CN108319927A - A kind of method of automatic identification disease - Google Patents

Abstract

The invention discloses a method for automatically identifying diseases, which is characterized in that it comprises the following steps: step 1) collecting ground penetrating radar data, and performing overlapping segmentation of the data; step 2) using row variance and a threshold for the data matrix formed after the segmentation method to extract disease location information; step 3) carry out histogram equalization processing to the data obtained in step 2); step 4) utilize the 28th principle to set the threshold, and carry out binarization to the image after histogram equalization; step 5) image Derivation is performed for each column, and the number of two phase types is calculated; Step 6) Select the phase type with the largest number as the image damage type. The beneficial effects achieved by the present invention: automatically locate the disease location, extract the disease characteristics, realize the common underground diseases of the expressway, save a lot of time and energy of the disease interpreters, save costs, and meet the pursuit of automatic disease identification in the ground penetrating radar industry The goal has great practical significance.

Description

A Method for Automatically Identifying Diseases

technical field

The invention relates to a method for automatically identifying diseases and belongs to the technical field of disease identification.

Background technique

In the interpretation of roadbed diseases of GPR signal expressway, the most widely used method at home and abroad is manual image interpretation. However, manual interpretation of images relies heavily on the experience of the interpreter and is highly subjective. When the amount of data is large, the manual interpretation cycle is very long and has a certain lag. How to automatically identify the type of disease and give feedback to save interpretation time has become one of the urgent needs of GPR signal road subgrade image interpretation. The present invention provides a new solution based on the automation of GPR data disease detection and classification. Other existing patents such as Patent No. CN104698503A use offset correction and Kirchhoff wave equation offset method to process preprocessed radar data, and then combine geological and environmental manual interpretation data. This method is computationally intensive and consumes a lot of manpower and time. The same patent number is CN105403883A, which uses the amplitude component to find the region of interest, and performs edge extraction and target hyperbola positioning on the region of interest. This method is suitable for searching for tubular targets, and the manual method is used to find the target body, which is time-consuming and labor-intensive. Another example is the patent number CN1595195A, which uses RBF neural network to automatically identify the target category of radar data. The premise is to analyze and extract the characteristics of the target, and a large amount of sample data is required. However, a large amount of sample data is usually difficult to obtain in the actual implementation process.

Contents of the invention

In order to solve the deficiencies of the prior art, the object of the present invention is to provide a method for automatically identifying the disease, automatically locating the position of the disease, extracting the characteristics of the disease, realizing the common diseases of the highway underground, and saving a lot of time and energy for the disease interpreters.

In order to achieve the above object, the present invention adopts the following technical solutions:

A method for automatically identifying diseases is characterized in that it comprises the steps of:

Step 1) collect ground penetrating radar data, and carry out data overlapping segmentation;

Step 2) adopt row variance and threshold value method to extract disease position information to the data matrix formed after segmentation;

Step 3) performing histogram equalization processing on the data obtained in step 2);

Step 4) utilize the 28th principle to set the threshold, and carry out binarization to the image after histogram equalization;

Step 5) deriving each column of the image, and calculating the quantity of the two phase types;

Step 6) Select a large number of phase types as image damage types.

Aforesaid a kind of method for automatically identifying disease, it is characterized in that, described step 2) obtains disease position according to row variance and threshold value method, the matrix of corresponding position is marked as I ', and the size of matrix I ' is P*N, element The gray value ranges from 0-255.

Aforesaid a kind of method for automatically identifying disease is characterized in that, the concrete content of described matrix I ' is:

21) Overlap and segment the entire data, divide a large amount of data into GPR measured data blocks with a matrix size of M*N, M is the number of samples per channel, N is the number of traces contained in each image after segmentation, and record one of the data The block is I, and I corresponds to the GPR data collected at a certain horizontal position in the actual environment;

22) The pixel value of the measured data I is mapped to the range of 0-255, and the normalized data is denoted as NI;

23) Calculate the row variance of the normalized data matrix NI. There are N data in one row. Calculate the N data variance to obtain a variance data, and calculate the row variance for each row. The data matrix has M rows in total, and finally get The row variance data size of the data matrix is M*1, forming a vector NI_v with a size of M*1, and each element in NI_v represents the row variance of each row of NI;

24) extracting a plurality of line segments whose image line variance is greater than the threshold T is denoted as M1, M2, ... Mp', and the number of lines of each line segment is denoted as m1, m2, ..., mp';

If there is a disease in a concrete area, the number of rows occupied by the disease is at least Among them, f is the antenna frequency, and dt is the sampling interval of two adjacent data on a trace;

If m1+m2+...+mp'<m', it is determined that the graph has no disease;

If m1+m2+...+mp'≥m', it is determined that the graph has disease;

For a diseased GPR image, find the largest number among m1, m2,...,mp′ and record it as _ma , a∈{1,2,...,p′}, and its corresponding line segment is M _a , M _a is determined to be the position of the disease to be found, that is, the row containing the disease extracted from the entire image, each row contains N data, that is, N columns, and the size of the M _a matrix is P*N, P is the row containing the disease The number of rows, record M _a as the matrix I'.

Aforesaid a kind of method for automatically identifying disease, it is characterized in that, described step 3) in the specific content is: for matrix I ', calculate the gray scale number probability distribution of the image of I ', carry out histogram equalization, the original The grayscale histogram of the image changes from a relatively concentrated distribution in the grayscale range of 0-255 to a uniform distribution in the grayscale range of 0-255, and the obtained image after histogram equalization is denoted as I ".

Aforesaid a kind of method for automatically identifying disease, it is characterized in that, the specific content of described step 4) is: carry out binarization process to image, the threshold value selected before binarization process is selected as follows: use 28 principles to get After histogram equalization, the two thresholds of the image I" are recorded as T2 and T3, and T2<T3. The elements with a gray value less than T2 in the image I" are changed to 0, and the elements with a gray value greater than T3 are changed to 255. The element whose degree value is between T2 and T3 becomes 127; the image matrix processed by binarization and 28th principle is recorded as Ik.

Aforesaid a kind of method for automatically identifying disease, it is characterized in that, the concrete content of described step 5) is: derivation is obtained N sequences to each column of matrix Ik, gets the first 6 non-zero values of each sequence, respectively It is recorded as Ik_p1, Ik_p2, Ik_p3, ..., Ik_pn; if the number of non-zero values in a certain sequence is less than 6, remove the phase sequence of the column; the pixel value in each column of matrix Ik is recorded as -1 from large to small, and the pixel Values are marked as 1 from small to large.

The aforementioned method for automatically identifying diseases is characterized in that the content of the step 6) is: only one element with the same sign as the adjacent elements is left, so that the adjacent elements of the phase must have different signs, and the final The standard phase sequence is recorded as Ik_pf1, Ik_pf2, Ik_pf3,...,Ik_pfn, if the number of sequence [-1 1 -1] is large, the disease is classified as water or steel; if the number of sequence [1 -1 1] is large, then Diseases are classified as voids or voids.

The beneficial effects achieved by the present invention: taking the GPR signal of the current highway underground concrete structure disease as the research object, an algorithm for automatically identifying the disease type of the GPR signal is designed; a combination of row variance, histogram equalization, and binarization is used. , first-order derivative algorithm, automatically locate the disease location, extract the disease characteristics, and realize the automatic identification of the common underground diseases of the expressway: void, air, and aquifer, with a certain degree of accuracy, saving a lot of time and time for the disease interpreter It is of great practical significance to save energy and save costs, which is in line with the pursuit goal of automatic disease identification in the ground penetrating radar industry.

Description of drawings

Fig. 1 is a detailed flow chart of disease location and identification in the present invention.

Detailed ways

The present invention will be further described below in conjunction with the accompanying drawings. The following examples are only used to illustrate the technical solution of the present invention more clearly, but not to limit the protection scope of the present invention.

A method for automatically identifying diseases that the present invention relates to comprises the following steps:

Step 1) collect ground penetrating radar data, and carry out data overlapping segmentation;

Step 2) adopt row variance and threshold method to extract disease position information to the data matrix formed after segmentation: obtain disease position according to row variance and threshold method, the matrix of corresponding position is denoted as I ', and the size of matrix I ' is P*N, The element gray value ranges from 0-255.

Step 3) carry out histogram equalization processing to the data obtained in step 2): for matrix I ', calculate the gray level probability distribution of the image of I ', carry out histogram equalization, the purpose is to make the gray level histogram of original image From a relatively concentrated gray-scale interval to a uniform distribution in the entire gray-scale range, the image after histogram equalization is recorded as I". The image contrast after histogram equalization is higher, and the texture characteristics of the disease are more obvious .

Step 4) Use the 28th principle to set the threshold, and binarize the image after histogram equalization:

The image is binarized, and the threshold selected before binarization is selected as follows:

Use the two-eighth principle to obtain the two thresholds of the image I" after histogram equalization and record them as T2 and T3, and T2<T3. Change the elements with a gray value less than T2 in the image I" to 0, and elements with a gray value greater than T3 becomes 255, and the element whose gray value is between T2 and T3 becomes 127. In this embodiment, T2=50 and T3=200.

Areas with pixel values of 0 and 255 reflect the location of the disease. The image matrix processed by binarization and the 28th principle is denoted as Ik. (Or T2 and T3 here are completely two thresholds obtained after comparing a large number of measured data)

Step 5) After step 4) the image pixel value is extremized, it is much simpler to obtain the phase information of the disease.

Derivate each column of the image and calculate the number of two phase types: derivate each column of the matrix Ik to obtain N sequences, take the first 6 non-zero values of each sequence, and record them as Ik_p1, Ik_p2, Ik_p3 ,..., Ik_pn; if the number of non-zero values in a sequence is less than 6, remove the phase sequence of the column. Since each column element of Ik has only three values (0, 127, 255), the pixel value from large to small is recorded as -1, and the pixel value from small to large is recorded as 1. Therefore, all elements of Ik_p1, Ik_p2, Ik_p3, ..., Ik_pn have only two cases: 1, -1.

Step 6) If the ideal Ik_p1 returned by this algorithm takes the first three elements as [-1 1 -1] or [1 -1 1], the adjacent elements must have different signs.

But for the measured data, there will often be 127 in the middle of the gray value transformation from 0 to 255. In step 4, the results are usually [-1 1 1 -1 -1 1], [1 -1 -1 1 1 -1]. In this case, we stipulate that the 2nd and 3rd elements, the 3rd and 4th elements need Merge, for this, the positive or negative of the product between the next element and the previous element is used to judge whether to merge. For example, Ik_p1=[-1 1 1 -1 -1 1], Ik_p1(1)*Ik_p1(2)<0, Ik_p1 retains the first and second elements; Ik_p1(2)*Ik_p1(3)>0, Ik_p1 merges The 2nd and 3rd elements. By analogy, the obtained phase adjacent elements must have different signs. Do this operation for Ik_p1, Ik_p2, Ik_p3,...,Ik_pn, and get the final standard phase sequence as Ik_pf1, Ik_pf2, Ik_pf3,...,Ik_pfn.

Select a large number of phase types as the image damage type: leave only one element with the same sign as adjacent elements, so that the phase adjacent elements must have different signs, and the final standard phase sequence is recorded as Ik_pf1, Ik_pf2, Ik_pf3, ..., Ik_pfn, if the number of sequence [-1 1 -1] is large, the disease is classified as water (rebar); if the number of sequence [1 -1 1] is large, the disease is classified as hollow or void.

The above is only a preferred embodiment of the present invention, it should be pointed out that for those of ordinary skill in the art, without departing from the technical principle of the present invention, some improvements and modifications can also be made. It should also be regarded as the protection scope of the present invention.

Claims (7)

1. A method for automatically identifying a disease, characterized in that it comprises the steps: Step 1) collect ground penetrating radar data, and carry out data overlapping segmentation; Step 2) adopt row variance and threshold value method to extract disease position information to the data matrix formed after segmentation; Step 3) performing histogram equalization processing on the data obtained in step 2); Step 4) utilize the 28th principle to set the threshold, and carry out binarization to the image after histogram equalization; Step 5) deriving each column of the image, and calculating the quantity of the two phase types; Step 6) Select a large number of phase types as image damage types. 2. a kind of method for automatically identifying disease according to claim 1, it is characterized in that, described step 2) obtains disease position according to row variance and threshold value method, the matrix of corresponding position is denoted as I ', the matrix I ' The size is P*N, and the gray value of the element is 0-255. 3. a kind of method for automatically identifying disease according to claim 2, is characterized in that, the concrete content of described matrix I ' is: 21) Overlap and segment the entire data, divide a large amount of data into GPR measured data blocks with a matrix size of M*N, M is the number of samples per channel, N is the number of traces contained in each image after segmentation, and record one of the data The block is I, and I corresponds to the GPR data collected at a certain horizontal position in the actual environment; 22) The pixel value of the measured data I is mapped to the range of 0-255, and the normalized data is denoted as NI; 23) Calculate the row variance of the normalized data matrix NI, there are N data in one row, calculate the N data variance, get a variance data, and calculate the row variance for each row, the data matrix has M rows, and finally get The row variance data size of the data matrix is M*1, forming a vector NI_v with a size of M*1, and each element in NI_v represents the row variance of each row of NI; 24) extracting a plurality of line segments whose image line variance is greater than the threshold T is denoted as M1, M2, ... Mp', and the number of lines of each line segment is denoted as m1, m2, ..., mp'; If there is a disease in a concrete area, the number of rows occupied by the disease is at least Among them, f is the antenna frequency, and dt is the sampling interval of two adjacent data on a trace; If m1+m2+...+mp'<m', it is determined that the graph has no disease; If m1+m2+...+mp'≥m', it is determined that the graph has disease; For a diseased GPR image, find the largest number among m1, m2,...,mp′ and record it as _ma , a∈{1,2,...,p′}, and its corresponding line segment is M _a , M _a is determined to be the position of the disease to be found, that is, the row containing the disease extracted from the entire image, each row contains N data, that is, N columns, and the size of the M _a matrix is P*N, P is the row containing the disease The number of rows, record M _a as the matrix I'. 4. a kind of method for automatically identifying disease according to claim 1, it is characterized in that, described step 3) in concrete content is: for matrix I ', calculate the gray scale number probability distribution of the image of I ', Perform histogram equalization, change the gray histogram of the original image from a relatively concentrated distribution in the gray scale range of 0-255 to a uniform distribution in the gray scale range of 0-255, and obtain the image after histogram equalization as I ". 5. a kind of method for automatically identifying disease according to claim 1, it is characterized in that, the specific content of described step 4) is: carry out binarization process to image, the threshold value selected before binarization process is as follows Select: Use the two-eighth principle to obtain the two thresholds of the image I" after histogram equalization and record them as T2 and T3, and T2<T3. Change the elements with a gray value less than T2 in the image I" to 0, and elements with a gray value greater than T3 becomes 255, and the element whose gray value is between T2 and T3 becomes 127; The image matrix processed by binarization and the 28th principle is denoted as Ik. 6. a kind of method for automatically identifying disease according to claim 5, it is characterized in that, the specific content of described step 5) is: obtain N sequences to each column derivation of matrix Ik, get the front of each sequence 6 non-zero values, respectively recorded as Ik_p1, Ik_p2, Ik_p3, ..., Ik_pn; if the number of non-zero values in a certain sequence is less than 6, then remove the phase sequence of the column; the pixel value in each column element of the matrix Ik ranges from large to The small value is recorded as -1, and the pixel value from small to large is recorded as 1. 7. A kind of method for automatic identification of disease according to claim 6, it is characterized in that, the content of described step 6) is: leave only one element of the same sign to adjacent element, make phase adjacent element It must be a different sign, and the final standard phase sequence is recorded as Ik_pf1, Ik_pf2, Ik_pf3,...,Ik_pfn. If the number of sequence [-1 1 -1] is large, the disease is classified as water or steel bar; if the sequence [1 -1 1], the disease is classified as hollow or empty.