Pipeline inspection system and method

Abstract

The invention provides a pipeline inspection system and method, the system comprises a control end, an inspection end and a rear end, the rear end generates a preset inspection instruction and sends the preset inspection instruction to the control end, and the preset inspection instruction at least comprises pipeline information of multiple sections of preset pipelines in a to-be-inspected area; the control end controls the inspection end to move in a to-be-inspected area based on a received preset inspection instruction, so that the inspection end collects a radar time sequence image corresponding to each section of preset pipeline; the rear end obtains a radar time sequence image corresponding to each segment of preset pipeline, and inputs a pre-trained pipeline fault recognition model to obtain a recognition result output by the pipeline fault recognition model; and the rear end determines the preset pipeline needing to be maintained based on the identification result and forms a corresponding maintenance task, so that the efficiency of pipeline troubleshooting is improved, and the inspection cost is reduced.

Description

Technical Field

[0001] The present application relates to the field of urban construction technology, and in particular to a pipeline inspection system and method. Background Art

[0002] Underground pipeline management is one of the important tasks to ensure the normal operation of urban infrastructure. However, due to various reasons, underground pipelines may fail, affecting the normal operation of the city, especially the underground pipelines in old urban areas, which have been in use for a long time and have serious aging materials, making them prone to pipeline failures.

[0003] Existing methods for troubleshooting pipeline faults are mostly manual regular inspections and maintenance. However, this method not only wastes manpower, but also has poor inspection results because it is impossible to dig on the ground to detect faults. Summary of the invention

[0004] The purpose of the embodiments of the present application is to provide a pipeline inspection system and method to improve the efficiency of pipeline fault detection and reduce inspection costs.

[0005] In a first aspect, the present invention provides a pipeline inspection system, the system includes a control end, an inspection end and a back end, wherein the back end generates a preset inspection instruction and sends it to the control end, the preset inspection instruction at least including pipeline information of multiple sections of preset pipelines in the area to be inspected; the control end controls the inspection end to move in the area to be inspected based on the received preset inspection instruction, so that the inspection end collects radar time-series images corresponding to each section of the preset pipeline; the back end obtains the radar time-series images corresponding to each section of the preset pipeline, and inputs a pre-trained pipeline fault recognition model to obtain the recognition result output by the pipeline fault recognition model; the back end determines the preset pipeline that needs to be repaired based on the recognition result and forms a corresponding maintenance task; wherein, for each preset pipeline, The back end obtains the recognition result output by the pipeline fault recognition model through the following steps: for all radar time series images corresponding to the preset pipeline, the first radar time series image and the second radar time series image are selected; the first radar time series image and the second radar time series image are subjected to wavelet transform to obtain the first reconstructed time series image and the second reconstructed time series image respectively; based on the difference between the first reconstructed time series image and the second reconstructed time series image, a difference image is determined; the spatial features of the difference image, the first reconstructed time series image and the second reconstructed time series image are extracted by the feature extractor of the pipeline fault recognition model; the spatial features are recognized by the classifier of the pipeline fault recognition model to output the corresponding pipeline fault recognition result.

[0006] In an optional implementation, the pipeline information includes at least a category identifier of each preset pipeline segment, and the category identifier is used to distinguish the geometric shapes between pipelines. The backend is also used to determine a corresponding pipeline fault identification model based on the category identifier of the preset pipeline.

[0007] In an optional embodiment, for each radar time series image, the back end converts the radar time series image into a reconstructed time series image in the following manner: performing grayscale conversion on the radar time series image; performing a two-dimensional discrete wavelet transform on the grayscale converted radar time series image to obtain a corresponding approximate coefficient matrix, a horizontal detail coefficient matrix, a vertical detail coefficient matrix, and a diagonal detail coefficient matrix; and reconstructing based on the obtained approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix, and the diagonal detail coefficient matrix to obtain a reconstructed time series image.

[0008] In an optional implementation, the back end constructs a wavelet coefficient structure based on the obtained approximate coefficient matrix, horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix; the back end performs an inverse wavelet transform on the wavelet coefficient structure to obtain a corresponding reconstructed image.

[0009] In an optional implementation, before the back end constructs the wavelet coefficient structure, it also includes setting the values ​​in the approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix that are less than the preset value to zero.

[0010] In an optional implementation, the back end obtains the approximation coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix, and the diagonal detail coefficient matrix in the following manner:

[0011]

[0012] Among them, A k , m is the value in the kth column and mth row of the approximate coefficient matrix, is the value on the kth column and mth row of the horizontal detail coefficient matrix, is the value on the kth column and mth row of the vertical detail coefficient matrix, is the value on the kth column and mth row of the diagonal detail coefficient matrix, are high-pass filter and low-pass filter respectively, P l , n is the gray value on the kth column and mth row of the radar time series image.

[0013] In an optional embodiment, the first radar time series image and the second radar time series image are two interval or adjacent radar time series images, and the recognition result output by the pipeline fault recognition model is used to indicate the pipeline defect type and defect level, and the back end forms a corresponding maintenance task in the following manner: determining whether a preset pipeline needs maintenance based on the pipeline defect type and defect level indicated by the recognition result; if so, forming a maintenance task based on the radar time series image of the preset pipeline and the corresponding recognition result.

[0014] In an optional implementation, the backend generates a preset inspection instruction in the following manner:

[0015] In response to a user clicking on a target area control through an inspection interface, an area to be inspected is determined; in response to a user clicking on a target pipeline category control through an inspection interface, a category identifier of a preset pipeline is determined; based on the area to be inspected and the category identifier of the preset pipeline, corresponding pipeline information of multiple sections of preset pipelines is searched from a pipeline network database, the pipeline information also including location data of each section of the preset pipeline; based on the category identifier of the preset pipeline and the corresponding pipeline information, a preset inspection instruction is generated.

[0016] In a second aspect, the present invention provides a pipeline inspection method, the method comprising:

[0017] The radar time series images corresponding to each preset pipeline are obtained, and a pre-trained pipeline fault recognition model is input to obtain the recognition result output by the pipeline fault recognition model; based on the recognition result, the preset pipeline that needs to be repaired is determined and a corresponding maintenance task is formed; wherein, for each preset pipeline, the recognition result output by the pipeline fault recognition model is obtained through the following steps: for all radar time series images corresponding to the preset pipeline, a first radar time series image and a second radar time series image are selected; the first radar time series image and the second radar time series image are subjected to wavelet transformation to obtain a first reconstructed time series image and a second reconstructed time series image respectively; based on the difference between the first reconstructed time series image and the second reconstructed time series image, a difference image is determined; the spatial features of the difference image, the first reconstructed time series image and the second reconstructed time series image are extracted by a feature extractor of the pipeline fault recognition model; the spatial features are recognized by a classifier of the pipeline fault recognition model to output the corresponding pipeline fault recognition result.

[0018] In an optional implementation, for each radar time series image, the back end converts the radar time series image into a reconstructed time series image by:

[0019] The radar time series image is gray-converted; the gray-converted radar time series image is subjected to a two-dimensional discrete wavelet transform to obtain a corresponding approximate coefficient matrix, a horizontal detail coefficient matrix, a vertical detail coefficient matrix and a diagonal detail coefficient matrix; and reconstruction is performed based on the obtained approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix to obtain a reconstructed time series image.

[0020] The present application provides a pipeline inspection system and method, wherein the back end generates a preset inspection instruction and sends it to the control end; the control end controls the inspection end to move in the inspection area based on the received preset inspection instruction, so that the inspection end collects radar time-series images corresponding to each preset pipeline section; the back end obtains the radar time-series images corresponding to each preset pipeline section, and inputs a pre-trained pipeline fault recognition model to obtain the recognition result output by the pipeline fault recognition model; the back end determines the preset pipeline that needs to be repaired based on the recognition result and forms a corresponding maintenance task; wherein, for each preset pipeline section, the back end obtains the pipeline fault recognition model output through the following steps The identification result is as follows: for all radar time series images corresponding to the preset pipeline, the first radar time series image and the second radar time series image are selected; the first radar time series image and the second radar time series image are subjected to wavelet transform to obtain the first reconstructed time series image and the second reconstructed time series image respectively; based on the difference between the first reconstructed time series image and the second reconstructed time series image, the difference image is determined; the spatial features of the difference image, the first reconstructed time series image, and the second reconstructed time series image are extracted by the feature extractor of the pipeline fault recognition model; the spatial features are recognized by the classifier of the pipeline fault recognition model to output the corresponding pipeline fault recognition result. The pipeline inspection system provided in the present application adopts radar scanning to realize non-contact pipeline inspection, and combines the AI ​​model to realize intelligent identification of pipeline faults and defect levels. The inspection process has a high degree of automation, which avoids ground excavation and manual inspection and investigation, which not only saves manpower and material resources, but also improves inspection efficiency and reduces inspection costs. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the drawings required for use in the embodiments of the present application will be briefly introduced below. It should be understood that the following drawings only show certain embodiments of the present application and therefore should not be regarded as limiting the scope. For ordinary technicians in this field, other related drawings can be obtained based on these drawings without paying creative work.

[0022] Figure 1 A schematic diagram of the structure of a pipeline inspection system provided in an embodiment of the present application.

[0023] Figure Number:

[0024] Back-end - 10, control end - 20, inspection end - 30. DETAILED DESCRIPTION

[0025] The technical solutions in the embodiments of the present application will be described below in conjunction with the drawings in the embodiments of the present application.

[0026] Figure 1A schematic diagram of the structure of a pipeline inspection system provided in an embodiment of the present application. A pipeline inspection system provided in an embodiment of the present application includes a control end, an inspection end and a back end.

[0027] Among them, the inspection end and the control end can be arranged in the area to be inspected. The control end can be a mobile electronic device such as a mobile phone or a remote control device of the inspection personnel or a non-mobile electronic device such as a computer. The inspection end here includes a collection module, a positioning module and a processing module. Among them, the collection module can be a drone or a robot equipped with a low-quality radar. The positioning module can be a GPS, etc. The processing module is used to respond to the instructions of the control end to realize the movement and data collection control of the inspection end.

[0028] The backend here can be arranged in a fixed area. It can be a server or a computer, etc. The backend is provided with a database for storing the official website information of each urban area. The backend is also configured with pipeline fault identification models for different pipeline categories. The pipeline fault identification model here can be generated based on ConvLSTM training. In a specific embodiment, the pipelines here are classified according to geometric shapes, which can specifically include tubular, hollow and layered.

[0029] The present application also provides an application corresponding to the pipeline inspection system and a visual interactive interface.

[0030] Users can select the area to be inspected and the pipeline category through the inspection interface provided by the control end or the back end to complete the dispatch of inspection tasks.

[0031] Specifically, the backend or control end can generate the preset inspection instructions in the following ways:

[0032] In response to the user clicking the target area control through the inspection interface, the area to be inspected is determined. In response to the user clicking the target pipeline category control through the inspection interface, the category identification of the preset pipeline is determined. Based on the area to be inspected and the category identification of the preset pipeline, the corresponding pipeline information of multiple sections of the preset pipeline is searched from the pipeline network database, and the pipeline information also includes the location data of each section of the preset pipeline. Based on the category identification of the preset pipeline and the corresponding pipeline information, a preset inspection instruction is generated.

[0033] The inspection interface can provide a map display area. Each city can be divided into urban areas and displayed in the form of area controls. Users can zoom in and out of the map and select the target area control. The inspection interface can also display pipeline category controls corresponding to pipeline categories. Users can select one or more pipeline categories as needed.

[0034] The backend can directly search the pipeline information of each preset pipeline in the area to be inspected in the pipeline network database through the determined area to be inspected and pipeline category identification. The pipeline information here can include the location data, material, diameter, category, etc. of the pipeline. The location data can specifically include the location coordinates of the pipeline centerline pile and weld, as well as the elevation data of the pipeline location and the buried depth of the pipeline in the ground, etc.

[0035] The back end generates a preset inspection instruction and sends it to the control end, and the preset inspection instruction at least includes pipeline information of multiple preset pipelines in the area to be inspected.

[0036] Based on the received preset inspection instructions, the control end controls the inspection end to move in the area to be inspected, so that the inspection end collects radar time-series images corresponding to each preset pipeline section.

[0037] For each preset pipeline section, the inspection end can move along the center line of the pipeline and collect corresponding radar time-series images at a specified frequency.

[0038] The radar here operates in B-can scanning mode.

[0039] The backend obtains the radar time series images corresponding to each preset pipeline and inputs the pre-trained pipeline fault recognition model to obtain the recognition results output by the pipeline fault recognition model. Specifically, the backend can determine the corresponding pipeline fault recognition model based on the category identification of the preset pipeline to obtain the recognition results. The recognition results output by the pipeline fault recognition model are used to indicate the pipeline defect type and defect level.

[0040] For each preset pipeline, the backend can obtain the identification result output by the pipeline fault identification model through the following steps:

[0041] For all radar time series images corresponding to the preset pipeline, the first radar time series image and the second radar time series image are selected. The first radar time series image and the second radar time series image are subjected to wavelet transform to obtain the first reconstructed time series image and the second reconstructed time series image respectively. Based on the difference between the first reconstructed time series image and the second reconstructed time series image, a difference image is determined. The spatial features of the difference image, the first reconstructed time series image, and the second reconstructed time series image are extracted by the feature extractor of the pipeline fault recognition model. The spatial features are recognized by the classifier of the pipeline fault recognition model to output the corresponding pipeline fault recognition result.

[0042] Based on the recognition results, the backend determines the preset pipelines that need to be repaired and forms corresponding maintenance tasks.

[0043] Specifically, the backend can form corresponding maintenance tasks in the following ways:

[0044] Based on the pipeline defect type and defect level indicated by the recognition result, it is determined whether the preset pipeline needs maintenance. If so, a maintenance task is formed based on the radar time-series image of the preset pipeline and the corresponding recognition result.

[0045] Here, pipeline defect types may include structural defects such as rupture, deformation, misalignment, disconnection, leakage, and functional defects such as scaling and obstruction. Different pipeline defect types may also correspond to different defect levels, which may be set according to the urgency of the required repair. For example, the defect levels may be divided into levels one to three from light to heavy.

[0046] Furthermore, for each pipeline defect type, whether maintenance is required is determined according to its defect level. For example, for leakage defects, maintenance is required when the defect level is level 1. For scaling defects, maintenance is required when the defect level is level 3.

[0047] The backend can send maintenance tasks to corresponding maintenance personnel to repair the pipeline in a timely manner.

[0048] A pipeline inspection system provided in the present application uses radar scanning to achieve non-contact pipeline inspection, and combines AI models to realize intelligent identification of pipeline faults and defect levels. The inspection process is highly automated, avoiding ground excavation and manual inspections. Maintenance tasks can also be automatically generated and issued, which not only saves manpower and material resources, but also improves inspection efficiency and reduces inspection costs.

[0049] In one embodiment of the present application, a method is proposed to recognize radar time-series images of pipelines based on an AI model to determine whether there are defects in the pipeline and the defect level.

[0050] As for the selection of AI models, this embodiment selects the convolutional long short-term memory network (ConvLSTM) as the initial model. The convolutional long short-term memory network combines the advantages of convolution operation and LSTM, which can extract features in space and model dependencies in time.

[0051] The following takes a tubular pipeline as an example to illustrate the training process of the pipeline fault identification model.

[0052] 1. Sample data collection

[0053] Collect a large number of radar time-series images of tubular, normal and defective pipelines. The radar time-series images here are obtained by scanning the geological radar in B-can mode. Among them, for pipelines with smaller diameters (less than 50cm), 400-900MHz antennas can be used. For pipelines with larger diameters, 200-400MHz antennas can be used.

[0054] 2. Preprocessing of sample data

[0055] The radar time series images in the sample are preprocessed by denoising and grayscale conversion.

[0056] For each radar time series image, wavelet transform is performed to obtain the corresponding reconstructed time series image. For each pipeline, based on two adjacent or one-step-spaced reconstructed time series images corresponding to the pipeline, the difference image corresponding to the pipeline is determined.

[0057] A sample set is constructed, wherein each sample in the sample set includes a first reconstructed time series image, a second reconstructed time series image, and a difference image corresponding to a pipeline.

[0058] Specifically, for each radar time series image, the radar time series image can be converted into a reconstructed time series image by the following method:

[0059] Convert the radar time series image into grayscale.

[0060] The radar time series image after grayscale conversion is subjected to two-dimensional discrete wavelet transform to obtain the corresponding approximate coefficient matrix, horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix.

[0061] Here we can obtain the approximate coefficient matrix, horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix as follows:

[0062]

[0063] Among them, A k , m is the value in the kth column and mth row of the approximate coefficient matrix, is the value on the kth column and mth row of the horizontal detail coefficient matrix, is the value on the kth column and mth row of the vertical detail coefficient matrix, is the value on the kth column and mth row of the diagonal detail coefficient matrix, are high-pass filter and low-pass filter respectively, P l , n is the gray value on the kth column and mth row of the radar time series image.

[0064] In a specific embodiment, the wavelet basis function here can use the Daubechies 4 wavelet basis function. Then, the 2D discrete wavelet transform is performed using the `pywt.dwt2` function to obtain an approximate coefficient matrix (cA), a horizontal detail coefficient matrix (cH), a vertical detail coefficient matrix (cV) and a diagonal detail coefficient matrix (cD).

[0065] Specifically, before constructing the wavelet coefficient structure, the method further includes setting the values ​​in the approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix that are smaller than the preset values ​​to zero.

[0066] In this way, the noise data that may exist in the image can be removed, thereby ensuring the quality of the reconstructed time series image to ensure accurate recognition.

[0067] Reconstruction is performed based on the obtained approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix to obtain a reconstructed time series image.

[0068] Then, a wavelet coefficient structure can be constructed based on the obtained approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix, and an inverse wavelet transform is performed on the wavelet coefficient structure to obtain a corresponding reconstructed image.

[0069] Here, we can use the two-dimensional discrete wavelet inverse transform to upsample the approximate coefficient matrix, horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix respectively. And perform low-pass and high-pass filtering on the upsampled results, and finally sum the results to obtain the wavelet coefficient structure, and finally convert it into a reconstructed image.

[0070] The reconstructed image can also be cropped or normalized to ensure that it is consistent with the size of the radar time series image.

[0071] The difference image here can be selected by subtracting the first reconstructed time series image with an earlier time stamp from the second reconstructed time series image with a later time stamp.

[0072] The images in the sample set are annotated, and the corresponding defect types and defect levels are marked. The images are divided into a training set (70%), a validation set (15%), and a test set (15%). Each set of samples includes the first reconstructed time series image, the second reconstructed time series image, and the corresponding difference image.

[0073] 3. Model Training

[0074] The ConvLSTM model here may include an input layer, an input gate, a forget gate, an output gate, a cell state, and an output layer.

[0075] The input data of the input layer is a three-dimensional tensor, including the length value of the two-dimensional image and the time series.

[0076] The input gate calculates and updates the current cell state based on the current input and the cell state at the previous moment through convolution operation and sigmoid activation function. t It can be expressed as:

[0077] i t =σg (W xi *X t +W hi *H t-1 +W ci ℃ t-1 +b i );

[0078] Among them, σ g is the sigmoid activation function, X t is the sample input of the current time step, H t-1 is the hidden state of the previous time step, C t-1 is the memory unit of the previous time step, W xi , W hi , W ci 、b i is the model parameter.

[0079] The forget gate uses convolution operations and sigmoid activation functions to calculate which information needs to be forgotten based on the current input and the cell state at the previous moment. t It can be expressed as:

[0080] f t =σ g (W xf *X t +W hf *H t-1 +W cf ℃ t-1 +b f );

[0081] Among them, W xf , W hf , W cf 、b f is the model parameter.

[0082] The output gate determines the current output based on the current input, the cell state at the previous moment, and the input gate through convolution operation and tanh activation function. t It can be expressed as:

[0083] o t =σ g (W xo *X t +W ho *H t-1 +W co ℃ t +b o );

[0084] Among them, W xo , W ho , Wco 、b o is the model parameter.

[0085] The cell state is updated and adjusted through the calculation results of the input gate, forget gate, and output gate. The cell state C at the current moment t It can be expressed as:

[0086] C t =f t ℃ t-1 +i t °tanh g (W xc *X t +W hc *H t-1 +b c );

[0087] Among them, W xc , W hc 、b c is the model parameter, * is the convolution, ° is the Hadamard product, tanh g is the hyperbolic tangent activation function.

[0088] The output layer converts the cell state into output results through the tanh activation function. Output layer H t It can be expressed as:

[0089] H t =o t °tanh g (C t ).

[0090] In each time step, the spatial information of the time series data can be captured through convolution operations, and the convolution kernel of each time step is shared, ensuring the consistency of spatial feature extraction.

[0091] The ConvLSTM model is trained using the samples in the labeled training set. The corresponding classification accuracy evaluation results are obtained by testing with the samples in the test set, so as to adjust the model parameters such as the convolution kernel size, step size, number of LSTM units, number of layers, etc., and optimize the model through regularization loss functions, etc., so as to obtain a trained pipeline fault recognition model.

[0092] In one embodiment of the present application, a pipeline inspection method is also provided, which can be implemented by a back-end, including: obtaining radar time series images corresponding to each preset pipeline, and inputting a pre-trained pipeline fault recognition model to obtain a recognition result output by the pipeline fault recognition model; based on the recognition result, determining the preset pipeline that needs to be repaired and forming a corresponding maintenance task; wherein, for each preset pipeline, the recognition result output by the pipeline fault recognition model is obtained by the following steps: for all radar time series images corresponding to the preset pipeline, selecting a first radar time series image and a second radar time series image; performing a wavelet transform on the first radar time series image and the second radar time series image to obtain a first reconstructed time series image and a second reconstructed time series image respectively; determining a difference image based on the difference between the first reconstructed time series image and the second reconstructed time series image; extracting spatial features of the difference image, the first reconstructed time series image, and the second reconstructed time series image by a feature extractor of the pipeline fault recognition model; and recognizing the spatial features by a classifier of the pipeline fault recognition model to output a corresponding pipeline fault recognition result.

[0093] The technical effects of the above method embodiments have been demonstrated in the system embodiments and will not be repeated here.

[0094] In the embodiments provided in the present application, it should be understood that the disclosed devices and methods can be implemented in other ways. The device embodiments described above are merely schematic. For example, the division of the units is only a logical function division. There may be other division methods in actual implementation. For example, multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some communication interfaces, and the indirect coupling or communication connection of the devices or units can be electrical, mechanical or other forms.

[0095] In addition, the units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0096] Furthermore, the functional modules in the various embodiments of the present application may be integrated together to form an independent part, or each module may exist separately, or two or more modules may be integrated to form an independent part.

[0097] It should be noted that if the function is implemented in the form of a software function module and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present application can essentially be embodied in the form of a software product, or the part that contributes to the prior art or the part of the technical solution. The computer software product is stored in a storage medium, including a number of instructions to enable 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 application. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM) random access memory (RAM), disk or optical disk, and other media that can store program codes.

[0098] In this document, relational terms such as first and second, etc. are used merely to distinguish one entity or operation from another entity or operation, but do not necessarily require or imply any such actual relationship or order between these entities or operations.

[0099] The above description is only an embodiment of the present application and is not intended to limit the protection scope of the present application. For those skilled in the art, the present application may have various modifications and variations. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present application shall be included in the protection scope of the present application.

Claims

1. A pipeline inspection system, characterized in that: The system includes a control end, an inspection end and a back end, wherein: The back end generates a preset inspection instruction and sends it to the control end, wherein the preset inspection instruction at least includes pipeline information of multiple preset pipelines in the area to be inspected; The control end controls the inspection end to move within the area to be inspected based on the received preset inspection instruction, so that the inspection end collects radar time-series images corresponding to each preset pipeline section; The backend obtains the radar time-series image corresponding to each preset pipeline segment, and inputs the pre-trained pipeline fault recognition model to obtain the recognition result output by the pipeline fault recognition model; The back end determines the preset pipelines that need to be repaired based on the recognition results and forms corresponding repair tasks; For each preset pipeline, the backend obtains the recognition result output by the pipeline fault recognition model through the following steps: For all radar time series images corresponding to the preset pipeline, a first radar time series image and a second radar time series image are selected; Performing wavelet transform on the first radar time series image and the second radar time series image to obtain a first reconstructed time series image and a second reconstructed time series image respectively; Determining a difference image based on a difference between the first reconstructed time-series image and the second reconstructed time-series image; Extracting spatial features of the difference image, the first reconstructed time series image, and the second reconstructed time series image by a feature extractor of the pipeline fault recognition model; The spatial features are identified by the classifier of the pipeline fault identification model to output the corresponding pipeline fault identification result.

2. The system according to claim 1, characterized in that The pipeline information at least includes a category identifier of each preset pipeline segment, and the category identifier is used to distinguish the geometric shapes between pipelines. The back end is also used to determine a corresponding pipeline fault identification model based on the category identifier of the preset pipeline.

3. The system according to claim 1, characterized in that For each radar time series image, the backend converts the radar time series image into a reconstructed time series image in the following way: Convert the radar time series image into grayscale; Perform a two-dimensional discrete wavelet transform on the grayscale-converted radar time series image to obtain a corresponding approximate coefficient matrix, a horizontal detail coefficient matrix, a vertical detail coefficient matrix, and a diagonal detail coefficient matrix; Reconstruction is performed based on the obtained approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix to obtain a reconstructed time series image.

4. The system according to claim 3, characterized in that The back end constructs the wavelet coefficient structure based on the obtained approximate coefficient matrix, horizontal detail coefficient matrix, vertical detail coefficient matrix and diagonal detail coefficient matrix; The back end performs inverse wavelet transform on the wavelet coefficient structure to obtain a corresponding reconstructed image.

5. The system according to claim 4, characterized in that Before the back end constructs the wavelet coefficient structure, it also includes setting the values ​​in the approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix that are less than the preset value to zero.

6. The system according to claim 3, characterized in that The backend obtains the approximation coefficient matrix, horizontal detail coefficient matrix, vertical detail coefficient matrix, and diagonal detail coefficient matrix in the following way: Among them, A k , m is the value on the kth column and mth row of the approximate coefficient matrix, is the value on the kth column and mth row of the horizontal detail coefficient matrix, is the value on the kth column and mth row of the vertical detail coefficient matrix, is the value on the kth column and mth row of the diagonal detail coefficient matrix, are high-pass filter and low-pass filter respectively, P l , n is the gray value on the kth column and mth row of the radar time series image.

7. The system according to claim 1, characterized in that The first radar time-series image and the second radar time-series image are two interval or adjacent radar time-series images. The recognition result output by the pipeline fault recognition model is used to indicate the pipeline defect type and defect level. The back end forms a corresponding maintenance task in the following manner: Determining whether a preset pipeline needs to be repaired based on the pipeline defect type and defect level indicated by the identification result; If so, a maintenance task is formed based on the radar time-series image of the preset pipeline and the corresponding recognition result.

8. The system according to claim 2, characterized in that The backend generates the preset inspection instructions in the following way: In response to the user's click operation on the target area control through the inspection interface, determine the area to be inspected; In response to a user clicking on a target pipeline category control through an inspection interface, a category identifier of a preset pipeline is determined; Based on the category identification of the area to be inspected and the preset pipeline, the pipeline information of the corresponding multiple sections of the preset pipeline is searched from the pipeline network database, and the pipeline information also includes the location data of each section of the preset pipeline; The preset inspection instruction is generated based on the category identification of the preset pipeline and the corresponding pipeline information.

9. A pipeline inspection method, characterized in that: The method comprises: Obtaining radar time-series images corresponding to each preset pipeline segment, and inputting them into a pre-trained pipeline fault recognition model to obtain recognition results output by the pipeline fault recognition model; Based on the recognition result, a preset pipeline that needs to be repaired is determined and a corresponding repair task is formed; For each preset pipeline section, the identification result output by the pipeline fault identification model is obtained through the following steps: For all radar time series images corresponding to the preset pipeline, a first radar time series image and a second radar time series image are selected; Performing wavelet transform on the first radar time series image and the second radar time series image to obtain a first reconstructed time series image and a second reconstructed time series image respectively; Determining a difference image based on a difference between the first reconstructed time-series image and the second reconstructed time-series image; Extracting spatial features of the difference image, the first reconstructed time series image, and the second reconstructed time series image by a feature extractor of the pipeline fault recognition model; The spatial features are identified by the classifier of the pipeline fault identification model to output the corresponding pipeline fault identification result.

10. The method according to claim 9, characterized in that For each radar time series image, the backend converts the radar time series image into a reconstructed time series image in the following way: Convert the radar time series image into grayscale; Perform a two-dimensional discrete wavelet transform on the grayscale-converted radar time series image to obtain a corresponding approximate coefficient matrix, a horizontal detail coefficient matrix, a vertical detail coefficient matrix, and a diagonal detail coefficient matrix; Reconstruction is performed based on the obtained approximate coefficient matrix, the horizontal detail coefficient matrix, the vertical detail coefficient matrix and the diagonal detail coefficient matrix to obtain a reconstructed time series image.

Images