A shield segment detection method and system using image features and feature enhancement

Abstract

The application discloses a shield segment detection method and system using image features and feature enhancement, and the method comprises the following steps: a target image sample is preprocessed to obtain a standard target image sample; a shield segment detection model with image feature recognition capability is constructed, the shield segment detection model comprises a backbone feature extraction network, a strengthened feature extraction network, a classifier and a regressor, and the backbone feature extraction network comprises a CONV-C2f module; the standard target image sample is input into the shield segment detection model, the shield segment detection model is iteratively trained, and a target shield segment detection model is obtained; an image to be detected containing a shield segment is input into the target shield segment detection model, and the target shield segment detection model outputs a detection result corresponding to the image to be detected. The image is divided into different hierarchical sizes, features are extracted after convolution, and then the features are fused, so that the category of the segment can be more accurately verified and judged according to the features.

Description

TECHNICAL FIELD

[0001] The present application belongs to the technical field of image detection, and particularly relates to a shield segment detection method and system using image features and feature enhancement. BACKGROUND

[0002] Shield construction is a widely used efficient construction method in tunnel construction, and shield segments are the main components of tunnel lining. Accurate identification of shield segments is of great significance for automatic assembly, quality detection and safety assurance in the construction process. However, shield segments may be affected by factors such as illumination changes, background interference and complex textures in actual construction environment, making it difficult for traditional identification methods to meet the requirements of high precision and real-time performance. SUMMARY

[0003] The present application provides a shield segment detection method and system using image features and feature enhancement, which solves the technical problem that traditional identification methods cannot meet the requirements of high precision and real-time performance.

[0004] In a first aspect, the present application provides a shield segment detection method using image features and feature enhancement, comprising:

[0005] Obtaining a target image sample and pre-processing the target image sample to obtain a standard target image sample;

[0006] Constructing a shield segment detection model with image feature recognition capability, the shield segment detection model comprising a backbone feature extraction network, a strengthened feature extraction network, a classifier and a regressor, the backbone feature extraction network comprising a CONV-C2f module, and the expression of the CONV-C2f module being:

[0007]

[0008] ,

[0009] ,

[0010] ,

[0011] ,

[0012] ,

[0013] In the formula, is the final output feature map, is the channel number compressed by 1x1 convolution, is the feature map after splicing the feature maps of the main branch and the auxiliary branch, is the feature map of the main branch, and the spliced feature map, the processed feature map for the main branch, the feature map directly transmitted for the auxiliary branch, processed by n Bottleneck modules, processed by n Bottleneck modules, the feature map after the channel is expanded by 1x1 convolution, 1x1 convolution operation is performed on, the input feature map;

[0014] inputting the standard target image sample into the shield segment detection model, iteratively training the shield segment detection model, and obtaining a target shield segment detection model;

[0015] inputting an image to be detected containing a shield segment into the target shield segment detection model, and the target shield segment detection model outputs a detection result corresponding to the image to be detected.

[0016] In a second aspect, the present application provides a shield segment detection system using image features and feature enhancement, comprising:

[0017] The acquisition module is configured to acquire a target image sample, and pre-process the target image sample to obtain a standard target image sample.

[0018] A shield segment detection model with the ability to identify image features is constructed, and the shield segment detection model comprises a backbone feature extraction network, a strengthened feature extraction network, a classifier, and a regressor. The backbone feature extraction network comprises a CONV-C2f module, and the expression of the CONV-C2f module is:

[0019]

[0020]

[0021]

[0022]

[0023]

[0024]

[0025] In the formula, the final output feature map,​​​​​​​​​ To compress the number of channels using 1x1 convolution, This is the feature map obtained by concatenating the feature maps of the main branch and the secondary branch. To be and The stitched feature map The feature map after processing of the main branch. The feature map is directly passed to the secondary branch. for Processed through n Bottleneck modules for Processed through n Bottleneck modules To divide the feature map into two parts, This is the feature map after being expanded through a 1x1 convolution. To Perform a 1x1 convolution operation. The input feature map;

[0026] The training module is configured to input the standard target image samples into the shield tunnel segment detection model, and iteratively train the shield tunnel segment detection model to obtain the target shield tunnel segment detection model.

[0027] The output module is configured to input the image to be detected, which contains shield tunnel segments, into the target shield tunnel segment detection model, and the target shield tunnel segment detection model outputs the detection result corresponding to the image to be detected.

[0028] Thirdly, an electronic device is provided, comprising: at least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to perform the steps of the shield tunnel segment detection method utilizing image features and feature enhancement according to any embodiment of the present invention.

[0029] Fourthly, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the steps of the shield tunnel segment detection method utilizing image features and feature enhancement according to any embodiment of the present invention.

[0030] This application proposes a shield tunnel segment detection method and system that utilizes image features and feature enhancement. It proposes a network architecture that enhances feature extraction capabilities by dividing the image into different levels of size, extracting features after convolution, and then fusing the features to more accurately determine the segment category based on feature verification. Attached Figure Description

[0031] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments description. Obviously, the drawings in the following description are some embodiments of the present application, and other drawings can be obtained by those skilled in the art without any creative effort on the basis of these drawings.

[0032] Figure 1 A flow chart of a shield segment detection method using image features and feature enhancement provided by an embodiment of the present application is shown in

[0033] Figure 2 A principle block diagram of a shield segment detection method using image features and feature enhancement provided by an embodiment of the present application is shown in

[0034] Figure 3 A structure block diagram of a shield segment detection system using image features and feature enhancement provided by an embodiment of the present application is shown in

[0035] Figure 4 A structure diagram of an electronic device provided by an embodiment of the present application is shown in DETAILED DESCRIPTION

[0036] In order to make the objects, technical solutions and advantages of the embodiments of the present application clearer, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some embodiments of the present application, but not all the embodiments. Based on the embodiments in the present application, all other embodiments obtained by those skilled in the art without any creative effort belong to the protection scope of the present application.

[0037] Please refer to Figure 1 , which shows a flow chart of a shield segment detection method using image features and feature enhancement provided by the present application.

[0038] As shown in Figure 1 , the shield segment detection method using image features and feature enhancement specifically includes the following steps:

[0039] Step S101, obtaining a target image sample, and pre-processing the target image sample to obtain a standard target image sample.

[0040] In this step, the target image sample is input into a preset image processor, the image processor performs a deblurring operation and a brightening operation on the target image sample, and outputs a standard target image sample corresponding to the target image sample.

[0041] Step S102, a shield segment detection model with image feature recognition capability is constructed, the shield segment detection model includes a backbone feature extraction network, a strengthened feature extraction network, a classifier and a regressor, the backbone feature extraction network includes a CONV-C2f module.

[0042] In this step, the backbone feature extraction network is composed of a CONV-C2f module and a channel attention module, the shield segment detection model automatically performs a forward propagation operation, the input shield segment detection model sequentially passes through a plurality of conv-c2f modules, three effective feature layers pass through the channel attention module, after passing through the channel attention module, feature enhancement operation is performed in the strengthened feature network, and finally input into the classifier and the regressor to perform position labeling.

[0043] The expression of the CONV-C2f module is:

[0044]

[0045] ,

[0046] ,

[0047] ,

[0048] ,

[0049] ,

[0050] In the formula, is the final output feature map, is the channel number compressed by 1x1 convolution, is the feature map after splicing the feature maps of the main branch and the auxiliary branch, is the feature map after splicing and , is the feature map after processing the main branch, is the feature map directly transmitted by the auxiliary branch, is processed by n Bottleneck modules, is processed by n Bottleneck modules, is the feature map divided into two parts, is the feature map after expanding the channel by 1x1 convolution, is the 1x1 convolution operation on , is the input feature map;

[0051] It should be noted that the backbone feature extraction network also contains a channel attention module, and the expression for the weighting operation via the channel attention module is:

[0052] ,

[0053] ,

[0054] ,

[0055] ,

[0056] In the formula, is the feature map after spatial attention weighting, is the output of the spatial attention module, represents element-wise multiplication, is the feature map after channel attention weighting, is the output of the channel attention module, is the Sigmoid activation function, used to normalize the output to the range [0, 1], is the feature map average value in the channel dimension, is the feature map maximum value in the channel dimension, is the global average pooling operation, is the feature after the average pooling operation, which is passed through two fully connected layers to realize feature dimension reduction and dimension increase, is the global maximum pooling operation, is the feature after the maximum pooling operation, which is passed through two fully connected layers to realize feature dimension reduction and dimension increase.

[0057] The bounding box regression loss function in the shield segment detection model is:

[0058] ,

[0059] In the formula, is the loss function, is the area of the real bounding box, is the area of the predicted bounding box, is the minimum closed region of the predicted bounding box and the real bounding box, is the intersection over union between the predicted bounding box and the real bounding box.

[0060] The class loss is used to measure the prediction accuracy of the model for the target class. This model usually uses a multi-class cross-entropy loss function. The formula for the class loss is:

[0061] ,

[0062] where, is the class loss function, is the number of grid cells, is the number of bounding boxes for each grid cell, is the indicator function, which is 1 when the jth bounding box of the ith grid contains the object, otherwise 0, is the One-Hot encoding of the ground truth label is the predicted probability for class j;

[0063] Step S103, input the standard target image sample into the shield segment detection model, iteratively train the shield segment detection model, and obtain a target shield segment detection model.

[0064] Step S104, input an image to be detected containing a shield segment into the target shield segment detection model, and the target shield segment detection model outputs a detection result corresponding to the image to be detected.

[0065] In this step, the detection result contains the position of the shield segment in the image to be detected and the type of the segment.

[0066] In this embodiment, please refer to Figure 2 The principle of shield segment detection using image features and feature enhancement includes:

[0067] Preprocessing, usually the image needs to be preprocessed to the size required by the model input (e.g. 640x640). Preprocessing includes adjusting image size, normalization, etc.

[0068] Load the trained model into memory, and prepare the image to pass through the model for forward propagation. The model will identify and classify the possible objects in the image and generate the corresponding bounding boxes and class labels.

[0069] Parse the detection results output by the model (including bounding box coordinates, class confidence, etc.) into a visual form for easy understanding.

[0070] In summary, the method of the present application proposes a network architecture that enhances feature extraction capability, divides the image into different levels of size, extracts features after convolution, and then fuses the features to more accurately determine the type of segment according to the features.

[0071] In one specific embodiment, the simulation experiment of the present application is implemented under PYTORCH, and NVIDIA(R) Tesla(R) V100 GPU is used for training and testing in the experiment. In order to facilitate comparison, the same shield segment dataset is used. By comparing with other three representative image restoration algorithms, the superior performance of the method of the present application is verified. The five algorithms are FAST-RCNN, SSD, YOLOV8n in turn. For the quality evaluation index, several common indexes are adopted: mAP@0.5 / %, FLOPs, Params. The experimental results on the shield segment dataset are shown in Table 1.

[0072] Table 1 Ablation experiment results of backbone network attention mechanism

[0073] ,

[0074] Table 2 Comparison experiment results

[0075] ,

[0076] As shown in Table 2, the MAP@0.5 of the method in this paper is improved by 12.6%, 25.6%, and 6.4% compared with Fast-RCNN, SSD, and YOLOv8n respectively. At the same time, the present application has achieved significant reduction in the amount of parameters (Params) and the total amount of floating point operations (FLOPS). By comparing different attention modules, the CBAM attention mechanism maintains similar model parameter amount and computational amount compared with other candidate modules, but shows advantages in the key indicator of mAP@0.5. The introduction of CBAM improves the model performance without affecting the model resource occupation.

[0077] Please refer to Figure 3 , which shows a structure block diagram of a shield segment detection system using image features and feature enhancement according to the present application.

[0078] As shown in Figure 3 , the shield segment detection system 200 includes an acquisition module 210, a construction module 220, a training module 230, and an output module 240.

[0079] The acquisition module 210 is configured to acquire target image samples and pre-process the target image samples to obtain standard target image samples.

[0080] The construction module 220 is configured to construct a shield segment detection model with the ability to identify image features, and the shield segment detection model includes a backbone feature extraction network, a strengthened feature extraction network, a classifier, and a regressor. The backbone feature extraction network includes a C2f module and a CONV module. The C2f module fuses features extracted by the CONV module, and a fusion expression is as follows:

[0081]

[0082]

[0083]

[0084]

[0085]

[0086]

[0087] In the formula, is a feature map of a final output, is a channel number compressed by 1x1 convolution, is a feature map obtained by splicing features of a main branch and a sub-branch, is a feature map obtained by splicing and , is a feature map of the main branch after processing, is a feature map directly transmitted by the sub-branch, is processed by n Bottleneck modules, is processed by n Bottleneck modules, is a feature map divided into two parts, is a feature map expanded in a channel after 1x1 convolution, is a 1x1 convolution operation on , is an input feature map.

[0088] The training module 230 is configured to input the standard target image sample into the shield segment detection model, iteratively train the shield segment detection model, and obtain a target shield segment detection model.

[0089] The output module 240 is configured to input an image to be detected containing a shield segment into the target shield segment detection model, and the target shield segment detection model outputs a detection result corresponding to the image to be detected. ​​​​​

[0090] It should be understood that Figure 3 The modules and references described in the document Figure 1 The steps described in the text correspond to those in the method described above. Therefore, the operations, features, and corresponding technical effects described above also apply to the method described in the text. Figure 3 The various modules in the document will not be described in detail here.

[0091] In other embodiments, the present invention also provides a computer-readable storage medium having a computer program stored thereon, wherein when the program instructions are executed by a processor, the processor performs the shield tunnel segment detection method utilizing image features and feature enhancement in any of the above method embodiments.

[0092] In one embodiment, the computer-readable storage medium of the present invention stores computer-executable instructions, which are configured as follows:

[0093] Acquire target image samples and preprocess the target image samples to obtain standard target image samples;

[0094] A shield tunnel segment detection model capable of recognizing image features is constructed. The model includes a backbone feature extraction network, an enhanced feature extraction network, a classifier, and a regressor. The backbone feature extraction network comprises a C2f module and a CONV module. The C2f module fuses the features extracted by the CONV module, where the fusion expression is:

[0095]

[0096] ,

[0097] ,

[0098] ,

[0099] ,

[0100] ,

[0101] In the formula, The final output feature map, To compress the number of channels using 1x1 convolution, This is the feature map obtained by concatenating the feature maps of the main branch and the secondary branch. To be and The stitched feature map The feature map after processing of the main branch. The feature map is directly passed to the secondary branch. for Processed through n Bottleneck modules for Processed through n Bottleneck modules To divide the feature map into two parts, This is the feature map after being expanded through a 1x1 convolution. To Perform a 1x1 convolution operation. The input feature map;

[0102] The standard target image sample is input into the shield tunnel segment detection model, and the shield tunnel segment detection model is iteratively trained to obtain the target shield tunnel segment detection model.

[0103] The image to be detected, containing tunnel segments, is input into the target tunnel segment detection model, and the target tunnel segment detection model outputs the detection result corresponding to the image to be detected.

[0104] Computer-readable storage media may include a program storage area and a data storage area, wherein the program storage area may store an operating system and an application program required for at least one function; the data storage area may store data created based on the use of the tunnel segment inspection system utilizing image features and feature enhancement. Furthermore, the computer-readable storage medium may include high-speed random access memory, and may also include memory, such as at least one disk storage device, flash memory device, or other non-volatile solid-state storage device. In some embodiments, the computer-readable storage medium may optionally include memory remotely disposed relative to a processor, which can be connected via a network to the tunnel segment inspection system utilizing image features and feature enhancement. Examples of such networks include, but are not limited to, the Internet, corporate intranets, local area networks, mobile communication networks, and combinations thereof.

[0105] Figure 4 This is a schematic diagram of the structure of the electronic device provided in the embodiment of the present invention, such as... Figure 4 As shown, the device includes a processor 310 and a memory 320. The electronic device may also include an input device 330 and an output device 340. The processor 310, memory 320, input device 330, and output device 340 can be connected via a bus or other means. Figure 4Taking a bus connection as an example, the memory 320 is the computer-readable storage medium described above. The processor 310 executes various server functions and data processing by running non-volatile software programs, instructions, and modules stored in the memory 320, thereby implementing the shield tunnel segment detection method utilizing image features and feature enhancement as described in the above method embodiment. The input device 330 can receive input digital or character information and generate key signal inputs related to user settings and function control of the shield tunnel segment detection system utilizing image features and feature enhancement. The output device 340 may include a display device such as a screen.

[0106] The aforementioned electronic device can execute the method provided in the embodiments of the present invention, and has the corresponding functional modules and beneficial effects for executing the method. Technical details not described in detail in this embodiment can be found in the method provided in the embodiments of the present invention.

[0107] In one implementation, the above-described electronic device is applied to a tunnel segment inspection system utilizing image features and feature enhancement, for a client, and includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the instructions being executed by the at least one processor to enable the at least one processor to:

[0108] Acquire target image samples and preprocess the target image samples to obtain standard target image samples;

[0109] A shield tunnel segment detection model capable of recognizing image features is constructed. The model includes a backbone feature extraction network, an enhanced feature extraction network, a classifier, and a regressor. The backbone feature extraction network comprises a C2f module and a CONV module. The C2f module fuses the features extracted by the CONV module, where the fusion expression is:

[0110]

[0111] ,

[0112] ,

[0113] ,

[0114] ,

[0115] ,

[0116] In the formula, The final output feature map, To compress the number of channels using 1x1 convolution, This is the feature map obtained by concatenating the feature maps of the main branch and the secondary branch. To be and The stitched feature map The feature map after processing of the main branch. The feature map is directly passed to the secondary branch. for Processed through n Bottleneck modules for Processed through n Bottleneck modules To divide the feature map into two parts, This is the feature map after being expanded through a 1x1 convolution. To Perform a 1x1 convolution operation. The input feature map;

[0117] The standard target image sample is input into the shield tunnel segment detection model, and the shield tunnel segment detection model is iteratively trained to obtain the target shield tunnel segment detection model.

[0118] The image to be detected, containing tunnel segments, is input into the target tunnel segment detection model, and the target tunnel segment detection model outputs the detection result corresponding to the image to be detected.

[0119] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., including several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods of various embodiments or some parts of embodiments.

[0120] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for detecting tunnel lining segments using image features and feature enhancement, characterized in that, include: Acquire target image samples and preprocess the target image samples to obtain standard target image samples; A shield tunnel segment detection model capable of recognizing image features is constructed. The model includes a backbone feature extraction network, an enhanced feature extraction network, a classifier, and a regressor. The backbone feature extraction network includes a CONV-C2f module, wherein the expression of the CONV-C2f module is: ， ， ， ， ， In the formula, The final output feature map, To compress the number of channels using 1x1 convolution, This is the feature map obtained by concatenating the feature maps of the main branch and the secondary branch. To be and The stitched feature map The feature map after processing of the main branch. The feature map is directly passed to the secondary branch. for Processed through n Bottleneck modules for Processed through n Bottleneck modules To divide the feature map into two parts, This is the feature map after being expanded through a 1x1 convolution. To Perform a 1x1 convolution operation. The input feature map; The standard target image sample is input into the shield tunnel segment detection model, and the shield tunnel segment detection model is iteratively trained to obtain the target shield tunnel segment detection model. The image to be detected, containing tunnel segments, is input into the target tunnel segment detection model, and the target tunnel segment detection model outputs the detection result corresponding to the image to be detected.

2. The shield tunnel segment detection method using image features and feature enhancement according to claim 1, characterized in that, The backbone feature extraction network also includes a channel attention module, and the expression for the weighting operation via the channel attention module is: ， ， In the formula, This is the feature map after spatial attention weighting. This is the output of the spatial attention module. This indicates element-wise multiplication. This is the feature map after channel attention weighting. This is the output of the channel attention module.

3. The shield tunnel segment detection method using image features and feature enhancement according to claim 2, characterized in that, in, The expression for the output of the computational spatial attention module is: ， The expression for calculating the output of the channel attention module is: ， In the formula, The Sigmoid activation function is used to normalize the output to the range [0,1]. For feature map The average value in the channel dimension. For feature map The maximum value in the channel dimension. This is a global average pooling operation. To achieve dimensionality reduction and dimensionality increase of the features after average pooling through two fully connected layers, This is a global max pooling operation. To achieve dimensionality reduction and dimensionality increase of the features after max pooling operation, two fully connected layers are used.

4. The method for detecting tunnel lining segments using image features and feature enhancement as described in claim 1, characterized in that, The bounding box regression loss function in the target shield tunnel segment detection model is: ， In the formula, For loss function, The area of ​​the actual bounding box. To predict the area of ​​the bounding box, To predict the minimum closure region between the bounding box and the true bounding box, To predict the intersection-union ratio between the bounding box and the true bounding box.

5. The method for detecting tunnel lining segments using image features and feature enhancement according to claim 1, characterized in that, The preprocessing of the target image samples to obtain standard target image samples includes: The target image sample is input into a preset image processor, which performs deblurring and brightening operations on the target image sample and outputs a standard target image sample corresponding to the target image sample.

6. A shield tunnel segment inspection system utilizing image features and feature enhancement, characterized in that, include: The acquisition module is configured to acquire target image samples and preprocess the target image samples to obtain standard target image samples; The construction module is configured to build a shield tunnel segment detection model with the ability to recognize image features. The shield tunnel segment detection model includes a backbone feature extraction network, an enhanced feature extraction network, a classifier, and a regressor. The backbone feature extraction network includes a CONV-C2f module, wherein the expression of the CONV-C2f module is: ， ， ， ， ， In the formula, The final output feature map, To compress the number of channels using 1x1 convolution, This is the feature map obtained by concatenating the feature maps of the main branch and the secondary branch. To be and The stitched feature map The feature map after processing of the main branch. The feature map is directly passed to the secondary branch. for Processed through n Bottleneck modules for Processed through n Bottleneck modules To divide the feature map into two parts, This is the feature map after being expanded through a 1x1 convolution. To Perform a 1x1 convolution operation. The input feature map; The training module is configured to input the standard target image samples into the shield tunnel segment detection model, and iteratively train the shield tunnel segment detection model to obtain the target shield tunnel segment detection model. The output module is configured to input the image to be detected, which contains shield tunnel segments, into the target shield tunnel segment detection model, and the target shield tunnel segment detection model outputs the detection result corresponding to the image to be detected.

7. An electronic device, characterized in that, include: At least one processor, and a memory communicatively connected to the at least one processor, wherein the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the program is executed by the processor, it implements the method described in any one of claims 1 to 5.

Images