CN104143115A - Technological method for achieving soil water content classified identification through geological radar technology - Google Patents

Abstract

The invention discloses a technological method for achieving soil water content classified identification through a geological radar technology. The technological method comprises the four parts of data preprocessing, soil water content parameter extraction, neural network classified identification and result display. Data preprocessing comprises the steps of null line correction, wavelet transformation and lowpass filtering. Soil water content parameter extraction comprises the steps that the power spectrum of reflection signals is solved with an autoregression moving average spectrum estimation method, data normalization is conducted, feature vectors and feature values of the power spectrum are extracted with a principle component analyzing method, and a sample feature vector database is constructed. Neural network classified identification comprises the steps that a neural network is trained through sample feature vectors, the trained mature network is used for conducting classified identification on data to be identified. Result display comprises the step that classified results are mapped and displayed. According to the technological method for achieving soil water content classified identification through the geological radar technology, automatic fast classified identification of the water content of soil is achieved, and a guiding function is performed on land detection and land rehabilitation.

Description

Geological radar technology is realized the technical method of soil water-containing Classification and Identification

Technical field

The present invention relates to Georadar Data and process and interpretation technique, particularly a kind of geological radar technology is realized the technical method of soil water-containing Classification and Identification.

Background technology

Geological radar technology is a science of carrying out buried target detection by transmitting high-frequency impulse electromagnetic wave (frequency range is at 106 ~ 109Hz).The features such as that geological radar has is simple to operate, detection accuracy is high, not damaged, picking rate are fast, are current engineering detecting and the most active technical method of prospecting, and the application in Geotechnical Engineering is increasingly extensive.

Geological radar data is processed and is belonged in theory the category that digital signal processing is explained.Integrated interpretation method mainly contains direct interpretation procedure and indirect interpretation method.Direct method is by the raw data of radar detection being done to the processing of some routines, according to the power of radar appearance, phase characteristic, the characteristic informations such as the variation of lineups, again in conjunction with drilling data and other relevant geologic informations, directly reflected signal is made quantitatively and qualitative interpretation, but the geologic condition in the face of more complicated, be difficult to the correct external appearance characteristic of explaining after being complicated, and complicated and changeable due to underground medium, when often having, echoed signal becomes, the feature of non-stationary and randomness, so adopting the analysis and processing method of random signal analyzes data information, extract the proper vector of the self structure of surveying thing, choose principal ingredient wherein, and preserve complicated Nonlinear Mapping information with neural network model, realize accordingly automatic classification identification, this interpretation methods can be good at the principal character that thing is surveyed in reflection, and can realize automatic explanation, rapidly and efficiently.

Summary of the invention

The shortcoming that the object of the invention is to cannot accurately realize fast for geologic radar detection soil water-containing classification, provides a kind of geological radar technology to realize the technical method of soil water-containing Classification and Identification.

The technical scheme that the present invention solves its technical matters employing is:

Geological radar technology is realized the technical method of soil water-containing Classification and Identification, it is characterized in that comprising following concrete steps:

1) data pre-service:

The pretreated input data of data are the raw data that geological radar collects, and first Georadar Data are carried out to zero line correction, then carry out wavelet transformation, finally carry out low-pass filtering, and its output data are for completing pretreated data;

2) soil water-containing information extraction:

The step 1) of usining completes pre-service output data afterwards as the input data of this step, it is carried out to autoregressive moving average and ask power spectrum, carry out again data normalization, finally carry out principal component analysis (PCA), extract soil physical property infomation, deposit in sample database, it is soil water-containing characteristic that this step completes later output data;

3) neural network classification identification:

Using step 2) complete the soil water-containing characteristic of output after processing as the input data of this step, the data of depositing in sample database are inputted to neural metwork training neural network to be trained, then with the ripe network of training, complete the Classification and Identification of data to be sorted, the output data after this step completes are classification number under data to be identified;

4) result shows:

After the step 3) of usining completes and processes, under the data to be identified of output, classification number, as the input data of this step, carries out color range modulation to classification results, draws out search coverage classification of soils result figure.

Described zero line is proofreaied and correct: every track data that geological radar is collected carries out zero line correction, removes the signal drift noise of instrument self; Described wavelet transformation is: the data after zero line is proofreaied and correct are carried out wavelet transformation, filtering high frequency saltus step noise and extraneous high frequency interference noise again; Described low-pass filtering: the data after wavelet transformation are carried out to low-pass filtering again, near the energy direct current of filtered signal.

Described autoregressive moving average asks power spectrum to be: the every track data in Georadar Data after pre-service is carried out to the variation of autoregressive moving average power spectrum, try to achieve power spectrum information; Described data normalization is: power spectrum data is normalized; Described principal component analysis (PCA) is: the data after normalization are carried out to principal component analysis (PCA); Described extraction soil physical property infomation is: using the principal character vector after principal component analysis (PCA) as soil physical property infomation; Described sample database is: the soil physical property infomation obtaining is stored in property data base.

Described neural network training is: the characteristic input neural network in property data base is trained, until network training is ripe; Described neural network classification is identified as: the ripe neural network of radar data input training of the classification to be predicted after pretreated is carried out to Classification and Identification, provide affiliated classification number.

Described color range is modulated to: the classification results that step 3) is obtained carries out color range modulation, the corresponding color value of each class; Described drafting classification of soils result figure is: utilize the color range modulating that classification results one-tenth figure is shown, image can intuitively show search coverage soil water-containing classification schematic diagram.

The invention has the beneficial effects as follows: based on geological radar technology, a kind of technical method of soil water-containing Classification and Identification is being provided aspect soil water-containing Classification and Identification, compare with traditional boring with sampling method of testing, saved cost, realized the detection and identify on yardstick among a small circle, and by neural network, realized fast automatic identification, greatly improved efficiency.For actual detection provides a kind of technological means of precise and high efficiency.

Accompanying drawing explanation

Fig. 1 is that data is processed explanation FB(flow block).

Fig. 2 is geologic radar detection curve map, and wherein (a) is raw data curve; (b) for zero line is proofreaied and correct rear data and curves; (c) be data and curves after wavelet transformation; (d) be data and curves after low-pass filtering.

Fig. 3 is power spectrum conversion comparison diagram, the power spectrum chart that wherein (a) obtains for classical Fourier transform, the power spectrum chart (b) obtaining for autoregressive moving average power Spectral Estimation.

Fig. 4 is the corresponding power spectrum chart of different water cut, and wherein (a), for water percentage is the power spectrum chart of 20% correspondence, (b) for water percentage is the power spectrum chart of 30% correspondence, for water percentage, is (c) power spectrum chart of 40% correspondence.

Fig. 5 is geologic radar detection original section image.

Fig. 6 is the pretreated profile image of geological radar.

Fig. 7 is imaging color range table.

Fig. 8 is the classification sectional view of geologic radar detection soil water-containing.

Embodiment

Below in conjunction with drawings and Examples, the present invention is described in further detail:

Fig. 1 is the data processing explanation FB(flow block) that geological radar technology of the present invention is realized soil water-containing Classification and Identification technical method.The concrete implementation detail of each step is as follows:

1. data pre-service

The pretreated object of data is before soil water-containing information extraction, to carry out the removal of various undesired signals, guarantees the reliability that soil water-containing response characteristic is extracted.Fig. 5 is the raw data profile image that geological radar gathers, and Fig. 6 is through pretreated Ground Penetrating Radar image, is specifically implemented as follows:

First the Georadar Data collecting is carried out to zero line correction, remove the signal drift noise of instrument self, Fig. 2 (a) original signal, Fig. 2 (b) is the signal after signal zero line is proofreaied and correct, the method of implementing is as follows: calculate signal average, by original signal value, deduct required average.

Secondly, the curve after zero line is proofreaied and correct carries out wavelet transformation, and compacting high frequency noise, uses Moret wavelet function, and scale parameter is 2, and transformation results is referring to Fig. 2 (c).

Finally, the signal after wavelet transformation is carried out to low-pass filtering, get rid of the interference of signal DC component, low-pass filtering parameter is 25MHz, and after filtering, result is referring to Fig. 2 (d).

2. soil physical property infomation extracts

First to pretreated Georadar Data signal carrying out autoregressive moving average spectrum estimates. for meeting the equation of difference

There is spectral density , wherein , with , , be called " rational expression spectral density ".Here adopt Cadzow Zymography in order to reduce number of parameters and to realize spectra calculation.Rational expression spectral density is rewritten into , and be decomposed into , wherein , all the other , between pass be:

Stationary process power spectrum is written as:

$p_x\left(z\right)=\underset{k=-\∞}{\overset{+\∞}{\mathrm{\Σ}}}{R}_x\left(k\right)z^{-k}=\underset{k=0}{\overset{+\∞}{\mathrm{\Σ}}}r\left(k\right)z^{-k}+\underset{k=-\∞}{\overset{0}{\mathrm{\Σ}}}r(-k)z^{-k},$ Wherein and be coefficient of autocorrelation can obtain: , this formula left and right is with taking advantage of , and relatively both sides, with the coefficient of power item, can obtain: .So far autocorrelation function and AR exponent number and parameter determine, coefficient also can determine, thereby power spectrum can be in the hope of, realized autoregressive moving average spectrum and estimated.This method is more obvious to the performance of radio-frequency component, within the scope of effective bandwidth, has strengthened the recognition feature of radio-frequency component, higher than the resolution characteristic of classical Fourier transform.The resolution of the radio-frequency component of the power spectrum that the radio-frequency component of the power spectrum chart that (b) figure of Fig. 3 obtains by autoregressive moving average obtains by classical Fourier transform than (a) figure is high.

Water cut has a great impact the dynamic characteristic of Electromagnetic Wave Propagation, on power spectrum, there is obvious performance, because water-bearing media electric conductivity strengthens, therefore medium electro-magnetic wave absorption coefficient strengthens, especially more obvious to the absorption of high-frequency signal, radio-frequency component occupation rate declines, and contrary low-frequency component occupancy volume increases, and can embody from the side media water-bearing rate feature.Utilize geological radar simulation tool (GPRMAX) to set up model herein, obtain the detection data of different water cut soil, reflection wave is analyzed, the (a) and (b) of Fig. 4, (c) are respectively that the water percentage that simulates is 20%, 30% and 40% power spectrum chart, and its low-frequency component accounting increases successively.

Secondly power spectrum information is normalized, first finds out maximin wherein, with treating that normalized numerical value deducts minimum value and obtains divided by the difference of maximin.Normalization can be eliminated the impact of data on neural network from dimension, reduces the fluctuation of network.

Then carry out principal component analysis (PCA), the data after normalization are extracted to major component.Specific implementation is: the sample matrix X to N dimension (experiment is 1024 dimensions), the covariance matrix S of first step compute matrix X; Second step calculates the latent vector e1 of covariance matrix S, e2 ... the eigenvalue of eN, eigenvalue, is chosen wherein latent vector corresponding to the larger eigenvalue of general impacts to little sequence by large, takes here overall accumulative total contribution degree is reached to 85% latent vector corresponding to eigenvalue; Among the space that the 3rd step data for projection is opened to eigenvector.Principal component analysis (PCA) has realized data compression effectively, kicks out of redundant data, at angle of statistics, holds the difference between signal, parses major influence factors from multiple factors.

Finally characteristic is deposited in to sample database storage administration.

3. neural network classification identification

First by the data input neural network training network in sample database, until network reaches frequency of training (setting 100000 times) or the systematic error of regulation, reach permissible value (setting 0.00001), think that network training is ripe.

Then by the ripe network of data input training of the search coverage to be identified of handling well, the result of record sort identification.

4. result shows

The classification results of output is 1,2,3,4,5,6 these 6 classes, corresponding soil moisture content is more than 0 ~ 10%, 10% ~ 20%, 20% ~ 30%, 30% ~ 40%, 40% ~ 50%, 50% this 6 kind respectively, stipulate the corresponding a kind of specific color of a classification number, modulation color range, as shown in Figure 7.

Then by the color range of modulation, become figure to show the Output rusults of search coverage, provide soil water-containing Classification and Identification sectional view, as shown in Figure 8.

Claims (5)

1. geological radar technology is realized the technical method of soil water-containing Classification and Identification, it is characterized in that comprising following concrete steps:

1) data pre-service

The pretreated input data of data are the raw data that geological radar collects, and first Georadar Data are carried out to zero line correction, then carry out wavelet transformation, finally carry out low-pass filtering, and its output data are for completing pretreated data;

2) soil water-containing information extraction

The step 1) of usining completes pre-service output data afterwards as the input data of this step, it is carried out to autoregressive moving average and ask power spectrum, carry out again data normalization, finally carry out principal component analysis (PCA), extract soil physical property infomation, deposit in sample database, it is soil water-containing characteristic that this step completes later output data;

3) neural network classification identification

Using step 2) complete the soil water-containing characteristic of output after processing as the input data of this step, the data of depositing in sample database are inputted to neural metwork training neural network to be trained, then with the ripe network of training, complete the Classification and Identification of data to be sorted, the output data after this step completes are classification number under data to be identified;

4) result shows

After the step 3) of usining completes and processes, under the data to be identified of output, classification number, as the input data of this step, carries out color range modulation to classification results, draws out search coverage classification of soils result figure.

2. geological radar technology according to claim 1 is realized the technical method of soil water-containing Classification and Identification, it is characterized in that, in step 1), described zero line is proofreaied and correct and is: every track data that geological radar is collected carries out zero line correction, removes the signal drift noise of instrument self; Described wavelet transformation is: the data after zero line is proofreaied and correct are carried out wavelet transformation, filtering high frequency saltus step noise and extraneous high frequency interference noise again; Described low-pass filtering: the data after wavelet transformation are carried out to low-pass filtering again, near the energy direct current of filtered signal.

3. geological radar technology according to claim 1 is realized the technical method of soil water-containing Classification and Identification, step 2) in, described autoregressive moving average asks power spectrum to be: the every track data in Georadar Data after pre-service is carried out to the variation of autoregressive moving average power spectrum, try to achieve power spectrum information; Described data normalization is: power spectrum data is normalized; Described principal component analysis (PCA) is: the data after normalization are carried out to principal component analysis (PCA); Described extraction soil physical property infomation is: using the principal character vector after principal component analysis (PCA) as soil physical property infomation; Described sample database is: the soil physical property infomation obtaining is stored in property data base.

4. geological radar technology according to claim 1 is realized the technical method of soil water-containing Classification and Identification, in step 3), described neural network training is: the characteristic input neural network in property data base is trained, until network training is ripe; Described neural network classification is identified as: the ripe neural network of radar data input training of the classification to be predicted after pretreated is carried out to Classification and Identification, provide affiliated classification number.

5. geological radar technology according to claim 1 is realized the technical method of soil water-containing Classification and Identification, and in step 4), described color range is modulated to: the classification results that step 3) is obtained carries out color range modulation, the corresponding color value of each class; Described drafting classification of soils result figure is: utilize the color range modulating that classification results one-tenth figure is shown, image can intuitively show search coverage soil water-containing classification schematic diagram.