A method and system for processing turbidity disease images of underwater bridge structures

Abstract

The present invention discloses a method and system for processing turbidity disease images of underwater bridge structures, and relates to the technical field of underwater bridge detection. The present invention comprises: acquiring underwater defect images of bridge piers, and numbering each acquired image to form a turbidity image dataset of underwater bridge structures; further dividing the turbidity image dataset into a training set and a test set, wherein the training set includes TrainA and TrainB, and the test set includes TestA and TestB; constructing a CycleGAN-deturbidity model for adaptive deturbidity removal of underwater bridge structure images; inputting the training set into the deturbidity removal model CycleGAN-deturbidity for training to obtain the model weights. The present invention enhances the texture extraction capability of underwater images by introducing TEM and PTFEM into the original CycleGAN model, and further enhances the model's perception capability of images by adding perceptual loss to the loss function. This method can achieve effective deturbidity removal of underwater bridge structure images under different turbidities and effective restoration of pier images.

Description

Technical Field

[0001] The present invention relates to the technical field of underwater bridge detection, and in particular to a method and system for processing turbidity disease images of underwater bridge structures. Background Art

[0002] A large number of bridges around the world have entered the stage of large-scale operation and maintenance. Bridges can be divided into land bridges and water-crossing bridges according to their environment. Water-crossing bridges include bridges across rivers, streams and straits. The inspection and monitoring of bridge superstructures, including beams, pavements, pylons and cables, has attracted widespread attention. However, the underwater parts of water-crossing bridges have received relatively less attention, and the means to detect the underwater structural status are quite scarce. In order to achieve precise detection of underwater structural defects of bridges, the current mainstream method still relies on professional divers using underwater cameras to manually collect images and determine defects.

[0003] The reliability of underwater structural defect detection generally depends on the quality of underwater imagery. However, idealized low-turbidity datasets are often used in underwater optical image defect recognition of bridges. In reality, the turbidity of water bodies in which cross-water bridges are located typically exceeds 10 NTU, resulting in poor optical imaging quality. The target structure is easily obscured by particles in the water, thus affecting defect recognition. To this end, we propose a method and system for processing turbidity images of underwater bridge structures. Summary of the Invention

[0004] The purpose of the present invention is to provide a method and system for processing turbidity disease images of underwater bridge structures, which can effectively remove turbidity from underwater bridge structure images under different turbidity levels and effectively restore bridge pier images.

[0005] To achieve the above-mentioned object, the present invention provides the following technical solution: a method for processing turbidity disease images of underwater bridge structures, comprising the following steps:

[0006] Acquire underwater defect images of bridge piers and number each acquired image to form a turbidity image dataset of underwater bridge structures;

[0007] The turbid image dataset is further divided into a training set and a test set with a division ratio of 4:1. The training set includes TrainA and TrainB, and the test set includes TestA and TestB.

[0008] Constructing the CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures;

[0009] Input the training set into the deturbidity model CycleGAN-deturbidity for training to obtain the model weights;

[0010] The trained weights are tested on TestA to obtain the de-turbidity image of the underwater structure of the bridge.

[0011] Furthermore, the underwater defect images of the bridge piers include defect images of multiple bridge piers, defect images of different water turbidity levels, and defect images of different shooting distances.

[0012] Furthermore, each acquired image is numbered to form a turbidity image dataset of the underwater structure of the bridge, and is numbered according to a number combination. The specific rule is: turbidity level - pier number - angle - distance - different water depths.

[0013] Furthermore, the minimum turbidity of the water quality turbidity level is set to no water, the maximum turbidity is set to 30 NTU, and a turbidity level is set every 5 NTU, which are 0 to 7 levels respectively.

[0014] Furthermore, the bridge structure underwater image adaptive de-turbidity model includes a generator, a discriminator, and a loss function. Specifically, the CycleGAN model is used as the baseline model, ResNet is used as the basic architecture of the generator, and PatchGAN is used as the basic architecture of the discriminator.

[0015] The loss function is LSGAN loss, cycle consistency loss and perceptual loss as the overall loss function of CycleGAN-deturbidity:

[0016]

[0017] The perceptual loss weight coefficient γ is 0.1, G A , G B Represents generators A and B, D A 、D B Represents discriminators A and B, That is, LSGAN loss, is the cycle consistency loss function, for perceived loss;

[0018] Furthermore, a texture enhancement module TEM and a pyramid feature extraction module PTFEM are introduced into the generator network of CycleGAN. The texture enhancement module TEM adopts 1-dQCO, while the pyramid feature extraction module PTFEM adopts 2-dQCO.

[0019] TEM converts the feature map into a quantization coding map E and statistical features D through 1-dQCO, where D plays the role of a histogram. D is used to obtain a new quantization level L. Each new level is obtained by perceiving the statistical information of the initial level, so it can be regarded as a graph, where each quantization level can be regarded as a node in the graph. The reconstructed quantization level L' is then assigned to each pixel using the quantization coding map E to obtain the output R.

[0020] PTFEM inputs the feature map into 2-dQCO to obtain C×N×N co-occurrence statistical features F, where C represents the number of channels and N represents the number of quantization levels. It then generates the texture features of the region through the MLP layer and level-wise average.

[0021] In addition, PTFEM adopts a pyramid structure to transmit the input feature map from four different scales [1, 2, 4, 8]. In each branch, the feature map is divided into a different number of sub-regions. Each sub-region is fully utilized by the texture feature extraction unit to fully utilize the texture representation corresponding to the region. Subsequently, each feature map is upsampled and spliced ​​by channel for output.

[0022] Furthermore, the generator structure is divided into a basic network and a texture feature extraction branch;

[0023] The base network includes 6 ResNetBlocks, which are used to extract basic features of the image. The texture feature extraction branch downsamples the first and second layer features from the backbone network ResNet to the same size as the base network output and then concatenates them in series to generate shallow features. These features are then input into the TEM module to enhance texture details, and the PTFEM module extracts multi-scale statistical texture information.

[0024] Finally, the output features of the basic network, TEM and PTFEM are concatenated together, and the final decongested result image is obtained after upsampling and a convolution layer.

[0025] According to one aspect of the present invention, a system for processing underwater turbidity damage data for bridges is provided, comprising:

[0026] Acquisition device: used to obtain underwater defect images of bridge piers and number each acquired image to form a turbidity image dataset of underwater bridge structures;

[0027] A division device: used to further divide the turbid image data set into a training set and a test set, the training set includes TrainA and TrainB, and the test set includes TestA and TestB;

[0028] Model building device: used to build the CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures;

[0029] Training device: used to input the training set into the deturbidity model CycleGAN-deturbidity for training to obtain the model weights;

[0030] Test output device: used to test the trained weights on TestA to obtain the de-turbidity image of the underwater structure of the bridge.

[0031] Furthermore, the collection device includes a glass fiber reinforced plastic transparent water tank, an underwater camera is installed inside the glass fiber reinforced plastic transparent water tank, a bridge pier with apparent defects is placed on the other side, a multi-parameter water quality sensor is also installed inside the water tank, and a control unit is provided outside the water tank;

[0032] The underwater camera controls the acquisition frequency of camera images and stores image data through the control unit, and the multi-parameter water quality sensor controls its sampling frequency and stores turbidity data through the control unit.

[0033] Furthermore, the underwater defect image of the bridge pier is obtained using the acquisition device. The specific steps are as follows:

[0034] (1) Prepare aqueous solutions with different turbidity levels, and place the aqueous solutions into a glass fiber reinforced plastic transparent water tank and mix them evenly;

[0035] (2) Make concrete scaled cylindrical bridge piers with different surface defects, mark the defect points every 60°, and then place the concrete scaled bridge piers into the device;

[0036] (3) Install an underwater camera and a multi-parameter water quality sensor in the aqueous solution of a glass fiber reinforced plastic transparent water tank, and use the turbidity data measured by the multi-parameter water quality sensor to control the increase and decrease of soil to achieve the turbidity range requirements of different turbidity levels;

[0037] (4) During actual sampling, the bridge pier is rotated so that the defect mark point faces the underwater camera, and the underwater camera is used to capture defect images of multiple bridge piers, defect images of different water turbidity levels, and defect images at different shooting distances.

[0038] The present invention has at least the following beneficial effects:

[0039] 1. This paper proposes a standard turbidity dataset focusing on underwater bridge structures, aiming to support underwater image deturbidity tasks and provide benchmark data for subsequent underwater image deturbidity research.

[0040] 2. The CycleGAN-deturbidity model proposed in this paper can realize the conversion of features between one image domain and another, that is, it can realize the transformation from the high-turbidity image domain to the low-turbidity image domain.

[0041] 3. The introduction of TEM, PTFEM and perceptual loss helps to restore the texture and features of concrete bridge piers and their defects, and ultimately achieves effective deturbidity of underwater structure images of bridges under different turbidity levels and effective restoration of bridge pier images.

[0042] Of course, any product implementing the present invention does not necessarily need to achieve all of the advantages described above at the same time. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] Figure 1 is a flow chart of the method of the present invention;

[0044] Figure 2 It is a three-dimensional schematic diagram of the collection device of the present invention;

[0045] Figure 3 This is a diagram of the QCO operator network structure used in TEM and PTFEM described in the present invention;

[0046] Figure 4 This is a diagram of the generator network structure of the CycleGAN-deturbidity deturbidity model described in the present invention;

[0047] Figure 5 It is a schematic diagram of the results of a specific embodiment of underwater image dehazing using the method of the present invention. DETAILED DESCRIPTION

[0048] The following will be combined with the accompanying drawings in the embodiments of the present disclosure to clearly and completely describe the technical solutions in the embodiments of the present disclosure. Obviously, the embodiments described are only part of the embodiments of the present disclosure, not all of the embodiments. Based on the embodiments of the present disclosure, all other embodiments obtained by ordinary technicians in this field without making any creative efforts are within the scope of protection of the present disclosure.

[0049] See also Figure 1-5 The present invention provides a technical solution: a method for processing turbidity disease images of underwater bridge structures, comprising the following steps:

[0050] S1 obtains underwater defect images of the pier 4 and numbers each image obtained to form a turbid image dataset of the underwater structure of the bridge;

[0051] It should be noted that the number of piers 4 is set to 10, which are numbered from 1 to 10. The underwater defect images of piers 4 include multiple defect images of piers 4, defect images of different water turbidity levels, and defect images at different shooting distances.

[0052] Furthermore, each collected image is numbered according to a number combination, with the following rule: turbidity level - pier 4 - angle - distance - different depths of immersion in water. For example, in the image number "01000101", '0' indicates that the turbidity level is 0, '1' indicates that pier 4 is 1, '000' indicates that the image was taken at a 40° angle to the pier, '10' indicates that the camera was 10 cm away from the surface of pier 4 during the actual shooting, and the last '1' indicates that the camera was immersed in water at a shallow depth. For the technical solution of this embodiment, a total of 2880 images were collected, of which 360 were underwater images from different piers 4 at each turbidity level, and each pier 4 had 288 images from different turbidity levels.

[0053] S2. The turbid image dataset is further divided into a training set and a test set with a division ratio of 4:1. The training set includes TrainA and TrainB, and the test set includes TestA and TestB.

[0054] Furthermore, the images collected from piers 4 No. 1 to 8 are used as a training set, and the images collected from piers 4 No. 9 and 10 are used as a test set; the image data with turbidity levels of 0, 1, and 2 are divided into TrainB and TestB, and the image data with turbidity levels of 3 to 7 are divided into TrainA and TestA;

[0055] S3. Constructing a CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures;

[0056] S4. The training set is input into the deturbidity model CycleGAN-deturbidity for training to obtain the model weights;

[0057] S5. Test the trained weights on TestA to obtain a de-turbidified image of the underwater structure of the bridge.

[0058] Constructing the CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures, specifically:

[0059] (1): CycleGAN model is used as the baseline model, ResNet is used as the basic architecture of the generator, and PatchGAN is used as the basic architecture of the discriminator;

[0060] (2): The loss function is LSGAN loss, cycle consistency loss and perceptual loss as the overall loss function of CycleGAN-deturbidity:

[0061]

[0062] The perceptual loss weight coefficient γ is 0.1, G A , G B Represents generators A and B, D A 、D B Represents discriminators A and B, That is, LSGAN loss, is the cycle consistency loss function, for perceived loss;

[0063] (3): The texture enhancement module TEM and the pyramid feature extraction module PTFEM are introduced into the generator network of CycleGAN. TEM adopts 1-d QCO and PTFEM adopts 2-d QCO. The structure of QCO is shown in the attached figure. Figure 3 TEM converts the feature map into a quantization encoding map E and statistical features D through a 1-d QCO, where D acts as a histogram. Subsequently, D is used to derive a new quantization level L. Each new level is obtained by perceiving the statistics of the initial level, so it can be viewed as a graph, where each quantization level can be considered a node. The reconstructed quantization level L' is then assigned to each pixel using the quantization encoding map E, resulting in the output R. PTFEM first inputs the feature map into a 2-d QCO to obtain a C×N×N co-occurrence statistical feature F, where C represents the number of channels and N represents the number of quantization levels. Texture features for that region are then generated through an MLP layer and level-wise average. Furthermore, PTFEM employs a pyramidal structure, passing the input feature map at four different scales: [1, 2, 4, 8]. In each branch, the feature map is partitioned into a different number of subregions, each of which is then processed through a texture feature extraction unit to fully utilize the corresponding texture representation. Each resulting feature map is then upsampled and concatenated channel-wise for output.

[0064] (4): The improved generator structure is shown in the attached Figure 4, divided into a basic network and a texture feature extraction branch; the basic network is mainly composed of 6 ResNetBlocks, which is used to extract the basic features of the image; the texture feature extraction branch downsamples the first and second layer features from the backbone network ResNet to the same size as the basic network output and then concatenates them in series to generate shallow features; subsequently, these features are input into the TEM module to enhance texture details, and the multi-scale statistical texture information is extracted through the PTFEM module; finally, the output features of the basic network, TEM and PTFEM are cascaded together, and the final dehazing result image is obtained after upsampling and a convolution layer.

[0065] According to one aspect of the present invention, a system for processing underwater turbidity damage data for bridges is provided, comprising:

[0066] Acquisition device: used to acquire underwater defect images of the bridge pier 4 and number each acquired image to form a turbidity image dataset of the bridge underwater structure;

[0067] A division device: used to further divide the turbid image data set into a training set and a test set, the training set includes TrainA and TrainB, and the test set includes TestA and TestB;

[0068] Model building device: used to build the CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures;

[0069] Training device: used to input the training set into the deturbidity model CycleGAN-deturbidity for training to obtain the model weights;

[0070] Test output device: used to test the trained weights on TestA to obtain the de-turbidity image of the underwater structure of the bridge.

[0071] Further, such as Figure 2 As shown, the acquisition device is mainly aimed at the underwater structure of the bridge, focusing on the acquisition of images of the underwater structure of the bridge with different turbidity, so as to provide a benchmark data set for the deturbidity task. The acquisition device includes a fiberglass transparent water tank 1, a Gopro11 underwater camera 2 is installed inside the fiberglass transparent water tank 1, a bridge pier 4 with apparent defects is placed on the other side, an S600-M multi-parameter water quality sensor 3 is also installed inside the water tank, and a control unit 5 is arranged outside the water tank. The control unit 5 can be a computer device such as a desktop computer, a laptop computer or a cloud server, and the control unit 5 includes but is not limited to a processor and a memory. For example, the control unit 5 can also include input and output devices, network access devices and buses, etc.

[0072] The underwater camera 2 controls the acquisition frequency of camera images and stores image data through the control unit 5 , and the multi-parameter water quality sensor 3 controls its sampling frequency and stores turbidity data through the control unit 5 .

[0073] Furthermore, an underwater defect image of the pier 4 is acquired using an acquisition device. The specific steps are as follows:

[0074] (1) Prepare aqueous solutions with different turbidity levels, and put the aqueous solutions into a glass fiber reinforced plastic transparent water tank 1 and mix them evenly;

[0075] To create turbid water environments with different turbidity levels, the water turbidity was first graded, with the maximum turbidity set at 30 NTU and a turbidity level set every 5 NTU. Secondly, an aqueous solution was prepared using surface soil from the riverbed of the Yangtze River Basin. An underwater camera 2 and a multi-parameter water quality sensor 3 were placed in the aqueous solution in a fiberglass reinforced plastic transparent water tank 1. The turbidity in the fiberglass transparent water tank 1 was uniformly distributed by stirring the aqueous solution. The turbidity data measured by the multi-parameter water quality sensor 3 was used to control the increase and decrease of soil to achieve the turbidity range requirements for different turbidity levels, as shown in the following table:

[0076] Table 1 Water turbidity classification

[0077]

[0078] (2) Making a concrete scaled cylindrical bridge pier 4 with different surface defects, marking the defect points every 60°, and then placing the concrete scaled bridge pier 4 into the device;

[0079] Specifically, a scaled-down concrete cylindrical bridge pier 4 with different surface defects was fabricated. This consisted of 10 manually cast scaled-down cylindrical concrete bridge piers 4, all measuring 500mm (D) x 500mm (H). The surfaces of the piers 4 contained multiple surface defects of varying shapes, types, and sizes, including exposed reinforcement, circular defects, square defects, and irregular defects. These defects were primarily created by pasting foam of varying shapes onto the inner surface of the formwork. The purpose of designing defects of varying shapes, positions, and sizes was to increase the diversity of the collected underwater images of the piers 4, enrich the dataset, and thus improve the generalization capability of the de-turbidity model.

[0080] (3) Install an underwater camera 2 and a multi-parameter water quality sensor 3 in the aqueous solution of a glass fiber reinforced plastic transparent water tank 1, and measure the turbidity data through the multi-parameter water quality sensor 3;

[0081] (4) During actual sampling, the bridge pier 4 is rotated so that the defect mark point faces the underwater camera 2, and the underwater camera 2 is used to capture defect images of multiple bridge piers 4, defect images of different water turbidity levels, and defect images at different shooting distances.

[0082] Furthermore, underwater defect images were collected at different distances between the camera and the pier 4 under test. Due to the refractive index in water, the field of view becomes narrower when collecting underwater images at the same distance compared to that in air. Therefore, the shooting distance between the underwater camera 2 and the pier 4 under test at turbidity level 0 is smaller than that at other turbidity levels. When the turbidity level is 0, the shooting distances between the camera and the pier 4 are set to 10 cm, 20 cm, and 30 cm, respectively. When the shooting distance is 10 cm, in order to ensure that the captured image can cover the surface of the pier 4 at this angle, the height of the top of the underwater camera 2 from the upper plane of the wooden support is 28.2 cm. , 40.2cm and 52.8cm respectively; similarly, when the shooting distance is 20 cm, the heights are 37.5cm and 47.5cm respectively; when the shooting distance is 30 cm, the height is 48cm; when the turbidity level is greater than 0, the distances between the camera and the pier 4 are set to 10cm, 20cm, 30cm respectively; when the shooting distance is 23cm, the heights of the top of the underwater camera 2 from the upper plane of the wooden support are 29cm, 40cm and 51cm respectively; similarly, when the shooting distance is 33cm, the heights are 35cm and 45cm respectively; when the shooting distance is 43cm, the height is 40cm.

[0083] like Figure 5 As shown in Figure 3, the de-turbidity effect of the underwater structure images of the bridge at different turbidity levels is quite obvious, and the texture and contour of the pier 4 and the defect are effectively restored.

[0084] In summary, the technical solution of the present invention constructs an underwater structure turbidity dataset through the proposed underwater turbidity image acquisition device, introduces TEM and PTFEM into the original CycleGAN model to enhance the texture extraction capability of underwater images, and adds perceptual loss to the loss function to further improve the model's image perception ability. This method can achieve effective deturbidity of underwater structure images of bridges under different turbidities and effective restoration of pier 4 images.

[0085] It should be noted that, in this document, relational terms such as first and second, etc., are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. Moreover, the terms "comprises," "comprising," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that includes a list of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or apparatus.

[0086] For those of ordinary skill in the art, the specific meanings of the above terms in the present invention can be understood according to the specific circumstances. When an element is referred to as being "assembled on", "installed on", "fixed on" or "set on" another element, it can be directly on the other element or there can be a central element. When an element is considered to be "connected" to another element, it can be directly connected to the other element or there can be a central element at the same time. The terms "vertical", "horizontal", "up", "down", "left", "right" and similar expressions used herein are for illustrative purposes only and are not intended to be the only embodiment.

[0087] While embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions, and variations may be made to these embodiments without departing from the principles and spirit of the invention, and that the scope of the invention is defined by the appended claims and their equivalents.

[0088] Throughout this specification, references to terms such as "one embodiment," "example," or "specific example" indicate that a specific feature, structure, material, or characteristic described in conjunction with that embodiment or example is included in at least one embodiment or example of the present disclosure. In this specification, schematic representations of these terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in any one or more embodiments or examples.

Claims

1. A method for processing turbidity disease images of underwater bridge structures, characterized in that: The following steps are involved: Acquire underwater defect images of bridge piers and number each acquired image to form a turbidity image dataset of underwater bridge structures; The turbid image dataset is further divided into a training set and a test set with a division ratio of 4:

1. The training set includes TrainA and TrainB, and the test set includes TestA and TestB. Constructing the CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures; Input the training set into the deturbidity model CycleGAN-deturbidity for training to obtain the model weights; The trained weights are tested on TestA to obtain the de-turbidity image of the underwater structure of the bridge; The adaptive de-clouding model for underwater bridge structure images includes a generator, a discriminator, and a loss function. Specifically, the CycleGAN model is used as the baseline model, ResNet is used as the basic architecture of the generator, and PatchGAN is used as the basic architecture of the discriminator. The loss function is LSGAN loss, cycle consistency loss and perceptual loss as the overall loss function of CycleGAN-deturbidity: The perceptual loss weight coefficient γ is 0.1, G A , G B Represents generators A and B, D A 、D B Represents discriminators A and B, That is, LSGAN loss, is the cycle consistency loss function, for perceived loss; The texture enhancement module TEM and the pyramid feature extraction module PTFEM are introduced into the generator network of CycleGAN. The texture enhancement module TEM adopts 1-dQCO, while the pyramid feature extraction module PTFEM adopts 2-dQCO. TEM converts the feature map into a quantization coding map E and statistical features D through 1-dQCO, where D plays the role of a histogram. D is used to obtain a new quantization level L. Each new level is obtained by perceiving the statistical information of the initial level, so it can be regarded as a graph, where each quantization level can be regarded as a node in the graph. The reconstructed quantization level L' is then assigned to each pixel using the quantization coding map E to obtain the output R. PTFEM inputs the feature map into 2-dQCO to obtain C×N×N co-occurrence statistical features F, where C represents the number of channels and N represents the number of quantization levels; In addition, PTFEM adopts a pyramid structure to transmit the input feature map from four different scales [1, 2, 4, 8]. In each branch, the feature map is divided into a different number of sub-regions. Each sub-region is passed through the MLP layer and level-wise average to generate the texture features of the region. Subsequently, each feature map is upsampled and spliced ​​by channel. The generator structure is divided into a basic network and a texture feature extraction branch; The base network includes 6 ResNetBlocks, which are used to extract basic features of the image. The texture feature extraction branch downsamples the first and second layer features from the backbone network ResNet to the same size as the base network output and then concatenates them in series to generate shallow features. These features are then input into the TEM module to enhance texture details, and the PTFEM module extracts multi-scale statistical texture information. Finally, the output features of the basic network, TEM and PTFEM are concatenated together, and the final decongested result image is obtained after upsampling and a convolution layer.

2. The method for processing turbidity damage images of underwater bridge structures according to claim 1, characterized in that: The underwater defect images of the bridge piers include defect images of multiple bridge piers, defect images of different water turbidity levels, and defect images of different shooting distances.

3. The method for processing turbidity damage images of underwater bridge structures according to claim 2, characterized in that: Each acquired image is numbered to form a turbidity image dataset of underwater bridge structures. The images are numbered according to the following numerical combinations: turbidity level - pier number - angle - distance - different water depths.

4. The method for processing turbidity damage images of underwater bridge structures according to claim 3, characterized in that: The minimum turbidity of the water quality turbidity level is set to no water, the maximum turbidity is set to 30 NTU, and a turbidity level is set every 5 NTU, ranging from 0 to 7.

5. A bridge underwater turbidity disease data processing system, used to implement the bridge underwater structure turbidity disease image processing method according to any one of claims 1 to 4, characterized in that: include: Acquisition device: used to obtain underwater defect images of bridge piers and number each acquired image to form a turbidity image dataset of underwater bridge structures; A division device: used to further divide the turbid image data set into a training set and a test set, the training set includes TrainA and TrainB, and the test set includes TestA and TestB; Model building device: used to build the CycleGAN-deturbidity model for adaptive deturbidity of underwater images of bridge structures; Training device: used to input the training set into the deturbidity model CycleGAN-deturbidity for training to obtain the model weights; Test output device: used to test the trained weights on TestA to obtain the de-turbidity image of the underwater structure of the bridge.

6. The bridge underwater turbidity disease data processing system according to claim 5 is characterized by: The collection device includes a glass fiber reinforced plastic transparent water tank, an underwater camera is installed inside the glass fiber reinforced plastic transparent water tank, a bridge pier with apparent defects is placed on the other side, a multi-parameter water quality sensor is also installed inside the water tank, and a control unit is set outside the water tank; The underwater camera controls the acquisition frequency of camera images and stores image data through the control unit, and the multi-parameter water quality sensor controls its sampling frequency and stores turbidity data through the control unit.

7. The bridge underwater turbidity disease data processing system according to claim 6 is characterized by: The acquisition device is used to obtain underwater defect images of bridge piers. The specific steps are as follows: (1) Prepare aqueous solutions with different turbidity levels, and put them into a glass fiber reinforced plastic transparent water tank to mix them evenly; (2) Make concrete scaled cylindrical bridge piers with different surface defects, mark the defect points every 60°, and then place the concrete scaled bridge piers into the device; (3) Install an underwater camera and a multi-parameter water quality sensor in the water solution of the glass fiber reinforced plastic transparent water tank. The turbidity data measured by the multi-parameter water quality sensor is used to control the increase and decrease of soil to achieve the turbidity range requirements of different turbidity levels; (4) During actual sampling, the bridge pier was rotated so that the defect mark was facing the underwater camera. The underwater camera was used to capture defect images of multiple bridge piers, defect images of different water turbidity levels, and defect images at different shooting distances.

Images