Pipeline dredging auxiliary pricing method based on image recognition technology
By using deep learning algorithms based on image recognition technology and Hough circle detection, pipeline silt is automatically identified and dredging costs are calculated, solving the problems of low efficiency and large errors in traditional manual calculations, and achieving efficient and accurate pipeline dredging cost management.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Patents(China)
- Current Assignee / Owner
- YANGTZE THREE GORGES TECHNOLOGY & ECONOMY DEVELOPMENT CO LTD
- Filing Date
- 2024-06-07
- Publication Date
- 2026-06-09
AI Technical Summary
In the existing technology, the cost calculation of municipal pipeline dredging projects mainly relies on traditional manual methods, which have problems of low efficiency and large errors, affecting the accuracy and efficiency of the calculation.
A pipeline dredging auxiliary pricing method based on image recognition technology is adopted. A target detection model is constructed through deep learning algorithm and combined with Hough circle detection algorithm to automatically identify pipeline silt and calculate dredging cost. The model is optimized by EfficientDet deep learning algorithm and Soft-NMS algorithm to achieve efficient and accurate cost calculation.
It improves the accuracy and efficiency of pipeline dredging cost calculation, reduces human error, and builds an intelligent pipeline dredging cost calculation system that supports visualization of the calculation process and result verification. It is applicable to cost formulas for different projects.
Smart Images
Figure CN118644682B_ABST
Abstract
Description
Technical Field
[0001] This invention belongs to the field of municipal pipeline engineering cost management, specifically involving a pipeline dredging auxiliary pricing method based on image recognition technology. Background Technology
[0002] As a core component of urban infrastructure, municipal pipeline systems play a crucial role in transporting water resources, disposing of waste, and preventing flooding. Over time, silt, garbage, and other debris inevitably accumulate inside these pipelines, severely hindering the normal flow of water and potentially leading to urban flooding and other disasters, negatively impacting the quality of life for city residents and the sustainable development of the city. Furthermore, the accumulation of pollutants within the pipelines can also cause environmental pollution problems, posing a potential threat to the ecological environment. Therefore, regular inspection, repair, and maintenance of municipal pipelines are particularly urgent.
[0003] Pipeline dredging is a key aspect of pipeline maintenance. Removing and clearing silt from the inside of pipelines effectively prevents sewage overflow, ensures unobstructed flow, and thus guarantees the normal operation of the city. Before carrying out pipeline repair work, dredging is necessary to make the internal condition of the pipeline clearly visible. This allows for further assessment of the pipeline's operational status and determination of appropriate repair methods. Dredging extends the lifespan of pipelines and lays the foundation for subsequent repair work.
[0004] Pipeline dredging cost calculation is a crucial component of municipal engineering management. However, current pipeline dredging cost calculations still largely rely on traditional manual methods, which suffer from errors in judgment and low efficiency. Before calculating the dredging volume, staff need to visually assess the pipeline's condition before and after dredging. This method is not only time-consuming and labor-intensive but also highly prone to errors, affecting the accuracy and efficiency of pipeline dredging cost calculations. Summary of the Invention
[0005] This invention addresses the characteristics and needs of municipal pipeline dredging project pricing by proposing an image recognition network-based auxiliary pricing method for pipeline dredging. This method, through the introduction of deep learning technology, assists financial managers in cost calculation, reducing errors in manual calculations and further improving the accuracy of dredging cost calculations. It provides an intelligent and efficient solution for the cost management of municipal pipeline dredging projects, thus solving the problems of low efficiency and large errors associated with traditional manual calculation methods.
[0006] To achieve the above-mentioned technical features, the objective of this invention is as follows: a pipeline dredging auxiliary pricing method based on image recognition technology, comprising the following implementation steps:
[0007] Step S1: Use CCTV inspection equipment to capture pipeline video, extract pipeline cross-section images containing silt from the video, select pipeline cross-section images with suitable angle and clarity through image screening and sampling, preprocess the images to improve image quality by removing noise, enhancing contrast and sharpening edges, and use annotation tools to annotate to obtain a dataset for training the target detection network.
[0008] Step S2: Based on the target detection network, construct a target detection model for pipe cross-section images, train the model using the dataset obtained in step S1, adjust the model parameters to optimize the model recognition effect, and improve the model accuracy until it meets the target detection requirements.
[0009] Step S3: Use the target detection model trained in step S2 to identify the pipe cross-section image, identify the silt and mark its target detection box, and use the Hough circle detection algorithm to obtain the pipe cross-section circle parameter set.
[0010] Step S4: Based on the target detection box information of the silt obtained in step S3, determine the Hough circle where the endpoint of the detection box is located, and use the above information to calculate the height ratio of the silt cross section to the pipe cross section to determine the degree of siltation in the pipe.
[0011] Step S5: Substitute the obtained data into the pipeline dredging pricing formula for different pipe diameters and different degrees of siltation to further obtain the pipeline dredging cost.
[0012] The CCTV inspection equipment mentioned in step S1 is a common model, and the collected pipe cross-section images are taken from the vertical pipe cross-section.
[0013] The object detection network is based on the EfficientDet deep learning algorithm and includes a weighted bidirectional feature pyramid network module for feature fusion. This module provides rich contextual information for multi-scale and high-level semantic information processing through top-down and bottom-up paths. A composite scaling strategy for network optimization is employed, simultaneously adjusting the network depth, width, and input resolution to achieve optimal resource utilization efficiency. It also includes a Soft-NMS algorithm to reduce overlapping detection boxes and a decoder module for output classification and regression, transforming the extracted and fused feature maps into accurate bounding boxes and class probabilities to complete the object detection task.
[0014] IoU and GIOU are used as quality assessment criteria to more accurately measure the degree of overlap between the detected bounding box and the ground truth bounding box.
[0015] The method for identifying pipeline sediment in step S3 is as follows: the target detection model is called to identify the image, and the corresponding detection box is drawn at the same time as the sediment is identified, so as to intuitively display its area in the image; the method for detecting Hough circles in the pipeline image is as follows: the circle detection algorithm based on the improved Hough gradient transform is called to process the image and identify the circular structure in the pipeline cross section.
[0016] The method for determining the Hough circle containing the two ends of the bottom of the pipeline sediment detection frame in step S4 is as follows: For each detected Hough circle, calculate the distance from the center of each circle to the two ends of the detection frame, and determine whether it is equal to the radius of the Hough circle. If the distance from the center of the circle to the two ends of the sediment detection frame is equal to or close to its radius, then the Hough circle is the Hough circle containing the sediment.
[0017] The method for determining the degree of siltation in the pipeline in step S4 is as follows: using the coordinate information of the center of the Hough circle and the coordinate information of the two ends of the bottom of the siltation detection frame, the average distance of the line connecting the two ends relative to the center of the circle is calculated, and the ratio of the average distance to the diameter is obtained to further evaluate the degree of siltation in the pipeline.
[0018] The method for calculating the cost in step S5 is as follows: There are specific pricing formulas for different pipe diameters and different degrees of siltation. The obtained percentage data, pipe diameter data and pipe section length are substituted into the corresponding pricing formulas to finally calculate the cost of pipeline dredging.
[0019] The pipeline dredging cost calculation model is connected to the front end of the cost calculation system via a network. It can display the acquired image data, cost calculation process, and calculation results on the front-end interface. This not only assists financial management personnel in quickly calculating the pipeline dredging cost, but also facilitates the management personnel in verifying the calculation process and results, providing double assurance for the accuracy of cost calculation.
[0020] The present invention has the following beneficial effects:
[0021] 1. Using deep learning algorithms to obtain information on pipe silt and assess the degree of siltation can effectively eliminate the subjective and uncertain factors of manual visual identification, and the algorithm has a high recognition accuracy.
[0022] 2. The target detection algorithm can be upgraded quickly. Data augmentation and other methods can effectively expand the dataset used to train the algorithm. As the training data increases and the model is updated, the accuracy of target detection can be further improved.
[0023] 3. It can connect the front-end and management terminal to form an intelligent calculation system for pipeline dredging costs. The identification results and calculation process can be automatically sent to the front-end, making the calculation process visible and facilitating financial management personnel to review the calculation process, thereby further ensuring the accuracy and reliability of the calculation results.
[0024] 4. The specific cost calculation process can be flexibly applied according to the pricing formulas specified for different projects, making it highly applicable. Attached Figure Description
[0025] The present invention will be further described below with reference to the accompanying drawings and embodiments.
[0026] Figure 1 This is a schematic diagram of the technical process of the present invention.
[0027] Figure 2 This is a schematic diagram of the Hough circle detection process.
[0028] Figure 3 This is a schematic diagram of target detection for pipeline cross-sections and Hough circles.
[0029] In the figure: 1: silt detection frame; 2: Hough circle containing endpoints M and N; 3: silt; M and N are the endpoints of the silt detection frame; C is the center of the Hough circle. Detailed Implementation
[0030] The technical solutions adopted in the embodiments of the present invention will be described in detail below with reference to the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments included in the present invention. Those skilled in the art can combine the embodiments described herein with other embodiments. All other implementation methods based on the embodiments of the present invention, modified and refined without creative effort, should fall within the scope of protection of the present invention.
[0031] like Figure 1 As shown, a pipeline dredging auxiliary pricing method based on image recognition technology includes the following steps:
[0032] Step S1: Use QV and CCTV equipment to capture images of pipes containing silt. Through image screening and sampling, select pipe cross-section images with suitable angles and clarity. Use annotation tools to annotate the images to obtain a training dataset.
[0033] Step S2: Construct a pipeline cross-section target detection model based on the target detection network, train the model using the dataset constructed in step S1, and improve the model performance by iteratively adjusting parameters and optimizing training strategies, thereby optimizing the model accuracy to the level required for application.
[0034] Step S3: Use the trained target detection model to identify the pipe cross-section image, identify the silt and mark its detection box, M and N are the bottom two ends of the detection box; use Hough circle detection to obtain the pipe cross-section circle group, calculate the distance between the center C of each Hough circle and the two ends M and N of the detection box, and determine the Hough circle where the endpoint of the detection box is located by comparing the radius of the Hough circle with the distance.
[0035] Step S4: Using the pixel information of M, N and the center of the Hough circle obtained in step S3, calculate the average distance d from the line connecting points M and N to the center of the Hough circle. By calculating the ratio of the height of the silt to the diameter of the pipe section, determine the degree of siltation in the pipe.
[0036] Step S5: Substitute the siltation level information from step S4 into the corresponding pipe diameter and siltation ratio pricing formula to calculate the final pipeline dredging cost.
[0037] Preferably, the municipal pipeline network described in this invention consists of circular pipes. Before shooting, the pipe sections are divided based on the distribution and direction of the pipes. The QV and CCTV equipment travel and shoot in a direction perpendicular to the circular cross-section of the pipe, effectively avoiding problems such as poor data acquisition results caused by pipe bends, undulations, etc. during the data acquisition process.
[0038] Preferably, in step S1, when constructing the training dataset for the object detection network, in order to ensure the quality of the dataset, the cropped photos need to be screened and preprocessed. By removing noise, enhancing contrast, and sharpening edges, a high-quality and high-accuracy training dataset can be obtained, avoiding the impact of insufficient light, excessive image noise, and other factors on the model training process.
[0039] Preferably, in step S2, an EfficientDet network is used to construct an object detection model, and deep learning algorithms are used to perform deep analysis on the input image. In the feature extraction stage, the model uses EfficientNet as the backbone network, extracting useful visual features from the input image through an effective feature extraction mechanism. These features are then fed to the Bi-directional Feature Pyramid Network (BiFPN) module for feature fusion to capture multi-scale contextual information. For network optimization, a Compound Scaling strategy is employed, simultaneously adjusting the network depth, width, and input resolution to achieve optimal resource utilization efficiency. This includes the Soft-NMS algorithm for reducing overlapping detection boxes, which more effectively reduces overlapping detection boxes to improve object detection accuracy. A decoder module for output classification and regression transforms the extracted and fused feature maps into accurate object boxes and class probabilities to complete the object detection task. Furthermore, IoU (Intersection over Union) and GIOU (Generalized Intersection over Union) are used as quality evaluation criteria to more accurately measure the degree of overlap between the detected boxes and the ground truth boxes.
[0040] During the prediction phase, the model decodes the fused features using a decoder and makes a target classification decision for each pixel, thereby generating a label map containing the siltation detection boxes. The decoding process includes upsampling to restore the resolution of the feature map and feature fusion to obtain more accurate detection results.
[0041] Preferably, model accuracy can be improved by adjusting training parameters and optimizing the training dataset. With algorithm updates and upgrades, training with existing datasets can yield more efficient object detection models. The key steps in building an object detection model are as follows:
[0042] (1) Weighted feature fusion: In BiFPN, feature maps of different scales are fused by weighted summation, which can usually be expressed as:
[0043] ;
[0044] in: It outputs the feature map. , For the input feature map, , Its corresponding weighting coefficient.
[0045] (2) Hybrid scaling: EfficientDet uses a hybrid scaling method to unify the model's size configuration, specifically involving adjustments to network width, depth, and resolution, which can usually be expressed as:
[0046] ;
[0047] in: The symbols represent the scaled network width, depth, and resolution, respectively. , , It is the raw data. , , It is the scaling factor.
[0048] (3) Output of object detection: Predicting the location and class of the object from the feature map, which usually involves anchor boxes and non-maximum suppression (NMS). Its loss function consists of two parts, namely classification loss and detection box regression loss, which can usually be expressed as:
[0049] ;
[0050] in: It is the total loss. It is classification loss. It is the detection box regression loss.
[0051] Preferably, in step S3, the target model outputs the bounding box of the sediment and its corresponding category label, identifying the location of the detected sediment, drawing the detection box of the sediment on the original image, and simultaneously obtaining the coordinate information of the endpoints of the detection box. Hough circle detection can be performed simultaneously with target detection. Hough circle detection can directly utilize the HoughCircles function in OpenCV, ensuring that the coordinate system of the target detection and the coordinate system of the Hough circle detection are in the same coordinate system, facilitating further analysis and data processing. The main process of Hough circle detection is as follows: Figure 2 As shown. Its key steps include the following aspects:
[0052] (1) Edge detection: Edge detection is performed using the Canny algorithm.
[0053] (2) Gradient calculation: For each pixel in the image ( x , y The Sobel operator is used to calculate the magnitude of the gradient. G And direction Theta.
[0054] ;
[0055] in, , They are points ( x , y )exist x and y Gradient of direction.
[0056] (3) Hough transform voting: for each point in the edge image ( x , y ), for all points in Hough space along its gradient direction Theta ( ) a , b To vote:
[0057] ;
[0058] (4) Accumulator space: The Hough space is a two-dimensional accumulator array N( a , b ), used to record each possible center of the circle ( a , b The number of votes is calculated. For each center of a vote, the value at the corresponding position in the accumulator is incremented.
[0059] (5) Extraction of center and radius: by searching the accumulator space N( a , b The location of the circle's center is determined by finding the local maxima in the circle.a , b ), the corresponding radius r c It can be calculated using the following formula:
[0060] ;
[0061] Here, (x,y) are the original edge points used during voting.
[0062] Preferably, the method for determining the Hough circle containing the endpoints M and N of the detection frame in step S4 is as follows:
[0063] Based on the shooting angle and the nature of the detection frame, the two bottom endpoints of the detection frame are the connection points between the silt and the pipe cross-section circle. Step S3 can obtain the coordinates M (X) of the endpoints of the two detection frames. a ,Y a ), N (X b ,Y b The Hough Circles transform returns the coordinates of the centers and radii of all detected circles. In OpenCV, the HoughCircles function returns a triple array (x...). i ,y i ,r i ), where (x i ,y i (r) represents the coordinates of the center of the circle. i Let be the radius. The distances between the center of the circle and its two endpoints are:
[0064] ;
[0065] .
[0066] Further calculate the difference between the two distances and the radius of the Hough circle. Repeat this calculation process. When the difference is minimized, the circle is the Hough circle containing the silt, with its center at C and its coordinates at (x, y, y). c ,y c ), radius is r c .
[0067] Preferably, in step S4, the method for determining the degree of siltation is as follows: calculate the average distance d from the line connecting points M and N to the center of the Hough circle, and then calculate the ratio of the siltation height to the pipe cross-sectional diameter. The formula for calculating the average distance d is:
[0068] ;
[0069] Furthermore, it is stipulated that the line MN is below the center of this Hough circle. d c For positive, above the center of the Hough circle dc If the value is negative, then the formula for calculating the degree of siltation is:
[0070] ;
[0071] That is, the ratio of the deposition thickness to the pipe diameter.
[0072] In step S5, different pricing formulas are used for different pipe diameters and degrees of siltation. For example, "For pipe diameters between 300mm and 600mm, with 1 / 4 of the pipe silted up, the comprehensive unit price is 65.73 yuan / meter." The price of the dredging project can be calculated by automatically selecting and substituting the corresponding pricing formula according to the actual situation of the inspected pipe section.
[0073] Preferably, based on the auxiliary pricing method provided by the present invention, a pipeline dredging auxiliary pricing system can be constructed by connecting the cost calculation management front-end and management terminal using a wireless network, realizing visualization of the calculation process, enabling engineers to check and verify the calculation results, and further, by optimizing and updating the algorithm, the system functions can be continuously improved and expanded.
[0074] It should be noted that the above are merely specific embodiments of the present invention, and their purpose is to enable other people skilled in the art to understand the specific content and implementation method of the present invention. The embodiments described are only the preferred embodiments of the present invention and are not equal to all embodiments. Any equivalent changes or modifications made to the embodiments within the scope of the claims of the present invention should fall within the protection scope of the present invention.
Claims
1. A pipeline dredging auxiliary pricing method based on image recognition technology, characterized in that, The implementation steps include the following: Step S1: Use CCTV inspection equipment to capture pipeline video, extract pipeline cross-section images containing silt from the video, select pipeline cross-section images with suitable angle and clarity through image screening and sampling, preprocess the images to improve image quality by removing noise, enhancing contrast and sharpening edges, and use annotation tools to annotate to obtain a dataset for training the target detection network. Step S2: Based on the target detection network, construct a target detection model for pipe cross-section images, train the model using the dataset obtained in step S1, adjust the model parameters to optimize the model recognition effect, and improve the model accuracy until it meets the target detection requirements. Step S3: Use the target detection model trained in step S2 to identify the pipe cross-section image, identify the silt and mark its target detection box, and use the circular detection algorithm of Hough gradient transform to obtain the circular parameter set of the pipe cross-section. Step S4: Based on the target detection box information of the silt obtained in step S3, determine the Hough circle where the endpoint of the detection box is located, and use the above information to calculate the height ratio of the silt cross section to the pipe cross section to determine the degree of siltation in the pipe. Step S5: Substitute the obtained data into the pipeline dredging pricing formula for different pipe diameters and different degrees of siltation to further obtain the pipeline dredging cost; The target detection network is based on the EfficientDet deep learning algorithm and includes a weighted bidirectional feature pyramid network module for feature fusion. This module provides rich contextual information for multi-scale and high-level semantic information processing through top-down and bottom-up paths. For network optimization, a composite scaling strategy is adopted to achieve optimal resource utilization efficiency by simultaneously adjusting the network depth, width and input resolution. It includes the Soft-NMS algorithm for reducing overlapping detection boxes; and a decoder module for classification and regression of the output results, which transforms the extracted and fused feature maps into accurate target boxes and class probabilities to complete the target detection task.
2. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 1, characterized in that: The CCTV inspection equipment mentioned in step S1 is a common model, and the collected pipe cross-section images are taken from the vertical pipe cross-section.
3. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 1, characterized in that: IoU and GIOU are used as quality assessment criteria to more accurately measure the degree of overlap between the detected bounding box and the ground truth bounding box.
4. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 3, characterized in that: The method for identifying pipe sediment in step S3 is as follows: call the target detection model to identify the image, and draw the corresponding detection box at the same time as identifying the sediment to intuitively display its area in the image; the method for detecting Hough circles in the pipe image is as follows: call the circle detection algorithm of Hough gradient transform to process the image and identify the circular structure in the pipe cross section.
5. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 4, characterized in that: The method for determining the Hough circle containing the two ends of the bottom of the pipeline sediment detection frame in step S4 is as follows: For each detected Hough circle, calculate the distance from the center of each circle to the two ends of the detection frame, and determine whether it is equal to the radius of the Hough circle. If the distance from the center of the circle to the two ends of the sediment detection frame is equal to its radius, then the Hough circle is the Hough circle containing the sediment.
6. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 5, characterized in that: The method for determining the degree of siltation in the pipeline in step S4 is as follows: using the coordinate information of the center of the Hough circle and the coordinate information of the two ends of the bottom of the siltation detection frame, the average distance of the line connecting the two ends relative to the center of the circle is calculated, and the ratio of the average distance to the diameter is obtained to further evaluate the degree of siltation in the pipeline.
7. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 6, characterized in that: The method for calculating the cost in step S5 is as follows: There are specific pricing formulas for different pipe diameters and different degrees of siltation. The obtained percentage data, pipe diameter data and pipe section length are substituted into the corresponding pricing formulas to finally calculate the cost of pipeline dredging.
8. The pipeline dredging auxiliary pricing method based on image recognition technology according to claim 1, characterized in that: The pipeline dredging cost calculation model is connected to the front end of the cost calculation system via a network. It can display the acquired image data, cost calculation process, and calculation results on the front-end interface. This not only assists financial management personnel in quickly calculating the pipeline dredging cost, but also facilitates the management personnel in verifying the calculation process and results, providing double assurance for the accuracy of cost calculation.