Intelligent identification method and system for railway tunnel face construction

Abstract

The invention provides an intelligent recognition method and system for railway tunnel face construction, and the method comprises the steps: constructing a training data set based on an image of a construction position collected by image collection equipment in history, and carrying out the pre-training of an intelligent recognition model through the training data set; an image, collected in real time, of the construction position is input into a pre-trained intelligent recognition model, the intelligent recognition model outputs an identification image, and an operation subject matter is identified in the identification image through a marking frame; on the basis of the recognition image output within a preset time period, the existence duration of the operation subject matter is calculated; and based on comparison between the existence duration and a preset threshold duration, whether footage exists in the construction position in a preset time period is judged. According to the scheme, a manager does not need to carry out real-time supervision on a construction site in real time, the footage condition of the tunnel face can be automatically identified, and the footage condition of construction can be objectively reflected.

Description

Technical Field

[0001] The present invention relates to the technical field of tunnel construction, and in particular to an intelligent identification method and system for railway tunnel face construction. Background Art

[0002] During the construction of railway tunnels, a massive and complex project, accurate identification and scientific management of the tunnel face are undoubtedly key elements in ensuring safety and high quality throughout the entire construction process. A tunnel face, a term widely known in the mining and tunneling fields, also known as the face of the subsurface, specifically refers to the working surface that continuously advances and is directly exposed to the surface or rock formation during underground operations such as coal mining, mining, or tunneling.

[0003] During railway tunnel construction, the condition of the tunnel face acts as a mirror, accurately and instantly reflecting the tunnel's progress. As the tunnel face progresses, the tunnel's outline gradually emerges, allowing for a quantifiable display of project progress. Furthermore, the stability of the tunnel face is directly linked to the mechanical state of the surrounding rock and serves as a crucial indicator for assessing tunnel structural safety. Signs of instability at the tunnel face, such as loosening of the surrounding rock and expansion of cracks, pose a direct threat to the safety of construction workers and the long-term stability of the tunnel. Furthermore, tunnel face management involves the selection and adjustment of construction methods. Different geological conditions, surrounding rock grades, and groundwater conditions all require the construction team to flexibly adopt appropriate excavation methods, support measures, and drainage schemes. Through careful observation and accurate identification of the tunnel face, the construction team can rapidly respond to geological changes, optimize construction plans, and ensure that construction safety and quality remain under constant control.

[0004] At railway tunnel construction sites, managers need to understand the construction status in real time to ensure progress and safety. Traditional methods mainly rely on manual inspections and records, which are not only inefficient but also highly subjective. Summary of the Invention

[0005] In view of this, an embodiment of the present invention provides an intelligent identification method for railway tunnel face construction to eliminate or improve one or more defects existing in the prior art.

[0006] One aspect of the present invention provides a method for intelligent identification of railway tunnel face construction, the method comprising the following steps:

[0007] Based on the images of construction locations collected historically by image acquisition equipment, a training dataset is constructed and used to pre-train the intelligent recognition model;

[0008] Inputting the real-time captured image of the construction location into a pre-trained intelligent recognition model, the intelligent recognition model outputting an identification image in which the work object is identified by a marking box;

[0009] Calculating the existence time of the operation target object based on the recognition image output within a preset time period;

[0010] Based on the comparison between the existence time and a preset threshold time, it is determined whether there is any progress at the construction location within the preset time period.

[0011] Using the above scheme, this scheme first constructs a training data set and pre-trains the intelligent recognition model. The pre-trained intelligent recognition model is used to identify the work targets in the real-time incoming images, and then the duration of the crop targets in the images over a period of time is counted to determine whether there is progress. This scheme does not require management personnel to conduct real-time supervision at the construction site. It can automatically identify the progress of the tunnel face and objectively reflect the progress of the construction.

[0012] In some embodiments of the present invention, if the work object is one, in the step of determining whether there is progress at the construction location within a preset time period based on comparing the existence duration with a preset threshold duration, if the existence duration is greater than the preset threshold duration, it is determined that there is progress at the construction location within the preset time period.

[0013] In some embodiments of the present invention, if there are multiple types of work objects, a threshold duration is set for each type of work object. In the step of determining whether there is progress at the construction location within a preset time period based on the comparison of the existence duration with the preset threshold duration, the existence duration of each crop object is compared with the corresponding threshold duration. If the existence duration of any crop object is greater than the corresponding threshold duration, it is determined whether there is progress at the construction location within the preset time period.

[0014] In some embodiments of the present invention, the operation target objects include people and slag trucks. In the step of calculating the existence duration of the operation target objects based on the recognition images output within a preset time period:

[0015] Extracting identification images containing people and slag trucks within the preset time period, and calculating the number of identification images containing people and slag trucks respectively;

[0016] Based on the total number of recognition images and the number of recognition images in which people and slag trucks are present within a preset time period, the presence time of the people and slag trucks is calculated respectively.

[0017] In some embodiments of the present invention, the sampling frequencies of the images in the preset time period are the same. In the step of respectively calculating the presence duration of the person and the slag truck based on the total number of recognized images and the number of recognized images containing the person and the slag truck in the preset time period:

[0018] Based on the total number of recognition images within a preset time period and the number of recognition images containing people and slag trucks, respectively, the ratio of the recognition images of people and slag trucks to the total number of recognition images within the preset time period is calculated;

[0019] The presence time of the person and the slag truck are calculated based on the total duration of the preset time period and the ratio value.

[0020] In some embodiments of the present invention, in the step of constructing a training data set based on images of construction locations historically captured by an image acquisition device, images are extracted from the video stream of the historically captured construction locations at preset time intervals to construct a training data set, and a marking box is used to mark the work objects in the image as label data.

[0021] In some embodiments of the present invention, in the step of pre-training the intelligent recognition model using the training data set, the loss function value is calculated using the following formula:

[0022]

[0023] Among them, x and y are the horizontal and vertical coordinates of the mark box output by the intelligent recognition model; w and h are the width and height of the mark box output by the intelligent recognition model; x gt and y gt are the horizontal and vertical coordinates of the marking box of the label data; W i and H i W is the width and height of the intersection of the marked box of the label data and the marked box output by the intelligent recognition model; g and H g They are the total width and total height of the overlapping marked boxes of the label data and the marked boxes output by the intelligent recognition model, L represents the loss function value, and w gt Indicates the width of the marker box of the label data, h gt The height of the marker box representing the label data.

[0024] In some embodiments of the present invention, the step of pre-training the intelligent recognition model using the training data set further includes:

[0025] Perform sparse training and pruning on intelligent recognition models;

[0026] Use the pruned model weights as pre-training weights W to generate the model weight matrix E mask , set the weights corresponding to the pruned channels to 0 and the rest to 1, W1 = E mask W, obtain the compressed intelligent recognition model W1, and send the compressed intelligent recognition model to the edge intelligent terminal at the railway tunnel construction site.

[0027] In some embodiments of the present invention, in the steps of performing sparse training and pruning on the intelligent recognition model, L1 sparsification is used to perform sparse training.

[0028] The second aspect of the present invention also provides an intelligent identification system for railway tunnel face construction, which includes a computer device, the computer device including a processor and a memory, the memory storing computer instructions, and the processor being used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps implemented by the method described above.

[0029] The third aspect of the present invention further provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps implemented by the aforementioned railway tunnel face construction intelligent identification method.

[0030] Additional advantages, objects, and features of the present invention will be described in part in the following description and will become apparent to those skilled in the art after studying the following or may be learned by practice of the present invention. The objects and other advantages of the present invention may be particularly pointed out and attained in the description and drawings.

[0031] Those skilled in the art will understand that the purposes and advantages that can be achieved by the present invention are not limited to the above specific descriptions, and the above and other purposes that can be achieved by the present invention will be more clearly understood based on the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0032] The drawings described herein are used to provide a further understanding of the present invention, constitute a part of this application, and do not constitute a limitation of the present invention.

[0033] Figure 1 This is a schematic diagram of an implementation method of the railway tunnel face construction intelligent identification method of the present invention. DETAILED DESCRIPTION

[0034] In order to make the purpose, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the embodiments and the accompanying drawings. Here, the exemplary embodiments of the present invention and their descriptions are used to explain the present invention, but are not intended to limit the present invention.

[0035] It should also be noted that, in order to avoid obscuring the present invention due to unnecessary details, the accompanying drawings only show structures and / or processing steps closely related to the solutions according to the present invention, while other details that are not closely related to the present invention are omitted.

[0036] like Figure 1As shown, the present invention proposes an intelligent identification method for railway tunnel face construction, the method comprising the following steps:

[0037] Step S100, based on images of construction locations historically captured by image acquisition equipment, a training data set is constructed, and the intelligent recognition model is pre-trained using the training data set;

[0038] During the specific implementation process, the image acquisition device is a camera, which is deployed on-site at construction locations in different regions.

[0039] During the specific implementation process, video frames are extracted at intervals of 10 seconds to obtain image sequences to form the basic training data set for the model. Slag truck images are screened from the public data set and integrated into the training data set to further expand the data set and improve its quality.

[0040] Step S200: inputting the real-time captured image of the construction location into a pre-trained intelligent recognition model, wherein the intelligent recognition model outputs a marking image, wherein the work object is marked by a marking frame;

[0041] In a specific implementation process, the step of inputting the real-time collected image of the construction location into the pre-trained intelligent recognition model further includes pre-processing the image, wherein the pre-processing includes size adjustment, noise removal and image enhancement;

[0042] In the resizing step, the image is scaled to a preset size;

[0043] In the step of noise removal, spatial domain filtering and / or frequency domain filtering are used;

[0044] Spatial filtering: Use filtering techniques such as median filtering or Gaussian filtering to remove noise from images. Median filtering is mainly used to remove salt and pepper noise, while Gaussian filtering is used to remove Gaussian noise.

[0045] Frequency domain filtering: Operations are performed in the frequency domain after Fourier transform, such as low-pass filters to eliminate high-frequency noise;

[0046] In the image enhancement step, a high-pass filter (such as the Laplacian operator and the Sobel edge detection operator) can be used to highlight the edge information in the image and enhance the details of the image.

[0047] Step S300, calculating the existence time of the operation target object based on the recognition image output within a preset time period;

[0048] During the specific implementation process, the original image collected within the preset time period is obtained by uniformly sampling using a preset sampling frequency, and a corresponding recognition image is obtained. Based on the proportion of the recognition image in which the operation object exists, the existence duration of the operation object is determined.

[0049] Step S400 : Based on the comparison between the existence duration and a preset threshold duration, it is determined whether there is any progress at the construction location within a preset time period.

[0050] In the specific implementation process, this solution includes a cloud model training module, a cloud model compression module, an edge computing module, a data statistics module, and a visualization module;

[0051] The cloud model training module is used to pre-train the intelligent recognition model using the training data set;

[0052] The cloud model compression module is used to perform sparse training, pruning and network fine-tuning on the intelligent recognition model;

[0053] The edge computing module processes:

[0054] Video acquisition: Edge video inference servers are deployed at tunnel construction sites A and B to read real-time video streams from the DVRs in the on-site video surveillance room.

[0055] Video Inference: A highly compressed face human and vehicle detection model is deployed on the video inference server to detect people and slag trucks in real time, screening videos with people and slag trucks for 10 consecutive seconds.

[0056] The processing of the data statistics module includes:

[0057] Build structured tables: Build a video stream table to record primary keys, creation time, etc.; a video file table to record video start and end times, etc.; a person identification table to record the number of people, video addresses, etc.; a slag truck identification table to record device type, video addresses, etc.

[0058] Statistics are performed based on the structured table to calculate the total time when people are present at the construction site, the number of times slag trucks operate, and the operation time. A threshold for the time when people are present and the operation time of slag trucks are set. If the total time when people are present is greater than or the operation time of slag trucks is greater than, the tunnel face is judged to have progress; otherwise, the tunnel face is judged to have no progress.

[0059] The visualization module compares the existence duration with a preset threshold duration to determine whether there is any progress at the construction location within a preset time period, and presents the progress through a display device.

[0060] In the specific implementation process, the visualization module includes a video display area and a statistical information display area;

[0061] Video display area: Based on the work point dimension, it displays the video analysis details of the tunnel face in different sections, including information such as section, work point, tunnel face, time without personnel, time without equipment, and video analysis results;

[0062] Statistical information display area: Based on the time dimension, the personnel hours and equipment hours of different sections are displayed by day, month and year.

[0063] Using the above scheme, this scheme first constructs a training data set and pre-trains the intelligent recognition model. The pre-trained intelligent recognition model is used to identify the work targets in the real-time incoming images, and then the duration of the crop targets in the images over a period of time is counted to determine whether there is progress. This scheme does not require management personnel to conduct real-time supervision at the construction site. It can automatically identify the progress of the tunnel face and objectively reflect the progress of the construction.

[0064] In some embodiments of the present invention, if the work object is one, in the step of determining whether there is progress at the construction location within a preset time period based on comparing the existence duration with a preset threshold duration, if the existence duration is greater than the preset threshold duration, it is determined that there is progress at the construction location within the preset time period.

[0065] Using the above scheme, the crop target is the equipment or workers used for tunnel face construction. If the operation target is of one type, the existence duration of this type of operation target is compared with the corresponding threshold duration. If it is greater than the threshold duration, it means that there is a certain operation time, and it is determined that there is progress.

[0066] In some embodiments of the present invention, if there are multiple types of work objects, a threshold duration is set for each type of work object. In the step of determining whether there is progress at the construction location within a preset time period based on the comparison of the existence duration with the preset threshold duration, the existence duration of each crop object is compared with the corresponding threshold duration. If the existence duration of any crop object is greater than the corresponding threshold duration, it is determined whether there is progress at the construction location within the preset time period.

[0067] In some embodiments of the present invention, the operation target objects include people and slag trucks. In the step of calculating the existence duration of the operation target objects based on the recognition images output within a preset time period:

[0068] Extracting identification images containing people and slag trucks within the preset time period, and calculating the number of identification images containing people and slag trucks respectively;

[0069] Based on the total number of recognition images and the number of recognition images in which people and slag trucks are present within a preset time period, the presence time of the people and slag trucks is calculated respectively.

[0070] In some embodiments of the present invention, the sampling frequencies of the images in the preset time period are the same. In the step of respectively calculating the presence duration of the person and the slag truck based on the total number of recognized images and the number of recognized images containing the person and the slag truck in the preset time period:

[0071] Based on the total number of recognition images within a preset time period and the number of recognition images containing people and slag trucks, respectively, the ratio of the recognition images of people and slag trucks to the total number of recognition images within the preset time period is calculated;

[0072] The presence time of the person and the slag truck are calculated based on the total duration of the preset time period and the ratio value.

[0073] In some embodiments of the present invention, in tunnel face video analysis tasks, the actual input is a video stream, and cameras are deployed at construction sites in different regions. Deploying the underlying algorithms required for image recognition applications in an edge computing environment requires not only consideration of detection accuracy, but also requires low latency and a small model size to ensure real-time detection of tunnel face video streams at the edge. Therefore, this patent selects a reasonable network model and a highly compatible model compression technology, fully considering the requirements of actual application scenarios and improving the reliability of engineering applications.

[0074] In some embodiments of the present invention, in the step of constructing a training data set based on images of construction locations historically captured by an image acquisition device, images are extracted from the video stream of the historically captured construction locations at preset time intervals to construct a training data set, and a marking box is used to mark the work objects in the image as label data.

[0075] In the specific implementation process, Labelimg is used to annotate the dataset, and the three categories of people, slag trucks, and construction equipment in the images are labeled as people, truck, or other respectively.

[0076] In some embodiments of the present invention, in the step of pre-training the intelligent recognition model using the training data set, the loss function value is calculated using the following formula:

[0077]

[0078] Among them, x and y are the horizontal and vertical coordinates of the mark box output by the intelligent recognition model; w and h are the width and height of the mark box output by the intelligent recognition model; x gt and y gt are the horizontal and vertical coordinates of the marking box of the label data; W i and H i W is the width and height of the intersection of the marked box of the label data and the marked box output by the intelligent recognition model;g and H g They are the total width and total height of the overlapping marked boxes of the label data and the marked boxes output by the intelligent recognition model, L represents the loss function value, and w gt Indicates the width of the marker box of the label data, h gt The height of the marker box representing the label data.

[0079] In a specific implementation process, the intelligent recognition model may be a YOLOv8 model, and an improved YOLOv8 is used to perform target detection model training on a labeled training set to obtain training weights.

[0080] Specifically, the YOLOv8 model combines computer vision with deep learning, achieving simultaneous prediction of classification and bounding boxes in a single pass of the neural network, greatly speeding up the process and improving real-time detection capabilities. YOLOv8 has been improved on the basis of YOLOv5, introducing features such as a new backbone network (a modified version of the CSPDarknet53 architecture), an anchor-free detection head, and multi-scale object detection. Its uniqueness lies in providing unparalleled speed and accuracy performance while maintaining a streamlined design, suitable for different applications and easy to adapt to various hardware platforms. In addition, YOLOv8 also supports a variety of computer vision tasks, such as instance segmentation and key point detection, and is widely used in security monitoring, autonomous driving, smart home, industrial automation and other fields. In general, YOLOv8 has demonstrated strong competitiveness in the field of computer vision with its efficiency, accuracy and versatility.

[0081] In a specific implementation process, the intelligent recognition model may also be a Faster R-CNN model and an SSD model;

[0082] Faster R-CNN is a highly efficient object detection model developed based on R-CNN and Fast R-CNN. By introducing the Region Proposal Network (RPN), this model significantly improves the speed and accuracy of object detection, becoming a benchmark method in modern object detection tasks. Faster R-CNN consists of two components, the RPN and Fast R-CNN, sharing convolutional features that make the entire detection process highly efficient. Although its computational and memory requirements are high, and its real-time performance is inferior to that of later models, its excellent accuracy has led to its widespread application in fields such as autonomous driving, video surveillance, and medical image analysis.

[0083] SSD (Single Shot MultiBox Detector) is a single-stage target detection model characterized by its ability to balance detection speed and accuracy. The SSD model extracts features through a convolutional neural network and makes predictions on feature maps of multiple scales to accommodate the detection of objects of different sizes. It uses an anchor mechanism, where each point on a feature map is used as the center point of a candidate region, directly regressing the target's category and position. SSD also introduces multiple prior boxes of different scales to cover targets of different sizes and aspect ratios, thereby improving detection accuracy. Due to its efficient design, the SSD model has achieved good performance in a variety of computer vision tasks such as target detection, pedestrian detection, and vehicle detection, and has broad application prospects.

[0084] In some embodiments of the present invention, the step of pre-training the intelligent recognition model using the training data set further includes:

[0085] Perform sparse training, pruning, and network fine-tuning on intelligent recognition models;

[0086] In the step of network fine-tuning, the pruned model weights are used as pre-training weights W to generate the model weight matrix E mask , set the weights corresponding to the pruned channels to 0 and the rest to 1, W1 = E mask W, obtain the compressed intelligent recognition model W1, and send the compressed intelligent recognition model to the edge intelligent terminal at the railway tunnel construction site.

[0087] In the specific implementation process, during the pruning process, a global pruning threshold t is set, and the γ factors are sorted from small to large. Channels smaller than the threshold t will be pruned to obtain a sparse and compact network.

[0088] In some embodiments of the present invention, in the steps of performing sparse training and pruning on the intelligent recognition model, L1 sparsification is used to perform sparse training.

[0089] In the specific implementation process, by adding L1 sparse regularization term It is added to the loss function to compete with the original loss function, penalizing the case where most parameters are not zero, and encouraging the model to learn sparse parameters during training. The sparse formula is:

[0090]

[0091] in, is the loss function of the target detection model; W is the training weight; x', y are the input and output of the detection target respectively; λ is the sparsity factor, and γ is the scaling factor corresponding to the channel of the convolutional layer.

[0092] In the specific implementation process, the benefits of sparse training are mainly reflected in:

[0093] Improve model generalization ability: By introducing sparsity constraints during training, the optimization algorithm will try to make parameters sparse while reducing conventional losses, preventing the model from relying too much on input features, thereby reducing model complexity and improving generalization ability.

[0094] Reduce model complexity: Sparse training can reduce the number of model parameters and computational complexity, making the model more concise.

[0095] The benefits of pruning include:

[0096] Reducing model parameters: Pruning reduces the number of model parameters by removing unimportant weights or neurons, thereby simplifying the model structure.

[0097] Reduced storage and computing requirements: Since the pruned model has fewer parameters, the required storage space and computing resources will also be reduced accordingly, which is especially important for deploying deep learning models in resource-constrained environments.

[0098] Improved inference speed: Pruning can reduce the amount of computation required by the model, thereby improving the model's inference speed, which is critical for application scenarios that require real-time response.

[0099] In summary, sparse training and pruning have significant advantages in improving model efficiency, reducing resource consumption, and enhancing model generalization capabilities.

[0100] An embodiment of the present invention also provides an intelligent identification system for railway tunnel face construction, which includes a computer device, wherein the computer device includes a processor and a memory, wherein the memory stores computer instructions, and the processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps implemented by the method described above.

[0101] An embodiment of the present invention further provides a computer-readable storage medium having a computer program stored thereon. When executed by a processor, the computer program implements the steps of the aforementioned railway tunnel face construction intelligent identification method. The computer-readable storage medium can be a tangible storage medium, such as a random access memory (RAM), internal memory, read-only memory (ROM), electrically programmable ROM, electrically erasable programmable ROM, register, floppy disk, hard disk, removable storage disk, CD-ROM, or any other form of storage medium known in the art.

[0102] It should be understood by those skilled in the art that the various exemplary components, systems and methods described in conjunction with the embodiments disclosed herein can be implemented in hardware, software or a combination of the two. Whether it is specifically performed in hardware or software depends on the specific application and design constraints of the technical solution. Professional and technical personnel can use different methods to implement the described functions for each specific application, but such implementation should not be considered to be beyond the scope of the present invention. When implemented in hardware, it can be, for example, an electronic circuit, an application specific integrated circuit (ASIC), appropriate firmware, a plug-in, a function card, etc. When implemented in software, the elements of the present invention are programs or code segments that are used to perform the required tasks. The program or code segment can be stored in a machine-readable medium, or transmitted on a transmission medium or a communication link via a data signal carried in a carrier.

[0103] It should be understood that the present invention is not limited to the specific configurations and processes described above and illustrated in the figures. For the sake of brevity, a detailed description of known methods is omitted. In the above embodiments, several specific steps are described and illustrated as examples. However, the method of the present invention is not limited to the specific steps described and illustrated. Those skilled in the art may make various changes, modifications, and additions, or change the order of the steps after understanding the spirit of the present invention.

[0104] In the present invention, features described and / or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, and / or combined with or replace features of other embodiments.

[0105] The foregoing description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Those skilled in the art will readily appreciate that various modifications and variations to the present invention are possible. Any modifications, equivalent substitutions, or improvements made within the spirit and principles of the present invention are intended to be within the scope of protection of the present invention.

Claims

1. A railway tunnel face construction intelligent identification method, characterized in that: The steps of the method include: Based on the images of construction locations collected historically by image acquisition equipment, a training dataset is constructed and used to pre-train the intelligent recognition model; Inputting the real-time captured image of the construction location into a pre-trained intelligent recognition model, the intelligent recognition model outputting an identification image in which the work object is identified by a marking box; Calculating the existence time of the operation target object based on the recognition image output within a preset time period; Based on the comparison between the existence time and a preset threshold time, it is determined whether there is any progress at the construction location within the preset time period.

2. The intelligent identification method for railway tunnel face construction according to claim 1 is characterized in that: If the work object is one type, in the step of determining whether there is progress at the construction location within a preset time period based on comparing the existence duration with a preset threshold duration, if the existence duration is greater than the preset threshold duration, it is determined that there is progress at the construction location within the preset time period.

3. The intelligent identification method for railway tunnel face construction according to claim 1 is characterized in that: If there are multiple types of work objects, a threshold duration is set for each type of work object. In the step of determining whether there is progress at the construction location within a preset time period based on the comparison of the existence duration with the preset threshold duration, the existence duration of each crop object is compared with the corresponding threshold duration. If the existence duration of any crop object is greater than the corresponding threshold duration, it is determined whether there is progress at the construction location within the preset time period.

4. The intelligent identification method for railway tunnel face construction according to claim 1 is characterized in that: The operation target objects include people and slag trucks. In the step of calculating the existence duration of the operation target objects based on the recognition images output within a preset time period: Extracting identification images containing people and slag trucks within the preset time period, and calculating the number of identification images containing people and slag trucks respectively; Based on the total number of recognition images and the number of recognition images in which people and slag trucks are present within a preset time period, the presence time of the people and slag trucks is calculated respectively.

5. The intelligent identification method for railway tunnel face construction according to claim 4 is characterized in that: The sampling frequency of the images in the preset time period is the same. In the step of respectively calculating the presence time of the person and the slag truck based on the total number of recognized images and the number of recognized images in which the person and the slag truck are present in the preset time period: Based on the total number of recognition images within a preset time period and the number of recognition images containing people and slag trucks, respectively, the ratio of the recognition images of people and slag trucks to the total number of recognition images within the preset time period is calculated; The presence time of the person and the slag truck are calculated based on the total duration of the preset time period and the ratio value.

6. The intelligent identification method for railway tunnel face construction according to any one of claims 1 to 5, characterized in that: In the step of constructing a training data set based on images of construction locations historically captured by image acquisition equipment, images are extracted from the video stream of the historically captured construction locations at preset time intervals to construct a training data set, and marking frames are used to mark the work objects in the images as label data.

7. The intelligent identification method for railway tunnel face construction according to claim 6, characterized in that: In the step of pre-training the intelligent recognition model using the training data set, the loss function value is calculated using the following formula: Among them, x and y are the horizontal and vertical coordinates of the mark box output by the intelligent recognition model; w and h are the width and height of the mark box output by the intelligent recognition model; x gt and y gt are the horizontal and vertical coordinates of the marking box of the label data; W i and H i W is the width and height of the intersection of the marked box of the label data and the marked box output by the intelligent recognition model; g and H g They are the total width and total height of the overlapping marked boxes of the label data and the marked boxes output by the intelligent recognition model, L represents the loss function value, and w gt Indicates the width of the marker box of the label data, h gt The height of the marker box representing the label data.

8. The intelligent identification method for railway tunnel face construction according to claim 7, characterized in that: The steps of pre-training the intelligent recognition model using the training data set also include: Perform sparse training and pruning on intelligent recognition models; Use the pruned model weights as pre-training weights W to generate the model weight matrix E mask , set the weights corresponding to the pruned channels to 0 and the rest to 1, W1 = E mask W, obtain the compressed intelligent recognition model W1, and send the compressed intelligent recognition model to the edge intelligent terminal at the railway tunnel construction site.

9. The intelligent identification method for railway tunnel face construction according to claim 8, characterized in that: In the steps of sparse training and pruning the intelligent recognition model, L1 sparsification is used for sparse training.

10. An intelligent identification system for railway tunnel face construction, characterized in that: The system includes a computer device, which includes a processor and a memory. The memory stores computer instructions. The processor is used to execute the computer instructions stored in the memory. When the computer instructions are executed by the processor, the system implements the steps implemented by the method according to any one of claims 1 to 9.

Images