Railway concrete structure deterioration identification method based on semi-supervised learning

Abstract

The invention discloses a railway concrete structure deterioration identification method based on semi-supervised learning. The method comprises the following steps: establishing a railway concrete structure deterioration identification model based on semi-supervised learning; and carrying out deterioration identification on the railway concrete structure based on the railway concrete structure deterioration identification model. The invention further discloses a corresponding system, electronic equipment and a computer readable storage medium, the pressure of data annotation is relieved, a semi-supervised learning method is applied to railway concrete structure deterioration detection, semi-supervised learning semantic segmentation is adopted, only part of image data needs to be marked, and the detection efficiency is improved. An optimal segmentation model is learned from the marked and unmarked mixed samples, so that the pixel level marking work is greatly reduced; according to the method, massive degradation data resources of a tunnel scene are fully utilized, an application method of a semi-supervised semantic segmentation algorithm in sample training is determined, a recognition model is established by adopting a transfer learning technology, and automatic identification of apparent degradation of concrete structures such as bridges and ballastless tracks is realized.

Description

Technical Field

[0001] The invention relates to the technical field of railway detection and intelligent recognition of apparent defects of railway concrete structures, and in particular to a method for identifying railway concrete structure deterioration based on semi-supervised learning. Background Art

[0002] In recent years, traffic safety accidents caused by apparent deterioration of concrete structures in railway scenarios have occurred frequently. Existing detection methods can detect cracks and water leakage in tunnel linings, cracks or damage in ballastless tracks, cracks, broken blocks, or honeycomb surfaces in bridges. The apparent deterioration types of concrete structures in different scenarios are similar, and tracking their development trends can help to prevent various unexpected risks in a timely manner and ensure the safety of railway operations.

[0003] With the development of machine vision technology, automated detection methods have gradually replaced manual inspection methods. With the introduction of fully convolutional networks, image semantic segmentation has achieved a significant improvement in recognition accuracy. As one of the basic tasks of machine vision, semantic segmentation has been applied to deterioration detection in the railway field. For example, the intelligent detection system for apparent defects in tunnel linings developed by the China Railway Construction Institute can be installed on a rail car to quickly collect apparent images of tunnel linings and use artificial intelligence technology to automatically identify cracks, falling blocks and water leaks, significantly reducing the intensity of manual work and the time occupied by skylights, and improving detection efficiency. However, although the image semantic segmentation method based on convolutional neural networks has achieved amazing performance, the following problems are still found in actual applications:

[0004] (1) Model training requires a large amount of pixel-level annotated data, but obtaining sufficient, high-quality pixel-level annotated data is time-consuming and laborious, requiring a lot of manpower and time.

[0005] (2) There are relatively few high-quality images of concrete surface deterioration, which cannot meet the sample requirements of deep learning.

[0006] Therefore, there is a technical demand to explore new structural deterioration identification methods and corresponding systems to solve the current problem of insufficient samples caused by the large workload of training and annotation and the relatively small number of high-quality disease photos of concrete surface deterioration. Summary of the invention

[0007] The purpose of the present invention is to provide a method for identifying railway concrete structure deterioration based on semi-supervised learning to solve the current problems of insufficient samples caused by large workload of training and annotation and relatively few high-quality disease photos of concrete apparent deterioration.

[0008] On the one hand, in order to alleviate the pressure of data labeling, the semi-supervised learning method is applied to the task of railway concrete structure deterioration detection. Semi-supervised learning semantic segmentation is used, which only requires labeling part of the image data, and then learns the optimal segmentation model from the mixed samples of labeled and unlabeled samples, thereby greatly reducing the pixel-level labeling work.

[0009] Secondly, in the context of apparent deterioration detection of railway concrete structures, in order to address the problems of large workload in training and annotation and insufficient samples, especially the relatively small number of high-quality disease photos of apparent deterioration of concrete, the idea of ​​transfer learning is used to fully utilize the massive deterioration data resources of tunnel scenes, and the application method of semi-supervised semantic segmentation algorithm in sample training is determined. According to the similar apparent deterioration characteristics of concrete structures such as tunnels, bridges, and ballastless tracks, the massive data obtained by the tunnel lining apparent detection system is fully utilized, and the transfer learning technology is used to establish a recognition model to realize the automatic identification of apparent deterioration of concrete structures such as bridges and ballastless tracks, effectively reducing the training data annotation work, while ensuring the identification of degradation with similar features in other scenarios.

[0010] A first aspect of the present invention is to provide a method for identifying railway concrete structure degradation based on semi-supervised learning, comprising:

[0011] S1, establish a railway concrete structure degradation identification model based on semi-supervised learning;

[0012] S2, performing degradation identification of railway concrete structures based on the railway concrete structure degradation identification model.

[0013] Preferably, the S1 comprises:

[0014] S11, determining a basic model for establishing the railway concrete structure degradation identification model; the basic model is a U-Net network architecture, which is composed of a symmetrical encoder and decoder, the encoder is composed of a series of convolutional layers and pooling layers, and is used to extract features from an input image; the decoder is composed of a series of deconvolutional layers and upsampling layers, and is used to reconstruct the features extracted by the encoder into a segmentation result of the same size as the original input image; there is also a jump connection between the encoder and the decoder, which is used to directly connect the high-level feature map in the encoder to the corresponding layer in the decoder, and use deconvolution and upsampling to avoid information loss and resolution reduction;

[0015] S12, establishing a primary railway concrete structure degradation identification model based on the basic model;

[0016] S13, evaluating the primary railway concrete structure degradation identification model based on the evaluation index to obtain the railway concrete structure degradation identification model.

[0017] Preferably, the S12 includes:

[0018] (1) performing data preprocessing and partitioning to determine a sample set for training the railway concrete structure degradation identification model; wherein the sample set comprises: dividing each data set into two parts, namely a labeled data set and an unlabeled data set, dividing the labeled data set into two parts, namely a first labeled data set L1 and a second labeled data set L2, the first labeled data set L1 and the second labeled data set L2 are independently used for the first training model F1 and the second training model F2 to obtain a first prediction result and a second prediction result; at the same time, using the unlabeled data set as the input of the first training model F1 and the second training model F2 to obtain a third prediction result and a fourth prediction result; cross-adding the third prediction result and the fourth prediction result to the first labeled data set L1 and the second labeled data set L2 as a sample pair for the next round of model training of the first training model F1 and the second training model F2; the data set is a massive amount of data obtained by using a tunnel lining surface detection system;

[0019] (2) determining a loss function of the model, wherein the loss function of the model includes two parts, namely, a cross pseudo-supervised loss function and an unsupervised loss function; the cross pseudo-supervised loss function uses a pseudo-segmentation map of one network as a supervisory signal of another network; the unsupervised loss function uses a consistency loss, i.e., calculating the consistency between the third prediction result and the fourth prediction result of the unlabeled data set;

[0020] (3) initializing the basic model, including: initializing the backbones of two segmentation networks of the U-Net network architecture using weights pre-trained on ImageNet, and randomly initializing the segmentation heads;

[0021] (4) Training the initialized model based on the loss function and training strategy of the model to obtain a primary railway concrete structure degradation identification model, wherein the training strategy includes using a standard pixel-level cross entropy loss to supervise the annotated data and the enhanced data.

[0022] Preferably, the training is performed using an optimizer, the optimizer is an Adam optimizer, the initial learning rate is set to 0.001, the momentum is 0.9, the weight decay is 0.0005, and the learning rate is adjusted according to a polynomial decay strategy.

[0023] Preferably, the enhanced data includes: enhancing the standard pixel-level cross entropy loss supervised annotation data by random horizontal flipping and multi-scale training.

[0024] Preferably, the evaluation indicator is model accuracy.

[0025] Preferably, S2 includes:

[0026] S21, acquiring multiple surface images of the railway concrete structure as degradation identification samples;

[0027] S22, inputting the degradation identification sample into the railway concrete structure degradation identification model for degradation identification.

[0028] The second aspect of the present invention is to provide a railway concrete structure deterioration identification system based on semi-supervised learning, which is used to implement the method of the first aspect, comprising:

[0029] A model building module (101) is used to build a railway concrete structure degradation identification model based on semi-supervised learning;

[0030] A degradation identification module (102) is used to identify the degradation of the railway concrete structure based on the railway concrete structure degradation identification model.

[0031] A third aspect of the present invention provides an electronic device, comprising a processor and a memory, wherein the memory stores a plurality of instructions, and the processor is configured to read the instructions and execute the method described in the first aspect.

[0032] A fourth aspect of the present invention provides a computer-readable storage medium, wherein the computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read by a processor and execute the method described in the first aspect.

[0033] Beneficial effects of the method and system of the present invention:

[0034] (1) In order to alleviate the pressure of data labeling, the semi-supervised learning method is applied to the railway concrete structure deterioration detection task. Semi-supervised learning semantic segmentation is adopted. Only part of the image data needs to be labeled, and then the optimal segmentation model is learned from the mixed samples of labeled and unlabeled samples, which greatly reduces the pixel-level labeling work.

[0035] (2) In the context of apparent deterioration detection of railway concrete structures, in order to address the problems of large workload of training and annotation and insufficient samples, especially the relatively small number of high-quality disease photos of apparent deterioration of concrete, the idea of ​​transfer learning is used to fully utilize the massive deterioration data resources of tunnel scenes, determine the application method of semi-supervised semantic segmentation algorithm in sample training, and according to the similar deterioration characteristics of concrete structures such as tunnels, bridges, and ballastless tracks, make full use of the massive data obtained by the tunnel lining surface detection system, and use transfer learning technology to establish a recognition model to realize the automatic identification of apparent deterioration of concrete structures such as bridges and ballastless tracks, effectively reducing the workload of training data annotation, while ensuring the identification of degradation with similar features in other scenes. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the related technologies, the drawings required for use in the specific embodiments or the related technical descriptions will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0037] Figure 1 A schematic diagram of the principle of semi-supervised learning based on the pseudo-label method according to the prior art;

[0038] Figure 2 A schematic diagram of a semi-supervised learning principle based on a generative adversarial network according to the prior art;

[0039] Figure 3 A schematic diagram of a semi-supervised learning principle based on consistency learning according to the prior art;

[0040] Figure 4 A schematic flow chart of a method for identifying railway concrete structure deterioration based on semi-supervised learning according to an embodiment of the present invention;

[0041] Figure 5 A schematic flow chart of step S1 of a method for identifying railway concrete structure deterioration based on semi-supervised learning according to an embodiment of the present invention;

[0042] Figure 6 A schematic diagram of a U-Net network structure provided according to an embodiment of the present invention;

[0043] Figure 7 A schematic flow chart of step S12 of a method for identifying railway concrete structure deterioration based on semi-supervised learning according to an embodiment of the present invention;

[0044] Figure 8A schematic flow chart of step S2 of a method for identifying railway concrete structure deterioration based on semi-supervised learning according to an embodiment of the present invention;

[0045] Fig. 9 This is a diagram of the architecture of a railway concrete structure degradation identification system based on semi-supervised learning according to an embodiment of the present invention; Fig.10 A structural diagram of an electronic device provided according to an embodiment of the present invention. DETAILED DESCRIPTION

[0046] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] In the description of the present invention, it should be noted that the terms "center", "upper", "lower", "left", "right", "vertical", "horizontal", "inner", "outer", etc., indicating the orientation or positional relationship, are based on the orientation or positional relationship shown in the drawings, and are only for the convenience of describing the present invention 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 limiting the present invention. In addition, the terms "first", "second", and "third" are used for descriptive purposes only, and cannot be understood as indicating or implying relative importance.

[0048] In the description of the present invention, it should be noted that, 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 indirectly connected through an intermediate medium, or it can be the internal communication of two components. For ordinary technicians in this field, the specific meanings of the above terms in the present invention can be understood according to specific circumstances.

[0049] On the one hand, in order to alleviate the pressure of data labeling, the semi-supervised learning method is applied to the railway concrete structure deterioration detection task, and semi-supervised learning semantic segmentation is adopted. It only needs to label part of the image data, and then learn the optimal segmentation model from the mixed samples of labeled and unlabeled, thereby greatly reducing the pixel-level labeling work.

[0050] Secondly, in the context of apparent deterioration detection of railway concrete structures, in order to address the problems of large workload in training and annotation and insufficient samples, especially the relatively small number of high-quality disease photos of apparent deterioration of concrete, the idea of ​​transfer learning is used to fully utilize the massive deterioration data resources of tunnel scenes, and the application method of semi-supervised semantic segmentation algorithm in sample training is determined. According to the similar apparent deterioration characteristics of concrete structures such as tunnels, bridges, and ballastless tracks, the massive data obtained by the tunnel lining apparent detection system is fully utilized, and the transfer learning technology is used to establish a recognition model to realize the automatic identification of apparent deterioration of concrete structures such as bridges and ballastless tracks, effectively reducing the training data annotation work, while ensuring the identification of degradation with similar features in other scenarios.

[0051] Semi-supervised learning mainly includes the following four methods:

[0052] (1) Pseudo-labeling method

[0053] like Figure 1 As shown in the figure, in semi-supervised learning, the pseudo-label method is a strategy that uses a model to generate pseudo-labels for unlabeled data. This method generates predicted labels as pseudo-labels by training a model on labeled data and then applying the model to unlabeled data. The core idea of ​​this method is to assume that the model's learning of labeled data can be generalized to unlabeled data, thereby expanding the model's training data set. First, the model is preliminarily trained using labeled data. Then, the model is applied to unlabeled data to obtain pseudo-labels. These pseudo-labels are added to the original training data to form a larger training set. Next, the model is retrained using the expanded data set. This iterative process can be repeated many times to gradually improve the model performance.

[0054] The advantage of the pseudo-label method is that it effectively utilizes unlabeled data and improves the generalization ability of the model. However, it also faces the challenge that the pseudo-label may contain noise, because the model's prediction on unlabeled data is not absolutely accurate. Therefore, when applying the pseudo-label method, the quality of the pseudo-label needs to be handled with caution, and the impact of noise can be reduced by threshold screening or other techniques. Overall, the pseudo-label method is a simple and effective strategy in semi-supervised learning for making full use of unlabeled data to improve model performance.

[0055] (2) Generative Adversarial Networks

[0056] like Figure 2 As shown in Figure 1, the Generative Adversarial Network (GAN) consists of two networks, the Generator G and the Discriminator F. The training process of the network is carried out in an adversarial manner. The Generator learns the distribution of the target data, generates synthetic images from random noise, and generates images similar to the target distribution in an attempt to confuse the Discriminator; while the Discriminator continuously improves its ability to distinguish between real and fake images and improves the segmentation effect.

[0057] (3) Consistency Learning

[0058] like Figure 3 As shown in Figure 1, the core idea of ​​consistency learning is to encourage the model to have similar outputs for the same sample after different transformations. The representative framework is Mean Teacher. The weight of the teacher network is calculated by the exponential moving average (EMA) of the student network weight, which ultimately forces the prediction consistency of the student network and the teacher network.

[0059] (4) Integration methods

[0060] The above semi-supervised learning methods have their own characteristics and perform well in different segmentation tasks.

[0061] Now, based on these methods, integrated methods have emerged that combine multiple semi-supervised segmentation ideas, such as adaptive balanced learning (AEL) and CPS (Cross Pseudo Supervision) that combine consistency learning and pseudo-labeling methods. These methods have achieved better results in some scenarios. Therefore, model diversity may be a key factor in achieving good results.

[0062] Embodiment 1

[0063] like Figure 4 As shown, this embodiment provides a railway concrete structure degradation identification method based on semi-supervised learning, comprising:

[0064] S1, establish a railway concrete structure degradation identification model based on semi-supervised learning;

[0065] S2, performing degradation identification of railway concrete structures based on the railway concrete structure degradation identification model.

[0066] like Figure 5 As shown, as a preferred embodiment, the S1 includes:

[0067] S11, determining a basic model for establishing the railway concrete structure degradation identification model;

[0068] In this embodiment, the S11 includes:

[0069] Generally, the semi-supervised learning method can use any basic semantic segmentation model. In order to better explore the role of the semi-supervised learning method, the basic model used in the embodiment of the present invention is the U-Net network architecture.

[0070] like Figure 6As shown, U-Net is a popular convolutional neural network (CNN) architecture specifically used for image segmentation tasks. The network architecture of U-Net in this embodiment has a symmetrical "U" shape, consisting of a symmetrical encoder (contraction path) and a decoder (expansion path). The encoder consists of a series of convolutional layers and pooling layers, which are used to extract features from the input image; the decoder consists of a series of deconvolutional layers and upsampling layers, which are used to reconstruct the features extracted by the encoder into a segmentation result of the same size as the original input image. In U-Net, there is also a skip connection between the encoder and the decoder, which is used to directly connect the high-level feature map in the encoder to the corresponding layer in the decoder to help the decoder better restore the segmentation result; in addition, U-Net also uses deconvolution and upsampling techniques to avoid problems such as information loss and resolution reduction. The structural characteristics of U-Net enable the network architecture to consider global and local information at the same time, and effectively use skip connections to improve segmentation accuracy. This design of U-Net makes it perform well in processing small objects and details in images.

[0071] S12, establishing a primary railway concrete structure degradation identification model based on the basic model;

[0072] like Figure 7 As shown, as a preferred implementation, the S12 includes:

[0073] (1) performing data preprocessing and partitioning to determine a sample set for training the railway concrete structure degradation identification model; wherein the sample set comprises: dividing each data set into two parts, namely a labeled data set and an unlabeled data set, dividing the labeled data set into two parts, namely a first labeled data set L1 and a second labeled data set L2, the first labeled data set L1 and the second labeled data set L2 are independently used for the first training model F1 and the second training model F2 to obtain a first prediction result and a second prediction result; at the same time, the unlabeled data set is used as the input of the first training model F1 and the second training model F2 to obtain a third prediction result and a fourth prediction result; the third prediction result and the fourth prediction result are cross-added to the first labeled data set L1 and the second labeled data set L2 as sample pairs for the next round of model training of the first training model F1 and the second training model F2;

[0074] In this embodiment, each data set is divided into a small amount of labeled data and a large amount of unlabeled data to form a labeled data set and an unlabeled data set, with the labeled data set: unlabeled data set = 1:9, and the three different scenarios maintain the same allocation ratio.

[0075] (2) determining a loss function of the model, wherein the loss function of the model includes two parts, namely, a cross pseudo-supervised loss function and an unsupervised loss function; the cross pseudo-supervised loss function uses a pseudo-segmentation map of one network as a supervisory signal of another network; the unsupervised loss function uses a consistency loss, i.e., calculating the consistency between the third prediction result and the fourth prediction result of the unlabeled data set;

[0076] (3) initializing the basic model, including: initializing the backbones of two segmentation networks of the U-Net network architecture using weights pre-trained on ImageNet, and randomly initializing the segmentation heads;

[0077] (4) Training the initialized model based on the loss function and training strategy of the model to obtain a primary railway concrete structure degradation identification model, wherein the training strategy includes using a standard pixel-level cross entropy loss to supervise the annotated data and the enhanced data.

[0078] As a preferred implementation, the training is performed using an optimizer, the optimizer is an Adam optimizer, the initial learning rate is set to 0.001, the momentum is 0.9, the weight decay is 0.0005, and the learning rate is adjusted according to a polynomial decay strategy.

[0079] As a preferred implementation, the enhanced data includes: enhancing the standard pixel-level cross entropy loss supervised annotation data by means of random horizontal flipping and multi-scale training.

[0080] S13, evaluating the primary railway concrete structure degradation identification model based on the evaluation index to obtain the railway concrete structure degradation identification model.

[0081] As a preferred embodiment, the evaluation index is the model accuracy. Of course, those skilled in the art should know that any other index suitable for evaluating the effectiveness and usability of the neural network model is within the protection scope of the present invention.

[0082] In this embodiment, the semi-supervised experimental results and conclusions of step S1 are shown in Table 1.

[0083] Table 1 Comparison results of semi-supervised experiments

[0084]

[0085] In the semi-supervised experiment of this embodiment, the U-Net model showed good performance in the image segmentation tasks of the three railway scenes. The test results show that the model achieved an accuracy of 82.70% on the tunnel lining test set, 81.02% on the ballastless track test set, and 78.80% on the bridge test set, with an improvement of +2.5, +1.8 and +0.9. Combined with the proportion of the number of unlabeled samples in the three scenes, it is concluded that the greater the proportion, the more obvious the improvement of the unsupervised learning effect. At the same time, the results of semi-supervised learning can well reflect the generalization and robustness of the model in the face of unknown samples. In general, the semi-supervised learning method provides a powerful tool for the field of railway deterioration disease image segmentation, which is of great significance to improving the accuracy of disease identification and the quality of engineering maintenance.

[0086] like Figure 8 As shown, as a preferred embodiment, the S2 includes:

[0087] S21, acquiring multiple surface images of the railway concrete structure as degradation identification samples;

[0088] S22, inputting the degradation identification sample into the railway concrete structure degradation identification model for degradation identification.

[0089] Embodiment 2

[0090] like Fig. 9 As shown, this embodiment provides a railway concrete structure degradation identification system based on semi-supervised learning, which is used to implement the method of embodiment 1, including:

[0091] A model building module 101 is used to build a railway concrete structure degradation identification model based on semi-supervised learning;

[0092] The degradation identification module 102 is used to identify the degradation of the railway concrete structure based on the railway concrete structure degradation identification model.

[0093] The present invention also provides a memory storing a plurality of instructions, wherein the instructions are used to implement the method as in the first embodiment.

[0094] like Fig.10 As shown, the present invention also provides an electronic device, including a processor 301 and a memory 302 connected to the processor 301, the memory 302 stores a plurality of instructions, and the instructions can be loaded and executed by the processor so that the processor can execute the method as in the first embodiment.

[0095] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit it. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or replace some or all of the technical features therein with equivalents. However, these modifications or replacements do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying railway concrete structure degradation based on semi-supervised learning, characterized in that: include: S1, establish a railway concrete structure degradation identification model based on semi-supervised learning; S2, performing degradation identification of railway concrete structures based on the railway concrete structure degradation identification model.

2. The method for identifying railway concrete structure degradation based on semi-supervised learning according to claim 1 is characterized in that: The S1 includes: S11, determining a basic model for establishing the railway concrete structure degradation identification model; the basic model is a U-Net network architecture, which is composed of a symmetrical encoder and decoder, the encoder is composed of a series of convolutional layers and pooling layers, and is used to extract features from an input image; the decoder is composed of a series of deconvolutional layers and upsampling layers, and is used to reconstruct the features extracted by the encoder into a segmentation result of the same size as the original input image; there is also a jump connection between the encoder and the decoder, which is used to directly connect the high-level feature map in the encoder to the corresponding layer in the decoder, and use deconvolution and upsampling to avoid information loss and resolution reduction; S12, establishing a primary railway concrete structure degradation identification model based on the basic model; S13, evaluating the primary railway concrete structure degradation identification model based on the evaluation index to obtain the railway concrete structure degradation identification model.

3. The method for identifying railway concrete structure degradation based on semi-supervised learning according to claim 2 is characterized in that: The S12 includes: (1) performing data preprocessing and partitioning to determine a sample set for training the railway concrete structure degradation identification model; wherein the sample set comprises: dividing each data set into two parts, namely a labeled data set and an unlabeled data set, dividing the labeled data set into two parts, namely a first labeled data set L1 and a second labeled data set L2, the first labeled data set L1 and the second labeled data set L2 are independently used for the first training model F1 and the second training model F2 to obtain a first prediction result and a second prediction result; at the same time, using the unlabeled data set as the input of the first training model F1 and the second training model F2 to obtain a third prediction result and a fourth prediction result; cross-adding the third prediction result and the fourth prediction result to the first labeled data set L1 and the second labeled data set L2 as a sample pair for the next round of model training of the first training model F1 and the second training model F2; the data set is a massive amount of data obtained by using a tunnel lining surface detection system; (2) determining a loss function of the model, wherein the loss function of the model includes two parts, namely, a cross pseudo-supervised loss function and an unsupervised loss function; the cross pseudo-supervised loss function uses a pseudo-segmentation map of one network as a supervisory signal of another network; the unsupervised loss function uses a consistency loss, i.e., calculating the consistency between the third prediction result and the fourth prediction result of the unlabeled data set; (3) initializing the basic model, including: initializing the backbones of two segmentation networks of the U-Net network architecture using weights pre-trained on ImageNet, and randomly initializing the segmentation heads; (4) Training the initialized model based on the loss function and training strategy of the model to obtain a primary railway concrete structure degradation identification model, wherein the training strategy includes using a standard pixel-level cross entropy loss to supervise the annotated data and the enhanced data.

4. The method for identifying railway concrete structure degradation based on semi-supervised learning according to claim 3 is characterized in that: The training is performed using an optimizer, which is an Adam optimizer. The initial learning rate is set to 0.001, the momentum is 0.9, the weight decay is 0.0005, and the learning rate is adjusted according to a polynomial decay strategy.

5. The method for identifying railway concrete structure degradation based on semi-supervised learning according to claim 4, characterized in that: The enhanced data includes: enhancing the standard pixel-level cross entropy loss supervised annotation data by random horizontal flipping and multi-scale training.

6. The method for identifying railway concrete structure degradation based on semi-supervised learning according to claim 5, characterized in that: The evaluation index is model accuracy.

7. The method for identifying railway concrete structure degradation based on semi-supervised learning according to claim 6, characterized in that: The S2 includes: S21, acquiring multiple surface images of the railway concrete structure as degradation identification samples; S22, inputting the degradation identification sample into the railway concrete structure degradation identification model for degradation identification.

8. A railway concrete structure degradation identification system based on semi-supervised learning, used to implement any of the methods of claims 1-7, characterized in that: include: A model building module (101) is used to build a railway concrete structure degradation identification model based on semi-supervised learning; A degradation identification module (102) is used to identify the degradation of the railway concrete structure based on the railway concrete structure degradation identification model.

9. An electronic device, characterized in that: The method comprises a processor and a memory, wherein the memory stores a plurality of instructions, and the processor is used to read the instructions and execute the method according to any one of claims 1 to 7.

10. A computer-readable storage medium, characterized in that: The computer-readable storage medium stores a plurality of instructions, and the plurality of instructions can be read by a processor to execute the method according to any one of claims 1 to 7.

Images