Concrete drain pipe crack automatic identification method, system, device and storage medium

Abstract

This invention discloses an automatic identification method, system, device, and storage medium for cracks in concrete drainage pipes. The method includes: acquiring original pipe images captured by a ground inspection robot and preprocessing them to obtain preprocessed original pipe images; detecting cracks in the preprocessed original pipe images using an improved HST-Net object detection algorithm to obtain crack locations; identifying and recording the distance of the crack location relative to the starting point in the current pipe inspection task using an improved CTPN character detection algorithm and CTC character correction algorithm; and exporting the identified data and storing it in a cloud database. This method, through the improved HST-Net object detection algorithm, achieves automatic crack identification in concrete drainage pipe images, improving detection efficiency and accuracy; it also utilizes the improved CTPN character detection algorithm and CTC character correction algorithm to effectively identify and correct text information related to crack locations, greatly improving the accuracy of character recognition and correction.

Description

Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method, system, device and storage medium for automatic identification of cracks in concrete drainage pipes. Background Technology

[0002] Concrete drainage pipes are a crucial component of urban infrastructure, undertaking vital functions such as rainwater drainage and sewage transport. However, due to long-term use or the impact of external environmental factors, these pipes are prone to cracking, affecting their normal function and potentially leading to serious safety hazards. Therefore, regular inspection and maintenance of concrete drainage pipes are essential. Traditional inspection methods typically rely on manual patrols. While this method is intuitive, it is inefficient and limited by the technical skills of the inspectors and environmental conditions, making it difficult to guarantee the comprehensiveness and accuracy of the inspection.

[0003] In recent years, with the rapid development of image processing and artificial intelligence technologies, image recognition-based crack detection methods have gradually attracted attention. Existing automatic detection methods typically include steps such as image acquisition, preprocessing, feature extraction, and crack identification. These methods acquire images of the inside of pipes using cameras, and then use computer vision technology to process and analyze the images to automatically detect cracks. Common image processing techniques include grayscale conversion, threshold segmentation, and edge detection, while target detection algorithms include traditional template matching and modern deep learning methods such as YOLO and SSD.

[0004] While existing automatic crack detection methods have improved the efficiency and accuracy of crack detection to some extent, they still have some shortcomings. First, image preprocessing is cumbersome, requiring multiple processing steps to achieve satisfactory results, which increases the system's complexity and computational load. Second, existing crack detection algorithms perform poorly under complex backgrounds and noise interference, easily leading to false positives or false negatives. Furthermore, most methods focus primarily on crack identification rather than providing effective means to record and analyze the specific location and changes of cracks, limiting their application in practical engineering. Summary of the Invention

[0005] To address the aforementioned deficiencies or improvement needs of existing technologies, the present invention aims to provide an automatic method, system, device, and storage medium for identifying cracks in concrete drainage pipes. This method utilizes an improved HST-Net target detection algorithm to automatically identify cracks in concrete drainage pipe images. This reduces reliance on manual inspection and improves detection efficiency and accuracy. The algorithm can accurately locate the minimum bounding rectangle coordinates of the crack's outline and effectively handle complex backgrounds and noise interference, ensuring high-accuracy crack detection. The method also employs an improved CTPN character detection algorithm and CTC character correction algorithm to effectively identify and correct textual information related to crack locations. These algorithms, combined with a region proposal network and a bidirectional LSTM network, can accurately detect character boundaries and perform precise correction via CTC, significantly improving the accuracy of character recognition and correction. Particularly in the distance calculation of crack locations, these improvements ensure the reliability and accuracy of the data information.

[0006] According to a first aspect of the present invention, an automatic identification method for cracks in concrete drainage pipes is provided, comprising:

[0007] S100: Acquire the original pipeline image captured by the ground inspection robot, and preprocess it to obtain the preprocessed original pipeline image.

[0008] S200. The preprocessed original pipeline image is cracked using the improved HST-Net target detection algorithm to locate the coordinates of the minimum outline bounding rectangle of the crack and obtain the crack location.

[0009] S300. The distance of the crack location relative to the starting point in the current pipeline inspection task is identified and recorded by using the improved CTPN character detection algorithm and CTC character correction algorithm.

[0010] S400: Export the identified data and store it in the cloud database.

[0011] Further, the preprocessing in step S100 includes: performing mean blurring on the original pipeline image to smooth noise and details in the image; applying mean filtering to further remove random noise in the image; converting the color image to a grayscale image using grayscale processing; using an adaptive threshold segmentation method to separate the crack area from the background area in the image to enhance the contrast of the crack; using the Canny edge detection algorithm to extract edge information in the image and accurately locate the crack outline; and performing morphological transformation to remove noise interference and connect broken crack segments to make the crack features clearer and more coherent.

[0012] Furthermore, the improved HST-Net object detection algorithm includes:

[0013] The backbone network consists of top convolutional layers, the first C3 module, the MACSP multi-head self-attention module, and the SPPF module, which are used to move from initial feature extraction to higher-level feature representation.

[0014] The neck network consists of a graph layer, an upsampling layer, a second C3 module, and an environment enhancement module, which are used for feature fusion, scale transformation, and enhancement processing.

[0015] The head network, including the detection layer and the third C3 module, is used to perform the final processing and detection output of the fused and enhanced feature maps.

[0016] Furthermore, the top-level convolutional layer is used to extract low-level features through top-level operations, while effectively reducing the size of the feature map through stride and padding operations; the first C3 module is used to perform feature fusion after each texture layer, integrating feature information on different geometric shapes to enhance the model's expressive power; the MACSP multi-head self-attention module is used to weight and summarize features at different locations through multiple attention heads, capturing important features in the global scope and improving the network's feature extraction efficiency; the SPPF module is used to further process features using multiple pooling operations to ensure that effective features are extracted from the input image.

[0017] Furthermore, the pattern layer and the upsampling layer are responsible for adjusting the size and structure of the feature maps and enhancing these features; the second C3 module is used to stitch the fused feature maps together to retain more detailed information and contextual relationships; the environment enhancement module is used to enhance the importance of features through a reweighting mechanism, further enhancing the feature representation capability of the model and ensuring the effectiveness of features passed to the head network.

[0018] Furthermore, the detection layer is used to perform specific detection and classification operations using predefined anchor boxes and the number of categories; the third C3 module is used to integrate multi-size features from the head network through corresponding graphic and stitching operations, thereby making full use of image information of different sizes.

[0019] Furthermore, the improved CTPN character detection algorithm includes: extracting features from the input image through multiple layers to capture edge and texture information in the image; using an improved region proposal network to generate text candidate boxes through a sliding mechanism and multi-geometric anchor boxes; introducing multiple LSTM networks to perform feature fusion and sequence modeling on the text candidate boxes to capture the contextual module relationships of the text; and connecting and correcting the detected unknown characters through reverse connections to ensure that the generated text is complete and without loss.

[0020] The improved CTC character correction algorithm includes: extracting the input image text through a feature extraction layer using a multi-layer structure and pooling operations to generate a highly abstract feature map; decoding the feature map by referencing the CTC loss function through a CTC decoding layer to align the input and output sequences and solve the problem of character order and content correction; and capturing the sequence information of the feature arrangement in the bidirectional LSTM processing sequence layer to capture the contextual relationship.

[0021] According to a second aspect of the present invention, an automatic crack identification system for concrete drainage pipes is provided, comprising:

[0022] The image acquisition and preprocessing module is used to acquire the original pipeline images captured by the ground inspection robot and preprocess them to obtain the preprocessed original pipeline images.

[0023] The crack detection module is used to detect cracks in the preprocessed original pipeline image using an improved HST-Net target detection algorithm, locate the coordinates of the minimum outline bounding rectangle of the crack, and obtain the crack location.

[0024] The distance recognition module is used to identify and record the distance of the crack location relative to the starting point in the current pipeline inspection task by using the improved CTPN character detection algorithm and CTC character correction algorithm.

[0025] The wireless communication module is used to send the crack images and distance information detected by the crack detection module and the distance recognition module to the cloud for storage.

[0026] According to a third aspect of the present invention, an automatic identification device for cracks in concrete drainage pipes is provided, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor loads and executes the computer program to implement the above-described method for automatic identification of cracks in concrete drainage pipes.

[0027] According to a fourth aspect of the present invention, a computer-readable storage medium is provided, wherein a computer program is stored in the computer-readable storage medium, the computer program being loaded and executed by a processor to implement the above-described method for automatic identification of cracks in concrete drainage pipes.

[0028] The beneficial effects of this invention are:

[0029] 1. This method achieves automatic crack identification in concrete drainage pipe images through an improved HST-Net target detection algorithm. It reduces reliance on manual inspection and improves detection efficiency and accuracy. This algorithm can not only accurately locate the coordinates of the minimum bounding rectangle of the crack's outline, but also effectively handle complex backgrounds and noise interference, ensuring high-accuracy crack detection.

[0030] 2. This method utilizes an improved CTPN character detection algorithm and CTC character correction algorithm to effectively identify and correct textual information related to crack locations. These algorithms, combining a region proposal network and a bidirectional LSTM network, accurately detect character boundaries and perform precise correction via CTC, significantly improving the accuracy of character recognition and correction. Particularly in the distance calculation of crack locations, these improvements ensure the reliability and accuracy of the data.

[0031] 3. The method of this application includes a multi-step image preprocessing workflow, utilizing techniques such as mean blurring, median filtering, grayscale conversion, adaptive threshold segmentation, and morphological transformation to significantly improve image quality. These preprocessing steps not only effectively remove noise and random noise from the image but also preserve the edge features of the cracks, enhancing the contrast between the cracks and the background, making subsequent image processing and crack detection more efficient and accurate.

[0032] 4. This application provides a comprehensive system including image acquisition, preprocessing, crack detection, and distance recognition. By integrating multiple modules working collaboratively, it achieves intelligent inspection of concrete drainage pipelines. Each module in the system employs advanced algorithms and technologies, which not only improves inspection efficiency but also significantly reduces the manual burden of inspection work, thereby enhancing the level of intelligence in pipeline maintenance.

[0033] 5. This application's system, through a wireless communication module, can transmit the results of crack detection and distance identification to a cloud server in real time for storage and analysis. This provides timely and reliable data support for pipeline maintenance, facilitating managers to take timely remedial measures and avoid potential safety hazards.

[0034] 6. This application also provides a readable storage medium storing a program that can be executed by a processor to implement the automatic identification method of the present invention. Through this form of program storage and retrieval, the system can be easily updated and maintained, enhancing the system's operability and practicality.

[0035] Additional aspects and advantages of this application will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of this application. Attached Figure Description

[0036] The above and / or additional aspects and advantages of this application will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein:

[0037] Figure 1 This is a flowchart of an automatic identification method for cracks in concrete drainage pipes according to an embodiment of the present invention;

[0038] Figure 2 This is a diagram of the improved HST-Net structure in an embodiment of the present invention;

[0039] Figure 3 This is a power supply circuit diagram of the power system in an automatic crack identification device for concrete drainage pipes according to an embodiment of the present invention;

[0040] Figure 4 This is a reset circuit diagram of the reset circuit system in an automatic crack identification device for concrete drainage pipes according to an embodiment of the present invention. Detailed Implementation

[0041] The present invention will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, the accompanying drawings show only the parts relevant to the present invention, and not all of the structures.

[0042] Those skilled in the art will understand that, unless otherwise stated, the singular forms “a,” “an,” “the,” and “the” used herein may also include the plural forms. It should be further understood that the word “comprising” as used in the specification of this application means the presence of the stated features, integers, steps, operations, elements, and / or components, but does not exclude the presence or addition of one or more other features, integers, steps, operations, elements, components, and / or combinations thereof.

[0043] It will be understood by those skilled in the art that, unless otherwise defined, all terms used herein (including technical and scientific terms) have the same meaning as commonly understood by one of ordinary skill in the art to which this application pertains. It should also be understood that terms such as those defined in general dictionaries should be understood to have the same meaning as in the context of the prior art, and should not be interpreted in an idealized or overly formal sense unless specifically defined as in the embodiments of this application.

[0044] This invention provides an automatic identification method, system, device, and storage medium for cracks in concrete drainage pipes. The method utilizes an improved HST-Net target detection algorithm to automatically identify cracks in concrete drainage pipe images. This reduces reliance on manual inspection and improves detection efficiency and accuracy. The algorithm can accurately locate the coordinates of the minimum bounding rectangle of the crack's outline and effectively handle complex backgrounds and noise interference, ensuring high-accuracy crack detection. The method also employs an improved CTPN character detection algorithm and CTC character correction algorithm to effectively identify and correct textual information related to crack locations. These algorithms, combined with a region proposal network and a bidirectional LSTM network, can accurately detect character boundaries and perform precise correction via CTC, significantly improving the accuracy of character recognition and correction. Particularly in the distance calculation of crack locations, these improvements ensure the reliability and accuracy of the data.

[0045] Example 1:

[0046] like Figure 1 As shown, this embodiment of the invention provides a method for automatic identification of cracks in concrete drainage pipes, including:

[0047] S100: Acquire the original pipeline image captured by the ground inspection robot, and preprocess it to obtain the preprocessed original pipeline image.

[0048] The preprocessing includes: applying mean blurring to the original pipeline image to smooth out noise and details; applying mean filtering to further remove random noise from the image; converting the color image to a grayscale image using grayscale processing; using an adaptive threshold segmentation method to separate the crack area from the background area in the image to enhance the contrast of the crack; using the Canny edge detection algorithm to extract edge information in the image and accurately locate the crack outline; and performing morphological transformation to remove noise interference and connect broken crack segments to make the crack features clearer and more coherent.

[0049] S200. The preprocessed original pipeline image is cracked using the improved HST-Net target detection algorithm to locate the coordinates of the minimum outline bounding rectangle of the crack and obtain the crack location.

[0050] like Figure 2 As shown, the improved HST-Net object detection algorithm comprises a backbone network, a neck network, and a head network. In the improved HST-Net backbone network, each top layer and the C3 module play a specific role, moving from initial feature extraction to more abstract high-level feature representations. The top convolutional layers extract low-level features, such as edges and textures, through top-level operations, while effectively reducing the size of the feature maps through stride and padding operations. The C3 module performs feature fusion after each texture layer, integrating feature information across different geometries to enhance the model's expressive power. The MACSP multi-head self-attention module weights and summarizes features from different locations using multiple attention heads, capturing important features globally and improving the network's feature extraction efficiency. Finally, the SPPF module further processes the features using multiple pooling operations to ensure the effectiveness of the features extracted from the input image.

[0051] It's important to note that in the improved HST-Net, the neck network acts as a bridge connecting the backbone and head networks. Its main functions include feature fusion, scaling, and enhancement. First, the graph layer and upsampling layer adjust the size and structure of the feature maps and enhance them. By concatenating the upsampled feature maps with feature maps from earlier layers in the backbone network, the neck network combines multi-level feature information. Subsequently, the C3 module processes these fused feature maps, preserving more detailed information and contextual relationships. Simultaneously, the context enhancement module uses a reweighting mechanism to increase the importance of features, further enhancing the model's feature representation capabilities and ensuring that the features passed to the head network are efficient and effective.

[0052] It should be noted that the task of the improved HST-Net head network is to perform the final processing and detection output of the fused and enhanced feature maps. Through corresponding graphing and stitching operations, multi-scale features from the head network are integrated, thus fully utilizing image information of different sizes. The C3 module plays a crucial role in this process, performing further supplementary operations to finalize the feature processing and ensure that the network focuses on important features. Finally, the detection layer uses predefined anchor boxes and a number of classes to perform specific detection and classification operations. This process includes not only location but also class determination, thereby achieving accurate identification and annotation of cracks in the preprocessed concrete drainage pipe image. With this design, HST-Net can efficiently detect various types of cracks in complex environments.

[0053] S300. The distance of the crack location relative to the starting point in the current pipeline inspection task is identified and recorded by using the improved CTPN character detection algorithm and CTC character correction algorithm.

[0054] It should be noted that the improved CTPN character detection algorithm and CTC character correction algorithm are described as follows:

[0055] The improved CTPN character detection algorithm is specifically designed for accurately detecting and extracting distance information to relevant crack locations in images of concrete drainage pipe inspections. First, the algorithm utilizes multiple convolutional layers to extract features from the input image, capturing edge and texture information, which helps in the initial identification of text boundaries. Based on this, an improved region proposal network is used to generate text candidate boxes through a sliding mechanism and anchor boxes of various geometric shapes. To improve detection accuracy, multiple bidirectional LSTM networks are introduced to perform feature fusion and sequence modeling on the text candidate boxes, thereby capturing the contextual relationships of the text. Finally, a back-connection step is used to connect and correct detected unknown characters, ensuring that the generated text is complete and without omissions. This series of improvements significantly enhances the continuity and accuracy of the CTPN algorithm in the text detection process.

[0056] Following character detection, the improved CTC character correction algorithm corrects the detected crack location text information to accurately calculate the distance between the crack location and the pipeline starting point. First, a feature extraction layer extracts the input image text using a multi-layer structure and pooling operations, generating a highly abstract feature map. Next, the CTC decoding layer decodes the feature map using the CTC loss function, aligning the input and output sequences and addressing character order and content correction issues. To enhance consistency, a sequence processing layer processes the feature arrangement sequence information using a bidirectional LSTM to capture contextual relationships. Additionally, a character correction module further corrects and concatenates the resulting characters, generating a continuous and accurate character sequence description. Finally, a location information parsing module transforms the character sequence into specific location information, ensuring accurate measurement of the distance between the crack location and the pipeline starting point. The CTC character correction algorithm ensures high accuracy in character recognition and localization, playing a crucial role in recording crack location information during pipeline inspection tasks.

[0057] S400: Export the identified data and store it in the cloud database.

[0058] Example 2:

[0059] This invention provides an automatic crack identification system for concrete drainage pipes, comprising:

[0060] The image acquisition and preprocessing module is used to acquire the original pipeline images captured by the ground inspection robot and preprocess them to obtain the preprocessed original pipeline images.

[0061] The crack detection module is used to detect cracks in the preprocessed original pipeline image using an improved HST-Net target detection algorithm, locate the coordinates of the minimum outline bounding rectangle of the crack, and obtain the crack location.

[0062] The distance recognition module is used to identify and record the distance of the crack location relative to the starting point in the current pipeline inspection task by using the improved CTPN character detection algorithm and CTC character correction algorithm.

[0063] The wireless communication module is used to send the crack images and distance information detected by the crack detection module and the distance recognition module to the cloud for storage.

[0064] Example 3:

[0065] This invention provides an automatic crack identification device for concrete drainage pipes, comprising:

[0066] The main controller uses an STM32F407VGT6 control chip, which is based on the ARM 32-bit Cortex-M4 architecture, has abundant resources, and operates at a frequency of up to 168MHz. The main controller is responsible for overall system management, providing functions such as data processing, communication, and machine vision control. By controlling the system's inspection operations, data collection, and transmission, the main controller achieves comprehensive control over the movement and work of the inspection robot. The main controller also includes a memory, a processor, and a computer program stored in the memory and executable on the processor. When the processor loads and executes the computer program, it implements all the steps in the automatic identification method for cracks in concrete drainage pipes described in Embodiment 1.

[0067] The drive system uses the TB6612FNG, which works in conjunction with the I / O ports of the STM32 microcontroller. This drive module can drive not only DC motors but also other high-power devices such as stepper motors and solenoid valves. The drive circuit design is simple, easy to integrate, and ensures the stability and reliability of the system.

[0068] The power supply system provides stable voltage and current to ensure the normal operation of all hardware components. This system uses a 12V DC power supply as the input. In this design, an AMS1117-3.3 voltage regulator chip is used to convert 12V to 3.3V to meet the power requirements of the STM32 main controller. Different power requirements are handled by the appropriate voltage regulator chip, providing stable 3.3V and 5V voltage outputs. The power supply circuit is as follows: Figure 3 As shown.

[0069] The USB-to-serial system uses the CP2102 chip for USB-to-serial conversion. This module not only facilitates information exchange and program downloading but also powers the control board, enhancing the system's flexibility and functionality.

[0070] The reset circuit system is designed to prevent system crashes or program errors, restoring the system to its initial state when necessary to ensure continuous operation. The design is simple and effective, facilitating integration into the overall hardware system. The reset circuit diagram is shown below. Figure 4 As shown.

[0071] The wireless communication system utilizes the ESP32-WROOM-32D module, which supports wireless communication in the 2.4GHz to 2.5GHz frequency band. It features a built-in TCP / IP protocol stack, enabling rapid wireless data transmission and system connection. Its high compatibility and built-in antenna optimize system communication performance, supporting various application scenarios.

[0072] The sensing system comprises multiple sensors for data acquisition and environmental perception. In this design, the image sensors used are the OV5640 visible light image sensor and the FLIR Lepton infrared image sensor. The OV5640 visible light image sensor has 5 megapixels, supports multiple image formats and high frame rate output, ensuring real-time image data acquisition and high-quality output. The FLIR Lepton infrared image sensor is used for temperature detection and environmental perception, suitable for low-light scenarios, and helps to efficiently identify and analyze the surrounding environment. This sensing system module can also dynamically adjust the vehicle's movement state to achieve intelligent inspection and crack detection.

[0073] Example 4:

[0074] This invention provides a computer-readable storage medium storing a computer program, which is loaded and executed by a processor to implement an automatic identification method for cracks in concrete drainage pipes as described in Embodiment 1.

[0075] It should be understood that although the steps in the flowcharts of the accompanying figures are shown sequentially as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless explicitly stated herein, there is no strict order restriction on the execution of these steps, and they can be executed in other orders. Moreover, at least some steps in the flowcharts of the accompanying figures may include multiple sub-steps or multiple stages. These sub-steps or stages are not necessarily completed at the same time, but can be executed at different times, and their execution order is not necessarily sequential, but can be performed alternately or in turn with other steps or at least some of the sub-steps or stages of other steps.

[0076] The above description is only a partial embodiment of this application. It should be noted that for those skilled in the art, several improvements and modifications can be made without departing from the principle of this application, and these improvements and modifications should also be considered within the scope of protection of this application.

Claims

1. A method for automatic identification of cracks in concrete drainage pipes, characterized in that, include: S100. Acquire the original pipeline image captured by the ground inspection robot and preprocess it to obtain the preprocessed original pipeline image. The preprocessing includes: performing mean blurring on the original pipeline image to smooth noise and details in the image; applying mean filtering to further remove random noise in the image; converting the color image to a grayscale image using grayscale processing; using an adaptive threshold segmentation method to separate the crack area from the background area in the image to enhance the contrast of the crack; using the Canny edge detection algorithm to extract edge information in the image and accurately locate the crack outline; and performing morphological transformation to remove noise interference and connect broken crack segments to make the crack features clearer and more coherent. S200. The preprocessed original pipe image is cracked using an improved HST-Net object detection algorithm to locate the coordinates of the minimum bounding rectangle of the crack and obtain the crack location. The improved HST-Net object detection algorithm includes: a backbone network part, including a top convolutional layer, a first C3 module, a MACSP multi-head self-attention module, and an SPPF module, used to extract features from the initial features to higher-level feature representations; a neck network part, including a graph layer, an upsampling layer, a second C3 module, and an environment enhancement module, used for feature fusion, scale transformation, and enhancement processing; and a head network part, including a detection layer and a third C3 module, used for final processing and detection output of the fused and enhanced feature maps. S300: The distance of the crack location relative to the starting point in the current pipeline inspection task is identified and recorded using an improved CTPN character detection algorithm and a CTC character correction algorithm. The improved CTPN character detection algorithm includes: extracting features from the input image through multiple layers to capture edge and texture information in the image; using an improved region proposal network to generate text candidate boxes through a sliding mechanism and multi-geometric anchor boxes; introducing multiple LSTM networks to perform feature fusion and sequence modeling on the text candidate boxes to capture the contextual module relationship of the text; connecting and correcting the detected unknown characters through reverse connections to ensure that the generated text is complete and without loss. The improved CTC character correction algorithm includes: extracting the input image text through a feature extraction layer using a multi-layer structure and pooling operations to generate a highly abstract feature map; decoding the feature map through a CTC decoding layer by referencing the CTC loss function to align the input and output sequences and solve the problem of character order and content correction; and capturing the sequence information of the feature arrangement in the bidirectional LSTM sequence processing layer to capture the contextual relationship. S400: Export the identified data and store it in the cloud database.

2. The method for automatic identification of cracks in concrete drainage pipes according to claim 1, characterized in that, The top-level convolutional layer is used to extract low-level features through top-level operations, while effectively reducing the size of the feature map through stride and padding operations; the first C3 module is used to perform feature fusion after each texture layer, integrating feature information on different geometric shapes to enhance the model's expressive power; the MACSP multi-head self-attention module is used to weight and summarize features at different locations through multiple attention heads, capturing important features in the global scope and improving the network's feature extraction efficiency; the SPPF module is used to further process features using multiple pooling operations to ensure that effective features are extracted from the input image.

3. The method for automatic identification of cracks in concrete drainage pipes according to claim 1, characterized in that, The graph layer and the upsampling layer are responsible for adjusting the size and structure of the feature maps and enhancing these features; the second C3 module is used to stitch the fused feature maps together to retain more detailed information and contextual relationships; the environment enhancement module is used to enhance the importance of features through a reweighting mechanism, further enhancing the feature representation ability of the model and ensuring the effectiveness of features passed to the head network.

4. The method for automatic identification of cracks in concrete drainage pipes according to claim 1, characterized in that, The detection layer is used to perform specific detection and classification operations using predefined anchor boxes and number of categories; the third C3 module is used to integrate multi-size features from the head network through corresponding graphic and stitching operations, thereby making full use of image information of different sizes.

5. An automatic crack identification system for concrete drainage pipes, characterized in that, include: The image acquisition and preprocessing module is used to acquire raw pipeline images captured by the ground inspection robot and preprocess them to obtain preprocessed raw pipeline images. The preprocessing includes: applying mean blurring to the raw pipeline images to smooth noise and details in the images; applying mean filtering to further remove random noise in the images; converting the color images to grayscale images using grayscale processing; using an adaptive threshold segmentation method to separate the crack areas in the images from the background areas to enhance the contrast of the cracks; using the Canny edge detection algorithm to extract edge information in the images and accurately locate the crack contours; and performing morphological transformations to remove noise interference and connect broken crack segments to make the crack features clearer and more coherent. A crack detection module is used to detect cracks in the preprocessed original pipe image using an improved HST-Net object detection algorithm, locating the coordinates of the minimum bounding rectangle of the crack to obtain the crack location. The improved HST-Net object detection algorithm includes: a backbone network part, including a top convolutional layer, a first C3 module, a MACSP multi-head self-attention module, and an SPPF module, used for moving from preliminary feature extraction to higher-level feature representation; a neck network part, including a graph layer, an upsampling layer, a second C3 module, and an environment enhancement module, used for feature fusion, scale transformation, and enhancement processing; and a head network part, including a detection layer and a third C3 module, used for final processing and detection output of the fused and enhanced feature maps. A distance recognition module is used to identify and record the distance of the crack location relative to the starting point in the current pipeline inspection task using an improved CTPN character detection algorithm and a CTC character correction algorithm. The improved CTPN character detection algorithm includes: extracting features from the input image through multiple layers to capture edge and texture information; using an improved region proposal network to generate text candidate boxes through a sliding mechanism and multi-geometric anchor boxes; introducing multiple LSTM networks to perform feature fusion and sequence modeling on the text candidate boxes to capture the contextual module relationships of the text; and connecting and correcting detected unknown characters through reverse connections to ensure that the generated text is complete and without loss. The improved CTC character correction algorithm includes: extracting the input image text through a feature extraction layer using a multi-layer structure and pooling operations to generate a highly abstract feature map; decoding the feature map using a CTC decoding layer with reference to the CTC loss function to align the input and output sequences and solve the problem of character order and content correction; and capturing the sequence information of feature arrangement in the bidirectional LSTM sequence processing layer to capture the contextual relationships. The wireless communication module is used to send the crack images and distance information detected by the crack detection module and the distance recognition module to the cloud for storage.

6. An automatic crack identification device for concrete drainage pipes, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that, When the processor loads and executes the computer program, it implements the automatic identification method for cracks in concrete drainage pipes as described in any one of claims 1-4.

Images