Train information acquisition method and device, computer equipment and storage medium

Abstract

The invention relates to a train information acquisition method and device, computer equipment and a storage medium. The method comprises the following steps: acquiring a target railway monitoring image in which a train is shot from an initial railway monitoring image; inputting the target railway monitoring image into a pre-trained train number identification model, and identifying a train number information image area containing train number information of the train from the target railway monitoring image through a train number detection network in the train number identification model; inputting the train number information image area into a train number identification network in a train number identification model, and outputting train number information of the train through the train number identification network; and generating train information of the train by using the train number information. According to the method, vehicle number information identification can be realized through the vehicle number detection network and the vehicle number identification network, so that the accuracy and the identification efficiency of vehicle number information identification are improved, and the intelligence of vehicle number identification is improved.

Description

Technical Field

[0001] The present application relates to the field of railway technology, and in particular to a train information collection method, apparatus, computer equipment, storage medium, and computer program product. Background Art

[0002] With the development of railway technology, a technology for collecting train information has emerged. This technology relies on the recognition of train number information. By identifying the train number, the train information can be aggregated using the train number.

[0003] Traditionally, train car numbers can be identified through RFID systems. However, RFID systems require trains to be equipped with electronic tags, which are susceptible to interference from the surrounding electromagnetic environment and vibrations from the train. Long-term use in harsh environments can easily damage and cause loss, leading to missed reports. Furthermore, if the electronic tag is damaged, its information may be incomplete, and the car label may not match the actual car number.

[0004] In addition, train number recognition can also be solved through methods such as image processing, machine learning, and deep learning. However, image processing is easily affected by complex lighting and angle changes, and cannot effectively identify train numbers in harsh environments and complex backgrounds. Machine learning is only effective in identifying train numbers at low speeds or when stationary, and is less effective in tasks involving moving train numbers with limited samples. Deep learning, on the other hand, requires a large number of samples and higher hardware configuration to train the model. Therefore, existing train information collection methods are not intelligent enough for train number recognition. Summary of the Invention

[0005] Based on this, it is necessary to provide a train information collection method, device, computer equipment, computer-readable storage medium and computer program product that can improve the intelligence level of train number recognition in response to the above technical problems.

[0006] In a first aspect, the present application provides a train information collection method, comprising:

[0007] Acquire a target railway monitoring image in which a train is captured from the initial railway monitoring image;

[0008] Inputting the target railway monitoring image into a pre-trained train number recognition model, and identifying a train number information image region containing the train number information from the target railway monitoring image through a train number detection network in the train number recognition model;

[0009] Inputting the vehicle number information image area into the vehicle number recognition network in the train vehicle number recognition model, and outputting the vehicle number information of the train through the vehicle number recognition network;

[0010] The train information of the train is generated using the train number information.

[0011] In one embodiment, the vehicle number detection network in the train vehicle number recognition model is used to identify the vehicle number information image area containing the vehicle number information of the train from the target railway monitoring image, including: obtaining a feature map corresponding to the target railway monitoring image through a first feature extraction module in the vehicle number detection network; inputting the feature map into a differentiable prediction head in the vehicle number detection network, and obtaining a probability map and a dynamic threshold map corresponding to the feature map through the differentiable prediction head; wherein the probability map is used to characterize the probability that each pixel in the feature map belongs to the vehicle number information image area, and the dynamic threshold map is used to characterize the probability threshold of each pixel in the feature map corresponding to the vehicle number information image area; based on the probability map and the dynamic threshold map, a differentiable binary map corresponding to the feature map is generated, and the vehicle number information image area is obtained based on the differentiable binary map; the differentiable binary map is used to characterize the prediction result of whether each pixel in the feature map belongs to the vehicle number information image area.

[0012] In one embodiment, obtaining the feature map corresponding to the target railway monitoring image through the first feature extraction module in the vehicle number detection network includes: performing multi-scale feature extraction on the target railway monitoring image through the ResNet50 backbone network in the first feature extraction module to obtain the multi-scale feature map corresponding to the target railway monitoring image; performing multi-scale feature fusion on the multi-scale feature map through each feature fusion layer of the pyramid structure in the first feature extraction module to obtain a fused feature map corresponding to each feature fusion layer; upsampling and cascading operations on each of the fused feature maps to obtain a feature map corresponding to the target railway monitoring image.

[0013] In one embodiment, the outputting of the train number information through the train number recognition network includes: obtaining a multidimensional feature map corresponding to the train number information image area through a second feature extraction module in the train number recognition network; inputting the multidimensional feature map into a head network in the train number recognition network, and obtaining the train number character sequence features through a CTC head and a SAR head included in the head network; and outputting the train number information based on the train number character sequence features.

[0014] In one embodiment, the multidimensional feature map corresponding to the vehicle number information image area is obtained through the second feature extraction module in the vehicle number recognition network, including: through the MobileNetv2 network in the second feature extraction module, through the standard convolutional network and the depth-separable convolutional network of the MobileNetv2 network, to obtain the original feature map corresponding to the vehicle number information image area; using the SE module of the MobileNetv2 network to perform secondary feature calibration on the original feature map to obtain an enhanced feature map; inputting the enhanced feature map into the pooling layer in the second feature extraction module to obtain the multidimensional feature map corresponding to the vehicle number information image area.

[0015] In one embodiment, the initial railway monitoring image is a railway monitoring image taken according to a preset period; obtaining a target railway monitoring image in which a train is captured from the initial railway monitoring image includes: inputting the initial railway monitoring image into a pre-trained target detection network to obtain a train detection result corresponding to the initial railway monitoring image; and using the initial railway monitoring image as the target railway monitoring image when the train detection result indicates that the initial railway monitoring image contains a train.

[0016] In one embodiment, the method of generating the train information of the train using the vehicle number information includes: performing vehicle number screening on the vehicle number information using a regularized expression, and performing vehicle number correction processing on the vehicle number information that passes the vehicle number screening to obtain corrected vehicle number information; obtaining the speed information and the running track status information of the train based on the target railway monitoring image; and integrating the corrected vehicle number information, the speed information and the running track status information as the train information of the train.

[0017] In a second aspect, the present application further provides a train information collection device, comprising:

[0018] A monitoring image acquisition module is used to acquire a target railway monitoring image in which a train is captured from the initial railway monitoring image;

[0019] a vehicle number region detection module, configured to input the target railway monitoring image into a pre-trained train vehicle number recognition model, and identify the vehicle number information image region containing the vehicle number information of the train from the target railway monitoring image through the vehicle number detection network in the train vehicle number recognition model;

[0020] a vehicle number information recognition module, configured to input the vehicle number information image area into a vehicle number recognition network in the train vehicle number recognition model, and output the vehicle number information of the train through the vehicle number recognition network;

[0021] The train information collection module is used to generate the train information of the train using the train number information.

[0022] In a third aspect, the present application further provides a computer device comprising a memory and a processor, wherein the memory stores a computer program, and the processor implements the steps of the method described in any one of the embodiments of the first aspect when executing the computer program.

[0023] In a fourth aspect, the present application further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the method described in any one of the embodiments of the first aspect.

[0024] In a fifth aspect, the present application also provides a computer program product, comprising a computer program, which, when executed by a processor, implements the steps of the method described in any one of the embodiments of the first aspect.

[0025] The above-mentioned train information collection method, device, computer equipment, storage medium and computer program product obtain a target railway monitoring image with a train captured from an initial railway monitoring image; input the target railway monitoring image into a pre-trained train car number recognition model, and identify the car number information image area containing the train car number information from the target railway monitoring image through the car number detection network in the train car number recognition model; input the car number information image area into the car number recognition network in the train car number recognition model, and output the train car number information through the car number recognition network; and generate train information of the train using the car number information. The present application collects and captures target railway monitoring images of trains, and inputs the target railway monitoring images into a pre-trained train car number recognition model. The model may include a car number detection network and a car number recognition network, wherein the car number detection network can identify the car number information image area containing the train car number information, and then the car number recognition network can identify the train car number information from the car number information image area to generate train information. In this way, car number information recognition can be realized through the car number detection network and the car number recognition network, thereby improving the accuracy and efficiency of car number information recognition, and thus improving the intelligence of car number recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] In order to more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following briefly introduces the drawings required for use in the embodiments or related technical descriptions. Obviously, the drawings described below are only some embodiments of the present application. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0027] Figure 1 1 is a flow chart of a train information collection method according to an embodiment;

[0028] Figure 2 A schematic diagram of a process for identifying a vehicle license plate information image area in one embodiment;

[0029] Figure 3 Schematic diagram of a process for obtaining a feature map corresponding to a target railway monitoring image in one embodiment;

[0030] Figure 4 1. A schematic diagram of a process for outputting train number information in one embodiment;

[0031] Figure 5 A schematic diagram of a process for obtaining a feature map corresponding to a vehicle license plate information image region in one embodiment;

[0032] Figure 6 A schematic diagram of a process for generating train information of a train in one embodiment;

[0033] Figure 7 This is a technical flow chart of a system for collecting information on a trackside of a freight railway train in one embodiment;

[0034] Figure 8 A schematic diagram of a network framework of a freight train vehicle number detection and recognition algorithm in one embodiment;

[0035] Figure 9 This is a flowchart of a system for collecting information on a freight railway trackside in one embodiment;

[0036] Figure 10 This is a structural block diagram of a train information collection device in one embodiment;

[0037] Figure 11 FIG. 1 is a diagram showing the internal structure of a computer device in one embodiment. DETAILED DESCRIPTION

[0038] In order to make the purpose, technical solutions and advantages of this application more clear, the following further describes this application in detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain this application and are not intended to limit this application.

[0039] In one embodiment, Figure 1 As shown, a train information collection method is provided. This embodiment uses the method applied to a terminal as an example for illustration. It is understandable that the method can also be applied to a server, or to a system including a terminal and a server, and implemented through interaction between the terminal and the server. In this embodiment, the method includes the following steps:

[0040] Step S101 : acquiring a target railway monitoring image in which a train is captured from an initial railway monitoring image.

[0041] The initial railway monitoring image may be a railway monitoring image captured by a railway monitoring device, and the target railway monitoring image refers to a railway monitoring image in which a train passes through the initial railway monitoring image. Specifically, the terminal may capture the initial railway monitoring image captured by the railway monitoring device, and then perform image screening on the initial railway monitoring image to select the target railway monitoring image in which a train passes through.

[0042] Step S102: Input the target railway monitoring image into a pre-trained train number recognition model, and identify the train number information image area containing the train number information from the target railway monitoring image through the train number detection network in the train number recognition model.

[0043] Among them, the train number recognition model refers to a neural network model used to identify train number information. The model can be pre-trained and can include two parts: a number detection network and a number recognition network. The number information image area refers to the image area part where the number information is recorded in the target railway monitoring image. This part can be output by the number detection network, and the number detection network can be a text area recognition network.

[0044] Specifically, after the terminal obtains the target railway monitoring image, it can input the target railway monitoring image into a pre-trained train number recognition model, and through the train number detection network in the train number recognition model, output the image area in the target railway monitoring image that records the train number information, that is, output the train number information image area.

[0045] Step S103: input the vehicle number information image area into the vehicle number recognition network in the train vehicle number recognition model, and output the train vehicle number information through the vehicle number recognition network.

[0046] The vehicle number recognition network is a neural network model used to identify vehicle number information. The network can be a text information recognition network. Specifically, after the vehicle number information image area is obtained through the vehicle number detection network output, the vehicle number information image area can be further input into the vehicle number recognition network in the train vehicle number recognition model. The network can identify the train number from the vehicle number information image area containing the vehicle number information.

[0047] Step S104: Generate train information of the train using the train number information.

[0048] After obtaining the train number information, the terminal can use the above train number information to generate the train information of the train, thereby completing the collection of train information.

[0049] In the above-mentioned train information collection method, a target railway monitoring image with a train captured is obtained from an initial railway monitoring image; the target railway monitoring image is input into a pre-trained train vehicle number recognition model, and a vehicle number detection network in the train vehicle number recognition model is used to identify a vehicle number information image area containing the train vehicle number information from the target railway monitoring image; the vehicle number information image area is input into a vehicle number recognition network in the train vehicle number recognition model, and the vehicle number information of the train is output through the vehicle number recognition network; and the train information of the train is generated using the vehicle number information. The present application collects a target railway monitoring image with a train captured, and inputs the target railway monitoring image into a pre-trained train vehicle number recognition model, which may include a vehicle number detection network and a vehicle number recognition network, wherein the vehicle number detection network may identify a vehicle number information image area containing the train vehicle number information, and then the vehicle number recognition network may identify the train vehicle number information from the vehicle number information image area, thereby generating train information. In this way, vehicle number information recognition can be achieved through the vehicle number detection network and the vehicle number recognition network, thereby improving the accuracy and efficiency of vehicle number information recognition, and further improving the intelligence of vehicle number recognition.

[0050] In one embodiment, Figure 2 As shown, step S102 may further include:

[0051] Step S201: Obtain a feature map corresponding to the target railway monitoring image through the first feature extraction module in the vehicle number detection network.

[0052] The first feature extraction module refers to a feature extraction module included in the vehicle number detection network, which can be used to extract feature information corresponding to the target railway monitoring image, thereby obtaining a feature map corresponding to the target railway monitoring image.

[0053] Specifically, the terminal may first input the target railway monitoring image into the first feature extraction module in the vehicle number detection network, and the first feature extraction module extracts feature information corresponding to the target railway monitoring image, thereby outputting a feature map corresponding to the target railway monitoring image.

[0054] In step S202, the feature map is input into the differentiable prediction head in the vehicle license plate detection network, and a probability map and a dynamic threshold map corresponding to the feature map are obtained through the differentiable prediction head; wherein the probability map is used to represent the probability that each pixel in the feature map belongs to the vehicle license plate information image area, and the dynamic threshold map is used to represent the probability threshold of each pixel in the feature map corresponding to the vehicle license plate information image area.

[0055] The differentiable prediction head refers to the prediction head of the differentiable network. The output of the prediction head can include two branches, namely a probability map and a dynamic threshold map. The probability map can be used to predict the probability of each pixel belonging to the text area, and the dynamic threshold map is formed according to the dynamic probability of each pixel. The dynamic probability refers to the dynamic probability threshold of the pixel belonging to the text area.

[0056] Specifically, after the terminal obtains the feature map of the target railway monitoring image through step S201, it can also input the feature map into the differentiable prediction head in the vehicle number detection network, and output the corresponding probability map and dynamic threshold map through the two branches of the differentiable prediction head respectively. The probability map is mainly used to characterize the probability that each pixel in the feature map belongs to the vehicle number information image area, while the dynamic threshold map is mainly used to characterize the probability threshold of each pixel in the feature map corresponding to the vehicle number information image area.

[0057] Step S203, based on the probability map and the dynamic threshold map, a differentiable binary map corresponding to the feature map is generated, and the vehicle license plate information image area is obtained based on the differentiable binary map; the differentiable binary map is used to represent the prediction result of whether each pixel in the feature map belongs to the vehicle license plate information image area.

[0058] The differentiable binary map can be used to determine whether the pixel is text or background, that is, it can be used to characterize whether each pixel in the feature map belongs to the vehicle license plate information image area. The differentiable binary map can be obtained by binarizing the predicted probability map and the threshold map. For example, it can be obtained by comparing each pixel in the probability map with the corresponding pixel in the threshold map. If the probability corresponding to a certain pixel in the probability map reaches the dynamic probability threshold corresponding to the pixel in the dynamic threshold map, then the pixel can be considered to belong to the vehicle license plate information image area, and the corresponding value of the differentiable binary map is 1. If the probability corresponding to a certain pixel in the probability map does not reach the dynamic probability threshold corresponding to the pixel in the dynamic threshold map, then the pixel can be considered not to belong to the vehicle license plate information image area, and the corresponding value of the differentiable binary map is 0. Therefore, the differentiable binary map can be used to characterize whether each pixel in the feature map belongs to the vehicle license plate information image area.

[0059] Specifically, after the differentiable prediction head outputs the probability map and the dynamic threshold map, the terminal can also use the probability map and the dynamic threshold map to generate the corresponding differentiable binary map. Since it represents whether each pixel in the feature map belongs to the vehicle license plate information image area, the terminal can then use the differentiable binary map to output the vehicle license plate information image area.

[0060] In this embodiment, the terminal can also extract the feature map through the feature extraction module of the vehicle license plate detection network, and use the differentiable prediction head in the vehicle license plate detection network to output the corresponding probability map and dynamic threshold map, thereby generating a differentiable binary map based on the probability map and the dynamic threshold map to output the vehicle license plate information image area. In this way, the efficiency and accuracy of the output of the vehicle license plate information image area can be improved.

[0061] Furthermore, if Figure 3 As shown, step S201 may further include:

[0062] Step S301: Perform multi-scale feature extraction on the target railway monitoring image through the ResNet50 backbone network in the first feature extraction module to obtain a multi-scale feature map corresponding to the target railway monitoring image.

[0063] In this embodiment, the first feature extraction module may include a ResNet50 backbone network, which can be used to implement multi-scale feature extraction. The terminal may first input the target railway monitoring image into the ResNet50 backbone network to implement multi-scale feature extraction, thereby obtaining a multi-scale feature map corresponding to the target railway monitoring image.

[0064] Step S302 : performing multi-scale feature fusion on the multi-scale feature map through each feature fusion layer of the pyramid structure in the first feature extraction module to obtain a fused feature map corresponding to each feature fusion layer.

[0065] At the same time, the first feature extraction module may also include a feature pyramid structure, which may include multiple feature fusion layers of different scales, which are used to perform feature fusion of different scales on the multi-scale feature map, and the fused feature map is a feature map obtained by fusing the multi-scale feature maps output by different feature fusion layers in the pyramid structure.

[0066] Specifically, the terminal can input the multi-scale feature map obtained by the output of the ResNet50 backbone network into the pyramid structure in the first feature extraction module, and then perform multi-scale feature fusion through the pyramid structure to enhance the model's adaptability to text-sized polygons, thereby obtaining a fused feature map corresponding to each feature fusion layer.

[0067] Step S303 : performing upsampling and cascade operation processing on each fused feature map to obtain a feature map corresponding to the target railway monitoring image.

[0068] After obtaining each fused feature map, each fused feature map can be further upsampled to output each fused feature map to the same size, and then the upsampled feature maps are cascaded to finally generate a feature map corresponding to the target railway monitoring image. The size of the feature map can be one-quarter of the original image size.

[0069] In this embodiment, a ResNet50 backbone network is used to extract multi-scale feature maps. Multi-scale feature fusion is then implemented using a pyramid structure to enhance the model's adaptability to text-sized polygons. The feature maps are then upsampled and output to the same size, followed by a cascade operation to generate a feature map that is one-quarter the size of the original image. This approach improves the accuracy of feature extraction during the vehicle license plate detection phase.

[0070] In one embodiment, Figure 4 As shown, step S103 may further include:

[0071] Step S401: obtaining a multi-dimensional feature map corresponding to the vehicle number information image area through the second feature extraction module in the vehicle number recognition network.

[0072] The second feature extraction module refers to the feature extraction module included in the vehicle number recognition network. The feature extraction module can be used to extract feature information corresponding to the vehicle number information image area, thereby obtaining a multi-dimensional feature map corresponding to the vehicle number information image area.

[0073] Specifically, the terminal can first input the vehicle number information image area into the second feature extraction module in the vehicle number recognition network, and the second feature extraction module extracts the feature information corresponding to the vehicle number information image area, thereby outputting a multidimensional feature map corresponding to the vehicle number information image area.

[0074] In step S402, the multidimensional feature graph is input into the head network in the vehicle number recognition network, and the train number character sequence features are obtained through the CTC head and SAR head included in the head network.

[0075] The head network in the vehicle license plate recognition network can be composed of a multi-head attention mechanism. The head network can include a CTC (Connectionist Temporal Classification) head and a SAR (Sequence-to-Sequence Attention-based Recognizer) head. The CTC head and the SAR head are used to output the vehicle license plate character sequence features. The CTC head can handle the problem of inconsistent input and output lengths and automatically align the input sequence with the target sequence. The SAR head uses an encoder-decoder structure and an attention mechanism to selectively focus on different parts of the input sequence during the decoding process to generate the output sequence.

[0076] Specifically, after the terminal extracts the multidimensional feature map corresponding to the vehicle number information image area through the second feature extraction module in the vehicle number recognition network, it can also input the multidimensional feature map into the head network in the vehicle number recognition network. The head network may include a CTC head and a SAR head, thereby obtaining the train number character sequence characteristics through the output of the CTC head and the SAR head.

[0077] Step S403: output the train number information based on the train number character sequence feature.

[0078] After obtaining the vehicle license plate character sequence features, the vehicle license plate recognition network can output the vehicle license plate information based on the vehicle license plate character sequence features. For example, the vehicle license plate character sequence features can be first input into the fully connected layer to realize feature mapping, and then the feature mapping results can be decoded through the decoders corresponding to CTC and SAR to output the train license plate information.

[0079] In this embodiment, the second feature extraction module in the vehicle number recognition network can also be used to extract a multidimensional feature map corresponding to the vehicle number information image area, and then the head network in the vehicle number recognition network is used to obtain the vehicle number character sequence features through the CTC head and SAR head in the head network, and then the vehicle number information is output based on the vehicle number character sequence features. In this way, the efficiency and accuracy of the vehicle number information output can be improved.

[0080] Further, if Figure 5 As shown, step S401 may further include:

[0081] Step S501: Obtain the original feature map corresponding to the vehicle license plate information image area through the MobileNetv2 network in the second feature extraction module, the standard convolutional network of the MobileNetv2 network, and the depthwise separable convolutional network.

[0082] In this embodiment, the second feature extraction module may include a MobileNetv2 network, that is, feature extraction is implemented through the MobileNetv2 network, and the MobileNetv2 network may include two feature extraction networks, namely a standard convolutional network and a depth-wise separable convolutional network.

[0083] Specifically, the terminal can first input the vehicle license plate information image area into the standard convolutional network for convolution operation, and then perform batch normalization operation. The processing result of the batch normalization operation is then input into the depthwise separable convolutional network. The network can first perform spatial filtering on each input channel independently through depthwise convolution, and then use point-by-point convolution to change the channel dimension, thereby obtaining the original feature map corresponding to the vehicle license plate information image area.

[0084] Step S502: Use the SE module of the MobileNetv2 network to perform secondary feature calibration on the original feature map to obtain an enhanced feature map.

[0085] The SE (Squeeze-and-Excitation) module refers to the compression and excitation network module. This module is a novel network structural unit that enhances the representation capability of the convolutional neural network (CNN) by adaptively recalibrating channel feature responses. In this embodiment, in addition to the standard convolutional network and the depthwise separable convolutional network, the MobileNetv2 network can also include an SE module. The SE module can be used for feature secondary calibration, improving the network's representation capability and performance by learning the importance of each channel, thereby effectively improving the effect of convolution.

[0086] Specifically, after obtaining the original feature map corresponding to the vehicle license plate information image area, the original feature map can also be input into the SE module of the MobileNetv2 network. The SE module performs secondary feature calibration on the original feature map to obtain a feature map after feature enhancement, that is, an enhanced feature map.

[0087] Step S503: Input the enhanced feature map into the pooling layer in the second feature extraction module to obtain a multi-dimensional feature map corresponding to the vehicle license plate information image area.

[0088] In this embodiment, the second feature extraction module may further include a pooling layer, which may be an average pooling layer. Specifically, after obtaining the enhanced feature map, the enhanced feature map may be input into the pooling layer in the second feature extraction module, and the pooling layer outputs the final feature map of the vehicle number information image area, that is, a multi-dimensional feature map corresponding to the vehicle number information image area is obtained.

[0089] In this embodiment, an enhanced feature map can also be obtained through the MobileNetv2 network in the second feature extraction module, where the MobileNetv2 network can include a standard convolutional network, a depth-wise separable convolutional network and an SE module. After obtaining the enhanced feature map, the final multi-dimensional feature map is obtained through the pooling layer in the second feature extraction module. In this way, the feature extraction of vehicle license plate information is realized by using the enhanced version of MobileNetv2, thereby improving the accuracy of vehicle license plate information recognition.

[0090] In one embodiment, the initial railway monitoring image is a railway monitoring image taken according to a preset period; step S101 may further include: inputting the initial railway monitoring image into a pre-trained target detection network to obtain a train detection result corresponding to the initial railway monitoring image; when the train detection result indicates that the initial railway monitoring image contains a train, using the initial railway monitoring image as the target railway monitoring image.

[0091] Among them, the target detection network refers to a neural network used to detect target objects, such as train objects. In order to reduce the long-term operating load of vehicle number detection and recognition and reduce unnecessary false alarms, in this embodiment, after the terminal collects the real-time monitoring image captured by the railway monitoring equipment, that is, the initial railway monitoring image, it can also filter the initial railway monitoring image to obtain the target railway monitoring image with the train captured, and the screening method is achieved through a pre-trained target detection network.

[0092] Specifically, the railway monitoring equipment can capture monitoring images according to a preset period to obtain an initial railway monitoring image, which can then be transmitted to the terminal. After obtaining the initial railway monitoring image, the terminal can input the initial railway monitoring image into a pre-trained target detection network. The target detection network can be trained using the YOLO series of target detection algorithms. Through the target detection network, the train detection result corresponding to the initial railway monitoring image can be output. The train detection result can indicate whether a train is captured in the initial railway monitoring image. Afterwards, if the train detection result indicates that a certain initial railway monitoring image contains a train, the terminal can use the initial railway monitoring image as a target railway monitoring image that contains a train.

[0093] In this embodiment, after the railway monitoring equipment obtains the initial railway monitoring image by shooting according to a preset period, the train detection result corresponding to the initial railway monitoring image can also be output through a pre-trained target detection network, so that the train detection result can be used to filter out the target railway monitoring image from the initial railway monitoring image for identifying the train number information. In this way, the long-term operation load of the train number detection and recognition can be reduced, and the occurrence of unnecessary false alarms can be reduced, thereby further improving the accuracy of the train number information recognition and reducing the burden of the train number information recognition.

[0094] In one embodiment, Figure 6 As shown, step S104 may further include:

[0095] Step S601: The vehicle number information is screened by using a regularized expression method, and the vehicle number information that passes the screening is corrected to obtain the corrected vehicle number information.

[0096] Corrected vehicle license plate information refers to vehicle license plate information that has undergone correction processing. The correction process can be roughly divided into two stages. First, the recognition results are screened, filtering out those that failed recognition and retaining the string results that meet the vehicle license plate characteristics. After the vehicle license plate screening process, the recognizable vehicle license plate is preserved to the greatest extent possible, thereby retaining the string results that meet the vehicle license plate characteristics. For example, the vehicle license plate information recognition results whose result length matches the correct vehicle license plate are retained as the vehicle license plate information that passed the vehicle license plate screening. The vehicle license plate information that passed the screening is then corrected to obtain the corrected vehicle license plate information, namely the corrected vehicle license plate information.

[0097] Specifically, after obtaining the vehicle number information, the terminal can first use regularized expression to filter the identified vehicle number information to retain the character string results that meet the vehicle number characteristics. Compared with the conditional statements of the character segment class to perform logical AND or NOT judgments to perform vehicle number screening, the vehicle number screening method in this embodiment uses a regularized expression method. This method can improve the program running speed and reduce other missing recognition results. After the completely correct and subsequently correctable formats are summarized through the regularized expression method, subsequent correction processing is performed to obtain the corrected vehicle number information.

[0098] Step S602: Acquire the train speed information and track status information based on the target railway monitoring image.

[0099] The train speed information refers to the train's speed, which can be estimated by measuring the speed of the target railway monitoring image. The track status information refers to the type of track the train is traveling on. For example, it can be used to determine whether the train is traveling on a light-load track or a heavy-load track. Specifically, after obtaining the target railway monitoring image, the terminal can also use the target railway monitoring image to analyze the train's motion trajectory and position changes in the video, thereby calculating the train's speed in real time and obtaining the speed information. It can also determine whether the train is traveling on a light-load track or a heavy-load track, that is, perform detection and classification of the near- and far-track driving conditions, thereby obtaining the train's track status information.

[0100] In step S603, the corrected vehicle number information, vehicle speed information, and track status information are integrated as the train information of the train.

[0101] After obtaining the corrected train number information, train speed information and track status information, the corrected train number information, speed information and track status information can be integrated as the train information of the train.

[0102] In this embodiment, after the vehicle number information is identified, the vehicle number information can be screened and corrected to obtain the corrected vehicle number information, and the train speed information and the running track status information can be further obtained based on the target railway monitoring image, so that the corrected vehicle number information, speed information and running track status information can be integrated as the train information of the train. In this way, the accuracy and completeness of the train information can be improved.

[0103] In one embodiment, a freight railway train information trackside collection system is also provided, which can be used to collect train information of freight railway trains. The system uses an intelligent camera to communicate with the information collection host to identify train information and realize the task of online detection and identification of train car numbers. Specifically, it is first necessary to build outdoor all-weather monitoring equipment, which mainly includes a dome camera and a front-end host. Then, based on the image collected by the dome camera, the target detection model is used to detect in real time whether there is a train passing in the image. When a train passes, the train image is saved. The saved train image is subjected to car number detection and identification at regular intervals. Finally, the detection results are appropriately screened and corrected to obtain the car number result that conforms to the correct train number format. Then, it is compared with the frequency table and uploaded to the cloud together with other train information to complete the car number identification task. The overall workflow can be as follows Figure 7 shown.

[0104] The algorithm for the freight train trackside information collection system in this embodiment consists of two parts: freight train positioning information calculation and vehicle number detection and recognition algorithm. These algorithms use machine learning and deep learning methods to complete the information collection and recognition tasks.

[0105] To calculate freight train positioning information, the YOLO series of object detection algorithms, combined with a dome camera, achieve real-time detection. This approach is driven by the unique characteristics of the railway environment. Trains operate on two tracks, one for the uplink and one for the downlink. Trains typically enter the station to load coal and then return after being fully loaded. Therefore, trains appear periodically, with each pass lasting approximately 2-4 minutes. Furthermore, the footage captures trains stopping and waiting. Therefore, using YOLO to detect the presence of trains in the image in real time before deciding whether to perform vehicle number recognition reduces the long-term operational burden of vehicle number detection and recognition, and reduces unnecessary false alarms. While the object detection algorithm can determine train speed and direction, vehicle number recognition requires text detection and recognition.

[0106] The freight train number detection and recognition algorithm is a two-stage text recognition algorithm. The specific network framework is as follows: Figure 8As shown in the figure, detection and recognition are separated into two independent tasks: the detection network obtains the coordinates of the text box region in the image, and the recognition model identifies the text box region image. Therefore, compared to single-stage text recognition, the two-stage method detects text boxes more accurately and achieves higher recognition accuracy.

[0107] Vehicle number detection stage. This embodiment utilizes the Differentiable Binarization text detection algorithm framework and its differentiable operator method to detect freight train vehicle numbers. The image is first fed into the ResNet50 backbone network for multi-scale feature extraction. The extracted features are then fed into a pyramid structure for multi-scale feature fusion to enhance the model's adaptability to text polygons of varying sizes. The feature map is then upsampled and output to a uniform size. A cascade operation then generates a feature map and feature layer that is a quarter of the original image. The prediction head of the differentiable network then generates a probability map and a threshold map for the model's predicted text regions. The probability map predicts the probability of each pixel belonging to a text region, while the threshold map forms a dynamic threshold map based on the dynamic probabilities of each pixel. Finally, the predicted probability map and threshold map are binarized to produce a differentiable binary map. Each pixel in the probability map is compared with the corresponding pixel in the threshold map to determine whether it is text or background. Finally, the differentiable binary map undergoes a series of post-processing operations to determine the text box position, completing text detection.

[0108] During the license plate recognition phase, this embodiment uses an enhanced MobileNetv2 network to extract features from the license plate area, improving the efficiency and speed of feature extraction. The enhanced MobileNetv2 network introduces depthwise separable convolutions and SE (Squeeze-and-Excitation) modules to enhance network performance and efficiency. This network architecture primarily consists of three parts: a standard convolutional network followed by batch normalization, followed by depthwise separable convolutions. This involves two steps: first, depthwise convolutions independently filter each input channel spatially, and then pointwise convolutions are used to transform the channel dimensions. This structure effectively reduces the number of parameters and computational overhead. The SE module is then optionally used for feature recalibration, improving the network's expressiveness and performance by learning the importance of each channel, effectively enhancing the convolution effect. The final convolution layer of the feature extraction network has a stride of 1 in height and 2 in width, and the final pooling layer uses average pooling. The head network consists of a multi-head attention mechanism, primarily consisting of a CTC (Connectionist Temporal Classification) head and a SAR (Sequence-to-Sequence Attention-based Recognizer) head. CTC can handle mismatches between input and output lengths, automatically aligning the input sequence with the target sequence. SAR, through an encoder-decoder structure and attention mechanism, selectively focuses on different parts of the input sequence during decoding to generate the output sequence. The loss function is designed using a weighted combination of CTC, SAR, and a loss function called GTC (Global Temporal Consistency) to ensure global temporal consistency.

[0109] Vehicle number correction stage. First, the recognition results are screened using a regularized expression method, filtering out the results that failed to be recognized and retaining the string results that meet the vehicle number characteristics to facilitate subsequent vehicle number correction. After the vehicle number screening process, the vehicle number retains the recognizable front vehicle number to the greatest extent possible. However, if the conditional statement of the character segment class is used to perform logical AND or NOT judgments, it will reduce the program running speed and easily miss other missing recognition results. Therefore, in the first screening step, the regularized expression method is used to summarize the completely correct and subsequently correctable formats. Here, the subsequently correctable format refers to the situation where the length of the recognition result is inconsistent with the correct vehicle number and there is confusion between the numbers and letters in the vehicle number, under the premise that all recognizable string features are not completely lost.

[0110] The overall workflow of the freight railway train information trackside collection system in this embodiment is as follows: Figure 9 As shown. The system starts from the program and performs the following steps:

[0111] Step 1: Real-time video target monitoring:

[0112] After the program is started, the first step is to monitor the target in real time through the dome camera's monitoring screen. In this stage, the YOLO target detection network is used to detect whether there is a train passing through the screen.

[0113] Step 2: Check if a train is passing by:

[0114] The system analyzes video frames in real time to determine if a train is passing. If no train is detected, the system continues monitoring the video and waits for the next detection cycle. If a train is detected (i.e., the model's confidence that the object is a train exceeds a set threshold), the system proceeds to the next step.

[0115] Step 3: Save the passing vehicle image:

[0116] When a train passes by, the system captures and saves multiple images of the train, which will serve as basic data for subsequent processing.

[0117] Step 4, calculation of vehicle speed and driving conditions:

[0118] While saving the images, the system estimates the train's speed and detects and classifies its near- and far-track conditions. By analyzing the train's trajectory and position changes in the video, the system can calculate the train's speed in real time and determine whether it is traveling on lightly loaded or heavily loaded tracks.

[0119] Step 5: Vehicle license plate area detection:

[0120] The system preprocesses multiple stored train images before feeding them into the text detection module. This module is responsible for locating the train number area within the image. Preprocessing steps include denoising, contrast enhancement, and image resizing to improve text detection accuracy.

[0121] Step 6: Vehicle license plate area recognition:

[0122] After detecting the vehicle license plate area, the system feeds the image of this area into the text recognition module for vehicle license plate recognition. This module uses a deep learning algorithm to identify the characters in the vehicle license plate and outputs a preliminary recognition result.

[0123] Step 7: Correct the recognition results:

[0124] After obtaining the preliminary recognition results, the system will make a preliminary judgment: if the license plate format is correct, a frequency comparison will be performed to output the result with the highest number of occurrences in the batch of recognition results. The frequency comparison mechanism is based on the recognition results of multiple images. By counting the license plates with the highest frequency, the final result is determined, thereby effectively reducing the impact of single-frame recognition errors and improving overall recognition accuracy. If there is an error in the initially recognized license plate number, the system will use the license plate number dictionary and license plate number correction mechanism to correct it. The license plate number dictionary contains all possible license plate number formats and contents. By comparing with the legal license plates in the dictionary, the system can correct erroneous characters in the recognition. The license plate number correction mechanism includes operations such as character replacement, character completion, and character rearrangement to ensure that the final output license plate number is accurate.

[0125] Step 8: Output the integrated train information:

[0126] After vehicle identification and correction is complete, the system integrates train speed, near- and far-track conditions, vehicle number, and vehicle model information and uploads it to the cloud. The cloud system, with its robust storage and computing capabilities, enables further analysis and processing of uploaded data, providing comprehensive and accurate data support for railway operations management. This information not only monitors train operating status in real time but also provides valuable insights for subsequent dispatch optimization and safety management.

[0127] Step 9, the program ends:

[0128] After all data has been processed, the program returns to the monitoring state and continues the next round of detection cycle, ensuring efficient and accurate detection and identification of each passing train.

[0129] This embodiment utilizes the YOLOv5 object detection model for real-time train detection and speed estimation, and improves the model's generalization through data augmentation strategies. This significantly improves the real-time and accuracy of train detection and recognition, enabling adaptation to complex railway environments. In terms of vehicle number detection, the differentiable detection network enables focus on the vehicle number region of interest, improving the accuracy of vehicle number region detection. Furthermore, by combining the SVTR algorithm with an enhanced version of MobileNetv2, this embodiment significantly improves the speed and effectiveness of feature extraction in vehicle number recognition. Through a dual-stage information collection and vehicle number correction mechanism, the accuracy and robustness of vehicle number recognition are further enhanced, addressing environmental issues such as complex lighting and angle variations, and significantly improving the overall system performance. Therefore, this embodiment not only improves the real-time and accuracy of detection and recognition for train information collection and vehicle number recognition, but also enhances the system's adaptability and robustness in complex environments, providing more reliable and efficient technical support for railway operations management.

[0130] It should be understood that, although the various steps in the flowcharts involved in the various embodiments described above are displayed in sequence according to the instructions of the arrows, these steps are not necessarily executed in sequence in the order indicated by the arrows. Unless otherwise specified herein, there is no strict order restriction on the execution of these steps, and these steps can be executed in other orders. Moreover, at least a portion of the steps in the flowcharts involved in the various embodiments described above can include multiple steps or multiple stages, and these steps or stages are not necessarily executed and completed at the same time, but can be executed at different times, and the execution order of these steps or stages is not necessarily to be carried out in sequence, but can be executed in turn or alternately with other steps or at least a portion of steps or stages in other steps.

[0131] Based on the same inventive concept, embodiments of the present application also provide a train information collection device for implementing the aforementioned train information collection method. The solution provided by this device is similar to the solution described in the aforementioned method. Therefore, the specific limitations of one or more train information collection device embodiments provided below can be found in the aforementioned limitations of the train information collection method and will not be further elaborated here.

[0132] In one embodiment, Figure 10 As shown, a train information collection device is provided, comprising: a monitoring image acquisition module 1001, a vehicle number area detection module 1002, a vehicle number information recognition module 1003 and a train information collection module 1004, wherein:

[0133] The monitoring image acquisition module 1001 is used to acquire a target railway monitoring image with a train captured therein from the initial railway monitoring image;

[0134] The train number region detection module 1002 is configured to input the target railway monitoring image into a pre-trained train number recognition model, and identify the train number information image region containing the train number information from the target railway monitoring image through the train number detection network in the train number recognition model;

[0135] The vehicle number information recognition module 1003 is used to input the vehicle number information image area into the vehicle number recognition network in the train vehicle number recognition model, and output the train vehicle number information through the vehicle number recognition network;

[0136] The train information collection module 1004 is used to generate train information of the train using the train number information.

[0137] In one embodiment, the vehicle number area detection module 1002 is further used to obtain a feature map corresponding to the target railway monitoring image through the first feature extraction module in the vehicle number detection network; the feature map is input into the differentiable prediction head in the vehicle number detection network, and a probability map and a dynamic threshold map corresponding to the feature map are obtained through the differentiable prediction head; wherein the probability map is used to represent the probability that each pixel in the feature map belongs to the vehicle number information image area, and the dynamic threshold map is used to represent the probability threshold of each pixel in the feature map corresponding to the vehicle number information image area; based on the probability map and the dynamic threshold map, a differentiable binary map corresponding to the feature map is generated, and the vehicle number information image area is obtained based on the differentiable binary map; the differentiable binary map is used to represent the prediction result of whether each pixel in the feature map belongs to the vehicle number information image area.

[0138] In one embodiment, the vehicle number area detection module 1002 is further used to perform multi-scale feature extraction on the target railway monitoring image through the ResNet50 backbone network in the first feature extraction module to obtain a multi-scale feature map corresponding to the target railway monitoring image; perform multi-scale feature fusion on the multi-scale feature map through each feature fusion layer of the pyramid structure in the first feature extraction module to obtain a fusion feature map corresponding to each feature fusion layer; upsample and cascade operation are performed on each fusion feature map to obtain a feature map corresponding to the target railway monitoring image.

[0139] In one embodiment, the vehicle number information recognition module 1003 is further used to obtain a multidimensional feature map corresponding to the vehicle number information image area through the second feature extraction module in the vehicle number recognition network; input the multidimensional feature map into the head network in the vehicle number recognition network, and obtain the train vehicle number character sequence features through the CTC head and SAR head included in the head network; based on the vehicle number character sequence features, output the train vehicle number information.

[0140] In one embodiment, the vehicle license plate information recognition module 1003 is further used for a standard convolutional network and a depthwise separable convolutional network to obtain an original feature map corresponding to the vehicle license plate information image area; the SE module of the MobileNetv2 network is used to perform secondary feature calibration on the original feature map to obtain an enhanced feature map; the enhanced feature map is input into the pooling layer in the second feature extraction module to obtain a multidimensional feature map corresponding to the vehicle license plate information image area.

[0141] In one embodiment, the initial railway monitoring image is a railway monitoring image taken according to a preset period; the monitoring image acquisition module 1001 is further used to input the initial railway monitoring image into a pre-trained target detection network to obtain a train detection result corresponding to the initial railway monitoring image; when the train detection result indicates that the initial railway monitoring image contains a train, the initial railway monitoring image is used as the target railway monitoring image.

[0142] In one embodiment, the train information acquisition module 1004 is further used to perform vehicle number screening on the vehicle number information using a regularized expression method, and perform vehicle number correction processing on the vehicle number information that passes the vehicle number screening to obtain corrected vehicle number information; obtain the train speed information and the running track status information based on the target railway monitoring image; and integrate the corrected vehicle number information, speed information and running track status information as the train information of the train.

[0143] Each module in the aforementioned train information collection device may be implemented in whole or in part through software, hardware, or a combination thereof. Each module may be embedded in or independent of a processor within a computer device in the form of hardware, or may be stored in a memory within the computer device in the form of software, so that the processor can call and execute the corresponding operations of each module.

[0144] In one embodiment, a computer device is provided. The computer device may be a terminal, and its internal structure diagram may be as follows: Figure 11 As shown. The computer device includes a processor, memory, an input / output interface, a communication interface, a display unit, and an input device. The processor, memory, and input / output interface are connected via a system bus, and the communication interface, display unit, and input device are connected to the system bus via the input / output interface. The processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and internal memory. The non-volatile storage medium stores an operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs in the non-volatile storage medium. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals via wired or wireless means, and the wireless means can be achieved via Wi-Fi, mobile cellular networks, NFC (near-field communication), or other technologies. When executed by the processor, the computer program implements a train information collection method. The display unit of the computer device is used to form a visual image and can be a display screen, a projection device, or a virtual reality imaging device. The display screen can be a liquid crystal display screen or an electronic ink display screen, and the input device of the computer device can be a touch layer covering the display screen, or a button, trackball or touchpad set on the computer device casing, or an external keyboard, touchpad or mouse.

[0145] Those skilled in the art will understand that Figure 11 The structure shown in the figure is only a block diagram of a part of the structure related to the solution of the present application, and does not constitute a limitation on the computer device to which the solution of the present application is applied. The specific computer device may include more or fewer components than shown in the figure, or combine certain components, or have a different component arrangement.

[0146] In one embodiment, a computer device is further provided, including a memory and a processor. The memory stores a computer program, and the processor implements the steps in the above method embodiments when executing the computer program.

[0147] In one embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in the above-mentioned method embodiments are implemented.

[0148] In one embodiment, a computer program product is provided, including a computer program, which implements the steps in the above method embodiments when executed by a processor.

[0149] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use and processing of relevant data must comply with relevant regulations.

[0150] Those skilled in the art will appreciate that all or part of the processes in the above-mentioned embodiments can be implemented by instructing the relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the above-mentioned embodiments. In particular, any reference to memory, database, or other media used in the embodiments provided in this application can include at least one of non-volatile and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetic random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can take various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM). The databases involved in the various embodiments provided herein may include at least one of a relational database and a non-relational database. Non-relational databases may include, but are not limited to, distributed databases based on blockchains. The processors involved in the various embodiments provided herein may be, but are not limited to, general-purpose processors, central processing units (CPUs), graphics processing units (GPUs), digital signal processors (DSPs), programmable logic devices (PLDs), data processing logic devices based on quantum computing, and the like.

[0151] The technical features of the above embodiments can be combined arbitrarily. To make the description concise, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0152] The above-described embodiments merely represent several implementation methods of the present application. While the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the present application. It should be noted that a person of ordinary skill in the art may make various modifications and improvements without departing from the spirit of the present application, and these modifications and improvements fall within the scope of protection of the present application. Therefore, the scope of protection of the present application shall be determined by the appended claims.

Claims

1. A train information collection method, characterized in that: The method comprises: Acquire a target railway monitoring image in which a train is captured from the initial railway monitoring image; Inputting the target railway monitoring image into a pre-trained train number recognition model, and identifying a train number information image region containing the train number information from the target railway monitoring image through a train number detection network in the train number recognition model; Inputting the vehicle number information image area into the vehicle number recognition network in the train vehicle number recognition model, and outputting the vehicle number information of the train through the vehicle number recognition network; The train information of the train is generated using the train number information.

2. The method according to claim 1, characterized in that The step of identifying a vehicle number information image region containing the vehicle number information of the train from the target railway monitoring image through the vehicle number detection network in the train vehicle number recognition model includes: Obtaining a feature map corresponding to the target railway monitoring image through a first feature extraction module in the vehicle number detection network; Inputting the feature map into a differentiable prediction head in the vehicle number detection network, and obtaining a probability map and a dynamic threshold map corresponding to the feature map through the differentiable prediction head; wherein the probability map is used to represent the probability that each pixel in the feature map belongs to the vehicle number information image region, and the dynamic threshold map is used to represent the probability threshold of each pixel in the feature map corresponding to the vehicle number information image region; Based on the probability map and the dynamic threshold map, a differentiable binary map corresponding to the feature map is generated, and the vehicle license plate information image area is obtained based on the differentiable binary map; the differentiable binary map is used to represent the prediction result of whether each pixel in the feature map belongs to the vehicle license plate information image area.

3. The method according to claim 2, characterized in that The step of obtaining a feature map corresponding to the target railway monitoring image by using a first feature extraction module in the vehicle number detection network includes: Performing multi-scale feature extraction on the target railway monitoring image through the ResNet50 backbone network in the first feature extraction module, obtaining a multi-scale feature map corresponding to the target railway monitoring image; Performing multi-scale feature fusion on the multi-scale feature map through each feature fusion layer of the pyramid structure in the first feature extraction module to obtain a fused feature map corresponding to each feature fusion layer; Upsampling and cascade operation processing are performed on each of the fused feature maps to obtain a feature map corresponding to the target railway monitoring image.

4. The method according to claim 1, wherein Outputting the train number information through the train number recognition network includes: Obtaining a multi-dimensional feature map corresponding to the vehicle number information image area through a second feature extraction module in the vehicle number recognition network; Inputting the multidimensional feature graph into the head network in the vehicle number recognition network, and obtaining the vehicle number character sequence feature of the train through the CTC head and SAR head included in the head network; Based on the train number character sequence feature, the train number information of the train is output.

5. The method according to claim 4, characterized in that The second feature extraction module in the vehicle number recognition network is used to obtain a multi-dimensional feature map corresponding to the vehicle number information image area, including: Obtaining an original feature map corresponding to the vehicle license plate information image region through the MobileNetv2 network in the second feature extraction module, and through the standard convolutional network and the depthwise separable convolutional network of the MobileNetv2 network; Performing a secondary feature calibration on the original feature map using the SE module of the MobileNetv2 network to obtain an enhanced feature map; The enhanced feature map is input into the pooling layer in the second feature extraction module to obtain a multidimensional feature map corresponding to the vehicle license plate information image area.

6. The method according to claim 1, characterized in that The initial railway monitoring image is a railway monitoring image taken according to a preset period; The step of obtaining a target railway monitoring image in which a train is captured from the initial railway monitoring image includes: Inputting the initial railway monitoring image into a pre-trained object detection network to obtain a train detection result corresponding to the initial railway monitoring image; When the train detection result indicates that the initial railway monitoring image contains a train, the initial railway monitoring image is used as the target railway monitoring image.

7. The method according to any one of claims 1 to 6, characterized in that The generating the train information of the train by using the train number information includes: The vehicle number information is screened by using a regularized expression method, and the vehicle number information that passes the screening is corrected to obtain corrected vehicle number information; Acquiring the train speed information and track status information according to the target railway monitoring image; The corrected vehicle number information, the vehicle speed information and the travel track status information are integrated as the train information of the train.

8. A train information collection device, characterized in that: The device comprises: A monitoring image acquisition module is used to acquire a target railway monitoring image in which a train is captured from an initial railway monitoring image; a vehicle number region detection module, configured to input the target railway monitoring image into a pre-trained train vehicle number recognition model, and identify the vehicle number information image region containing the vehicle number information of the train from the target railway monitoring image through the vehicle number detection network in the train vehicle number recognition model; a vehicle number information recognition module, configured to input the vehicle number information image region into a vehicle number recognition network in the train vehicle number recognition model, and output the vehicle number information of the train through the vehicle number recognition network; The train information collection module is used to generate the train information of the train using the train number information.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, wherein: When the processor executes the computer program, the steps of the method according to any one of claims 1 to 7 are implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images