A tunnel deformation detection method and system based on artificial intelligence

Abstract

The present invention relates to the field of artificial intelligence technology, and in particular to a tunnel deformation detection method and system based on artificial intelligence. The method obtains a displacement distribution map based on the similarity of pixel points in two adjacent frames of tunnel images. In combination with the characteristics of the image, a deep neural network is constrained by a reconstruction loss function, a perspective displacement constraint loss function, and a deformation displacement constraint loss function. The displacement distribution map is processed by a deep neural network, and a deformation displacement distribution map and a perspective displacement distribution map are output. The displacement information of the pixel points is analyzed according to the deformation displacement distribution map. The present invention uses a deep neural network to remove the displacement caused by the camera perspective in the image, and accurately detects the tunnel deformation information.

Description

Technical Field

[0001] The present invention relates to the field of artificial intelligence technology, and in particular to an artificial intelligence-based tunnel deformation detection method and system. Background Art

[0002] When tunnels are excavated, the stress distribution of the surrounding rock is changed, which will cause deformation of the tunnel, such as the settlement deformation of the arch crown, the deformation of the arch foot caused by the eccentric pressure, and the deformation of the side wall caused by the eccentric pressure. During tunnel construction, it is necessary to conduct timely and effective detection of tunnel deformation to avoid construction accidents.

[0003] In the prior art, sensors are usually installed at multiple locations on the tunnel to sense the displacement change at the current location. Using sensors to detect tunnel deformation requires installing sensors at multiple locations in the tunnel for detection, which is not only costly but also has a limited detection range. In order to increase the detection range, more sensors need to be installed, which increases the cost investment.

[0004] In order to reduce the cost, the image processing method is used to detect the characteristics of tunnel deformation by comparing the difference of tunnel images in different time periods. However, during the image analysis process, the image itself may shift in perspective due to the problem of camera parameters and setting position, resulting in errors and affecting the accuracy of the detection results. Summary of the invention

[0005] In order to solve the above technical problems, the purpose of the present invention is to provide a tunnel deformation detection method and system based on artificial intelligence. The technical solutions adopted are as follows:

[0006] The present invention proposes a tunnel deformation detection method based on artificial intelligence, the method comprising:

[0007] Acquire multiple frames of tunnel images continuously according to a preset sampling time; use a frame of the tunnel image corresponding to a target time as a target image; and use a frame of the tunnel image before the target image as a comparison image;

[0008] Obtaining color distribution features and texture distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image; obtaining a displacement probability of each pixel point in the comparison image at other positions according to the pixel value difference between each pixel point of the comparison image and all pixels in the target image, the similarity of the color distribution features, and the similarity of the texture distribution features, and obtaining a displacement distribution map according to the displacement probability;

[0009] The displacement distribution map of all pixels in the contrast image is input into a pre-trained deep neural network, and a plurality of deformation displacement distribution maps and a plurality of viewing angle displacement distribution maps are output; the loss function of the deep neural network includes a reconstruction loss function, a viewing angle displacement constraint loss function and a deformation displacement constraint loss function; the reconstruction loss function is obtained according to the deformation displacement distribution map and the difference between the viewing angle displacement distribution map and the displacement distribution map; all the viewing angle displacement distribution maps are superimposed and averaged to obtain an average viewing angle displacement distribution map, and a viewing angle displacement constraint loss function is obtained according to the difference between each viewing angle displacement distribution map and the average viewing angle displacement distribution map; the complexity of the color distribution features and the texture distribution features at other pixel points in the neighborhood range of the pixel point in the contrast image is used as the deformation displacement weight, and the deformation displacement constraint loss function is obtained according to the difference between the deformation displacement distribution map and the deformation displacement distribution map of other pixel points in the neighborhood range corresponding to the pixel point in the contrast image and the deformation displacement weight;

[0010] The displacement information generated at the target moment is obtained according to each of the deformation displacement distribution diagrams.

[0011] Furthermore, obtaining the color distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image includes:

[0012] The image within the neighborhood is converted into a grayscale image; the grayscale image is filtered to obtain a low-frequency filtered image; the first occurrence probability of each grayscale value is obtained according to the grayscale histogram of the low-frequency filtered image; and the vector formed by the first occurrence probability is used as the color distribution feature.

[0013] Furthermore, obtaining the texture distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image includes:

[0014] Subtract the grayscale image from the low-frequency filtered image to obtain a high-frequency filtered image; obtain the gradient direction of each pixel in the high-frequency filtered image to obtain a gradient histogram; obtain a second occurrence probability of each gradient direction based on the gradient histogram; and use the vector composed of the second occurrence probabilities as the texture distribution feature.

[0015] Furthermore, the step of obtaining the displacement probability of each pixel point in the comparison image at other positions according to the pixel value difference between each pixel point of the comparison image and all pixel points in the target image, the similarity of the color distribution features and the similarity of the texture distribution features includes:

[0016] The displacement probability is obtained according to the displacement probability formula; the displacement probability formula includes:

[0017]

[0018] Wherein, D(p,q) is the displacement probability from the p-th pixel in the comparison image to the q-th pixel in the target image, sim() is the cosine similarity function, and v1 p is the color distribution feature of the p-th pixel in the comparison image, v1 q is the color distribution feature of the qth pixel in the target image, v2 p is the texture distribution feature of the p-th pixel in the comparison image, v2 q is the texture distribution feature of the qth pixel in the target image, f2(q) is the pixel value information of the qth pixel in the target image, and f1(p) is the pixel value information of the pth pixel in the comparison image.

[0019] Further, obtaining the reconstruction loss function according to the deformation displacement distribution map and the difference between the viewing angle displacement distribution map and the displacement distribution map comprises:

[0020]

[0021] Wherein, Loss1 is the reconstruction loss function, N is the number of pixels in the displacement distribution map, is the pixel value of the hth pixel point on the viewing angle displacement distribution map corresponding to the i-th pixel point in the comparison image, is the pixel value of the hth pixel point on the deformation displacement distribution map corresponding to the i-th pixel point in the comparison image, I i (h) is the pixel value of the h-th pixel on the displacement distribution map corresponding to the i-th pixel in the comparison image.

[0022] Furthermore, obtaining the viewing angle displacement constraint loss function according to the difference between each viewing angle displacement distribution map and the average viewing angle displacement distribution map comprises:

[0023] The first entropy of the viewing angle displacement distribution map and the second entropy of the deformation displacement distribution map are obtained, and the viewing angle displacement constraint loss function is obtained by combining the difference between each viewing angle displacement distribution map and the average viewing angle displacement distribution map; the viewing angle displacement constraint loss function includes:

[0024]

[0025] Among them, Loss2 is the perspective displacement constraint loss function, is the viewing angle displacement distribution map corresponding to the ith pixel point in the comparison image, K is the number of pixels in the comparison image, is the viewing angle displacement distribution map corresponding to the k-th pixel point in the comparison image, is the entropy of the deformation displacement distribution map corresponding to the i-th pixel point in the comparison image, is the entropy of the viewing angle displacement distribution map corresponding to the i-th pixel point in the comparison image.

[0026] Furthermore, the taking the complexity of the color distribution features and the texture distribution features at other pixel points in the neighborhood range of the pixel point corresponding to the deformation displacement distribution map as the deformation displacement weight includes:

[0027] The entropy of the color distribution feature and the texture distribution feature is used as the complexity, and the deformation displacement weight is obtained according to the deformation displacement weight formula; the deformation displacement weight formula includes:

[0028] w i,m =exp(H(v2 m )+H(v1 m ))

[0029] Among them, w i,m is the deformation displacement weight of the mth pixel in the neighborhood of the ith pixel in the comparison image, H() is the entropy calculation formula, v2 m is the texture distribution feature of the mth pixel in the neighborhood of the ith pixel in the comparison image, v1 m is the color distribution feature of the mth pixel point in the neighborhood of the ith pixel point in the comparison image.

[0030] Furthermore, obtaining the deformation displacement constraint loss function according to the difference between the deformation displacement distribution map and the deformation displacement distribution maps of other pixels in the neighborhood range of the corresponding pixel point and the deformation displacement weight comprises:

[0031]

[0032] Among them, Loss3 is the deformation displacement constraint loss function, V is the number of pixels in the neighborhood, and w i,m is the deformation displacement weight of the mth other pixel point within the neighborhood of the ith pixel point in the comparison image, is the deformation displacement distribution map corresponding to the ith pixel point in the comparison image, It is the deformation displacement distribution map corresponding to the mth other pixel point within the neighborhood range of the ith pixel point in the comparison image.

[0033] Further, obtaining the displacement information generated at the target time according to each of the deformation displacement distribution diagrams includes:

[0034] The coordinate information of the pixel point with the largest pixel value on the deformation displacement distribution map is used as the displacement coordinate of the corresponding pixel point on the comparison image; the coordinates of the pixel point on the comparison image and the corresponding displacement coordinates constitute a displacement vector; and the displacement information is obtained according to the displacement vector.

[0035] The present invention also proposes an artificial intelligence-based tunnel deformation detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein when the processor executes the computer program, the system implements any one of the steps of the artificial intelligence-based tunnel deformation detection method.

[0036] The present invention has the following beneficial effects:

[0037] 1. The embodiment of the present invention constructs a displacement distribution map through the pixel value information, color distribution characteristics and texture distribution characteristics at each pixel point position on two tunnel images in different time periods. The displacement distribution map is split into a deformation displacement distribution map and a perspective displacement distribution map through a deep neural network. Only the information in the deformation displacement distribution map is analyzed, and accurate displacement information is obtained while saving detection costs.

[0038] 2. The deep neural network in the embodiment of the present invention accurately obtains the deformation displacement distribution map and the perspective displacement distribution map through the loss functions of the three parts. The reconstruction loss function can ensure the accuracy of the obtained deformation displacement distribution map and the perspective displacement distribution map relative to the displacement distribution map. Because the camera position is fixed, the displacement deviations of all pixel positions in the image are the same. The perspective displacement constraint loss function uses this feature to constrain the perspective displacement distribution map, thereby ensuring the accuracy of the perspective displacement distribution map. The deformation displacement constraint loss function uses the deformation displacement weight to make the network pay more attention to areas with complex textures, suppressing the errors generated in areas with uniform color and texture distribution. BRIEF DESCRIPTION OF THE DRAWINGS

[0039] In order to more clearly illustrate the technical solutions and advantages in the embodiments of the present invention or the prior art, the drawings required for use in the embodiments or the prior art descriptions are briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0040] Figure 1 A flow chart of a tunnel deformation detection method based on artificial intelligence provided by an embodiment of the present invention. DETAILED DESCRIPTION

[0041] In order to further explain the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following is a detailed description of the specific implementation, structure, features and effects of a tunnel deformation detection method and system based on artificial intelligence proposed by the present invention in combination with the accompanying drawings and preferred embodiments. In the following description, different "one embodiment" or "another embodiment" does not necessarily refer to the same embodiment. In addition, specific features, structures, or characteristics in one or more embodiments may be combined in any suitable form.

[0042] Unless defined otherwise, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention belongs.

[0043] The following is a detailed description of a tunnel deformation detection method and system based on artificial intelligence provided by the present invention in conjunction with the accompanying drawings.

[0044] See also Figure 1 , which shows a flow chart of a tunnel deformation detection method based on artificial intelligence provided by an embodiment of the present invention, the method comprising:

[0045] Step S1: acquiring multiple frames of tunnel images continuously according to a preset sampling time; taking a frame of tunnel image corresponding to a target moment as a target image; and taking a frame of tunnel image before the target image as a comparison image.

[0046] A camera is deployed in the area inside the tunnel where deformation needs to be detected, and multiple frames of tunnel images are acquired according to a preset sampling time. In an embodiment of the present invention, the image data acquired by the camera is RGB image data, and the sampling time is set to two hours, that is, one frame of tunnel image is acquired every two hours.

[0047] If the area to be detected is deformed and displaced, the deformation of the tunnel can be obtained through image difference, with a frame of tunnel image corresponding to the target moment for which displacement information is to be obtained as the target image, and the frame before the target image as the comparison image. During the deformation process, there will be pixels in the comparison image that move to a certain position in the target image.

[0048] In the embodiment of the present invention, after the tunnel image is obtained, a low-pass filter is performed on the tunnel image using a Gaussian kernel of size 3*3 to ensure image quality.

[0049] Step S2: obtaining color distribution features and texture distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image; obtaining the displacement probability of each pixel point in the comparison image at other positions according to the pixel value difference between each pixel point of the comparison image and all pixels in the target image, the similarity of the color distribution features and the similarity of the texture distribution features, and obtaining a displacement distribution map according to the displacement probability.

[0050] If the tunnel deforms within a certain period of time, the comparison image will show that a pixel point p moves to the position of a pixel point q in the target image. Therefore, the similarity between two pixels can be judged based on the feature information between the pixels. The more similar, the greater the probability that the pixel point p moves to the position of the pixel point q.

[0051] In order to avoid large probability errors, when analyzing the similarity between two pixels, it is necessary to consider not only the pixel value information, but also the information around the pixel, that is, to obtain the color distribution characteristics and texture distribution characteristics of each pixel in the preset neighborhood range on the comparison image and the target image, specifically including:

[0052] Convert the image within the neighborhood to a grayscale image. Filter the grayscale image to obtain a low-frequency filtered image. Obtain the first occurrence probability of each grayscale value based on the grayscale histogram of the low-frequency filtered image. The grayscale histogram can be regarded as a one-dimensional vector, that is, the first occurrence probability constitutes the one-dimensional vector, the position of each element in the vector corresponds to a grayscale value, and the size of the element is the first occurrence probability. Because only low-frequency color information is retained in the low-frequency filtered image, and the high-frequency texture information is removed, the vector composed of the first occurrence probability is used as the color distribution feature.

[0053] In the embodiment of the present invention, the size of the neighborhood range is 17*17. The filtering adopts Gaussian filtering method, and the size of the Gaussian kernel is 11*11.

[0054] Subtract the grayscale image from the low-frequency filtered image to obtain a high-frequency filtered image. Only complex texture information is retained in the high-frequency filtered image. The gradient direction of each pixel in the high-frequency filtered image is obtained, and a gradient histogram is obtained after statistics. The second occurrence probability of each gradient direction is obtained according to the gradient histogram. Similar to the color distribution feature, the vector composed of the second occurrence probability is used as the texture distribution feature.

[0055] The displacement probability of each pixel point in the comparison image at other positions is obtained according to the pixel value difference between each pixel point of the comparison image and all the pixels in the target image, the similarity of the color distribution characteristics and the similarity of the texture distribution characteristics, including:

[0056] The displacement probability is obtained according to the displacement probability formula; the displacement probability formula includes:

[0057]

[0058] Where D(p,q) is the displacement probability from the p-th pixel in the comparison image to the q-th pixel in the target image, sim() is the cosine similarity function, and v1 p To compare the color distribution characteristics of the p-th pixel in the image, v1 q is the color distribution feature of the qth pixel in the target image, v2 p To compare the texture distribution characteristics of the p-th pixel in the image, v2 q is the texture distribution feature of the qth pixel in the target image, f2(q) is the pixel value information of the qth pixel in the target image, and f1(p) is the pixel value information of the pth pixel in the comparison image.

[0059] Displacement probability formula, using exp(-‖f2(q)-f1(p)‖2) to represent the similarity of the pixel value information of two pixels, ‖‖2 is the symbol for the calculation of the two norm. The cosine similarity sim() is used to calculate the similarity of color distribution features and texture distribution features, sim(v1 p ,v1 q ) is larger, indicating that the color distribution around the two pixels is more similar. p ,v2 q ) is larger, the more similar the texture distribution around the two pixels is. The more similar the pixel value information is, the more similar the color distribution characteristics and the texture distribution characteristics are, the greater the corresponding displacement probability is.

[0060] It should be noted that, in the embodiment of the present invention, the tunnel image is RGB image data, so the pixel value information f2(q) and f1(p) are a three-dimensional vector, and each dimension of the vector corresponds to the R, G, and B channels of the image.

[0061] The displacement probability of each pixel in the comparison image to the position of all pixels in the target image is obtained to obtain a displacement distribution map. The pixel value of the pixel in the displacement distribution map represents the probability of the corresponding pixel in the comparison image moving to a certain position in the displacement distribution map.

[0062] Step S3: Input the displacement distribution map of all pixels in the comparison image into a pre-trained deep neural network, and output multiple deformation displacement distribution maps and multiple viewing angle displacement distribution maps.

[0063] Because the displacement effect in the image is produced in two aspects, one is the deformation caused by the tunnel itself reflected in the image, and the other is the deformation caused by the camera parameters and camera perspective. In the process of displacement analysis, the deformation caused by the perspective is the error information, and the displacement distribution map needs to be accurately split into the deformation displacement distribution map and the perspective displacement distribution map.

[0064] The displacement distribution map of all pixels in the contrast image is obtained, and the displacement distribution map is processed by a deep neural network to output multiple deformation displacement distribution maps and multiple perspective displacement distribution maps. The task of the deep neural network is to split the displacement distribution map into a deformation displacement distribution map and a perspective displacement distribution map. Therefore, the network is an unsupervised training network, and it is necessary to form a loss function through multiple constraints to fit the network. The loss function Loss of the deep neural network includes the reconstruction loss function Loss1, the perspective displacement constraint loss function loss2, and the deformation displacement constraint loss function Loss3, that is: Loss = Loss1 + loss2 + Loss3.

[0065] The purpose of the reconstruction loss function is to constrain the output deformation displacement distribution and perspective displacement distribution map as a whole to achieve the reconstruction effect. The displacement of the tunnel in the image is composed of its own deformation and the camera perspective, that is, the deformation displacement distribution map and the perspective displacement distribution map together constitute the displacement distribution map. Therefore, the reconstruction loss function is obtained according to the difference between the deformation displacement distribution map and the perspective displacement distribution map and the displacement distribution map. The specific form of the reconstruction loss function Loss1 is:

[0066]

[0067] Wherein, Loss1 is the reconstruction loss function, N is the number of pixels in the displacement distribution map, is the pixel value of the hth pixel point on the viewing angle displacement distribution map corresponding to the i-th pixel point in the comparison image, is the pixel value of the hth pixel point on the deformation displacement distribution map corresponding to the i-th pixel point in the comparison image, I i (h) is the pixel value of the h-th pixel on the displacement distribution map corresponding to the i-th pixel in the comparison image.

[0068] The reconstruction loss function is constrained by the pixel value of each pixel in the image, and the expectation is As much as possible with I i (h) are equal, that is, the loss value of the reconstruction loss function during network training needs to be as small as possible.

[0069] The perspective displacement is caused by the camera, so the error of the camera for each pixel in the image is the same. Based on this feature, all the perspective displacement distribution maps are superimposed and averaged to obtain the average perspective displacement distribution map. The perspective displacement constraint loss function is obtained according to the difference between each perspective displacement distribution map and the average perspective displacement distribution map. The specific form of the perspective displacement constraint loss function Loss2 is:

[0070]

[0071] Among them, Loss2 is the perspective displacement constraint loss function, is the distribution diagram of the viewing angle displacement corresponding to the i-th pixel in the comparison image, K is the number of pixels in the comparison image, is the distribution diagram of the viewing angle displacement corresponding to the kth pixel in the comparison image, is the entropy of the deformation displacement distribution map corresponding to the i-th pixel in the comparison image, is the entropy of the viewing angle displacement distribution map corresponding to the i-th pixel in the comparison image.

[0072] The perspective displacement constraint loss function expects that the difference between the perspective displacement distribution map corresponding to the i-th pixel and the average perspective distribution map is as small as possible, that is, each perspective displacement distribution map is as consistent as possible. In order to ensure that each pixel has a clear displacement caused by the deformation of the tunnel itself and the displacement caused by the camera perspective, the entropy of the deformation displacement distribution map and the perspective displacement distribution map is introduced. By making the sum of the entropies as small as possible, the characteristics of the deformation displacement distribution map and the perspective displacement distribution map are clarified.

[0073] There are some areas in the tunnel image where the pixel values ​​are evenly distributed and the color and texture are also evenly distributed. When the pixel points on the contrast image belong to such areas, in order to avoid the large displacement probability of the pixel points in this area in the deformation displacement distribution, resulting in errors, the deformation displacement distribution map needs to pay more attention to the positions with complex colors and textures in the surrounding areas. Therefore, the color distribution characteristics and the complexity of the texture distribution characteristics at other pixel points in the neighborhood range of the pixel points in the contrast image are used as the deformation displacement weights, which specifically include:

[0074] The entropy of color distribution features and texture distribution features is used as the complexity, and the deformation displacement weight is obtained according to the deformation displacement weight formula. The deformation displacement weight formula includes:

[0075] w i,m =exp(H(v2 m )+H(v1 m ))

[0076] Among them, w i,m is the deformation displacement weight of the mth pixel in the neighborhood of the i-th pixel in the comparison image, H() is the entropy calculation formula, v2m To compare the texture distribution characteristics of the mth pixel in the neighborhood of the i-th pixel in the image, v1 m To compare the color distribution characteristics of the mth pixel in the neighborhood of the i-th pixel in the image.

[0077] The larger the deformation displacement weight is, the greater the texture complexity in the corresponding surrounding neighborhood is, and the more attention should be paid to the information here, and the information here has greater reference significance; on the contrary, the possibility of misidentification is greater for areas with simple color texture, so the reference significance is smaller.

[0078] The deformation displacement constraint loss function is obtained based on the difference between the deformation displacement distribution map and the deformation displacement distribution maps of other pixels in the neighborhood of the pixel point in the corresponding comparison image and the deformation displacement weight. The specific form of the deformation displacement constraint loss function is:

[0079]

[0080] Among them, Loss3 is the deformation displacement constraint loss function, V is the number of pixels in the neighborhood, and w i,m To compare the deformation displacement weights of the mth pixel in the neighborhood of the i-th pixel in the image, is the deformation displacement distribution map corresponding to the i-th pixel in the comparison image, It is the deformation displacement distribution map corresponding to the mth pixel point in the neighborhood of the i-th pixel point in the comparison image.

[0081] Because the actual deformation process of the tunnel is the movement of one area to another, the deformation displacement distribution of adjacent pixels needs to be ensured during the pixel-by-pixel analysis. Figure 1 Therefore, the deformation displacement constraint loss function is expected to be As small as possible, and use the deformation displacement weight as the weight, paying more attention to the complex texture position In the network fitting process, the larger the deformation displacement weight, the corresponding Needs to be smaller to satisfy the constraints.

[0082] The network output is constrained by the loss function of the deep neural network, and the network training is completed after training and verification of the training data. In the embodiment of the present invention, the deep neural network uses an encoding-decoding structure, and the specific training method is a conventional technical means, which will not be repeated here.

[0083] Step S4: Obtain the displacement information generated at the target time according to each deformation displacement distribution diagram.

[0084] Only the deformation displacement distribution map output by the network is retained to analyze the deformation. Each deformation displacement distribution map corresponds to the displacement probability distribution of a pixel in the comparison image, and the coordinate information of the pixel with the largest pixel value on the deformation displacement distribution map is used as the displacement coordinate of the corresponding pixel on the comparison image. The coordinates of the pixel on the comparison image and the corresponding displacement coordinates constitute a displacement vector. The displacement information is obtained according to the displacement vector. The displacement vector can analyze a variety of displacement information such as the displacement distance and displacement direction of the pixel, which can be obtained according to the specific task needs and is not restricted here.

[0085] By obtaining the displacement information of each pixel, it can be used to assist in safety warning or construction and maintenance of subsequent tunnel construction projects.

[0086] In summary, the embodiment of the present invention obtains a displacement distribution map based on the similarity of pixel points in two adjacent frames of tunnel images. The deep neural network is constrained by the reconstruction loss function, the perspective displacement constraint loss function and the deformation displacement constraint loss function in combination with the characteristics of the image. The displacement distribution map is processed by the deep neural network, and a deformation displacement distribution map and a perspective displacement distribution map are output. The displacement information of the pixel points is analyzed according to the deformation displacement distribution map. The embodiment of the present invention uses a deep neural network to remove the displacement caused by the camera perspective in the image, and accurately detects the deformation information of the tunnel.

[0087] The present invention also proposes an artificial intelligence-based tunnel deformation detection system, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor executes the computer program, any step of an artificial intelligence-based tunnel deformation detection method is implemented.

[0088] It should be noted that the sequence of the embodiments of the present invention described above is for description only and does not represent the advantages and disadvantages of the embodiments. The above describes specific embodiments of this specification. Other embodiments are within the scope of the appended claims. In some cases, the actions or steps recorded in the claims can be performed in an order different from that in the embodiments and still achieve the desired results. In addition, the processes depicted in the drawings do not necessarily require the specific order or continuous order shown to achieve the desired results. In some embodiments, multitasking and parallel processing are also possible or may be advantageous.

[0089] The various embodiments in this specification are described in a progressive manner, and the same or similar parts between the various embodiments can be referenced to each other, and each embodiment focuses on the differences from other embodiments.

[0090] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A tunnel deformation detection method based on artificial intelligence, characterized in that: The method comprises: Acquire multiple frames of tunnel images continuously according to a preset sampling time; use a frame of the tunnel image corresponding to a target time as a target image; and use a frame of the tunnel image before the target image as a comparison image; Obtaining color distribution features and texture distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image; obtaining a displacement probability of each pixel point in the comparison image at other positions according to the pixel value difference between each pixel point of the comparison image and all pixels in the target image, the similarity of the color distribution features, and the similarity of the texture distribution features, and obtaining a displacement distribution map according to the displacement probability; The displacement distribution map of all pixels in the contrast image is input into a pre-trained deep neural network, and a plurality of deformation displacement distribution maps and a plurality of viewing angle displacement distribution maps are output; the loss function of the deep neural network includes a reconstruction loss function, a viewing angle displacement constraint loss function and a deformation displacement constraint loss function; the reconstruction loss function is obtained according to the deformation displacement distribution map and the difference between the viewing angle displacement distribution map and the displacement distribution map; all the viewing angle displacement distribution maps are superimposed and averaged to obtain an average viewing angle displacement distribution map, and a viewing angle displacement constraint loss function is obtained according to the difference between each viewing angle displacement distribution map and the average viewing angle displacement distribution map; the complexity of the color distribution features and the texture distribution features at other pixel points in the neighborhood range of the pixel point in the contrast image is used as the deformation displacement weight, and the deformation displacement constraint loss function is obtained according to the difference between the deformation displacement distribution map and the deformation displacement distribution map of other pixel points in the neighborhood range corresponding to the pixel point in the contrast image and the deformation displacement weight; Obtaining displacement information generated at the target moment according to each of the deformation displacement distribution diagrams; The obtaining of the reconstruction loss function according to the deformation displacement distribution map and the difference between the viewing angle displacement distribution map and the displacement distribution map comprises: Wherein, Loss1 is the reconstruction loss function, N is the number of pixels in the displacement distribution map, is the pixel value of the hth pixel point on the viewing angle displacement distribution map corresponding to the i-th pixel point in the comparison image, is the pixel value of the hth pixel point on the deformation displacement distribution map corresponding to the i-th pixel point in the comparison image, I i (h) is the pixel value of the h-th pixel on the displacement distribution map corresponding to the i-th pixel in the comparison image.

2. The tunnel deformation detection method based on artificial intelligence according to claim 1 is characterized in that: The step of obtaining the color distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image comprises: The image within the neighborhood is converted into a grayscale image; the grayscale image is filtered to obtain a low-frequency filtered image; the first occurrence probability of each grayscale value is obtained according to the grayscale histogram of the low-frequency filtered image; and the vector formed by the first occurrence probability is used as the color distribution feature.

3. The tunnel deformation detection method based on artificial intelligence according to claim 2 is characterized in that: The step of obtaining the texture distribution features within a preset neighborhood range of each pixel point on the comparison image and the target image comprises: Subtract the grayscale image from the low-frequency filtered image to obtain a high-frequency filtered image; obtain the gradient direction of each pixel in the high-frequency filtered image to obtain a gradient histogram; obtain a second occurrence probability of each gradient direction based on the gradient histogram; and use the vector composed of the second occurrence probabilities as the texture distribution feature.

4. The tunnel deformation detection method based on artificial intelligence according to claim 1 is characterized in that: The step of obtaining the displacement probability of each pixel point in the comparison image at other positions according to the pixel value difference between each pixel point in the comparison image and all pixel points in the target image, the similarity of the color distribution features and the similarity of the texture distribution features comprises: The displacement probability is obtained according to the displacement probability formula; the displacement probability formula includes: Wherein, D(p,q) is the displacement probability from the p-th pixel in the comparison image to the q-th pixel in the target image, sim() is the cosine similarity function, and v1 p is the color distribution feature of the p-th pixel in the comparison image, v1 q is the color distribution feature of the qth pixel in the target image, v2 p is the texture distribution feature of the p-th pixel in the comparison image, v2 q is the texture distribution feature of the qth pixel in the target image, f2(q) is the pixel value information of the qth pixel in the target image, and f1(p) is the pixel value information of the pth pixel in the comparison image.

5. The tunnel deformation detection method based on artificial intelligence according to claim 1 is characterized in that: The obtaining of the viewing angle displacement constraint loss function according to the difference between each viewing angle displacement distribution map and the average viewing angle displacement distribution map comprises: The first entropy of the viewing angle displacement distribution map and the second entropy of the deformation displacement distribution map are obtained, and the viewing angle displacement constraint loss function is obtained by combining the difference between each viewing angle displacement distribution map and the average viewing angle displacement distribution map; the viewing angle displacement constraint loss function includes: Among them, Loss2 is the perspective displacement constraint loss function, is the viewing angle displacement distribution map corresponding to the ith pixel point in the comparison image, K is the number of pixels in the comparison image, is the viewing angle displacement distribution map corresponding to the k-th pixel point in the comparison image, is the entropy of the deformation displacement distribution map corresponding to the i-th pixel point in the comparison image, is the entropy of the viewing angle displacement distribution map corresponding to the i-th pixel point in the comparison image.

6. The tunnel deformation detection method based on artificial intelligence according to claim 1 is characterized in that: The method of using the complexity of the color distribution features and the texture distribution features at other pixel points within the neighborhood range of the pixel point corresponding to the deformation displacement distribution map as the deformation displacement weight includes: The entropy of the color distribution feature and the texture distribution feature is used as the complexity, and the deformation displacement weight is obtained according to the deformation displacement weight formula; the deformation displacement weight formula includes: w i,m =exp(H(v2 m )+H(v1 m )) Among them, w i,m is the deformation displacement weight of the mth pixel in the neighborhood of the ith pixel in the comparison image, H() is the entropy calculation formula, v2 m is the texture distribution feature of the mth pixel in the neighborhood of the ith pixel in the comparison image, v1 m is the color distribution feature of the mth pixel point in the neighborhood of the ith pixel point in the comparison image.

7. The tunnel deformation detection method based on artificial intelligence according to claim 6 is characterized in that: The step of obtaining the deformation displacement constraint loss function according to the difference between the deformation displacement distribution map and the deformation displacement distribution maps of other pixels in the neighborhood range of the corresponding pixel point and the deformation displacement weight comprises: Among them, Loss3 is the deformation displacement constraint loss function, V is the number of pixels in the neighborhood, and w i,m is the deformation displacement weight of the mth other pixel point within the neighborhood of the ith pixel point in the comparison image, is the deformation displacement distribution map corresponding to the ith pixel point in the comparison image, It is the deformation displacement distribution map corresponding to the mth other pixel point within the neighborhood range of the ith pixel point in the comparison image.

8. The tunnel deformation detection method based on artificial intelligence according to claim 1 is characterized in that: The obtaining of the displacement information generated at the target time according to each of the deformation displacement distribution diagrams comprises: The coordinate information of the pixel point with the largest pixel value on the deformation displacement distribution map is used as the displacement coordinate of the corresponding pixel point on the comparison image; the coordinates of the pixel point on the comparison image and the corresponding displacement coordinates constitute a displacement vector; and the displacement information is obtained according to the displacement vector.

9. A tunnel deformation detection system based on artificial intelligence, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 8 are implemented.

Images