Reinforcing mesh recognition method based on RuNet semantic segmentation network

Abstract

The invention belongs to the technical field of building construction quality detection, and particularly relates to a reinforcing mesh recognition method based on a ReuNet semantic segmentation network, and the method comprises the following steps: S1, a construction acceptance worker operates a terminal to start up, enters an operation interface through a handheld terminal, selects a corresponding function module, and carries out the recognition of the reinforcing mesh; the functional module comprises an anchor rod detection module, a steel bar detection module and a welding seam detection module; s2, performing image acquisition on the corresponding anchor rod, the reinforcing steel bar or the welding seam through the selected function module, and preprocessing the acquired image through a data processing module arranged in the terminal; and S3, when the network connection between the terminal and the server is normal, the terminal transmits the preprocessed image to the server in real time through a video stream transmission protocol, and a built-in ReuNet semantic segmentation network of the server carries out semantic segmentation and identification analysis on the received image to obtain a detection result and stores the detection result.

Description

Technical Field

[0001] This invention belongs to the field of building construction quality inspection technology, and in particular relates to a rebar mesh recognition method based on ReuNet semantic segmentation network. Background Technology

[0002] Anchors and reinforcing bars are the main load-bearing components in concrete structures and are key links in the quality acceptance of concealed works. According to the "Code for Design of Concrete Reinforcement in Railway Engineering" (TB 10064-2019) [1], there are numerical requirements for the spacing, anchorage and joints of reinforcing bars. Manual measurement and counting are often inefficient and have large errors. Compared with the traditional manual reinforcing bar acceptance method, the use of image recognition technology can quickly and accurately process a large amount of image data, greatly improve the efficiency and accuracy of reinforcing bar quantity detection, and reduce human error. The existing method of using image recognition technology to count reinforcing bars is to convert continuous grayscale images into binary images by setting a threshold to achieve the segmentation of foreground and background. This method can only show the edge contour and cannot show the image details and texture. Moreover, the application scenario is limited by the background environment. When facing complex background image recognition, the effect is not outstanding [2]. The application of image recognition technology in anchor and reinforcing bar detection is only the detection of apparent cracks [3]. Therefore, this paper proposes an automatic identification method for the acceptance of reinforcing bars in railway concealed works and establishes the automatic identification system for the acceptance of reinforcing bars, which supports handheld video terminals and is easy to shoot.

[0003] Traditional steel mesh identification mainly relies on manual visual inspection and measurement. This method not only consumes a lot of manpower and time, but is also easily affected by the experience and sense of responsibility of the inspectors, as well as environmental factors such as on-site lighting and dust. This makes it easy to miss or misidentify, resulting in difficulty in guaranteeing the accuracy of the inspection.

[0004] With the development of computer vision technology, image recognition technology has been introduced into some construction acceptance processes. However, existing technologies still have significant shortcomings: on the one hand, commonly used image recognition models (such as basic CNN models and ordinary UNet models) suffer from insufficient feature extraction and blurred boundary segmentation when processing images with dense textures and complex cross structures, such as steel mesh, making it difficult to accurately identify key parameters such as the quantity, spacing, and diameter of steel bars; on the other hand, the network environment at construction sites is complex, and network interruptions and unstable signals are common. Existing detection systems mostly rely on stable networks for data transmission and processing, and data loss and interruption of the detection process are prone to occur when the network is abnormal, resulting in low efficiency of acceptance work. In view of this, we propose a steel mesh recognition method based on the ReuNet semantic segmentation network. Summary of the Invention

[0005] The purpose of this invention is to address the aforementioned technical problems by providing a rebar mesh recognition method based on the ReuNet semantic segmentation network, thereby resolving the issues raised in the background section.

[0006] In view of this, the present invention provides a rebar mesh recognition method based on the ReuNet semantic segmentation network, comprising the following steps: Step S1: The construction acceptance personnel turn on the terminal, enter the operation interface through the handheld terminal and select the corresponding function module. The function module includes the anchor bolt detection module, the rebar detection module and the weld seam detection module. Step S2: The selected function module acquires images of the corresponding anchor rod, rebar or weld seam, and the data processing module built into the terminal preprocesses the acquired images. Step S3: When the network connection between the terminal and the server is normal, the terminal transmits the preprocessed image to the server in real time through the video streaming protocol. The server's built-in ReuNet semantic segmentation network performs semantic segmentation and recognition analysis on the received image, obtains the detection results, and stores them. Step S4: When the network connection between the terminal and the server is abnormal, the terminal saves the collected raw image data and preprocessing results to the local cache, and the terminal's data cache management module records the status of this data item as "not uploaded". Step S5: The terminal's real-time transmission module periodically checks the network connection status. When the network is detected to be restored, the data upload process is triggered to transmit the data in the local cache with the status of "not uploaded" to the server. Step S6: If the network is interrupted again during the data upload process, the terminal records the progress of the uploaded data. After the network is restored, it continues to upload the remaining data until all "unuploaded" data is transmitted. After receiving the data, the server performs recognition and analysis through the ReuNet semantic segmentation network and stores the detection results.

[0007] In the above technical solution, the anchor bolt detection module further includes an anchor bolt image enhancement unit. The anchor bolt image enhancement unit performs adaptive contrast adjustment and edge sharpening processing on the rust and surface stain areas in the acquired anchor bolt image to highlight the contour features of the anchor bolt. The adaptive contrast adjustment dynamically adjusts the enhancement coefficient based on the difference between the grayscale mean of the anchor bolt area and the grayscale mean of the background area.

[0008] In the above technical solution, the rebar detection module further includes a rebar image acquisition optimization unit. This unit can automatically adjust the exposure parameters and white balance of the terminal camera according to the ambient light intensity. When the light intensity is lower than 500 lux, the supplementary lighting function is automatically turned on to ensure that the clarity of the rebar texture details in the acquired rebar image is not less than 80%. When the rebar detection module performs rebar recognition, it can distinguish the rebar category through the ReuNet semantic segmentation network. The category includes plain round rebar and ribbed rebar.

[0009] In the above technical solution, further, the joint detection module in step S1 includes a weld seam image preprocessing unit. This unit performs Gaussian filtering denoising on the acquired weld seam image to remove high-frequency noise in the image, and then uses an adaptive threshold segmentation algorithm to separate the weld seam region from the background region. The adaptive threshold is calculated based on the local gray-scale mean and variance of the weld seam region and dynamically adjusted.

[0010] Furthermore, the above technical solution also includes a terminal data processing module, which performs image preprocessing including image denoising, contrast enhancement, and size normalization.

[0011] In the above technical solution, the preliminary preprocessing of the terminal data processing module further includes: extracting straight line features in the image based on the edge detection algorithm and removing non-straight line interference areas; Adaptive brightness correction is applied to the image to eliminate the impact of uneven lighting at the construction site on the identification of steel mesh.

[0012] In the above technical solution, further, the ReuNet semantic segmentation network in step S3 is a network model based on UNet improvement. The network includes an encoding module, a decoding module and a skip connection module. The encoding module uses a residual structure to extract image features, and the decoding module performs weighted fusion of features through an attention mechanism.

[0013] In the above technical solution, further, in step S3, the detection results obtained by the server through the ReuNet semantic segmentation network include the location information of the steel mesh, the number of steel bars, the spacing of the steel bars, the length of the anchor bolts, and the type of weld defects. The video stream transmission protocol is either RTSP or WebRTC.

[0014] In the above technical solution, the data cache management module in step S4 is further used to manage the locally cached image data. When the cache space is insufficient, the earliest uploaded data is deleted according to the order of data storage time.

[0015] In the above technical solution, further, in step S6, after receiving the detection results returned by the server, the terminal can visualize the detection results on the operation interface, including marking the positions of the reinforcing bars, anchor bolts, and welds and the detection results on the original image using different colored bounding boxes or masks. When storing the detection results, they are classified and stored according to the detection time and terminal number for easy subsequent query and statistics. The beneficial effects of this invention are: 1. This rebar mesh recognition method based on ReuNet semantic segmentation network applies ReuNet semantic segmentation network to rebar mesh recognition. By introducing attention mechanism and model optimization, it improves the accuracy and robustness of rebar mesh recognition and can effectively identify rebar mesh under different working conditions.

[0016] 2. The rebar mesh recognition method based on the ReuNet semantic segmentation network uses a preprocessing module on the terminal to denoise and enhance the image, reducing interference from irrelevant information and improving the server's processing efficiency. At the same time, by combining real-time transmission and local caching mechanisms, reliable data transmission and processing are ensured even when network conditions are unstable, enhancing the practicality of the method.

[0017] 3. This rebar mesh recognition method based on the ReuNet semantic segmentation network utilizes the online update function of the ReuNet model to continuously adapt to new construction environments, thereby improving recognition performance and extending the lifespan of the method. The comprehensive storage and application mechanism for detection results facilitates the management and analysis of construction acceptance data, providing data support for improving construction quality. Attached Figure Description

[0018] Figure 1 This is a flowchart of the present invention. Detailed Implementation

[0019] The technical solutions of the embodiments of this application will be clearly described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application are within the scope of protection of this application.

[0020] In the description of this application, it should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the exemplary embodiments according to this application. For ease of description, the dimensions of the various parts shown in the drawings are not drawn to actual scale. Techniques, methods, and devices known to those skilled in the art may not be discussed in detail, but where appropriate, such techniques, methods, and devices should be considered part of the specification. In all examples shown and discussed herein, any specific values ​​should be interpreted as merely exemplary and not as limitations. Therefore, other examples of exemplary embodiments may have different values. It should be noted that similar reference numerals and letters in the following drawings denote similar items; therefore, once an item is defined in one drawing, it need not be further discussed in subsequent drawings.

[0021] Example 1: Please see Figure 1 As shown, this embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network.

[0022] Includes the following steps: Step S1: The construction acceptance personnel turn on the terminal, enter the operation interface through the handheld terminal and select the corresponding function module. The function modules include the anchor bolt detection module, the rebar detection module and the weld seam detection module. Step S2: The selected function module acquires images of the corresponding anchor rod, rebar or weld seam, and the data processing module built into the terminal preprocesses the acquired images. Step S3: When the network connection between the terminal and the server is normal, the terminal transmits the preprocessed image to the server in real time through the video streaming protocol. The server's built-in ReuNet semantic segmentation network performs semantic segmentation and recognition analysis on the received image, obtains the detection results, and stores them. Step S4: When the network connection between the terminal and the server is abnormal, the terminal saves the collected raw image data and preprocessing results to the local cache, and the terminal's data cache management module records the status of this data item as "not uploaded". Step S5: The terminal's real-time transmission module periodically checks the network connection status. When the network is detected to be restored, the data upload process is triggered to transmit the data in the local cache with the status of "not uploaded" to the server. Step S6: If the network is interrupted again during the data upload process, the terminal records the progress of the uploaded data. After the network is restored, it continues to upload the remaining data until all "unuploaded" data is transmitted. After receiving the data, the server performs recognition and analysis through the ReuNet semantic segmentation network and stores the detection results.

[0023] Example 2: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0024] In step S1, the anchor bolt detection module includes an anchor bolt image enhancement unit. The anchor bolt image enhancement unit performs adaptive contrast adjustment and edge sharpening processing on the rust and surface stain areas in the acquired anchor bolt image to highlight the contour features of the anchor bolt. The adaptive contrast adjustment dynamically adjusts the enhancement coefficient based on the difference between the grayscale mean of the anchor bolt area and the grayscale mean of the background area.

[0025] Example 3: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0026] In step S1, the rebar detection module includes a rebar image acquisition and optimization unit. This unit can automatically adjust the exposure parameters and white balance of the terminal camera according to the ambient light intensity. When the light intensity is lower than 500 lux, it automatically turns on the supplementary lighting function to ensure that the clarity of the rebar texture details in the acquired rebar image is not less than 80%. When the rebar detection module performs rebar recognition, it can distinguish the rebar categories through the ReuNet semantic segmentation network. The categories include plain round rebar and ribbed rebar.

[0027] Example 4: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0028] In step S1, the joint detection module includes a weld seam image preprocessing unit. This unit performs Gaussian filtering on the acquired weld seam image to remove high-frequency noise from the image. Then, it uses an adaptive threshold segmentation algorithm to separate the weld seam region from the background region. The adaptive threshold is calculated based on the local gray-scale mean and variance of the weld seam region and is dynamically adjusted.

[0029] Example 5: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0030] It also includes a terminal data processing module, which performs image preprocessing including image denoising, contrast enhancement, and size normalization.

[0031] The preliminary preprocessing of the terminal data processing module includes: extracting straight line features from the image based on the edge detection algorithm and removing non-straight line interference areas; Adaptive brightness correction is applied to the image to eliminate the impact of uneven lighting at the construction site on the identification of steel mesh.

[0032] Example 6: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0033] In step S3, the ReuNet semantic segmentation network is a network model based on an improved UNet. This network includes an encoding module, a decoding module, and a skip connection module. The encoding module uses a residual structure to extract image features, and the decoding module uses an attention mechanism to perform weighted fusion of features. The residual structure used in the encoding module of the ReuNet semantic segmentation network is the residual block design of ResNet, especially the BasicBlock structure of ResNet. The core formula of the residual structure is: Output = F(x) + x, where x is the input feature map, F(x) is the residual mapping after convolution transformation, and + indicates element-wise addition (skip connection).

[0034] It is worth noting that the attention module workflow consists of the following four key steps: Step 1: Global Context Extraction Input feature map (x) → Global average pooling → 1×1 convolution (dimensionality reduction) → ReLU activation → 1×1 convolution (dimensionality increase) → Attention vector (g); The feature map is compressed from [B,H,W,C] to [B,1,1,C] using global average pooling. After processing through two 1×1 convolutional layers (including ReLU activation), a global attention vector is generated; Step 2: Attention weight generation; Attention vector (g) → Softmax activation → Normalized weight vector (α); The global vector is normalized using Softmax to generate weights in the range (0,1); The length of the weight vector α is equal to the number of channels C of the input feature map; Step 3: Feature Enhancement Input feature map (x) × weight vector (α) → enhanced feature map; Broadcast the weight vector α to the spatial dimension and perform element-wise multiplication with the original feature map; Enhanced feature maps highlight channels and regions with higher weights while suppressing background information; Step 4: Feature Fusion and Output Enhanced feature map + high-level semantic features → ReLU activation → output feature map; The enhanced feature map is residually connected to the high-level semantic features passed from the encoder; The final feature map is output through ReLU activation and used for subsequent upsampling and segmentation.

[0035] In step S3, the detection results obtained by the server through the ReuNet semantic segmentation network include the location information of the steel mesh, the number of steel bars, the spacing between steel bars, the length of the anchor bolts, and the type of weld defects. The video stream transmission protocol is either RTSP or WebRTC.

[0036] Example 7: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0037] In step S4, the data cache management module is also used to manage the locally cached image data. When the cache space is insufficient, the earliest uploaded data is deleted according to the order of data storage time.

[0038] Example 8: This embodiment provides a rebar mesh recognition method based on the ReuNet semantic segmentation network. In addition to the technical solutions of the above embodiments, it also has the following technical features.

[0039] In step S6, after receiving the detection results returned by the server, the terminal can visualize the detection results on the operation interface, including marking the positions of the steel bars, anchor rods and welds and the detection results on the original image with different colored bounding boxes or masks. When storing the detection results, they are classified and stored according to the detection time and terminal number for easy subsequent query and statistics.

[0040] This invention: Image acquisition and preprocessing stage The acceptance personnel start the handheld terminal, enter the operation interface, and select the target function module (anchor bolt inspection, rebar inspection, or weld joint inspection); The terminal adaptively adjusts the acquisition parameters based on the selected module: the rebar detection module adjusts exposure and white balance according to the ambient light intensity, and automatically activates supplemental lighting when the light intensity is below 500 lux; the anchor bolt detection module prepares for subsequent image enhancement processing; the weld seam detection module presets Gaussian filter parameters. Acceptance personnel acquire images of the target area using the terminal camera, and the terminal data processing module simultaneously performs preprocessing: extracting line features to remove non-target interference, adaptive brightness correction, noise reduction, contrast enhancement, and size normalization; Network status assessment and data processing stage The terminal monitors the network connection status with the server in real time to determine whether data transmission conditions are met. If the network connection is normal: the terminal transmits the preprocessed image to the server in real time as a video stream via RTSP or WebRTC protocol; the server starts the ReuNet semantic segmentation network, extracts features through residual structure and fuses features through attention mechanism, completes semantic segmentation and recognition analysis, generates detection results containing information on location, quantity, size, and defect type, and stores them; If the network connection is interrupted: the terminal automatically stores the original image data and preprocessing results into the local cache, and the data cache management module marks the status of the data item as "not uploaded"; if the cache space is insufficient, the earliest uploaded data is deleted according to the storage time to ensure the storage of new data; Network recovery and data resume phase The terminal's real-time transmission module periodically checks the network connection status and continuously monitors network recovery. When network recovery is detected, the data upload process is automatically triggered, prioritizing the transmission of data in the local cache that is in the "not uploaded" state; If the network is interrupted again during the upload process: the terminal immediately records the progress of the uploaded data and pauses the transmission; after the network is restored, the remaining data will continue to be uploaded based on the recorded progress to avoid duplicate transmission; Server final processing stage After the server has received all the "unuploaded" data, it starts the ReuNet semantic segmentation network to perform batch recognition and analysis on the remaining data. The network outputs the detection results, which include: the location, quantity, and spacing of the reinforcing mesh, the length of the anchor bolts, and the type of weld defects. These results are then merged and archived with the results of the processed data.

[0041] Results Presentation and Data Archiving Phase The server sends the complete detection results back to the terminal, which then displays them visually on the user interface: the target location is marked with different colored bounding boxes or masks, detailed parameters are displayed in a floating manner, and abnormal items are highlighted with special colors.

[0042] The terminal categorizes and stores the test results for easy subsequent querying, statistics, and traceability of engineering quality.

[0043] The embodiments of this application have been described above with reference to the accompanying drawings. Unless otherwise specified, the embodiments and features in the embodiments of this application can be combined with each other. This application is not limited to the specific embodiments described above. The specific embodiments described above are merely illustrative and not restrictive. Those skilled in the art can make many other forms under the guidance of this application without departing from the spirit and scope of the claims, and all of these forms are within the protection scope of this application.

Claims

1. A method for identifying rebar mesh based on ReuNet semantic segmentation network, characterized in that, Includes the following steps: Step S1: The construction acceptance personnel turn on the terminal, enter the operation interface through the handheld terminal and select the corresponding function module. The function module includes the anchor bolt detection module, the rebar detection module and the weld seam detection module. Step S2: The selected function module acquires images of the corresponding anchor rod, rebar or weld seam, and the data processing module built into the terminal preprocesses the acquired images. Step S3: When the network connection between the terminal and the server is normal, the terminal transmits the preprocessed image to the server in real time through the video streaming protocol. The server's built-in ReuNet semantic segmentation network performs semantic segmentation and recognition analysis on the received image, obtains the detection results, and stores them. Step S4: When the network connection between the terminal and the server is abnormal, the terminal saves the collected raw image data and preprocessing results to the local cache, and the terminal's data cache management module records the status of this data item as "not uploaded". Step S5: The terminal's real-time transmission module periodically checks the network connection status. When the network is detected to be restored, the data upload process is triggered to transmit the data with the status of "not uploaded" in the local cache to the server. Step S6: If the network is interrupted again during the data upload process, the terminal records the progress of the uploaded data. After the network is restored, it continues to upload the remaining data until all "unuploaded" data is transmitted. After receiving the data, the server performs recognition and analysis through the ReuNet semantic segmentation network and stores the detection results.

2. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, In step S1, the anchor bolt detection module includes an anchor bolt image enhancement unit. The anchor bolt image enhancement unit performs adaptive contrast adjustment and edge sharpening processing on the rust and surface stain areas in the acquired anchor bolt image to highlight the contour features of the anchor bolt. The adaptive contrast adjustment dynamically adjusts the enhancement coefficient based on the difference between the grayscale mean of the anchor bolt area and the grayscale mean of the background area.

3. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, In step S1, the rebar detection module includes a rebar image acquisition and optimization unit. This unit can automatically adjust the exposure parameters and white balance of the terminal camera according to the ambient light intensity. When the light intensity is lower than 500 lux, it automatically turns on the supplementary lighting function to ensure that the clarity of the rebar texture details in the acquired rebar image is not less than 80%. When the rebar detection module performs rebar recognition, it can distinguish the rebar categories through the ReuNet semantic segmentation network. The categories include plain round rebar and ribbed rebar.

4. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, The joint detection module in step S1 includes a weld seam image preprocessing unit. This unit performs Gaussian filtering denoising on the acquired weld seam image to remove high-frequency noise from the image. Then, it uses an adaptive threshold segmentation algorithm to separate the weld seam region from the background region. The adaptive threshold is calculated based on the local gray-scale mean and variance of the weld seam region and is dynamically adjusted.

5. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, It also includes a terminal data processing module, which performs image preprocessing including image denoising, contrast enhancement, and size normalization.

6. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 5, characterized in that, The preliminary preprocessing of the terminal data processing module includes: extracting straight line features from the image based on the edge detection algorithm and removing non-straight line interference areas; Adaptive brightness correction is applied to the image to eliminate the impact of uneven lighting at the construction site on the identification of steel mesh.

7. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, In step S3, the ReuNet semantic segmentation network is a network model based on UNet. This network includes an encoding module, a decoding module, and a skip connection module. The encoding module uses a residual structure to extract image features, and the decoding module uses an attention mechanism to perform weighted fusion of features.

8. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, In step S3, the detection results obtained by the server through the ReuNet semantic segmentation network include the location information of the steel mesh, the number of steel bars, the spacing between steel bars, the length of the anchor bolts, and the type of weld defects. The video stream transmission protocol is either RTSP or WebRTC.

9. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, In step S4, the data cache management module is also used to manage the locally cached image data. When the cache space is insufficient, the earliest uploaded data is deleted according to the order of data storage time.

10. The rebar mesh recognition method based on ReuNet semantic segmentation network according to claim 1, characterized in that, In step S6, after receiving the detection results returned by the server, the terminal can visualize the detection results on the operation interface, including marking the positions of the reinforcing bars, anchor rods and welds and the detection results on the original image with different colored bounding boxes or masks. When storing the detection results, they are classified and stored according to the detection time and terminal number for easy subsequent query and statistics.

Images