AI identification method and system for pipeline defects, medium and product

Abstract

A pipeline defect AI identification method and system, a medium and a product relate to the field of artificial intelligence and image processing, and the method comprises the following steps: controlling a detection robot to advance in a target pipeline in a first illumination mode, obtaining first image data and inputting the first image data to a defect identification model to obtain a defect position and a defect type; when the defect type is a preset type, calling a preset flash parameter to trigger flash, and acquiring a defect preview image of the defect position; inputting the defect preview image into a defect verification model corresponding to the defect type to obtain a defect existence probability; when the defect existence probability exceeds a preset probability threshold value, supplementary illumination parameters are determined according to the defect type, the detection robot is controlled to obtain second image data in a corresponding second illumination mode, and a defect detection result is determined; and after a defect detection report is generated based on the defect detection result, adjusting back to the first illumination mode. According to the invention, the cruising ability of pipeline detection can be improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the field of artificial intelligence and image processing, and particularly relates to an AI identification method and system for pipeline defects, a medium and a product. BACKGROUND

[0002] With the rapid development of urban infrastructure, underground pipe network systems are increasingly large and complex, and the safe operation of pipelines has become an important issue in urban management. During the long-term operation of the pipeline system, various types of damage such as cracks, corrosion, and deformation may occur. If these damages are not discovered and addressed in a timely manner, they may lead to serious accidents such as pipeline leaks and ruptures, affecting the normal operation of the city and threatening public safety.

[0003] In related technologies, pipeline detection mainly uses pipeline detection robot technology. The robot is equipped with lighting and CCTV camera systems for internal detection. The robot needs to be powered by an outdoor mobile power supply in remote working conditions, and uses its own lighting equipment and camera system to collect images of the inner wall of the pipeline. To ensure the reliability of the detection, the system uses a high image resolution to capture the subtle changes in the inner wall of the pipeline. During the movement of the detection robot in the pipeline, image data is continuously collected and transmitted in real time to the control terminal for analysis. Related technologies use an intelligent lighting control system to adapt to different detection scenarios through automatic exposure adjustment and light source control.

[0004] However, due to the variety of pipeline defect types, the lighting control strategy in related technologies tends to use a relatively high baseline lighting intensity to ensure the detection rate of different defect types. Although this lighting strategy improves the detection reliability, it causes unnecessary energy consumption when detecting pipe sections with no defects or only macro defects that do not require high lighting intensity. SUMMARY

[0005] The present application provides an AI identification method and system for pipeline defects, a medium and a product, which can improve the endurance of pipeline detection.

[0006] In a first aspect, the present application provides an AI identification method for pipeline defects, applied to a pipeline detection system. The method comprises: controlling a detection robot to travel in a target pipeline in a first lighting mode, and obtaining first image data; inputting the first image data into a defect identification model to obtain a defect position and a defect type of a suspected defect area; when the defect type is a preset type, calling preset flash parameters corresponding to the defect type, obtaining a defect preview image of the defect position while triggering flash in the preset flash parameters; inputting the defect preview image into a defect verification model corresponding to the defect type to obtain a defect existence probability; when the defect existence probability exceeds a preset probability threshold, determining a supplementary lighting parameter according to the defect type, and controlling the detection robot to obtain second image data in a corresponding second lighting mode based on the supplementary lighting parameter; inputting the first image data and the second image data into the defect identification model to obtain a defect detection result; and after generating a defect detection report based on the defect detection result, controlling the detection robot to adjust back to the first lighting mode.

[0007] In the above embodiment, the pipeline detection system performs regular detection in the energy-saving first lighting mode, triggers flash preview and supplementary lighting only when a suspected defect is found; the defect verification model is used for secondary confirmation, which avoids unnecessary supplementary lighting, and adjusts to the special second lighting mode for accurate detection in time after confirming the defect; the first lighting mode is restored immediately after detection is completed, which realizes precise control of lighting energy consumption and greatly improves the endurance of the detection robot.

[0008] In combination with some embodiments of the first aspect, in some embodiments, the step of inputting the first image data into the defect identification model to obtain the defect position and the defect type of the suspected defect area specifically comprises: performing standardization processing of brightness and contrast on the first image data to generate a standardized image; performing defect feature extraction on the standardized image based on a deep convolutional neural network to determine the suspected defect area and the defect position of the suspected defect area; and identifying and classifying the suspected defect area based on a support vector machine to determine the defect type of the suspected defect area.

[0009] In the above embodiment, the pipeline detection system performs standardization processing of brightness and contrast on the image data, which eliminates the image difference under different lighting conditions; the deep convolutional neural network is used to extract defect features, and the support vector machine is used for classification and identification, which constructs an accurate and reliable two-stage defect identification process, and significantly improves the accuracy and reliability of defect detection.

[0010] In some embodiments of the first aspect, after the step of inputting the first image data and the second image data into the defect identification model to obtain the defect detection result, the method further comprises: extracting a defect type and a geometric appearance parameter of the target defect in the defect detection result; calculating a defect influence coefficient of the target defect according to the defect type and the geometric appearance parameter in combination with a preset pipeline parameter; and calculating a defect quantification score of the target pipeline according to the defect influence coefficients of all defects in the pre-detection section of the target pipeline.

[0011] In the above embodiments, the pipeline network detection system calculates the defect influence coefficient based on the defect type and the geometric appearance parameter, and comprehensively evaluates in combination with the pipeline parameter, thereby realizing the quantification evaluation of the defect degree of the pipeline, providing a scientific basis for the pipeline maintenance decision, and effectively guiding the reasonable allocation of the maintenance priority.

[0012] In some embodiments of the first aspect, after the step of calculating the defect quantification score of the target pipeline according to the defect influence coefficients of all defects in the pre-detection section of the target pipeline, the method further comprises: when the defect quantification score is lower than a preset score threshold, calculating a third illumination mode of the detection robot in the post-detection section according to the defect types of all defects in the pre-detection section; and controlling the detection robot to travel in the target pipeline in the third illumination mode and collect images.

[0013] In the above embodiments, the pipeline network detection system dynamically adjusts the subsequent detection strategy according to the defect quantification score of the pre-detection section, and adopts the third illumination mode when the defect degree is large, thereby avoiding frequent illumination mode switching and improving the detection efficiency.

[0014] In some embodiments of the first aspect, when the defect type is the preset type, the step of acquiring the defect preview image of the defect position while triggering the flash in the preset flash parameter comprises: when the defect type is the preset type, determining the reflectivity and the gray value of the suspected defect area, and calculating a minimum illumination energy threshold; determining the flash intensity and the duration according to the minimum illumination energy threshold and the defect type, generating an intensity time sequence and a trigger time sequence of the flash illumination; triggering the flash according to the intensity time sequence and the trigger time sequence, and acquiring the defect preview image of the defect position.

[0015] In the above embodiments, the pipeline network detection system calculates the minimum illumination energy threshold based on the reflectivity and the gray value of the defect area, accurately controls the flash intensity and the duration, ensures the quality of the preview image, avoids excessive illumination, and optimizes the energy utilization efficiency.

[0016] In some embodiments of the first aspect, after the step of inputting the first image data and the second image data into the defect identification model to obtain the defect detection result, the method further comprises: collecting odometer data and an image shooting angle of the detection robot during travel; and calculating defect positioning data of the target defect according to the odometer data, the image shooting angle, and a defect position of the target defect in the defect detection result.

[0017] In the above embodiments, the pipe network detection system combines the odometer data and the image shooting angle to achieve accurate positioning of the defect position, thereby improving the accuracy of defect position marking and providing a reliable position reference for subsequent maintenance work.

[0018] In some embodiments of the first aspect, after the step of calculating the defect positioning data of the target defect according to the odometer data, the image shooting angle, and the defect position of the target defect in the defect detection result, the method further comprises: obtaining a pipe image captured by the detection robot during travel, extracting pipe structure feature points including joints and welds in the pipe image, and determining point distribution features of the pipe structure feature points; obtaining radius parameters of a pipe cross section and material change region data along the pipe; combining the odometer data, the point distribution features, the radius parameters, and the material change region data to determine positioning reference information of each image position in the pipe image; and correcting the defect positioning data according to the positioning reference information.

[0019] In the above embodiments, the pipe network detection system extracts the pipe structure feature points and the material change information to establish a complete positioning reference system, corrects the defect positioning data in combination with multi-dimensional positioning parameters, and further improves the accuracy of defect positioning, thereby providing strong support for accurate maintenance.

[0020] In the second aspect, the embodiments of the present application provide a pipe network detection system, which comprises: one or more processors and a memory; the memory is coupled to the one or more processors, and the memory is configured to store computer program code, the computer program code comprising computer instructions, and the one or more processors invoke the computer instructions to enable the pipe network detection system to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0021] In the third aspect, the embodiments of the present application provide a computer program product comprising instructions, which, when the computer program product is executed on the pipe network detection system, enable the pipe network detection system to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0022] In a fourth aspect, the embodiments of the present application provide a computer readable storage medium, including instructions, when the instructions run on the pipe network detection system, causing the pipe network detection system to perform the method described in the first aspect and any possible implementation manner of the first aspect.

[0023] It can be understood that the pipe network detection system provided in the second aspect, the computer program product provided in the third aspect, and the computer storage medium provided in the fourth aspect are all used to execute the method provided in the embodiments of the present application. Therefore, the beneficial effects that can be achieved are referred to the beneficial effects in the corresponding method, which will not be described here.

[0024] The one or more technical solutions provided in the embodiments of the present application have at least the following technical effects or advantages: 1. Since the multi-level lighting control strategy based on AI is adopted, including the energy-saving lighting mode in the conventional detection, the flash preview mode in the suspected defect, and the special lighting mode after confirming the defect, the precise regulation and on-demand distribution of lighting energy consumption are realized. The problem of excessive energy consumption caused by the continuous use of high-intensity lighting by the detection robot to ensure the detection rate in the prior art is effectively solved, thereby significantly improving the endurance of the detection equipment. The system performs preliminary screening on the image through the defect recognition model, triggers the flash preview only when a suspected defect is found, and enables the special lighting mode only after the secondary confirmation by the defect verification model. This multi-level lighting control mechanism ensures the detection quality while maximizing the energy consumption.

[0025] 2. Since the quantitative evaluation mechanism based on defect features is adopted, including the comprehensive analysis of the defect type, geometric appearance parameters, and pipeline parameters, a scientific pipeline state evaluation system is established. The problem of lack of basis for maintenance decision caused by the lack of unified evaluation standard in the prior art is effectively solved, thereby realizing the reasonable allocation of maintenance resources and the improvement of maintenance efficiency. By calculating the influence coefficient of each defect and combining the pipeline characteristics for weighted calculation, a quantitative score that objectively reflects the overall state of the pipeline is obtained, providing a reliable basis for the priority ranking and maintenance scheme of the pipeline maintenance.

[0026] 3. Since the intelligent flash parameter control strategy is adopted, including the lighting energy calculation based on the reflection characteristics of the defect area and the precise timing control mechanism, the optimization of the flash lighting effect is realized. The problem of energy waste or unstable image quality caused by the fixed flash lighting parameters in the prior art is effectively solved, thereby realizing the high efficiency and high quality of the preview image acquisition. By analyzing the reflectivity and gray value characteristics of the defect area, the minimum required lighting energy is calculated, and the intensity and duration of the flash are accurately controlled accordingly, ensuring that a clear defect preview image is obtained at the minimum energy consumption. BRIEF DESCRIPTION OF DRAWINGS

[0027] Figure 1 is a flowchart of an AI identification method for pipeline defects in an embodiment of the present application; Figure 2 is another flowchart of an AI identification method for pipeline defects in an embodiment of the present application; Figure 3 is a schematic diagram of an entity device structure of a pipeline network detection system in an embodiment of the present application. DETAILED DESCRIPTION

[0028] The terms used in the following embodiments of the present application are only for the purpose of describing specific embodiments and are not intended to be limiting of the present application. As used in the specification, the singular forms "a," "an" and "the" are intended to include the plural forms as well, unless the context clearly indicates otherwise. It will be further understood that the terms "and / or" as used herein refer to any or all possible combinations of one or more of the associated listed items.

[0029] Hereinafter, the terms "first" and "second" are only for the purpose of description and cannot be understood as implying or suggesting relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined with "first" and "second" can explicitly or implicitly include one or more of the features, and in the description of the embodiments of the present application, the meaning of "a plurality of" is two or more, unless otherwise specified.

[0030] To facilitate understanding, the application scenarios of the embodiments of the present application are introduced as follows.

[0031] A certain city is carrying out large-scale underground pipeline network census work. Traditional pipeline network detection methods mainly rely on manual visual inspection or simple camera recording, which is low in detection efficiency and easy to miss defects. Especially in the pipeline environment with complex lighting conditions, it is difficult for the detection personnel to accurately identify the early defects such as fine cracks and corrosion points on the pipe wall. At the same time, due to the lack of standardized detection process and intelligent analysis means, the detection results often depend on the experience judgment of the detection personnel, which has strong subjectivity and poor repeatability. In addition, the detection process needs to frequently adjust the lighting equipment, which not only increases the operation time, but also easily leads to missed detection and false detection, and cannot meet the requirements of large-scale pipeline census work on detection efficiency and accuracy.

[0032] In the related art, the detection of pipeline inner wall defects can be realized by using a detection robot with fixed lighting. This way needs manual real-time observation of images and judgment of defects, which is low in detection efficiency and easy to be affected by subjective factors. At the same time, due to the fixed lighting parameters, it is difficult to adapt to the imaging needs of different types of defects, which is easy to cause missed detection and false detection. The following introduces the scene of using the AI identification method for pipeline defects in the related art.

[0033] A certain detection unit uses a detection robot with fixed lighting for pipeline inspection. The robot is equipped with LED ring lights that illuminate at a constant power and captures pipeline images through high-definition cameras. Detection personnel view images in real-time through a remote control system, marking suspicious areas for closer observation. However, in actual detection, it is found that the fixed lighting mode is difficult to adapt to the imaging needs of different types of defects. For example, for crack-type defects, the existing lighting intensity is often insufficient to highlight the fine crack features; while for corrosion-type defects, uniform lighting can easily cause glare interference, affecting the identification of defect boundaries. Due to the lack of intelligent analysis and adaptive control capabilities, detection personnel need to frequently manually adjust camera parameters, greatly reducing detection efficiency.

[0034] However, by using the AI identification method for pipeline defects in the embodiments of the present application, through the combination of preliminary scanning in the first lighting mode and fine imaging in the targeted flash lighting, efficient and accurate defect detection is achieved, not only significantly improving detection efficiency, but also effectively reducing energy consumption. Especially in complex pipeline environments, the system can adaptively adjust the lighting parameters according to the defect features to ensure the best quality of defect images. The following describes the scene using the AI identification method for pipeline defects in the present application.

[0035] A certain pipeline network operation unit uses the intelligent detection system of the present application for pipeline detection. The system first performs preliminary scanning in the energy-saving first lighting mode, and when a suspected defect is detected, it automatically switches to the targeted flash mode for fine imaging. For example, when a suspected crack is detected, the system immediately calculates the optimal flash parameters and triggers the flash at an accurately controlled timing, successfully capturing clear crack details. At the same time, the system's deep learning model can automatically analyze image features, accurately identify defect types and assess their severity. This intelligent detection method not only improves detection efficiency, but also significantly reduces the rate of missed defects, providing reliable data support for pipeline maintenance decisions.

[0036] As can be seen, by using the AI identification method for pipeline defects in the embodiments of the present application, intelligent defect detection is achieved while effectively solving the problem of unstable image quality under traditional fixed lighting, thereby significantly improving detection efficiency and accuracy; through precise analysis of defect features by a deep learning model, combined with adaptive lighting control, the objectivity and reliability of the detection results are ensured.

[0037] For ease of understanding, the method provided by the present embodiment is described in the following flow. Please refer to Figure 1 , a flowchart of the AI identification method for pipeline defects in the embodiments of the present application.

[0038] S101、Control the detection robot to travel in the target pipeline in a first lighting mode, and acquire first image data.

[0039] In the formula, the detection robot represents an autonomous mobile device for traveling inside a pipeline and performing detection, including a driving system, a lighting system, and an image acquisition system; the first lighting mode represents a basic lighting configuration adopted by the detection robot, mainly including lighting intensity, lighting angle, and energy consumption, etc.; the target pipeline represents a pipeline section to be detected; and the first image data represents a sequence of images of the inner wall of the pipeline acquired by the detection robot under the first lighting mode.

[0040] When the pipeline detection system starts a detection task, it needs to preliminarily scan the target pipeline to acquire basic image data. Specifically, the pipeline detection system first controls the detection robot to enter the entrance of the target pipeline, starts the LED light source under the first lighting mode, sets the basic lighting intensity (usually 200-300 lumens) and the lighting angle (usually 120-degree fan-shaped irradiation) suitable for regular detection. The detection robot travels in the pipeline at a stable speed (usually 0.1-0.2 m / s), while continuously acquiring images of the inner wall of the pipeline through the high-definition camera carried, generating first image data with a resolution not less than 1920×1080 pixels.

[0041] In some embodiments, the lighting control and image acquisition of the detection robot can be realized in various ways: alternatively, the pipeline detection system automatically adjusts the parameter configuration of the first lighting mode according to the diameter and material characteristics of the pipeline, including: first acquiring the basic parameter information of the pipeline, then calculating the optimal lighting distance and irradiation angle, then determining the lighting intensity according to the reflection characteristics of the pipe wall, and finally generating a sequence of lighting control instructions. Alternatively, the pipeline detection system adopts an adaptive exposure control strategy, including: real-time monitoring of image brightness distribution, dynamic adjustment of camera exposure parameters, synchronous optimization of lighting output power, and ensuring stable image quality. It can be understood that other ways can also be used to realize the motion control and data acquisition of the detection robot, such as dynamically adjusting the travel speed according to the degree of pipeline curvature, etc., which are not limited here.

[0042] In actual application, the detection robot may be affected in detection effect due to insufficient battery energy. For this purpose, the pipeline detection system can adopt an intelligent energy-saving strategy: real-time monitoring of battery power and remaining detection mileage, dynamic calculation of maximum allowed lighting power; predicting the lighting needs of subsequent pipeline sections according to historical detection data, and reasonably allocating the remaining power; when detecting insufficient power, automatically reducing the energy consumption of unnecessary lighting to ensure the completion of the prescribed detection task. When the power is sufficient, the system will appropriately increase the lighting intensity to improve the image quality.

[0043] S102、Input the first image data into a defect recognition model to obtain the defect position and defect type of the suspected defect region.

[0044] wherein the defect recognition model represents a deep learning model for analyzing the pipeline image and recognizing defects, including an image preprocessing module, a feature extraction module, and a classification recognition module; the suspected defect region represents an image region in the first image data that is preliminarily identified as possibly having a defect; the defect position represents spatial coordinate information of the suspected defect region in the pipeline; and the defect type represents a defect category preliminarily determined according to a defect morphological feature, such as a crack, corrosion, deformation, etc.

[0045] After obtaining the first image data, the pipe network detection system needs to perform preliminary defect detection and classification. Specifically, the pipe network detection system first performs preprocessing on the first image data, including image enhancement, noise removal, and scale normalization operations. Then the processed image is input into the pre-trained defect recognition model, which extracts image features through a multi-layer convolutional neural network, locates the suspected defect region by combining the attention mechanism, and finally outputs the position coordinates and type probability distribution of each suspected defect region through the full connection layer.

[0046] In some embodiments, the recognition and positioning of defects can be achieved in various ways: alternatively, the pipe network detection system adopts a two-stage detection strategy, including: first generating candidate regions through a region proposal network, then using a classification network to finely classify the candidate regions, and finally screening out the final suspected defect region through a non-maximum suppression algorithm. Alternatively, the pipe network detection system combines a multi-scale feature fusion method, including: constructing a feature pyramid network, extracting feature maps at different scales, enhancing the detection ability of the model through feature fusion, and achieving accurate recognition of defects of different sizes. It can be understood that other ways can also be used to realize the detection and classification of defects, such as introducing a domain adaptation module to handle the feature differences of different pipeline materials, etc., which are not limited here.

[0047] It should be noted that the defect recognition model is trained using a large number of labeled pipeline image datasets, including normal pipeline images and images with different types of defects. Each image in the training data is labeled with accurate position coordinates, range bounding box, defect type, severity, and other information. The training uses an end-to-end supervised learning method, using a cross-entropy loss function to evaluate classification errors, and using IoU (intersection over union) and mean average precision (mAP) to evaluate positioning accuracy. The network parameters are optimized through the backpropagation algorithm until the performance indicators of the model on the validation set meet the preset requirements. The model uses a two-stage detection architecture, the first stage being a feature extraction network based on a deep convolutional neural network, which extracts multi-scale features of the image through multiple convolution operations and pooling operations; the second stage includes a region proposal network and a classification network, the region proposal network generates candidate defect regions, and the classification network classifies and regresses the positions of these regions. The model also integrates an attention mechanism that can adaptively focus on key areas in the image. In actual detection, the model receives a standardized pipeline image as input and outputs the position coordinates, bounding box, type probability distribution, and confidence score of each detected defect region. For each identified defect, the model also outputs geometric feature parameters for subsequent severity evaluation. The model's inference process supports real-time processing, allowing it to quickly analyze image streams captured by detection robots.

[0048] In actual applications, the complexity of the pipeline environment may result in false positives in the preliminary recognition results. To address this, the pipe network detection system uses a multi-feature verification mechanism: it establishes a defect feature library based on the characteristics of pipeline materials, verifies the reliability of the recognition results through feature matching; it trains an anomaly detection model using historical detection data to filter out recognition results that do not conform to statistical rules; when suspicious recognition results are detected, it triggers subsequent fine detection processes to ensure the accuracy of the detection results.

[0049] S103、In the case of a defect type being a preset type, a preset flash parameter corresponding to the defect type is called, and a defect preview image of the defect position is acquired while triggering the flash with the preset flash parameter.

[0050] Among them, the preset type represents a set of defect categories that require fine detection; the preset flash parameter represents the flash illumination parameters pre-configured for different defect types, including flash intensity, duration, and trigger timing; the defect preview image represents a high-quality image of the defect local area acquired under flash illumination.

[0051] After initially identifying suspected defects, the pipeline inspection system needs to perform detailed imaging of the localized areas to obtain more details about the defects. Specifically, when the identified defect type belongs to a preset type, the pipeline inspection system calls the corresponding flash parameter configuration, controls the flash of the inspection robot to be triggered at a specified time sequence, and simultaneously captures instantaneous images of the defect area through a high-speed camera module to ensure clear defect details are obtained under optimal lighting conditions.

[0052] In some embodiments, flash control and image acquisition can be achieved in various ways: Optionally, the pipeline inspection system employs an intelligent exposure control strategy, including: real-time analysis of the reflection characteristics of the defect area, dynamic adjustment of flash intensity and duration, optimization of exposure parameter configuration, and ensuring image quality. Optionally, the pipeline inspection system uses a multi-flash combination scheme, including: designing multiple flash sequences, acquiring multiple images under different lighting conditions, and generating a high dynamic range defect image through image fusion technology. It is understood that other methods can also be used to achieve fine imaging, such as combining light field camera technology to obtain depth information of the defect, etc., which are not limited here.

[0053] In practical applications, flash illumination can cause energy loss and acquisition delay. To address this, the pipeline inspection system implements intelligent trigger control: establishing a mapping relationship between defect types and optimal flash parameters to avoid over-illumination; using predictive algorithms to pre-calculate flash timing to reduce the waiting time of the inspection robot; and, under energy-constrained conditions, rationally allocating flash resources through a priority management mechanism to ensure the precise inspection of critical defects.

[0054] S104. Input the defect preview image into the defect verification model corresponding to the defect type to obtain the probability of defect existence.

[0055] Among them, the defect verification model refers to a deep learning model specifically designed to verify a particular type of defect, which is trained separately for different defect types; the defect existence probability represents the confidence score of the presence of the target type defect in the defect preview image, with a value ranging from 0 to 1; and the defect type correspondence represents the mapping relationship between different defect types and their dedicated verification models.

[0056] After acquiring defect preview images, the pipeline inspection system needs to perform precise defect verification. Specifically, the system selects a corresponding dedicated verification model based on the identified defect type and inputs the defect preview image into the model for analysis. The verification model calculates the probability value of the existence of a specified type of defect in the target area through fine-grained feature extraction and multi-dimensional feature analysis. This probability value reflects the reliability of the defect identification result and is used for dynamic adjustment of subsequent detection strategies.

[0057] In some embodiments, defect verification can be implemented in various ways: optionally, the pipe network detection system adopts an ensemble learning method, including: deploying multiple verification models with different structures, respectively calculating defect probabilities, and obtaining the final existence probability through weighted fusion to improve the reliability of the verification result. Optionally, the pipe network detection system combines knowledge distillation technology, including: using a large-scale pre-training model to extract general features, and migrating knowledge to a lightweight verification model to improve verification efficiency while ensuring accuracy. It can be understood that other ways can also be used to implement defect verification, such as introducing an expert rule system to assist verification, etc., which are not limited here.

[0058] It should be noted that the verification model is trained for each specific defect type, and the training data is a high-quality defect image sample confirmed by experts. The training label includes the accurate contour, internal structure feature and severity rating of the defect. The training process pays special attention to the robustness of the model, simulates the defect performance under different lighting conditions and imaging angles through data enhancement technology. The precision-recall curve (PR curve) and F1 score are used as evaluation criteria to ensure the high specificity of the model. The model uses a lightweight deep learning network structure, focusing on extracting discriminative features of specific types of defects. The network includes a feature extraction module and a verification module. The feature extraction module uses depth separable convolution to reduce computation, and the verification module calculates the probability of defect existence through a multi-layer perceptron. The model can also evaluate the uncertainty of the prediction result. The verification model receives the suspected defect local image obtained under flash preview as input, and outputs the probability value of the existence of the target type defect in the region and the uncertainty estimate of the prediction. When the prediction probability exceeds the preset threshold and the uncertainty is low, the system will trigger the subsequent fine detection process. The prediction result of the model is also used to optimize the flash parameters and supplementary lighting strategy.

[0059] In actual application, the feature difference of different types of defects may affect the verification accuracy. In this regard, the pipe network detection system adopts an adaptive verification strategy: a defect feature knowledge base is constructed, which contains typical feature descriptions of different types of defects; the feature extraction parameters are dynamically adjusted according to the defect type to highlight the key features; when the feature matching degree is insufficient, the verification threshold is automatically adjusted to ensure the reliability of the verification result.

[0060] S105、In some embodiments, defect verification can be implemented in various ways: optionally, the pipe network detection system adopts an ensemble learning method, including: deploying multiple verification models with different structures, respectively calculating defect probabilities, and obtaining the final existence probability through weighted fusion to improve the reliability of the verification result. Optionally, the pipe network detection system combines knowledge distillation technology, including: using a large-scale pre-training model to extract general features, and migrating knowledge to a lightweight verification model to improve verification efficiency while ensuring accuracy. It can be understood that other ways can also be used to implement defect verification, such as introducing an expert rule system to assist verification, etc., which are not limited here.

[0061] Wherein, the preset probability threshold represents the confidence threshold for triggering fine detection; the supplementary lighting parameter represents the fine detection lighting configuration determined according to the defect type, including lighting intensity, angle distribution and energy control, etc.; the second lighting mode represents a high-quality lighting scheme for fine detection.

[0062] After completing the defect verification, the pipe network detection system needs to conduct fine detection on high-probability defects. Specifically, when the probability of defects exceeds a preset threshold, the system calculates the optimal supplementary lighting parameters according to the detection requirements of the defect type. These parameters determine the specific configuration of the second lighting mode, including light source selection, intensity distribution, and illumination angle, etc. The detection robot switches to the second lighting mode according to these parameters and collects high-quality image data.

[0063] In some embodiments, fine detection can be achieved in various ways: optionally, the pipe network detection system uses a multi-light source cooperative lighting strategy, including: analyzing defect geometric features, planning the spatial layout of multiple light sources, highlighting defect features through phase coordination between light sources, and achieving all-around fine imaging. Optionally, the pipe network detection system adopts adaptive lighting control, including: real-time monitoring of image quality indicators, dynamic adjustment of lighting parameters, establishment of a feedback optimization mechanism, and ensuring optimal collection of image data. It can be understood that other ways can also be used to achieve fine detection, such as combining structured light technology to obtain three-dimensional information of defects, etc., which are not limited here.

[0064] In actual applications, different defects have different requirements for lighting conditions. In this regard, the pipe network detection system realizes intelligent lighting configuration: establishes a mapping relationship between defect types and optimal lighting schemes; dynamically adjusts the lighting strategy according to defect features; under energy constraints, ensures the detection quality of key defects through lighting resource scheduling.

[0065] S106, input the first image data and the second image data into a defect recognition model to obtain a defect detection result.

[0066] Among them, the defect detection result represents the final defect judgment information obtained after comprehensive analysis of the first image data and the second image data, including the accurate position of the defect, the type, the severity, and the geometric parameters, etc.; the data fusion analysis represents the process of cooperative processing of image data obtained under two lighting modes; the detection confidence represents the reliability evaluation index of the final detection result.

[0067] After obtaining the image data under the two lighting modes, the pipe network detection system needs to conduct comprehensive analysis to obtain the final conclusion. Specifically, the pipe network detection system first registers and preprocesses the two groups of image data to ensure the accuracy of the spatial correspondence. Then the processed image data is input into the defect recognition model at the same time, and the model extracts multi-scale features and fuses cross-modal features, fully utilizes the complementary information under different lighting conditions, and finally outputs high-confidence defect detection results. The detection result not only contains the basic information of the defect, but also includes detailed geometric parameters and damage evaluation data.

[0068] In some embodiments, data fusion analysis can be implemented in various ways: optionally, the pipe network detection system adopts a hierarchical fusion strategy, including: image alignment and enhancement at the pixel level, multi-dimensional feature extraction and fusion at the feature level, and comprehensive detection results of multiple sub-models at the decision level to generate the final defect judgment. Optionally, the pipe network detection system uses an attention-guided fusion method, including: calculating the importance weight of the image area under different lighting conditions, guiding the feature fusion process according to the weight, and highlighting the contribution of key information. It can be understood that other ways can also be used to implement result fusion, such as introducing a time sequence analysis module to process the associated information between consecutive frames, etc., which is not limited here.

[0069] In actual application, the image quality difference under different lighting modes may affect the fusion effect. In this regard, the pipe network detection system implements adaptive fusion control: an image quality evaluation mechanism is established to score the reliability of data from different sources; the fusion weight is dynamically adjusted according to the quality score; when the data quality of a certain mode is insufficient, the weight contribution of other data sources is increased to ensure the stability of the fusion result.

[0070] S107、After generating a defect detection report based on the defect detection result, the detection robot is controlled to adjust back to the first lighting mode.

[0071] Among them, the defect detection report represents a standardized document recording the detection process and results, including defect information, detection parameters, evaluation conclusions, etc.; the lighting mode adjustment represents the process of the detection robot returning to the basic lighting state of energy saving; the detection task state represents the management information of the current detection progress and subsequent detection plan.

[0072] After completing the detection analysis of the current defect, the pipe network detection system needs to adjust the lighting mode in time and prepare for subsequent detection. Specifically, the pipe network detection system first generates a detection report in a standard format according to the detection result, containing complete description information and evaluation suggestions of the defect. After completing the report generation, the system controls the detection robot to reduce the lighting power and returns to the first lighting mode with lower energy consumption to prepare for the detection of subsequent pipe sections. This dynamic lighting mode adjustment strategy can effectively balance the detection effect and energy consumption.

[0073] In some embodiments, mode switching management can be achieved in various ways: optionally, the pipe network detection system adopts a gradual adjustment strategy, including: calculating the difference between the current lighting parameters and the target parameters, designing a smooth transition curve, avoiding sudden changes affecting image acquisition through step-by-step adjustment, and ensuring detection continuity. Optionally, the pipe network detection system uses a predictive adjustment method, including: analyzing the pipe features of subsequent detection sections, planning the switching time of the lighting mode in advance, and achieving optimal allocation of lighting resources. It can be understood that other ways of implementing lighting adjustment can also be used, such as dynamically optimizing the lighting strategy according to the remaining battery power, etc., which are not limited here.

[0074] In practical applications, frequent lighting mode switching can affect system stability. To this end, the pipe network detection system implements intelligent mode management: a buffer mechanism for lighting mode switching is established to avoid repeated switching within a short period of time; the working state of each component of the system is monitored to ensure coordinated operation during mode switching; when an abnormal condition is detected, a protection strategy is automatically started to ensure safe and reliable operation of the detection equipment.

[0075] In the above embodiments, the basic lighting control and defect detection process is mainly described. In practical applications, the system can also optimize the lighting strategy based on historical detection data, establish a correlation model between defect features and lighting parameters, and achieve a more intelligent detection process. At the same time, through multi-source data fusion and positioning calibration technology, the accuracy of defect location positioning is further improved. The scene of this embodiment is supplemented as follows.

[0076] The intelligent detection system continuously accumulates experience and optimizes performance in long-term operation. The system analyzes historical detection data to establish a mapping relationship between defect features and optimal lighting parameters, and can automatically predict the best lighting strategy according to the pipe material and environmental conditions. For example, when a change in pipe material is detected, the system will smoothly adjust the lighting parameters to ensure the continuity of detection quality. In addition, the system can also dynamically plan the detection path according to the defect distribution characteristics, and preferentially check high-risk areas. Through this continuously optimized intelligent detection method, not only the detection efficiency is further improved, but also the optimal allocation of detection resources is realized, laying a foundation for large-scale pipe network intelligent operation and maintenance.

[0077] After combining the above scenarios, the method provided by the present embodiment is further described in more detail. Please refer to Figure 2 , another flowchart of the AI recognition method for pipe defects in the embodiments of the present application.

[0078] S201, control the detection robot to travel in the target pipe in a first lighting mode, and acquire first image data.

[0079] Referring to step S101, the pipe network detection system controls the detection robot to enter the target pipe, enables the energy-saving first lighting mode, continuously travels in the pipe, and collects first image data of the inner wall of the pipe.

[0080] S202, input the first image data into the defect recognition model to obtain the defect position and defect type of the suspected defect area.

[0081] Referring to step S102, the pipe network detection system inputs the collected first image data into the pre-trained defect recognition model, analyzes and processes the image, and identifies the area position where the defect may exist and the preliminary judged defect type.

[0082] In some embodiments, the pipe network detection system adopts a two-stage image processing and defect recognition process to improve detection accuracy; that is, the pipe network detection system performs standardization processing of brightness and contrast on the first image data to generate a standardized image; based on a deep convolutional neural network, the standardized image is subjected to defect feature extraction to determine the suspected defect area and the defect position of the suspected defect area; based on a support vector machine, the suspected defect area is identified and classified to determine the defect type of the suspected defect area.

[0083] Among them, the standardization processing represents a preprocessing process of normalizing the brightness and contrast of the image; the standardized image represents an image with uniform brightness and contrast characteristics after preprocessing; the deep convolutional neural network represents a multi-layer neural network model for image feature extraction; the suspected defect area represents an image area where the defect may exist which is preliminarily identified; the support vector machine represents a machine learning model for defect type classification; the defect type represents a defect category label obtained after classifying the suspected defect area.

[0084] After obtaining the first image data, the pipe network detection system needs to perform multi-stage intelligent analysis and processing. Specifically, the pipe network detection system first performs standardization preprocessing on the image data, converts the images collected under different lighting conditions into standardized images with consistent characteristics through histogram equalization and gamma correction, etc. Then, a pre-trained deep convolutional neural network model is used to extract features from the standardized image, multi-scale features of the image are extracted through multi-layer convolution operation and pooling operation, and the suspected defect area is located by combining the attention mechanism. Finally, the extracted features are input into a support vector machine classifier, the suspected defect area is accurately classified based on the trained decision boundary, and a defect type label is output.

[0085] In some embodiments, image analysis and defect recognition can be achieved in various ways: optionally, the pipe network detection system adopts a multi-stage processing strategy, including: first, adaptive threshold segmentation is performed to obtain preliminary candidate regions, then morphological operations are used to optimize the region boundaries, then multi-dimensional feature vectors are extracted, and finally the classification accuracy is improved through an ensemble learning method. Optionally, the pipe network detection system uses a deep learning enhancement method, including: constructing a feature pyramid network to extract multi-scale features, applying a region proposal network to generate candidate boxes, accurately positioning the defect area through an instance segmentation network, and combining transfer learning to improve the model generalization ability. It can be understood that other ways can also be used to realize image analysis, such as a hybrid method combining traditional image processing algorithms and deep learning models, etc., which are not limited here.

[0086] In practical applications, the complex lighting conditions inside the pipeline can cause unstable image preprocessing results. In this regard, the pipe network detection system implements an adaptive preprocessing strategy: an image quality evaluation model is established to evaluate the preprocessing effect in real time; the processing parameters are dynamically adjusted according to the image features; when preprocessing abnormalities are detected, a multi-scale fusion method is used to improve the processing robustness. For example, for strong light reflection areas, the system will automatically reduce the contrast enhancement degree; for dark details, the local enhancement intensity is increased.

[0087] It should be noted that the training data of this image quality evaluation model comes from a large number of pipeline detection images with expert quality scores. The training labels include objective indicators such as image sharpness, contrast, signal-to-noise ratio, and comprehensive subjective scores. The training process uses a multi-task learning method to simultaneously optimize multiple quality evaluation indicators. Mean squared error (MSE) and rank consistency are used as loss functions to ensure that the evaluation results of the model are consistent with manual judgments. The model uses a multi-branch network structure, including an image feature extraction branch and a quality evaluation branch. The feature extraction branch uses a lightweight convolutional network to extract multi-dimensional features of the image, and the quality evaluation branch calculates various quality indicators through a fully connected layer. The model integrates traditional image quality evaluation algorithms such as sharpness evaluation and noise estimation to form an end-to-end quality evaluation system. The model evaluates the quality of the captured images in real time during the detection process, with the input being the original image and the output including multiple dimensions of quality scores and a comprehensive score. These evaluation results are used to dynamically adjust the camera parameters and lighting parameters to ensure image capture quality. When image quality abnormalities are detected, the model triggers a re-capture mechanism to ensure the reliability of the detection data.

[0088] S203、In the case of a defect type being a preset type, the reflectivity and gray value of the suspected defect area are determined, and the minimum illumination energy threshold is calculated.

[0089] Wherein, the reflectivity represents the index of the pipeline surface's ability to reflect light; the gray value represents the brightness level of the pixel points in the image; the minimum illumination energy threshold represents the lowest light energy level required to obtain a clear defect image; and the optical characteristic parameter represents a physical quantity describing the response characteristics of the material to light.

[0090] After confirming the defect type, the pipeline network detection system needs to evaluate the optical characteristics of the target area to optimize the lighting strategy. Specifically, the pipeline network detection system first analyzes the image data of the suspected defect area, extracts the surface reflectivity characteristics and gray value distribution. Then, according to the imaging requirements of the defect type and combined with the optical characteristics of the material, the minimum illumination energy threshold that can highlight the defect characteristics is calculated. This threshold serves as a reference value for subsequent flash parameter setting, ensuring that the detection requirements are met while avoiding excessive illumination.

[0091] In some embodiments, optical characteristic evaluation can be achieved in various ways: optionally, the pipeline network detection system uses a region adaptive analysis method, including: dividing multiple local areas to calculate reflectivity characteristics, constructing a reflectivity gradient map, identifying material change boundaries, and generating a regional optical characteristic distribution map. Optionally, the pipeline network detection system uses multi-scale photometric measurement technology, including: analyzing surface optical characteristics at different spatial scales, establishing a hierarchical characteristic description model, and extracting main influencing factors. It can be understood that other ways can also be used to implement characteristic evaluation, such as introducing spectral analysis technology to evaluate the wavelength response characteristics of the material, etc., which are not limited here.

[0092] In practical applications, the complexity of the pipeline surface state may affect the accuracy of the optical characteristic evaluation. To this end, the pipeline network detection system implements a robust evaluation strategy: establishes a surface state classification model to identify factors such as contamination and scaling; dynamically adjusts the evaluation parameters according to the surface state; when the evaluation result is unstable, a conservative strategy is used to set the lighting parameters to ensure detection reliability.

[0093] S204, determine the flash intensity and duration according to the minimum illumination energy threshold and the defect type, and generate the intensity time sequence and trigger time sequence of the flash lighting.

[0094] Wherein, the flash intensity represents the instantaneous power level of the flash lighting; the duration represents the illumination duration period of a single flash; the intensity time sequence represents the change curve of the flash intensity over time; and the trigger time sequence represents the time control sequence of the flash trigger.

[0095] After obtaining the minimum illumination energy threshold, the pipe network detection system needs to design a detailed flash illumination scheme. Specifically, the system first calculates the flash intensity value that can meet the imaging requirements according to the minimum illumination energy threshold and the defect type characteristics. Then the illumination duration suitable for this type of defect is determined, and a complete timing scheme containing intensity changes and trigger timing is generated. This scheme needs to ensure that the flash illumination is precisely synchronized with image acquisition to obtain the best quality defect image.

[0096] In some embodiments, flash control can be achieved in various ways: optionally, the pipe network detection system adopts a multi-level intensity control strategy, including: designing a three-section intensity curve of preheating-main illumination-afterglow, optimizing energy utilization efficiency, and improving imaging quality. Optionally, the pipe network detection system uses an intelligent trigger optimization method, including: analyzing the motion state of the detection robot, predicting the optimal trigger timing, and realizing motion compensation and image stabilization. It can be understood that other ways of implementing flash control can also be used, such as dynamically adjusting flash parameters according to environmental temperature, etc., which are not limited here.

[0097] In practical applications, the timing control accuracy of flash illumination directly affects image quality. To this end, the pipe network detection system implements high-precision timing management: establishes a hardware-level synchronous trigger mechanism to ensure microsecond-level synchronization of flash and image acquisition; monitors the response characteristics of flash devices and compensates for device delays; when detecting synchronization abnormalities, automatically re-plans the timing scheme.

[0098] S205, triggering the flash with the intensity timing and the trigger timing, and acquiring a defect preview image of the defect position.

[0099] Among them, intensity timing triggering means performing flash intensity control according to a preset time sequence; trigger timing execution means activating the flash device according to the planned time point; the defect preview image means a high-quality local image obtained under the condition of precisely controlled flash illumination.

[0100] After completing the flash parameter design, the pipe network detection system needs to precisely execute flash control and acquire images. Specifically, the system first loads the designed intensity timing and trigger timing to the flash control module to ensure that the execution is ready. Then when the detection robot reaches the specified position, the flash trigger is executed according to the timing scheme, and high-speed image acquisition is started at the same time to capture the transient image of the defect area under the best illumination condition. The entire process needs to ensure microsecond-level control accuracy to ensure precise synchronization of flash illumination and image acquisition.

[0101] In some embodiments, precise imaging can be achieved in various ways: optionally, the pipe network detection system adopts a multi-frame synthesis technique, including: high-speed acquisition of multiple frames of images in a single flash process, image reconstruction using a super-resolution algorithm, and improvement of the detail resolution capability. Optionally, the pipe network detection system uses real-time feedback control, including: monitoring the flash output characteristics, dynamically adjusting the control parameters, compensating for environmental factors, and ensuring stable lighting effects. It can be understood that other ways can also be used to achieve precise imaging, such as combining light field camera technology to obtain depth information, etc., which are not limited here.

[0102] In practical applications, the instability of flash lighting can cause fluctuations in image quality. In this regard, the pipe network detection system implements an adaptive control strategy: a flash characteristic model is established to predict the response behavior of the device; fluctuations are dynamically compensated based on real-time monitoring data; when an abnormal response is detected, the control parameters are automatically adjusted or a backup solution is triggered.

[0103] S206, input the defect preview image into the defect verification model corresponding to the defect type to obtain a defect existence probability.

[0104] Referring to step S104, the pipe network detection system will input the acquired defect preview image into the verification model specially trained for this type of defect, analyze the image features through deep learning algorithm, and calculate the probability value of the existence of the target type of defect in this area.

[0105] S207, when the defect existence probability exceeds a preset probability threshold, determine the supplementary lighting parameters according to the defect type, and control the detection robot to acquire second image data in the corresponding second lighting mode based on the supplementary lighting parameters.

[0106] Referring to step S105, when the defect existence probability exceeds the preset threshold, the pipe network detection system will set appropriate supplementary lighting parameters according to the detection requirements of the defect type, control the detection robot to switch to the corresponding second lighting mode to reacquire image data.

[0107] S208, input the first image data and the second image data into the defect recognition model to obtain a defect detection result.

[0108] Referring to step S106, the pipe network detection system will input the image data acquired in the two lighting modes into the defect recognition model at the same time, analyze the features from multiple angles and multiple dimensions, and obtain the final defect detection result.

[0109] S209, extract the defect type and geometric appearance parameters of the target defect in the defect detection result.

[0110] Wherein, the target defect represents a confirmed defect that needs to be quantitatively evaluated; the defect type represents an attribute classification of the defect; and the geometric appearance parameter represents a quantitative index describing the morphological characteristics of the defect, including size, depth, shape, etc.

[0111] After obtaining the defect detection result, the pipe network detection system needs to extract key parameters for subsequent evaluation. Specifically, the system first locates the accurate position of the target defect from the detection result, and then extracts the type label and detailed geometric feature parameters of the defect. These parameters include the spatial size, depth distribution, shape characteristics, etc. of the defect, providing basic data support for subsequent impact evaluation.

[0112] In some embodiments, parameter extraction can be achieved in various ways: optionally, the pipe network detection system adopts a multi-dimensional feature analysis method, including: constructing a defect feature vector, extracting morphological features, texture features and statistical features, and establishing a complete defect description model. Optionally, the pipe network detection system uses depth estimation technology, including: reconstructing the three-dimensional structure of the defect using multi-view images, calculating geometric parameters, and evaluating the spatial distribution characteristics of the defect. It can be understood that other ways can also be used to achieve parameter extraction, such as using structured light scanning to obtain high-precision topographic data, etc., which are not limited here.

[0113] In practical applications, the complex morphology of the defect may affect the accuracy of parameter extraction. In this regard, the pipe network detection system implements robust parameter extraction: establishes a defect morphology classification system, selects appropriate feature extraction methods for different types; uses multi-source data fusion to improve parameter estimation accuracy; and when there is uncertainty in parameter extraction, uses interval estimation to represent the result.

[0114] S210, according to the defect type and the geometric appearance parameter, combining the preset pipe parameters to calculate the defect impact coefficient of the target defect.

[0115] Wherein, the preset pipe parameters represent a set of engineering parameters describing the basic characteristics of the pipe, including pipe diameter, wall thickness, material strength, etc.; the defect impact coefficient represents a scoring index quantifying the influence degree of a single defect on the safety of the pipe; and the geometric appearance parameter represents the quantitative characteristics of the defect, such as shape, size, depth, etc.

[0116] After obtaining the defect features, the pipe network detection system needs to evaluate the hazard degree of a single defect. Specifically, the system first reads the preset pipe basic parameters, including the design parameters and material performance indicators of the pipe. Then, combined with the type characteristics and geometric parameters of the defect, the influence degree of the defect on the structural integrity of the pipe is calculated through a mechanical model. Finally, a standardized defect impact coefficient is generated, which comprehensively considers the severity of the defect and the carrying capacity of the pipe.

[0117] In some embodiments, the influence coefficient calculation can be implemented in various ways: optionally, the pipe network detection system adopts a multi-factor evaluation method, including: constructing a comprehensive evaluation model considering stress concentration, strength reduction and life loss, calculating the contribution value of each influencing factor, and obtaining the weighted fusion influence coefficient. Optionally, the pipe network detection system uses numerical simulation technology, including: establishing a locally refined finite element model, analyzing the stress distribution change caused by defects, and evaluating the structural safety margin. It can be understood that other ways can also be used to realize influence evaluation, such as damage tolerance analysis based on fracture mechanics, etc., which are not limited here.

[0118] In practical applications, the coupling effect of defects and pipeline parameters may affect the evaluation accuracy. For this purpose, the pipe network detection system implements an evaluation strategy considering the coupling effect: a defect-pipeline interaction model is established to analyze the influence of geometric and material nonlinearity; the evaluation parameters are adjusted according to the service conditions; when abnormal coupling phenomena are found, a refined evaluation process is started.

[0119] S211、According to the defect influence coefficients of all defects in the pre-detection section of the target pipeline that has been detected, the quantitative score of the defects of the target pipeline is calculated.

[0120] Wherein, the pre-detection section represents the pipeline section that has been detected; the quantitative score of the defects represents a comprehensive index reflecting the overall health condition of the pipeline; and the influence coefficient set represents the influence coefficient dataset of all defects in a specific pipe section.

[0121] After obtaining the influence coefficient of a single defect, the pipe network detection system needs to evaluate the overall condition of the pipeline. Specifically, the system first collects the influence coefficient data of all defects found in the pre-detection section, and establishes a defect distribution database. Then, considering the spatial distribution characteristics and interaction effects of defects, the overall quantitative score of the pipeline is calculated through a scoring model. This score reflects the degree of cumulative damage of the pipeline and can be used to guide the dynamic adjustment of the detection strategy.

[0122] In some embodiments, the quantitative score can be implemented in various ways: optionally, the pipe network detection system adopts an analytic hierarchy process, including: establishing a multi-level scoring index system, considering defect density, distribution law and cumulative effect, and obtaining the final score through hierarchical weighting. Optionally, the pipe network detection system uses statistical inference technology, including: analyzing the statistical characteristics of defect distribution, evaluating the degradation trend of the pipeline, and predicting the probability of potential risks. It can be understood that other ways can also be used to realize quantitative scoring, such as probability evaluation based on reliability theory, etc., which are not limited here.

[0123] In practical applications, the mutual influence between defects can cause scoring deviation. To this end, the pipe network detection system implements a scoring mechanism considering group effect: a defect group analysis model is established to evaluate the combined effect of defect clusters; the distribution characteristics of defects in the axial and circumferential directions of the pipeline are considered; when a highly clustered defect group is detected, the scoring weight of the section is increased.

[0124] In some embodiments, the pipe network detection system dynamically adjusts the lighting strategy based on defect scoring to improve detection efficiency; that is, when the defect quantification score is lower than the preset scoring threshold, the pipe network detection system calculates the third lighting mode of the detection robot in the post-detection section according to the defect types of all defects in the pre-detection section; controls the detection robot to travel in the target pipeline in the third lighting mode and collect images.

[0125] Among them, the defect quantification score represents a comprehensive index reflecting the overall health status of the pipeline; the preset scoring threshold represents the scoring threshold value triggering the adjustment of the lighting strategy; the pre-detection section represents the pipeline section that has completed detection; the post-detection section represents the pipeline section to be detected; the third lighting mode represents a comprehensive enhanced lighting configuration scheme for efficient detection of severe defect areas.

[0126] After completing the pre-detection section detection and calculating the quantification score, the pipe network detection system needs to optimize the subsequent detection strategy according to the damage degree of the pipeline. Specifically, the pipe network detection system first compares the calculated defect quantification score with the preset threshold value, and when the score is lower than the threshold value, it indicates that the pipeline damage degree is large. Then the system analyzes all defect types and their distribution characteristics found in the pre-detection section, designs a comprehensive lighting scheme that takes into account the detection needs of multiple types of defects, and forms a third lighting mode. This mode has high lighting intensity and optimized light source configuration, which can meet the imaging needs of different types of defects at the same time, avoiding frequent switching of lighting modes. Finally, the detection robot is controlled to continue to perform the detection task of the subsequent section using this mode, improving the detection efficiency.

[0127] In some embodiments, lighting strategy optimization can be achieved in various ways: optionally, the pipe network detection system uses a severity-based lighting configuration method, which includes: analyzing the severity distribution of defects in the pre-detection section, determining the lighting needs of the main defect types, designing a combination of lighting parameters that can meet multiple detection needs at the same time, and achieving a balance between detection efficiency and lighting effect. Optionally, the pipe network detection system uses predictive optimization technology, which includes: establishing a defect spread model to predict the damage degree of the subsequent section, designing an adaptive lighting scheme, and adjusting the detection strategy in advance to avoid frequent adjustments during the detection process. It can be understood that other ways can also be used to realize lighting optimization, such as a zoned lighting control strategy based on the health status of the pipeline, etc., which are not limited here.

[0128] In practical applications, the spatial correlation of pipe damage degree can affect the effectiveness of the lighting strategy. To this end, the pipe network detection system implements zoned adaptive lighting control: a spatial distribution model of pipe damage degree is established to analyze the continuity characteristics of defect severity; a reasonable lighting mode switching threshold is set according to the damage degree; when a significant change in damage degree is detected, the lighting strategy is adjusted in a timely manner to adapt to the new detection requirements. For example, when entering a severely damaged area, the system will smoothly transition to a higher intensity lighting mode, ensuring detection effectiveness while avoiding the delay caused by frequent switching.

[0129] In some embodiments, the pipe network detection system can use a multi-source data fusion positioning method to accurately locate defects; that is, the pipe network detection system can collect odometer data and image shooting angles of the detection robot during travel; according to the odometer data, image shooting angles, and defect location of the corresponding target defect in the defect detection result, the defect positioning data of the target defect is calculated.

[0130] Among them, the odometer data represents the travel distance and speed information of the detection robot; the image shooting angle represents the spatial orientation parameter of the camera; the defect positioning data represents the coordinate information describing the accurate spatial position of the defect in the pipeline; the target defect represents the confirmed defect that needs to be accurately positioned.

[0131] During the travel of the detection robot, the pipe network detection system needs to collect motion parameters in real time for defect positioning. Specifically, the pipe network detection system first records the displacement, speed, and acceleration data during travel through the odometer sensor mounted on the robot. At the same time, the cloud angle of the camera is collected through the attitude sensor to obtain accurate orientation information when the image is taken. Then, combined with these motion parameters and relative position information in the defect detection result, the absolute spatial coordinates of the target defect in the pipeline are calculated through coordinate transformation and error compensation, and standardized defect positioning data is generated.

[0132] In some embodiments, accurate positioning can be achieved in various ways: optionally, the pipe network detection system uses a multi-sensor fusion method, including: integrating an inertial measurement unit to obtain attitude data, deploying an optical encoder to measure the rotation angle, using a visual odometer to estimate the relative displacement, and fusing multi-source data through a Kalman filter algorithm. Optionally, the pipe network detection system uses visual positioning technology, including: extracting the corresponding relationship of feature points in consecutive images, reconstructing the camera motion trajectory, and combining structure light ranging data to achieve high-precision defect spatial positioning. It can be understood that other ways of measuring position can also be used, such as ultrasonic or laser ranging positioning, etc., which are not limited here.

[0133] In practical applications, the slip and jitter of the robot can cause cumulative errors in the odometer data. To this end, the pipe network detection system implements a multi-source data correction mechanism: a motion state evaluation model is established to detect abnormal motion in real time; image feature matching is used for position correction; when the cumulative error exceeds the threshold, the global position relocation process is triggered. For example, when the robot slips, the system will temporarily switch to a vision-based positioning method to ensure positioning accuracy.

[0134] Further, in some embodiments, the pipe network detection system will build a complete positioning reference system to avoid inaccurate defect positioning caused by single factors such as robot travel problems, i.e., the pipe network detection system will obtain pipe images taken by the detection robot during travel, extract pipe structure feature points including joints and welds in the pipe images, determine the point distribution characteristics of the pipe structure feature points; obtain the radius parameter of the pipe cross section and the material change area data along the pipe; combine the odometer data, point distribution characteristics, radius parameters, and material change area data to determine the positioning reference information of each image position in the pipe image; and correct the defect positioning data according to the positioning reference information.

[0135] Among them, the pipe structure feature point represents a fixed landmark component on the pipe, such as a joint, a weld, etc.; the point distribution characteristic represents the spatial arrangement rule of the feature point; the radius parameter represents the geometric dimension information of the pipe cross section; the material change area data represents the material property change information along the pipe; the positioning reference information represents the reference data set for position calibration; and the positioning correction represents the process of accurately calibrating the defect position based on the reference information.

[0136] After obtaining the preliminary positioning result, the pipe network detection system needs to establish a complete positioning reference system for accurate calibration. Specifically, the pipe network detection system first analyzes the pipe image sequence collected by the detection robot, identifies permanent markers such as joints and welds on the pipe through image processing algorithms, extracts the position information of these feature points and analyzes their distribution rules. At the same time, the basic parameter information of the pipe is obtained, including the cross-sectional radius and material distribution data. Then, the odometer data, feature point distribution, pipe parameters, and other multi-source information are fused to build a complete spatial reference coordinate system. Finally, based on this reference system, the previously calculated defect position is calibrated to obtain a more accurate positioning result.

[0137] In some embodiments, the positioning reference system can be constructed in various ways: Optionally, the pipeline network detection system adopts a multi-level reference point management strategy, including: establishing a hierarchical feature point database, performing reliability rating on feature points, constructing topological relationships between feature points, and realizing position calibration based on multiple references. Optionally, the pipeline network detection system uses dynamic reference update technology, including: real-time evaluation of the effectiveness of reference points, dynamic adjustment of reference point weights, and establishing a sliding reference window to ensure the real-time performance and accuracy of the positioning reference system. It is understood that other methods can also be used to construct the reference system, such as feature point detection and matching methods based on deep learning, etc., which are not limited here.

[0138] In practical applications, the complex structure of pipelines can lead to uneven distribution of reference feature points. To address this, the pipeline network detection system implements adaptive reference point management: it establishes a feature point quality assessment model to calculate the reliability score of each feature point; in areas with sparse feature points, it increases the weight contribution of image features; and when a section lacks reliable reference points, it constructs virtual reference points through interpolation. For example, in pipe bends or material transition zones, the system comprehensively uses multiple types of reference features and dynamically adjusts their weights based on their reliability to ensure the continuity and stability of positioning accuracy. Furthermore, the system records and analyzes historical data for positioning corrections to optimize reference point selection strategies and calibration algorithm parameters, continuously improving the performance of the positioning system.

[0139] S212. After generating a defect detection report based on the defect detection results, control the detection robot to switch back to the first lighting mode.

[0140] Referring to step S107, after generating the defect detection report, the pipeline inspection system will control the inspection robot to return to the energy-saving first lighting mode and continue to perform the inspection tasks of subsequent pipeline sections.

[0141] It's important to note that generating a defect detection report requires integrating multi-dimensional detection data: First, key information is extracted from the defect detection results, including the precise spatial location, type classification, geometric dimensions (such as length, width, and depth), and severity rating of the defect. Then, combined with basic pipeline parameters (such as pipe diameter, wall thickness, and material), the impact coefficient of each defect is calculated to assess its potential impact on the pipeline's structural integrity. The system also analyzes the spatial distribution characteristics of defects, identifying high-risk areas and areas with high defect density. Finally, this information is organized according to a standardized format to generate a comprehensive detection report containing text descriptions, data charts, and defect images.

[0142] The lighting mode adjustment is a fine transition process. The pipe network detection system first evaluates the difference between the current lighting parameters and the first lighting mode, and then designs a smooth parameter transition curve to avoid sudden changes in light intensity that interfere with subsequent detection. While reducing the intensity of the main lighting, the system will monitor the image quality indicators in real time to ensure that the basic image acquisition quality is maintained during the transition process. If the detection robot is about to enter a new detection section, the system will calculate the optimal lighting parameter configuration in advance to achieve seamless switching. For different types of lighting devices (such as LED light sources, auxiliary lighting, etc.), the system will coordinate the timing of power changes to ensure smooth energy transition.

[0143] The pipe network detection system will ensure the traceability and comparability of the detection results through a standardized report generation process. Through a reasonable lighting mode switching strategy, the continuity of the detection quality is ensured, and the optimization of energy utilization is achieved, providing a reliable foundation for subsequent detection tasks and historical data analysis. For example, after a section of pipe detection is completed, the report generated by the system not only contains the crack location and size data, but also predicts the development trend of the crack based on the pipe service time and load conditions. At the same time, the system will adjust the lighting strategy in advance according to the expected conditions of the subsequent detection section, such as appropriately increasing the baseline lighting intensity before entering a curved section of the pipe to ensure the continuous and efficient performance of the detection work.

[0144] In the embodiments of the present application, the detection method combining multi-mode cooperative lighting and intelligent analysis can adaptively adjust the lighting strategy according to the defect characteristics, and realize accurate defect recognition and positioning. Specifically, the system first uses the energy-saving first lighting mode for preliminary scanning, and automatically switches to the targeted flash mode for fine imaging when a suspected defect is found. Through deep learning model analysis of image features and multi-source data fusion technology for position calibration, the system effectively solves the problems of unstable image quality, low detection efficiency, and results dependent on human experience in traditional fixed lighting methods; the system can also continuously optimize the detection strategy through historical data analysis, establish the mapping relationship between defect characteristics and optimal lighting parameters, and thus realize the intelligentization, standardization and high efficiency of the detection process. Not only does it improve the detection accuracy and reduce energy consumption, but it also provides reliable technical support for large-scale pipe network intelligent operation and maintenance.

[0145] The pipe network detection system in the embodiments of the present application will be described from the perspective of hardware processing. Please refer to Figure 3 , which is a schematic diagram of an entity device structure of the pipe network detection system in the embodiments of the present application.

[0146] It should be noted that Figure 3 The structure of the pipe network detection system shown is only an example and should not limit the function and use range of the embodiments of the present application.

[0147] like Figure 3 As shown, the pipeline monitoring system includes a CPU 301, which can perform various appropriate actions and processes according to a program stored in ROM 302 or a program loaded from storage section 308 into RAM 303, such as executing the methods described in the above embodiments. RAM 303 also stores various programs and data required for system operation. The CPU 301, ROM 302, and RAM 303 are interconnected via bus 304. I / O interface 305 is also connected to bus 304.

[0148] The following components are connected to I / O interface 305: input section 306 including audio input devices, push-button switches, etc.; output section 307 including liquid crystal display (LCD) and audio output devices, indicator lights, etc.; storage section 308 including hard disks, etc.; and communication section 309 including network interface cards such as LAN (Local Area Network) cards, modems, etc. Communication section 309 performs communication processing via a network such as the Internet. Drive 310 is also connected to I / O interface 305 as needed. Removable media 311, such as disks, optical disks, magneto-optical disks, semiconductor memories, etc., are installed on drive 310 as needed so that computer programs read from them can be installed into storage section 308 as needed.

[0149] In particular, according to embodiments of the present invention, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of the present invention include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing computer programs for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 309, and / or installed from removable medium 311. When the computer program is executed by CPU 301, it performs the various functions defined in the present invention.

[0150] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of the present invention. Each block in a flowchart or block diagram may represent a module, program segment, or portion of code, which contains one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those shown in the drawings.

[0151] Specifically, the pipe network detection system of the embodiment includes a processor and a memory, and the memory stores a computer program. When the computer program is executed by the processor, the AI identification method of the pipe defect provided in the above embodiment is implemented.

[0152] As another aspect, the application further provides a computer readable storage medium. The storage medium can be included in the pipe network detection system described in the above embodiments, or can exist independently without being assembled into the pipe network detection system. The storage medium carries one or more computer programs. When the one or more computer programs are executed by a processor of the pipe network detection system, the pipe network detection system implements the AI identification method of the pipe defect provided in the above embodiments.

[0153] The above embodiments are only used to illustrate the technical solutions of the present application, but not limit the present application; although the present application has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that they can still modify the technical solutions recorded in the foregoing embodiments, or make equivalent replacements to some technical features; and these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present application.

[0154] In the above embodiments, according to the context, the term "when" can be interpreted as meaning "if" or "after" or "in response to determining" or "in response to detecting". Similarly, according to the context, the phrase "upon determining" or "if detecting (the stated condition or event)" can be interpreted as meaning "if determining" or "in response to determining" or "upon detecting (the stated condition or event)" or "in response to detecting (the stated condition or event)".

Claims

1. An AI identification method for pipeline defects, characterized in that, The method is applied to a pipe network detection system, and the method comprises: controlling a detection robot to travel in a target pipeline in a first illumination mode to obtain first image data; inputting the first image data into a defect identification model to obtain a defect position and a defect type of a suspected defect region; when the defect type is a preset type, calling a preset flash parameter corresponding to the defect type, and obtaining a defect preview image of the defect position while triggering a flash in the preset flash parameter; inputting the defect preview image into a defect verification model corresponding to the defect type to obtain a defect existence probability; when the defect existence probability exceeds a preset probability threshold, determining a supplementary illumination parameter according to the defect type, and controlling the detection robot to obtain second image data in a corresponding second illumination mode based on the supplementary illumination parameter; inputting the first image data and the second image data into the defect identification model to obtain a defect detection result; after generating a defect detection report based on the defect detection result, controlling the detection robot to adjust back to the first illumination mode.

2. The method of claim 1, wherein, The step of inputting the first image data into a defect identification model to obtain a defect position and a defect type of a suspected defect region specifically comprises: performing standardization processing of brightness and contrast on the first image data to generate a standardized image; performing defect feature extraction on the standardized image based on a deep convolutional neural network to determine a suspected defect region and a defect position of the suspected defect region; performing identification classification on the suspected defect region based on a support vector machine to determine a defect type of the suspected defect region.

3. The method of claim 1, wherein, After the step of inputting the first image data and the second image data into the defect identification model to obtain a defect detection result, the method further comprises: extracting a defect type and a geometric appearance parameter of a target defect in the defect detection result; calculating a defect influence coefficient of the target defect according to the defect type and the geometric appearance parameter in combination with a preset pipeline parameter; calculating a defect quantitative score of the target pipeline according to defect influence coefficients of all defects in a pre-detection section of the target pipeline that has been detected.

4. The method of claim 3, wherein, After the step of calculating a defect quantitative score of the target pipeline according to defect influence coefficients of all defects in a pre-detection section of the target pipeline, the method further comprises: when the defect quantitative score is lower than a preset score threshold, calculating a third illumination mode of the detection robot in a post-detection section according to defect types of all defects in the pre-detection section; controlling the detection robot to travel in a target pipeline in the third illumination mode and collect images.

5. The method of claim 1, wherein, The step of, when the defect type is a preset type, calling a preset flash parameter corresponding to the defect type, and obtaining a defect preview image of the defect position while triggering a flash in the preset flash parameter specifically comprises: when the defect type is a preset type, determining reflectivity and a gray value of the suspected defect region, and calculating a minimum illumination energy threshold; determine a flash intensity and a duration according to the minimum illumination energy threshold and the defect type, generate an intensity timing and a trigger timing of flash illumination; trigger a flash with the intensity timing and the trigger timing, and acquire a defect preview image of the defect position.

6. The method of claim 1, wherein, After the step of inputting the first image data and the second image data into the defect identification model to obtain a defect detection result, the method further comprises: acquiring odometer data and an image shooting angle of the detection robot during travel; calculating defect positioning data of a target defect according to the odometer data, the image shooting angle, and a defect position of the target defect in the defect detection result.

7. The method of claim 6, wherein, After the step of calculating defect positioning data of a target defect according to the odometer data, the image shooting angle, and a defect position of the target defect in the defect detection result, the method further comprises: acquiring a pipeline image shot by the detection robot during travel, extracting pipeline structure feature points including a joint and a welding point in the pipeline image, and determining point distribution characteristics of the pipeline structure feature points; acquiring a radius parameter of a pipeline cross section and material change region data of the pipeline along a pipeline length; combining the odometer data, the point distribution characteristics, the radius parameter, and the material change region data to determine positioning reference information of each image position in the pipeline image; correcting the defect positioning data according to the positioning reference information.

8. A pipe network detection system characterised by, The pipeline network detection system comprises one or more processors and a memory; the memory is coupled to the one or more processors, the memory is configured to store computer program code, the computer program code comprises computer instructions, and the one or more processors invoke the computer instructions to enable the pipeline network detection system to perform the method according to any one of claims 1-7.

9. A computer-readable storage medium comprising instructions, characterized in that, The instructions enable the pipeline network detection system to perform the method according to any one of claims 1-7 when the instructions run on the pipeline network detection system.

10. A computer program product, characterised in that, The computer program product enables the pipeline network detection system to perform the method according to any one of claims 1-7 when the computer program product runs on the pipeline network detection system.

Images