Method, device and equipment for detecting corrosion damage of switch bed plate and storage medium

Abstract

This invention belongs to the technical field of turnout slide plate quality assessment, and discloses a method, apparatus, equipment, and storage medium for detecting corrosion damage of turnout slide plates. The method includes: acquiring an image of the turnout slide plate; inputting the image of the turnout slide plate into a pre-trained corrosion damage detection model; if corrosion damage occurs in the turnout slide plate, the output result is "corrosion damage has occurred in the turnout slide plate"; the corrosion damage detection model is obtained by training a VGG-16+ improved ResNet50 network model using images of turnout slide plates containing corrosion damage as a training dataset. This invention achieves high accuracy in detecting corrosion of turnout slide plates, significantly reduces detection time, and features high recognition accuracy and fast corrosion detection speed.

Description

Technical Field

[0001] This invention belongs to the technical field of turnout slide plate quality assessment, specifically relating to methods, devices, equipment, and storage media for detecting corrosion damage to turnout slide plates. Background Technology

[0002] As one of the fundamental sectors of the national economy, railways have always received high attention. The turnout slide plate is a key component ensuring safe railway operation. As a crucial structure enabling trains to switch tracks or cross over, it is considered one of the weakest links in the track system. The slide plate is an important component of the turnout, providing support for the switch rail and frog rail, and directly affecting the safe operation of high-speed trains.

[0003] Turnout slides are exposed to the elements year-round, making them highly susceptible to corrosion from wind, sand, rain, snow, and debris from trains. Therefore, the consequences of corrosion are significant. Failure to detect safety hazards in turnout slides in a timely manner will severely impact train safety. Currently, the main method for detecting and identifying metal corrosion is traditional manual inspection. This method is overly reliant on human experience, highly random, time-consuming, labor-intensive, and has a low accuracy rate. Summary of the Invention

[0004] To address the problems existing in the prior art, this invention provides a method, apparatus, equipment, and storage medium for detecting corrosion damage to turnout slide plates. The accuracy of the measured corrosion of turnout slide plates is high, the detection time is greatly shortened, and it features high identification accuracy and fast corrosion detection speed.

[0005] To solve the above-mentioned technical problems, the present invention is achieved through the following technical solution:

[0006] A method for detecting corrosion damage to turnout slide plates, comprising:

[0007] Acquire an image of the turnout slide plate;

[0008] The image of the turnout slide plate is input into a pre-trained corrosion damage detection model. If corrosion damage occurs on the turnout slide plate, the output result is that the turnout slide plate has corrosion damage. The corrosion damage detection model is obtained by training a VGG-16+ improved ResNet50 network model using images of turnout slide plates with corrosion damage as the training dataset.

[0009] Furthermore, the VGG-16+ improved ResNet50 network model includes fully connected layers and feature vectors. The feature vectors are formed by concatenating the feature vectors obtained from VGG-16 through global pooling and the feature vectors obtained from improved ResNet50 through global pooling.

[0010] Furthermore, the VGG-16+ improved ResNet50 network model uses the Softmax function for classification.

[0011] Furthermore, the improved ResNet50 includes BN residual blocks, ReLU residual blocks, Conv residual blocks, BN residual blocks, ReLU residual blocks, and Conv residual blocks connected in sequence.

[0012] Furthermore, the improved ResNet50 also includes an SENet feature extraction module and three nonlinear activation layers.

[0013] Furthermore, the input layer size of the improved ResNet50 is 400×400.

[0014] Furthermore, the step of using images of turnout slide plates containing corrosion damage as a training dataset to train the VGG-16+ improved ResNet50 network model specifically involves:

[0015] Obtain a training dataset, which includes several sets of training data. Each set of training data includes: an image of a turnout slide plate that has suffered corrosion damage and a label that identifies the turnout slide plate image as having suffered corrosion damage.

[0016] The VGG-16+ improved ResNet50 network model was trained using the training dataset to obtain the corrosion damage detection model.

[0017] A device for detecting corrosion damage to turnout slide plates includes:

[0018] The acquisition module is used to acquire images of the turnout slide plate;

[0019] The detection module is used to input the image of the turnout slide plate into a pre-trained corrosion damage detection model. If corrosion damage occurs on the turnout slide plate, the output result is that the turnout slide plate has corrosion damage. The corrosion damage detection model is obtained by training a VGG-16+ improved ResNet50 network model using images of turnout slide plates with corrosion damage as the training dataset.

[0020] An apparatus includes a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the method for detecting corrosion damage to a turnout slide plate.

[0021] A computer-readable storage medium storing a computer program that, when executed by a processor, implements the steps of the method for detecting corrosion damage to a turnout slide plate.

[0022] Compared with the prior art, the present invention has at least the following beneficial effects:

[0023] This invention provides a method for detecting corrosion damage to turnout slide plates. For detecting corrosion damage to turnout slide plates, an image of the turnout slide plate is input into a pre-trained corrosion damage detection model. If corrosion damage occurs, the output result is "Corrosion damage has occurred." The corrosion damage detection model uses images of turnout slide plates with corrosion damage as a training dataset, training a VGG-16 + improved ResNet50 network model. Combining the VGG-16 and improved ResNet50 models, the feature vectors obtained through global pooling are fused using concatenation. Fully connected layers ensure a sufficient number of features, improving the training speed and recognition accuracy of the model. The recognition accuracy is high, reaching nearly 99% on a corrosion experimental dataset of turnout slide plates. Compared to traditional detection methods, this invention achieves high accuracy in detecting corrosion of turnout slide plates, significantly reduces detection time, and features high recognition accuracy and fast corrosion detection speed.

[0024] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0025] To more clearly illustrate the technical solutions in the specific embodiments of the present invention, the drawings used in the description of the specific embodiments will be briefly introduced below. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0026] Figure 1 This is a flowchart of a method for detecting corrosion damage to turnout slide plates according to the present invention.

[0027] Figure 2 This is a flowchart of the training process for the VGG-16+ improved ResNet50 network model of this invention.

[0028] Figure 3 Images of samples collected at different corrosion times.

[0029] Figure 4 This is a flowchart of the SE feature extraction module.

[0030] Figure 5 Flowchart for the fusion model.

[0031] Figure 6 The graph shows the results of 100 iterations with different learning rates.

[0032] Figure 7 The graph shows the results of 100 iterations with different batch sizes.

[0033] Figure 8 This is a graph showing the changes in model accuracy and loss rate.

[0034] Figure 9 This is a graph comparing the accuracy of different iteration numbers.

[0035] Figure 10 Confusion matrix diagram for test results.

[0036] Figure 11 The graphs show the accuracy and loss rates of different models.

[0037] Figure 12 Accuracy graphs for different models. Detailed Implementation

[0038] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0039] As a specific embodiment of the present invention, such as Figure 1 As shown, a method for detecting corrosion damage to turnout slide plates specifically includes the following steps:

[0040] S1. Acquire images of the turnout slide plate. For example, images of the turnout slide plate are acquired using a camera and an electron microscope.

[0041] S2. Input the image of the turnout slide plate into the pre-trained corrosion damage detection model. If the turnout slide plate is corroded, the output result is that the turnout slide plate is corroded.

[0042] The corrosion damage detection model is obtained by training a VGG-16+ improved ResNet50 network model using images of turnout slide plates containing corrosion damage as the training dataset.

[0043] Specifically, the VGG-16+ improved ResNet50 network model includes fully connected layers and feature vectors. The feature vectors are formed by concatenating the feature vectors obtained from VGG-16 through global pooling and the feature vectors obtained from improved ResNet50 through global pooling. The VGG-16+ improved ResNet50 network model uses the Softmax function for classification.

[0044] Specifically, the improved ResNet50 includes BN residual blocks, ReLU residual blocks, Conv residual blocks, BN residual blocks, ReLU residual blocks, and Conv residual blocks connected in sequence; the improved ResNet50 also includes an SENet feature extraction module and three nonlinear activation layers; the input layer size of the improved ResNet50 is 400×400.

[0045] Specifically, such as Figure 2 As shown, the VGG-16+ improved ResNet50 network model is trained using images of turnout slide plates with corrosion damage as the training dataset, as detailed below:

[0046] a. Obtain a training dataset, which includes several sets of training data. Each set of training data includes: an image of a turnout slide plate that has suffered corrosion damage and a label that identifies the turnout slide plate image as having suffered corrosion damage.

[0047] b. Train the VGG-16+ improved ResNet50 network model using the training dataset to obtain the corrosion damage detection model.

[0048] Regarding the acquisition of the training dataset in step a, for example, images of the turnout slide plate at different time points during the corrosion process are collected using a camera and an electron microscope. The collected images are then enhanced, resized, normalized, and labeled to obtain a sufficient and standardized image sample dataset. The image sample dataset is then divided into a training dataset and a validation dataset according to the data ratio.

[0049] The sample dataset includes corrosion image sample data of turnout slide plates at 0h, 2h, 6h, 24h, 48h, and 72h.

[0050] The training dataset includes 60% of the corrosion image samples of turnout slide plates at 0h, 60% of the corrosion image samples of turnout slide plates at 2h, 60% of the corrosion image samples of turnout slide plates at 6h, 60% of the corrosion image samples of turnout slide plates at 24h, 60% of the corrosion image samples of turnout slide plates at 48h, and 60% of the corrosion image samples of turnout slide plates at 72h.

[0051] The test dataset includes 40% of the corrosion image samples of turnout slide plates at 0h, 40% at 2h, 40% at 6h, 40% at 24h, 40% at 48h, and 40% at 72h.

[0052] Regarding the VGG-16+ improved ResNet50 network model in step b, it's important to note that ResNet50 can only extract sufficient erosion features in subsequent convolutional layers if the input image size is large enough. Because the ResNet50 model is deep and has many parameters, an excessively large input image size is detrimental to the computation of subsequent neural networks. Therefore, the ResNet50 network model was adjusted based on image features. The input layer was designed with a 400×400 erosion image, and a small convolutional filter was used to integrate three non-linear activation layers instead of a single non-linear activation layer. This increased the discriminative power, reduced the number of network parameters, and decreased the computational cost. Residual blocks can address the issue of declining prediction performance as the number of network layers increases. However, simply changing the training parameters of the normal network causes the model to converge quickly, and the prediction performance decreases further with increasing network layers. Therefore, the branches of the ResNet50 residual blocks are improved by changing the original Conv-BN-Relu order to BN-Relu-Conv-BN-Relu-Conv. Furthermore, an SENet feature extraction module is added to the ResNet50 network model to allocate weights to each channel. The feature extraction process is as follows:

[0053] First, a global average pooling operation is performed on the input image of size H×W×C to obtain a 1×1×C feature map, where W, H, and C represent the width, height, and number of channels of the feature map of the turnout slide plate corrosion image, respectively. Second, the number of channels of the feature map is reduced to 1 / r of the original number through two fully connected layers, and then increased to the original number of channels through the ReLU activation function. Then, the normalized weights are obtained through the sigmoid function. Finally, the normalized weights are weighted onto the feature of each channel through the scale operation, resulting in a feature map of size H'×W'×C' for the turnout slide plate surface corrosion image.

[0054] The feature vectors obtained from the two models, VGG-16 and the improved ResNet50, through global pooling are fused using Concatenate, then a fully connected layer is used to ensure a sufficient number of features, and finally Softmax is used for classification.

[0055] For example, a VGG-16+ improved ResNet50 network model is trained using a training dataset. After training, the validation dataset is fed into the trained VGG-16+ improved ResNet50 network model to obtain corrosion loss detection results, which are then output. The specific process of corrosion loss detection is as follows:

[0056] ① The preprocessed corrosion image data of the turnout slide plate is sent to the VGG-16 and the improved ResNet50 network models respectively, and is not directly classified.

[0057] ② The feature vectors are fused by adding their respective GAP layers and then adding a six-classification layer to classify the fused feature vectors.

[0058] ③ The fused model achieved an accuracy of 0.99 in Acc, Recall, Precision, and F1 scores. After 100 training iterations, the VGG-16+ improved ResNet50 performed well, with an accuracy of 98.98%.

[0059] ④ The VGG-16+ improved ResNet50 network model is significantly better than other models in terms of test set accuracy, with an accuracy rate of 98.98%, which verifies that the method can quickly and accurately identify the corrosion status of turnout slide plate steel plates.

[0060] Example

[0061] To demonstrate the effectiveness of this method, a simulation experiment was first conducted using Q235 steel as the substrate in a neutral salt spray environment. The substrate material dimensions were 100mm × 100mm × 10mm. The cladding layer material consisted of Cu powder and Ni powder, mixed in a 9:1 mass ratio. The Cu powder used was spherical with a smooth surface, a particle size of 38μm, and a purity of 99%. The Ni powder used was also spherical with a particle size of 45μm and a purity of 99%.

[0062] Table 1: Main experimental instruments used before and after the neutral salt spray experiment

[0063] Equipment Name Specifications and Models Technical parameters Planetary ball mill YXQM-4L Power: 0.75KW High-speed laser cladding machine tool ZKZM-JC41 — ultrasonic cleaner KQ3200DE Frequency: 40kHz Salt spray test chamber FQY025 Salt spray deposition rate: 1-2 ml / 80 cm·h air compressor EV51 Power: 1.5KW Electronic balance DHC-5000 Accurate to 0.01g Digital thickness gauge — Accurate to 0.01mm Scanning electron microscope digital camera Canon EOS 80D Resolution: 6000 x 4000 pixels

[0064] The preparation process of the corrosion material is as follows: After completing the chemical plating of graphite powder, the nickel-plated graphite is made into copper-based graphite self-lubricating material by powder metallurgy. Then, sodium carboxymethyl cellulose binder is uniformly mixed with the cladding powder (copper-based graphite composite self-lubricating material) into a paste and applied to the surface of the substrate Q235 steel plate. Laser cladding is performed with parameters of laser power 1000W, scanning speed 200mm / s, spot diameter 2.5mm and overlap rate 50%.

[0065] The process of the salt spray corrosion test is as follows: a 5% sodium chloride solution is continuously sprayed in the form of a spray tower, so that the salt spray settles onto six prepared test specimens (100mm×100mm×10mm). Finally, samples are taken at different time intervals according to the experimental settings for subsequent image information acquisition.

[0066] like Figure 2 As shown, after the samples reached the expected salt spray corrosion time, they were removed and dried. After complete drying, images of the untreated corroded surface were captured using a camera. A total of 711 images were captured during the experiment, and the files were in *.JPG format.

[0067] The following comparative experiments on different corrosion detection methods are used to verify the method of the present invention.

[0068] Step 1: The preprocessed turnout slide plate image data are fed into the improved ResNet50, VGG-16, and Xception pre-trained networks, respectively. The training process of the neural network is considered as a parameter optimization problem. During training, the input samples are iteratively processed to achieve the effect of cyclically updating the weights and bias parameters in the network. Learning rates of 0.005, 0.001, and 0.0001 and batch sizes of 16, 32, and 64 are set. The parameters most suitable for this model are selected by analyzing the recognition rate after 100 iterations, and the iteration results are obtained. Figure 6 and Figure 7 As shown in the figure, the iteration results show that after 100 training iterations, the accuracy of both the training and test sets tends to level off with the number of iterations. The curves in the figure indicate that with a learning rate of 0.005 and a batch size of 32, the accuracy is higher, the curve is smoother, and the model exhibits better performance. The training parameter settings for the experimental model are shown in Table 2.

[0069] Table 2: Model Training Parameters

[0070] Step 2: The confusion matrix obtained by testing the experimental model on the test set is presented. A total of 16 images in the test set were incorrectly predicted. Among 213 images from 11:48 AM on May 24th, the model correctly identified one as an erosion image from 9:00 AM that day. Among 219 erosion images from 9:48 AM on May 24th, one was incorrectly identified as an erosion image from 9:00 AM that day, while the rest were correctly identified. Among 207 erosion images from 9:48 AM on May 25th, nine were incorrectly identified as erosion images from the same time on May 26th, while the rest were correctly identified. Among 233 erosion images from 9:48 AM on May 26th, five were incorrectly identified as erosion images from other time points, while the rest were correctly identified. The results show that the experimental model has a high accuracy rate, but there is still room for improvement.

[0071] Step 3: The preprocessed corrosion image data of the turnout slide plate is fed into VGG-16, Xception, and the improved ResNet50 network respectively. Instead of direct classification, they are fused through their respective GAP layers and by permutation and combination. This experiment selects the two fusion models with better classification performance. Finally, a six-classification layer is added to classify the fused feature vectors. The results show that the fused model achieves an index of 0.99 in all aspects, significantly higher than the improved models in steps 1 and 2, indicating that the model is more suitable for detecting corrosion images of turnout slide plates.

[0072] Step 4: To obtain a model with better generalization ability, this experiment trains the models from Steps 1, 2, and 3 100 times each, and then plots the accuracy and loss rate curves, as shown below. Figure 8 As shown, the horizontal axis represents the number of model iterations, the vertical axis of the left graph represents accuracy, and the vertical axis of the right graph represents loss rate. Different colored curves represent different models. Through horizontal comparison, the training results of two fusion models and three models after 100 iterations were analyzed. Figure 11 and Figure 12 As shown, among the five convolutional neural network models, the VGG-16+ improved ResNet50 has the best performance, with an accuracy of 98.98%. To improve the model's training speed and reduce its training time...

[0073] Table 3: Experimental Results of Different Models

[0074]

[0075]

[0076] As shown in Table 3, after the same number of training iterations, the improved ResNet50 demonstrated better recognition performance as a single model, while the Xception model had the lowest validation set accuracy among all models, reaching only 58.83%. However, the accuracy significantly improved when the improved ResNet50 was fused with the Xception model. In addition, as... Figure 12 As shown, the fusion-improved model described in this invention significantly outperforms other models in validation set accuracy, achieving an accuracy of 98.98%, which is 3.16% higher than VGG-16, 1.02% higher than the improved ResNet50, 40.15% higher than Xception, and 2.89% higher than Xception + improved ResNet50. Therefore, the improved model VGG-16 + improved ResNet50 described in this invention has outstanding advantages over the other four methods.

[0077] This embodiment provides a device for detecting corrosion damage to turnout slide plates, including:

[0078] The acquisition module is used to acquire images of the turnout slide plate;

[0079] The detection module is used to input the image of the turnout slide plate into a pre-trained corrosion damage detection model. If corrosion damage occurs on the turnout slide plate, the output result is that the turnout slide plate has corrosion damage. The corrosion damage detection model is obtained by training a VGG-16+ improved ResNet50 network model using images of turnout slide plates with corrosion damage as the training dataset.

[0080] In one embodiment of the present invention, a computer device is provided, comprising a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions to achieve a corresponding method flow or corresponding function. The processor described in this embodiment of the present invention can be used to implement the operation of a method for detecting corrosion damage to turnout slide plates.

[0081] In one embodiment of the present invention, a method for detecting corrosion damage to turnout slide plates, if implemented as a software functional unit and sold or used as an independent product, can be stored in a computer-readable storage medium. Based on this understanding, all or part of the processes in the methods of the above embodiments of the present invention can also be implemented by a computer program instructing related hardware. The computer program can be stored in a computer-readable storage medium, and when executed by a processor, it can implement the steps of the various method embodiments described above. The computer program includes computer program code, which can be in the form of source code, object code, executable files, or certain intermediate forms. The computer-readable storage medium includes permanent and non-permanent, removable and non-removable media, and information storage can be implemented by any method or technology. The information can be computer-readable instructions, data structures, program modules, or other data.

[0082] The computer storage medium can be any available medium or data storage device that a computer can access, including but not limited to magnetic storage (e.g., floppy disks, hard disks, magnetic tapes, magneto-optical disks (MOs)), optical storage (e.g., CDs, DVDs, BDs, HVDs), and semiconductor storage (e.g., ROMs, EPROMs, EEPROMs, non-volatile memory (NAND flash), solid-state drives (SSDs)).

[0083] Those skilled in the art will understand that embodiments of this application can be provided as methods, systems, or computer program products. Therefore, this application can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, this application can take the form of a computer program product embodied on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0084] This application is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of this application. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, create a machine for implementing the flowchart illustrations. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0085] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0086] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0087] Finally, it should be noted that the above-described embodiments are merely specific implementations of the present invention, used to illustrate the technical solutions of the present invention, and not to limit it. The scope of protection of the present invention is not limited thereto. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that any person skilled in the art can still modify or easily conceive of changes to the technical solutions described in the foregoing embodiments within the technical scope disclosed in the present invention, or make equivalent substitutions for some of the technical features; and these modifications, changes, or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention, and should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. A method for detecting corrosion damage of a frog sliding bed plate, characterized in that, The method comprises the following steps: acquiring an image of a turnout sliding bed plate; inputting the image of the turnout sliding bed plate into a pre-trained corrosion damage detection model, and if the turnout sliding bed plate has corrosion damage, outputting a result that the turnout sliding bed plate has corrosion damage; the corrosion damage detection model is obtained by training a VGG-16 and an improved ResNet50 network model using images of turnout sliding bed plates having corrosion damage as a training data set; the VGG-16 and the improved ResNet50 network model comprise a full connection layer and a feature vector, the feature vector is formed by using Concatenate to fuse features of a feature vector obtained by global pooling of the VGG-16 and a feature vector obtained by global pooling of the improved ResNet50; the VGG-16 and the improved ResNet50 network model use a Softmax function for classification; the improved ResNet50 comprises BN residual blocks, Relu residual blocks, Conv residual blocks, BN residual blocks, Relu residual blocks and Conv residual blocks connected in sequence; the improved ResNet50 further comprises an SENet feature extraction module and three nonlinear activation layers; the input layer size of the improved ResNet50 is 400*400; the VGG-16 and the improved ResNet50 network model are trained using images of turnout sliding bed plates having corrosion damage as a training data set, and the training is as follows: acquiring a training data set, the training data set comprising a plurality of groups of training data, each group of training data comprising an image of a turnout sliding bed plate having corrosion damage and a label identifying the image of the turnout sliding bed plate as having corrosion damage; training the VGG-16 and the improved ResNet50 network model using the training data set to obtain the corrosion damage detection model.

2. A device for detecting corrosion damage of a frog sliding bed plate, characterized in that, The method comprises the following steps: an acquiring module for acquiring an image of a turnout sliding bed plate; a detection module for inputting the image of the turnout sliding bed plate into a pre-trained corrosion damage detection model, and if the turnout sliding bed plate has corrosion damage, outputting a result that the turnout sliding bed plate has corrosion damage; the corrosion damage detection model is obtained by training a VGG-16 and an improved ResNet50 network model using images of turnout sliding bed plates having corrosion damage as a training data set; the VGG-16 and the improved ResNet50 network model comprise a full connection layer and a feature vector, the feature vector is formed by using Concatenate to fuse features of a feature vector obtained by global pooling of the VGG-16 and a feature vector obtained by global pooling of the improved ResNet50; the VGG-16 and the improved ResNet50 network model use a Softmax function for classification; the improved ResNet50 comprises BN residual blocks, Relu residual blocks, Conv residual blocks, BN residual blocks, Relu residual blocks and Conv residual blocks connected in sequence; the improved ResNet50 further comprises an SENet feature extraction module and three nonlinear activation layers; The input layer size of the improved ResNet50 is 400*400; The improved ResNet50 is trained by using the image of the turnout slide bed plate with corrosion damage as the training data set, and the specific steps are as follows: Obtaining a training data set, the training data set includes a plurality of groups of training data, and each group of training data includes: an image of a turnout slide bed plate with corrosion damage and a label identifying the turnout slide bed plate image as having corrosion damage; The VGG-16 and the improved ResNet50 network model are trained by using the training data set, and the corrosion damage detection model is obtained.

3. An apparatus comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein, The processor executes the computer program to realize the steps of the turnout slide bed plate corrosion damage detection method in claim 1.

4. A computer-readable storage medium storing a computer program, the computer program comprising instructions that, when executed by a computer, cause the computer to perform the method of any one of claims 1 to 3. The computer program is executed by the processor to realize the steps of the turnout slide bed plate corrosion damage detection method in claim 1.

Images