, includes:

[0124] Randomly sample all vehicle data to obtain multiple sampling sets;

[0125] For each sampling set, all vehicle data in the sampling set are sorted from earliest to latest based on the acquisition time to obtain the vehicle data sequence corresponding to the sampling set;

[0126] The vehicle data sequence and the defect detection results corresponding to each vehicle data are determined as the data sequence to be predicted;

[0127] The data sequence to be predicted is input into a trained LSTM neural network to obtain the future predicted defects corresponding to the sampling set; optionally, the LSTM neural network is trained using a training dataset that includes multiple training vehicle sensor data sequences and corresponding historical defect detection result annotations and future defect detection result annotations; the future predicted defects include defect location, defect type and defect occurrence probability.

[0128] As can be seen, through the above optional embodiments, the potential future defect risks of the vehicle to be identified can be predicted based on the vehicle sensor data sorted by time in the sampling results and the corresponding defect detection results. This allows for accurate prediction of vehicle defects, assists in realizing prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provides accurate data reference for vehicle production or maintenance, and improves vehicle safety.

[0129] As an optional embodiment, the step described above, determining the vehicle defect of the vehicle to be identified based on the defect detection results and future predicted defects, according to defect association judgment rules, includes:

[0130] Based on association rules, the defect association corresponding to each future predicted defect is determined.

[0131] Filter out all future predicted defects whose defect correlation is greater than the correlation threshold, and obtain at least one high-probability future defect;

[0132] The objective function is set to maximize the number of defect results included in the defect set; optionally, the defect results are defect detection results or high-probability future defects.

[0133] The specified conditions include:

[0134] The defect set must include at least one high-probability future defect;

[0135] The distance between the defect locations corresponding to any two defect results in the defect set is less than the distance threshold.

[0136] The area of ​​the minimum envelope shape corresponding to the location of the defect in the defect set is less than the area threshold.

[0137] The average probability of defect occurrence corresponding to all defect results in the defect set is greater than the probability threshold;

[0138] Based on the dynamic programming algorithm, the optimal defect set is obtained by iterative calculation according to the objective function and the constraints, and the set of defects of interest corresponding to the vehicle to be identified is obtained.

[0139] The set of defects of interest is determined as the vehicle defects of the vehicle to be identified.

[0140] As can be seen, through the above optional embodiments, a large number of defect detection results can be aggregated and calculated based on a preset objective function and limiting conditions to determine the concentrated defect set in the vehicle, so as to accurately characterize the defect status of the vehicle, realize the prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0141] As an optional embodiment, the step above, determining the defect correlation corresponding to each future predicted defect based on association rules, includes:

[0142] For each predicted future defect, calculate the first occurrence count of the defect type in all defect detection results.

[0143] Calculate the second occurrence number of the defect location in all defect detection results for the predicted future defect;

[0144] Calculate the first weight, which is proportional to the number of times the first occurrence occurs;

[0145] Calculate the second weight, which is proportional to the number of times the second occurrence occurs;

[0146] The defect correlation corresponding to the predicted defect is obtained by multiplying the first weight, the second weight, and the probability of occurrence of the predicted defect in the future.

[0147] As can be seen, through the above optional embodiments, the correlation between the predicted future defects and the existing defect detection results can be effectively measured based on the number of times the defects appear in the existing defect detection results. This facilitates the subsequent screening of more accurate high-probability future defects, and improves the accuracy and efficiency of vehicle defect identification by using prediction results and joint judgment based on multiple neural networks. This provides accurate data references for vehicle production or maintenance and improves vehicle safety.

[0148] Example 2

[0149] Please see Figure 2 , Figure 2 This is a schematic diagram of the structure of a vehicle defect detection system based on a neural network, as disclosed in an embodiment of the present invention. Wherein, Figure 2 The described neural network-based vehicle defect detection system can be applied to data processing systems / data processing equipment / data processing servers (wherein, the server includes local processing servers or cloud processing servers). For example... Figure 2 As shown, the neural network-based vehicle defect detection system may include:

[0150] The acquisition module 201 is used to acquire multiple vehicle data of the vehicle to be identified.

[0151] The first prediction module 202 is used to determine the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data.

[0152] The second prediction module 203 is used to predict future defects of the vehicle to be identified based on all vehicle data and the corresponding defect detection results.

[0153] The determination module 204 is used to determine the vehicle defects of the vehicle to be identified based on the defect detection results and future predicted defects, and based on the defect association judgment rules.

[0154] As can be seen, the above-described embodiments of the invention can determine the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data, and then predict the future defects of the vehicle based on the vehicle data and the corresponding defect detection result. Based on the defect association judgment rule, the vehicle defects are comprehensively determined. Thus, the accuracy and efficiency of vehicle defect identification can be improved based on the prediction results and joint judgment of multiple neural networks, providing accurate data reference for vehicle production or maintenance and improving vehicle safety.

[0155] As an optional embodiment, the vehicle data types are in-vehicle video recording data, historical driving operation data, historical sensor recording data, and real-time vehicle test data; the historical sensor recording data is obtained by the sensors set on the vehicle to be identified operating during historical time periods.

[0156] As can be seen, the specific content of the vehicle data is defined through the above optional embodiments, which can fully characterize the features of the vehicle in its historical driving operations. This data can then be used to accurately predict vehicle defects, assist in realizing prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0157] As an optional embodiment, the first prediction module determines the specific method of the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data, including:

[0158] For each vehicle data, a predictive neural network corresponding to the data type is determined from a pre-defined algorithm model library; optionally, the predictive neural network is trained using a training dataset that includes training vehicle data corresponding to multiple data types of the vehicle data and corresponding defect annotations.

[0159] The vehicle data is input into a predictive neural network to obtain the defect detection results corresponding to the vehicle data; the defect detection results include the defect location, defect type, and defect occurrence probability.

[0160] As can be seen, through the above optional embodiments, the defect detection results corresponding to the vehicle data can be predicted by the predictive neural network corresponding to the data type of the vehicle data, so as to accurately predict vehicle defects, assist in realizing the prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0161] As an optional embodiment, when the data type of the vehicle data is in-vehicle video recording data, historical sensor recording data, or real-time vehicle test data, the network type of the predictive neural network is an RNN neural network; when the data type of the vehicle data is historical driving operation data, the network type of the predictive neural network is a composite network; the composite network includes an operation screening network and an operation prediction network; the operation screening network includes multiple classifier algorithm models, an LSTM neural network, and an output judgment model. The classifier algorithm models are used to predict the hazard level of each driving operation in the historical driving operation data. The LSTM neural network is used to predict the next operation for any two consecutive driving operations with a hazard level greater than a first hazard threshold, and compare the operation similarity between the predicted next operation and the actual next driving operation; the output judgment model is used to identify all driving operations with a hazard level greater than a second hazard threshold or an operation similarity greater than a first similarity threshold as dangerous driving operations; the operation prediction network is used to predict the corresponding defect detection results based on the dangerous driving operations.

[0162] As can be seen, the above optional embodiments define the network type and architecture function of the predictive neural network, enabling accurate prediction of vehicle defects represented by vehicle data of different types. This allows for precise prediction of vehicle defects, assists in improving the accuracy and efficiency of vehicle defect identification by combining prediction results and joint judgments based on multiple neural networks, provides accurate data references for vehicle production or maintenance, and enhances vehicle safety.

[0163] As an optional embodiment, the RNN neural network or operation prediction network includes a feature extraction network and a classification network; each RNN neural network or operation prediction network is jointly trained through the following steps:

[0164] Multiple training datasets containing training vehicle data of different data types are used as a unified dataset. The unified feature extraction network and the unified classification network are trained on the unified dataset until convergence, resulting in the trained unified feature extraction network and unified classification network.

[0165] For each data type, obtain multiple historical data corresponding to that data type; optionally, the historical data can be training vehicle data or vehicle data obtained at historical time points.

[0166] Based on the complexity algorithm, calculate the data complexity corresponding to each historical data point;

[0167] Calculate the average of the data complexity for all historical data to obtain the type complexity for that data type;

[0168] Based on the predefined correspondence between complexity and distillation parameters, determine the distillation parameters corresponding to the type complexity of the data type; optionally, the distillation parameters include at least one of distillation step size, distillation dataset size, distillation loss function parameters, and distillation model architecture parameters.

[0169] Based on the distillation parameters, the trained unified feature extraction network and unified classification network are distilled to obtain the basic neural network model corresponding to this data type.

[0170] Based on the training dataset corresponding to this data type, the basic neural network model obtained by distillation is trained in a targeted manner until convergence, resulting in a trained RNN neural network or operational prediction network corresponding to this data type.

[0171] As can be seen, through the above optional embodiments, it is possible to achieve unified training of basic networks for multiple data types based on a unified dataset to obtain a unified basic network with more relevant prediction results. Then, based on the calculation of the complexity corresponding to the data type, the corresponding network model distillation is achieved to obtain a more accurate and reasonable prediction network for each data type, so as to accurately predict vehicle defects. This helps to improve the accuracy and efficiency of vehicle defect identification by using prediction results and joint judgment based on multiple neural networks, providing accurate data references for vehicle production or maintenance, and improving vehicle safety.

[0172] As an optional embodiment, the second prediction module predicts the specific method by which it predicts future defects of the vehicle to be identified based on all vehicle data and corresponding defect detection results, including:

[0173] Randomly sample all vehicle data to obtain multiple sampling sets;

[0174] For each sampling set, all vehicle data in the sampling set are sorted from earliest to latest based on the acquisition time to obtain the vehicle data sequence corresponding to the sampling set;

[0175] The vehicle data sequence and the defect detection results corresponding to each vehicle data are determined as the data sequence to be predicted;

[0176] The data sequence to be predicted is input into a trained LSTM neural network to obtain the future predicted defects corresponding to the sampling set; optionally, the LSTM neural network is trained using a training dataset that includes multiple training vehicle sensor data sequences and corresponding historical defect detection result annotations and future defect detection result annotations; the future predicted defects include defect location, defect type and defect occurrence probability.

[0177] As can be seen, through the above optional embodiments, the potential future defect risks of the vehicle to be identified can be predicted based on the vehicle sensor data sorted by time in the sampling results and the corresponding defect detection results. This allows for accurate prediction of vehicle defects, assists in realizing prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provides accurate data reference for vehicle production or maintenance, and improves vehicle safety.

[0178] As an optional embodiment, the determining module determines the specific method of identifying vehicle defects of the vehicle to be identified based on defect detection results and future predicted defects, according to defect association judgment rules, including:

[0179] Based on association rules, the defect association corresponding to each future predicted defect is determined.

[0180] Filter out all future predicted defects whose defect correlation is greater than the correlation threshold, and obtain at least one high-probability future defect;

[0181] The objective function is set to maximize the number of defect results included in the defect set; optionally, the defect results are defect detection results or high-probability future defects.

[0182] The specified conditions include:

[0183] The defect set must include at least one high-probability future defect;

[0184] The distance between the defect locations corresponding to any two defect results in the defect set is less than the distance threshold.

[0185] The area of ​​the minimum envelope shape corresponding to the location of the defect in the defect set is less than the area threshold.

[0186] The average probability of defect occurrence corresponding to all defect results in the defect set is greater than the probability threshold;

[0187] Based on the dynamic programming algorithm, the optimal defect set is obtained by iterative calculation according to the objective function and the constraints, and the set of defects of interest corresponding to the vehicle to be identified is obtained.

[0188] The set of defects of interest is determined as the vehicle defects of the vehicle to be identified.

[0189] As can be seen, through the above optional embodiments, a large number of defect detection results can be aggregated and calculated based on a preset objective function and limiting conditions to determine the concentrated defect set in the vehicle, so as to accurately characterize the defect status of the vehicle, realize the prediction results and joint judgment based on multiple neural networks to improve the accuracy and efficiency of vehicle defect identification, provide accurate data reference for vehicle production or maintenance, and improve vehicle safety.

[0190] As an optional embodiment, the determining module determines the specific method by which it determines the defect association corresponding to each future predicted defect based on association rules, including:

[0191] For each predicted future defect, calculate the first occurrence count of the defect type in all defect detection results.

[0192] Calculate the second occurrence number of the defect location in all defect detection results for the predicted future defect;

[0193] Calculate the first weight, which is proportional to the number of times the first occurrence occurs;

[0194] Calculate the second weight, which is proportional to the number of times the second occurrence occurs;

[0195] The defect correlation corresponding to the predicted defect is obtained by multiplying the first weight, the second weight, and the probability of occurrence of the predicted defect in the future.

[0196] As can be seen, through the above optional embodiments, the correlation between the predicted future defects and the existing defect detection results can be effectively measured based on the number of times the defects appear in the existing defect detection results. This facilitates the subsequent screening of more accurate high-probability future defects, and improves the accuracy and efficiency of vehicle defect identification by using prediction results and joint judgment based on multiple neural networks. This provides accurate data references for vehicle production or maintenance and improves vehicle safety.

[0197] Example 3

[0198] Please see Figure 3 , Figure 3 This is another vehicle defect detection system based on neural networks disclosed in the embodiments of the present invention. Figure 3 The described neural network-based vehicle defect detection system is applied in a data processing system / data processing equipment / data processing server (wherein, the server includes a local processing server or a cloud processing server). For example... Figure 3 As shown, the neural network-based vehicle defect detection system may include:

[0199] Memory 301 storing executable program code;

[0200] Processor 302 coupled to memory 301;

[0201] The processor 302 calls the executable program code stored in the memory 301 to execute the steps of the neural network-based vehicle defect detection method described in Embodiment 1.

[0202] Example 4

[0203] This invention discloses a computer read storage medium that stores a computer program for electronic data interchange, wherein the computer program causes a computer to execute the steps of the neural network-based vehicle defect detection method described in Embodiment 1.

[0204] Example 5

[0205] This invention discloses a computer program product, which includes a non-transitory computer-readable storage medium storing a computer program, and the computer program is operable to cause a computer to perform the steps of the neural network-based vehicle defect detection method described in Embodiment 1.

[0206] The foregoing has described specific embodiments of this specification; other embodiments are within the scope of the appended claims. In some cases, the actions or steps described in the claims may be performed in a different order than those shown in the embodiments and may still achieve the desired result. Furthermore, the processes depicted in the drawings do not necessarily have to follow the specific or sequential order shown to achieve the desired result. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0207] The systems, devices, modules, or units described in the above embodiments can be implemented by computer chips or entities, or by products with certain functions. A typical implementation device is a computer. Specifically, a computer can be, for example, a personal computer, laptop computer, cellular phone, camera phone, smartphone, personal digital assistant, media player, navigation device, email device, game console, tablet computer, wearable device, or any combination of these devices.

[0208] For ease of description, the above devices are described in terms of function, divided into various units. Of course, in implementing this specification, the functions of each unit can be implemented in one or more software and / or hardware components.

[0209] Those skilled in the art will understand that the embodiments of this specification can be provided as methods, systems, or computer program products. Therefore, the embodiments of this specification can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the embodiments of this specification can take the form of a computer program product implemented on one or more computer-usable storage media (including, but not limited to, disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0210] This specification is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this specification. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0211] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0212] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0213] In a typical configuration, a computing device includes one or more processors (CPU), input / output interfaces, network interfaces, and memory.

[0214] Memory may include non-persistent storage in computer-readable media, such as random access memory (RAM) and / or non-volatile memory, such as read-only memory (ROM) or flash RAM. Memory is an example of computer-readable media.

[0215] Computer-readable media includes both permanent and non-permanent, removable and non-removable media that can store information using any method or technology. Information can be computer-readable instructions, data structures, modules of programs, or other data. Examples of computer storage media include, but are not limited to, phase-change memory (PRAM), static random access memory (SRAM), dynamic random access memory (DRAM), other types of random access memory (RAM), read-only memory (ROM), electrically erasable programmable read-only memory (EEPROM), flash memory or other memory technologies, CD-ROM, digital versatile optical disc (DVD) or other optical storage, magnetic tape, magnetic disk storage or other magnetic storage devices, or any other non-transferable medium that can be used to store information accessible by a computing device. As defined herein, computer-readable media does not include transient computer-readable media, such as modulated data signals and carrier waves.

[0216] It should also be noted that the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitation, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0217] This specification can be described in the general context of computer-executable instructions that are executed by a computer, such as program modules. Generally, program modules include routines, programs, objects, components, data structures, etc., that perform a specific task or implement a specific abstract data type. This specification can also be practiced in distributed computing environments, where tasks are performed by remote processing devices connected via a communication network. In distributed computing environments, program modules can reside in local and remote computer storage media, including storage devices.

[0218] The various embodiments in this specification are described in a progressive manner. Similar or identical parts between embodiments can be referred to interchangeably. Each embodiment focuses on describing the differences from other embodiments. In particular, the system embodiments are basically similar to the method embodiments, so the description is relatively simple; relevant parts can be referred to the descriptions in the method embodiments.

[0219] Finally, it should be noted that the vehicle defect detection method and system based on neural networks disclosed in the embodiments of the present invention are merely preferred embodiments of the present invention and are only used to illustrate the technical solutions of the present invention, not to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A vehicle defect detection method based on neural networks, characterized in that, The method includes: Acquire data from multiple vehicles to be identified; Based on the neural network corresponding to the data type of each vehicle data, determine the defect detection result corresponding to each vehicle data. Based on all the vehicle data and the corresponding defect detection results, predict the future defects of the vehicle to be identified; Based on the defect detection results and the predicted future defects, and using defect association judgment rules, the vehicle defects of the vehicle to be identified are determined, including: For each of the predicted future defects, calculate the first occurrence count of the defect type of the predicted future defect in all the defect detection results; Calculate the second occurrence number of the defect location of the predicted future defect in all the defect detection results; Calculate a first weight that is proportional to the first occurrence frequency; Calculate the second weight that is proportional to the second occurrence frequency; Calculate the product of the first weight, the second weight, and the probability of occurrence of the future predicted defect to obtain the defect correlation corresponding to the future predicted defect; Filter out all the predicted future defects whose correlation with the defect is greater than the correlation threshold to obtain at least one high-probability future defect; The objective function is set to maximize the number of defect results included in the defect set; the defect results are the defect detection results or the high-probability future defects. The specified limitations include: The defect set includes at least one of the high-probability future defects; The distance between the defect locations corresponding to any two defect results in the defect set is less than a distance threshold. The area of ​​the minimum envelope shape corresponding to the location of the defect location for all the aforementioned defect results in the defect set is less than the area threshold. The average probability of occurrence of all the aforementioned defects in the defect set is greater than the probability threshold. Based on the dynamic programming algorithm, the optimal defect set is obtained by iterative calculation according to the objective function and the limiting conditions, and the set of defects of concern corresponding to the vehicle to be identified is obtained. The set of defects of concern is determined as the vehicle defects of the vehicle to be identified.

2. The vehicle defect detection method based on neural networks according to claim 1, characterized in that, The vehicle data types include in-vehicle video recording data, historical driving operation data, historical sensor recording data, and real-time vehicle test data; the historical sensor recording data is obtained by sensors installed on the vehicle to be identified operating during historical time periods.

3. The vehicle defect detection method based on neural networks according to claim 2, characterized in that, The step of determining the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data includes: For each type of vehicle data, a predictive neural network corresponding to the data type is determined from a preset algorithm model library; the predictive neural network is trained using a training dataset that includes multiple training vehicle data corresponding to the data type of the vehicle data and corresponding defect annotations. The vehicle data is input into the predictive neural network to obtain the defect detection result corresponding to the vehicle data; the defect detection result includes the defect location, defect type and defect occurrence probability.

4. The vehicle defect detection method based on neural networks according to claim 3, characterized in that, When the vehicle data type is in-vehicle video data, historical sensor record data, or real-time vehicle test data, the predictive neural network type is an RNN neural network; when the vehicle data type is historical driving operation data, the predictive neural network type is a composite network; the composite network includes an operation filtering network and an operation prediction network; the operation filtering network includes multiple classifier algorithm models, an LSTM neural network, and an output judgment model, the classifier algorithm models are used to predict the hazard level of each driving operation in the historical driving operation data, the LSTM neural network is used to predict the next operation for any two consecutive driving operations with a hazard level greater than a first hazard threshold, and compare the operation similarity between the predicted next operation and the actual next driving operation; the output judgment model is used to determine all driving operations with a hazard level greater than a second hazard threshold or an operation similarity greater than a first similarity threshold as dangerous driving operations; the operation prediction network is used to predict the corresponding defect detection result based on the dangerous driving operations.

5. The vehicle defect detection method based on neural networks according to claim 4, characterized in that, The RNN neural network or the operation prediction network includes a feature extraction network and a classification network; each RNN neural network or operation prediction network is jointly trained through the following steps: Multiple training datasets containing training vehicle data of different data types are used as a unified dataset. The unified feature extraction network and the unified classification network are trained on the unified dataset until convergence, resulting in the trained unified feature extraction network and unified classification network. For each data type, obtain multiple historical data corresponding to that data type; the historical data is the training vehicle data or the vehicle data obtained at historical time points; Based on the complexity algorithm, calculate the data complexity corresponding to each of the historical data; Calculate the average of the data complexity corresponding to all the historical data to obtain the type complexity corresponding to the data type; Based on the pre-defined correspondence between complexity and distillation parameters, the distillation parameters corresponding to the type complexity of the data type are determined; the distillation parameters include at least one of distillation step size, distillation dataset size, distillation loss function parameters, and distillation model architecture parameters. Based on the distillation parameters, the trained unified feature extraction network and unified classification network are distilled to obtain the neural network basic model corresponding to this data type. The neural network base model obtained by distillation is trained in a targeted manner based on the training dataset corresponding to the data type until convergence, so as to obtain the trained RNN neural network or the operation prediction network corresponding to the data type.

6. The vehicle defect detection method based on neural networks according to claim 1, characterized in that, The step of predicting future defects of the vehicle to be identified based on all the vehicle data and the corresponding defect detection results includes: Randomly sample all the vehicle data to obtain multiple sampling sets; For each of the sampling sets, all the vehicle data in the sampling set are sorted from earliest to latest based on the acquisition time to obtain the vehicle data sequence corresponding to the sampling set; The vehicle data sequence and the defect detection result corresponding to each vehicle data are determined as the data sequence to be predicted; The data sequence to be predicted is input into a trained LSTM neural network to obtain the future predicted defects corresponding to the sampling set; the LSTM neural network is trained using a training dataset that includes multiple training vehicle sensor data sequences and corresponding historical defect detection result annotations and future defect detection result annotations; the future predicted defects include defect location, defect type and defect occurrence probability.

7. A vehicle defect detection system based on neural networks, characterized in that, The system executes the vehicle defect detection method based on neural networks as described in any one of claims 1-6, the system comprising: The acquisition module is used to acquire multiple vehicle data of the vehicle to be identified; The first prediction module is used to determine the defect detection result corresponding to each vehicle data according to the neural network corresponding to the data type of each vehicle data. The second prediction module is used to predict the future defects of the vehicle to be identified based on all the vehicle data and the corresponding defect detection results. The determination module is used to determine the vehicle defects of the vehicle to be identified based on the defect detection results and the predicted future defects, and according to the defect association judgment rules.

8. A vehicle defect detection system based on neural networks, characterized in that, The system includes: Memory containing executable program code; A processor coupled to the memory; The processor calls the executable program code stored in the memory to execute the vehicle defect detection method based on neural networks as described in any one of claims 1-6.

Images