Bridge cable sheath disease detection method and system

Abstract

The invention relates to a bridge cable sheath disease detection method. The method comprises the following steps: acquiring a to-be-detected bridge cable sheath image; identifying the to-be-detected bridge cable sheath image through a semantic segmentation model obtained through pre-training to obtain a cable sheath disease type and a separately stripped disease area grey-scale map; selecting a specified disease analysis model based on the cable sheath disease type; and analyzing the gray-scale map of the disease area through a pre-trained disease analysis model to obtain the disease degree of the cable sheath. According to the bridge inhaul cable sheath disease detection method provided by the invention, disease detection of the inhaul cable sheath is carried out in an image recognition mode, so that the recognition accuracy is improved; and in combination with the characteristic of the disease condition easily occurring in the cable sheath, the disease type identification and disease degree analysis of the image are divided into two stages, so that the method is more targeted, and the identification accuracy is higher.

Description

Technical Field

[0001] The present invention relates to the technical field of disease identification, and in particular to a method and system for detecting diseases of a bridge cable sheath. Background Art

[0002] In order to protect the cables of cable-stayed bridges from corrosion and damage, PE (polyethylene) is usually used to wrap the cables as cable sheaths. Cable sheaths can provide corrosion protection and mechanical protection, have the advantages of being waterproof and low in friction, are easy to maintain, help extend the life of the cables, and ensure their stable performance in various environments. However, since bridge cables are subject to tension, their strength, heat resistance, and aging resistance are limited, and cable sheaths may be damaged by aging, cracking, mechanical damage, and other conditions.

[0003] In the related art, the detection method of bridge cable sheath defects usually relies on manual visual inspection. Some large or extra-large bridges also need to use a hanging basket or cable lifting device to send maintenance personnel, tools or instruments to any position on the inclined cable to complete the inspection. The inspection efficiency is low, a lot of manpower and material resources are consumed, and there are great safety hazards. Moreover, compared with other types of defects such as deformation and wear, the aging and other defects unique to the cable sheath are more difficult to distinguish, and the recognition accuracy of manual visual inspection is low. Summary of the invention

[0004] The present application aims to solve the technical problem that the related art bridge cable sheath disease detection method relies on manual visual inspection, resulting in low detection efficiency and low identification accuracy.

[0005] The present application provides a method for detecting bridge cable sheath defects, comprising the following steps:

[0006] Collect images of the cable sheath of the bridge to be inspected;

[0007] The image of the bridge cable sheath to be inspected is identified by using a pre-trained semantic segmentation model to obtain the type of cable sheath damage and a grayscale image of the separately stripped damage area;

[0008] Selecting a designated disease analysis model based on the cable sheath disease type;

[0009] The grayscale image of the damaged area is analyzed by using a pre-trained damage analysis model to obtain the damage degree of the cable sheath.

[0010] In one embodiment, the pre-training process of the semantic segmentation model includes:

[0011] The types and degrees of damage in the bridge cable sheath image set are semantically annotated, and the annotated bridge cable sheath image set is divided into a training set and a validation set;

[0012] A semantic segmentation model is constructed and trained based on the training set and the validation set to obtain the types of cable sheath damage and grayscale images of the separately stripped damage areas.

[0013] In one embodiment, the semantic segmentation model uses Deeplab-v3+ including an encoder module and a decoder module as a basic model.

[0014] In one embodiment, a first pixel cross entropy loss function is used in the training process, and the first pixel cross entropy loss function is expressed as:

[0015]

[0016] Among them, c represents the number of categories; y i represents the image encoding vector, p i Represents the probability that the predicted sample belongs to i.

[0017] In one embodiment, the pre-training process of the disease analysis model includes:

[0018] Build a convolutional neural network model;

[0019] Analyze each pixel of the grayscale image of the diseased area, calculate the weight corresponding to the grayscale value, and obtain a weight matrix describing the grayscale;

[0020] Determining the degree of damage to the cable sheath based on the weight matrix;

[0021] The above process is repeated to traverse the grayscale images of the diseased areas corresponding to different types of cable sheath diseases, and the disease analysis models corresponding to different types of cable sheath diseases are obtained through training.

[0022] In one embodiment, the disease types correspond one-to-one to the disease analysis models.

[0023] In one embodiment, a second pixel cross entropy loss function is used in the training process, and the second pixel cross entropy loss function is expressed as:

[0024]

[0025] Among them, s is the true value of the disease degree, width and height represent the width and height of the image respectively, and p ij represents the gray value of the pixel in the i-th row and j-th column, w pre (p ij ) represents the weight corresponding to the predicted gray value.

[0026] In one embodiment, the bridge cable sheath disease detection method further includes: determining a maintenance plan according to the degree of the cable sheath disease.

[0027] The embodiment of the present application further provides a bridge cable sheath defect detection system, which uses any of the above-mentioned bridge cable sheath defect detection methods, and the bridge cable sheath defect detection system includes:

[0028] The image acquisition module is configured to: acquire images of the cable sheaths of the bridge to be inspected;

[0029] The damage type recognition module is configured to: recognize the image of the bridge cable sheath to be detected by using a pre-trained semantic segmentation model to obtain the damage type of the cable sheath and a grayscale image of the damaged area that is separately stripped off;

[0030] The damage degree analysis module is configured to: select a designated damage analysis model based on the damage type of the cable sheath; and analyze the grayscale image of the damage area using the pre-trained damage analysis model to obtain the damage degree of the cable sheath.

[0031] In one embodiment, the image acquisition module is integrated with a local storage module, and the damage type identification module and the damage degree analysis module are located in a cloud server. The image of the bridge cable sheath to be inspected acquired by the image acquisition module is stored in the local storage module. After the image of the bridge cable sheath to be inspected reaches a set amount, it is transmitted to the cloud server for identification by the damage type identification module and analysis by the damage degree analysis module.

[0032] The beneficial effects brought by the technical solution provided in the embodiments of the present application include:

[0033] The present application provides a bridge cable sheath disease detection method. For the bridge cable sheath image to be detected, a semantic segmentation model is first used for preliminary identification to confirm the disease type and separately extract the grayscale image of the diseased area, and then the corresponding specified disease analysis model is used to analyze the degree of disease on the grayscale image of the diseased area. The disease detection of the cable sheath by image recognition avoids the problem of low detection efficiency caused by relying on manual or engineering machinery to identify the disease type and degree in the related technology, and improves the recognition accuracy; more specifically, combined with the characteristics of the disease conditions that are prone to occur in the cable sheath, the image disease type recognition and disease degree analysis are divided into two stages, which is more targeted and has higher recognition accuracy. It also improves the scalability of recognition and robustness to severe weather, has strong adaptability, and can be extended to other fields. BRIEF DESCRIPTION OF THE DRAWINGS

[0034] In order to more clearly illustrate the technical solutions in the embodiments of the present application, the drawings required for use in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0035] Figure 1 The present invention is a flow chart of a bridge cable sheath disease detection method according to an embodiment of the present invention.

[0036] Figure 2 Schematic diagram of the installation position of the image acquisition module in one embodiment of the present invention.

[0037] Figure 3 4 is a network structure diagram of a semantic segmentation model in one embodiment of the present invention.

[0038] Figure 4 4 is a network structure diagram of a disease analysis model in one embodiment of the present invention.

[0039] Figure 5 The flowchart of analyzing the degree of disease in one embodiment of the present invention.

[0040] Figure 6 Schematic diagram of different maintenance solutions in one embodiment of the present invention.

[0041] Figure 7 The figure is a workflow diagram of each module in one embodiment of the present invention.

[0042] In the figure: 1. Cable sheath; 2. Camera. DETAILED DESCRIPTION

[0043] In order to enable those skilled in the art to better understand the solution of the present application, the technical solution in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only part of the embodiments of the present application, not all of the embodiments. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of this application.

[0044] like Figure 1 As shown, Figure 1 The present invention is a flow chart of a bridge cable sheath disease detection method according to an embodiment of the present invention.

[0045] This embodiment provides a bridge cable sheath disease detection method, comprising the following steps:

[0046] Step S1, collecting images of the cable sheaths of the bridge to be inspected;

[0047] Step S2, using the pre-trained semantic segmentation model to identify the image of the bridge cable sheath to be inspected, so as to obtain the type of cable sheath disease and the grayscale image of the diseased area that is separately stripped off;

[0048] Step S3, selecting a designated disease analysis model based on the cable sheath disease type;

[0049] Step S4: Analyze the grayscale image of the damaged area using the pre-trained damage analysis model to obtain the damage degree of the cable sheath.

[0050] The present embodiment provides a bridge cable sheath disease detection method. For the bridge cable sheath image to be detected, a semantic segmentation model is first used for preliminary identification to confirm the disease type and separately extract the grayscale image of the diseased area, and then the corresponding specified disease analysis model is used to analyze the degree of disease on the grayscale image of the diseased area. The disease detection of the cable sheath by image recognition avoids the problem of low detection efficiency caused by relying on manual or engineering machinery to identify the disease type and degree in the related technology, and improves the recognition accuracy. More specifically, combined with the characteristics of the disease conditions that are prone to occur in the cable sheath, the image disease type recognition and disease degree analysis are divided into two stages, which is more targeted and has higher recognition accuracy. It also improves the scalability of recognition and the robustness to severe weather. It has strong adaptability and can be extended to other fields.

[0051] Each step is described and explained in detail below.

[0052] like Figure 2 As shown, Figure 2 Schematic diagram of the installation position of the image acquisition module in one embodiment of the present invention.

[0053] Take a cable-stayed bridge as an example. The cable-stayed bridge has two bridge columns on one side. Cables are symmetrically installed on each bridge column. The outer layer of the cable is wrapped with a cable sheath 1, which is made of PE (polyethylene) material. A spherical camera 2 is installed about 3m on the bridge deck in the bridge column area of ​​the cable-stayed bridge as an image acquisition module. There are four cameras on the bridge deck on both sides. The rotation and zoom of the camera 2 are used to capture images of cables at multiple positions, so as to cover the entire bridge cable area and capture the bridge cable sheath in all directions.

[0054] The actual size of the shooting area of ​​camera 2 is pre-set, for example, the camera resolution is 2560px*1440px; for example, the number of cables in the image of the bridge cable sheath to be detected is not less than 1 and not more than two, so as to ensure the accuracy of the recognition result, comprehensively consider the portability of the current bridge equipment installation, and the characteristics of image recognition technology, which is closer to the actual application scenario and more reasonable.

[0055] In one embodiment, in step S2, the pre-training process of the semantic segmentation model includes:

[0056] Step S21, semantically annotating the types and degrees of damage in the bridge cable sheath image set, and dividing the annotated bridge cable sheath image set into a training set and a verification set.

[0057] Step S22: construct a semantic segmentation model, and train the semantic segmentation model based on the training set and the validation set to obtain the cable sheath disease type and the grayscale image of the diseased area that is separately stripped out.

[0058] Specifically, in step S21, 600 existing bridge cable sheath pictures are collected as a bridge cable sheath image set, and the bridge cable sheath image set is semantically annotated using labelme software. The annotation information is the type of disease and the degree of disease. The disease types include aging, cracking, and mechanical damage. The disease types refer to the standards, as shown in Tables 1 to 3.

[0059] Table 1 HDPE sheath aging assessment

[0060]

[0061]

[0062] Table 2 HDPE sheath cracking assessment

[0063] Scale Qualitative description 1 intact 2 HDPE sheath has micro cracks, but the cracks do not extend to the steel wire 3 The HDPE sheath has obvious cracks, but the cracks do not extend to the steel wire 4 The HDPE sheath has serious cracks, and the cracks extend to the steel wire

[0064] Table 3 Mechanical damage assessment of HDPE sheath

[0065]

[0066] The annotated bridge cable sheath image set is divided into a 4:1 ratio, with 480 images as training sets and 120 images as validation sets.

[0067] In one embodiment, in step S22, the semantic segmentation model uses Deeplab-v3+ including an encoder module and a decoder module as a basic model.

[0068] In one embodiment, a first pixel cross entropy loss function is used in the training process, and the first pixel cross entropy loss function is expressed as:

[0069]

[0070] Among them, c represents the number of categories; y i represents the image encoding vector, which takes 1 if the category is the same as the labeled category, otherwise takes 0; p i Represents the probability that the predicted sample belongs to i.

[0071] Specifically, Figure 3 As shown in the figure, the semantic segmentation model is an encoder-decoder structure. Encoder part: After the image enters the backbone network (DCNN), two feature layers are obtained. The shallow feature layer directly enters the decoder for channel compression (through 1×1 convolution) to reduce the proportion of the low-level layer; the deep feature enters the encoder's atrous spatial pyramid pooling (ASPP) module. The ASPP module includes 1 1×1 convolution, 3 3×3 atrous convolutions of different ratios (6, 12, 18), and a global pooling operation of the image. These operations are followed by a 1×1 convolution. The features obtained by the encoder have richer information and higher proportion. Decoder part: The low-level features extracted from the encoder are spliced ​​with the sampled high-level semantic feature map. This operation ensures that high-level semantic information and low-level spatial information are used at the same time, which not only retains the global context but also restores the boundary details.

[0072] Through the above scheme, the semantic segmentation model uses Deeplab-v3+ as the basic model and is trained. Only one semantic segmentation model needs to be trained. The design of this structure enables it to perform well in processing complex scenes and obtaining rich contextual information. Compared with other network structures, DeepLab v3+ is more able to balance accuracy and computational efficiency in disease recognition tasks, especially when processing images with complex backgrounds and subtle diseases. The training model uses the first pixel cross entropy loss function in the training process, which can provide accurate and stable learning objectives and has strong adaptability.

[0073] Since different types of diseases often have different characteristics, it is necessary to determine the type of disease before analyzing the degree of the disease, so that the corresponding disease analysis model can be used for subsequent analysis. Through the above scheme, the image of the bridge cable sheath to be detected is recognized based on the pre-trained semantic segmentation model, which can accurately determine the specific type of disease and provide key information for the subsequent analysis of the degree of disease; if the type of disease does not exist in the image of the bridge cable sheath to be detected, the detection process ends. If a certain type of disease exists, the diseased area is separated out separately, and a grayscale image of the diseased area is provided to facilitate the subsequent disease analysis model to analyze the degree of disease, and it can also reduce image noise and improve analysis efficiency. By solving complex tasks step by step, an efficient combination of detection and analysis is achieved.

[0074] In one embodiment, there is a one-to-one correspondence between the disease types and the disease analysis models.

[0075] Through the above scheme, since there are many types of cable sheath disease, a designated disease analysis model is trained accordingly for each type of cable sheath disease, which is more targeted and has higher recognition accuracy.

[0076] In one embodiment, in step S4, the pre-training process of the disease analysis model includes:

[0077] Step S41, constructing a convolutional neural network model;

[0078] Step S42, analyzing each pixel of the grayscale image of the diseased area, calculating the weight corresponding to its grayscale value, to obtain a weight matrix describing the grayscale;

[0079] Step S43, determining the degree of damage of the cable sheath based on the weight matrix;

[0080] The above process is repeated to traverse the grayscale images of the diseased areas corresponding to different types of cable sheath diseases, and the disease analysis models corresponding to different types of cable sheath diseases are obtained through training.

[0081] In one embodiment, a second pixel cross entropy loss function is used in the training process, and the second pixel cross entropy loss function is expressed as:

[0082]

[0083] Among them, s is the true value of the disease degree, width and height represent the width and height of the image respectively, and p ij represents the gray value of the pixel in the i-th row and j-th column, w pre (p ij ) represents the weight corresponding to the predicted gray value.

[0084] The training model uses the second pixel cross entropy loss function to act on the training process, which can provide accurate and stable learning objectives and has strong adaptability. A special loss function is designed for each disease type to optimize pixel weights. This targeted optimization can better focus on the key feature areas of the disease, thereby improving the accuracy of detection and analysis.

[0085] Specifically, in step S2, the damage degree of the existing bridge cable sheath image set has been semantically annotated, and a training set and a verification set have been provided. Figure 4 As shown in the figure, the disease analysis model uses a convolutional neural network, which is a deep learning method. Its advantage is that it can automatically extract image features at different levels and has strong adaptability. Traditional SVM and GBDT methods need to be combined with manual feature extraction technology, which is not as effective and adaptable as convolutional neural networks. The new Transformer network has a feature extraction capability that is not inferior to convolutional neural networks, but the amount of calculation is relatively large. Therefore, on the whole, convolutional neural networks are a choice that balances efficiency and accuracy.

[0086] A learning strategy is constructed, with the input being the grayscale image of the damaged area processed by the semantic segmentation model. After analyzing the damage degree, the output is a weight matrix describing various grayscales, and an analysis report is generated to guide the maintenance of the cable sheath.

[0087] like Figure 5 As shown, when the grayscale image of the diseased area is analyzed by the disease analysis model, the process is: fitting a function w, the function description is: f = w (p);

[0088] Among them, p represents the gray value, f represents the weight of the gray value p in the current disease type, and the disease degree is described as:

[0089]

[0090] Among them, s is the degree of damage, width and height represent the width and height of the image respectively, p ij Represents the grayscale value of the pixel in the i-th row and j-th column.

[0091] Based on the above function, each pixel of the grayscale image of the diseased area is analyzed, and the weight corresponding to its grayscale value is calculated. The function output is a floating point number not less than 0, which represents the disease weight of this grayscale value under the disease type. The weights are accumulated to obtain a weight matrix describing the grayscale; the degree of cable sheath disease is determined based on the weight matrix; the above process is repeated to traverse the grayscale images of the diseased area corresponding to different cable sheath disease types, and the disease analysis models corresponding to different cable sheath disease types are trained.

[0092] Through the above scheme, a convolutional neural network is used to train a disease analysis model for each disease type to deeply analyze the extent and impact of each disease, ensuring accurate analysis and evaluation of the disease extent and improving the credibility and efficiency of detection.

[0093] In one embodiment, the bridge cable sheath disease detection method further includes: step S5, determining a maintenance plan according to the degree of the cable sheath disease.

[0094] Specifically, Figure 6 As shown, targeted maintenance plans are provided according to the degree of damage to the cable sheath, such as whether it needs repair or replacement. The damage degree of the cable sheath is divided into 4 levels, namely no damage, mild damage, moderate damage and severe damage, and there are different maintenance plans for each level. The maintenance plans and reasons for different damage degrees are as follows:

[0095] No damage: At this time, the cable sheath damage degree s value is 0, indicating that the cable sheath has not received any damage and does not require any treatment. This condition generally occurs in the cable area that has just completed maintenance.

[0096] Mild damage: At this time, the damage degree s of the cable sheath is between 0 and 1. The cable sheath has suffered a certain degree of damage, but the degree of damage is relatively light and within the tolerable range of the cable sheath. It will not threaten the health of the cables inside the cable sheath. Therefore, it is judged that the cable sheath is in good health and does not need to be treated.

[0097] Moderate damage: At this time, the damage level s of the cable sheath is between 1 and 2. The cable sheath has been seriously damaged, but it will not threaten the health of the cable inside the cable sheath for the time being. However, it has exceeded the tolerable range of the cable sheath. Therefore, it is determined that the health of the cable sheath has been affected to a certain extent. In order to prevent the moderate damage from deteriorating to severe damage, the cable sheath in the diseased area needs to be maintained or replaced in time. This step is crucial to maintaining and extending the service life of the cable and its sheath.

[0098] Severe damage: At this time, the cable sheath damage degree s value is greater than 2, indicating that the cable sheath material has been severely damaged, and there is a high probability that the cables inside the cable sheath in this area have also been damaged. Therefore, it is determined that the health of the cable sheath has been severely affected, which may cause abnormal health of the bridge as a whole. Therefore, it is necessary not only to replace the cable sheath area in time, but also to further perform health inspections inside the area. This step is crucial to maintaining the health of the bridge.

[0099] Through the above scheme, different maintenance plans are provided according to different degrees of damage to ensure the long-term reliability and performance of equipment or structures, help to effectively manage and maintain assets, and minimize potential risks and costs.

[0100] The embodiment of the present application also provides a bridge cable sheath disease detection system, which uses the above-mentioned bridge cable sheath disease detection method. The bridge cable sheath disease detection system includes:

[0101] The image acquisition module is configured to: acquire images of the cable sheaths of the bridge to be inspected;

[0102] The damage type recognition module is configured to: recognize the image of the bridge cable sheath to be inspected by using the pre-trained semantic segmentation model to obtain the damage type of the cable sheath and the grayscale image of the separately stripped damage area;

[0103] The damage degree analysis module is configured to: select a specified damage analysis model based on the damage type of the cable sheath; analyze the grayscale image of the damage area through the pre-trained damage analysis model to obtain the damage degree of the cable sheath.

[0104] The functions of the image acquisition, disease type recognition and disease degree analysis modules have been described in the above method and will not be repeated here.

[0105] In one embodiment, the image acquisition module is integrated with a local storage module, the damage type recognition module and the damage degree analysis module are located in a cloud server, and the images of the cable sheath of the bridge to be inspected acquired by the image acquisition module are stored in the local storage module. After the images of the cable sheath of the bridge to be inspected reach a set amount, they are transmitted to the cloud server for identification by the damage type recognition module and analysis by the damage degree analysis module.

[0106] Specifically, there is a transmission network: used to transmit instructions and data between different devices, which is divided into wired transmission network and wireless transmission network. Figure 7 As shown, the image acquisition module is integrated with a local storage module. The images of the bridge cable sheath to be detected collected by the image acquisition module are transmitted to the local storage module through a wired transmission network and stored. When the images of the bridge cable sheath to be detected reach a set amount (such as the storage capacity of the local storage module is full), they are transmitted to the cloud server through a wireless transmission network. The cloud server serves as a cloud storage module, and the disease type identification module and the disease degree analysis module serve as data analysis modules. The program runs on the cloud server. After the analysis is completed, the analysis results are saved to the cloud server through a wireless transmission network.

[0107] In the description of the present application, it should be noted that the terms "upper", "lower", etc. indicate orientations or positional relationships based on the orientations or positional relationships shown in the accompanying drawings, and are only for the convenience of describing the present application and simplifying the description, rather than indicating or implying that the device or element referred to must have a specific orientation, be constructed and operated in a specific orientation, and therefore cannot be understood as a limitation on the present application. Unless otherwise clearly specified and limited, the terms "installed", "connected", and "connected" should be understood in a broad sense, for example, it can be a fixed connection, a detachable connection, or an integral connection; it can be a mechanical connection or an electrical connection; it can be a direct connection, or it can be an indirect connection through an intermediate medium, or it can be a connection between the two elements. For those of ordinary skill in the art, the specific meanings of the above terms in the present application can be understood according to the specific circumstances.

[0108] It should be noted that, in this application, relational terms such as "first" and "second" are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, the elements defined by the sentence "comprise a ..." do not exclude the presence of other identical elements in the process, method, article or device including the elements.

[0109] In the description of the embodiments of the present application, unless otherwise specified, “ / ” means or, for example, A / B can mean A or B; the “and / or” in the text is merely a description of the association relationship of associated objects, indicating that three relationships may exist, for example, A and / or B can mean: A exists alone, A and B exist at the same time, and B exists alone. In addition, in the description of the embodiments of the present application, “multiple” refers to two or more than two.

[0110] The above is only a specific implementation of the present application, so that those skilled in the art can understand or implement the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest range consistent with the principles and novel features applied for herein.

Claims

1. A bridge cable sheath disease detection method, characterized in that: The following steps are involved: Collect images of the cable sheath of the bridge to be inspected; The image of the bridge cable sheath to be inspected is identified by using a pre-trained semantic segmentation model to obtain the type of cable sheath damage and a grayscale image of the separately stripped damage area; Selecting a designated disease analysis model based on the cable sheath disease type; The grayscale image of the damaged area is analyzed by a pre-trained damage analysis model to obtain the damage degree of the cable sheath.

2. The bridge cable sheath disease detection method according to claim 1, characterized in that: The pre-training process of the semantic segmentation model includes: The types and degrees of damage in the bridge cable sheath image set are semantically annotated, and the annotated bridge cable sheath image set is divided into a training set and a validation set; A semantic segmentation model is constructed and trained based on the training set and the validation set to obtain the types of cable sheath damage and grayscale images of the separately stripped damage areas.

3. The bridge cable sheath disease detection method according to claim 2, characterized in that: The semantic segmentation model uses Deeplab-v3+ including an encoder module and a decoder module as a basic model.

4. The bridge cable sheath disease detection method according to claim 3, characterized in that: The first pixel cross entropy loss function is used in the training process. The first pixel cross entropy loss function is expressed as: Among them, c represents the number of categories; y i represents the image encoding vector, p i Represents the probability that the predicted sample belongs to i.

5. The bridge cable sheath disease detection method according to claim 2, characterized in that: The pre-training process of the disease analysis model includes: Build a convolutional neural network model; Analyze each pixel of the grayscale image of the diseased area, calculate the weight corresponding to the grayscale value, and obtain a weight matrix describing the grayscale; Determining the degree of damage to the cable sheath based on the weight matrix; The above process is repeated to traverse the grayscale images of the diseased areas corresponding to different types of cable sheath diseases, and the disease analysis models corresponding to different types of cable sheath diseases are obtained through training.

6. The bridge cable sheath disease detection method according to claim 5, characterized in that: The disease types correspond one to one with the disease analysis models.

7. The bridge cable sheath disease detection method according to claim 5, characterized in that: The second pixel cross entropy loss function is used in the training process, and the second pixel cross entropy loss function is expressed as: Among them, s is the true value of the disease degree, width and height represent the width and height of the image respectively, and p ij represents the gray value of the pixel in the i-th row and j-th column, w pre (p ij ) represents the weight corresponding to the predicted gray value.

8. The bridge cable sheath disease detection method according to claim 1, characterized in that: The bridge cable sheath disease detection method further includes: determining a maintenance plan according to the degree of disease of the cable sheath.

9. A bridge cable sheath defect detection system, using the bridge cable sheath defect detection method according to any one of claims 1 to 8, characterized in that: The bridge cable sheath disease detection system comprises: The image acquisition module is configured to: acquire images of the cable sheaths of the bridge to be inspected; The damage type recognition module is configured to: recognize the image of the bridge cable sheath to be detected by using a pre-trained semantic segmentation model to obtain the damage type of the cable sheath and a grayscale image of the damaged area that is separately stripped off; The damage degree analysis module is configured to: select a designated damage analysis model based on the damage type of the cable sheath; and analyze the grayscale image of the damage area using the pre-trained damage analysis model to obtain the damage degree of the cable sheath.

10. The bridge cable sheath defect detection system according to claim 9, characterized in that: The image acquisition module is integrated with a local storage module, and the damage type recognition module and the damage degree analysis module are located in the cloud server. The image of the bridge cable sheath to be detected acquired by the image acquisition module is stored in the local storage module. After the image of the bridge cable sheath to be detected reaches a set amount, it is transmitted to the cloud server for identification by the damage type recognition module and analysis by the damage degree analysis module.

Images