Municipal road management method based on geological radar monitoring technology

Abstract

The present invention relates to the field of road management technology, and in particular to a municipal road management method based on geological radar monitoring technology. The municipal road management method based on geological radar monitoring technology includes: obtaining road geological images, road geological image segmentation operations, road geological image mask operations, filter parameter adaptive adjustment and road defect detection. The present invention obtains road geological images through geological radar, and segments different areas in the road geological images based on noise influence, and then adaptively selects the most suitable filter parameters for the segmented road geological area images, performs different filtering operations, improves the effect of the road geological image filtering operation, and thus improves the accuracy of subsequent municipal road defect detection.

Description

Technical Field

[0001] The present invention relates to the technical field of road management, and in particular to a municipal road management method based on geological radar monitoring technology. Background Art

[0002] With the acceleration of urbanization, the construction and maintenance of municipal roads have become an important part of urban infrastructure management. The long-term use of municipal roads and the influence of environmental factors, such as traffic load, climate change, groundwater infiltration, etc., often lead to various structural defects such as cracks, potholes, and subsidence on the roads. In order to detect and repair these defects in a timely manner, ground penetrating radar (GPR) technology has gradually become an important monitoring method in municipal road management. By emitting and receiving electromagnetic waves, ground penetrating radar can non-destructively detect the surface and deep structures of roads, provide detailed information on the interior of the road, and help detect potential defects in the road, such as cavities, cracks, and subsidence of the base layer.

[0003] However, when penetrating road materials, geological radar signals are easily interfered by various noises, including system noise, environmental noise, and ground clutter. These noises will seriously affect the quality of radar signals, thereby reducing the accuracy of defect detection. In order to improve the quality of radar signals, various filters are usually used to pre-process radar data. However, in practical applications, traditional filters (such as simple mean filtering or Gaussian filtering) are not effective when dealing with complex noise. Specifically, while removing noise, filters often weaken or blur the real signal of road defects, especially when the frequency characteristics of noise and defect signals are similar, it is difficult for filters to effectively distinguish between the two.

[0004] In this case, although the noise is partially suppressed, the filter simultaneously weakens the defect signal, making it difficult to accurately identify important defect features in the subsequent defect detection process, thereby reducing the accuracy of defect detection. Summary of the invention

[0005] The present invention acquires road geological images through geological radar, and segments different areas in the road geological images based on the influence of noise, and then adaptively selects the most suitable filter parameters for the segmented road geological area images, performs different filtering operations, and improves the effect of the filtering operation on the road geological image, thereby improving the accuracy of subsequent municipal road defect detection.

[0006] Municipal road management methods based on geological radar monitoring technology include:

[0007] Obtain road geological images;

[0008] Send the road geological image to the regional segmentation network for processing, and output a number of road geological regional images and their corresponding noise types;

[0009] For any road geological region image, a mask image to be processed corresponding to the road geological region image is constructed. The specific construction method is to construct a mask weight matrix to be processed. The size of the mask weight matrix to be processed is consistent with the size of the road geological image. In the mask weight matrix to be processed, the positions corresponding to the road geological region image mapping are all 1, and the other positions are all 0. The Hadamard product of the road geological region image and the mask weight matrix to be processed is used as the mask image to be processed corresponding to the road geological region image.

[0010] For any mask image to be processed, the mask image to be processed is sent to the image feature extraction model for processing, and a local feature vector is output. Then, the local feature vector corresponding to the mask image to be processed is spliced ​​with the noise type of the road geological area image corresponding to the mask image to be processed, and noise analysis data is constructed. The noise analysis data is sent to the filter parameter output model for processing, and the optimal filter parameter set corresponding to the mask image to be processed is output. The mask image to be processed is filtered by the optimal filter parameter set corresponding to the mask image to be processed, and a filtered road area image is constructed.

[0011] All filtered road area images are added and then standardized to output a processed road geological image. The processed road geological image is then sent to a road defect detection model for processing, and the road defect type is output. Relevant staff are reminded based on the road defects to achieve management of municipal roads.

[0012] As a preferred aspect, the region segmentation network is established based on the U-net network, and training is performed on the region segmentation network, specifically including the following steps:

[0013] A number of noise segmentation training samples are obtained, where the noise segmentation training samples are road geological images, and the noise segmentation training samples are labeled by noise type; all labeled noise segmentation training samples are combined into a noise segmentation training set, and then the noise segmentation training set is sent to a parameter-initialized regional segmentation network for training, a regional segmentation loss value is calculated, and it is determined whether the regional segmentation loss value is within a first preset range. If the regional segmentation loss value is within the first preset range, the trained regional segmentation network is output; otherwise, the regional segmentation network is continuously trained using the noise segmentation training set.

[0014] As a preferred aspect, the filter parameter output model is established based on the BP neural network model, and training is performed on the filter parameter output model, specifically including the following steps:

[0015] A plurality of filter parameter training samples are obtained, wherein the filter training samples include noise analysis data and a corresponding optimal filter parameter set, and the filter parameter training samples are obtained by simulation and calculation based on a swarm optimization algorithm, and all the filter training samples are combined into a filter training set. The filter training set is then sent to a filter parameter output model with parameter initialization for training, and a filter parameter loss value is calculated to determine whether the filter parameter loss value is within a second preset range. If the filter parameter loss value is within the second preset range, the trained filter parameter output model is output; otherwise, the filter parameter output model is continuously trained using the filter training set.

[0016] As a preferred aspect, the filter parameter training samples are obtained by simulating and calculating the swarm optimization algorithm, which specifically includes the following steps:

[0017] A road geological image is randomly selected, a noisy road geological image and its corresponding noise type are obtained, the road geological image is sent to the image feature extraction model for processing, a local feature vector is output, and then the optimal filter parameter set corresponding to the road geological image is obtained by simulation calculation through the swarm optimization algorithm; then the local feature vector and the noise type corresponding to the road geological image are combined into noise analysis data, and the noise analysis data and its corresponding optimal filter parameter set are combined into a filter parameter training sample.

[0018] As a preferred aspect, the optimal filter parameter set corresponding to the road geological image is obtained by simulation calculation using a group optimization algorithm, which specifically includes the following steps:

[0019] S1: Construct N analog filter parameter sets, and combine all analog filter parameter sets into a simulation population set, and the storage form of the analog filter parameter set is {G 1 , G 2 , G 3 , …, G m , …, G M}, where m=1, 2, 3, …, M, and G m is the filter parameter, M is the total number of filter parameters, for example, if the filter is 3×3, then M is 9, and the maximum number of iterations W is set;

[0020] S2: Let w=1, w is used to record the number of iterations;

[0021] S3: Calculate the fitness of each analog filter parameter set, and select the analog filter parameter set with the largest fitness among all analog filter parameter sets and record it as U best , and then simulate the filter parameter set U best Stored in the optimal analog filter parameter set library;

[0022] S4: Based on the fitness of the analog filter parameter set, the analog filter parameter set is selected by a roulette algorithm until 0.5N analog filter parameter sets are selected, and then the selected 0.5N analog filter parameter sets are formed into a parent set, and the remaining 0.5N analog filter parameter sets are formed into a candidate set; based on the candidate set and the analog filter parameter set U best Construct a parent set, and perform a recombination operation on the analog filter parameter sets in the parent set and the parent set. The recombination operation specifically includes randomly selecting an analog filter parameter set from the parent set, randomly selecting an analog filter parameter set from the parent set, and no longer selecting the selected analog filter parameter set. A random integer μ between (1, M) is generated by a random function, and the filter parameters at position μ in the two selected analog filter parameter sets are swapped to generate 0.5N new analog filter parameter sets. The parent set and the 0.5N new analog filter parameter sets are then combined into a transition population set, and a mutation operation is performed on the transition population set. The mutation operation specifically includes selecting analog filter parameter sets one by one from the transition population set, and generating a random number η between (0, 1) by a random function, and judging whether "η>P c "Is it true? c is the mutation threshold, generally 0.7. If “η>P c " is established, a random integer α between (1, M) is generated by a random function, and the filter parameter at position α in the selected analog filter parameter set is mutated, that is, the filter parameter at position α in the selected analog filter parameter set is randomly replaced with another random value within the range. If "η>P c "Not true, no operation; the transition population set after the mutation operation is recorded as the new simulation population set;

[0023] S5: Determine whether "w<W" is true. If "w<W" is true, it means that the maximum number of iterations has not been reached, and return to S3; if "w<W" is not true, it means that the maximum number of iterations has been reached, and enter S6;

[0024] S6: Select the analog filter parameter set with the largest fitness in the optimal analog filter parameter set library as the optimal filter parameter set for output.

[0025] As a preferred aspect, the fitness of each simulation filter parameter set is calculated, which specifically includes the following steps: filtering the noisy road geological image through the simulation filter parameter set, outputting the simulated road geological image, and then using the structural similarity index between the simulated road geological image and the road geological image as the fitness of the simulation filter parameter set.

[0026] As a preferred aspect, based on the candidate set and the simulation filter parameter set Ubest Constructing the parent generation set specifically includes the following steps: traversing the analog filter parameter sets in the candidate set, and for each selected analog filter parameter set, updating the selected analog filter parameter set by the following formula: m =G m +β(u m -G m ), in the updating process, the continuous value change is realized, which makes up for the defect that the genetic algorithm can only change discretely, where β is a random number between 0 and 1, u m is the simulation filter parameter set U best The mth filter parameter in the candidate set is traversed until all analog filter parameter sets in the candidate set are traversed, and all updated analog filter parameter sets are used to construct a parent set.

[0027] As a preferred aspect, the road defect detection model is established based on the Faster-RCNN model, and training is performed on the road defect detection model, specifically including the following steps:

[0028] A plurality of road defect detection training samples are obtained, the road defect detection training samples include road geological images, the road defect detection training samples are annotated by road defects, all the annotated road defect detection training samples are formed into a road defect detection training set, and then the road defect detection training set is sent to a parameter-initialized road defect detection model for training, a road defect loss value is calculated, and it is determined whether the road defect loss value is within a third preset range. If the road defect loss value is within the third preset range, the trained road defect detection model is output; otherwise, the road defect detection model is continuously trained by the road defect detection training set.

[0029] The present invention has the following advantages:

[0030] The present invention acquires road geological images through geological radar, and segments different areas in the road geological images based on the influence of noise, and then adaptively selects the most suitable filter parameters for the segmented road geological area images, performs different filtering operations, and improves the effect of the filtering operation on the road geological image, thereby improving the accuracy of subsequent municipal road defect detection. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] Figure 1 This is a flow chart of the municipal road management method based on geological radar monitoring technology adopted in an embodiment of the present invention. DETAILED DESCRIPTION

[0032] In order to enable persons skilled in the art to better understand the technical solutions in the present invention, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention.

[0033] Example 1, a municipal road management method based on geological radar monitoring technology, such as Figure 1 As shown, including:

[0034] Step Q1: Acquire road geological images;

[0035] Obtain road geological images through geological radar. Road geological images are images that are processed based on underground data obtained by geological radar. They are not visible light images in the traditional sense, but images generated after data processing by recording the interaction between radar waves and underground media through physical phenomena such as reflection and scattering of electromagnetic wave signals;

[0036] Step Q2: road geological image segmentation operation;

[0037] The road geological image is sent to the regional segmentation network for processing. The regional segmentation network can segment the areas in the road geological image that are affected by different noises. The noise influence here includes the environmental noise in the radar data acquisition process, as well as ground clutter and scattering noise, etc. These noises will affect the acquired road geological image, thereby reducing the accuracy of subsequent road detection. Several road geological regional images and their corresponding noise types are output. These road geological regional images represent the areas in the road geological image that are mainly affected by any type of noise. The noise type is generally stored in binary unique hot encoding;

[0038] Step Q3: road geological image mask operation;

[0039] For any road geological region image, a mask image to be processed corresponding to the road geological region image is constructed. The specific construction method is to construct a mask weight matrix to be processed. The size of the mask weight matrix to be processed is consistent with the size of the road geological image. In the mask weight matrix to be processed, the positions corresponding to the road geological region image mapping are all 1, and the remaining positions are all 0. The Hadamard product of the road geological region image and the mask weight matrix to be processed is used as the mask image to be processed corresponding to the road geological region image. Since the output road geological region image is not necessarily a regular rectangle, it is inconvenient to directly perform subsequent processing on the road geological region image. Therefore, the mask image to be processed is constructed by masking to facilitate subsequent operations.

[0040] Step Q4: adaptive adjustment of filter parameters;

[0041] For any mask image to be processed, the mask image to be processed is sent to the image feature extraction model for processing. The image feature extraction model is established based on a pre-trained convolutional neural network, and a local feature vector is output. The local feature vector can reflect the local image features of the road geological area image corresponding to the mask image to be processed, such as texture, edge, gradient and other information. Then, the local feature vector corresponding to the mask image to be processed is spliced ​​with the noise type of the road geological area image corresponding to the mask image to be processed to construct noise analysis data. The noise analysis data is sent to the filter parameter output model for processing, and the optimal filter parameter set corresponding to the mask image to be processed is output. The filter parameter output model can adaptively select the most suitable filter parameters corresponding to the mask image to be processed according to the local image features and noise type corresponding to the mask image to be processed, so as to improve the filtering operation effect for noise. The mask image to be processed is filtered by the optimal filter parameter set corresponding to the mask image to be processed to construct a filtered road area image. It should be noted that the filtering operation actually performs a convolution operation, and the convolution operation corresponding to the filtering operation is a padded convolution operation, which can ensure that the output filtered road area image is consistent in size with the mask image to be processed.

[0042] Step Q5: road defect detection;

[0043] After performing the addition operation on all filtered road area images, the normalization operation is performed to output the processed road geological image, and the denoising operation on the road geological image is completed. The processed road geological image is then sent to the road defect detection model for processing, and the road defect type is output. Based on the road defects, reminders are given to relevant staff, such as SMS reminders or APP information reminders, to achieve the management of municipal roads;

[0044] The present invention acquires road geological images through geological radar, and segments different areas in the road geological images based on the influence of noise, and then adaptively selects the most suitable filter parameters for the segmented road geological area images, performs different filtering operations, and improves the effect of the filtering operation on the road geological image, thereby improving the accuracy of subsequent municipal road defect detection.

[0045] The region segmentation network is established based on the U-net network. The training of the region segmentation network includes the following steps:

[0046] A number of noise segmentation training samples are obtained, which are road geological images actually obtained by geological radar, and the noise segmentation training samples are labeled by noise type. It should be noted that the labeling process can be performed manually, and the labeling method is based on pixel points, that is, the noise type to which each pixel point in the road geological image belongs is labeled; all labeled noise segmentation training samples are composed of a noise segmentation training set, and then the noise segmentation training set is sent to a parameter-initialized region segmentation network for training, and the region segmentation loss value is calculated using a cross entropy loss function to determine whether the region segmentation loss value is within a first preset range. The first preset range is determined based on experience and is used to characterize the accuracy of the region segmentation network. If the region segmentation loss value is within the first preset range, the trained region segmentation network is output; otherwise, the region segmentation network continues to be trained using the noise segmentation training set.

[0047] The filter parameter output model is established based on the BP neural network model, and the training of the filter parameter output model includes the following steps:

[0048] A number of filter parameter training samples are obtained, the filter training samples include noise analysis data and the corresponding optimal filter parameter set, and the filter parameter training samples are obtained by simulation calculation according to the group optimization algorithm. The process of obtaining the filter parameter training samples by simulation calculation of the group optimization algorithm reflects the optimization process for the most suitable filter parameters. All filter training samples are composed of a filter training set, and then the filter training set is sent to the parameter-initialized filter parameter output model for training, and the filter parameter loss value is calculated. The calculation method also uses the cross entropy loss function to determine whether the filter parameter loss value is within a second preset range. The second preset range is determined based on experience and is used to characterize the accuracy of the filter parameter output model. If the filter parameter loss value is within the second preset range, the trained filter parameter output model is output; otherwise, the filter parameter output model continues to be trained through the filter training set.

[0049] The filter parameter training samples are obtained by simulating and calculating the swarm optimization algorithm, which specifically includes the following steps:

[0050] A road geological image is randomly selected to obtain a noisy road geological image and its corresponding noise type corresponding to the road geological image. The road geological image here is a road geological image that is not affected by noise during the experiment. The road geological image is mainly affected by one type of noise through technical means and then the noisy road geological image corresponding to the road geological image is obtained again through geological radar. The road geological image is sent to the image feature extraction model for processing, and the local feature vector is output. Then the optimal filter parameter set corresponding to the road geological image is obtained by simulation calculation through the swarm optimization algorithm; then the local feature vector and the noise type corresponding to the road geological image are composed of noise analysis data, and the noise analysis data and its corresponding optimal filter parameter set are composed of a filter parameter training sample.

[0051] The optimal filter parameter set corresponding to the road geological image is obtained by simulation calculation using a swarm optimization algorithm, specifically including the following steps:

[0052] S1: Construct N analog filter parameter sets, and combine all analog filter parameter sets into a simulation population set, and the storage form of the analog filter parameter set is {G 1 , G 2 , G 3 , …, G m , …, G M}, where m=1, 2, 3, …, M, and G m is the filter parameter, M is the total number of filter parameters, for example, if the filter is 3×3, then M is 9, and the maximum number of iterations W is set; when constructing the filter parameters, the values ​​are randomly selected within the range of the filter parameters;

[0053] S2: Let w=1, w is used to record the number of iterations;

[0054] S3: Calculate the fitness of each analog filter parameter set, and select the analog filter parameter set with the largest fitness among all analog filter parameter sets and record it as U best , and then simulate the filter parameter set U best Stored in the optimal analog filter parameter set library;

[0055] Calculating the fitness of each simulation filter parameter set specifically includes the following steps: performing a filtering operation on the noisy road geological image through the simulation filter parameter set, outputting a simulated road geological image, and then using the structural similarity index between the simulated road geological image and the road geological image as the fitness of the simulation filter parameter set;

[0056] S4: Based on the fitness of the analog filter parameter set, the analog filter parameter set is selected by a roulette algorithm, specifically, the fitness of all analog filter parameter sets is formed into a building disk from small to large, and the position on the roulette disk is determined according to the generated random number of 0-1, and then the analog filter parameter set is selected according to the position on the roulette disk until 0.5N analog filter parameter sets are selected, and then the selected 0.5N analog filter parameter sets are formed into a parent set, and the remaining 0.5N analog filter parameter sets are formed into a candidate set; based on the candidate set and the analog filter parameter set U best Construct a parent set, and perform a recombination operation on the analog filter parameter sets in the parent set and the parent set. The recombination operation specifically includes randomly selecting an analog filter parameter set from the parent set, randomly selecting an analog filter parameter set from the parent set, and no longer selecting the selected analog filter parameter set. A random integer μ between (1, M) is generated by a random function, and the filter parameters at position μ in the two selected analog filter parameter sets are swapped to generate 0.5N new analog filter parameter sets. The parent set and the 0.5N new analog filter parameter sets are then combined into a transition population set, and a mutation operation is performed on the transition population set. The mutation operation specifically includes selecting analog filter parameter sets one by one from the transition population set, and generating a random number η between (0, 1) by a random function, and judging whether "η>P c "Is it true? c is the mutation threshold, generally 0.7. If “η>P c " is established, a random integer α between (1, M) is generated by a random function, and the filter parameter at position α in the selected analog filter parameter set is mutated, that is, the filter parameter at position α in the selected analog filter parameter set is randomly replaced with another random value within the range. If "η>P c "Not true, no operation; the transition population set after the mutation operation is recorded as the new simulation population set;

[0057] Based on the candidate set and the simulated filter parameter set U best Constructing the parent generation set specifically includes the following steps: traversing the analog filter parameter sets in the candidate set, and for each selected analog filter parameter set, updating the selected analog filter parameter set by the following formula: m =G m +β(u m -G m ), in the updating process, the continuous value change is realized, which makes up for the defect that the genetic algorithm can only change discretely, where β is a random number between 0 and 1, u m is the simulation filter parameter set U bestThe mth filter parameter in the candidate set is traversed until all analog filter parameter sets in the candidate set are traversed, and all updated analog filter parameter sets are used to construct a parent set;

[0058] S5: Determine whether "w<W" is true. If "w<W" is true, it means that the maximum number of iterations has not been reached, and return to S3; if "w<W" is not true, it means that the maximum number of iterations has been reached, and enter S6;

[0059] S6: Select the analog filter parameter set with the largest fitness in the optimal analog filter parameter set library as the optimal filter parameter set for output.

[0060] The road defect detection model is established based on the Faster-RCNN model. The training of the road defect detection model includes the following steps:

[0061] A number of road defect detection training samples are obtained, the road defect detection training samples include road geological images, the road defect detection training samples are annotated by road defects, the road defects include potholes, rutting, spalling, roadbed settlement and voids, all the annotated road defect detection training samples are formed into a road defect detection training set, and then the road defect detection training set is sent to a parameter-initialized road defect detection model for training, a road defect loss value is calculated, and it is determined whether the road defect loss value is within a third preset range, the third preset range is determined based on experience, and is used to characterize the accuracy of the road defect detection model. If the road defect loss value is within the third preset range, the trained road defect detection model is output; otherwise, the road defect detection model is continuously trained using the road defect detection training set.

[0062] It should be understood that those skilled in the art can make improvements or changes based on the above description, and all such improvements and changes should fall within the scope of protection of the appended claims of the present invention. Parts not described in detail in this specification belong to the prior art known to those skilled in the art.

Claims

1. A municipal road management method based on geological radar monitoring technology, characterized in that: include: Obtain road geological images; Send the road geological image to the regional segmentation network for processing, and output a number of road geological regional images and their corresponding noise types; For any road geological region image, a mask image to be processed corresponding to the road geological region image is constructed. The specific construction method is to construct a mask weight matrix to be processed. The size of the mask weight matrix to be processed is consistent with the size of the road geological image. In the mask weight matrix to be processed, the positions corresponding to the road geological region image mapping are all 1, and the remaining positions are all 0. The Hadamard product of the road geological region image and the mask weight matrix to be processed is used as the mask image to be processed corresponding to the road geological region image. For any mask image to be processed, the mask image to be processed is sent to the image feature extraction model for processing, and a local feature vector is output. Then, the local feature vector corresponding to the mask image to be processed is spliced ​​with the noise type of the road geological area image corresponding to the mask image to be processed, and noise analysis data is constructed. The noise analysis data is sent to the filter parameter output model for processing, and the optimal filter parameter set corresponding to the mask image to be processed is output. The mask image to be processed is filtered by the optimal filter parameter set corresponding to the mask image to be processed, and a filtered road area image is constructed. All filtered road area images are added and then standardized to output a processed road geological image. The processed road geological image is then sent to a road defect detection model for processing, and the road defect type is output. Reminders are given to relevant staff based on the road defects to achieve management of municipal roads. The optimal filter parameter set corresponding to the road geological image is obtained by simulation calculation using a swarm optimization algorithm, specifically including the following steps: S1: Construct N analog filter parameter sets, and combine all analog filter parameter sets into a simulation population set, and the storage form of the analog filter parameter set is {G1, G2, G3, ..., G m , …, G M }, where m=1, 2, 3, …, M, and G m is the filter parameter, M is the total number of filter parameters, and the maximum number of iterations W is set; S2: Let w=1, w is used to record the number of iterations; S3: Calculate the fitness of each analog filter parameter set, and select the analog filter parameter set with the largest fitness among all analog filter parameter sets and record it as U best , and then simulate the filter parameter set U best Stored in the optimal analog filter parameter set library; S4: Based on the fitness of the analog filter parameter set, the analog filter parameter set is selected by a roulette algorithm until 0.5N analog filter parameter sets are selected, and then the selected 0.5N analog filter parameter sets are formed into a parent set, and the remaining 0.5N analog filter parameter sets are formed into a candidate set; based on the candidate set and the analog filter parameter set U best Construct a parent set, and perform a recombination operation on the analog filter parameter sets in the parent set and the parent set. The recombination operation specifically includes randomly selecting an analog filter parameter set from the parent set, randomly selecting an analog filter parameter set from the parent set, and the selected analog filter parameter set is no longer selected. A random integer μ between (1, M) is generated by a random function, and the filter parameters at position μ in the two selected analog filter parameter sets are exchanged to generate 0.5N new analog filter parameter sets. The parent set and the 0.5N new analog filter parameter sets are then combined into a transition population set, and a mutation operation is performed on the transition population set. The mutation operation specifically includes selecting analog filter parameter sets one by one from the transition population set, and generating a random number η between (0, 1) by a random function. It is judged that "η>P c "Is it true? c is the mutation threshold, if "η>P c " is established, a random integer α between (1, M) is generated by a random function, and the filter parameter at position α in the selected analog filter parameter set is mutated, that is, the filter parameter at position α in the selected analog filter parameter set is randomly replaced with another random value within the range. If "η>P c "Not true, no operation; the transition population set after the mutation operation is recorded as the new simulation population set; S5: Determine whether "w<W" holds. If "w<W" holds, it means that the maximum number of iterations has not been reached, and return to S3; if "w<W" does not hold, it means that the maximum number of iterations has been reached, and enter S6; S6: Selecting the analog filter parameter set with the largest fitness in the optimal analog filter parameter set library as the optimal filter parameter set for output; Calculating the fitness of each simulation filter parameter set specifically includes the following steps: performing a filtering operation on the noisy road geological image through the simulation filter parameter set, outputting a simulated road geological image, and then using the structural similarity index between the simulated road geological image and the road geological image as the fitness of the simulation filter parameter set; Based on the candidate set and the simulated filter parameter set U best Constructing the parent generation set specifically includes the following steps: traversing the analog filter parameter sets in the candidate set, and for each selected analog filter parameter set, updating the selected analog filter parameter set by the following formula: m =G m +β(u m -G m ), where β is a random number between 0 and 1, and u m is the simulation filter parameter set U best The mth filter parameter in the candidate set is traversed until all analog filter parameter sets in the candidate set are traversed, and all updated analog filter parameter sets are used to construct a parent set.

2. The municipal road management method based on geological radar monitoring technology according to claim 1 is characterized in that: The region segmentation network is established based on the U-net network. The training of the region segmentation network includes the following steps: A number of noise segmentation training samples are obtained, where the noise segmentation training samples are road geological images, and the noise segmentation training samples are labeled by noise type; all labeled noise segmentation training samples are combined into a noise segmentation training set, and then the noise segmentation training set is sent to a parameter-initialized regional segmentation network for training, a regional segmentation loss value is calculated, and it is determined whether the regional segmentation loss value is within a first preset range. If the regional segmentation loss value is within the first preset range, the trained regional segmentation network is output; otherwise, the regional segmentation network is continuously trained using the noise segmentation training set.

3. The municipal road management method based on geological radar monitoring technology according to claim 2 is characterized in that: The filter parameter output model is established based on the BP neural network model, and the training of the filter parameter output model includes the following steps: A plurality of filter parameter training samples are obtained, wherein the filter training samples include noise analysis data and a corresponding optimal filter parameter set, and the filter parameter training samples are obtained by simulation and calculation based on a swarm optimization algorithm, and all the filter training samples are combined into a filter training set. The filter training set is then sent to a filter parameter output model with parameter initialization for training, and a filter parameter loss value is calculated to determine whether the filter parameter loss value is within a second preset range. If the filter parameter loss value is within the second preset range, the trained filter parameter output model is output; otherwise, the filter parameter output model is continuously trained using the filter training set.

4. The municipal road management method based on geological radar monitoring technology according to claim 3 is characterized in that: The filter parameter training samples are obtained by simulating and calculating the swarm optimization algorithm, which specifically includes the following steps: A road geological image is randomly selected, a noisy road geological image and its corresponding noise type are obtained, the road geological image is sent to the image feature extraction model for processing, a local feature vector is output, and then the optimal filter parameter set corresponding to the road geological image is obtained by simulation calculation through the swarm optimization algorithm; then the local feature vector and the noise type corresponding to the road geological image are combined into noise analysis data, and the noise analysis data and its corresponding optimal filter parameter set are combined into a filter parameter training sample.

5. The municipal road management method based on geological radar monitoring technology according to claim 4 is characterized in that: The road defect detection model is established based on the Faster-RCNN model. The training of the road defect detection model includes the following steps: A plurality of road defect detection training samples are obtained, the road defect detection training samples include road geological images, the road defect detection training samples are annotated by road defects, all the annotated road defect detection training samples are formed into a road defect detection training set, and then the road defect detection training set is sent to a parameter-initialized road defect detection model for training, a road defect loss value is calculated, and it is determined whether the road defect loss value is within a third preset range. If the road defect loss value is within the third preset range, the trained road defect detection model is output; otherwise, the road defect detection model is continuously trained by the road defect detection training set.

Images