阅读说明：本技术 极点特征聚类的公路隐藏病害识别系统及其实施方法 (Pole characteristic clustering road hidden disease identification system and implementation method thereof ) 是由 周丽军 刘博� 吴宏涛 孟颖 孙贝 薛春明 周晓旭 梁玉荣 段英杰 李永胜 刘晓 于 2019-09-27 设计创作，主要内容包括：本发明提供了一种极点特征聚类的公路隐藏病害识别系统及其实施方法,所述识别系统包括数据采集系统、数据预处理平台、极点特征聚类分析平台和输出平台。本发明提供的极点特征聚类分析设计方案,实现了公路隐藏病害的自动识别,有效解决了地质雷达无损检测中回波数据量大、人工解读困难等问题。本发明采用奇异值分解方法对回波信号进行降维,而且去除了噪声干扰。同时,本发明通过解卷积过程获取目标的冲激响应,实现了标识目标属性的极点特征提取,对不同属性目标进行聚类分析,构建了极点特征数据库,有效识别了不同介电属性的公路隐藏病害,解决了地质雷达数据解析难以实现自动化的问题。(The invention provides a highway hidden disease recognition system based on pole feature clusters and an implementation method thereof, wherein the recognition system comprises a data acquisition system, a data preprocessing platform, a pole feature cluster analysis platform and an output platform.)

The road hidden disease identification system based on pole feature clustering is characterized by comprising a data acquisition platform, a data preprocessing platform, a pole feature clustering analysis platform and an output platform;

the data acquisition platform comprises a geological radar, a radar support and a data acquisition control device, wherein the geological radar is arranged on the data acquisition platform in a vehicle-mounted mode or a suspension mode so as to form a preset distance between the geological radar and the ground, and the data acquisition platform forms a plane wave signal to acquire disease data;

the data preprocessing platform comprises a filtering module, a self-adaptive gain adjusting module and a singular value dimension reduction module, and is used for carrying out preliminary processing on the disease data to obtain a characteristic vector and a characteristic value of singular value decomposition;

the pole feature clustering analysis platform comprises a feature space model module, a clustering algorithm module and a pole feature database module, reconstructs echo signals by using optimal feature values and feature vectors, extracts impulse responses from the echo signals, calculates poles according to the impulse responses, constructs a pole feature space, obtains feature space distribution information of different diseases through a clustering algorithm, and establishes a pole feature database;

and the output platform displays a disease identification interface and outputs the cluster type and dielectric property of the disease in the pole characteristic space.

2. The system for identifying hidden diseases on roads with polar feature clustering according to claim 1, wherein the data acquisition platform further comprises an on-board device and a mileage calibration device, wherein an accommodating space is arranged inside the on-board device, the accommodating space is used for accommodating the data acquisition and processing system, the devices are connected and fixed through screws, the mileage calibration device is fixedly arranged on a tire, the radar support and the on-board device are connected and fixed through screws, and the radar support and the geological radar are connected and fixed through plastic screws.

3. The system for identifying hidden road diseases based on pole feature clustering according to claim 1, wherein the filtering module is used for filtering out high-frequency clutter and retaining echo data near a central frequency;

the self-adaptive gain adjusting module is used for adjusting the gain of each part in the echo according to the whole echo energy, enhancing the gain of the target echo part and weakening the gain of the direct wave part;

and the singular value dimensionality reduction module is used for carrying out singular value decomposition on the signal and selecting an optimal characteristic value and a corresponding characteristic vector.

4. The system for identifying hidden road diseases based on pole feature clustering according to claim 1, wherein the calculation formula of the coverage area of the plane waves reaching the ground is as follows:

s＝(a+2×0.7×tanθ)×(b+2×0.7tanθ) (1)

where a is the length of the radar, b is the width of the radar, and θ is the radar antenna lobe angle.

5. The pole-feature-clustered road hidden disease identification system according to claim 1, wherein the preset distance between the geological radar and the ground is 29 cm to 50 cm, or the preset distance between the geological radar and the ground is 70cm to 79 cm.

6, implementation method of road hidden disease recognition system of pole feature clustering, characterized in that, the recognition system includes data acquisition platform, data preprocessing platform, pole feature clustering analysis platform and output platform, the data acquisition platform includes geological radar, radar support, data acquisition control device, the data preprocessing platform includes filtering module, adaptive gain adjustment module, singular value dimension reduction module, the pole feature clustering analysis platform includes feature space model module, clustering algorithm module, pole feature database module;

the implementation method comprises the following steps:

the geological radar is arranged on the data acquisition platform in a vehicle-mounted mode or a suspension mode so as to form a preset distance between the geological radar and the ground, and the data acquisition platform forms a plane wave signal to acquire disease data;

the data preprocessing platform carries out primary processing on the disease data to obtain a characteristic vector and a characteristic value of singular value decomposition;

the pole feature clustering analysis platform reconstructs echo signals by using the optimal feature values and the feature vectors, extracts impulse responses from the echo signals, calculates poles according to the impulse responses, constructs pole feature spaces, obtains feature space distribution information of different diseases through a clustering algorithm, and establishes a pole feature database;

and the output platform displays a disease identification interface and outputs the cluster type and dielectric property of the disease in the pole characteristic space.

7. The implementation method of the pole feature clustering road hidden disease identification system according to claim 6, wherein the data preprocessing platform performs preliminary processing on the disease data, and the step of obtaining the eigenvectors and eigenvalues of singular value decomposition comprises:

the filtering module filters high-frequency clutter and retains echo data near the central frequency;

the self-adaptive gain adjusting module adjusts the gain of each part in the echo according to the whole echo energy, enhances the gain of the target echo part and weakens the gain of the direct wave part;

and the singular value dimensionality reduction module carries out singular value decomposition on the signal and selects an optimal characteristic value and a corresponding characteristic vector.

8. The implementation method of the pole feature clustered road hidden disease identification system as claimed in claim 6, wherein the step of forming plane wave signals by the data acquisition platform to acquire disease data comprises:

adjusting the azimuth angle of the geological radar until the echo peak value is maximum so that the radar antenna and the hidden diseases are in a common polarization direction, wherein the calculation formula of the azimuth angle is as follows:

9. the implementation method of the pole feature clustering road hidden disease identification system according to claim 7, wherein the singular value dimensionality reduction module performs singular value decomposition on the signal, and the step of selecting the optimal eigenvalue and the corresponding eigenvector comprises:

the singular value dimensionality reduction module carries out singular value decomposition on the signal to obtain a characteristic value, and a calculation formula of the characteristic value is as follows:

constructing a Hankel matrix Y for echo data Y (t), wherein a diagonal matrix sigma consists of characteristic values of Y;

arranging the characteristic values according to the sequence from large to small to obtain the ratio of each characteristic value to the maximum characteristic value;

selecting characteristic values with the ratio larger than 0.01, and determining the number of large characteristic values;

and obtaining a corresponding feature vector according to the determined large feature value.

10. The implementation method of the pole feature clustering road hidden disease identification system according to claim 6, wherein the pole feature clustering analysis platform reconstructs echo signals by using the optimal feature values and feature vectors, extracts impulse responses therefrom, calculates poles according to the impulse responses, and constructs a pole feature space by the steps of:

collecting background echo signals without targets;

and (3) extracting impulse response by using a deconvolution method, wherein the convolution process is as follows:

y_b(t)*h(t)＝y_t(t) (4)

wherein the background echo signal is y_b(t) target echo response is y_t(t), the impulse response is h (t);

the impulse response h (t) is written as follows:

wherein h is^op(t) represents the early response of the impulse response, the second part of the right side of the above formula is the late response of the impulse response, t₀Is the start time of the late response, R_iComplex amplitude, s, representing the ith resonance state_i＝α_i+j2πf_iThe pole representing the ith resonance state, α_iAnd f_iRespectively attenuation factor and resonance frequency;

according to attenuation factor α_iTo the resonance frequency f_iThe pole feature space is formed as follows:

[A，F]＝{(α₁，f₁)，(α₂，f₂)，…，(α_M，f_M)} (6)

wherein M is the number of available poles obtained by reconstructing the data.

Technical Field

The invention relates to the technical field of signal processing, in particular to a road hidden disease identification system based on pole feature clustering and an implementation method thereof.

Background

At present, the nondestructive detection of hidden diseases under the highway mainly depends on a Ground Penetrating Radar (GPR) technology, namely, high-frequency electromagnetic waves are transmitted to an underground space, a dielectric difference between an underground object and the environment is utilized to process a received echo, and the distribution condition of an underground medium is extracted.

In recent years, artificial intelligence technology is used in ground penetrating radar road disease detection by beginners. If an artificial neural network multilayer model is established, the method is used for automatically diagnosing the road diseases; processing ground penetrating radar echo signals by using a depth dictionary learning method, and distinguishing a plurality of buried targets with different shapes; the vector quantization neural network method is used for identifying the railway subgrade diseases; and automatically identifying the abnormal edge of the radar image by using a clustering method. These B-scan image based methods also focus mainly on physical models, most of which identify hyperbolic features of radar images, and progress in identifying dielectric properties of targets is very limited. A learner can effectively identify water damage and cavities through different kernel matching tracking algorithms and can identify the soil-based disease category of urban roads, and the algorithm consumes more time and responds more when the length of a measuring line is increased only based on detection of a B-scan image.

With the development of time domain pulse radar, a target identification method based on radar echo transient response is greatly developed, and compared with an imaging technology, the target identification method has the advantage that only a single transient response of a target time domain measurement is required. According to the SEM technology, the target depends on Complex Natural Resonance (CNR) of a late part of a transient response echo, namely pole characteristics, the distribution rule of the pole characteristics is only related to the target attribute, and the performance can be well applied to the field of target identification. In recent years, these techniques have also been applied to subsurface target identification for GPR exploration, such as identifying targets of different shapes or different orientations using pole differences. Estimating and identifying an underground target by utilizing a transient resonance phenomenon, wherein the impulse response of the target is often required to be obtained; studying the impulse response of the target and the contained internal resonance current from the perspective of a time domain integral equation, and indicating that the transient scattering characteristic of the target can be completely represented by the impulse response; obtaining impulse response of the target by using a deconvolution method, performing singular value decomposition on the impulse response to obtain CNR of the target late response, and researching the resonance behavior of the semi-space dielectric target; the relation between CNR and target geometric property is respectively researched based on the internal resonance and the external resonance of transient response. The above documents, however, lack in-depth study of resonance and target dielectric characteristics.

In summary, most of geological radar data analysis at present depends on a B-scan image, so that the data acquisition process is complex, the data volume is large, clutter interference is large, the target position is judged mainly through a hyperbolic peak in the B-scan image in the existing data analysis, and the precision is low; and targets of different materials can not be distinguished by using a hyperbolic peak value, so that the targets of different media are difficult to identify.

Disclosure of Invention

In order to solve the limitations and defects of the prior art, the invention provides a road hidden disease identification system with pole feature clusters, which comprises a data acquisition platform, a data preprocessing platform, a pole feature cluster analysis platform and an output platform;

the data acquisition platform comprises a geological radar, a radar support and a data acquisition control device, wherein the geological radar is arranged on the data acquisition platform in a vehicle-mounted mode or a suspension mode so as to form a preset distance between the geological radar and the ground, and the data acquisition platform forms a plane wave signal to acquire disease data;

the data preprocessing platform comprises a filtering module, a self-adaptive gain adjusting module and a singular value dimension reduction module, and is used for carrying out preliminary processing on the disease data to obtain a characteristic vector and a characteristic value of singular value decomposition;

the pole feature clustering analysis platform comprises a feature space model module, a clustering algorithm module and a pole feature database module, reconstructs echo signals by using optimal feature values and feature vectors, extracts impulse responses from the echo signals, calculates poles according to the impulse responses, constructs a pole feature space, obtains feature space distribution information of different diseases through a clustering algorithm, and establishes a pole feature database;

and the output platform displays a disease identification interface and outputs the cluster type and dielectric property of the disease in the pole characteristic space.

Optionally, the data acquisition platform further includes a vehicle-mounted device and a mileage calibration device, an accommodation space is provided inside the vehicle-mounted device, the accommodation space is used for placing a data acquisition and processing system, the devices are connected and fixed by screws, the mileage calibration device is fixedly arranged on a tire, the radar support is connected and fixed with the vehicle-mounted device by screws, and the radar support is connected and fixed with the geological radar by plastic screws.

Optionally, the filtering module is configured to filter out high-frequency clutter and retain echo data near the center frequency;

the self-adaptive gain adjusting module is used for adjusting the gain of each part in the echo according to the whole echo energy, enhancing the gain of the target echo part and weakening the gain of the direct wave part;

and the singular value dimensionality reduction module is used for carrying out singular value decomposition on the signal and selecting an optimal characteristic value and a corresponding characteristic vector.

Optionally, the calculation formula of the coverage area of the plane wave reaching the ground is as follows:

s＝(a+2×0.7×tan θ)×(b+2×0.7×tan θ) (1)

where a is the length of the radar, b is the width of the radar, and θ is the radar antenna lobe angle.

Optionally, the preset distance between the geological radar and the ground is 29 cm to 50 cm, or the preset distance between the geological radar and the ground is 70cm to 79 cm.

The invention also provides an implementation method of the road hidden disease identification system with pole feature clusters, wherein the identification system comprises a data acquisition platform, a data preprocessing platform, a pole feature cluster analysis platform and an output platform, the data acquisition platform comprises a geological radar, a radar bracket and a data acquisition control device, the data preprocessing platform comprises a filtering module, an adaptive gain adjustment module and a singular value dimension reduction module, and the pole feature cluster analysis platform comprises a feature space model module, a clustering algorithm module and a pole feature database module;

the implementation method comprises the following steps:

the geological radar is arranged on the data acquisition platform in a vehicle-mounted mode or a suspension mode so as to form a preset distance between the geological radar and the ground, and the data acquisition platform forms a plane wave signal to acquire disease data;

the data preprocessing platform carries out primary processing on the disease data to obtain a characteristic vector and a characteristic value of singular value decomposition;

the pole feature clustering analysis platform reconstructs echo signals by using the optimal feature values and the feature vectors, extracts impulse responses from the echo signals, calculates poles according to the impulse responses, constructs pole feature spaces, obtains feature space distribution information of different diseases through a clustering algorithm, and establishes a pole feature database;

and the output platform displays a disease identification interface and outputs the cluster type and dielectric property of the disease in the pole characteristic space.

Optionally, the step of performing preliminary processing on the disease data by the data preprocessing platform to obtain a singular value decomposition eigenvector and a singular value includes:

the filtering module filters high-frequency clutter and retains echo data near the central frequency;

the self-adaptive gain adjusting module adjusts the gain of each part in the echo according to the whole echo energy, enhances the gain of the target echo part and weakens the gain of the direct wave part;

and the singular value dimensionality reduction module carries out singular value decomposition on the signal and selects an optimal characteristic value and a corresponding characteristic vector.

Optionally, the step of acquiring the disease data by the data acquisition platform forming a plane wave signal includes:

adjusting the azimuth angle of the geological radar until the echo peak value is maximum so that the radar antenna and the hidden diseases are in a common polarization direction, wherein the calculation formula of the azimuth angle is as follows:

optionally, the singular value dimensionality reduction module performs singular value decomposition on the signal, and the step of selecting the optimal eigenvalue and the corresponding eigenvector includes:

the singular value dimensionality reduction module carries out singular value decomposition on the signal to obtain a characteristic value, and a calculation formula of the characteristic value is as follows:

constructing a Hankel matrix Y for echo data Y (t), wherein a diagonal matrix sigma consists of characteristic values of Y;

arranging the characteristic values according to the sequence from large to small to obtain the ratio of each characteristic value to the maximum characteristic value;

selecting characteristic values with the ratio larger than 0.01, and determining the number of large characteristic values;

and obtaining a corresponding feature vector according to the determined large feature value.

Optionally, the pole feature clustering analysis platform reconstructs an echo signal by using the optimal feature value and the feature vector, extracts an impulse response from the echo signal, calculates a pole according to the impulse response, and constructs a pole feature space, including:

collecting background echo signals without targets;

and (3) extracting impulse response by using a deconvolution method, wherein the convolution process is as follows:

y_b(t)*h(t)＝y_t(t) (4)

wherein the background echo signal is y_b(t) target echo response is y_t(t), the impulse response is h (t);

the impulse response h (t) is written as follows:

wherein h is^op(t) represents the early response of the impulse response, the second part of the right side of the above formula is the late response of the impulse response, t₀Is the start time of the late response, R_iComplex amplitude, s, representing the ith resonance state_i＝α_i+j2πf_iThe pole representing the ith resonance state, α_iAnd f_iRespectively attenuation factor and resonance frequency;

according to attenuation factor α_iTo the resonance frequency f_iThe pole feature space is formed as follows:

[A，F]＝{(α₁，f₁)，(α₂，f₂)，...，(α_M，f_M)} (6)

wherein M is the number of available poles obtained by reconstructing the data.

The invention has the following beneficial effects:

the invention provides a highway hidden disease recognition system based on pole feature clusters and an implementation method thereof, wherein the recognition system comprises a data acquisition system, a data preprocessing platform, a pole feature cluster analysis platform and an output platform.

Drawings

Fig. 1 is a schematic structural diagram of a pole feature clustering road hidden disease identification system provided in embodiment of the present invention.

Fig. 2 is a schematic overall construction diagram of a pole feature clustering road hidden disease identification system provided in embodiment of the present invention.

Fig. 3 is an error distribution diagram of an air crack reconstructed echo of the pole feature clustering road hidden disease identification system provided in embodiment of the present invention.

Fig. 4 is an error distribution diagram of a water-filled crack reconstruction echo of the pole feature clustering road hidden disease identification system provided in embodiment of the present invention.

Fig. 5 is a pole feature space distribution diagram of the pole feature clustering road hidden disease identification system according to the second embodiment of the present invention.

Fig. 6 is a pole cluster distribution and test sample cluster result diagram of the pole feature clustered road hidden disease identification system provided in the second embodiment of the present invention.

The device comprises a reference numeral 1, a vehicle-mounted device, 2, a geological radar, 3, a mileage calibration device, 4, a data acquisition control system, 5, a radar support, 6, a data preprocessing system, 7, a pole characteristic clustering analysis platform, 8, an output platform, 9, pole distribution of a background, 10, pole distribution of an air crack, 11, pole distribution of a water filling crack, 12, a clustering category ①, 13, a clustering category ②, 14, a clustering category ③, 15, a clustering category ④, 16, a pole belonging to a category ①, 17, a pole belonging to a category ②, 18, a pole belonging to a category ③, 19, a pole belonging to a category ④, 20, pole distribution of a test sample air crack, and 21, pole distribution of a test sample water filling crack.

Detailed Description

In order to make those skilled in the art better understand the technical solution of the present invention, the highway hidden disease identification system with polar feature clustering and the implementation method thereof provided by the present invention are described in detail below with reference to the accompanying drawings.