A deep foundation pit crack detection method and device for deep foundation pit tunneling

Abstract

The present invention relates to the technical field of image optimization and recognition, and particularly relates to a deep foundation pit crack detection method and device for deep foundation pit tunneling. The method obtains a deep foundation pit image and the motion vector of an image acquisition device. Blurred images in the deep foundation pit image are screened out, and rigid edges are identified. Using the positional relationship between other edges and rigid edges, the stability of pixel value changes on the path, and the degree of association between the edge vector and the motion vector, smear edges are screened out. Furthermore, a two-dimensional blur kernel is constructed, and the blur kernel value is adaptively set using the gradient information and position information of pixel values within the two-dimensional blur kernel. The blurred image is de-blurred through the two-dimensional blur kernel, and crack recognition is performed. By extracting and analyzing the feature information in the blurred image, quantifying the smear effect in the image, and then adaptively determining the blur kernel in the image and performing de-blurring, the present invention improves the quality of the deep foundation pit image and the accuracy of the crack recognition result.

Description

Technical Field

[0001] The present invention relates to the technical field of image optimization and recognition, and particularly to a method and device for detecting deep foundation pit cracks for deep foundation pit tunneling. Background Art

[0002] Deep foundation pit engineering is a key link in infrastructure construction. During the tunneling process of deep foundation pits, due to factors such as soil stress changes, groundwater effects, and construction techniques, cracks may occur in the foundation pit structure. The occurrence of cracks not only reduces the stability and bearing capacity of the foundation pit but may also trigger serious safety accidents such as soil collapse and groundwater leakage, posing a huge threat to the safety of surrounding buildings, underground pipelines, and construction workers' lives and property.

[0003] In the prior art, image processing means can be used to identify and locate cracks by monitoring images in deep foundation pits. However, during the construction process, vibrations caused by large construction equipment such as excavator operations, pile driver vibrations, and earthwork transport vehicles are inevitable environmental factors. The generation of vibrations will cause the image acquisition device to vibrate as well, resulting in blurred ghosting in the images. Moreover, due to the randomness of vibrations, the degree of image blurring and the direction of ghosting vary randomly at different times. If a fixed blur kernel is used to deblur the images, the deblurring effect of the images will be poor, affecting the accuracy of crack identification and detection. Summary of the Invention

[0004] In order to solve the technical problem in the prior art that using a fixed blur kernel to deblur deep foundation pit images results in a poor deblurring effect and affects the accuracy of crack identification, the purpose of the present invention is to provide a method and device for detecting deep foundation pit cracks for deep foundation pit tunneling. The specific technical solutions adopted are as follows:

[0005] The present invention proposes a method for detecting deep foundation pit cracks for deep foundation pit tunneling, and the method includes:

[0006] Obtain a deep foundation pit image collected by an image acquisition device in the deep foundation pit; the image acquisition device is integrated with a sensor for detecting motion vectors generated by vibrations.

[0007] Obtain a blurred image in the deep foundation pit image when vibrations occur; according to the position of each edge and the saliency of gradient information in the blurred image, select a rigid edge.

[0008] For each edge, obtain the first ghosting probability of each edge according to the positional relationship between the edge and the rigid edge, and the stability of pixel value changes on the path; obtain the second ghosting probability of each edge according to the correlation degree between the edge vector from the rigid edge to the edge and the motion vector; screen out the ghosting edges from all edges according to the first ghosting probability and the second ghosting probability.

[0009] Obtain the trailing distance and trailing direction between the rigid edge and the trailing edge; set the size of the two-dimensional blur kernel according to the trailing distance. Within the two-dimensional blur kernel, determine whether the pixel is within the trailing range, and obtain the blur kernel value at the pixel position according to the gradient of the pixel and the size of the two-dimensional blur kernel; use the two-dimensional blur kernel to deblur the blurred image.

[0010] Perform crack identification on the deblurred deep foundation pit image.

[0011] Further, the method for obtaining the blurred image includes:

[0012] Convolve the deep foundation pit image using the Laplace operator to obtain a judgment image. If the energy value in the judgment image is lower than a preset energy threshold, use the deep foundation pit image corresponding to the judgment image as the blurred image.

[0013] Further, the method for screening the rigid edge includes:

[0014] For each edge, use the distance between the edge and the center point of the image as the reference distance, and use the ratio of the length of the edge to the reference distance as the position and shape parameter of the edge; use the ratio of the overall gradient amplitude on the edge to the gray variance as the first gradient significance; use the perpendicular degree between the overall gradient direction on the edge and the tangent direction of the edge as the second gradient significance; obtain the rigidity degree of each edge according to the position and shape parameter, the first gradient significance, and the second gradient significance; select the edge with the maximum rigidity degree in the blurred image as the rigid edge.

[0015] Further, the method for obtaining the first trailing probability includes:

[0016] For each edge, use the included angle between the edge trend of the edge and the rigid edge as the position included angle; map each point on the edge to each point on the rigid edge, and use the path between the two mapped points as the mapping path; obtain the gradient amplitude sequence and the gradient direction angle sequence of the pixel points on the mapping path; obtain the change stability according to the differences between the elements in the gradient amplitude sequence and the gradient direction angle sequence; obtain the first trailing probability according to the change stability and the position included angle.

[0017] Further, the method for obtaining the second trailing probability includes:

[0018] For each edge, the sum vector of the vectors formed by the mapping paths of the edges pointing in the rigid edge direction is used as the edge vector between the rigid edge and the edge; taking the sum of the magnitudes of the edge vector and the motion vector as the denominator and the magnitude of the difference vector between the edge vector and the motion vector as the numerator, the second ghosting probability is obtained.

[0019] Further, the method for screening ghosting edges includes:

[0020] Taking the sum value of the first ghosting probability and the second ghosting probability as the overall ghosting probability of each edge, and selecting the edge with the largest overall ghosting probability as the ghosting edge of the rigid edge.

[0021] Further, the method for obtaining the ghosting distance and the ghosting direction includes:

[0022] Taking the magnitude of the edge vector between the rigid edge and the ghosting edge as the ghosting distance, and taking the direction of the edge vector between the rigid edge and the ghosting edge as the ghosting direction.

[0023] Further, the method for obtaining the blur kernel value includes:

[0024] For a pixel point in the two-dimensional blur kernel, taking the line segment from the pixel point to the center point of the two-dimensional blur kernel as the judgment line segment. If the projection length of the judgment line segment in the ghosting direction is less than or equal to the ghosting distance, it is determined that the pixel point is within the ghosting range. Taking the reciprocal of the ghosting distance as the distance weight, multiplying the negative correlation mapping and normalization of the gradient amplitude of the pixel point by the distance weight to obtain the blur kernel value at the position of the pixel point; if it is determined that the pixel point is not within the ghosting range, setting the blur kernel value at the position of the pixel point to 0.

[0025] Further, a pre-trained crack detection neural network model is used to identify cracks in the de-blurred deep foundation pit image.

[0026] The present invention also provides a deep foundation pit crack detection device for deep foundation pit tunneling, and the device includes:

[0027] A deep foundation pit information collection module, configured to obtain a deep foundation pit image collected by an image collection device in the deep foundation pit; the image collection device is integrated with a sensor for detecting the motion vector generated by vibration;

[0028] A first deep foundation pit image analysis module, configured to obtain a blurred image in the deep foundation pit image when vibration occurs; and screen out a rigid edge according to the position of each edge and the saliency of the gradient information in the blurred image;

[0029] The second deep foundation pit image analysis module is used to obtain the first smear probability of each edge according to the positional relationship between the edge and the rigid edge, as well as the stability of the pixel values on the path; obtain the second smear probability of each edge according to the correlation degree between the edge vector from the rigid edge to the edge and the motion vector; screen out the smeared edges from all edges according to the first smear probability and the second smear probability.

[0030] The image enhancement module is used to obtain the first smear probability of each edge according to the positional relationship between the edge and the rigid edge, as well as the stability of the pixel values on the path; obtain the second smear probability of each edge according to the correlation degree between the edge vector from the rigid edge to the edge and the motion vector; screen out the smeared edges from all edges according to the first smear probability and the second smear probability.

[0031] The crack recognition module is used to perform crack recognition on the deblurred deep foundation pit image.

[0032] The present invention has the following beneficial effects:

[0033] The present invention first takes into account the characteristics of high contrast and high saliency information of the rigid edges in the deep foundation pit image, which can be used to analyze the degree of smearing. Therefore, the present invention selects a rigid edge in the blurred image for smearing analysis according to the position and gradient information of each edge. The smeared edge should have an approximately parallel positional relationship with the rigid edge, and the smearing path can be regarded as the effect of the rigid edge blurring. Therefore, the pixel value information on the smearing path should follow certain rules, so the first smear probability can be obtained; further combined with the actual motion vector of the device feedback by the sensor, the second smear probability is determined by using the correlation between the edge vector and the motion vector, and then the smeared edge of the rigid edge is determined. Using the smear distance and smear direction, the two-dimensional blur kernel at each position in the blurred image can be adaptively determined. Deblurring the blurred image with the adaptive two-dimensional blur kernel can obtain a deep foundation pit image with significant feature information, which is then applied to crack recognition. The present invention extracts and analyzes the feature information in the blurred image, quantifies the smearing effect in the image, and then adaptively determines the blur kernel in the image and performs deblurring, improving the quality of the deep foundation pit image and the accuracy of the crack recognition result. Description of the Drawings

[0034] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the following will briefly introduce the drawings required for the description of the embodiments or the prior art. Obviously, the following drawings are only some embodiments of the present invention. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0035] Figure 1 Flowchart of a deep foundation pit crack detection method provided by an embodiment of the present invention for deep foundation pit tunneling;

[0036] Figure 2 Schematic diagram of mapping of a rigid edge to other edges provided by an embodiment of the present invention;

[0037] Figure 3 A blurred image provided by an embodiment of the present invention;

[0038] Figure 4 Provided by an embodiment of the present invention Figure 3 Deblurring effect diagram. Detailed implementation manners

[0039] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the intended invention purpose, the following, in conjunction with the accompanying drawings and preferred embodiments, details a deep foundation pit crack detection method and device according to the present invention, its specific implementation manners, structures, features and effects. In the following description, different "one embodiment" or "another embodiment" do not necessarily refer to the same embodiment. In addition, the specific features, structures or characteristics in one or more embodiments can be combined in any suitable form.

[0040] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by those of ordinary skill in the technical field to which the present invention belongs.

[0041] The following specifically describes the specific solutions of a deep foundation pit crack detection method and device provided by the present invention in conjunction with the accompanying drawings.

[0042] Please refer to Figure 1 , which shows a flowchart of a deep foundation pit crack detection method provided by an embodiment of the present invention for deep foundation pit tunneling. The method includes:

[0043] Step S1: Obtain a deep foundation pit image collected by an image acquisition device in the deep foundation pit; the image acquisition device is integrated with a sensor for detecting motion vectors generated by vibrations.

[0044] In the embodiments of the present invention, the image acquisition device is integrated with sensors such as an acceleration sensor and a gyroscope for detecting the vibration intensity and direction. After the sensors collect the high-frequency data of the vibration, the Kalman filtering algorithm can be used to denoise and integrate the acceleration data. Combining the rotation angle information measured by the gyroscope can determine the actual spatial position and attitude of the image acquisition device at each moment. By comparing the position and attitude of the image acquisition device at adjacent moments, it can be determined whether vibration has occurred and the motion vector can be determined. The specific data processing algorithm is a well-known technical means for those skilled in the art and will not be elaborated here. It should be noted that, for the convenience of subsequent image processing operations, after obtaining the motion vector of the device in reality, it needs to be converted into the image coordinate system to facilitate the subsequent correlation analysis of the edge vectors.

[0045] The image acquisition device can be used to acquire the deep foundation pit images in the deep foundation pit, and at the same time, the motion vector at each moment can be acquired during vibration, where the motion vector has the same time sequence as the deep foundation pit image.

[0046] Step S2: Obtain the blurred image in the deep foundation pit image when vibration occurs; select a rigid edge according to the position of each edge and the saliency of the gradient information in the blurred image.

[0047] During the vibration process, since the image acquisition device will move due to vibration, the captured image will have motion blur. The motion blur is manifested as the attenuation of the overall gradient in the image. Because of the motion blur, the significant edge information in the image is weakened, resulting in unclear gradients and smoothed edge details, that is, the high-frequency components in the image will decrease. Therefore, the blurred image is selected from the deep foundation pit images, that is, the moment corresponding to the blurred image is the moment when obvious vibration occurs in the deep foundation pit.

[0048] Preferably, in the embodiments of the present invention, the blurred image in the deep foundation pit image can be automatically selected. Since the high-frequency information in the blurred image will be significantly reduced, the embodiments of the present invention use the Laplace operator to perform convolution on the deep foundation pit image to obtain a judgment image. For the judgment image, if the deep foundation pit image corresponding to the judgment image is a blurred image, its high-frequency components are less, the pixel values change smoothly, and the obtained judgment image will have a smaller energy value. Therefore, the embodiments of the present invention set an energy threshold. If the energy value in the judgment image is lower than the preset energy threshold, the deep foundation pit image corresponding to the judgment image is used as the blurred image. In the embodiments of the present invention, the energy threshold is set to 1200. The specific method of Laplace operator convolution and image energy acquisition is a well-known technical means for those skilled in the art and will not be elaborated here.

[0049] It should be noted that in the embodiments of the present invention, if it is determined that the deep foundation pit image at a certain moment is a blurred image, the motion vector collected by the corresponding image acquisition device can be called for subsequent step analysis.

[0050] In a blurred image, the trailing shadow therein is essentially a linear diffusion of the original edge in the vibration direction. The intensity of the diffusion is positively correlated with the vibration amplitude and frequency, and the diffusion direction is also close to the vibration direction. In order to accurately analyze the degree of trailing shadow in the blurred image, it is necessary to analyze with a representative edge with significant information, that is, the rigid edge extracted in the embodiments of the present invention. The rigid edge is the edge with prominent information under the original engineering structure in the deep foundation pit, which has an obvious high contrast. Before and after the vibration occurs, the rigid edge is not a clear and significant edge in the image. The trailing shadow will cause a blurring effect on the edge. The originally insignificant edge will become more blurred under the influence of the trailing shadow. The edge formed by the trailing shadow is more distorted relative to the original edge and is not easy to be used for reference analysis; while the rigid edge, because of its prominent edge features, has strong information reference in the image, and the blurring and distortion effects caused by the trailing shadow on it are easier to observe. Therefore, the rigid edge can be used as a reference edge for analyzing the trailing shadow.

[0051] For each edge in the blurred image, the more it is located in the middle area and the longer the edge length in the blurred image, it indicates that the position and scale of this position and edge are more conducive to analyzing the degree of trailing shadow; and the greater the significance of the gradient information on the edge, it indicates that the edge information is more significant, and the more likely it is a rigid edge. Therefore, in the embodiments of the present invention, a rigid edge is selected according to the position of each edge in the blurred image and the significance of the gradient information for subsequent step analysis.

[0052] Preferably, in the embodiments of the present invention, the method for screening the rigid edge includes:

[0053] For each edge, the distance between the edge and the center point of the image is used as the reference distance, and the ratio of the edge length to the reference distance is used as the position and shape parameter of the edge. The smaller the reference distance, it indicates that the edge is closer to the center of the image. At the same time, the larger the edge length, it indicates that the edge is easier to observe the trailing shadow. The larger the position and shape parameter, the more likely the edge is a rigid edge.

[0054] The ratio of the overall gradient amplitude on the edge to the gray variance is used as the first gradient significance. That is, the smaller the gray variance, it indicates that the gray information on the edge is more stable. At the same time, the larger the overall gradient amplitude, it indicates that the edge is clearer. The larger the first gradient significance, it indicates that the edge is more likely to be a rigid edge.

[0055] The perpendicular degree between the overall gradient direction on the edge and the edge tangent direction is used as the second gradient saliency. That is, the second gradient saliency is analyzed based on the gradient direction. Since a rigid edge is an edge with relatively prominent gradient information, its gradient direction should be significantly perpendicular to the edge tangent, indicating that the edge pixel points at this position have significant gradient changes. Therefore, the greater the perpendicular degree, the greater the second gradient saliency, and the more likely this edge is a rigid edge.

[0056] Obtain the rigidity degree of each edge according to the position morphology parameter, the first gradient saliency, and the second gradient saliency; select the edge with the greatest rigidity degree in the blurred image as the rigid edge. In the embodiment of the present invention, after quantifying the position morphology parameter, the first gradient saliency, and the second gradient saliency respectively, the product of the three is used as the rigidity degree.

[0057] In the embodiment of the present invention, the overall gradient direction is the direction corresponding to the sum vector of the unit vectors corresponding to the gradient directions of each point on the edge. Similarly, the overall gradient amplitude is the average value of the gradient amplitudes of each point on the edge. Obtain the acute angle between the overall gradient direction and the edge tangent direction, and use the sine value of this angle as the perpendicular degree. That is, the closer this angle is to 90 degrees, the greater the sine value, and the greater the perpendicular degree, indicating that the gradient direction is more perpendicular to the edge tangent.

[0058] Step S3: For each edge, obtain the first ghosting probability of each edge according to the position relationship between the edge and the rigid edge, and the change stability of the pixel values on the path; obtain the second ghosting probability of each edge according to the correlation degree between the edge vector from the rigid edge to the edge and the motion vector; screen out the ghosting edges among all edges according to the first ghosting probability and the second ghosting probability.

[0059] After obtaining the rigid edge in step S2, the corresponding ghosting edge can be further determined. The ghosting edge is the edge after the end of the ghosting generated by the vibration of this rigid edge, that is, the ghosting edge is the boundary of the ghosting of the rigid edge. Since this boundary is generated by the ghosting of the rigid edge, it has an edge trend and pixel value characteristics similar to those of the rigid edge. And the ghosting edge and the rigid edge are the ghosting path, which can be regarded as generated by the rigid edge moving in the vibration direction. Then, the pixel value changes on the path should be relatively similar and stable.

[0060] Therefore, in order to screen out the ghosting edges of the rigid edge in the blurred image, for each edge, first obtain the first ghosting probability of each edge according to the position relationship between the edge and the rigid edge, and the change stability of the pixel values on the path. That is, the more parallel an edge is to the rigid edge, the more likely it is that this edge is the ghosting edge generated by the movement of the rigid edge in the vibration direction; the greater the change stability of the pixel values on the path, the more likely it is that this edge is the boundary of the ghosting of the rigid edge.

[0061] Further, in combination with the motion vectors obtained by the image acquisition device, since the motion vectors have a clear direction and magnitude, which represent the actual motion generated by the image acquisition device under the influence of vibration, and the image field of view will move with the movement of the device, thereby generating smear in the image. That is, the smear direction in the image is related to the movement direction of the device. Therefore, in the embodiments of the present invention, the edge vectors from the rigid edge to each edge are further obtained, and the degree of association between the edge vector and the motion vector is determined. The greater the degree of association, the greater the second smear probability of the edge, and the more likely the edge is a smear edge.

[0062] Combining the first smear probability and the second smear probability can screen out the smear edges among all edges.

[0063] Preferably, in the embodiments of the present invention, the method for obtaining the first smear probability includes:

[0064] For each edge, the angle between the edge trend of the edge and the rigid edge is used as the position angle. That is, the smaller the position angle, the more parallel the two edges are, and the more likely the edge is a smear edge of the rigid edge. It should be noted that the method for obtaining the angle between the edge trends can linearly fit the two edges respectively, obtain the angles of the fitted lines relative to the horizontal direction, and use the angle difference between the two fitted lines as the position angle.

[0065] Further, in order to facilitate the analysis of whether the area between the edge and the rigid edge is a smear area. Map each point on the edge to each point on the rigid edge, and the path between the two mapped points is used as the mapping path. Please refer to Figure 2 , which shows a schematic diagram of the mapping of a rigid edge and other edges provided by an embodiment of the present invention. Figure 2 The left edge i1 in is an edge that is relatively parallel to the rigid edge and has a relatively high probability of being a smear edge; the right edge i2 is an edge with a relatively large angle between the edge trend and the rigid edge and a relatively low probability of being a smear edge. Through mapping, the information change on the path can be analyzed to confirm whether it is the path in the smear area.

[0066] Obtain the gradient magnitude sequence and the gradient direction angle sequence of the pixel points on the mapping path. If the mapping path is the path of the smear area, the gradient information on the path is generated by the rigid edge, and the gradient magnitude and the gradient direction are relatively stable. Therefore, the change stability is obtained according to the differences between the elements in the gradient magnitude sequence and the gradient direction angle sequence. In the embodiments of the present invention, the standard deviations of the two sequences are multiplied and then negatively correlated and normalized to obtain the change stability.

[0067] The first smear probability is obtained based on the change stability and the position included angle. In the embodiments of the present invention, the greater the change stability and the smaller the position included angle, the greater the first smear probability of the edge. Therefore, after performing a negative correlation mapping and normalization on the value of the position included angle, it is multiplied by the change stability to obtain the first smear probability.

[0068] It should be noted that negative correlation mapping and normalization are mathematical means well-known to those skilled in the art. For example, it can be implemented through an exponential function with the natural constant as the base, taking the opposite number of the data as the power of the exponential function, and the function mapping result is the result after negative correlation mapping and normalization. Those skilled in the art can flexibly select other algorithms for negative correlation mapping and normalization, which will not be limited and elaborated here, and the subsequent content related to relevant algorithms in the embodiments of the present invention will not be elaborated either.

[0069] Preferably, in the embodiments of the present invention, the method for obtaining the second smear probability includes:

[0070] For each edge, since there are multiple mapping paths, the sum vector of the vectors formed by the mapping paths pointing from the edge to the rigid edge direction is used as the edge vector between the rigid edge and the edge.

[0071] If an obvious vibration occurs at a certain moment, there will be an obvious smear distance between the rigid edge and the smear edge in the blurred image, and the smear direction is significantly correlated with the vibration direction. To quantify this correlation, the edge vector in the embodiments of the present invention is the direction from the edge to the rigid edge, that is, if the edge is a smear edge, the direction of the edge vector should be opposite to the direction of the motion vector. Therefore, taking the sum of the magnitudes of the edge vector and the motion vector as the denominator and the magnitude of the difference vector between the edge vector and the motion vector as the numerator, the second smear probability is obtained. If the edge is a smear edge, the denominator and the numerator are equal, that is, the second smear probability is 1. On the contrary, if the edge is not a smear edge, the directions between the edge vector and the motion vector are not opposite, and the magnitude of the obtained difference vector will be less than the sum of the magnitudes of the two adjacent ones, that is, the numerator is less than the denominator, and the second smear probability becomes smaller. That is, the greater the second smear probability, the greater the correlation between the edge vector and the adjacent motion, and the more likely the edge is a smear edge.

[0072] In the embodiments of the present invention, after obtaining the first smear probability and the second smear probability, the sum value of the first smear probability and the second smear probability is used as the overall smear probability of each edge, and the edge with the largest overall smear probability is selected as the smear edge of the rigid edge.

[0073] Step S4: Obtain the smear distance and smear direction between the rigid edge and the smear edge; set the size of the two-dimensional blur kernel according to the smear distance. Within the two-dimensional blur kernel, determine whether the pixel is within the smear range, and obtain the blur kernel value at the pixel position according to the gradient of the pixel and the size of the two-dimensional blur kernel; use the two-dimensional blur kernel to deblur the blurred image.

[0074] In a blurred image, the rigid edge and the corresponding smear edge are determined. The distance between the two is the smear distance, and the direction from the rigid edge to the smear edge is the smear direction. The smear direction and smear distance can be used to characterize the smear degree in the blurred image. Thus, a two-dimensional blur kernel can be adaptively constructed. The size of the two-dimensional blur kernel can be set according to the smear distance. That is, if the smear length is L, the size of the two-dimensional blur kernel is L×L.

[0075] Furthermore, the blur kernel value at each position in the blurred image can be set for the two-dimensional blur kernel. Within the two-dimensional blur kernel, first, it is necessary to determine whether the pixel is within the smear range. If it is within the smear range, then the blur kernel value needs to be determined according to the corresponding gradient information in combination with the size of the two-dimensional blur kernel. If the gradient amplitude at the position of the pixel is large, the blur kernel weight at this position should be correspondingly reduced, so as to reduce the smoothing of this feature during the deblurring process and avoid information distortion. After setting each pixel position, an effective two-dimensional blur kernel can be determined to deblur the blurred image.

[0076] Specifically, in the embodiment of the present invention, the modulus of the edge vector between the rigid edge and the smear edge is used as the smear distance, and the direction of the edge vector between the rigid edge and the smear edge is used as the smear direction.

[0077] In an embodiment of the present invention, the method for obtaining the blur kernel value includes:

[0078] For a pixel within the two-dimensional blur kernel, the line segment from the pixel to the center point of the two-dimensional blur kernel is used as the judgment line segment. If the projection length of the judgment line segment in the smear direction is less than or equal to the smear distance, it is determined that the pixel is within the smear range. Under the linear change caused by the smear, the edge diffusion within the smear range is uniform. Therefore, the reciprocal of the smear distance is used as the distance weight. The gradient amplitude of the pixel is negatively correlated and normalized and then multiplied by the distance weight to obtain the blur kernel value at the pixel position. That is, the gradient information weight at the position of the pixel value obtained after negatively correlating and normalizing the gradient amplitude. The larger the gradient amplitude, the more the smoothing at this position needs to be reduced. By multiplying with the distance weight, the blur kernel value at the pixel position can be obtained.

[0079] If it is determined that the pixel is not within the smear range, the blur kernel value at the pixel position is set to 0.

[0080] In the embodiments of the present invention, after determining the two-dimensional blur kernel at each position in the blurred image, a non-blind area convolution operation can be performed on the blurred image according to the blur kernel to remove the blur information in the image, thereby obtaining a de-blurred image. It should be noted that the core of de-blurring lies in performing inverse convolution operation on the blurred image using the blur kernel to restore the original clear appearance of the image. Specifically, for the blurred image, the inverse process of performing convolution operation on it with the blur kernel. Since the blur of the image essentially means that each pixel point of the image diffuses in space according to the pattern of the blur kernel, de-blurring is to re-aggregate the diffused pixel information according to the reverse pattern of the blur kernel. Please refer to Figure 3 and Figure 4 , Figure 3 which is a blurred image provided by an embodiment of the present invention, Figure 4 is provided by an embodiment of the present invention Figure 3 de-blurring effect diagram. After de-blurring, a deep foundation pit image with significant feature information can be obtained.

[0081] Step S5: Identify cracks in the de-blurred deep foundation pit image.

[0082] After obtaining a deep foundation pit image with significant feature information, the crack detection technology in the prior art can be used to identify cracks. In the embodiments of the present invention, a pre-trained crack detection neural network model is used to identify cracks in the de-blurred deep foundation pit image. The crack detection neural network is a convolutional neural network, and its multiple convolutional layers and pooling layers are used to automatically extract rich texture, shape and other features in the image. Then, a large number of labeled image data containing normal areas and various crack conditions of the deep foundation pit are collected and used as a training set to train the neural network. During the training process, the neural network continuously adjusts its internal parameters to learn the feature patterns of cracks in the image, so that the prediction results output by the model are as close as possible to the labeled true crack conditions. When the model training is completed and reaches a certain accuracy, the de-blurred deep foundation pit image is input into the neural network model. Finally, the prediction results about whether there are cracks in the image, the positions and types of cracks, etc. are output. It can quickly and accurately judge the actual conditions of cracks in the deep foundation pit, provide a key basis for engineering safety assessment and subsequent processing, and effectively improve the efficiency and reliability of deep foundation pit crack monitoring.

[0083] In summary, the present invention obtains deep foundation pit images and the motion vectors of the image acquisition device. Blurred images in the deep foundation pit images are screened out, and rigid edges are identified. By using the positional relationship between other edges and the rigid edges, the stability of pixel value changes on the path, and the degree of correlation between the edge vectors and the motion vectors, the trailing edges are screened out. Based on the rigid edges and the corresponding trailing edges, the trailing distance and the trailing direction can be determined, and then a two-dimensional blur kernel is constructed. Moreover, the blur kernel values are adaptively set by using the gradient information and positional information of the pixel values within the two-dimensional blur kernel. The blurred images are deblurred by the two-dimensional blur kernel, and then crack identification is performed. By extracting and analyzing the feature information in the blurred images, the present invention quantifies the trailing effect in the images, and then adaptively determines the blur kernel in the images and performs deblurring, improving the quality of the deep foundation pit images and the accuracy of the crack identification results.

[0084] Based on the same inventive concept, the present invention also provides a deep foundation pit crack detection device for deep foundation pit tunneling, which comprises:

[0085] A deep foundation pit information acquisition module, configured to obtain deep foundation pit images acquired by an image acquisition device within the deep foundation pit; the image acquisition device is integrated with a sensor for detecting the motion vectors generated by vibrations;

[0086] A first deep foundation pit image analysis module, configured to obtain the blurred images in the deep foundation pit images when vibrations occur; and screen out a rigid edge according to the position of each edge in the blurred image and the saliency of the gradient information;

[0087] A second deep foundation pit image analysis module, configured to, for each edge, obtain the first trailing probability of each edge according to the positional relationship between the edge and the rigid edge, and the stability of the pixel values on the path; obtain the second trailing probability of each edge according to the degree of correlation between the edge vector from the rigid edge to the edge and the motion vector; and screen out the trailing edges from all the edges according to the first trailing probability and the second trailing probability;

[0088] An image enhancement module, configured to, for each edge, obtain the first trailing probability of each edge according to the positional relationship between the edge and the rigid edge, and the stability of the pixel values on the path; obtain the second trailing probability of each edge according to the degree of correlation between the edge vector from the rigid edge to the edge and the motion vector; and screen out the trailing edges from all the edges according to the first trailing probability and the second trailing probability;

[0089] A crack identification module, configured to perform crack identification on the deblurred deep foundation pit images.

[0090] It should be noted that the above-mentioned sequence of embodiments of the present invention is only for description and does not represent the superiority or inferiority of the embodiments. The processes depicted in the accompanying drawings do not necessarily require the specific order or sequential order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0091] Each embodiment in this specification is described in a progressive manner. For the same or similar parts among the embodiments, reference can be made to each other. Each embodiment focuses on the differences from other embodiments.

Claims

1. A deep foundation pit crack detection method for deep foundation pit tunneling, characterized in that, The method includes: Obtaining a deep foundation pit image collected by an image acquisition device in the deep foundation pit; the image acquisition device is integrated with a sensor for detecting a motion vector generated by vibration. Obtaining a blurred image in the deep foundation pit image when vibration occurs; screening out a rigid edge according to the position of each edge and the saliency of gradient information in the blurred image. For each edge, obtaining a first ghosting probability of each edge according to the positional relationship between the edge and the rigid edge, and the change stability of pixel values on the path; obtaining a second ghosting probability of each edge according to the correlation degree between the edge vector from the rigid edge to the edge and the motion vector; screening out ghosting edges from all edges according to the first ghosting probability and the second ghosting probability; the motion vector is collected by the image acquisition device and corresponds to the blurred image. Obtaining the ghosting distance and ghosting direction between the rigid edge and the ghosting edge; setting the size of a two-dimensional blur kernel according to the ghosting distance, determining whether a pixel point is within the ghosting range within the two-dimensional blur kernel, and obtaining the blur kernel value of the pixel point position according to the gradient of the pixel point and the size of the two-dimensional blur kernel; using the two-dimensional blur kernel to deblur the blurred image. Performing crack recognition on the deblurred deep foundation pit image.

2. The deep foundation pit crack detection method for deep foundation pit tunneling according to claim 1, wherein, The method for obtaining the blurred image includes: Performing convolution on the deep foundation pit image using a Laplacian operator to obtain a judgment image, and if the energy value in the judgment image is lower than a preset energy threshold, using the deep foundation pit image corresponding to the judgment image as the blurred image.

3. A method for detecting deep foundation pit cracks for deep foundation pit tunneling according to claim 1, characterized in that, The method for screening the rigid edge includes: For each edge, taking the distance between the edge and the center point of the image as a reference distance, and taking the ratio of the length of the edge to the reference distance as the position and shape parameter of the edge; taking the ratio of the overall gradient amplitude on the edge to the gray variance as the first gradient saliency; taking the perpendicular degree between the overall gradient direction on the edge and the tangent direction of the edge as the second gradient saliency; obtaining the rigidity degree of each edge according to the position and shape parameter, the first gradient saliency, and the second gradient saliency; selecting the edge with the largest rigidity degree in the blurred image as the rigid edge.

4. A deep foundation pit crack detection method for deep foundation pit tunneling according to claim 1, characterized in that, The method for obtaining the first ghosting probability includes: For each edge, taking the included angle between the edge trend between the edge and the rigid edge as the position included angle; mapping each point on the edge to each point on the rigid edge, and taking the path between the two mapped points as the mapping path; obtaining the gradient amplitude sequence and the gradient direction angle sequence of the pixel points on the mapping path; obtaining the change stability according to the difference between the elements in the gradient amplitude sequence and the gradient direction angle sequence; obtaining the first ghosting probability according to the change stability and the position included angle.

5. A method for detecting deep foundation pit cracks for deep foundation pit tunneling according to claim 4, characterized in that, The method for obtaining the second ghosting probability includes: For each edge, taking the sum vector of the vectors formed by the mapping paths of the edge pointing in the direction of the rigid edge as the edge vector from the rigid edge to the edge; taking the sum of the magnitudes of the edge vector and the motion vector as the denominator, and taking the magnitude of the difference vector between the edge vector and the motion vector as the numerator to obtain the second ghosting probability.

6. A method for detecting deep foundation pit cracks for deep foundation pit tunneling according to claim 1, characterized in that, The screening method for the trailing edge includes: Taking the sum value of the first trailing probability and the second trailing probability as the overall trailing probability of each edge, and selecting the edge with the maximum overall trailing probability as the trailing edge of the rigid edge.

7. A deep foundation pit crack detection method for deep foundation pit tunneling according to claim 5, characterized in that, The method for obtaining the trailing distance and trailing direction includes: Taking the modulus length of the edge vector between the rigid edge and the trailing edge as the trailing distance, and taking the direction of the edge vector between the rigid edge and the trailing edge as the trailing direction.

8. A deep foundation pit crack detection method for deep foundation pit tunneling according to claim 1, characterized in that, The method for obtaining the blurred kernel value includes: For a pixel point in the two-dimensional blurred kernel, taking the line segment from the pixel point to the center point of the two-dimensional blurred kernel as the judgment line segment. If the projection length of the judgment line segment in the trailing direction is less than or equal to the trailing distance, it is determined that the pixel point is within the trailing range. Taking the reciprocal of the trailing distance as the distance weight, multiplying the negative correlation mapping and normalization of the gradient amplitude of the pixel point by the distance weight to obtain the blurred kernel value at the position of the pixel point; if it is determined that the pixel point is not within the trailing range, setting the blurred kernel value at the position of the pixel point to 0.

9. A deep foundation pit crack detection method for deep foundation pit tunneling according to claim 1, characterized in that, Using a pre-trained crack detection neural network model to identify cracks in the de-blurred deep foundation pit image.

10. A deep foundation pit crack detection device for deep foundation pit tunneling, characterized in that, The device includes: A deep foundation pit information collection module for obtaining the deep foundation pit image collected by the image collection device in the deep foundation pit; the image collection device is integrated with a sensor for detecting the motion vector generated by vibration. A first deep foundation pit image analysis module for obtaining the blurred image in the deep foundation pit image when vibration occurs; screening out a rigid edge according to the position of each edge and the significance of the gradient information in the blurred image. A second deep foundation pit image analysis module for, for each edge, obtaining the first trailing probability of each edge according to the positional relationship between the edge and the rigid edge, and the stability of the pixel values on the path; obtaining the second trailing probability of each edge according to the correlation degree between the edge vector from the rigid edge to the edge and the motion vector; screening out the trailing edge among all edges according to the first trailing probability and the second trailing probability; the motion vector is collected by the image collection device and corresponds to the blurred image. An image enhancement module for obtaining the trailing distance and trailing direction between the rigid edge and the trailing edge; setting the size of the two-dimensional blurred kernel according to the trailing distance, and in the two-dimensional blurred kernel, determining whether the pixel point is within the trailing range, and obtaining the blurred kernel value at the position of the pixel point according to the gradient of the pixel point and the size of the two-dimensional blurred kernel; de-blurring the blurred image using the two-dimensional blurred kernel. A crack identification module for identifying cracks in the de-blurred deep foundation pit image.

Images