Faster RCNN-based roof tray identification method and system in complex environment of underground roadway

Abstract

The invention discloses a Faster RCNN-based roof tray identification method and system in an underground roadway complex environment, and relates to the field of underground roadway roof target identification, and the method comprises the steps: obtaining a to-be-identified tray image; and inputting the to-be-identified tray image into a pre-trained tray identification model, and outputting an identification result of the to-be-identified tray image. A feature extraction network, an RPN network branch and a detection network branch are used as a network structure of a tray identification model, feature extraction is performed on a tray image through the feature extraction network, and an extracted tray feature map is used as shared input of the RPN network branch and the detection network branch; according to the method, a tray image is acquired, then an RPN network branch is utilized to extract a suggested area as input of a detection network branch, so that an identification result of the tray image is acquired, environmental influences of dark underground light, dust and the like are overcome, the method has the characteristics of high autonomy and high precision, and technical support is provided for implementation of position and posture detection and automatic anchoring protection of tunneling equipment based on machine vision.

Description

Technical Field

[0001] This application relates to the field of roof target recognition in underground roadways, and in particular to a method and system for roof tray recognition in complex environments of underground roadways based on Faster RCNN. Background Technology

[0002] At a critical juncture in the global mining industry's transition towards safety, efficiency, and green practices, smart mines have become a strategic path to overcome the "high-risk, low-efficiency" dilemma of traditional mining. The detection of roof support plates in underground roadways is not only core to achieving vision-based pose detection for tunneling equipment, but also a key step in realizing machine vision-based roadway deformation detection and anchoring robots.

[0003] The underground tunnel environment is complex, characterized by narrow tunnels, high temperature and humidity, dim lighting, and high dust levels. Therefore, according to the requirements of detection accuracy, traditional visual methods are difficult to use for detecting the roof trays of the tunnels. Summary of the Invention

[0004] The purpose of this application is to provide a method and system for roof tray recognition in complex underground roadways based on Faster RCNN, which can overcome the influence of environmental factors such as dim lighting and dust in underground mines, and has the characteristics of strong autonomy and high accuracy.

[0005] To achieve the above objectives, this application provides the following solution: In a first aspect, this application provides a method for identifying roof pallets in complex underground roadways based on Faster R-CNN, the method comprising: Obtain the image of the tray to be identified; The image of the tray to be identified is input into a pre-trained tray recognition model, and the recognition result of the image of the tray to be identified is output. The tray recognition model includes a feature extraction network, an RPN network branch, and a detection network branch. The feature extraction network serves as a shared feature extractor for the RPN network branch and the detection network branch. The input of the RPN network branch is the output of the feature extraction network, and the input of the detection network branch includes the output of the feature extraction network and the output of the RPN network branch. The step of inputting the tray image to be identified into a pre-trained tray recognition model and outputting the recognition result of the tray image to be identified includes: The feature extraction network is used to perform feature extraction on the tray image to be identified, and a tray feature map is obtained. The tray feature map is processed using the RPN network branch to obtain multiple suggested regions; The pallet feature map and multiple suggested regions are input into the detection network branch to output the recognition result of the pallet image to be recognized. The recognition result includes the classification result of whether the pallet image to be recognized is a pallet and the position information of the pallet in the pallet image to be recognized.

[0006] Secondly, this application provides a roof pallet recognition system based on Faster R-CNN in complex underground roadways, the roof pallet recognition system based on Faster R-CNN in complex underground roadways comprising: Image acquisition unit, used to acquire images of the tray to be identified; An image recognition unit is used to input the image of the tray to be recognized into a pre-trained tray recognition model and output the recognition result of the image of the tray to be recognized. The tray recognition model includes a feature extraction network, an RPN network branch, and a detection network branch. The feature extraction network serves as a shared feature extractor for the RPN network branch and the detection network branch. The input of the RPN network branch is the output of the feature extraction network, and the input of the detection network branch includes the output of the feature extraction network and the output of the RPN network branch. The step of inputting the tray image to be identified into a pre-trained tray recognition model and outputting the recognition result of the tray image to be identified includes: The feature extraction network is used to perform feature extraction on the tray image to be identified, and a tray feature map is obtained. The tray feature map is processed using the RPN network branch to obtain multiple suggested regions; The pallet feature map and multiple suggested regions are input into the detection network branch to output the recognition result of the pallet image to be recognized. The recognition result includes the classification result of whether the pallet image to be recognized is a pallet and the position information of the pallet in the pallet image to be recognized.

[0007] Thirdly, this application provides a computer device, including: a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the steps of the Faster RCNN-based method for roof tray recognition in complex underground roadways as described above.

[0008] Fourthly, this application provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the steps of the Faster R-CNN-based roof tray recognition method for complex environments in underground roadways described above.

[0009] Fifthly, this application provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the Faster R-CNN-based roof tray recognition method for complex environments in underground roadways described above.

[0010] According to the specific embodiments provided in this application, the following technical effects are disclosed: This application provides a method and system for roof pallet recognition in complex underground roadways based on Faster R-CNN. The method uses a feature extraction network, an RPN network branch, and a detection network branch as the network structure of the pallet recognition model. The feature extraction network extracts features from the pallet image, and the extracted pallet feature map is used as a shared input to the RPN and detection network branches. Then, the RPN network branch extracts a proposed region, which is used as the input to the detection network branch, thus obtaining the pallet image recognition result. This method overcomes the influence of dim lighting and dust in underground environments, and features strong autonomy and high accuracy. It provides technical support for machine vision-based tunneling equipment pose detection and automatic anchoring. Attached Figure Description

[0011] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of this application. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0012] Figure 1 This is a flowchart illustrating a method for identifying roof trays in complex underground roadways based on Faster RCNN, according to one embodiment of this application. Figure 2 This is a flowchart illustrating a method for identifying roof trays in complex underground roadways based on Faster RCNN, according to one embodiment of this application. Figure 3 This is a schematic diagram illustrating the process of inputting the image of the pallet to be identified into a pre-trained pallet recognition model and outputting the recognition result of the image of the pallet to be identified in a complex environment of underground roadways based on Faster RCNN, as provided in an embodiment of this application. Figure 4 This document provides a flowchart of the training process for a pallet recognition model in a Faster R-CNN-based method for recognizing roof pallets in complex underground roadways, as part of an embodiment of this application. Detailed Implementation

[0013] The technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, and not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0014] To make the above-mentioned objectives, features and advantages of this application more apparent and understandable, the application will be further described in detail below with reference to the accompanying drawings and specific embodiments.

[0015] like Figures 1 to 2 As shown, this application provides a method for roof pallet recognition in complex environments of underground roadways based on Faster R-CNN, including the following steps S101 to S102. Wherein: Step S101: Obtain the image of the pallet to be identified; in this embodiment of the application, the image of the pallet to be identified is obtained by using an industrial camera to capture images of the pallet on the roof of the tunnel.

[0016] Step S102: Input the tray image to be identified into a pre-trained tray recognition model and output the recognition result of the tray image to be identified; wherein, the tray recognition model includes a feature extraction network, an RPN network branch, and a detection network branch, the feature extraction network serves as a shared feature extractor for the RPN network branch and the detection network branch, the input of the RPN network branch is the output of the feature extraction network, and the input of the detection network branch includes the output of the feature extraction network and the output of the RPN network branch.

[0017] In a preferred embodiment of this application, the feature extraction network adopts a trained convolutional neural network model. Specifically, the convolutional neural network model uses a VGG16 deep convolutional neural network to extract tray features, and the Fast RCNN network is constructed by the VGG16 deep convolutional neural network, the RPN network branch, and the detection network branch as the tray recognition model.

[0018] In step S101 of this application embodiment, the step of inputting the tray image to be identified into a pre-trained tray recognition model and outputting the recognition result of the tray image to be identified includes the following steps S201 to S203. Wherein: Step S201: Use the feature extraction network to perform feature extraction on the tray image to be identified to obtain a tray feature map.

[0019] Step S202: Process the tray feature map using the RPN network branch to obtain multiple suggested regions; specifically including the following steps S301 to S303. Wherein: Step S301: Use a sliding window to collect several candidate regions from the tray feature map. The sliding window includes several anchor boxes of different scales generated with the sliding window as the center as the anchor point. Each anchor box is used to collect a candidate region from the tray feature map.

[0020] Step S302: Determine whether the candidate region contains tray features. If the determination result is yes, the candidate region is identified as foreground; otherwise, the candidate region is identified as background. Wherein, the foreground indicates that the candidate region contains a preset target.

[0021] Step S303: Filter the candidate regions identified as foreground regions using nonmaximum suppression to obtain the proposed regions, and output the bounding box coordinates of the proposed regions.

[0022] By performing steps S301 to S303, this application acquires candidate regions by sliding windows of different sizes across the tunnel roof image from left to right and from top to bottom. Figure 3 As shown, in this embodiment, a 3x3 sliding window is selected, and k anchor bounding boxes of different scales are set with the center point of the sliding window as the anchor point. If the current window has a high probability, it is considered that the tray in the image has been detected. Then, non-maximum suppression (NMS) is used to filter and reduce redundant windows. The filtered proposal boxes (proposal regions) are sent to the subsequent ROI pooling layer. The ROI pooling layer pools the proposal regions of different scales to obtain a feature map of fixed length.

[0023] Step S203: Input the tray feature map and multiple proposed regions into the detection network branch to output the recognition result of the tray image to be recognized. The recognition result includes the classification result of whether the tray image to be recognized is a tray and the position information of the tray in the tray image to be recognized. Figure 3 As shown, for k suggested regions, the detection network branch performs a binary classification task, outputting 2k scores and outputting the coordinates of the k suggested regions as location information. The coordinates include four coordinate information, including the coordinates of the anchor point (x and y) and the size of the suggested region (width w and height h), thus obtaining 4k coordinates.

[0024] In step S203 of this application embodiment, the step of inputting the tray feature map and multiple suggested regions into the detection network branch to output the recognition result of the tray image to be recognized includes the following steps S401 to S402. Wherein: Step S401: Use the ROI pooling layer to pool the proposed regions so that the scale of the multiple proposed regions is consistent with the scale of the tray feature map.

[0025] Step S402: Calculate the probability vector of each suggested region belonging to the pre-classified region through a fully connected layer and a Softmax classifier, and output the classification result of each suggested region based on the probability vector; simultaneously, adjust the suggested regions using bounding box regression to obtain the location information of the suggested regions; it should be noted that since the tray recognition model in this application performs a binary classification task, when outputting the classification result of each suggested region based on the probability vector, the probability vector of the suggested region belonging to the pre-classified region is used as the predicted probability and compared with the preset target box predicted probability. When the predicted probability is greater than the target box predicted probability, the suggested region is considered to contain the target, i.e., the tray feature.

[0026] The classification result of the suggested region is the classification result of whether the pallet image to be identified is a pallet; the location information of the suggested region is the location information of the pallet in the pallet image to be identified.

[0027] like Figure 4 As shown in the embodiment of this application, the pre-training method for the tray recognition model includes the following steps S501 to S506. Wherein: Step S501: Construct a sample training set, which includes multiple pre-collected training images, wherein the training images are tray images; use a feature extraction network (... Figure 4 The ImageNet model in the training set extracts features from the images in the training set to obtain the tray feature image of the training set.

[0028] Step S502: Use the tray feature images from the training set to perform the training process of the initial RPN network branch to update the network parameters of the initial RPN network branch, thus obtaining the first RPN network branch, i.e., as shown below. Figure 4 Step 1 in the process.

[0029] Step S503: Freeze the network parameters of the first RPN network branch, and use the tray feature image of the training set and the output of the first RPN network branch to perform the training process of the initial detection network branch to update the network parameters of the initial detection network branch, thereby obtaining the first detection network branch, i.e., as shown below. Figure 4 Step 2 in the process.

[0030] Step S504: Freeze the network parameters of the first detection network branch, release the network parameters of the first RPN network branch, and use the tray feature image of the training set to perform the training process of the first RPN network branch to update the network parameters of the first RPN network branch, thereby obtaining the second RPN network branch, i.e., as shown below. Figure 4 Step 3 in the process.

[0031] Step S505: Release the network parameters of the first detection network branch, and use the tray feature image of the training set and the output of the second RPN network branch to perform the training process of the first detection network branch to update the network parameters of the first detection network branch, thereby obtaining the second detection network branch, i.e., as shown below. Figure 4 Step 4 in the process.

[0032] Step S506: Use the second RPN network branch as the RPN network branch of the pre-trained tray recognition model, and use the second detection network branch as the detection network branch of the pre-trained tray recognition model.

[0033] By implementing steps S501 to S506 above, the Faster RCNN network model in this application uses an alternating training method of training the RPN network branch and the detection network branch in turn during training, so that the entire model forms an end-to-end object detection capability and achieves better overall performance than training alone.

[0034] During the training process, which includes the initial RPN network branch and the first RPN network branch, the loss function used by the RPN network branch is: ; In the formula, For the logarithmic loss of the target and non-target, Let be the anchor index, representing the first... An anchor point box generated based on the anchor point; For the first The predicted probability of each anchor box; For the first The predicted probability of the target box for each anchor point box; For regression loss; For the first Boundary regression box of each anchor point box; The target bounding box coordinate vector; The coordinates, width, and height of the center point of the sliding window.

[0035] It should be noted that the number of tunnel roof pallet images collected in the training set is no less than 5000, and the training set images are labeled with pallet information, including the target bounding box prediction probability and the target bounding box coordinate vector. The target bounding box prediction probability is a binary probability. When the target bounding box prediction probability is 1, it means that the training set image contains the target to be identified (pallet), and when the target bounding box prediction probability is 0, it means that the training set image does not contain the target to be identified. The target bounding box coordinate vector is the location information of the target in the training set image containing the target to be identified. In this application, the training learning rate is set to 0.0001 and the number of iterations is 20000.

[0036] By implementing steps S101 to S102 above, this application uses a feature extraction network, an RPN network branch, and a detection network branch as the network structure of the pallet recognition model. The feature extraction network extracts features from the pallet image, and the extracted pallet feature map is used as the shared input of the RPN network branch and the detection network branch. Then, the RPN network branch is used to extract the proposed region as the input of the detection network branch, thereby obtaining the recognition result of the pallet image. This overcomes the influence of the environment such as dim lighting and dust in the mine, and has the characteristics of strong autonomy and high accuracy, providing technical support for the realization of machine vision-based tunneling equipment pose detection and automatic anchoring.

[0037] Based on the same inventive concept, this application also provides a roof pallet recognition system for complex underground roadways using Faster R-CNN. This system includes an image acquisition unit and an image recognition unit. Image acquisition unit, used to acquire images of the tray to be identified; An image recognition unit is used to input the image of the tray to be recognized into a pre-trained tray recognition model and output the recognition result of the image of the tray to be recognized. The tray recognition model includes a feature extraction network, an RPN network branch, and a detection network branch. The feature extraction network serves as a shared feature extractor for the RPN network branch and the detection network branch. The input of the RPN network branch is the output of the feature extraction network, and the input of the detection network branch includes the output of the feature extraction network and the output of the RPN network branch. The step of inputting the tray image to be identified into a pre-trained tray recognition model and outputting the recognition result of the tray image to be identified includes: The feature extraction network is used to perform feature extraction on the tray image to be identified, and a tray feature map is obtained. The tray feature map is processed using the RPN network branch to obtain multiple suggested regions; The pallet feature map and multiple suggested regions are input into the detection network branch to output the recognition result of the pallet image to be recognized. The recognition result includes the classification result of whether the pallet image to be recognized is a pallet and the position information of the pallet in the pallet image to be recognized.

[0038] In one exemplary embodiment, a computer device is also provided, including a memory and a processor, wherein the memory stores a computer program, and the processor executes the computer program to implement the steps in the above-described method embodiments.

[0039] In one exemplary embodiment, a computer-readable storage medium is provided storing a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0040] In one exemplary embodiment, a computer program product is provided, including a computer program that, when executed by a processor, implements the steps in the above-described method embodiments.

[0041] 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, data stored, data displayed, 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 the relevant data must comply with relevant regulations.

[0042] Those skilled in the art will understand that all or part of the processes in the above embodiments can be implemented by a computer program instructing related hardware. The computer program can be stored in a non-volatile computer-readable storage medium. When executed, the computer program can include the processes of the embodiments described above. Any references to memory, databases, 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 many forms, such as Static Random Access Memory (SRAM) or Dynamic Random Access Memory (DRAM).

[0043] The databases involved in the embodiments provided in this application may include at least one type of relational database and non-relational database. Non-relational databases may include, but are not limited to, blockchain-based distributed databases. The processors involved in the embodiments provided in this application may be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, quantum computing-based data processing logic devices, etc., and are not limited to these.

[0044] The technical features of the above embodiments can be combined in any way. For the sake of brevity, 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.

[0045] This document uses specific examples to illustrate the principles and implementation methods of this application. The descriptions of the above embodiments are only for the purpose of helping to understand the methods and core ideas of this application. Furthermore, those skilled in the art will recognize that, based on the ideas of this application, there will be changes in the specific implementation methods and application scope. Therefore, the content of this specification should not be construed as a limitation of this application.

Claims

1. A method for identifying roof pallets in complex underground roadways based on Faster R-CNN, characterized in that, The method for roof pallet recognition in complex underground roadways based on Faster R-CNN includes: Obtain the image of the tray to be identified; The image of the tray to be identified is input into a pre-trained tray recognition model, and the recognition result of the image of the tray to be identified is output. The tray recognition model includes a feature extraction network, an RPN network branch, and a detection network branch. The feature extraction network serves as a shared feature extractor for the RPN network branch and the detection network branch. The input of the RPN network branch is the output of the feature extraction network, and the input of the detection network branch includes the output of the feature extraction network and the output of the RPN network branch. The step of inputting the tray image to be identified into a pre-trained tray recognition model and outputting the recognition result of the tray image to be identified includes: The feature extraction network is used to perform feature extraction on the tray image to be identified, and a tray feature map is obtained. The tray feature map is processed using the RPN network branch to obtain multiple suggested regions; The pallet feature map and multiple suggested regions are input into the detection network branch to output the recognition result of the pallet image to be recognized. The recognition result includes the classification result of whether the pallet image to be recognized is a pallet and the position information of the pallet in the pallet image to be recognized.

2. The method for roof pallet recognition in complex underground roadways based on Faster R-CNN according to claim 1, characterized in that, The process of using the RPN network branch to process the tray feature map yields multiple proposed regions, including: A sliding window is used to collect several candidate regions from the tray feature map. The sliding window includes several anchor boxes of different scales generated with the sliding window as the center as the anchor point. Each anchor box is used to collect a candidate region from the tray feature map. Determine whether the candidate region contains a tray feature. If the determination result is yes, the candidate region is identified as foreground; otherwise, the candidate region is identified as background. The foreground indicates that the candidate region contains a preset target. The candidate regions identified as foreground are filtered by nonmaximum suppression to obtain the proposed regions, and the bounding box coordinates of the proposed regions are output.

3. The method for roof tray recognition in complex underground roadways based on Faster R-CNN according to claim 2, characterized in that, The step of inputting the tray feature map and multiple proposed regions into the detection network branch to output the recognition result of the tray image to be recognized includes: The proposed regions are pooled using an ROI pooling layer to ensure that the scale of multiple proposed regions is consistent with the scale of the tray feature map. The probability vector of each proposed region belonging to the pre-class is calculated by using a fully connected layer and a Softmax classifier, and the classification result of each proposed region is output according to the probability vector; at the same time, the proposed regions are adjusted by using bounding box regression to obtain the position information of the proposed regions. The classification result of the suggested region is the classification result of whether the pallet image to be identified is a pallet; the location information of the suggested region is the location information of the pallet in the pallet image to be identified.

4. The method for roof pallet recognition in complex underground roadways based on Faster R-CNN according to claim 1, characterized in that, The pre-training method for the tray recognition model includes: A sample training set is constructed, which includes multiple pre-collected training set images, wherein the training set images are tray images; a feature extraction network is used to extract features from the training set images to obtain tray feature images of the training set; The training process of the initial RPN network branch is performed using the tray feature images of the training set to update the network parameters of the initial RPN network branch, thereby obtaining the first RPN network branch; Freeze the network parameters of the first RPN network branch, and use the tray feature image of the training set and the output of the first RPN network branch to perform the training process of the initial detection network branch to update the network parameters of the initial detection network branch, thereby obtaining the first detection network branch. Freeze the network parameters of the first detection network branch, release the network parameters of the first RPN network branch, and use the tray feature image of the training set to perform the training process of the first RPN network branch to update the network parameters of the first RPN network branch, thereby obtaining the second RPN network branch. Release the network parameters of the first detection network branch, and use the tray feature image of the training set and the output of the second RPN network branch to perform the training process of the first detection network branch to update the network parameters of the first detection network branch, thereby obtaining the second detection network branch; The second RPN network branch is used as the RPN network branch of the pre-trained tray recognition model, and the second detection network branch is used as the detection network branch of the pre-trained tray recognition model.

5. The method for roof tray recognition in complex underground roadways based on Faster R-CNN according to claim 4, characterized in that, The loss function of the RPN network branch is: ； In the formula, For the logarithmic loss of the target and non-target, Let be the anchor index, representing the first... An anchor point box generated based on the anchor point; For the first The predicted probability of each anchor box; For the first The predicted probability of the target box for each anchor point box; For regression loss; For the first Boundary regression box of each anchor point box; The target bounding box coordinate vector; The coordinates, width, and height of the center point of the sliding window.

6. The method for roof tray recognition in complex underground roadways based on Faster R-CNN according to claim 5, characterized in that, The training learning rate for the tray recognition model is set to 0.0001 and the number of iterations is 20000 during the pre-training process.

7. A roof pallet recognition system based on Faster R-CNN in complex underground roadways, characterized in that, The Faster R-CNN-based roof tray recognition system for complex underground roadways includes: Image acquisition unit, used to acquire images of the tray to be identified; An image recognition unit is used to input the image of the tray to be recognized into a pre-trained tray recognition model and output the recognition result of the image of the tray to be recognized. The tray recognition model includes a feature extraction network, an RPN network branch, and a detection network branch. The feature extraction network serves as a shared feature extractor for the RPN network branch and the detection network branch. The input of the RPN network branch is the output of the feature extraction network, and the input of the detection network branch includes the output of the feature extraction network and the output of the RPN network branch. The step of inputting the tray image to be identified into a pre-trained tray recognition model and outputting the recognition result of the tray image to be identified includes: The feature extraction network is used to perform feature extraction on the tray image to be identified, and a tray feature map is obtained. The tray feature map is processed using the RPN network branch to obtain multiple suggested regions; The pallet feature map and multiple suggested regions are input into the detection network branch to output the recognition result of the pallet image to be recognized. The recognition result includes the classification result of whether the pallet image to be recognized is a pallet and the position information of the pallet in the pallet image to be recognized.

8. A computer device, comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that the processor executes the computer program to implement the Faster R-CNN-based method for roof tray recognition in complex underground roadways as described in any one of claims 1-6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that, When executed by a processor, the computer program implements the method for roof tray recognition in complex environments of underground roadways based on Faster R-CNN as described in any one of claims 1-6.

10. A computer program product, comprising a computer program, characterized in that, When executed by a processor, the computer program implements the method for roof tray recognition in complex environments of underground roadways based on Faster R-CNN as described in any one of claims 1-6.

Images