Intra-table combined rapid lossless grading detection method and device

Abstract

The invention provides an in-surface combined rapid lossless grading detection method and device. The method comprises the following steps: acquiring a surface photographic image and a ground penetrating radar profile image of a filling layer of a work area to be detected; based on a grading detection model, according to the surface layer photographic image and the ground penetrating radar profile image, grading of the filling layer is detected, and a grading detection result is obtained; wherein the grading detection model is obtained by training a convolutional neural network by taking a sample surface photographic image and a sample ground penetrating radar profile image obtained in a sampling area as training samples, taking grading data actually measured in the sampling area as a training label and taking a fusion feature as a training feature; the fusion feature is determined based on a residual feature extracted from the sample ground penetrating radar profile image and a convolution feature extracted from the sample surface photographic image. According to the method, the grading detection efficiency and accuracy of the large-scale heap dam are improved, the filling layer does not need to be damaged, and the construction safety is improved.

Description

Technical Field

[0001] The present invention relates to the field of deep learning technology, and in particular to a fast non-destructive gradation detection method and device for intra-table combination. Background Art

[0002] At present, the main body of the earth dam is built with stone filling and an impermeable body, which is a type of earth dam. During the construction period of the earth dam, the construction quality of the rockfill body is the key to the quality control of the dam, which is directly related to the safety of project construction and operation. An important indicator for measuring the quality of the rockfill body of the dam is the grading of the rockfill body.

[0003] Grading refers to the distribution relationship of various particle sizes in the dam-building rockfill. During the dam-building process, it is necessary to fill the dam body with rockfill materials that meet the grading requirements of regulations and specifications. Therefore, after a layer of dam material is filled at the construction site, a random inspection will be carried out on this layer of dam-building rockfill to check whether the grading curve of the inspected samples is within the reasonable range required by regulations and specifications.

[0004] In the related technologies, the pit test method is still the main technical method for on-site gradation detection. For large-scale high dams and reservoirs, due to the large amount of dam filling, high construction intensity and long construction period, high-frequency pit test gradation detection not only affects the construction progress; but also the random sampling through the pit test method is very sparse relative to the volume of the rockfill layer, and cannot well represent the overall gradation of the current filling layer; in addition, pit test sampling is destructive to the existing filling layer, causes great damage to the dam rockfill body, and affects the safety of dam construction. Summary of the invention

[0005] The present invention provides a table-in-table combined rapid non-destructive gradation detection method and device, which are used to solve the defects of the prior art that the pit measurement method is used to obtain the gradation data of a large-scale dam, the random sampling cannot characterize the overall gradation of the current filling layer, and the dam rockfill body is damaged during the measurement process, resulting in low accuracy, efficiency and safety in obtaining the gradation data of the large-scale dam, thereby improving the gradation detection efficiency and accuracy of the large-scale dam and improving the construction safety at the same time.

[0006] The present invention provides a rapid non-destructive gradation detection method combined with an internal table, comprising: Obtain surface photographic images and ground penetrating radar profile images of the filling layer in the work area to be inspected; Based on the gradation detection model, the gradation of the filling layer is detected according to the surface photography image and the ground penetrating radar profile image to obtain a gradation detection result; Among them, the gradation detection model is obtained by training the convolutional neural network with sample surface photographic images and sample ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and fusion features as training features; the fusion features are determined based on the residual features extracted from the sample ground penetrating radar profile images and the convolution features extracted from the sample surface photographic images.

[0007] According to the table-joint rapid non-destructive gradation detection method provided by the present invention, the gradation detection model is trained by the following steps: Preprocessing the sample surface photographic image to obtain a surface three-dimensional data volume; preprocessing the ground penetrating radar profile image to obtain a profile three-dimensional data volume; Performing convolution calculation and residual calculation on the cross-section three-dimensional data volume in sequence to obtain the residual feature; performing convolution calculation on the surface three-dimensional data volume to obtain the convolution feature; The residual feature and the convolution feature are fused to obtain the fused feature, and the loss value of the convolutional neural network is calculated according to the fused feature according to a preset loss function, and the weight parameters of the convolutional neural network are updated according to the loss value. When the maximum number of iterations is met or the convolutional neural network converges, the gradation detection model is obtained.

[0008] According to the surface-inside combined rapid non-destructive gradation detection method provided by the present invention, the preprocessing of the sample surface photographic image to obtain the surface three-dimensional data volume includes: Converting the sample surface photographic image into a binary image, and performing a number arrangement process on the binary image to obtain a surface three-dimensional data volume; The preprocessing of the ground penetrating radar profile image to obtain a three-dimensional profile data volume includes: The ground penetrating radar profile image is normalized, and the normalized data is arranged with the measurement line as the dimension to obtain a three-dimensional profile data volume.

[0009] According to the fast non-destructive gradation detection method for in-table combination provided by the present invention, the convolutional neural network includes a first feature extraction module, a second feature extraction module and a feature fusion module; The first feature extraction module includes at least one first convolution layer and at least one residual module; the first feature extraction module is used to perform convolution calculation on the three-dimensional data of the cross-section through the at least one first convolution layer to obtain a convolution calculation result, and perform residual calculation on the convolution calculation result through the at least one residual module to obtain the residual feature; The second feature extraction module includes at least one second convolution layer; the second feature extraction module is used to perform convolution calculation on the surface three-dimensional data volume through the at least one second convolution layer to obtain the convolution feature; A feature fusion module is used to fuse the residual feature and the convolution feature to obtain the fused feature.

[0010] According to the table-in-table combined rapid non-destructive gradation detection method provided by the present invention, after obtaining the gradation detection result, the method further includes: Based on the preset grading curve and the grading test result, the quality of the stone filling in the work area to be tested is evaluated to obtain an evaluation result.

[0011] The present invention also provides a rapid non-destructive gradation detection device combined with a table, comprising: An image acquisition module is used to acquire a surface photographic image and a ground penetrating radar profile image of the filling layer in the work area to be inspected; A detection module detects the gradation of the filling layer according to the surface photography image and the ground penetrating radar profile image based on a gradation detection model to obtain a gradation detection result; Among them, the gradation detection model is obtained by training the convolutional neural network with sample surface photographic images and sample ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and fusion features as training features; the fusion features are determined based on the residual features extracted from the sample ground penetrating radar profile images and the convolution features extracted from the sample surface photographic images.

[0012] According to the fast non-destructive gradation detection device for in-table combination provided by the present invention, the device also includes: The quality assessment module is used to assess the quality of the stone materials filled in the work area to be inspected based on a preset gradation curve and the gradation test result after obtaining the gradation test result, so as to obtain an assessment result.

[0013] The present invention also provides an electronic device, comprising a memory, a processor and a computer program stored in the memory and executable on the processor, wherein when the processor executes the program, the fast non-destructive grading detection method for in-table combination as described in any one of the above is implemented.

[0014] The present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, and when the computer program is executed by a processor, the fast non-destructive grading detection method for in-table combination as described in any one of the above is implemented.

[0015] The present invention also provides a computer program product, comprising a computer program, wherein when the computer program is executed by a processor, the computer program implements any of the above-mentioned methods for rapid non-destructive gradation detection by in-table combination.

[0016] The table-in-table combined rapid non-destructive gradation detection method and device provided by the present invention use sample surface photographic images and ground penetrating radar profile images obtained in the sampling area as training samples, use gradation data actually measured in the sampling area as training labels, and use fusion features as training features to train a gradation detection model to detect the gradation of the filling layer based on the surface photographic images and the ground penetrating radar profile images to obtain a gradation detection result, thereby improving the gradation detection efficiency and accuracy of large-scale dams, without damaging the filling layer, and improving construction safety. BRIEF DESCRIPTION OF THE DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are 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.

[0018] Figure 1 It is one of the flow diagrams of the rapid non-destructive gradation detection method combined within the table provided by the present invention.

[0019] Figure 2 This is the second flow chart of the rapid non-destructive gradation detection method combined within the table provided by the present invention.

[0020] Figure 3 It is a schematic diagram of the process of obtaining fusion features provided by the present invention.

[0021] Figure 4 It is a structural schematic diagram of the rapid non-destructive gradation detection device combined with the table provided by the present invention.

[0022] Figure 5 It is a structural schematic diagram of the electronic device provided by the present invention. DETAILED DESCRIPTION

[0023] In order to make the purpose, technical solution and advantages of the present invention clearer, the technical solution of the present invention will be clearly and completely described below in conjunction with the drawings of the present invention. Obviously, the described embodiments are part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0024] Combine the following Figure 1-Figure 4The invention describes a rapid non-destructive gradation detection method and device combined within a table.

[0025] Figure 1 This is one of the flow diagrams of the rapid non-destructive gradation detection method provided by the present invention, such as Figure 1 As shown in the table, the rapid non-destructive gradation detection method combined in the table includes the following steps: Step 110: Acquire a surface photographic image and a ground penetrating radar profile image of the filling layer of the work area to be inspected.

[0026] In this step, the work area to be inspected can be a large-scale pile dam or a small or medium-sized pile dam.

[0027] In this step, the surface photographic image is obtained by photographing the surface of the filling layer in the detection area; for example, the photographic tool may be an industrial camera or a mobile terminal with a photographing function.

[0028] In this step, the ground penetrating radar profile image is a radar reverse profile in the depth direction obtained by scanning the surface of the filling layer in the area using a ground penetrating radar that transmits and receives electromagnetic waves; for example, a ground penetrating radar can be used to scan the filling layer of the work area to be inspected to obtain a corresponding ground penetrating radar profile image.

[0029] In this embodiment, after acquiring the surface photography image and the ground penetrating radar profile image, the image can be preprocessed to improve the image quality. The preprocessing includes normalizing the image, removing outliers, etc.

[0030] Step 120, based on the gradation detection model, the gradation of the fill layer is detected according to the surface photographic image and the ground penetrating radar profile image to obtain the gradation detection result; wherein the gradation detection model is obtained by training the convolutional neural network with the sample surface photographic image and the sample ground penetrating radar profile image obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and the fusion feature as the training feature; the fusion feature is determined based on the residual feature extracted from the sample ground penetrating radar profile image and the convolution feature extracted from the sample surface photographic image.

[0031] In this step, the gradation detection model is obtained through laboratory stage training; wherein, the model to be trained can be a convolutional neural network; specifically, the convolutional neural network includes two cascaded network branches, one branch is used to input the surface photographic image, and perform convolution calculation and residual calculation on the surface photographic image to obtain the corresponding residual feature; the other branch is used to input the ground penetrating radar profile image, and perform convolution calculation on the ground penetrating radar profile image to obtain the corresponding convolution feature.

[0032] For example, the first branch may include multiple convolutional layers and multiple residual modules to implement convolution calculation and residual calculation of surface photographic images; the second branch may include multiple convolutional layers to implement convolution calculation of ground penetrating radar profile images.

[0033] In this embodiment, the model to be trained may also be a deep neural network, a residual neural network, etc.

[0034] In this embodiment, the gradation data actually measured in the sampling area can be obtained by a pit measurement method. For example, random sampling is first performed at the filling material site, and a screening machine is used to sieve and divide the particles into particle size ranges. The mass of the rockfill materials in each particle size range is weighed, and the gradation data of the rockfill materials is calculated and analyzed to obtain the corresponding gradation data as the training label corresponding to the current input image (including the surface photography image and the ground penetrating radar profile image).

[0035] In this embodiment, the trained gradation detection model can be used to perform non-destructive gradation detection during the construction of a large-scale dam.

[0036] In this embodiment, the filling layer of the work area to be inspected is scanned by a ground penetrating radar to obtain a ground penetrating radar profile image, and the surface of the filling layer is photographed by a camera to obtain a surface photographic image; the ground penetrating radar profile image and the surface photographic image are input into a trained grading detection model for detection to obtain a grading detection result, and the quality of the stone filling in the work area to be inspected is evaluated based on the grading detection result.

[0037] The fast non-destructive gradation detection method for in-table combination provided by the embodiment of the present invention uses the sample surface photographic images and ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and the gradation detection model trained with fusion features as training features to detect the gradation of the filling layer according to the surface photographic images and the ground penetrating radar profile images to obtain the gradation detection result, thereby improving the gradation detection efficiency and accuracy of large-scale dams, without damaging the filling layer, and improving the construction safety.

[0038] In some embodiments, the gradation detection model is trained by the following steps: preprocessing the sample surface photographic image to obtain a surface three-dimensional data volume; preprocessing the sample ground penetrating radar profile image to obtain a profile three-dimensional data volume; performing convolution calculation and residual calculation on the profile three-dimensional data volume in sequence to obtain residual features; performing convolution calculation on the surface three-dimensional data volume to obtain convolution features; fusing the residual features and the convolution features to obtain fused features, and calculating the loss value of the convolutional neural network according to the fused features according to a preset loss function, and updating the weight parameters of the convolutional neural network according to the loss value, and obtaining the gradation detection model when the maximum number of iterations is met or the convolutional neural network converges.

[0039] In this embodiment, the preprocessing of the sample surface photographic images and ground penetrating radar profile images may include: converting the two-dimensional images into three-dimensional data volumes, expanding the number of sample surface photographic images by data enhancement to avoid overfitting of the model training, and enhancing the image quality by normalizing the sample surface photographic images.

[0040] Specifically, preprocessing the sample surface photographic image to obtain a surface three-dimensional data volume includes: converting the sample surface photographic image into a binary image, and arranging the binary image by number to obtain a surface three-dimensional data volume; preprocessing the sample ground penetrating radar profile image to obtain a profile three-dimensional data volume includes: normalizing the ground penetrating radar profile image, and arranging the normalized data with the measurement line as the dimension to obtain a profile three-dimensional data volume.

[0041] In this embodiment, the loss function may include a loss function, for example, the loss function may be a cross entropy loss function, a norm loss function, or a mean square error loss function, etc.; it may also be a combination of multiple loss functions.

[0042] For example, the loss function A=x × X+y × Y ; in, X is the cross entropy loss function, x for X The weight factor of Y is the mean square error loss function, y for Y The weight factor of .

[0043] In this embodiment, the training set (including sample surface photographic images and sample ground penetrating radar profile images) is collected and produced by an on-site experimental site; specifically, an experimental site is selected near the on-site material yard, an experimental pit of a certain depth is dug, and several experimental areas are divided; the same rockfill material as that used for dam body filling is used, and different screening, filling, and rolling processes are carried out in different areas until the experimental pit is filled; the surface of the experimental pit is photographed to obtain an apparent image, a binary image is generated through image processing, and a ground penetrating radar is used to scan it to obtain a radar profile, and then the input data of the sample is obtained after data regularization; the actual gradation is determined by the pit measurement method as the label data of the sample at that location.

[0044] Figure 2 This is the second flow chart of the rapid non-destructive gradation detection method provided by the present invention. Figure 2In the illustrated embodiment, rockfill materials are collected in a material yard to make an experimental pit for making a training set, the actual gradation curve of the partition is recorded before the partition is filled, original input data is obtained through surface photography and radar scanning, and standardized input data is obtained through image preprocessing and data regularization. The convolutional neural network is trained based on the training data set collected and produced in the experimental pit to obtain a gradation detection model suitable for gradation detection of the dam. After a layer of filling is completed on site, surface photography and radar scanning are also used to collect detection data, which are input into the network after regularization processing to directly obtain the predicted gradation, that is, the gradation of the current filling layer.

[0045] The fast non-destructive gradation detection method with intra-table combination provided in the embodiment of the present invention obtains residual features by successively performing convolution calculation and residual calculation on the surface three-dimensional data volume corresponding to the sample surface photographic image; performing convolution calculation on the profile three-dimensional data volume corresponding to the sample ground penetrating radar profile image to obtain convolution features; then fusing the residual features and the convolution features, and training the convolution neural network according to the fused features according to a preset loss function to obtain a corresponding gradation detection model, thereby realizing the automated detection of the gradation of large-scale pile dams and improving the detection efficiency.

[0046] In some embodiments, the convolutional neural network includes a first feature extraction module, a second feature extraction module and a feature fusion module; the first feature extraction module includes at least one first convolutional layer and at least one residual module; the first feature extraction module is used to perform convolution calculation on the cross-sectional three-dimensional data through at least one first convolutional layer to obtain a convolution calculation result, and perform residual calculation on the convolution calculation result through at least one residual module to obtain a residual feature; the second feature extraction module includes at least one second convolutional layer; the second feature extraction module is used to perform convolution calculation on the surface three-dimensional data body through at least one second convolutional layer to obtain a convolution feature; the feature fusion module is used to perform feature fusion on the residual feature and the convolution feature to obtain a fusion feature.

[0047] In this embodiment, the feature fusion module can be composed of one or more fully connected layers; specifically, the residual features output by the first feature extraction module and the convolutional features output by the second feature extraction module are input into two fully connected layers connected in sequence, the first fully connected layer concatenates the residual features and the convolutional features to obtain fused features, and the second fully connected layer performs gradation detection based on the fused features to obtain corresponding gradation detection results.

[0048] In this embodiment, the first feature extraction module may also include a fully connected layer, which is used to convert the residual features output by the residual module into a data structure of a specified dimension; the second feature extraction module may also include a fully connected layer, which is used to convert the convolution features output by the convolution layer into the data structure of the above-mentioned specified dimension, so as to facilitate the splicing of the two features with the same number of channels.

[0049] Figure 3 is a schematic diagram of the process of obtaining fusion features provided by the present invention. Figure 3 In the embodiment shown, the first feature extraction module includes two first convolutional layers (corresponding to convolutional layers 1 and 2 above), 7 residual modules (corresponding to Resnet Modules one to seven) and one fully connected layer; the second feature extraction module includes five second convolutional layers (corresponding to convolutional layers one to five below) and one fully connected layer; the feature fusion module includes two fully connected layers connected in sequence; the input radar ground penetrating radar profile image is input into the first feature extraction module, and the corresponding residual features are obtained after convolution calculation and residual calculation, and the residual features are input into the fully connected layer of the first feature extraction module, and the feature map extracted by the convolution layer is converted into a one-dimensional data structure, and finally output as a one-dimensional array composed of 30 elements, indicating the proportion of rockfill materials under different particle sizes; the input surface ground penetrating radar profile image is input into the second feature extraction module, and the corresponding convolution features are obtained after convolution calculation, and the convolution features are input into the fully connected layer of the second feature extraction module, and the feature map extracted by the convolution layer is converted into a one-dimensional data structure; finally, the fusion features are obtained according to the residual features and the convolution features through two fully connected layers connected in sequence, and the corresponding grading detection results are predicted according to the fusion features.

[0050] The fast non-destructive gradation detection method for intra-table joint provided by the embodiment of the present invention comprises a first feature extraction module, a second feature extraction module and a feature fusion module by setting a convolutional neural network; the first feature extraction module comprises at least one first convolutional layer and at least one residual module; the second feature extraction module comprises at least one second convolutional layer; the feature fusion module is used to fuse the residual features and the convolution features to obtain the fused features, thereby enhancing the constraint of the overall gradation and improving the accuracy of the gradation detection results.

[0051] In some embodiments, after obtaining the grading test result, the method further includes: evaluating the quality of the stone material filled in the work area to be tested based on a preset grading curve according to the grading test result to obtain an evaluation result.

[0052] In this embodiment, the preset grading curve can be drawn through the real grading data collected by the above-mentioned pit measurement method, which is used to reflect the relative content of each particle group in the soil. The uniformity of the soil or whether the gradation is good can be judged according to the slope of the curve; for example, a steep curve indicates that the particle sizes are similar and the soil particles are relatively uniform; a gentle curve indicates that the particle sizes are very different, the soil particles are uneven, and the gradation is good.

[0053] In this embodiment, according to the gradation test results output by the gradation test model and compared with the preset gradation curve, the relative content of each particle group in the soil of the work area corresponding to the gradation test results can be queried, thereby determining the quality of the stone filling in the work area.

[0054] In a feasible embodiment, since the dam rockfill construction is filled layer by layer with a layer thickness of about 1m, the conventional gradation detection is to find several pits after filling a layer, and obtain the gradation data of the stone in the pit to see whether it meets the requirements; this embodiment is to make a training data set on site through experimental pits or existing data. After the model is trained, after each layer is filled during the construction period, surface photography is performed to obtain a surface image, and a ground penetrating radar scan is performed to obtain a profile image, which is input into the gradation detection model for detection, and the predicted gradation data is output. Then, the quality of the current filling layer is evaluated, and then the next layer is filled, and the process is repeated.

[0055] It should be noted that the dam gradation detection method based on image recognition and radar detection in this embodiment includes the preparation of an early training data set, the training of a mid-term network, and the prediction of the gradation of each filling layer in the later construction.

[0056] The fast non-destructive gradation detection method combined with the table provided in the embodiment of the present invention can evaluate the quality of the filling stones in the inspection area according to the gradation detection results through the preset gradation curve, thereby improving the efficiency of the filling stone quality assessment.

[0057] The following is a description of the rapid non-destructive gradation detection device for combined in-table provided by the present invention. The rapid non-destructive gradation detection device for combined in-table described below and the rapid non-destructive gradation detection method for combined in-table described above can be referenced to each other.

[0058] Figure 4 Schematic diagram of the structure of the rapid non-destructive gradation detection device provided by the present invention, such as Figure 4 As shown, the rapid non-destructive gradation detection device combined in the table includes: an image acquisition module 410 and a detection module 420.

[0059] An image acquisition module 410 acquires a surface photographic image and a ground penetrating radar profile image of the filling layer of the work area to be inspected; A detection module 420 detects the gradation of the filling layer according to the surface photography image and the ground penetrating radar profile image based on the gradation detection model to obtain a gradation detection result; Among them, the gradation detection model is obtained by training the convolutional neural network with the sample surface photographic images and sample ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and the fusion features as training features; the fusion features are determined based on the residual features extracted from the sample ground penetrating radar profile images and the convolution features extracted from the sample surface photographic images.

[0060] The fast non-destructive gradation detection device with table-in-table combination provided in the embodiment of the present invention uses the sample surface photographic images and ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and the gradation detection model trained with fusion features as training features to detect the gradation of the filling layer according to the surface photographic images and the ground penetrating radar profile images to obtain the gradation detection result, thereby improving the gradation detection efficiency and accuracy of large-scale dams, without damaging the filling layer, and improving the construction safety.

[0061] In some embodiments, the rapid non-destructive gradation detection device combined in the table further includes: The quality assessment module is used to assess the quality of the stone materials filled in the inspection area based on the preset gradation curve and the gradation test results after obtaining the gradation test results to obtain the assessment results.

[0062] The fast non-destructive gradation detection device combined with the table provided in the embodiment of the present invention can evaluate the quality of the stone filling in the inspection area according to the gradation detection result through the preset gradation curve, thereby improving the efficiency of the filling stone quality assessment.

[0063] Figure 5 An example of a physical structure diagram of an electronic device is shown in FIG. Figure 5As shown, the electronic device may include: a processor 510, a communication interface 520, a memory 530 and a communication bus 540, wherein the processor 510, the communication interface 520 and the memory 530 communicate with each other through the communication bus 540. The processor 510 may call the logic instructions in the memory 530 to execute the fast non-destructive gradation detection method of the table combination, the method comprising: obtaining the surface photography image and the ground penetrating radar profile image of the filling layer of the work area to be detected; detecting the gradation of the filling layer according to the surface photography image and the ground penetrating radar profile image based on the gradation detection model to obtain the gradation detection result; wherein the gradation detection model is obtained by training the convolutional neural network with the sample surface photography image and the sample ground penetrating radar profile image obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and the fusion feature as the training feature; the fusion feature is determined based on the residual feature extracted from the sample ground penetrating radar profile image and the convolution feature extracted from the sample surface photography image.

[0064] In addition, the logic instructions in the above-mentioned memory 530 can be implemented in the form of a software functional unit and can be stored in a computer-readable storage medium when it is sold or used as an independent product. Based on such an understanding, the technical solution of the present invention, in essence, or the part that contributes to the prior art or the part of the technical solution, can be embodied in the form of a software product, and the computer software product is stored in a storage medium, including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, etc. Various media that can store program codes.

[0065] On the other hand, the present invention also provides a computer program product, which includes a computer program, which can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer can execute the intra-table combined rapid non-destructive gradation detection method provided by the above methods, the method including: obtaining a surface photographic image and a ground penetrating radar profile image of the filling layer of the work area to be detected; detecting the gradation of the filling layer according to the surface photographic image and the ground penetrating radar profile image based on a gradation detection model to obtain a gradation detection result; wherein the gradation detection model is obtained by training a convolutional neural network using sample surface photographic images and sample ground penetrating radar profile images obtained in a sampling area as training samples, using gradation data actually measured in the sampling area as training labels, and using fusion features as training features; the fusion features are determined based on the residual features extracted from the sample ground penetrating radar profile images and the convolution features extracted from the sample surface photographic images.

[0066] On the other hand, the present invention also provides a non-transitory computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, is implemented to execute the fast non-destructive gradation detection method provided by the above-mentioned methods in a table-joint manner, the method comprising: obtaining a surface photographic image and a ground penetrating radar profile image of a filling layer in a work area to be detected; detecting the gradation of the filling layer according to the surface photographic image and the ground penetrating radar profile image based on a gradation detection model to obtain a gradation detection result; wherein the gradation detection model is obtained by training a convolutional neural network using sample surface photographic images and sample ground penetrating radar profile images obtained in a sampling area as training samples, using gradation data actually measured in the sampling area as training labels, and using fusion features as training features; the fusion features are determined based on residual features extracted from sample ground penetrating radar profile images and convolution features extracted from sample surface photographic images.

[0067] The device embodiments described above are merely illustrative, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules may be selected according to actual needs to achieve the purpose of the scheme of this embodiment. Ordinary technicians in this field can understand and implement it without paying creative labor.

[0068] Through the description of the above implementation methods, those skilled in the art can clearly understand that each implementation method can be implemented by means of software plus a necessary general hardware platform, and of course, can also be implemented by hardware. Based on this understanding, the above technical solution is essentially or the part that contributes to the prior art can be embodied in the form of a software product, and the computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, a disk, an optical disk, etc., including a number of instructions for a computer device (which can be a personal computer, a server, or a network device, etc.) to execute the methods described in each embodiment or some parts of the embodiments.

[0069] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A rapid non-destructive gradation detection method combined with a table, characterized in that: include: Obtain surface photographic images and ground penetrating radar profile images of the filling layer in the work area to be inspected; Based on the gradation detection model, the gradation of the filling layer is detected according to the surface photography image and the ground penetrating radar profile image to obtain a gradation detection result; Among them, the gradation detection model is obtained by training the convolutional neural network with sample surface photographic images and sample ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and fusion features as training features; the fusion features are determined based on the residual features extracted from the sample ground penetrating radar profile images and the convolution features extracted from the sample surface photographic images.

2. The rapid non-destructive gradation detection method combined with the table according to claim 1 is characterized in that: The gradation detection model is trained by the following steps: Preprocessing the sample surface photographic image to obtain a surface three-dimensional data volume; preprocessing the ground penetrating radar profile image to obtain a profile three-dimensional data volume; Performing convolution calculation and residual calculation on the three-dimensional data volume of the cross section in sequence to obtain the residual feature; Performing convolution calculation on the surface three-dimensional data volume to obtain the convolution feature; The residual feature and the convolution feature are fused to obtain the fused feature, and the loss value of the convolutional neural network is calculated according to the fused feature according to a preset loss function, and the weight parameters of the convolutional neural network are updated according to the loss value. When the maximum number of iterations is met or the convolutional neural network converges, the gradation detection model is obtained.

3. The rapid non-destructive gradation detection method combined with the table according to claim 2 is characterized in that: The preprocessing of the sample surface photographic image to obtain the surface three-dimensional data volume comprises: Converting the sample surface photographic image into a binary image, and performing a number arrangement process on the binary image to obtain a surface three-dimensional data volume; The preprocessing of the ground penetrating radar profile image to obtain a three-dimensional profile data volume includes: The ground penetrating radar profile image is normalized, and the normalized data is arranged with the measurement line as the dimension to obtain a three-dimensional profile data volume.

4. The rapid non-destructive gradation detection method of any one of claims 1-2, characterized in that: The convolutional neural network includes a first feature extraction module, a second feature extraction module and a feature fusion module; The first feature extraction module includes at least one first convolution layer and at least one residual module; the first feature extraction module is used to perform convolution calculation on the three-dimensional data of the cross-section through the at least one first convolution layer to obtain a convolution calculation result, and perform residual calculation on the convolution calculation result through the at least one residual module to obtain the residual feature; The second feature extraction module includes at least one second convolution layer; the second feature extraction module is used to perform convolution calculation on the surface three-dimensional data volume through the at least one second convolution layer to obtain the convolution feature; A feature fusion module is used to fuse the residual feature and the convolution feature to obtain the fused feature.

5. The rapid non-destructive gradation detection method combined with the table according to claim 1 is characterized in that: After obtaining the gradation detection result, the method further includes: Based on the preset grading curve and the grading test result, the quality of the stone filling in the work area to be tested is evaluated to obtain an evaluation result.

6. A rapid non-destructive gradation detection device combined with a table, characterized in that: include: An image acquisition module is used to acquire a surface photographic image and a ground penetrating radar profile image of the filling layer in the work area to be inspected; A detection module detects the gradation of the filling layer according to the surface photography image and the ground penetrating radar profile image based on a gradation detection model to obtain a gradation detection result; Among them, the gradation detection model is obtained by training the convolutional neural network with sample surface photographic images and sample ground penetrating radar profile images obtained in the sampling area as training samples, the gradation data actually measured in the sampling area as training labels, and fusion features as training features; the fusion features are determined based on the residual features extracted from the sample ground penetrating radar profile images and the convolution features extracted from the sample surface photographic images.

7. The rapid non-destructive gradation detection device combined with the table according to claim 6 is characterized in that: The device also includes: The quality assessment module is used to assess the quality of the stone materials filled in the work area to be inspected based on a preset gradation curve and the gradation test result after obtaining the gradation test result, so as to obtain an assessment result.

8. An electronic device 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 fast non-destructive gradation detection method for in-table combination as claimed in any one of claims 1 to 5 is implemented.

9. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the fast non-destructive gradation detection method for in-table combination as claimed in any one of claims 1 to 5 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the fast non-destructive gradation detection method for in-table combination as claimed in any one of claims 1 to 5 is implemented.

Images