CN111858342A - Fuzzy test data generation method based on intelligent traffic image input feature recognition - Google Patents
Fuzzy test data generation method based on intelligent traffic image input feature recognition Download PDFInfo
- Publication number
- CN111858342A CN111858342A CN202010712766.8A CN202010712766A CN111858342A CN 111858342 A CN111858342 A CN 111858342A CN 202010712766 A CN202010712766 A CN 202010712766A CN 111858342 A CN111858342 A CN 111858342A
- Authority
- CN
- China
- Prior art keywords
- feature
- data
- intelligent traffic
- input
- test
- Prior art date
- Legal status (The legal status is an assumption and is not a legal conclusion. Google has not performed a legal analysis and makes no representation as to the accuracy of the status listed.)
- Pending
Links
Images
Classifications
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3684—Test management for test design, e.g. generating new test cases
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06F—ELECTRIC DIGITAL DATA PROCESSING
- G06F11/00—Error detection; Error correction; Monitoring
- G06F11/36—Preventing errors by testing or debugging software
- G06F11/3668—Software testing
- G06F11/3672—Test management
- G06F11/3688—Test management for test execution, e.g. scheduling of test suites
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/043—Architecture, e.g. interconnection topology based on fuzzy logic, fuzzy membership or fuzzy inference, e.g. adaptive neuro-fuzzy inference systems [ANFIS]
-
- G—PHYSICS
- G06—COMPUTING; CALCULATING OR COUNTING
- G06N—COMPUTING ARRANGEMENTS BASED ON SPECIFIC COMPUTATIONAL MODELS
- G06N3/00—Computing arrangements based on biological models
- G06N3/02—Neural networks
- G06N3/04—Architecture, e.g. interconnection topology
- G06N3/045—Combinations of networks
Landscapes
- Engineering & Computer Science (AREA)
- Theoretical Computer Science (AREA)
- Physics & Mathematics (AREA)
- General Physics & Mathematics (AREA)
- General Engineering & Computer Science (AREA)
- Software Systems (AREA)
- Computing Systems (AREA)
- Health & Medical Sciences (AREA)
- Data Mining & Analysis (AREA)
- Evolutionary Computation (AREA)
- General Health & Medical Sciences (AREA)
- Molecular Biology (AREA)
- Biophysics (AREA)
- Biomedical Technology (AREA)
- Artificial Intelligence (AREA)
- Mathematical Physics (AREA)
- Life Sciences & Earth Sciences (AREA)
- Computational Linguistics (AREA)
- Quality & Reliability (AREA)
- Computer Hardware Design (AREA)
- Fuzzy Systems (AREA)
- Mathematical Analysis (AREA)
- Mathematical Optimization (AREA)
- Pure & Applied Mathematics (AREA)
- Computational Mathematics (AREA)
- Automation & Control Theory (AREA)
- Image Analysis (AREA)
Abstract
A fuzzy test data generation method based on intelligent traffic image input feature recognition is characterized in that feature analysis is carried out on data in an intelligent traffic scene, image data features in the intelligent traffic scene are described, and test image data generation aiming at the intelligent traffic scene is achieved by applying pixel disturbance with traffic semantics. The invention comprises three components: the system comprises an input feature extraction module, an image data variation module and a data diversity evaluation module. The input of the invention is intelligent traffic image data, feature extraction is carried out based on the input data, a candidate feature data set is generated, pixel disturbance information is added, and a fuzzy test case is generated. The invention has the following beneficial effects: by further analyzing the characteristics of the input data, test data with more semantics under an intelligent traffic scene is generated, and the problem of insufficient intelligent traffic test cases is solved.
Description
Technical Field
The invention belongs to the field of software testing, and particularly relates to image test data generation in an intelligent traffic scene. The method comprises the steps of performing feature analysis on data in an intelligent traffic scene, depicting image data features in the intelligent traffic scene, and applying pixel disturbance with traffic semantics to generate test image data aiming at the intelligent traffic scene.
Background
In recent years, intelligent software technology is rapidly developed and applied to the field of traffic driving, and a top-down multilayer architecture of an intelligent traffic system is penetrated. Unlike traditional software, the intelligent transportation industry is still in the beginning stage at present, and the existing technical standards are still in the development process. The problems of large-scale test verification, installation authentication and the like still need to be solved and promoted. The test data is effective content for verifying the quality of the system, however, the intelligent traffic scene is high in complexity, and a large amount of high-quality data needs a large amount of time and cost for acquisition and processing. Therefore, test data generation for intelligent traffic scenarios is referred to as an important hotspot problem.
In fact, the fuzzy testing technology is a technical means for effectively testing the system security, and can realize automatic generation and automatic testing of large-scale data. However, the input types in the intelligent traffic scene are various, and the structure is complex, so that the original semantic information of the test data disturbance generation method implemented in the traditional software test process may be damaged when the image data is processed.
Based on the background, the invention further analyzes the semantic features contained in the input and hopes to generate more semantic test data in an intelligent traffic scene. Therefore, the input generation method based on the image input feature recognition is realized by constructing the feature vector on the image data according to the activation condition of the feature map based on the feature extraction characteristic of the convolutional layer commonly used in the image recognition network, and reversely calculating the region and the value of the feature vector on the original input through the feature map.
Disclosure of Invention
The invention aims to solve the problems that: the method for generating the test data of the intelligent traffic software system is not mature, and particularly relates to an intelligent traffic software model realized by utilizing a deep learning technology. Verification of an intelligent software model requires a large amount of test data to achieve adequate testing of the model. However, existing test data generation methods often ignore the maintenance of semantic information of input data, which makes test data generated in these ways often dispute and may not be able to embody the domain characteristics of an intelligent traffic scenario. The invention starts from data collected in an intelligent traffic scene, and provides a characteristic transformation method suitable for the intelligent traffic scene by combining with input characteristic analysis, so as to realize the generation of new test data.
The technical scheme of the invention is as follows: a fuzzy test data generation method based on intelligent traffic image input feature recognition is characterized in that extraction is carried out according to features of training data, a candidate feature list set is constructed according to the features, and then test data generation through feature variation is achieved. The generation method comprises the following three modules/steps:
1) an input feature extraction module: the input feature extraction realized by the module mainly comprises two aspects: firstly, carrying out feature point recognition on individuals in training data based on a feature recognition algorithm to obtain a feature candidate set independent of a model and confidence coefficients of all feature points; and secondly, the feature graph after the first layer of convolution input in the deep learning model is used as the feature representation input in the model space.
2) An image data mutation module: the testing of the deep learning operator depends on a large amount of tests on the operator, and the module utilizes a variation-based fuzzy testing technology and introduces a highly customized intelligent traffic scene variation method to generate a large amount of effective test data based on the existing test data. The variation method further solves the value of the feature combination in the input space by carrying out disturbance variation on the feature combination identified in the previous image input, so as to realize the generation of test data.
3) A data diversity evaluation module: in order to ensure the diversity of the test data obtained by the image data mutation method, i.e. the test data are uniformly distributed in the input space and have no potential data bias, the method implements the measurement of the diversity of the test data based on the coverage condition of the feature space. Meanwhile, the method can guide the generation direction of the test data, and selects the features which are not covered at present to carry out variation.
The invention is characterized in that:
1. in the generation of intelligent transportation software test data, the technology for generating fuzzy test data based on image input characteristics is firstly provided
2. The first layer of convolution output characteristic value in the model is used as the input characteristic representation of the model space for the first time
3. First-pass feature coverage as an evaluation criterion and guide direction for test data diversity
Drawings
Fig. 1 is a general flow chart of the implementation of the present invention.
Figure 2 is a flow chart of key step 2.
Figure 3 is a flow chart of key step 2.
Detailed Description
The embodiments of the present invention are described below with reference to specific examples, and other advantages and effects of the present invention will be readily apparent to those skilled in the art from the disclosure of the present specification.
The method mainly adopts a fuzzy test technology for generating image test data, and relates to specific key technologies such as an image feature identification technology, a convolution feature identification technology, a variation method selection technology, a data variation technology and a diversity measurement technology.
1. Intelligent traffic input feature candidate set construction
In the invention, a model-independent feature recognition technology is combined with a feature map recognized by a first layer convolution kernel of a model structure to serve as a construction basic method of an intelligent traffic input feature candidate set. The model-independent image feature recognition technology is a general feature recognition technology in the image field, such as SIFT, HOG and the like. And the feature map output by the first layer of convolution kernel of the model structure contains specific semantic features in the intelligent traffic scene. A group of candidate feature sets are respectively identified through the two feature identification methods, and the union set of the candidate feature sets is the intelligent traffic input feature candidate set constructed by the project.
2. Initial data selection
The initial data selection strategy is used in the invention to further improve the execution efficiency of the fuzz test. During each iteration of the fuzzy test, the fuzzy test engine selects specific original data from the initial data set for variant data generation. The baseline for initial data selection is chosen randomly, i.e., one input at a time is drawn from the seed set by a random function. In the method, constraint solution is carried out by taking feature difference maximization as an object through a feature label covered by each input, and initial image data required by the iteration is selected.
3. Selection of variant methods
The method realizes the mutation of the initial data by using a mutation method aiming at the characteristics of the intelligent traffic field. The image generation method of the existing facies intelligent software directly modifies pixel points based on original input to generate new test data. The selection of different variation methods also has a direct connection to the effect of the current fuzz test. In the method, an appropriate mutation method in the current round is selected by the MC/MC method based on the performance of the mutation method in the historical execution process.
4. Test data generation
In order to enable the generated image data to have more real semantics, the method is different from the existing modification mode based on model gradient or image pixels, and the feature transformation of the test input is implemented on the basis of the feature candidate set constructed in the previous step. And reversely solving the actual value of the characteristic in the input space according to the characteristic result so as to construct new test data.
5. Data diversity assessment
In order to ensure the sufficiency and diversity of generated data, the invention further realizes the diversity of the data by adopting a characteristic coverage method. When testing deep learning models in the intelligent transportation field such as an automatic driving model, the existing method based on neuron coverage cannot verify whether test data is sufficient in the field. The method is used for evaluating the diversity of the test data by calculating the feature coverage rate of the constructed intelligent traffic field feature population and the feature combination covered by the current test data.
In the present embodiment, the intelligent transportation data set adopts the current open source automatic driving data set, including Udacity, RobotCar, Cityscape. And a plurality of deep learning models are constructed. In the feature extraction part, SIFT algorithm is adopted to carry out model-independent feature extraction, and the first layer convolution output in the network is used as the identified feature of the model space. Candidate features are uniformly selected from the two feature sets to carry out feature variation, only one feature item is disturbed each time, other features are not affected, and therefore new test data are generated. The measure of feature diversity will feed back in real time the coverage of the current newly generated test data over all feature options.
Claims (4)
1. A fuzzy test data generation method based on intelligent traffic image input feature recognition is characterized in that feature analysis is carried out on data in an intelligent traffic scene, image data features in the intelligent traffic scene are described, a candidate feature data set is constructed, a mutation operator is selected to generate a fuzzy test case by applying pixel disturbance with traffic semantics, test image data generation aiming at the intelligent traffic scene is achieved, and finally diversity evaluation is carried out on the generated case.
2. The method for constructing the candidate feature data set according to claim 1, wherein a model-independent feature recognition technology is adopted to be combined with a feature map recognized by a first layer convolution kernel of a model structure to serve as a construction basic method of the intelligent traffic input feature candidate set, SIFT and HOG technologies are adopted for model-independent image feature recognition, and a deep learning convolution kernel technology is adopted for intelligent traffic feature extraction.
3. The fuzzy test case data generation method according to claim 1, wherein the fuzzy test case set is generated by reversely solving the actual value of the feature in the input space according to the feature result based on the constructed feature candidate data set, and performing the feature transformation of the test input.
4. The diversity evaluation method of claim 1, wherein the evaluation of the diversity of the test data is performed by calculating the feature coverage of the feature combination of the constructed intelligent transportation domain feature population and the feature covered by the current test data, and the measurement of the diversity of the feature can feed back the coverage of the current newly generated test data in all feature options in real time.
Priority Applications (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010712766.8A CN111858342A (en) | 2020-07-23 | 2020-07-23 | Fuzzy test data generation method based on intelligent traffic image input feature recognition |
Applications Claiming Priority (1)
| Application Number | Priority Date | Filing Date | Title |
|---|---|---|---|
| CN202010712766.8A CN111858342A (en) | 2020-07-23 | 2020-07-23 | Fuzzy test data generation method based on intelligent traffic image input feature recognition |
Publications (1)
| Publication Number | Publication Date |
|---|---|
| CN111858342A true CN111858342A (en) | 2020-10-30 |
Family
ID=72949330
Family Applications (1)
| Application Number | Title | Priority Date | Filing Date |
|---|---|---|---|
| CN202010712766.8A Pending CN111858342A (en) | 2020-07-23 | 2020-07-23 | Fuzzy test data generation method based on intelligent traffic image input feature recognition |
Country Status (1)
| Country | Link |
|---|---|
| CN (1) | CN111858342A (en) |
Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169421A (en) * | 2017-04-20 | 2017-09-15 | 华南理工大学 | A kind of car steering scene objects detection method based on depth convolutional neural networks |
| CN107609602A (en) * | 2017-09-28 | 2018-01-19 | 吉林大学 | A kind of Driving Scene sorting technique based on convolutional neural networks |
| CN108985194A (en) * | 2018-06-29 | 2018-12-11 | 华南理工大学 | A kind of intelligent vehicle based on image, semantic segmentation can travel the recognition methods in region |
| CN109582993A (en) * | 2018-06-20 | 2019-04-05 | 长安大学 | Urban transportation scene image understands and multi-angle of view gunz optimization method |
| CN111091049A (en) * | 2019-11-01 | 2020-05-01 | 东南大学 | Road surface obstacle detection method based on reverse feature matching |
-
2020
- 2020-07-23 CN CN202010712766.8A patent/CN111858342A/en active Pending
Patent Citations (5)
| Publication number | Priority date | Publication date | Assignee | Title |
|---|---|---|---|---|
| CN107169421A (en) * | 2017-04-20 | 2017-09-15 | 华南理工大学 | A kind of car steering scene objects detection method based on depth convolutional neural networks |
| CN107609602A (en) * | 2017-09-28 | 2018-01-19 | 吉林大学 | A kind of Driving Scene sorting technique based on convolutional neural networks |
| CN109582993A (en) * | 2018-06-20 | 2019-04-05 | 长安大学 | Urban transportation scene image understands and multi-angle of view gunz optimization method |
| CN108985194A (en) * | 2018-06-29 | 2018-12-11 | 华南理工大学 | A kind of intelligent vehicle based on image, semantic segmentation can travel the recognition methods in region |
| CN111091049A (en) * | 2019-11-01 | 2020-05-01 | 东南大学 | Road surface obstacle detection method based on reverse feature matching |
Non-Patent Citations (1)
| Title |
|---|
| 闫施违: "自动驾驶系统测试用例生成技术的实证研究", 《中国优秀硕士学位论文全文数据库 (工程科技Ⅱ辑)》 * |
Similar Documents
| Publication | Publication Date | Title |
|---|---|---|
| CN109657552B (en) | Vehicle type recognition device and method for realizing cross-scene cold start based on transfer learning | |
| CN110533086B (en) | Semi-automatic image data labeling method | |
| Kong et al. | Detect and locate: Exposing face manipulation by semantic-and noise-level telltales | |
| CN108537119B (en) | Small sample video identification method | |
| US20200401503A1 (en) | System and Method for Testing Artificial Intelligence Systems | |
| JP7282212B2 (en) | Method for learning deep learning network by AI and learning device using the same | |
| CN111767927A (en) | Lightweight license plate recognition method and system based on full convolution network | |
| CN109409257A (en) | A kind of video timing motion detection method based on Weakly supervised study | |
| CN109766259A (en) | A kind of classifier test method and system based on compound transformation relationship | |
| CN114333062B (en) | Pedestrian re-recognition model training method based on heterogeneous dual networks and feature consistency | |
| CN113033410B (en) | Domain generalization pedestrian re-recognition method, system and medium based on automatic data enhancement | |
| CN111858342A (en) | Fuzzy test data generation method based on intelligent traffic image input feature recognition | |
| CN117516937A (en) | Rolling bearing unknown fault detection method based on multi-mode feature fusion enhancement | |
| CN111858347A (en) | Metamorphic test method based on correctness of intelligent traffic model | |
| CN117036904A (en) | Attention-guided semi-supervised corn hyperspectral image data expansion method | |
| CN116580176A (en) | Vehicle-mounted CAN bus anomaly detection method based on lightweight network MobileViT | |
| CN116563636A (en) | Synthetic aperture radar image generation method and system | |
| Xu et al. | Cyclic Self-Training With Proposal Weight Modulation for Cross-Supervised Object Detection | |
| CN113298112B (en) | Integrated data intelligent labeling method and system | |
| CN115356599A (en) | Multi-mode urban power grid fault diagnosis method and system | |
| Xiang et al. | Semi-supervised image classification via attention mechanism and generative adversarial network | |
| Park et al. | Robustness Evaluation of Stacked Generative Adversarial Networks using Metamorphic Testing | |
| CN115643153A (en) | Alarm correlation analysis method based on graph neural network | |
| CN114841266A (en) | Voltage sag identification method based on triple prototype network under small sample | |
| Yang et al. | NAM Net: Meta-network with normalization-based attention for few-shot learning |
Legal Events
| Date | Code | Title | Description |
|---|---|---|---|
| PB01 | Publication | ||
| PB01 | Publication | ||
| SE01 | Entry into force of request for substantive examination | ||
| SE01 | Entry into force of request for substantive examination |