CN-122020251-A - Quality evaluation method, device and equipment for single-period vehicle-mounted CAN network data

CN122020251ACN 122020251 ACN122020251 ACN 122020251ACN-122020251-A

Abstract

The invention relates to the technical field of image processing, in particular to a quality evaluation method, a device and equipment of single-period vehicle-mounted CAN network data, which comprises the steps of acquiring a standard vehicle-mounted CAN network data set D1 and single-period vehicle-mounted CAN network data D2 to be evaluated; identifying a period starting CAN_ID of D1, carrying out standardization processing according to the period starting CAN_ID, extracting a plurality of monocycle D1 data segments and monocycle D2 data segments, converting the monocycle D1 data segments into color bars according to a coloring classification system to obtain a reference color bar graph data set D1sts, converting the monocycle D2 data segments into color bars to obtain a color bar graph D2st to be evaluated, calculating a feature reference library by using an unsupervised image feature extractor, calculating the similarity of feature vectors to be evaluated and feature mean vectors, and judging the quality grade of D2 according to the similarity. The invention CAN improve the accuracy and efficiency of quality evaluation of single-period vehicle-mounted CAN network data.

Inventors

XIA XIAOFENG
LI QIMIN
LI GUANGYI
SANG JUN
CAI BIN
HU CHUNQIANG
XIANG HONG

Assignees

重庆大学

Dates

Publication Date: 20260512
Application Date: 20260130

Claims (10)

1. The quality evaluation method of the single-period vehicle-mounted CAN network data is characterized by comprising the following steps of: acquiring a standard vehicle-mounted CAN network data set D1 and single-period vehicle-mounted CAN network data D2 to be evaluated; Identifying a period starting CAN_ID of a standard vehicle-mounted CAN network data set D1; According to the cycle starting CAN_ID, carrying out standardization processing on the standard vehicle-mounted CAN network data set D1 and the single-cycle vehicle-mounted CAN network data D2 to be evaluated, and extracting a plurality of single-cycle D1 data segments of the standard vehicle-mounted CAN network data set D1 and single-cycle D2 data segments of the single-cycle vehicle-mounted CAN network data D2 to be evaluated; Converting a plurality of single-period D1 data segments into color bar graphs according to a pre-constructed CAN_ID coloring classification system to obtain a reference color bar graph data set D1sts, and converting single-period D2 data segments into color bar graphs to obtain a color bar graph D2st to be evaluated; performing feature extraction on the reference color bar graph dataset D1sts by using an unsupervised image feature extractor to obtain a reference feature vector, and calculating a feature mean vector of the reference feature vector to obtain a feature reference library; And extracting the feature vector to be evaluated of the color bar graph D2st to be evaluated by using an unsupervised image feature extractor, calculating the similarity between the feature vector to be evaluated and the feature mean value vector in the feature reference library, and judging the quality grade of the single-period vehicle-mounted CAN network data D2 to be evaluated according to the similarity.
2. The quality evaluation method of single-cycle vehicle CAN network data according to claim 1, wherein the cycle start can_id of the identification standard vehicle CAN network data set D1 includes: Carrying out statistical analysis on CAN_ID fields of a standard vehicle-mounted CAN network data set D1 by using a preset data analysis script to obtain the total number of CAN_ID fields, calculating the occurrence times and time interval standard deviations of each CAN_ID in the total number of CAN_ID fields in different time windows, and constructing periodic CAN_ID field data according to the occurrence times and the time interval standard deviations; And determining the CAN_ID corresponding to the period starting position in the period CAN_ID field data as the period starting CAN_ID.
3. The method for evaluating the quality of the monocycle vehicle-mounted CAN network data according to claim 1, wherein the normalizing the standard vehicle-mounted CAN network data set D1 and the monocycle vehicle-mounted CAN network data D2 to be evaluated according to the cycle start can_id and extracting a plurality of monocycle D1 data segments of the standard vehicle-mounted CAN network data set D1 and monocycle D2 data segments of the monocycle vehicle-mounted CAN network data D2 to be evaluated includes: the CAN_ID field and the DLC field in the standard vehicle-mounted CAN network data set D1 are reserved, and redundant fields except the CAN_ID field and the DLC field in the standard vehicle-mounted CAN network data set D1 are deleted; Dividing a standard vehicle-mounted CAN network data set D1 into a plurality of independent single-period D1 data segments according to a period starting CAN_ID; The CAN_ID field and the DLC field in the single-period vehicle-mounted CAN network data D2 to be evaluated are reserved, and redundant fields except the CAN_ID field and the DLC field in the single-period vehicle-mounted CAN network data D2 to be evaluated are deleted; and identifying a cycle starting CAN_ID in the single-cycle vehicle-mounted CAN network data D2 to be evaluated, positioning the single-cycle starting position according to the cycle starting CAN_ID in the single-cycle vehicle-mounted CAN network data D2 to be evaluated, and removing data before the first cycle starting CAN_ID and after the second cycle starting CAN_ID to obtain a complete single-cycle D2 data segment.
4. The method for evaluating the quality of the monocycle vehicle-mounted CAN network data according to claim 1, wherein before the standardized processing is performed on the standard vehicle-mounted CAN network data set D1 and the monocycle vehicle-mounted CAN network data D2 to be evaluated according to the cycle start CAN_ID, the method further comprises the steps of performing data inspection on the monocycle vehicle-mounted CAN network data D2 to be evaluated, and verifying the field integrity and the CAN_ID validity of the monocycle vehicle-mounted CAN network data D2 to be evaluated.
5. The quality evaluation method of single-cycle vehicle-mounted CAN network data according to claim 1, wherein the data checking of the single-cycle vehicle-mounted CAN network data D2 to be evaluated, verifying field integrity and can_id validity of the single-cycle vehicle-mounted CAN network data D2 to be evaluated, comprises: Traversing data records of single-period vehicle-mounted CAN network data D2 to be evaluated, checking whether each data record contains a CAN_ID field and a DLC field, counting the number of data records with the CAN_ID field or the DLC field missing, and if any data record has the CAN_ID field or the DLC field missing, judging that the D2 does not pass the check, and generating an evaluation conclusion of incomplete data field and unqualified quality; And checking at a second stage, according to the CAN_ID set in the standard vehicle-mounted CAN network data set D1, verifying whether all CAN_IDs in the single-period vehicle-mounted CAN network data D2 to be evaluated are in the effective range of the CAN_ID set, wherein verifying whether all CAN_IDs in the single-period vehicle-mounted CAN network data D2 to be evaluated are in the effective range of the CAN_ID set comprises comparing all CAN_IDs of the single-period vehicle-mounted CAN network data D2 to be evaluated with the CAN_ID set of the standard vehicle-mounted CAN network data set D1 through set operation, and if the CAN_IDs which do not belong to the standard vehicle-mounted CAN network data set D1 exist in the single-period vehicle-mounted CAN network data D2 to be evaluated, judging that the D2 does not pass the check, and generating an evaluation conclusion of invalid CAN_ID and unqualified quality.
6. The quality evaluation method of monocycle vehicle CAN network data according to claim 1, wherein the converting the plurality of monocycle D1 data segments into color bars to obtain the reference color bar graph data set D1sts and converting the monocycle D2 data segments into color bars to obtain the color bar graph D2st to be evaluated according to the pre-constructed can_id coloring classification system comprises: Inquiring CAN_ID types, CAN_ID ranges, occurrence periods and transmission frequencies of CAN_IDs in the CAN_ID coloring classification system in the single-period D1 data segment, distributing color bar preset colors to the CAN_IDs in the single-period D1 data segment, and distributing color bar lengths according to DLC field lengths in the single-period D1 data segment; inquiring CAN_ID types, CAN_ID ranges, occurrence periods and transmission frequencies of CAN_IDs in the CAN_ID coloring classification system in the single-period D2 data segment, distributing color bar preset colors to the CAN_IDs in the single-period D2 data segment, and distributing color bar lengths according to DLC field lengths in the single-period D2 data segment; The color bar graph comprises a plurality of color bars in the longitudinal direction, corresponding color bar heights and transverse color bar lengths, wherein the plurality of color bars in the longitudinal direction correspond to the maximum data amount of each single-period data segment, the height of each color bar is fixed, the color bar length is set in the transverse direction of the color bar graph according to the DLC field length of each single-period data segment, black is used for supplementing if the DLC field length is smaller than 8, the color of the color bar is determined according to a pre-constructed CAN_ID coloring classification system, and after the construction of a reference color bar graph dataset D1sts and a color bar graph D2st to be evaluated is completed, unified size scaling processing is carried out on the reference color bar graph dataset D1sts and the color bar graph D2st to be evaluated, so that a standard-size reference color bar graph dataset D1sts and a standard-size color bar graph D2st to be evaluated are obtained.
7. The quality evaluation method of single-cycle vehicle CAN network data according to claim 1, wherein the feature extraction of the reference color bar graph dataset D1sts by the unsupervised image feature extractor, the construction of the unsupervised image feature extractor, comprises: Step 1, loading a pre-trained MobileNetV backbone network, and removing a top-level classification head module of the MobileNetV backbone network; step 2, adding a global average pooling layer after MobileNetV backbone network output, and freezing weight parameters of the MobileNetV backbone network; step 3, adopting a full-connection layer with a linear activation function as a characteristic dimension reduction module; and (3) obtaining the unsupervised image feature extractor through the model processing in the step (1-3).
8. A quality evaluation device of single-cycle vehicle-mounted CAN network data, characterized in that the device is used for realizing the quality evaluation method of single-cycle vehicle-mounted CAN network data according to any one of claims 1 to 7, and the device comprises: the data acquisition module is used for acquiring a standard vehicle-mounted CAN network data set D1 and single-period vehicle-mounted CAN network data D2 to be evaluated; The color bar graph conversion module is used for identifying a cycle starting CAN_ID of the standard vehicle-mounted CAN network data set D1, carrying out standardization processing on the standard vehicle-mounted CAN network data set D1 and the single-cycle vehicle-mounted CAN network data D2 to be evaluated according to the cycle starting CAN_ID, extracting a plurality of single-cycle D1 data segments of the standard vehicle-mounted CAN network data set D1 and single-cycle D2 data segments of the single-cycle vehicle-mounted CAN network data D2 to be evaluated; The quality evaluation module is used for carrying out feature extraction on the reference color bar graph dataset D1sts by using the non-supervision image feature extractor to obtain a reference feature vector and calculating a feature average value vector of the reference feature vector to obtain a feature reference library, extracting the feature vector to be evaluated of the color bar graph D2st to be evaluated by using the non-supervision image feature extractor and calculating the similarity between the feature vector to be evaluated and the feature average value vector in the feature reference library, and judging the quality grade of the single-period vehicle-mounted CAN network data D2 to be evaluated according to the similarity.
9. An electronic device, the electronic device comprising: at least one processor, and A memory communicatively coupled to the at least one processor, wherein, The memory stores a computer program executable by the at least one processor to enable the at least one processor to perform the quality evaluation method of single cycle on-board CAN network data as claimed in any one of claims 1 to 7.
10. A computer-readable storage medium storing a computer program, wherein the computer program, when executed by a processor, implements the quality evaluation method of single-cycle on-vehicle CAN network data according to any one of claims 1 to 7.

Description

Quality evaluation method, device and equipment for single-period vehicle-mounted CAN network data Technical Field The invention belongs to the technical field of image processing, and particularly relates to a quality evaluation method, device and equipment for single-period vehicle-mounted CAN network data. Background With the rapid development of intelligent networking automobile technology, a vehicle-mounted controller area network (CAN, controller Area Network) is used as a nerve hub for internal communication of an automobile, and the generated data becomes a core resource for vehicle state monitoring, fault diagnosis and automatic driving algorithm training. In particular, the quality evaluation of the data generated by the vehicle-mounted CAN network is directly related to the reliability and safety of downstream data application. At present, in the field of data quality evaluation generated by a vehicle-mounted CAN network, the prior art generally relates to three main links of data processing, quality evaluation and deep learning model application, but a plurality of defects still exist in practical application. In the aspect of vehicle-mounted CAN network data processing, the prior art is mostly based on SQL, python and other general tools for basic analysis, including data screening, field extraction, redundant data rejection and other operations. Such methods, while capable of preliminary sorting of data and speculating data cycle attributes by identifying the distribution characteristics of can_ids, lack standardized procedures for extraction of "monocycle" data. Specifically, the identification method of the periodic distribution of the CAN_ID in the prior art is rough, and is difficult to accurately position the 'CYCLESTART CAN _ID' (cycle start ID), so that the segmentation interception of single-cycle data is not accurate enough. Meanwhile, due to the lack of unified data redundancy deletion and field screening specifications, key information is extremely easy to lose or invalid data is extremely easy to remain in the processing process, and the data base of subsequent quality evaluation is further affected. In the concrete implementation of data quality evaluation, the prior art mainly adopts two types of methods, namely a manual evaluation method based on rules, namely whether the data reach standards or not is checked manually by presetting the rules of CAN_ID integrity, transmission stability and the like, and a method based on simple statistical analysis, and judgment is carried out by calculating indexes such as the deletion rate, the abnormal value duty ratio and the like. However, the method has obvious limitations when facing large-scale data generation, namely, manual evaluation is seriously dependent on expert experience, subjectivity is strong, efficiency is extremely low, and the quick screening requirement cannot be met, while simple statistical analysis can only cover quality indexes of the surface, inherent reliability and consistency of the data cannot be deeply mined, and evaluation dimension is too single. In recent years, some prior art attempts to introduce deep learning models to assist in data quality evaluation have been made, but a supervised learning mode is often adopted, i.e. training is performed by relying on a large number of data samples with marked quality levels. The method has poor adaptability in the vehicle-mounted CAN network data scene, on one hand, the vehicle-mounted CAN network data lacks a unified quality marking standard, the cost for obtaining a large amount of high-quality marking data is extremely high, on the other hand, the general deep learning model is not optimized for the specificity (such as strict periodicity, time sequence, strong relevance of CAN_ID and data length DLC fields and the like) of the vehicle-mounted CAN network data, and the characteristic extraction pertinence is insufficient, so that the quality evaluation accuracy and efficiency of the final single-period vehicle-mounted CAN network data are low. Disclosure of Invention The invention provides a quality evaluation method, device and equipment for single-period vehicle-mounted CAN network data, which CAN improve the accuracy and efficiency of the quality evaluation of the single-period vehicle-mounted CAN network data. In order to achieve the above object, the present invention provides a quality evaluation method for single-cycle vehicle-mounted CAN network data, including: acquiring a standard vehicle-mounted CAN network data set D1 and single-period vehicle-mounted CAN network data D2 to be evaluated; Identifying a period starting CAN_ID of a standard vehicle-mounted CAN network data set D1; According to the cycle starting CAN_ID, carrying out standardization processing on the standard vehicle-mounted CAN network data set D1 and the single-cycle vehicle-mounted CAN network data D2 to be evaluated, and extracting a plurality of single-cycle D1 data segments of the sta