CN105469099A - Sparse-representation-classification-based pavement crack detection and identification method - Google Patents

Abstract

The present invention provides a pavement crack detection and recognition method based on sparse representation classification, by introducing a sparse representation classifier (Sparse? Representation-based? Classifier, namely SRC), selecting effective sub-block high-order moment features, avoiding The image is preprocessed and postprocessed, which simplifies the detection steps and improves the operating efficiency. Specifically, it includes: extraction and normalization of the feature vector of the training set (sub-block), extracting features from the test image into blocks and using SRC for classification, and identifying crack types according to the mapping code of the sub-block classification results. Compared with the traditional crack detection and recognition method, the method proposed by the invention has higher recognition accuracy and execution efficiency.

Description

Pavement Crack Detection and Recognition Method Based on Sparse Representation Classification

technical field

The invention relates to the fields of computer vision and pattern recognition, mainly uses machine learning methods for detection and recognition of road surface cracks, and in particular relates to a method for detection and recognition of road surface cracks based on sparse representation classification.

Background technique

Cracks are the most common disease on pavement, timely and accurate detection of pavement cracks is very important for the maintenance and management of high-load roads. Through artificial vision detection, a lot of manpower and material resources are required, and the detection results are subjective. The rapid development of computers has enabled people to use computers to complete automatic detection of road surface diseases.

Traditional pavement crack detection is based on image processing and analysis, and then some cross-field methods have also been proposed to describe and enhance crack characteristics in complex environments, introducing new ideas for crack detection. For example, the method combined with fuzzy set theory, the detection strategy based on artificial population, the detection algorithm using the minimum spanning tree of the target point, and the algorithm based on fractional differential. With the rapid development of pattern recognition technology, machine learning methods have also been applied to the detection and identification of pavement cracks.

Pattern recognition is also often called pattern classification. From the perspective of the nature of the problem and the method of solving the problem, pattern recognition is divided into supervised classification (Supervised Classification) and unsupervised classification (Unsupervised Classification). The main difference between the two lies in whether the category of each experimental sample is known in advance; generally speaking, supervised classification often requires a large number of samples of known categories.

In recent years, various pattern recognition methods have been applied to the detection and identification of pavement cracks. Due to the complex noise components of the actually collected pavement images, many methods require preprocessing to eliminate the influence of part of the noise. The effect depends heavily on image preprocessing.

Contents of the invention

The purpose of the present invention is to provide a pavement crack detection and recognition method based on sparse representation classification, so as to overcome the common shortcomings of low detection accuracy and long time consumption in traditional methods.

The above objects of the invention are achieved by the technical features of the independent claims, which the dependent claims develop in an alternative or advantageous manner.

In order to achieve the above object, the present invention proposes a pavement crack detection and recognition method based on sparse representation classification, which includes the following steps:

1) Select the training set, extract the feature vector set of the training set and normalize;

2) mark the sub-blocks of the training set, and divide them into non-crack sub-blocks and crack sub-blocks;

3) Divide the test picture into sub-blocks of specified size, extract the features of each sub-block and normalize;

4) According to the feature vector of the test set, use SRC to classify each sub-block of the image to obtain the sub-block

type matrix;

5) performing horizontal and vertical mapping coding on the sub-block type matrix and performing coding enhancement;

6) Use the code processed in step 5) to identify the type of fracture.

It should be understood that all combinations of the foregoing concepts, as well as additional concepts described in more detail below, may be considered part of the inventive subject matter of the present disclosure, provided such concepts are not mutually inconsistent. Additionally, all combinations of claimed subject matter are considered a part of the inventive subject matter of this disclosure.

The foregoing and other aspects, embodiments and features of the present teachings can be more fully understood from the following description when taken in conjunction with the accompanying drawings. Other additional aspects of the invention, such as the features and/or advantages of the exemplary embodiments, will be apparent from the description below, or learned by practice of specific embodiments in accordance with the teachings of the invention.

Description of drawings

The figures are not intended to be drawn to scale. In the drawings, each identical or nearly identical component that is illustrated in various figures may be represented by a like reference numeral. For purposes of clarity, not every component may be labeled in every drawing. Embodiments of the various aspects of the invention will now be described by way of example with reference to the accompanying drawings, in which:

Fig. 1 is a flowchart of pavement crack detection and identification based on sparse representation classification according to some embodiments of the present invention.

Figure 2 is a flowchart of fracture identification in some embodiments of the present invention.

Fig. 3 is an example diagram of mapping encoding and encoding enhancement of a sub-block classification matrix in some embodiments of the present invention. The middle is the sub-block type matrix, the bold part in the middle is the vertical code, the one with the reduced font is the horizontal code, and the outermost bold part is the new code after code enhancement.

Figure 4 is a schematic diagram of the crack identification effect, "o" indicates longitudinal cracks, "x" indicates network cracks, and "*" indicates transverse cracks.

Figure 5 is a schematic diagram of crack types.

detailed description

In order to better understand the technical content of the present invention, specific embodiments are given together with the attached drawings for description as follows.

Aspects of the invention are described in this disclosure with reference to the accompanying drawings, which show a number of illustrated embodiments. Embodiments of the present disclosure are not necessarily intended to include all aspects of the invention. It should be appreciated that the various concepts and embodiments described above, as well as those described in more detail below, can be implemented in any of numerous ways, since the concepts and embodiments disclosed herein are not limited to any implementation. In addition, some aspects of the present disclosure may be used alone or in any suitable combination with other aspects of the present disclosure.

According to the embodiment of the present invention, the pavement crack detection and recognition method based on sparse representation classification of the present invention introduces a sparse representation classifier and uses the high-order moment features of image sub-blocks as the basis for classifier classification, without the need for Image preprocessing has greatly improved execution efficiency and recognition accuracy. Its specific implementation mainly includes the extraction and normalization of the feature vectors of the training set (sub-block), extracting features from the test image into blocks and classifying them using SRC, and identifying crack types according to the mapping code of the sub-block classification results.

Some exemplary embodiments of the present invention will be described below with reference to the accompanying drawings.

According to an embodiment of the present invention, a fast pedestrian detection method based on object-likeness estimation is used to overcome the problem of too slow detection speed of the existing sliding window-based pedestrian detection method. As shown in Figure 1, the implementation of this method roughly includes the following six steps:

1) Select the training set, extract the feature vector set of the training set and normalize;

2) mark the sub-blocks of the training set, and divide them into non-crack sub-blocks and crack sub-blocks;

3) Divide the test picture into sub-blocks of specified size, extract the features of each sub-block and normalize;

4) According to the feature vector of the test set, use SRC to classify each sub-block of the image to obtain the sub-block

type matrix;

5) performing horizontal and vertical mapping coding on the sub-block type matrix and performing coding enhancement;

6) Use the code processed in step 5) to identify the type of fracture.

In the above method, the step 1) is specifically:

11) Select multiple pictures from the collected pictures and divide them into sub-blocks of specified size n×n, and the sub-blocks are divided into crack sub-blocks and non-crack sub-blocks;

When the sub-block is too large, local features (small cracks) may be ignored, resulting in the sub-blocks containing cracks not being detected, and the recall rate is reduced. If the sub-block is too small, in addition to the large amount of data to be processed, the efficiency will be reduced, and some noise will be misjudged as cracks, which will reduce the precision rate; in order to ensure the precision and recall rate, the sub-block size is set to 75*75.

12) Select m (m≥100) sub-blocks in the sub-blocks of step 11) as the training set, extract the feature vector (stdM3M4) of the sub-block, std is the standard deviation, and M3 and M4 are the third-order moment features of the sub-block image and fourth moment features;

Using the moment feature as the feature of the classifier makes the classifier quite robust to noise, does not require preprocessing, and improves operating efficiency.

13) normalize the feature vector of each sub-block;

In the above method, the step 12) is specifically:

121) Assuming that the sub-block is l(n×n), the standard deviation feature std extraction method of the sub-block is as formula (1);

$\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; d</span>}<span\; class="notranslate">\; =</span>\sqrt{\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}<span\; class="notranslate">\; -</span>{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}{{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; 1</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 1</span>}<span\; class="notranslate">\; )</span>$

in, $mea\mathrm{no}\mathrm{the\; s}=\frac{{\mathrm{\&Sigma;}}_{i=1}^{\mathrm{no}}{\mathrm{\&Sigma;}}_{j=1}^{\mathrm{no}}l(i,j)}{{\mathrm{no}}^2};$

123) The moment feature extraction method of the sub-block is as formula (2). When k=3, it is the third-order moment feature, and when k=4, it is the fourth-order moment feature.

${\mathrm{<span\; class="notranslate">\; m</span>}}_{\mathrm{<span\; class="notranslate">\; k</span>}}<span\; class="notranslate">\; =</span>\frac{{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{\mathrm{<span\; class="notranslate">\; \&Sigma;</span>}}_{\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; 1</span>}}^{\mathrm{<span\; class="notranslate">\; no</span>}}{<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; l</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; i</span>}<span\; class="notranslate">\; ,</span>\mathrm{<span\; class="notranslate">\; j</span>}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span>\mathrm{<span\; class="notranslate">\; m</span>}\mathrm{<span\; class="notranslate">\; e</span>}\mathrm{<span\; class="notranslate">\; a</span>}\mathrm{<span\; class="notranslate">\; no</span>}\mathrm{<span\; class="notranslate">\; the\; s</span>}<span\; class="notranslate">\; )</span>}^{\mathrm{<span\; class="notranslate">\; k</span>}}}{{\mathrm{<span\; class="notranslate">\; no</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 2</span>}<span\; class="notranslate">\; )</span>$

In the above method, the step 2) is specifically:

21) mark the sub-blocks of the test set, and divide them into non-crack sub-blocks and crack sub-blocks;

22) According to the labeled groundtruth, set the sub-block labels of each test set to 0 and 1, 0 indicates a non-cracked sub-block, and 1 indicates a cracked sub-block.

5. The method for detecting and identifying pavement cracks based on sparse representation classification according to claim 4, wherein said step 3) specifically comprises the following steps:

31) Divide the test picture P(W×L) into sub-blocks of specified size, the size of the sub-blocks is the same as in step 11), and obtain K=M×N (M=W/n, N=L/n) sub-blocks;

32) Extract the feature composition vector (stdM3M4) of each sub-block;

33) Eigenvector normalization;

In the above method, the step 4) is specifically:

41) For different types of road surfaces, extract the feature vector set of the training set according to step 1);

42) Use SRC to complete the classification of the K sub-blocks of the test picture.

43) According to the classification result of step 42), the sub-block type matrix p(M×N) is obtained, which represents the classification of each sub-block of the test picture, p(i, j) (i=0, 1...M; j=0 , 1...N) are 0 or 1, 0 means non-crack sub-block, 1 means crack sub-block.

As shown in Figure 2, in the above method, the step 5) is specifically:

51) For the sub-block type matrix p (M×N) in step 43), perform mapping coding in the horizontal direction and the vertical direction.

x _i represents the number of 1s in the i-th column of the sub-block type matrix, and y _j represents the number of 1s in the j-th row of the sub-block type matrix. As shown in the example in Figure 3;

52) Take x _i as an example, select a range X=[x _id x _i+d ], d is the enhanced distance parameter, perform encoding enhancement according to formula (3), and obtain a new horizontal encoding.

x _i =max(X)-min(X)(3)

As shown in Figure 3, in the example d selects 1;

53) The vertical coding enhancement method is similar to the horizontal coding.

As shown in Figure 4, in the above method, the step 1) is specifically:

61) Count the number sum of sub-block labels as 1;

62) If the sum is 0, the image type is a normal road surface without cracks, output the result, and the method ends;

63) If sum is not 0, find the standard deviation stdx and stdy of horizontal coding and vertical coding respectively, perform step 64) and step 65);

64) Calculate the angle θ between the two values in the Cartesian coordinate system and the horizontal axis by stdx and stdy, as in formula (4);

$\mathrm{<span\; class="notranslate">\; \&theta;</span>}<span\; class="notranslate">\; =</span>\mathrm{<span\; class="notranslate">\; arcsin</span>}<span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; the\; s</span>}\mathrm{<span\; class="notranslate">\; t</span>}\mathrm{<span\; class="notranslate">\; d</span>}\mathrm{<span\; class="notranslate">\; the\; y</span>}<span\; class="notranslate">\; /</span>\sqrt{{\mathrm{<span\; class="notranslate">\; stdy</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}<span\; class="notranslate">\; +</span>{\mathrm{<span\; class="notranslate">\; stdx</span>}}^{\mathrm{<span\; class="notranslate">\; 2</span>}}}<span\; class="notranslate">\; )</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; -</span><span\; class="notranslate">\; (</span>\mathrm{<span\; class="notranslate">\; 4</span>}<span\; class="notranslate">\; )</span>$

65) Perform crack type identification on the image type according to formula (5).

The biased cracks indicate that there are mostly cracks in this direction, while the network cracks have no obvious bias. As shown in Fig. 6, the first three pictures show partial lateral fractures, and the last picture shows network cracks.

To sum up, the present invention provides a pavement crack detection and recognition method based on sparse representation classification. Applying the method of machine learning to the detection and identification of pavement cracks generally has the disadvantages of low detection accuracy and long time consumption. To solve this problem, the present invention introduces a sparse representation classifier (Sparse Representation-based Classifier, namely SRC), by selecting an effective The sub-block high-order moment feature avoids image preprocessing and postprocessing, simplifies the detection steps, and improves operating efficiency. Specifically, it includes: extraction and normalization of the feature vector of the training set (sub-block), extracting features from the test image into blocks and using SRC for classification, and identifying crack types according to the mapping code of the sub-block classification results. Compared with the traditional crack detection vehicle identification method, the method proposed by the invention has higher identification accuracy and execution efficiency.

Although the present invention has been disclosed above with preferred embodiments, it is not intended to limit the present invention. Those skilled in the art of the present invention can make various changes and modifications without departing from the spirit and scope of the present invention. Therefore, the scope of protection of the present invention should be defined by the claims.

Claims (8)

1. A pavement crack detection and identification method based on sparse representation classification is characterized by comprising the following steps:

1) selecting a training set, extracting a feature vector set of the training set and normalizing;

2) labeling the sub-blocks of the training set, and dividing the sub-blocks into non-crack sub-blocks and crack sub-blocks;

3) dividing the test picture into subblocks with specified sizes, extracting the characteristics of each subblock and normalizing;

4) classifying each sub-block of the image by using SRC according to the characteristic vector of the test set to obtain a type matrix of the sub-block;

5) carrying out horizontal and longitudinal mapping coding on the sub-block type matrix and carrying out coding enhancement;

6) and identifying the crack type by using the code processed in the step 5).

2. The sparse representation classification-based pavement crack detection and identification method according to claim 1, wherein the step 1) specifically comprises the following steps:

11) selecting a plurality of pictures from the collected pictures to be divided into subblocks of n multiplied by n with specified sizes, wherein the subblocks are divided into crack subblocks and non-crack subblocks;

12) selecting M subblocks with the number being more than or equal to 100 from the subblocks in the step 11) as a training set, extracting a feature vector (stdM3M4) of the subblocks, wherein std is a standard deviation, and M3 and M4 are third-order moment features and fourth-order moment features of subblock images;

13) the feature vectors of each sub-block are normalized.

3. The pavement crack detection and identification method based on sparse representation classification as claimed in claim 2, wherein the step 12) comprises the following steps:

121) assuming that the subblock is l and the size of the subblock is n multiplied by n, the method for extracting the standard deviation feature std of the subblock is as shown in the formula (1);

$std=\sqrt{\frac{{\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&Sigma;}}_{j=1}^n{\left(l\right(i,j)-means)}^2}{n^2-1}}---\left(1\right)$

wherein, $means=\frac{{\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&Sigma;}}_{j=1}^{n}l\left(i,j\right)}{n^2};$

122) the moment feature extraction method of the sub-blocks is as in formula (2), wherein when k is 3, the moment feature is a third-order moment feature, and when k is 4, the moment feature is a fourth-order moment feature:

$m_k=\frac{{\mathrm{\&Sigma;}}_{i=1}^n{\mathrm{\&Sigma;}}_{j=1}^n{\left(l\right(i,j)-means)}^k}{n^2}.---\left(2\right)$

4. the pavement crack detection and identification method based on sparse representation classification as claimed in claim 3, wherein the step 2) specifically comprises the following steps:

21) labeling the sub-blocks of the test set, and dividing the sub-blocks into non-crack sub-blocks and crack sub-blocks;

22) the labels of the sub-blocks of each test set are set to 0 and 1 according to the labeled grountruth, with 0 representing a non-fractured sub-block and 1 representing a fractured sub-block.

5. The pavement crack detection and identification method based on sparse representation classification as claimed in claim 4, wherein the step 3) specifically comprises the following steps:

31) dividing the test picture P with the size of W multiplied by L into subblocks with the specified size, wherein the subblock size is the same as that in the step 11), and obtaining K multiplied by N subblocks, M multiplied by W/N, and N multiplied by L/N;

32) extracting a feature composition vector (stdM3M4) of each sub-block;

33) and normalizing the feature vector.

6. The pavement crack detection and identification method based on sparse representation classification as claimed in claim 5, wherein the step 4) specifically comprises the following steps:

41) for different types of pavements, extracting a feature vector set of a training set according to the step 1);

42) finishing classification of K sub-blocks of the test picture by using the SRC;

43) obtaining a sub-block type matrix p (M × N) according to the classification result in step 42), which represents the classification condition of each sub-block of the test picture, where p (i, j) is 0 or 1,0 represents a non-crack sub-block, 1 represents a crack sub-block, and i is 0,1, 2.

7. The pavement crack detection and identification method based on sparse representation classification as claimed in claim 6, wherein the step 5) specifically comprises the following steps:

51) mapping and coding the subblock type matrix p (M × N) in the step 43) in the horizontal direction and the vertical direction to obtain horizontal codes (x)₁x₂……x_N) And vertical coding (y)₁y₂……y_M)；

x_iIndicates the number of i-th column 1 in the subblock type matrix, y_jIndicating the number of 1's in the jth row of the subblock type matrix.

52) For x_iSelecting a range of X ═ X_i-dx_i+d]D is an enhancement distance parameter, and coding enhancement is carried out according to a formula (3) to obtain a new horizontal code:

x_i＝max(X)-min(X)(3)

53) vertical coding enhancement is performed in the same manner as horizontal coding described above.

8. The pavement crack detection and identification method based on sparse representation classification as claimed in claim 7, wherein the step 6) specifically comprises the following steps:

61) counting the number sum of sub-block labels of 1;

62) if sum is 0, the image type is a normal road surface, no crack exists, a result is output, and the method is ended;

63) if sum is not 0, finding the standard deviations stdx and stdy of the horizontal encoding and the vertical encoding, respectively, and performing step 64) and step 65);

64) and (3) calculating an included angle theta between the two values and a horizontal axis in the rectangular coordinate system through stdx and stdy, as shown in a formula (4):

$\mathrm{\&theta;}=\arcsin (stdy/\sqrt{{\mathrm{stdy}}^2+{\mathrm{stdx}}^2})---\left(4\right)$

65) and (3) identifying the type of the image according to a formula (5):