Image recognition and preprocessing method for tunnel drainage pipeline deposits based on artificial intelligence

Abstract

The invention provides an artificial intelligence-based image identification and preprocessing method for tunnel drainage pipeline deposits. According to the artificial intelligence-based image recognition and preprocessing method for the tunnel drainage pipeline deposits, a high-definition camera or an unmanned aerial vehicle is adopted to collect image data, and preprocessing and denoising are carried out, so that the quality of input data is ensured. And then, feature extraction is performed through a convolutional neural network, and a deposit region and other pipeline regions are accurately distinguished in combination with a U-Net semantic segmentation technology, so that the precision and robustness of image recognition are improved. The artificial intelligence-based image recognition and preprocessing method for the deposits in the tunnel drainage pipeline comprises the following steps of: a, acquiring image data of the tunnel drainage pipeline, and acquiring the image data through a high-definition camera or an unmanned aerial vehicle; and b, preprocessing the image data, removing noise and enhancing the image definition.

Description

Technical Field

[0001] The present invention relates to the technical field of image recognition and preprocessing of sediment in drainage pipelines, and specifically to an image recognition and preprocessing method for sediment in tunnel drainage pipelines based on artificial intelligence. Background Technique

[0002] The sediment treatment methods for tunnel drainage pipelines generally include two major categories of technologies: physical cleaning methods and chemical cleaning methods. Physical cleaning methods mainly use means such as water flow scouring and mechanical cleaning to remove sediment. Common physical cleaning methods include flushing the inside of the pipeline with a high-pressure water gun, mechanical scraper cleaning, and using a sewage suction truck to suck up sediment. The key principle of these methods is to decompose or remove the sediment physically to restore the pipeline to be unobstructed. Chemical cleaning methods, on the other hand, are to dissolve or decompose the solid substances in the pipeline by using chemical agents, such as using acidic or alkaline cleaning agents to remove the deposited minerals or sediments.

[0003] Although the application of modern technologies in the treatment of sediment in tunnel drainage pipelines has brought many advantages, there are still certain defects and challenges. The adaptability of existing technologies in extremely complex environments is still limited, which may lead to a decrease in the accuracy of data recognition or processing. The image recognition system requires a large amount of data for training and has high requirements for the real-time processing ability of data, which may be restricted by equipment aging and insufficient data processing ability in some old tunnel facilities. Summary of the Invention

[0004] Aiming at the deficiencies of the existing technology, the present invention provides an image recognition and preprocessing method for sediment in tunnel drainage pipelines based on artificial intelligence, which solves the problem that the adaptability in extremely complex environments is still limited, resulting in a decrease in the accuracy of data recognition or processing. The image recognition system requires a large amount of data for training and has high requirements for the real-time processing ability of data, and will be restricted by equipment aging and insufficient data processing ability.

[0005] To achieve the above objectives, the present invention is realized through the following technical solutions: An image recognition and preprocessing method for sediment in tunnel drainage pipelines based on artificial intelligence, including:

[0006] a. Obtain the image data of the tunnel drainage pipeline, and the image data is collected by a high-definition camera or a drone;

[0007] b. Preprocess the image data to remove noise and enhance the image clarity;

[0008] c. Use a convolutional neural network to extract features from the preprocessed image, obtain the important features in the pipeline image, use semantic segmentation technology to classify different regions in the image, and separate the sediment region from other regions;

[0009] d. Through model training, establish a neural network model for identifying sediment, and use big data training methods to improve the recognition accuracy;

[0010] e. Mark the sediment area in the image and output the position and size information of the sediment. The position marking can be achieved through a regression model:

[0011]

[0012] where p i is the true position, is the predicted position;

[0013] f. Repair or further process the identified sediment area to provide data support for subsequent pipeline maintenance;

[0014] g. Use an optimization algorithm to dynamically adjust the image recognition results to meet the recognition requirements in different complex environments;

[0015] h. Process the collected image data in real time to ensure that the system can still operate efficiently under high load, and use streaming data processing technology to ensure real-time performance;

[0016] i. Keep the long-term effectiveness and efficiency of the image recognition system by regularly updating the dataset and model;

[0017] j. Use data fusion technology to comprehensively process the data collected by different sensors to improve the accuracy and robustness of recognition;

[0018] k. In the case of equipment aging or insufficient data processing capacity, adopt an adaptive learning strategy to optimize the algorithm model and improve the recognition ability of the system.

[0019] Preferably, the mathematical model of the convolutional neural network is:

[0020] Y = f(W * X + b)

[0021] where Y is the output feature, W is the convolutional kernel weight, X is the input image, b is the bias term, f is the activation function, and * represents the convolution operation.

[0022] Preferably, the semantic segmentation technology uses the U-Net model for segmentation, and the model output is:

[0023]

[0024] where is the segmented image, and X is the input image.

[0025] Preferably, the loss function of the neural network model uses the cross-entropy loss function:

[0026]

[0027] Among them, y i is the true label, is the predicted value, and n is the number of samples.

[0028] Preferably, the optimization algorithm uses the Adam optimizer, and the update formula is:

[0029]

[0030] Among them, θ t is the current parameter, η is the learning rate, m t and v t are the first-order moment estimate and the second-order moment estimate respectively, and ∈ is the smoothing term.

[0031] Preferably, the data fusion technology is carried out by the weighted average method:

[0032]

[0033] Among them, Z is the fused result, w i is the weight, and X i is the sensor output.

[0034] Preferably, the semantic segmentation technology adopts a semantic segmentation network to effectively distinguish the silt area in the image from other areas.

[0035] Preferably, the neural network model enhances the generalization ability of the model through data augmentation technology and improves its adaptability in complex environments.

[0036] [[ID=4-6]]The present invention provides an image recognition and preprocessing method for silt in tunnel drainage pipes based on artificial intelligence. It has the following beneficial effects:

[0037] The image recognition and preprocessing method for silt in tunnel drainage pipes based on artificial intelligence uses a high-definition camera or a drone to collect image data, and performs preprocessing and denoising to ensure the quality of the input data. Then, feature extraction is carried out through a convolutional neural network, and combined with the U-Net semantic segmentation technology, the silt area is accurately distinguished from other pipe areas, improving the accuracy and robustness of image recognition. In addition, this method continuously trains and optimizes the neural network model through deep learning algorithms, can adapt to various complex environmental changes, and effectively solves the problem that the accuracy of the existing technology decreases in the case of equipment aging and insufficient data processing capacity.

[0038] This technical solution utilizes an optimization algorithm and an adaptive learning strategy to dynamically adjust the image recognition results, and ensures efficient operation even under high load conditions through real-time data processing technology. The adoption of data fusion technology further improves the accuracy and stability of the system, enabling the system to comprehensively judge and accurately identify the silt area under different sensor data inputs, providing accurate and timely decision-making support for pipeline maintenance. By regularly updating the dataset and the model, the present invention ensures the long-term effectiveness and efficiency of the system, thus realizing the full-life cycle optimization of the method for identifying and preprocessing the silt in the tunnel drainage pipeline. Brief Description of the Drawings

[0039] Figure 1 It is a schematic flowchart of the present invention. Detailed Embodiment

[0040] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative efforts shall fall within the protection scope of the present invention.

[0041] As Figure 1 shown, the embodiment of the present invention provides an image recognition and preprocessing method for silt in a tunnel drainage pipeline based on artificial intelligence, including: a. Obtaining image data of the tunnel drainage pipeline, and the image data is collected by a high-definition camera or a drone.

[0042] b. Preprocessing the image data to remove noise and enhance the image clarity.

[0043] c. Using a convolutional neural network to extract features from the preprocessed image, obtaining important features in the pipeline image, using semantic segmentation technology to classify different regions in the image, separating the silt area from other areas, and the semantic segmentation technology uses the U-Net model for segmentation, and the model output is:

[0044]

[0045] Wherein, is the segmented image, X is the input image, and the semantic segmentation technology uses a semantic segmentation network to effectively distinguish the silt area in the image from other areas.

[0046] d. Through model training, a neural network model for identifying silt is established, and the recognition accuracy is improved by using a big data training method. The mathematical model of the convolutional neural network is:

[0047] Y = f(W * X + b)

[0048] Among them, Y is the output feature, W is the convolutional kernel weight, X is the input image, b is the bias term, f is the activation function, * represents the convolution operation, and the loss function of the neural network model uses the cross-entropy loss function:

[0049]

[0050] Among them, y i is the true label, is the predicted value, n is the number of samples, and the neural network model enhances the generalization ability of the model through data augmentation technology to improve its adaptability in complex environments.

[0051] e. Mark the silt area in the image and output the position and size information of the silt. The position marking can be achieved through a regression model:

[0052]

[0053] Among them, p i is the true position, is the predicted position.

[0054] f. Repair or further process the identified silt area to provide data support for subsequent pipeline maintenance.

[0055] g. Use an optimization algorithm to dynamically adjust the image recognition results to meet the recognition requirements in different complex environments. The optimization algorithm uses the Adam optimizer, and the update formula is:

[0056]

[0057] Among them, θ t is the current parameter, η is the learning rate, m t and v t are the first-order moment estimate and the second-order moment estimate respectively, and ∈ is the smoothing term.

[0058] h. Process the collected image data in real time to ensure that the system can still operate efficiently under high load, and use streaming data processing technology to ensure real-time performance.

[0059] i. Keep the long-term effectiveness and efficiency of the image recognition system by regularly updating the dataset and the model.

[0060] j. Use data fusion technology to comprehensively process the data collected by different sensors to improve the accuracy and robustness of recognition. The data fusion technology is carried out by the weighted average method:

[0061]

[0062] Among them, Z is the fused result, wi is the weight, and X i is the sensor output.

[0063] k. In the case of equipment aging or insufficient data processing capacity, adopt an adaptive learning strategy to optimize the algorithm model and improve the recognition ability of the system.

[0064] Example 2:

[0065] Suppose that during a routine inspection of the tunnel drainage pipes in a certain city, problems with sediment deposits are found, and an artificial intelligence-based image recognition and preprocessing method is required for detection and repair. The following are the specific implementation steps, including the use and processing of data:

[0066] 1. Image acquisition

[0067] Equipment: Use a drone equipped with a high-definition camera to collect image data of the tunnel drainage pipes. Set the flight altitude and speed of the drone to ensure that every section of the entire pipe is covered.

[0068] Data volume: During this inspection, a total of 1000 images inside the tunnel were collected, with a resolution of 1920x1080 pixels.

[0069] 2. Image preprocessing

[0070] Noise removal: Use the Gaussian filtering algorithm to denoise the collected original images to eliminate the noise generated during image acquisition.

[0071] Image enhancement: Use histogram equalization technology to enhance the contrast of the images and improve the visibility of the pipe details.

[0072] Preprocessed data: After preprocessing, the noise in all 1000 images was significantly reduced, enhancing the detailed information of the images.

[0073] 3. Feature extraction and classification

[0074] Use a convolutional neural network: Use a pre-trained convolutional neural network to extract features from each preprocessed image. The CNN extracts texture, shape, and edge information in the images and generates a feature vector for each image.

[0075] Semantic segmentation: Use the U-Net model for semantic segmentation of the images. The model divides the images into sediment deposit areas and normal areas. Through training, the model can already accurately identify the sediment deposits in the images and label them as different categories.

[0076] Data volume and results: After processing, the sediment deposit areas in 1000 images were successfully segmented and labeled. Among them, approximately 200 images contained obvious sediment deposits, and the others were normal areas.

[0077] 4. Model Training and Accuracy Optimization

[0078] Training Data: By using 1000 images containing labeled sediment, the model conducted multiple rounds of training on the convolutional layer and pooling layer to improve the recognition accuracy.

[0079] Training Results: After training, the accuracy of the model on the validation set reached 95%, and the precision and recall were 92% and 89% respectively.

[0080] 5. Sediment Marking and Location

[0081] Regression Model: The regression model accurately calibrated the sediment position in each image and output the position coordinates and size information. For example, in an image containing sediment, the rectangular box of the sediment was between (50, 30) and (150, 80).

[0082] Output Data: The calibrated data included the specific position, area, etc. of the sediment. For example, the position of the sediment in Image 1 was (50, 30), and the area was 200 square pixels.

[0083] 6. Area Repair and Processing

[0084] Repair Process: After identifying the sediment area, the virtual repair technology was used to simulate the pipeline cleaning process, generating the repaired image and the prediction of the pipeline recovery status.

[0085] Data Example: The repaired image showed that the area originally blocked by sediment was clearly presented, and the pipeline returned to the normal drainage state.

[0086] 7. Optimization and Adaptation

[0087] Optimization Algorithm: The Adam optimizer was used to dynamically adjust the model and optimize the parameters in the image recognition process.

[0088] System Adaptability: The system adjusted the model parameters in real time through the feedback mechanism to adapt to the changes in different tunnel environments. For example, factors such as humidity and temperature in the tunnel may affect the image quality, and the system can automatically adjust to maintain high recognition accuracy.

[0089] 8. Real-time Data Processing

[0090] Streaming Data Processing: During the pipeline cleaning process, the collected image data was processed in real time. After each image processing, the system immediately output the sediment position and repair suggestions to guide the maintenance team to perform quick operations.

[0091] Data Volume: The system processes about 2 images per second to ensure real-time feedback to the pipeline maintenance team.

[0092] 9. Data Update and Long-Term Maintenance

[0093] Regularly update the dataset: After each maintenance, new image data is added to the dataset, and the neural network is retrained to improve the long-term stability of the model.

[0094] Training results: After each update, the accuracy of the recognition system increases by approximately 2-3%.

[0095] 10. Data Fusion

[0096] Fuse data from other sensors: In addition to image data, the system also fuses data from temperature, humidity, and pressure sensors, and performs data fusion through the weighted average method.

[0097] Improve accuracy: Through data fusion, the system can improve the recognition accuracy under different environmental conditions.

[0098] 11. Adaptive Learning

[0099] Adaptive optimization: When the device ages or the data processing capacity is insufficient, the system automatically adopts an adaptive learning strategy for optimization to ensure high recognition accuracy even under low computing resource conditions.

[0100] Implementation effects:

[0101] Recognition accuracy: Through the above steps, the system can successfully identify 200 images containing sediment out of 1000 images, and the recognition accuracy reaches 95%.

[0102] Pipeline maintenance: Through real-time image processing and marking, the maintenance team can quickly locate the position of the sediment and clean it in time, effectively preventing drainage system failures.

[0103] Although the embodiments of the present invention have been shown and described, it will be understood by those of ordinary skill in the art that various changes, modifications, substitutions, and variations can be made to these embodiments without departing from the principles and spirit of the present invention, and the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An image recognition and preprocessing method for sediment in tunnel drainage pipelines based on artificial intelligence, characterized in that Including: a. Obtain the image data of the tunnel drainage pipeline, and the image data is collected by a high-definition camera or a drone; b. Preprocess the image data to remove noise and enhance image clarity; c. Use a convolutional neural network to extract features from the preprocessed image, obtain important features in the pipeline image, use semantic segmentation technology to classify different regions in the image, and separate the silt region from other regions; d. Through model training, establish a neural network model for identifying silt, and use big data training methods to improve the recognition accuracy; e. Mark the silt region in the image and output the position and size information of the silt. The position marking can be achieved through a regression model: where p i is the true position, and is the predicted position; f. Repair or further process the identified silt region to provide data support for subsequent pipeline maintenance; g. Use an optimization algorithm to dynamically adjust the image recognition results to meet the recognition requirements in different complex environments; h. Process the collected image data in real time to ensure that the system can still operate efficiently under high load, and use streaming data processing technology to ensure real-time performance; i. Keep the long-term effectiveness and efficiency of the image recognition system by regularly updating the dataset and model; j. Use data fusion technology to comprehensively process the data collected by different sensors to improve the accuracy and robustness of recognition; k. In the case of equipment aging or insufficient data processing capacity, adopt an adaptive learning strategy to optimize the algorithm model and improve the recognition ability of the system.

2. The image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The mathematical model of the convolutional neural network is: Y = f(W * X + b) where Y is the output feature, W is the convolutional kernel weight, X is the input image, b is the bias term, f is the activation function, and * represents the convolution operation.

3. An image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The semantic segmentation technology uses the U-Net model for segmentation, and the model output is: Among them, is the segmented image, and X is the input image.

4. An image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The loss function of the neural network model adopts the cross-entropy loss function: Among them, y i is the true label, is the predicted value, and n is the number of samples.

5. The image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The optimization algorithm uses the Adam optimizer, and the update formula is: where, θ t is the current parameter, η is the learning rate, m t and v t are the first moment estimate and the second moment estimate respectively, and ∈ is the smoothing term.

6. The image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The data fusion technology is carried out by the weighted average method: Among them, Z is the result after fusion, and w i is the weight, and X i is the sensor output.

7. An image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The semantic segmentation technology adopts a semantic segmentation network to effectively distinguish the silt region in the image from other regions.

8. An image recognition and preprocessing method for sediment in a tunnel drainage pipeline based on artificial intelligence according to claim 1, characterized in that: The neural network model enhances the generalization ability of the model through data augmentation technology and improves the adaptability in complex environments.

Images