Intelligent method and system for evaluating used car based on deep learning

Abstract

The application provides a kind of intelligent evaluation method and system of second-hand car based on deep learning, it is related to vehicle value evaluation technical field.In which, the application first obtains the use life recorded by vehicle, the real-time total mileage data reported and the multivariate vibration and temperature signal collected;Second, based on depth image sequence and historical maintenance data record, identify and fuse out damage mark;Then based on use life, real-time total mileage data and multivariate vibration and temperature signal, extract vehicle condition mark;Then based on the serious level in damage mark, the state attenuation level of vehicle is modulated by directional weighting, and the modulated vehicle condition mark is generated;Subsequently, the state feedback optimization is carried out on the associated serious level, and the optimized damage mark is generated;Finally, based on the optimized damage mark and the modulated vehicle condition mark, the intelligent evaluation result of target vehicle is generated.The technical scheme provided by the application realizes the accurate evaluation of deep fusion vehicle condition through the bidirectional modulation and closed-loop optimization of damage and vehicle condition information.

Description

Technical Field

[0001] This application relates to the field of vehicle valuation technology, and in particular to a deep learning-based intelligent valuation method and system for used cars. Background Technology

[0002] As the scale of the used car market expands, the demand for rapid, objective, and accurate intelligent assessment of vehicle conditions is growing. Traditional methods that rely on the personal experience of appraisers are no longer sufficient to meet the needs of large-scale and highly consistent assessments.

[0003] The current technical solution is a used car evaluation method that combines vehicle exterior images with on-board diagnostic system data. This method collects two-dimensional images of the vehicle exterior to identify surface damage and reads fault codes and mileage data from the on-board diagnostic system. It calculates the deduction values ​​for exterior damage and mechanical condition separately, and finally adds the two scores according to preset weights to obtain a comprehensive evaluation score.

[0004] However, this method still has the following shortcomings. First, the assessment of appearance damage is based solely on visual features and does not consider the actual differences in the impact of the damage location on different mechanical systems of the vehicle, such as the engine and transmission. Second, the assessment of mechanical condition relies on a pre-set general attenuation model and fails to make targeted corrections to the degree of aging based on the damage history of specific vehicles. This results in inaccurate assessment of the implicit mechanical condition attenuation caused by accidents or repairs, and the assessment results are difficult to fully reflect the true impact of the complex relationship between damage and vehicle condition on the vehicle's value. Summary of the Invention

[0005] This application provides a deep learning-based intelligent evaluation method and system for used cars, which solves the problems of fragmented evaluation dimensions and failure to achieve in-depth collaborative analysis of damage appearance and internal vehicle condition in existing technologies.

[0006] Firstly, this application provides a deep learning-based intelligent evaluation method for used cars, including: The system acquires a depth image sequence of the target vehicle, historical maintenance data records, the service life recorded on the vehicle nameplate, real-time total mileage data reported by the on-board diagnostic system, and multi-dimensional vibration and temperature signals collected in real time by a sensor group deployed in the core transmission system of the vehicle. The depth image sequence includes the interior and exterior trim and chassis areas. Based on the depth image sequence and the historical maintenance data record, the damage imprint of the target vehicle is identified and fused. The damage imprint includes the spatial location distribution of the damage, the severity level associated with the depth, and the damage type label. Based on the service life, the real-time total mileage data, and the multi-element vibration and temperature signals, the vehicle condition imprint of the target vehicle is extracted. The vehicle condition imprint reflects the current operating status attenuation level of each core system of the vehicle. Based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, the state decay level of the corresponding vehicle system in the vehicle condition imprint is directionally weighted and modulated to generate the modulated vehicle condition imprint. Using the modulated vehicle condition imprint, the severity level associated with the damage type label in the damage imprint is optimized by state feedback to generate an optimized damage imprint. Based on the optimized damage imprint and the modulated vehicle condition imprint, an intelligent evaluation result for the target vehicle is generated collaboratively.

[0007] Optionally, based on the optimized damage imprint and the modulated vehicle condition imprint, an intelligent evaluation result for the target vehicle is collaboratively generated, including: From the optimized damage imprint, damage type labels and the optimized severity level corresponding to the damage type labels, as well as the spatial distribution of the damage, are extracted. Based on the preset damage type value influence weight table, the first evaluation sub-item of the damage on the overall condition of the target vehicle is calculated. From the modulated vehicle condition imprint, extract the modulated system state attenuation vectors corresponding to the engine system, transmission system and drive shaft bearing system respectively, and calculate the second evaluation sub-item of the vehicle system on the overall vehicle condition according to the preset attenuation vector performance influence coefficient table. Based on the preset damage system association mapping relationship, establish the influence association between the damage indicated by the damage type label and the engine system, transmission system and drive shaft bearing system; Based on the association between the damage type label and the impact on each vehicle system, and the current second assessment item for each vehicle system, calculate the third assessment item for the damage's impact on the overall vehicle condition. The first evaluation sub-item corresponding to each damage type label in the optimized damage imprint is fused with the third evaluation sub-item through an integration operation to obtain the final damage evaluation value of the damage type label. The final damage assessment values ​​of all damage type labels are summarized, and the second assessment sub-items of all vehicle systems are also summarized. The results are then processed through a preset joint assessment function to generate an intelligent assessment result.

[0008] Secondly, this application provides a deep learning-based intelligent evaluation system for used cars, including: The acquisition module is used to acquire the target vehicle's depth image sequence, historical maintenance data records, the service life recorded on the vehicle nameplate, the real-time total mileage data reported by the on-board diagnostic system, and multi-dimensional vibration and temperature signals collected in real time by the sensor group arranged in the vehicle's core transmission system. The depth image sequence includes the interior and exterior trim and chassis areas. The identification module is used to identify and fuse the damage imprint of the target vehicle based on the depth image sequence and the historical maintenance data record. The damage imprint includes the spatial location distribution of the damage, the severity level associated with the depth, and the damage type label. The extraction module is used to extract the vehicle condition imprint of the target vehicle based on the service life, the real-time total mileage data and the multi-dimensional vibration and temperature signals. The vehicle condition imprint reflects the current operating status attenuation level of each core system of the vehicle. The modulation module is used to perform directional weighted modulation on the state attenuation level of the corresponding vehicle system in the vehicle condition imprint based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, so as to generate the modulated vehicle condition imprint. The optimization module is used to optimize the severity level associated with the damage type label in the damage imprint using the modulated vehicle condition imprint, and generate an optimized damage imprint. The generation module is used to collaboratively generate an intelligent evaluation result for the target vehicle based on the optimized damage imprint and the modulated vehicle condition imprint.

[0009] Thirdly, this application provides a computing device, including a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a deep learning-based intelligent evaluation method for used cars as described in the first aspect above.

[0010] Fourthly, this application provides a computer storage medium storing a computer program, which, when executed by a computer, implements a deep learning-based intelligent evaluation method for used cars as described in the first aspect.

[0011] This application effectively solves the problem of fragmented assessment dimensions in the prior art by establishing a two-way modulation and optimization mechanism for damage imprints and vehicle condition imprints. Specifically, instead of simply weighting the appearance damage assessment and mechanical condition assessment as two independent steps, it first performs targeted weighted modulation on the state decay level of specific systems such as engines and transmissions based on the spatial distribution and severity level of the damage. This allows the mechanical system state assessment to be specifically corrected based on the location and degree of the specific damage. Then, this corrected vehicle condition imprint, which is closer to the actual operating state of the vehicle, is used to optimize the severity level of the damage. Thus, in terms of assessment logic, a deep correlation and collaborative analysis between the damage appearance and the internal vehicle condition is achieved, breaking down the barrier of the two not affecting each other in traditional methods.

[0012] These or other aspects of this application will become more apparent in the following description of the embodiments. Attached Figure Description

[0013] To more clearly illustrate the technical solutions in this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0014] Figure 1 A flowchart of a deep learning-based intelligent evaluation method for used cars provided in this application is shown; Figure 2 A schematic diagram of the structure of a deep learning-based intelligent evaluation system for used cars provided in this application is shown. Figure 3 A schematic diagram of the structure of a computing device provided in this application is shown. Detailed Implementation

[0015] To enable those skilled in the art to better understand the present application, the technical solution of the present application will be clearly and completely described below with reference to the accompanying drawings.

[0016] In some of the processes described in the specification, claims, and accompanying drawings of this application, multiple operations appearing in a specific order are included. However, it should be clearly understood that these operations may not be executed in the order they appear herein, or may be executed in parallel. The operation numbers, such as 101, 102, etc., are merely used to distinguish different operations and do not themselves represent any execution order. Furthermore, these processes may include more or fewer operations, and these operations may be executed sequentially or in parallel. It should be noted that the descriptions such as "first," "second," etc., in this document are used to distinguish different messages, devices, modules, etc., and do not represent a chronological order, nor do they limit "first" and "second" to different types.

[0017] The technical solutions of this application will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0018] Figure 1 This application provides a flowchart of a deep learning-based intelligent evaluation method for used cars, such as... Figure 1 As shown, the method includes: Step 101: Acquire the depth image sequence of the target vehicle, historical maintenance data records, the service life recorded on the vehicle nameplate, the real-time total mileage data reported by the on-board diagnostic system, and the multi-element vibration and temperature signals collected in real time by the sensor group arranged in the core transmission system of the vehicle. The depth image sequence includes the interior and exterior trim and chassis areas.

[0019] Optionally, step 101 may specifically include: Step 1011: Using multiple depth cameras positioned around the target vehicle, images are taken from different preset angles of multiple exterior surfaces, interior areas, and the structure under the lifted vehicle chassis of the target vehicle to obtain a depth image sequence.

[0020] Step 1012: Retrieve and download historical maintenance data records in chronological order from the de-identified vehicle maintenance blockchain network associated with the target vehicle. The historical maintenance data records include maintenance item descriptions and parts replacement entries.

[0021] Step 1013: Read the production date on the vehicle nameplate of the target vehicle and calculate the total number of days of use from the production date to the evaluation date, as the service life.

[0022] Step 1014: By accessing the on-board diagnostic system interface of the target vehicle, the vehicle's cumulative mileage value reported by the engine control unit is read from the non-volatile memory of the on-board diagnostic system as real-time total mileage data.

[0023] Step 1015: Install a triaxial vibration sensor and a temperature sensor in the core transmission system of the target vehicle. When the vehicle is idling and under preset load conditions, synchronously collect and record the triaxial acceleration values ​​of the triaxial vibration sensor at multiple time points and the temperature measurement values ​​of the temperature sensor at the corresponding time points to form multi-element vibration and temperature signals.

[0024] In this step, the depth image sequence refers to the set of images that reflect the three-dimensional geometry of the vehicle surface, captured from multiple fixed angles by depth cameras arranged in a surround configuration. The data in this sequence is essentially a dense three-dimensional point cloud, used to accurately describe the structural morphology of the vehicle's interior and exterior trim and chassis, and serves as the visual and geometric basis for identifying physical damage.

[0025] Historical maintenance data records refer to electronic archives of past vehicle maintenance events retrieved from a blockchain-based de-identified maintenance network, arranged in chronological order. These records document maintenance content and parts replacement history in structured text form. Their core value lies in providing credible and tamper-proof maintenance facts, which can be used to correlate and verify physical damage.

[0026] Service life refers to the precise total number of calendar days from the date the vehicle was manufactured to the date of evaluation, calculated from the production date recorded on the vehicle's nameplate. It is a scalar value used to quantify the degree of natural aging and material degradation of a vehicle due to the passage of time.

[0027] Real-time total mileage data refers to the cumulative mileage value of a vehicle, which is read directly from the vehicle's engine control unit through a standard diagnostic interface. It is a scalar value used to quantify the mechanical wear and component fatigue caused by actual driving use.

[0028] Multi-source vibration and temperature signals refer to the time-series dataset of physical signals synchronously collected by a sensor array installed in key parts of the core transmission system under two typical operating conditions: idling and simulated load. It contains multi-channel vibration acceleration and temperature information that reflects the real-time operating status of core systems such as engines and transmissions, and is used to evaluate their dynamic performance and health status.

[0029] A depth camera is a camera device that can acquire depth information of a scene. Its working principles, such as structured light and time-of-flight method, are well-known technologies in the field.

[0030] A vehicle nameplate is a plate affixed to a vehicle to record information such as the vehicle model and production date; it is a standard feature of vehicles.

[0031] Preset angles and preset load conditions refer to the technical parameters that are set in advance according to the standard operating procedures before the evaluation. Their specific values ​​can be adjusted according to different vehicle models and evaluation standards, and belong to the implementation details.

[0032] Idle speed refers to the operating state of an engine maintaining the lowest stable speed under no load, which is a standard operating condition for a vehicle.

[0033] In this step, the site is first prepared by installing multiple depth cameras around the target vehicle in the evaluation station according to the preset layout. At the same time, triaxial vibration sensors and temperature sensors are installed at the designated measurement points of the core transmission system, such as the engine mount and gearbox housing. Then, data acquisition is carried out in parallel or sequentially: the depth cameras are controlled to take pictures of the vehicle's exterior and interior from different preset angles, and the chassis is taken pictures after the vehicle is lifted to generate a sequence of depth images. Next, the vehicle is started, and data from all sensors are collected simultaneously under idling conditions. Then, the vehicle is placed on the chassis dynamometer and collected again under preset load conditions, thereby forming a multi-dimensional vibration and temperature signal containing two types of operating condition signals. At the same time, the acquisition of information system data is carried out simultaneously: then the evaluation program submits the identification code of the target vehicle to the de-identified vehicle maintenance blockchain network, queries and downloads its historical maintenance data records sorted by time; Subsequently, the production date on the vehicle nameplate is obtained through image recognition or manual input, and the total number of days up to the evaluation date is calculated as the service life. Finally, the cumulative mileage value in the engine control unit is read through the diagnostic equipment connected to the vehicle's on-board diagnostic system interface as real-time total mileage data. Ultimately, these five types of data together constitute the complete data foundation for subsequent analysis.

[0034] For example, for a target SUV vehicle with vehicle identification number LBV5S3100ES123456, at the evaluation station, five depth cameras arranged around the vehicle, including an exterior camera facing the left front, front, and right front, a top-down camera facing the interior, and a bottom-up camera facing the raised chassis, take pictures sequentially from five fixed angles to obtain a depth image sequence consisting of five 1280 by 720 pixel depth image files. Next, vibration and temperature sensors were installed on the right engine mount and gearbox housing of the vehicle. The vehicle was then allowed to idle for 3 minutes while the sensor data was collected synchronously. The vehicle was then placed on a chassis dynamometer and driven at a speed of 50 km / h with a certain load for 3 minutes. Data was collected synchronously again, thus forming a multi-dimensional vibration and temperature signal dataset containing multi-channel raw signals under both idling and load conditions. The evaluation process then submits the vehicle identification number to the de-identified vehicle maintenance blockchain network via an application programming interface (API) for querying. The network returns three records sorted by time, thus obtaining its historical maintenance data. The records include descriptions such as "right front door paint repair" on May 10, 2023, and "engine mount replacement" on January 15, 2024. Subsequently, the vehicle's nameplate image is read using an optical character recognition device, identifying the production date as August 15, 2022. Combined with the system's current evaluation date of March 29, 2026, the difference in days is calculated, resulting in the vehicle's service life being 1322 days. Finally, by connecting a universal diagnostic tool to the vehicle's OBD-II interface, sending and parsing standard diagnostic commands, the value 85500 was read from the engine control unit's memory, thus obtaining the vehicle's real-time total mileage data of 85500 kilometers; at this point, all five types of data required for the evaluation of the target vehicle have been obtained.

[0035] Step 102: Based on the depth image sequence and the historical maintenance data record, identify and fuse the damage imprint of the target vehicle. The damage imprint includes the spatial location distribution of the damage, the severity level associated with the depth, and the damage type label.

[0036] Optionally, step 102 may specifically include: Step 1021: Based on the depth image sequence, the visual structural information and three-dimensional geometric information of the surface of each component of the target vehicle are parsed out.

[0037] Step 1022: Based on the repair item descriptions and part replacement entries recorded in the historical repair data records in chronological order, parse out the component location identifiers and repair action descriptions associated with each repair event.

[0038] Step 1023: Associate the visual structural information, the three-dimensional geometric information, the parsed component location identifiers, and the maintenance action descriptions to identify candidate areas with maintenance traces on the physical structure of the target vehicle.

[0039] Step 1024: Based on the three-dimensional point cloud data of the corresponding position in the depth image sequence, determine the specific spatial location and contour range of the candidate region in the vehicle's three-dimensional coordinate system, as the spatial location distribution of the damage.

[0040] Step 1025: Combine the surface depression depth contained in the three-dimensional geometric information of the candidate region with the maintenance nature recorded in the associated maintenance action description to jointly determine the severity level of the depth association.

[0041] Step 1026: Determine the damage type label based on the texture and color anomaly patterns presented by the visual structure information of the candidate region, as well as the professional terms used in the associated repair project description.

[0042] Step 1027: Combine the spatial location distribution of damage corresponding to the candidate region, the severity level of depth association, and the damage type label to obtain the damage imprint of the target vehicle.

[0043] Visual structural information refers to image attribute data extracted from depth image sequences that describes the appearance features of vehicle parts, such as color, texture, and pattern continuity, including paint color difference, texture breakage, and splicing gaps between parts, and is used to help determine the type of damage.

[0044] Three-dimensional geometric information refers to data calculated from the three-dimensional point cloud data corresponding to the depth image sequence, which describes the spatial geometric attributes of the surface of vehicle parts, such as the surface depression depth, protrusion height, and contour deformation, and is used to quantify the severity of damage.

[0045] Component location identifiers refer to the names or codes of specific vehicle parts to which maintenance operations are performed, identified from the text descriptions in historical maintenance data records using text parsing technology. These identifiers include things like the right front door, left A-pillar, and engine hood, and are used to locate the text records on the vehicle's physical structure.

[0046] Maintenance action descriptions refer to action keywords extracted from the textual descriptions of historical maintenance data records that describe the nature of the maintenance work performed, such as sheet metal repair, replacement, painting, and repair. These keywords are used to help determine the repair history of the damage and any potential hidden problems.

[0047] Candidate regions refer to a local three-dimensional spatial range on the physical structure of the target vehicle that may have damage or repair marks, initially delineated by the algorithm. It is an intermediate result of the damage mark generation process and a target area for subsequent precise positioning, severity assessment and type determination.

[0048] The spatial distribution of damage refers to the spatial occupancy description of the candidate region in the unified three-dimensional coordinate system of the vehicle after the candidate region is precisely defined. The generated result can usually be represented as the corner coordinates of a three-dimensional bounding box or a sequence of polygon vertex coordinates, which clearly indicates where the damage is located and how large it is.

[0049] The severity level of depth correlation refers to a numerical index that quantifies the severity of damage or repair traces at a candidate region by combining the physical deformation depth of the surface of the candidate region derived from three-dimensional geometric information and the nature of the region’s historical repairs derived from repair action descriptions. This level reflects the physical severity of the damage.

[0050] Damage type label refers to a classification name assigned to the damage by combining the visual anomaly pattern of the candidate area surface from visual structural information and the project description keywords of the area's historical repairs. Such as scratches on paint, sheet metal repair marks, parts replacement marks, etc., this label describes what the damage is.

[0051] In this step, firstly, a unified 3D model of the vehicle is generated based on the depth image sequence through 3D point cloud registration and reconstruction technology. Then, computer vision algorithms such as feature extraction and segmentation are used to parse the appearance features describing the surface color and texture of each component (i.e., visual structure information) and the geometric features describing the surface shape and unevenness (i.e., 3D geometric information) from the model and its associated color texture information. At the same time, based on historical maintenance data records, named entity recognition and keyword extraction technology in natural language processing are used to parse the component location identifier indicating the maintenance part and the maintenance action description describing the maintenance operation type from each maintenance item description and part replacement entry. Next, the parsed visual structural information, 3D geometric information, and component location identifiers and repair action descriptions from the parsed text are spatially and semantically correlated. Specifically, the component location identifiers in the text are mapped to the corresponding components in the 3D model. Within the component region, areas that simultaneously satisfy visual structural anomalies such as color difference and texture discontinuity, and 3D geometric anomalies such as local depressions are searched. This locates one or more candidate regions with abnormal traces on the physical structure of the target vehicle. Then, for each candidate region, based on its corresponding 3D point cloud data, the spatial distribution boundary of the point cloud is calculated to determine the specific spatial coordinate range and contour shape of the candidate region in the vehicle's 3D coordinate system, generating the spatial location distribution of the damage. At the same time, combined with the quantified maximum or average depth of surface depressions in the 3D geometric information of the candidate region, and the degree of repair intervention implied in the associated repair action description (such as replacement usually being associated with more severe early damage than repair), a value is calculated through a preset mapping rule as the severity level of the depth association of the damage. Furthermore, based on the specific abnormal patterns presented by the visual structure information of the candidate area, such as regular linear color differences which may indicate scratches, irregular patches which may indicate rust, and the professional terms in the associated repair item descriptions, such as painting and sheet metal work, a descriptive name is determined through a classification model or rule, serving as the damage type label for the damage. Finally, the three attributes of the spatial location distribution of the damage corresponding to each candidate area, the severity level of the deep association, and the damage type label are combined to form a complete record. All such records are collected together to obtain the damage imprint of the target vehicle.

[0052] For example, for the target SUV vehicle with vehicle identification number LBV5S3100ES123456, a 3D model of the vehicle is first generated based on its five depth image sequences using a point cloud registration algorithm. From this model, visual structural information is extracted about a region in the right front door area that exhibits a significant color difference from the surrounding paint and a slightly rough texture. Simultaneously, the 3D geometric information of this region shows a slight dent with an average depth of approximately 1.5 mm. Furthermore, based on the "right front door paint repair" entry in its historical repair data records, the component location is identified as "right front door" through text parsing, and the repair action is described as "paint repair." Next, the "right front door" location identifier in the text is mapped onto the right front door component of the 3D model, and candidate regions on this component that coincide with the color difference and dent area are located. Then, for this candidate area, its spatial distribution is calculated based on its point cloud data to be an approximately rectangular region with center coordinates of approximately X=1200, Y=450, Z=800 mm (in the vehicle coordinate system), and an outline length of approximately 300 mm and a width of approximately 150 mm. Combining the average dent depth of 1.5 mm in this area with the repair action of "paint repair," and by consulting a preset severity mapping table, its depth-related severity level is determined to be 2 (the level range is assumed to be 1-5, with higher values ​​indicating greater severity). Simultaneously, based on the visual color difference pattern of this area and the keyword "paint repair," its damage type is labeled as "paint repair trace." Finally, the three pieces of information—"spatial distribution with center coordinates (1200, 450, 800), 300×150mm rectangle, severity level 2, and type label 'paint repair trace'"—are combined into a single data object. Similarly, other traces that may exist in the engine compartment area are processed. The collection of all these data objects constitutes the damage mark of the target vehicle.

[0053] For example, following the specific implementation of the previous step, for the target SUV vehicle with vehicle identification number LBV5S3100ES123456, firstly, a 3D model is generated based on its depth image sequence, and the presence of color difference and slight dent in the right front door area is analyzed; simultaneously, the right front door and paint repair are analyzed based on historical repair records; then, the text position is mapped onto the 3D model, and the area where the color difference and dent overlap is located as a candidate area; then, based on the point cloud data of the candidate area, its spatial location distribution is determined to be approximately a rectangle with center coordinates of 1200, 450, 800 mm and an outline of 300 mm by 150 mm; combined with the average dent depth of 1.5 mm and the paint repair description, the severity level of its depth association is determined to be 2 through preset rules; and based on visual color difference and repair keywords, its damage type label is determined to be paint repair trace; finally, the above spatial location distribution, severity level, and type label are combined into a data object, and all abnormal areas are processed in the same way. All the data objects obtained together constitute the damage mark of the vehicle.

[0054] Step 103: Based on the service life, the real-time total mileage data, and the multi-element vibration and temperature signals, extract the vehicle condition imprint of the target vehicle. The vehicle condition imprint reflects the current operating status attenuation level of each core system of the vehicle.

[0055] Optionally, step 103 may specifically include: Step 1031: Based on the time decay curve and mileage decay curve determined by vehicle industry experience data, and combined with the service life and the real-time total mileage data, calculate the basic aging coefficient.

[0056] Step 1032: Separate the first set of vibration and temperature signals collected under the target vehicle idling condition and the second set of vibration and temperature signals collected under the preset load condition from the multi-element vibration and temperature signals.

[0057] Step 1033: Analyze the vibration signal in the first set of vibration and temperature signals to obtain the first vibration spectrum of the engine in the idling state, and analyze the temperature signal in the first set of vibration and temperature signals to obtain the first temperature value of the engine in the idling state.

[0058] Step 1034: Analyze the vibration signal in the second set of vibration signals and temperature signals to obtain the second vibration spectrum of the transmission system under load, and analyze the temperature signal in the second set of vibration and temperature signals to obtain the second temperature value of the transmission system under load.

[0059] Step 1035: For the engine system, determine the spectral energy associated with the engine ignition frequency from the first vibration spectrum, determine the deviation from the engine standard operating temperature from the first temperature value, and combine the spectral energy and the deviation from the engine standard operating temperature to obtain the engine abnormality index.

[0060] Step 1036: For the transmission system, determine the spectral energy associated with the gear meshing frequency from the second vibration spectrum, determine the abnormal rise in transmission oil temperature from the second temperature value, and combine the spectral energy and the abnormal rise in transmission oil temperature to obtain the transmission abnormality index.

[0061] Step 1037: For the transmission shaft bearing system, determine the spectral energy associated with the bearing characteristic fault frequency from the second vibration spectrum, and combine it with the temperature value of the bearing housing position in the second temperature value to obtain the bearing abnormality index.

[0062] Step 1038: Using the basic aging coefficient, the engine abnormality index, the gearbox abnormality index, and the bearing abnormality index are weighted and modulated to generate system state decay vectors corresponding to the engine system, gearbox system, and drive shaft bearing system, respectively.

[0063] Step 1039: Combine the system state decay vector of the engine system, the system state decay vector of the transmission system, and the system state decay vector of the drive shaft bearing system to form a vehicle condition imprint that reflects the current operating state decay level of each core system of the vehicle.

[0064] In this step, vehicle condition imprint refers to a set of digital descriptions generated by analyzing the vehicle's usage time, mileage, and real-time vibration and temperature signals of the core transmission system. These descriptions are used to quantitatively characterize the current operating status attenuation level of each core mechanical system of the vehicle.

[0065] The basic aging coefficient refers to the general aging law derived from a large amount of vehicle industry experience data. Specifically, it is represented by time decay curves and mileage decay curves. It is a scalar coefficient obtained by interpolation or calculation by combining the specific service life and real-time total mileage data of the target vehicle.

[0066] Vibration spectrum refers to the spectrum obtained by performing a fast Fourier transform on the original acceleration time-domain signal collected by the vibration sensor, which converts the signal from the time domain to the frequency domain, representing the distribution of signal energy at different frequency components. The first vibration spectrum is used to analyze the vibration characteristics of the engine under idling conditions, and the second vibration spectrum is used to analyze the vibration characteristics of the transmission system under load conditions.

[0067] An anomaly index is a quantitative value obtained by extracting specific physical characteristics from the vibration spectrum and temperature signals of a particular system to characterize its abnormal state, and then fusing them according to preset rules. The larger the value, the greater the degree to which the system deviates from its healthy state.

[0068] The system state decay vector refers to the mathematical vector generated by weighted and fused modulation of anomaly indicators that reflect the degree of anomaly in a specific system, combined with a basic aging coefficient that represents the overall aging level of the vehicle, to ultimately describe the health status of the system. This vector is multidimensional, and its different dimensions can carry state information modulated by different features such as vibration and temperature. It is the basic unit that constitutes the vehicle condition imprint.

[0069] In this step, firstly, based on the standard time decay curve and mileage decay curve in the industry experience database, the specific service life and real-time total mileage data of the target vehicle are substituted into the corresponding curves for interpolation calculation to obtain the time decay factor and mileage decay factor. Then, these two factors are fused by weighted averaging or multiplication to calculate the basic aging coefficient of the vehicle. At the same time, from the collected multi-element vibration and temperature signals, according to the operating condition identifier carried in the signals, the data are separated into the first set of vibration and temperature signals under idling conditions and the second set of vibration and temperature signals under preset load conditions. Next, spectral analysis is performed on the vibration signal in the first group of signals to obtain the first vibration spectrum. The temperature signal in the first group of signals is statistically analyzed, such as by taking the average value, to obtain the first temperature value. The same processing is performed on the second group of signals to obtain the second vibration spectrum and the second temperature value. Then, for the engine system, the sum of spectral energy related to the engine ignition frequency and its harmonics is extracted from the first vibration spectrum. The absolute difference between the first temperature value and the engine standard operating temperature reference value is calculated. These two quantities are combined through a linear or nonlinear function to obtain the engine abnormality index. Subsequently, for the transmission system, spectral energy related to the theoretical meshing frequency of the gear is extracted from the second vibration spectrum, and the difference between the measured transmission oil temperature and the standard oil temperature reference value under the same working condition is calculated from the second temperature value. Similarly, the transmission abnormality index is obtained by combining these two quantities. For the drive shaft bearing system, spectral energy related to the theoretical fault characteristic frequency of the bearing is extracted from the second vibration spectrum, and the second temperature value at the bearing mounting point is directly obtained. The bearing abnormality index is obtained by combining these two quantities. Then, using the calculated base aging coefficient, the abnormal indicators of the engine, transmission, and bearing are modulated respectively. For example, each abnormal indicator is multiplied by the base aging coefficient to generate system state decay vectors corresponding to the three systems, which include aging background correction. Finally, the system state decay vectors of the engine system, transmission system, and drive shaft bearing system are arranged and combined in a predetermined order to form a complete data structure. This structure is the vehicle condition imprint that reflects the current operating state decay level of each core system of the vehicle.

[0070] For example, following the specific implementation of the previous step, for the target SUV vehicle, based on its 1322 days of service life and 85500 kilometers of real-time total mileage, the standard attenuation curve in industry experience data is queried, and its basic aging coefficient is calculated by interpolation. First, from the multi-element vibration and temperature signals collected from the vehicle, the first set of signals under idling conditions and the second set of signals under load conditions of 50 km / h are separated. Second, the spectral analysis of the idling vibration signal is performed to obtain the first vibration spectrum, and the average engine idling temperature is calculated as the first temperature value. Then, the spectral analysis of the load vibration signal is performed to obtain the second vibration spectrum, and the transmission oil temperature and bearing housing temperature are obtained as the second temperature values. Then, for the engine system, the spectral energy at its ignition operating frequency is extracted from the first vibration spectrum, and combined with the deviation of the first temperature value from the standard value, the engine abnormality index is calculated. Subsequently, for the transmission system, the spectral energy at the gear meshing frequency was extracted from the second vibration spectrum, and combined with the abnormal increase in transmission oil temperature to calculate the transmission anomaly index. Simultaneously, for the driveshaft bearing system, the spectral energy at the bearing characteristic fault frequency was extracted from the second vibration spectrum, and combined with the bearing housing temperature to calculate the bearing anomaly index. Then, the three anomaly indices were weighted and modulated using a basic aging coefficient to generate corresponding system state decay vectors. Finally, the system state decay vectors of the engine system, transmission system, and driveshaft bearing system were combined to form the vehicle's condition profile.

[0071] Step 104: Based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, perform directional weighted modulation on the state decay level of the corresponding vehicle system in the vehicle condition imprint to generate the modulated vehicle condition imprint.

[0072] Optionally, step 104 may specifically include: Step 1041: Analyze the damage imprint to obtain the spatial location distribution of the damage and the severity level of the depth associated with the spatial location distribution.

[0073] Step 1042: According to the preset mapping table, the location of the physical component indicated by the spatial location distribution of the damage is mapped to the target vehicle system affected by the location of the physical component. The target vehicle system is the engine system, transmission system or drive shaft bearing system that constitutes the vehicle condition imprint.

[0074] Step 1043: Based on the severity level of the depth association corresponding to the spatial location distribution of the damage, calculate the influence factor value through a preset conversion rule.

[0075] Step 1044: Locate all damages mapped to the target vehicle system and obtain all influence factor values ​​corresponding to the damages.

[0076] Step 1045: All influence factor values ​​mapped to the same target vehicle system are processed through a preset aggregation function to generate a comprehensive modulation coefficient.

[0077] Step 1046: Multiply the system state attenuation vector corresponding to the target vehicle system in the vehicle condition imprint with the calculated comprehensive modulation coefficient to obtain the modulated system state attenuation vector of the target vehicle system. Step 1047: After performing the multiplication operation on the system state attenuation vectors of all target vehicle systems in the vehicle condition imprint, recombine all the modulated system state attenuation vectors to generate the modulated vehicle condition imprint.

[0078] In this step, the modulated vehicle condition imprint refers to an updated set of vehicle condition descriptions obtained by modifying the generated vehicle condition imprints that reflect the baseline attenuation levels of each core system of the vehicle using the damage location and severity information recorded in the damage imprints. The result is an updated version of the vehicle condition imprints, in which the state attenuation vector of the target vehicle system has been weighted and adjusted by the damage information associated with it, so as to more closely reflect the real situation of the mechanical state attenuation that may be aggravated by the specific damage to the vehicle.

[0079] The physical component location refers to the specific part of the vehicle's physical structure that the spatial distribution of damage in the damage imprint precisely points to, such as the middle of the right front door panel or the rear edge of the left front fender. This is the spatial bridge connecting surface damage and the internal mechanical system.

[0080] The target vehicle system refers specifically to those core mechanical systems whose operating status, as described by the vehicle condition signature, may be affected by damage to specific physical components, such as the engine system, transmission system, or driveshaft bearing system. Damage to one physical component may affect one or more target vehicle systems.

[0081] The impact factor value refers to a numerical value obtained by transforming a single damage point into a target vehicle system through a pre-defined mathematical relationship, such as a linear function or a lookup table, based on the severity level associated with the depth of the damage. The higher the severity level, the larger the resulting impact factor value is usually.

[0082] The comprehensive modulation coefficient refers to a single coefficient obtained by taking all the influence factor values ​​corresponding to all the damages mapped onto a specific target vehicle system and processing them through a preset aggregation calculation method, such as taking the maximum value, summing, or weighted averaging. This coefficient represents the joint influence intensity of all related damages on the state of the target vehicle system and will be used to directly modulate the original state decay vector of the system.

[0083] In this step, the generated damage imprints are first read and parsed, and the spatial distribution data of the damage and the severity level data associated with each damage record are extracted one by one. Then, according to a predefined and stored mapping table, the physical component location described by the spatial distribution data of each damage, such as the front end of the right front longitudinal beam, is queried and mapped to one or more target vehicle systems that may be affected. For example, damage to the right front longitudinal beam may be mapped to the engine system. Then, for each damage, the impact factor value of the damage on the affected system is calculated according to its own severity level associated with the damage through a preset conversion rule. For example, a severity level of 3 may correspond to an impact factor value of 0.3. Next, for each target vehicle system in the vehicle condition imprint, all damages mapped to that system are found in the previous mapping results, and all influence factor values ​​corresponding to these damages are collected. These multiple influence factor values ​​belonging to the same target vehicle system are calculated using a preset aggregation function, such as taking the maximum value, to generate a comprehensive modulation coefficient for modulating the system's state decay vector. Then, the original system state decay vector of the target vehicle system in the vehicle condition imprint is multiplied by the calculated comprehensive modulation coefficient for that system to obtain a new system state decay vector after modulation. Finally, the above mapping, influence factor calculation, aggregation to generate comprehensive modulation coefficient and vector modulation operations are repeated for all target vehicle systems in the vehicle condition imprint. All the new system state decay vectors after modulation are recombined according to the original structure to generate the final modulated vehicle condition imprint after damage information-oriented weighted modulation.

[0084] For example, following the specific embodiment of the previous step, for the target SUV vehicle, its damage imprint is analyzed. First, the damage imprint records a damage located in the right front door area, with a corresponding depth-related severity level of 2. By querying a preset mapping table, the damage in the right front door area is mapped to the engine system, the target vehicle system. Second, based on the severity level of 2, the impact factor value of the damage on the engine system is calculated using a preset conversion rule. Next, in this example, the impact factor value is a value obtained by converting the severity level of the damage. Since only this one damage is mapped to the engine system, the comprehensive modulation coefficient corresponding to the engine system is this impact factor value. Then, the original system state decay vector of the engine system is obtained from the vehicle condition imprint. The vector is combined with the calculated comprehensive modulation coefficient to generate the modulated engine system state decay vector. Subsequently, assuming that the transmission system and driveshaft bearing system have no related damage mapping, their system state decay vectors remain unchanged. Finally, the modulated system state decay vectors of the engine system, transmission system, and driveshaft bearing system are recombined to generate the modulated vehicle condition imprint.

[0085] Step 105: Using the modulated vehicle condition imprint, perform state feedback optimization on the severity level associated with the damage type label in the damage imprint to generate an optimized damage imprint.

[0086] Optionally, step 105 may specifically include: Step 1051: Analyze the damage imprint to obtain damage type labels and the severity level of the depth association corresponding to each damage type label.

[0087] Step 1052: Based on the preset damage type system influence mapping table, determine the target vehicle system that is mainly affected by each damage type label. The target vehicle system is the engine system, transmission system, or drive shaft bearing system that constitutes the modulated vehicle condition imprint.

[0088] Step 1053: Extract the modulated system state decay vector corresponding to each target vehicle system from the modulated vehicle condition imprint.

[0089] Step 1054: The modulated system state attenuation vector corresponding to the target vehicle system mainly affected by the damage type label is used to calculate the system influence factor through a preset vector influence factor conversion function.

[0090] Step 1055: All system influence factors corresponding to the damage type label are merged into overall state optimization coefficients through a preset aggregation operation.

[0091] Step 1056: Combine the severity level of the depth association corresponding to the damage type label with the overall state optimization coefficient, and calculate the optimized severity level of the damage type label through a preset severity level correction rule.

[0092] Step 1057: Update the severity level associated with each damage type label in the damage imprint to the optimized severity level, while keeping the spatial distribution of damage and damage type labels in the damage imprint unchanged, so as to obtain the optimized damage imprint.

[0093] In this step, the optimized damage imprint refers to an updated set of digital damage descriptions obtained by re-evaluating and correcting the severity level of each damage in the original damage imprint after the vehicle core system's true state, as reflected in the modulated vehicle condition imprint and corrected by damage information. The result is an optimized version of the damage imprint, in which the severity level of the damage more accurately reflects the actual impact and severity of the damage under the current mechanical condition of the vehicle.

[0094] The Damage Type System Impact Mapping Table is a pre-defined data mapping table that defines the association between different damage type labels and the target vehicle systems that may be affected by them. For example, damage type paint repair marks may primarily map to the engine system, while structural component deformation may simultaneously map to both the engine system and the drive shaft bearing system. This table is used to establish a logical association between the nature of the damage and the affected mechanical system.

[0095] The system impact factor is a value calculated from the modulated system state decay vector of a target vehicle system in the modulated vehicle condition imprint using a preset transformation function. It quantifies the degree of corrective impact that the current actual decay state of the target vehicle system may have on the severity level of the associated damage. The worse the system state, the larger the calculated system impact factor is usually, which means that the damage may need to be reassessed as more severe.

[0096] The overall state optimization coefficient refers to a single coefficient obtained by combining the system influence factors of all target vehicle systems associated with a specific damage type label through a preset aggregation operation. It represents the overall feedback correction intensity of the comprehensive state of all related mechanical systems to the severity level of this type of damage.

[0097] The severity level correction rule refers to a preset calculation rule used to combine the original severity level of the deep association with the calculated overall state optimization coefficient to calculate the optimized severity level. For example, the rule can be to multiply the original level by the overall state optimization coefficient, or to make incremental adjustments to the original level based on the interval in which the coefficient is located.

[0098] In this step, the generated damage imprint is first parsed to extract each damage type label and its original severity level of deep association. Next, based on a pre-defined damage type system impact mapping table, one or more target vehicle systems primarily affected by each damage type label are queried and identified. These target vehicle systems are the engine system, transmission system, or driveshaft bearing system that constitutes the modulated vehicle condition imprint. Then, from the generated modulated vehicle condition imprint, the modulated system state decay vectors corresponding to each of these target vehicle systems are extracted. Subsequently, for each damage type label, the modulated system state decay vectors corresponding to each target vehicle system associated with it are used to calculate a system impact factor through a pre-defined vector impact factor transformation function. Then, all system impact factors corresponding to the same damage type label are merged into a single overall state optimization coefficient through a preset aggregation operation, such as taking the maximum value or calculating the average value. At the same time, for each damage type label, its original deep-association severity level and the calculated overall state optimization coefficient are combined with the preset severity level correction rule to calculate the optimized severity level of the damage type label. Finally, the original severity level associated with each damage type label in the damage imprint is updated to the calculated optimized severity level, while keeping the spatial distribution of damage and the damage type label attributes in the damage imprint unchanged, thus obtaining the final optimized damage imprint.

[0099] For example, following the specific implementation of the previous step, for the target SUV vehicle, firstly, its damage marks are analyzed to obtain the damage type label of one of the damages as paint repair marks, and its original depth-related severity level is 2; secondly, according to the preset damage type system impact mapping table, it is found that the target vehicle system mainly affected by the paint repair marks is the engine system. Next, the modulated system state attenuation vector of the engine system is extracted from the modulated vehicle condition imprint. The system influence factor of the engine system on the damage of paint repair marks is calculated by passing the vector through the preset vector influence factor transformation function. Then, since this damage is only related to the engine system, this factor is the overall state optimization coefficient; subsequently, the original severity level 2 and this overall state optimization coefficient are calculated using the preset severity level correction rules to obtain the optimized severity level of the paint repair mark; finally, the severity level of the damage in the damage mark is updated to this optimized value, while maintaining its spatial distribution.

[0100] Step 106: Based on the optimized damage imprint and the modulated vehicle condition imprint, a collaborative intelligent evaluation result for the target vehicle is generated.

[0101] Optionally, step 106 may specifically include: Step 1061: Extract the damage type label and the optimized severity level corresponding to the damage type label, as well as the spatial distribution of the damage, from the optimized damage imprint. Calculate the first assessment sub-item of the damage's impact on the overall condition of the target vehicle based on the preset damage type value influence weight table.

[0102] Step 1062: Extract the modulated system state attenuation vectors corresponding to the engine system, transmission system and drive shaft bearing system from the modulated vehicle condition imprint, and calculate the second evaluation sub-item of the vehicle system on the overall vehicle condition according to the preset attenuation vector performance influence coefficient table.

[0103] Step 1063: Based on the preset damage system association mapping relationship, establish the influence association between the damage indicated by the damage type label and the engine system, transmission system and drive shaft bearing system.

[0104] Step 1064: Based on the association between the damage type label and the impact on each vehicle system, and the current second assessment item of each vehicle system, calculate the third assessment item of the damage on the overall condition of the vehicle.

[0105] Step 1065: The first evaluation sub-item corresponding to each damage type label in the optimized damage imprint is fused with the third evaluation sub-item through an integration operation to obtain the final damage evaluation value of the damage type label.

[0106] Step 1066: Summarize the final damage assessment values ​​of all damage type labels, and simultaneously summarize the second assessment sub-items of all vehicle systems. Process the results using a preset joint assessment function to generate intelligent assessment results.

[0107] The second evaluation sub-item refers to an independent score calculated based on the preset attenuation vector performance influence coefficient table, for each core vehicle system engine, transmission, and drive shaft bearing system in the modulated vehicle condition imprint, according to its modulated system state attenuation vector, representing the impact of the current state of the system on the overall performance or condition of the vehicle. It reflects the direct impact of the health status of each core mechanical system on the overall condition of the vehicle.

[0108] Damage system association mapping refers to a pre-defined association rule used to establish a specific connection between the damage nature indicated by the damage type label and the engine system, transmission system, and drive shaft bearing system that may be affected by it. Unlike the mapping focus, this focuses more on the value or impact association during the assessment.

[0109] The third assessment sub-item refers to the score calculated based on the damage system correlation mapping relationship and combined with the second assessment sub-item of each relevant vehicle system. It represents the additional impact score of the damage on the overall condition of the vehicle due to its potential or actual impact on a specific mechanical system, reflecting the consequences indirectly caused by the damage through its impact on the mechanical system.

[0110] The final damage assessment value refers to the final, comprehensive impact score on the overall condition of the vehicle obtained by fusing the inherent impact of the first assessment sub-item and the indirect impact of the third assessment sub-item for each damage in the optimized damage imprint through a preset integration calculation.

[0111] The joint evaluation function refers to a pre-defined mathematical processing model or rule used to combine the final damage assessment scores of all damages with the summation scores of the second evaluation sub-items of all vehicle systems to generate a structured intelligent evaluation result. It defines how to combine the evaluation information from the damage side and the vehicle condition side into a unified conclusion.

[0112] In this step, firstly, the damage type label, optimized severity level, and spatial distribution of each damage are extracted from the optimized damage imprint. Then, based on a predefined damage type value influence weight table, which sets different weight coefficients for different types of damage, and combined with the optimized severity level of each damage, a first evaluation sub-item is calculated for each damage through weighted calculation or other mapping methods. Simultaneously, from the modulated vehicle condition imprint, the modulated system state attenuation vectors of the engine system, transmission system, and driveshaft bearing system are extracted. Then, based on another pre-defined attenuation vector performance influence coefficient table, which defines the influence coefficient of attenuation vectors of different states on the overall vehicle performance, a second evaluation sub-item is generated for each core vehicle system. Next, based on the preset damage system association mapping relationship, a specific impact association is established between the damage indicated by each damage type label and one or more vehicle systems. Then, for each damage, based on the various vehicle systems it is associated with and the current second evaluation sub-items of these vehicle systems, the third evaluation sub-item caused by the damage to the vehicle as a whole is calculated according to preset rules. Subsequently, the first evaluation sub-items and the third evaluation sub-items corresponding to each damage in the optimized damage imprint are fused through a preset integration operation, such as adding the two or weighted averaging, to obtain the final damage assessment value representing the comprehensive impact of the damage. Then, the final damage assessment values ​​of all damages are summarized to obtain a total damage assessment value. At the same time, the second assessment sub-items of all vehicle systems are summarized to obtain a total vehicle condition assessment value. Finally, these two total assessment values ​​are processed through a preset joint assessment function, which may be a weighted summation model or a more complex decision model, to generate the final intelligent assessment result. This result may include structured information such as the overall vehicle score, sub-item comments, and maintenance suggestions.

[0113] For example, following the specific implementation of the previous step, for the target SUV vehicle, firstly, damage information is extracted from its optimized damage imprint. This imprint contains one piece of damage, labeled as paint repair trace, with an optimized severity level of L. Assume that after optimization in step 105, the level is adjusted from 2 to L, and the location is on the right front door. Secondly, the damage type value influence weight table is queried, and the weight of the paint repair trace is W1. Combined with the severity level L, the first evaluation sub-item value of this damage is calculated as S1. Simultaneously, the state vectors of three systems are extracted from the modulated vehicle condition imprint: engine system 0.026, transmission system 0.22, and bearing system 0.074. Next, the attenuation vector performance influence coefficient table is queried, and the second evaluation sub-item for the engine system is calculated as C_e, for the transmission system as C_t, and for the bearing system as C_b. Then, according to the damage system association mapping relationship, the paint repair trace is associated with the engine system. Based on this correlation and the current second evaluation item C_e of the engine system, the third evaluation item value of the damage is calculated as S3. Then, the first evaluation item S1 and the third evaluation item S3 of the damage are integrated and merged to obtain the final damage evaluation value F1. Then, assuming that the vehicle has only this one damage, the total damage evaluation value is F1, and the total vehicle condition evaluation value is the sum of C_e, C_t, and C_b, C_total. Finally, the total damage evaluation value F1 and the total vehicle condition evaluation value C_total are input into a preset joint evaluation function to generate the intelligent evaluation result of the vehicle. This result can be an evaluation document containing a comprehensive score, a vehicle condition report, and a damage report.

[0114] Figure 2 This application provides a schematic diagram of the structure of a deep learning-based intelligent evaluation system for used cars, as shown below. Figure 2 As shown, the system includes: The acquisition module 21 is used to acquire the depth image sequence of the target vehicle, historical maintenance data records, the service life recorded on the vehicle nameplate, the real-time total mileage data reported by the on-board diagnostic system, and the multi-element vibration and temperature signals collected in real time by the sensor group arranged in the core transmission system of the vehicle. The depth image sequence includes the interior and exterior trim and chassis areas. The identification module 22 is used to identify and fuse the damage imprint of the target vehicle based on the depth image sequence and the historical maintenance data record. The damage imprint includes the spatial location distribution of the damage, the severity level associated with the depth, and the damage type label. Extraction module 23 is used to extract the vehicle condition imprint of the target vehicle based on the service life, the real-time total mileage data and the multi-element vibration and temperature signals. The vehicle condition imprint reflects the current operating status attenuation level of each core system of the vehicle. The modulation module 24 is used to perform directional weighted modulation on the state attenuation level of the corresponding vehicle system in the vehicle condition imprint based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, so as to generate the modulated vehicle condition imprint. Optimization module 25 is used to optimize the severity level associated with the damage type label in the damage imprint using the modulated vehicle condition imprint, and generate an optimized damage imprint. The generation module 26 is used to collaboratively generate an intelligent evaluation result for the target vehicle based on the optimized damage imprint and the modulated vehicle condition imprint.

[0115] Figure 2 The aforementioned deep learning-based intelligent evaluation system for used cars can perform... Figure 1 The implementation principle and technical effects of the deep learning-based intelligent evaluation method for used cars described in the illustrated embodiment will not be repeated here. The specific methods by which each module and unit of the deep learning-based intelligent evaluation system for used cars performs its operations have been described in detail in the embodiments related to this method, and will not be elaborated upon here.

[0116] In one possible design, Figure 2 The deep learning-based intelligent evaluation system for used cars shown in the embodiment can be implemented as a computing device, such as... Figure 3 As shown, the computing device may include a storage component 31 and a processing component 32; The storage component 31 stores one or more computer instructions, wherein the one or more computer instructions are invoked and executed by the processing component 32.

[0117] The processing component 32 is used for the above Figure 1 The embodiment describes a deep learning-based intelligent evaluation method for used cars.

[0118] The processing component 32 may include one or more processors to execute computer instructions to complete all or part of the steps in the above-described method. Alternatively, the processing component may be implemented as one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to perform the above-described method.

[0119] Storage component 31 is configured to store various types of data to support operations at the terminal. The storage component can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0120] Of course, computing devices may also include other components, such as input / output interfaces, display components, communication components, etc.

[0121] Input / output interfaces provide interfaces between processing components and peripheral interface modules, which can be output devices, input devices, etc.

[0122] The communication components are configured to facilitate wired or wireless communication between computing devices and other devices.

[0123] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of this application, and are not intended to limit them. Although this application 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 this application.

Claims

1. A deep learning-based intelligent evaluation method for used cars, characterized in that, include: The system acquires a depth image sequence of the target vehicle, historical maintenance data records, the service life recorded on the vehicle nameplate, real-time total mileage data reported by the on-board diagnostic system, and multi-dimensional vibration and temperature signals collected in real time by a sensor group deployed in the core transmission system of the vehicle. The depth image sequence includes the interior and exterior trim and chassis areas. Based on the depth image sequence and the historical maintenance data record, the damage imprint of the target vehicle is identified and fused. The damage imprint includes the spatial location distribution of the damage, the severity level associated with the depth, and the damage type label. Based on the service life, the real-time total mileage data, and the multi-element vibration and temperature signals, the vehicle condition imprint of the target vehicle is extracted. The vehicle condition imprint reflects the current operating status attenuation level of each core system of the vehicle. Based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, the state decay level of the corresponding vehicle system in the vehicle condition imprint is directionally weighted and modulated to generate the modulated vehicle condition imprint. Using the modulated vehicle condition imprint, the severity level associated with the damage type label in the damage imprint is optimized by state feedback to generate an optimized damage imprint. Based on the optimized damage imprint and the modulated vehicle condition imprint, an intelligent evaluation result for the target vehicle is generated collaboratively.

2. The method according to claim 1, characterized in that, Acquire depth image sequences of the target vehicle, historical maintenance data records, the service life recorded on the vehicle nameplate, real-time total mileage data reported by the on-board diagnostic system, and multi-dimensional vibration and temperature signals collected in real time by sensor arrays deployed in the vehicle's core drivetrain, including: By using multiple depth cameras positioned around the target vehicle, images are captured from different preset angles of multiple exterior surfaces, interior areas, and the structure under the lifted vehicle chassis, resulting in a sequence of depth images. Retrieve and download historical maintenance data records in chronological order from the de-identified vehicle maintenance blockchain network associated with the target vehicle. The historical maintenance data records include maintenance item descriptions and parts replacement entries. Read the production date on the vehicle nameplate of the target vehicle and calculate the total number of days of use from the production date to the evaluation date, as the service life; By accessing the on-board diagnostic system interface of the target vehicle, the cumulative mileage value reported by the engine control unit is read from the non-volatile memory of the on-board diagnostic system as real-time total mileage data. A triaxial vibration sensor and a temperature sensor are installed in the core transmission system of the target vehicle. When the vehicle is idling and under preset load conditions, the triaxial acceleration values ​​of the triaxial vibration sensor at multiple time points and the temperature measurement values ​​of the temperature sensor at the corresponding time points are collected and recorded simultaneously to form multi-dimensional vibration and temperature signals.

3. The method according to claim 1, characterized in that, Based on the depth image sequence and the historical maintenance data records, the damage marks of the target vehicle are identified and fused, including: Based on the depth image sequence, the visual structural information and three-dimensional geometric information of the surface of each component of the target vehicle are parsed out; Based on the repair project descriptions and parts replacement entries recorded in chronological order in the historical repair data records, the component location identifiers and repair action descriptions associated with each repair event are parsed out; The visual structural information, the three-dimensional geometric information, the parsed component location identifiers, and the maintenance action descriptions are correlated to identify candidate areas with maintenance traces on the physical structure of the target vehicle. Based on the three-dimensional point cloud data of the corresponding position in the depth image sequence, the specific spatial location and contour range of the candidate region in the vehicle's three-dimensional coordinate system are determined as the spatial location distribution of the damage. The severity level of the depth association is determined by combining the surface depression depth contained in the three-dimensional geometric information of the candidate region with the maintenance nature recorded in the associated maintenance action description. The damage type label is determined by combining the texture and color anomaly patterns presented by the visual structure information of the candidate region with the professional terminology used in the associated repair project description. The spatial location distribution of damage corresponding to the candidate region, the severity level associated with depth, and the damage type label are combined to obtain the damage imprint of the target vehicle.

4. The method according to claim 1, characterized in that, Based on the stated years of use, the real-time total mileage data, and the multi-element vibration and temperature signals, the vehicle condition imprint of the target vehicle is extracted. This imprint reflects the current operational degradation level of each core system of the vehicle, including: Based on the time decay curve and mileage decay curve determined by experience data in the vehicle industry, and combined with the service life and the real-time total mileage data, the basic aging coefficient is calculated. From the multi-element vibration and temperature signals, a first set of vibration and temperature signals collected under the target vehicle idling condition and a second set of vibration and temperature signals collected under the preset load condition are separated. The vibration signal in the first set of vibration and temperature signals is analyzed to obtain the first vibration spectrum of the engine in the idling state, and the temperature signal in the first set of vibration and temperature signals is analyzed to obtain the first temperature value of the engine in the idling state. The vibration signal in the second set of vibration and temperature signals is analyzed to obtain the second vibration spectrum of the transmission system under load. The temperature signal in the second set of vibration and temperature signals is analyzed to obtain the second temperature value of the transmission system under load. For the engine system, the spectral energy associated with the engine ignition frequency is determined from the first vibration spectrum, and the deviation from the engine standard operating temperature is determined from the first temperature value. By combining the spectral energy and the deviation from the engine standard operating temperature, an engine abnormality index is obtained. For the transmission system, the spectral energy associated with the gear meshing frequency is determined from the second vibration spectrum, and the abnormal rise in transmission oil temperature is determined from the second temperature value. Combining the spectral energy and the abnormal rise in transmission oil temperature, the transmission abnormality index is obtained. For the drive shaft bearing system, the spectral energy associated with the bearing characteristic fault frequency is determined from the second vibration spectrum, and combined with the temperature value of the bearing housing position in the second temperature value, the bearing abnormality index is obtained; Using the aforementioned basic aging coefficient, the engine abnormality index, the transmission abnormality index, and the bearing abnormality index are weighted and modulated respectively to generate system state decay vectors corresponding to the engine system, transmission system, and drive shaft bearing system, respectively. The system state decay vectors of the engine system, the transmission system, and the drive shaft bearing system are combined to form a vehicle condition imprint that reflects the current operating state decay level of each core system of the vehicle.

5. The method according to claim 1, characterized in that, Based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, the state decay level of the corresponding vehicle system in the vehicle condition imprint is directionally weighted and modulated to generate a modulated vehicle condition imprint, including: The damage imprint is analyzed to obtain the spatial distribution of the damage and the severity level associated with the depth corresponding to the spatial distribution; According to the preset mapping table, the physical component locations indicated by the spatial distribution of the damage are mapped to the target vehicle system affected by the physical component locations. The target vehicle system is the engine system, transmission system, or drive shaft bearing system that constitutes the vehicle condition imprint. Based on the severity level associated with the depth corresponding to the spatial location distribution of the damage, the impact factor value is calculated using a preset conversion rule. Find all damages mapped to the target vehicle system and obtain all impact factor values ​​corresponding to the damages; All influencing factor values ​​mapped to the same target vehicle system are processed through a preset aggregation function to generate a comprehensive modulation coefficient. The system state attenuation vector corresponding to the target vehicle system in the vehicle condition imprint is multiplied by the calculated comprehensive modulation coefficient to obtain the modulated system state attenuation vector of the target vehicle system. After performing the multiplication operation on the system state attenuation vectors of all target vehicle systems in the vehicle condition imprint, all modulated system state attenuation vectors are recombined to generate the modulated vehicle condition imprint.

6. The method according to claim 1, characterized in that, Using the modulated vehicle condition imprint, the severity level associated with the damage type label in the damage imprint is optimized using state feedback to generate an optimized damage imprint, including: The damage imprint is analyzed to obtain damage type labels and the severity level of the deep association corresponding to each damage type label; Based on the preset damage type system impact mapping table, the target vehicle system mainly affected by each damage type label is determined, and the target vehicle system is the engine system, transmission system or drive shaft bearing system that constitutes the modulated vehicle condition imprint. From the modulated vehicle condition imprint, extract the modulated system state decay vector corresponding to each target vehicle system; The modulated system state attenuation vector corresponding to the target vehicle system mainly affected by the damage type label is used to calculate the system influence factor through a preset vector influence factor conversion function. All system impact factors corresponding to the damage type labels are merged into an overall state optimization coefficient through a preset aggregation operation. The severity level of the depth association corresponding to the damage type label is combined with the overall state optimization coefficient, and the optimized severity level of the damage type label is calculated through a preset severity level correction rule. The severity level associated with each damage type label in the damage imprint is updated to the optimized severity level, while keeping the spatial distribution of damage and damage type labels in the damage imprint unchanged, so as to obtain the optimized damage imprint.

7. The method according to claim 1, characterized in that, Based on the optimized damage imprint and the modulated vehicle condition imprint, an intelligent evaluation result for the target vehicle is collaboratively generated, including: From the optimized damage imprint, damage type labels and the optimized severity level corresponding to the damage type labels, as well as the spatial distribution of the damage, are extracted. Based on the preset damage type value influence weight table, the first evaluation sub-item of the damage on the overall condition of the target vehicle is calculated. From the modulated vehicle condition imprint, extract the modulated system state attenuation vectors corresponding to the engine system, transmission system and drive shaft bearing system respectively, and calculate the second evaluation sub-item of the vehicle system on the overall vehicle condition according to the preset attenuation vector performance influence coefficient table. Based on the preset damage system association mapping relationship, establish the influence association between the damage indicated by the damage type label and the engine system, transmission system and drive shaft bearing system; Based on the association between the damage type label and the impact on each vehicle system, and the current second assessment item for each vehicle system, calculate the third assessment item for the damage's impact on the overall vehicle condition. The first evaluation sub-item corresponding to each damage type label in the optimized damage imprint is fused with the third evaluation sub-item through an integration operation to obtain the final damage evaluation value of the damage type label. The final damage assessment values ​​of all damage type labels are summarized, and the second assessment sub-items of all vehicle systems are also summarized. The results are then processed through a preset joint assessment function to generate an intelligent assessment result.

8. A deep learning-based intelligent evaluation system for used cars, characterized in that, include: The acquisition module is used to acquire the target vehicle's depth image sequence, historical maintenance data records, the service life recorded on the vehicle nameplate, the real-time total mileage data reported by the on-board diagnostic system, and multi-dimensional vibration and temperature signals collected in real time by the sensor group arranged in the vehicle's core transmission system. The depth image sequence includes the interior and exterior trim and chassis areas. The identification module is used to identify and fuse the damage imprint of the target vehicle based on the depth image sequence and the historical maintenance data record. The damage imprint includes the spatial location distribution of the damage, the severity level associated with the depth, and the damage type label. The extraction module is used to extract the vehicle condition imprint of the target vehicle based on the service life, the real-time total mileage data and the multi-dimensional vibration and temperature signals. The vehicle condition imprint reflects the current operating status attenuation level of each core system of the vehicle. The modulation module is used to perform directional weighted modulation on the state attenuation level of the corresponding vehicle system in the vehicle condition imprint based on the severity level of the spatial location distribution and depth correlation of the damage in the damage imprint, so as to generate the modulated vehicle condition imprint. The optimization module is used to optimize the severity level associated with the damage type label in the damage imprint using the modulated vehicle condition imprint, and generate an optimized damage imprint. The generation module is used to collaboratively generate an intelligent evaluation result for the target vehicle based on the optimized damage imprint and the modulated vehicle condition imprint.

9. A computing device, characterized in that, It includes a processing component and a storage component; the storage component stores one or more computer instructions; the one or more computer instructions are invoked and executed by the processing component to implement a deep learning-based intelligent evaluation method for used cars as described in any one of claims 1 to 7.

10. A computer storage medium, characterized in that, The system contains a computer program that, when executed by a computer, implements a deep learning-based intelligent evaluation method for used cars as described in any one of claims 1 to 7.

Images