A method for identifying the characteristics of interference factors in track inspection data

Abstract

The present invention relates to a method for identifying interference factor characteristics of track inspection data, including: obtaining the track inspection data to be measured; determining whether the interference category of the track inspection data to be measured is a gauge machine failure through pattern recognition; if not, inputting the track inspection data to be measured into a trained first recognition network and a second recognition network respectively to obtain corresponding first recognition results and second recognition results; determining whether there is interference and the interference category in the track inspection data to be measured according to the first recognition result and the second recognition result; wherein, both the first recognition network and the second recognition network are deep learning networks based on channel attention, and the deep learning network based on channel attention includes a cascaded feature extraction module and a classification module. The method of the present invention uses a deep learning network to identify the characteristics of interference factors in track inspection data with multiple channels and complex vibration conditions, improving the identification efficiency and accuracy of the interference factors in track inspection data.

Description

Technical Field

[0001] The present invention belongs to the technical field of track inspection data analysis, and particularly relates to a method for identifying interference factor characteristics of track inspection data. Background Art

[0002] The high-speed railway in our country has developed rapidly. While enjoying the benefits brought by rail transit, it is more important to ensure the safe operation of trains. To ensure the smooth operation of trains, the premise is to ensure the quality and safety of the track line. Among them, the track fault detection work is an important measure to ensure track safety at all times during the track operation state.

[0003] The track inspection vehicle is an important equipment for undertaking the daily track inspection tasks, and can detect the track elevation, level, cross level, etc. Track disease detection is an important task to ensure track safety. Various diseases on the track, such as cross level, unevenness, track wear, etc., can be judged through the geometric waveform data obtained by the track inspection vehicle. However, the track inspection vehicle will be affected by natural interference factors such as rain and sunlight or equipment failures such as laser failures and displacement gauge failures during operation, resulting in abnormal fluctuations in the track inspection data and being unable to be used for track disease analysis. Therefore, various abnormal data in the track inspection data need to be analyzed and eliminated before disease analysis.

[0004] At present, the analysis and elimination of abnormal data mainly rely on manual labor. The amount of track inspection data is huge, the workload is huge, and the requirements for the experience and professional knowledge of analysts are relatively high. Also, because the discrimination criteria of each person are not exactly the same, problems such as mislabeling and missing labeling will occur. Summary of the Invention

[0005] In order to solve the above problems existing in the prior art, the present invention provides a method for identifying interference factor characteristics of track inspection data. The technical problems to be solved by the present invention are realized through the following technical solutions:

[0006] The present invention provides a method for identifying interference factor characteristics of track inspection data, including:

[0007] Step 1: Obtain the track inspection data to be measured;

[0008] Step 2: Determine whether the interference category of the track inspection data to be measured is a track gauge machine failure through pattern recognition;

[0009] If not, input the track inspection data to be measured into the trained first recognition network and the second recognition network respectively to obtain the corresponding first recognition result and second recognition result;

[0010] Step 3: Determine whether there is interference and the interference category in the track inspection data to be measured according to the first recognition result and the second recognition result;

[0011] Among them, both the first recognition network and the second recognition network are deep learning networks based on channel attention. The deep learning network based on channel attention includes a cascaded feature extraction module and a classification module. The feature extraction module includes a plurality of cascaded feature extraction units, and each feature extraction unit includes a plurality of cascaded convolutional units and a channel attention module.

[0012] In an embodiment of the present invention, the feature extraction module includes 5 cascaded feature extraction units. The feature extraction unit includes 2 cascaded convolutional units and a channel attention module; each convolutional unit includes a cascaded convolutional layer and a Leaky ReLU activation function layer;

[0013] The channel attention module includes a cascaded global average pooling layer, a fully connected layer, a Leaky ReLU activation function layer, and a fully connected layer;

[0014] The channel attention module obtains an attention vector according to the feature matrix extracted by the convolutional unit connected to it, and multiplies the attention vector by the input feature matrix of the channel attention module and uses it as the output feature matrix of the feature extraction unit.

[0015] In an embodiment of the present invention, the convolution kernel size of the convolution layer of the convolutional unit in the first feature extraction unit is 1×7, and the convolution kernel size of the convolution layer of the convolutional unit in the remaining feature extraction units is 1×5.

[0016] In an embodiment of the present invention, the classification module includes a cascaded Dropout layer, a global average pooling layer, a fully connected layer, and a Softmax function layer, wherein the dropout probability of the Dropout layer is 0.5.

[0017] In an embodiment of the present invention, the interference categories of track inspection data include: rain interference, sunlight interference, rail grinding, laser failure, track gauge machine failure, displacement gauge failure, and ice and snow interference.

[0018] In an embodiment of the present invention, the track inspection data to be measured is detected by a GJ-6 type track inspection system. The track inspection data to be measured includes 7-channel data of left vertical alignment, right vertical alignment, left alignment, right alignment, track gauge, level, and cross level.

[0019] In an embodiment of the present invention, in step 2, if all the track gauge channel data of the track inspection data to be measured are 0, it is determined that the interference category of the track inspection data to be measured is a track gauge machine failure.

[0020] In an embodiment of the present invention, in step 2, the to-be-detected track inspection data is respectively input into the trained first recognition network and the second recognition network to obtain corresponding first recognition result and second recognition result, including:

[0021] Input the left track alignment channel data, right track alignment channel data and gauge channel data of the to-be-detected track inspection data into the trained first recognition network to obtain a first recognition result; input the left vertical alignment channel data, right vertical alignment channel data and gauge channel data of the to-be-detected track inspection data into the trained second recognition network to obtain a second recognition result.

[0022] In an embodiment of the present invention, the first recognition result is that the to-be-detected track inspection data is normal data, or the interference category of the to-be-detected track inspection data is rain and snow interference, sunlight interference, rail grinding or laser failure;

[0023] The second recognition result is that the to-be-detected track inspection data is normal data, or the interference category of the to-be-detected track inspection data is displacement gauge failure.

[0024] In an embodiment of the present invention, step 3 includes:

[0025] If both the first recognition result and the second recognition result are that the to-be-detected track inspection data is normal data, it is determined that there is no interference in the to-be-detected track inspection data;

[0026] Otherwise, it is determined that there is interference in the to-be-detected track inspection data, and the interference category is the interference category corresponding to the recognition result;

[0027] Wherein, if the interference category of the to-be-detected track inspection data is rain and snow interference, it is determined that the interference category is rain interference or ice and snow interference according to the geographical location and the season of the predicted section.

[0028] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0029] The track inspection data interference factor feature recognition method of the present invention aims at the problems existing in the current track inspection data validity analysis task, such as large amount of data, inconsistent manual classification standards, easy omission and mislabeling, and low recognition rate. A interference factor feature recognition method is proposed, which uses a deep learning network to recognize the interference factor features of track inspection data with multiple channels and complex vibration conditions, improves the recognition efficiency and accuracy of the interference factors of track inspection data, and thus improves the detection accuracy and efficiency of track diseases.

[0030] The above description is only an overview of the technical solution of the present invention. In order to understand the technical means of the present invention more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of the present invention more obvious and understandable, the following specifically gives preferred embodiments and, in conjunction with the accompanying drawings, is described in detail as follows. Description of the Drawings

[0031] Figure 1 It is a schematic diagram of a method for identifying interference factor characteristics of track inspection data provided by an embodiment of the present invention;

[0032] Figure 2 It is a schematic structural diagram of a deep learning network based on channel attention provided by an embodiment of the present invention. Detailed Embodiments

[0033] In order to further elaborate on the technical means and effects adopted by the present invention to achieve the predetermined invention purpose, the following, in conjunction with the accompanying drawings and specific embodiments, details a method for identifying interference factor characteristics of track inspection data proposed according to the present invention.

[0034] The foregoing and other technical contents, features and effects of the present invention can be clearly presented in the following detailed description in conjunction with the accompanying drawings. Through the description of the specific embodiments, a more in-depth and specific understanding of the technical means and effects adopted by the present invention to achieve the predetermined purpose can be obtained. However, the accompanying drawings are only for reference and illustration, and are not used to limit the technical solution of the present invention.

[0035] Embodiment 1

[0036] Please refer to Figure 1 , Figure 1 which is a schematic diagram of a method for identifying interference factor characteristics of track inspection data provided by an embodiment of the present invention. As shown in the figure, the method for identifying interference factor characteristics of track inspection data in this embodiment includes:

[0037] Step 1: Obtain the track inspection data to be measured;

[0038] In an optional embodiment, the track inspection data to be measured is obtained by using a GJ-6 type track inspection system.

[0039] Generally, the GJ-6 type track inspection system is mounted on a track inspection vehicle to collect track inspection data. However, during the operation of the track inspection vehicle, it will be affected by natural interference factors or equipment failures, resulting in abnormal fluctuations in the track inspection data. The interference categories of the track inspection data include: rain interference, sunlight interference, rail grinding, laser failure, track gauge machine failure, displacement gauge failure, and ice and snow interference.

[0040] The main track inspection data includes data from 9 channels, including left alignment, right alignment, left gauge, right gauge, gauge, superelevation, level, cross level and speed, etc. The speed of the track inspection vehicle has no connection with the generation mechanism of interference data, and the superelevation is obtained by calculating the data of the left and right alignment channels and has no direct relationship with interference factors. Therefore, in this embodiment, the 7-channel data of the track inspection data to be measured is used, that is, the channel data of left alignment, right alignment, left gauge, right gauge, gauge, level and cross level are used to identify interference factors.

[0041] Step 2: Determine whether the interference category of the track inspection data to be measured is a gauge machine failure through pattern recognition;

[0042] In an optional implementation manner, if the interference category of the track inspection data to be measured is a gauge machine failure, then the waveform of the gauge channel data is a straight line, that is, if all the gauge channel data of the track inspection data to be measured is 0, it is determined that the interference category of the track inspection data to be measured is a gauge machine failure.

[0043] If the interference category of the track inspection data to be measured is not a gauge machine failure, the track inspection data to be measured is respectively input into the trained first recognition network and the second recognition network to obtain corresponding first recognition results and second recognition results.

[0044] In an optional implementation manner, inputting the track inspection data to be measured into the trained first recognition network and the second recognition network respectively to obtain corresponding first recognition results and second recognition results includes:

[0045] Input the left gauge channel data, right gauge channel data and gauge channel data of the track inspection data to be measured into the trained first recognition network to obtain a first recognition result. Input the left alignment channel data, right alignment channel data and gauge channel data of the track inspection data to be measured into the trained second recognition network to obtain a second recognition result.

[0046] In this embodiment, the first recognition result is that the track inspection data to be measured is normal data, or the interference category of the track inspection data to be measured is rain and snow interference, sunlight interference, rail grinding or laser failure. The second recognition result is that the track inspection data to be measured is normal data, or the interference category of the track inspection data to be measured is displacement gauge failure.

[0047] It should be noted that due to the large number of original channels of track inspection data and the complex waveforms of each channel, during the feature extraction process of the classification algorithm, it is easily affected by the irrelevant information between channels, resulting in the inability to accurately identify and extract key interference features. Through the analysis of various interference factors, it is found that the affected channels of different categories are not the same, and the interference features of each category are not obvious or there are no category-related features in some channels. After a large number of observations and analyses of the data samples of each category, it is found that generally only the data of the three channels of left rail alignment, right rail alignment, and gauge will be affected in the samples of rain interference, sunlight interference, rail grinding, and laser failure; generally only the data of the three channels of left elevation, right elevation, and gauge will be affected in the samples of displacement gauge failure. In addition, the interference form of ice and snow interference is similar to that of rain interference, and ice and snow interference and rain interference can be combined into one type of interference.

[0048] Therefore, in this embodiment, by constructing two recognition networks and using five types of data sets including rain interference, sunlight interference, rail grinding, laser failure, and normal samples, and two types of data sets including displacement gauge failure and normal data to train the two recognition networks, the trained first recognition network and second recognition network are obtained.

[0049] In an optional implementation manner, both the first recognition network and the second recognition network are deep learning networks based on channel attention. Please refer to Figure 2 the schematic structural diagram of a deep learning network based on channel attention provided by the embodiment of the present invention shown in the figure. As shown in the figure, the deep learning network based on channel attention includes a cascaded feature extraction module and a classification module.

[0050] Among them, the feature extraction module includes a cascaded plurality of feature extraction units, and each feature extraction unit includes a cascaded plurality of convolutional units and a channel attention module.

[0051] In this embodiment, by setting a plurality of feature extraction units to extract the category features of different interference factors, and a channel attention module will be passed after every two layers of convolution to improve the network prediction accuracy and training stability.

[0052] Optionally, the feature extraction module includes 5 cascaded feature extraction units, and the feature extraction unit includes 2 cascaded convolutional units and a channel attention module; each convolutional unit includes a cascaded convolutional layer and a Leaky ReLU activation function layer. In this embodiment, the Leaky ReLU activation function layer can, to a certain extent, avoid the problem of gradient disappearance.

[0053] Optionally, the convolution kernel size of the convolution layer in the first feature extraction unit is 1×7, and the convolution kernel size of the convolution layer in the remaining feature extraction units is 1×5. The large-size convolution kernel can expand the receptive field, which is beneficial to extracting continuous features, while the small-size convolution kernel is more capable of extracting the detailed features of different interference factors.

[0054] In an optional embodiment, the channel attention module includes a cascaded global average pooling layer, a fully connected layer, a Leaky ReLU activation function layer, and a fully connected layer.

[0055] In this embodiment, the channel attention module obtains an attention vector based on the feature matrix extracted by the convolution unit connected to it, and multiplies the attention vector by the input feature matrix of the channel attention module to obtain the output feature matrix of the feature extraction unit. By using the channel attention module, the extracted features can be adaptively selectively learned, enabling the network's attention to focus on the feature channels useful for classification.

[0056] In this embodiment, after the track inspection data enters the feature extraction unit, the convolution layer extracts features from the input data, and the Leaky ReLU activation function layer activates the data features extracted by the convolution layer, that is, performs calculations on the feature values according to a fixed rule. The channel attention module is used to learn the relationships between channels and adaptively adjust the model's response to each channel. After passing through the channel attention module, the size of the features does not change, but each channel is assigned a weight adaptively learned by the network, which can significantly improve the classification results of the model.

[0057] In an optional embodiment, the classification module includes a cascaded Dropout layer, a global average pooling layer, a fully connected layer, and a Softmax function layer.

[0058] In this embodiment, the dropout probability of the Dropout layer is 0.5. The Dropout layer is located after the last channel attention module, and its main function is to avoid overfitting during the model training process. The global average pooling layer is located after the Dropout layer. The global average pooling layer can avoid overfitting to a certain extent and improve the interpretability of the network model. The fully connected layer is located after the global average pooling layer, and its main function is to reduce the high-dimensional features extracted to the dimension of the number of predicted categories. The Softmax activation function layer is located after the fully connected layer, and its main function is to receive the output of the fully connected layer, use mathematical calculations to unify the output values between 0 and 1 as the classification scores for each category, correspond the classification results to the category labels, and output the interference category corresponding to the category label with the highest classification score as the recognition result.

[0059] It should be noted that during the training of the recognition network, the data of the three channels of the left track direction, the right track direction, and the gauge of the sample data in the five types of datasets are input into the first recognition network for training; the data of the three channels of the left height, the right height, and the gauge of the sample data in the two types of datasets are input into the second recognition network for training.

[0060] Step 3: According to the first recognition result and the second recognition result, determine whether there is interference in the track inspection data to be measured and the interference category.

[0061] In an optional implementation manner, if both the first recognition result and the second recognition result are that the track inspection data to be measured is normal data, it is determined that there is no interference in the track inspection data to be measured; otherwise, it is determined that there is interference in the track inspection data to be measured, and the interference category is the interference category corresponding to the recognition result.

[0062] That is, if only one of the first recognition result and the second recognition result is that the track inspection data to be measured is normal data, it is determined that there is one type of interference in the track inspection data to be measured, and the interference category is the interference category corresponding to the other recognition result. Otherwise, it is determined that there are two types of interference in the track inspection data to be measured, and the interference categories are the interference categories corresponding to the two recognition results.

[0063] It should be noted that if the interference category of the track inspection data to be measured is rain and snow interference, then according to the geographical location and the season of the predicted section, it is determined that the interference category is rain interference or ice and snow interference.

[0064] The method for identifying the characteristics of interference factors in track inspection data in this embodiment aims at the problems currently existing in the task of analyzing the effectiveness of track inspection data, such as large data volume, inconsistent manual classification standards, easy omission and mislabeling, and low recognition rate. A method for identifying the characteristics of interference factors is proposed, which uses a deep learning network to identify the characteristics of interference factors in track inspection data with multiple channels and complex vibration conditions, improving the recognition efficiency and accuracy of the interference factors in track inspection data, thereby improving the accuracy and efficiency of track disease detection.

[0065] Embodiment 2

[0066] This embodiment illustrates the effect of the method for identifying the characteristics of interference factors in track inspection data of the present invention through a comparative experiment.

[0067] 1. Training environment and parameter configuration of the recognition network

[0068] Programming is carried out based on the Python language, and training and testing are carried out based on the Pytorch deep learning framework. The hardware device is a server based on the Linux 18.04 operating system with 64G of memory, a 12-core CPU, and a 24G video memory 3090 GPU. In this embodiment, the Adam optimizer is selected to train and optimize the model. The initial learning rate is set to 0.00001, the number of samples taken each time for training BatchSize is set to 32, the total number of iterations Epoch for training on five types of datasets is set to 1000, and the learning rate is reduced to 10% at the 950th Epoch. The total number of iterations Epoch for training on two types of datasets is set to 100, and the learning rate is reduced to 10% at the 90th Epoch.

[0069] Among them, both datasets are divided into a training set and a test set according to a ratio of 8:2. For convenience during training, the test set of the five-classification is further segmented, and according to a ratio of 7:3 for each category, it is divided into a validation set and a test set. The validation set is used to perform real-time testing on the model during training to select a model with better performance; while the test set of the two-classification is relatively small and is not segmented again.

[0070] 2. Experimental results and comparative experiments

[0071] 1) Comparative experiment on the number of channels

[0072] The original track inspection data mainly contains 9 channels, namely: left alignment, right alignment, left verticality, right verticality, gauge, superelevation, level, twist, and speed. Among them, the speed and superelevation signals have no direct connection with the influence mechanism of interference factors. After removing these two channels, there are still 7 channels left. Through a large number of observations and analyses of various types of datasets, it is found that for rain interference, sunlight interference, rail grinding, and laser faults, the signal channels affected are generally the left and right alignments and the gauge; while for displacement gauge faults, the signal channels affected are generally the left and right verticalities and the gauge.

[0073] Specifically, experiments are conducted to verify the above analysis results. Keeping the model structure unchanged, experiments are respectively carried out on the five-classification and two-classification datasets with 7 input channels and 3 input channels. The experimental results are shown in Table 1.1. When training on the five-classification dataset using three channels, the accuracy of the obtained model can reach 98.86%, which is 0.47% higher than the 98.39% accuracy obtained when training on the five-classification dataset using seven channels; when training on the two-classification dataset using three channels, the accuracy of the obtained model can reach 99.77%, which is 0.23% higher than the 99.54% accuracy obtained when training on the two-classification dataset using seven channels.

[0074] Table 1.1 Comparative experiment on input data with different numbers of channels

[0075]

[0076] 2) Comparative experiments of related methods

[0077] The detection models in related directions were selected and trained on the same track inspection dataset, and the experimental results were compared. The comparison results are shown in Table 1.2.

[0078] Table 1.2 Comparison of the effects of related methods on the five-class dataset

[0079]

[0080] In the comparative experiment, the experimental equipment remained unchanged; the five-class track inspection dataset used during the training of the recognition network of the present invention was selected as the dataset, with 8393 samples in the training set, 628 samples in the validation set, and 1471 samples in the test set; the Adam optimizer was uniformly adopted as the optimizer; the Focal Loss for sample imbalance was uniformly adopted as the loss function. Since the learning rate, number of iterations (Epoch), and batch size required for each model to achieve the best effect are different, these three hyperparameters were determined through experiments. Among them, the learning rate of the LSTM model and the Bi-LSTM model was set to 0.001, the batch size was set to 64, and the number of iterations was set to 450 Epochs, and the learning rate was reduced to 10% of the initial learning rate at the 445th Epoch; the learning rate of the ACNN-FD and Res-1D-Net models was set to 0.001, the batch size was set to 64, and the number of iterations was set to 1000 Epochs, and the learning rate was reduced to 10% of the initial learning rate at the 950th Epoch; the learning rate of the proposed RA-Net model in this paper was set to 0.00001, the batch size was set to 32, and the number of iterations was set to 1000 Epochs, and the learning rate was reduced to 10% of the initial learning rate at the 950th Epoch.

[0081] The experimental results were analyzed. Compared with the Long Short-Term Memory networks (LSTM), all evaluation indicators of the Bi-directional Long Short-Term Memory networks (Bi-LSTM) were improved. The accuracy of the ACNN-FD network was lower than that of the Bi-LSTM, but the average recall rate, average F1 score, and accuracy were significantly improved; the proposed RA-Net model in this paper achieved a leading effect in the calculation of all evaluation indicators.

[0082] It should be noted that in this text, 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 actual relationship or order between these entities or operations. Moreover, the term "comprising", "including" or any other variant is intended to cover non-exclusive inclusion, so that an article or device comprising a series of elements includes not only those elements but also other elements not expressly listed. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the article or device comprising said element. Similar words such as "connected" or "coupled" are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect.

[0083] The above content is a further detailed description of the present invention in combination with specific preferred embodiments, and it cannot be determined that the specific implementation of the present invention is only limited to these descriptions. For those of ordinary skill in the technical field to which the present invention pertains, without departing from the concept of the present invention, several simple deductions or substitutions can be made, and all should be regarded as belonging to the protection scope of the present invention.

Claims

1. A method for identifying the characteristics of interference factors in track inspection data, characterized in that Including: Step 1: Obtain track inspection data to be measured; Step 2: Determine whether the interference category of the track inspection data to be measured is a track gauge machine failure through pattern recognition; If not, input the track inspection data to be measured into the trained first recognition network and the second recognition network respectively to obtain corresponding first recognition results and second recognition results, including: input the left alignment channel data, right alignment channel data and track gauge channel data of the track inspection data to be measured into the trained first recognition network to obtain the first recognition result; input the left elevation channel data, right elevation channel data and track gauge channel data of the track inspection data to be measured into the trained second recognition network to obtain the second recognition result; Step 3: Determine whether there is interference and the interference category of the track inspection data to be measured according to the first recognition result and the second recognition result; Among them, both the first recognition network and the second recognition network are deep learning networks based on channel attention. The deep learning network based on channel attention includes a cascaded feature extraction module and a classification module. The feature extraction module includes a cascaded plurality of feature extraction units, and each feature extraction unit includes a cascaded plurality of convolutional units and a channel attention module.

2. The method for identifying the characteristics of interference factors in track inspection data according to claim 1, wherein The feature extraction module includes 5 cascaded feature extraction units. The feature extraction unit includes 2 cascaded convolutional units and a channel attention module; each convolutional unit includes a cascaded convolutional layer and a Leaky ReLU activation function layer; The channel attention module includes a cascaded global average pooling layer, a fully connected layer, a Leaky ReLU activation function layer and a fully connected layer; The channel attention module obtains an attention vector according to the feature matrix extracted by the convolutional unit connected to it, and multiplies the attention vector and the input feature matrix of the channel attention module and uses the result as the output feature matrix of the feature extraction unit.

3. The method for identifying the characteristics of track inspection data interference factors according to claim 2, wherein The convolutional kernel size of the convolutional layer of the convolutional unit in the first feature extraction unit is 1×7, and the convolutional kernel size of the convolutional layer of the convolutional unit in the remaining feature extraction units is 1×5.

4. The method for identifying the characteristics of track inspection data interference factors according to claim 1, wherein The classification module includes a cascaded Dropout layer, a global average pooling layer, a fully connected layer and a Softmax function layer, where the dropout probability of the Dropout layer is 0.

5.

5. The method for identifying the characteristics of track inspection data interference factors according to claim 1, characterized in that, The interference categories of track inspection data include: rain interference, sunlight interference, rail grinding, laser failure, track gauge machine failure, displacement gauge failure and ice and snow interference.

6. The method for identifying the characteristics of track inspection data interference factors according to claim 1, wherein The track inspection data to be measured is detected by a GJ-6 type track inspection system. The track inspection data to be measured includes 7-channel data of left elevation, right elevation, left alignment, right alignment, track gauge, level and cross level.

7. The method for identifying the characteristics of track inspection data interference factors according to claim 6, wherein In the step 2, if the track gauge channel data of the track inspection data to be measured are all 0, it is determined that the interference category of the track inspection data to be measured is a track gauge machine failure.

8. The method for identifying the characteristics of interference factors of track inspection data according to claim 1, wherein The first recognition result is that the track inspection data to be measured is normal data, or the interference category of the track inspection data to be measured is rain and snow interference, sunlight interference, rail grinding or laser failure; The second recognition result is that the track inspection data to be measured is normal data, or the interference category of the track inspection data to be measured is a displacement gauge failure.

9. The method for identifying the characteristics of track inspection data interference factors according to claim 8, wherein, The step 3 includes: If both the first recognition result and the second recognition result are that the track inspection data to be measured is normal data, it is determined that there is no interference in the track inspection data to be measured; Otherwise, it is determined that there is interference in the track inspection data to be measured, and the interference category is the interference category corresponding to the recognition result; Wherein, if the interference category of the track inspection data to be measured is rain and snow interference, it is determined that the interference category is rain interference or ice and snow interference according to the geographical location and the season of the predicted section.

Images