Building construction pollution degree detection method based on image recognition

Through the image recognition-based pollution detection method of building construction construction, multi-spectral cameras and deep learning models, combined with distributed camera networks and drones or ground robots, the real-time and comprehensive problems of environmental pollution monitoring on construction sites are solved, and the realization of the Accurate identification and classification of pollution sources improves the flexibility and response speed of environmental monitoring.

CN119064368APending Publication Date: 2024-12-03CHINA MCC17 GRP CO LTD
View PDF 0 Cites 2 Cited by

Patent Information

Application Number
CN202411241637.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2024-09-04
Publication Date
2024-12-03

AI Technical Summary

Technical Problem

The existing technology lacks real-time and comprehensive environmental pollution monitoring in construction projects. Traditional methods are difficult to cope with rapid changes in the construction site environment, and lack flexibility and adaptability, and cannot respond to potential pollution events in a timely or effective manner.

Method used

Using image recognition-based detection methods, the comprehensive light information of the construction site is obtained through multi-spectral cameras and high dynamic range imaging technology, and a distributed camera network is deployed and a drone or ground robot is equipped with cameras, combining deep learning models and image segmentation technology. Real-time analysis and identification of pollution sources and data processing and management through edge computing and distributed storage systems.

Benefits of technology

It realizes accurate identification and classification of pollution sources under different ambient light conditions, enhances the spatial coverage and flexibility of the monitoring system, can promptly capture dynamic changes and emergencies at the construction site, and provides more detailed pollution management support.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN119064368A_ABST
    Figure CN119064368A_ABST
Patent Text Reader

Abstract

The invention relates to the technical field of environment monitoring and building construction management, and discloses a building construction pollution degree detection method based on image recognition, and the method comprises the following steps: S1, installing a multispectral camera, and collecting the image data of a plurality of spectral segments, so as to obtain the comprehensive light information of a construction site; s2, synthesizing and processing the multispectral image by using a high dynamic range imaging computational photography technology to enhance details and quality of the image; s3, deploying a plurality of fixed cameras around the construction site to form a distributed network, and capturing image data at different angles to cover the whole construction area; and S4, deploying an unmanned aerial vehicle or a ground robot to carry a camera. According to the method, the quality and the detailed degree of an image are improved through the combined action of the use of a multispectral camera and high-dynamic-range imaging, so that the recognition of the pollution source is not limited to a certain spectral range, and the pollution source can be more accurately recognized and segmented from a complex construction scene through the training and application of an advanced model such as U-Net.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The invention relates to the technical field of environmental monitoring and building construction management, and in particular to a building construction pollution degree detection method based on image recognition. Background Art

[0002] Environmental monitoring is a vital part of construction projects, especially in cities and near densely populated areas. In existing technologies, environmental pollution monitoring mainly relies on data collection from fixed monitoring stations or regular manual inspections, which often fail to capture the pollution situation at the construction site in real time or comprehensively. For example, fixed monitoring stations may not cover all key areas of the construction site, while manual inspections are time-consuming and inefficient.

[0003] In addition, traditional monitoring methods usually only focus on specific types of pollution, such as dust, noise, etc., while ignoring the environmental impacts that may be caused by multiple pollution sources. These methods lack flexibility and adaptability, and are difficult to cope with the rapid changes in the construction site environment. Moreover, traditional monitoring technologies usually do not provide sufficient data analysis tools to analyze and respond to potential pollution incidents in real time, resulting in the implementation of pollution management measures not being timely or effective. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides a method for detecting construction pollution based on image recognition, which solves the problem of insufficient real-time and comprehensive monitoring of construction site environmental pollution in the prior art.

[0005] To achieve the above objectives, the present invention is implemented through the following technical solutions: A construction pollution detection method based on image recognition, comprising the following steps:

[0006] S1. Install a multispectral camera to collect image data in multiple spectral bands to obtain comprehensive light information at the construction site;

[0007] S2. Use high dynamic range imaging computational photography technology to synthesize and process multispectral images to enhance image details and quality;

[0008] S3. Deploy multiple fixed cameras around the construction site to form a distributed network to capture image data from different angles to cover the entire construction area;

[0009] S4. Deploy drones or ground robots equipped with cameras, adjust the shooting angle and position regularly or as needed, and obtain image data with more dynamic perspectives;

[0010] S5, using background subtraction algorithm to learn and remove the unchanged background information according to the dynamic changes of the construction site;

[0011] S6. Train and apply deep learning adaptive models to analyze collected image data to identify and extract features of potential pollution sources;

[0012] S7, combining multimodal data to identify pollution sources;

[0013] S8. Use image segmentation technology to classify and segment pollution sources and identify different types of pollutants;

[0014] S9. Deploy edge computing devices on site to process image and sensor data in real time and reduce data transmission latency.

[0015] S10. Use Hadoop and Ceph distributed storage systems to securely store and manage large amounts of image and sensor data.

[0016] Preferably, the step S2 specifically includes the following steps:

[0017] S2.1, Image capture: Set the multispectral camera to continuously capture the same scene at different exposure values;

[0018] S2.2, Image synthesis: using software to synthesize multiple exposure images into a single high dynamic range image;

[0019] S2.3, Post-processing: Adjust the color balance, contrast and clarity of the composite image.

[0020] Preferably, the step S4 specifically includes the following steps:

[0021] S4.1. Select the appropriate drone / robot: Select the appropriate drone or ground robot according to the operation requirements and environment;

[0022] S4.2. Install the camera: Install and fix the camera on the drone or robot;

[0023] S4.3, Programming and Debugging: Program drones or robots to perform scheduled or on-demand filming missions;

[0024] S4.4. Real-time monitoring and control: Set up a control center to receive and process image data sent back by drones or robots in real time.

[0025] Preferably, the step S5 specifically includes the following steps:

[0026] S5.1. Establish background model: Capture images of the construction site at different time periods to establish a background model;

[0027] S5.2, Algorithm deployment: Deploy the background subtraction algorithm in the processing system and apply it to the incoming image stream in real time;

[0028] S5.3. Continuous learning and updating: As construction progresses, the background model is continuously updated to adapt to environmental changes.

[0029] Preferably, the step S6 specifically includes the following steps:

[0030] S6.1. Collect and annotate training data: collect images from installed cameras and manually annotate them to train the model;

[0031] S6.2. Model training: Use labeled datasets to train deep learning models, such as convolutional neural networks;

[0032] S6.3. Model optimization and validation: Optimize and validate model performance using test data sets;

[0033] S6.4, Model deployment: Deploy the trained model on edge computing devices for real-time image analysis.

[0034] Preferably, the step S7 specifically includes the following steps:

[0035] S7.1. Integrate sensor data: Integrate data from different sensors;

[0036] S7.2. Data fusion and analysis: Use data fusion techniques to analyze image and sensor data to identify pollution patterns;

[0037] S7.3. Feature extraction and pollution identification: Extract key features from the fused data for identifying and classifying pollution sources.

[0038] Preferably, the step S8 specifically includes the following steps:

[0039] S8.1. Select and train a segmentation model: Select a suitable image segmentation model, such as U-Net, and train it with actual data;

[0040] S8.2. Perform image segmentation: Apply the trained segmentation model on real-time data to accurately segment the pollution sources in the image;

[0041] S8.3. Classification and labeling: Classify and label the segmented image areas according to the pollution type.

[0042] Preferably, the sensors in step S7.1 include a gas sensor, a temperature sensor and a humidity sensor.

[0043] The invention provides a method for detecting pollution level in building construction based on image recognition. It has the following beneficial effects:

[0044] 1. The present invention uses a multi-spectral camera and high dynamic range imaging to improve the quality and detail of the image, so that the identification of pollution sources is not limited to a certain spectral range, but can be carried out more accurately under different ambient light conditions.

[0045] 2. Through the training and application of advanced models such as U-Net, the present invention can more accurately identify and segment pollution sources from complex construction scenes, thereby providing more detailed classification and identification than traditional methods.

[0046] 3. The present invention provides the ability to dynamically adjust the shooting angle and position by deploying drones or ground robots equipped with cameras, greatly enhancing the spatial coverage and flexibility of the monitoring system. The dynamic adjustment can cope with the rapidly changing situation at the construction site and timely capture new pollution sources caused by changes in natural light, construction activities or emergencies. BRIEF DESCRIPTION OF THE DRAWINGS

[0047] Figure 1 It is a stereogram of the present invention. DETAILED DESCRIPTION

[0048] The technical solution of the present invention will be described clearly and completely below in conjunction with the accompanying drawings of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0049] Example:

[0050] Please refer to the attached Figure 1 The embodiment of the present invention provides a method for detecting pollution level of construction based on image recognition, comprising the following steps:

[0051] S1. Install a multispectral camera to collect image data in multiple spectral bands to obtain comprehensive light information at the construction site;

[0052] S2. Use high dynamic range imaging computational photography technology to synthesize and process multispectral images to enhance image details and quality;

[0053] S3. Deploy multiple fixed cameras around the construction site to form a distributed network to capture image data from different angles to cover the entire construction area;

[0054] S4. Deploy drones or ground robots equipped with cameras, adjust the shooting angle and position regularly or as needed, and obtain image data with more dynamic perspectives;

[0055] S5, using background subtraction algorithm to learn and remove the unchanged background information according to the dynamic changes of the construction site;

[0056] S6. Train and apply deep learning adaptive models to analyze collected image data to identify and extract features of potential pollution sources;

[0057] S7, combining multimodal data to identify pollution sources;

[0058] S8. Use image segmentation technology to classify and segment pollution sources and identify different types of pollutants;

[0059] S9. Deploy edge computing devices on site to process image and sensor data in real time and reduce data transmission latency.

[0060] S10. Use Hadoop and Ceph distributed storage systems to securely store and manage large amounts of image and sensor data.

[0061] Specifically, the multispectral camera in step S1 can capture spectra beyond the visible range of the human eye, such as infrared and ultraviolet light, which makes it possible to detect specific absorption and reflection characteristics of certain chemicals. For example, certain pollutants may show specific spectral characteristics under ultraviolet light. This step makes the monitoring system not limited to visible light, increasing the ability to identify and analyze environmental pollution, especially in low light or night conditions.

[0062] In step S2, HDR technology is used to increase the dynamic range of the image by synthesizing multiple images with different exposure levels, so that details of very bright and very dark areas can be displayed simultaneously in the same image.

[0063] This step improves image quality and makes details clearer, especially in high-contrast scenes, helping to more accurately identify and analyze pollution sources.

[0064] In step S3, by deploying cameras at different locations, a monitoring network covering the entire construction area is formed, which can obtain data from multiple angles and locations and reduce monitoring blind spots.

[0065] It ensures coverage without blind spots, improving the comprehensiveness of site monitoring and the integrity of data collection.

[0066] In step S4, drones or ground robots can move to locations that are difficult to reach manually or high-risk areas for flexible filming.

[0067] This step provides a dynamic and customized perspective, especially when it is necessary to focus on specific activities or examine contamination in hard-to-access areas.

[0068] In step S5, the algorithm is used to identify and remove the unchanged background parts in the image, and only the changed areas (such as moving equipment, personnel or smoke, etc.) are retained. This effectively reduces the data processing requirements, focuses on key change areas, and improves processing efficiency and pollution identification speed.

[0069] Step S6, through deep learning models learning from large amounts of data, is able to identify complex patterns and features in images for accurate identification of pollution sources.

[0070] It improves the accuracy and reliability of contamination identification and can automate and reduce the need for manual monitoring.

[0071] In step S7, the image data is combined with other sensor data (such as gas concentration, temperature and humidity, etc.) to provide a more comprehensive environmental analysis using data fusion technology. This enhances the detection of pollution indicators that may be overlooked by a single technology, verifies pollution events through multiple data sources, and improves the accuracy of diagnosis.

[0072] In step S8, advanced image processing techniques, such as deep learning segmentation networks, are used to accurately segment pollution sources from images and classify them, making the identification of different types of pollutants more accurate and systematic, facilitating further analysis and processing.

[0073] In step S9, edge computing devices are used to process data close to the data source, which reduces the need to transmit data to the central server and speeds up data processing. This significantly reduces the response time, enables real-time processing and response to emergency pollution incidents, and improves the response speed of the environmental monitoring system.

[0074] Step S10, by utilizing distributed storage technology, can effectively manage large amounts of data, support high concurrent access, and provide data redundancy and fault recovery functions. It ensures data security and high availability, supports long-term storage and fast access to large-scale data, and provides strong data support for environmental monitoring and subsequent analysis.

[0075] The S2 step specifically includes the following steps:

[0076] S2.1, Image capture: Set the multispectral camera to continuously capture the same scene at different exposure values;

[0077] S2.2, Image synthesis: using software to synthesize multiple exposure images into a single high dynamic range image;

[0078] S2.3, Post-processing: Adjust the color balance, contrast and clarity of the composite image.

[0079] Specifically, step S2.1 uses a multi-spectral camera to continuously shoot the same scene with different exposure settings (usually including normal exposure, overexposure, and underexposure) through multi-exposure image capture to capture scene details that may be lost under a single exposure setting, such as details in dark and bright areas.

[0080] In an overexposed image, details in the highlights will be clearer, while in an underexposed image, details in the shadows will be more apparent. This approach allows for a wider dynamic range, capturing details from deep shadows to highlights.

[0081] Multi-exposure technology enables a more comprehensive understanding of the scene, especially under complex lighting conditions, which enhances the visual information in the image and facilitates subsequent more accurate image analysis and processing.

[0082] The specific steps of this step are:

[0083] Set up a multispectral camera: Adjust the camera settings to continuously capture the same scene at different exposure values. Usually includes minimum exposure (underexposure), medium exposure (normal exposure), and high exposure (overexposure).

[0084] Time synchronization: Ensure that the capture times of the multispectral cameras are synchronized so that the images can be accurately synthesized.

[0085] Calculation formula:

[0086] Exposure formula: E = N 2 / t

[0087] Among them, E is the exposure, N is the aperture value, and t is the exposure time (seconds).

[0088] Use this formula to adjust exposure parameters to ensure quality images captured under different lighting conditions.

[0089] In step S2.2, HDR image synthesis uses an algorithm to synthesize multiple images with different exposures into a high dynamic range image, which involves weighting and synthesizing the pixel values ​​of images with different exposures to form an image with more information between the maximum and minimum brightness.

[0090] Calculation formula: A basic method for synthesizing HDR images is to use a weighted average method, where the value V of each pixel is calculated by the following formula:

[0091]

[0092] Among them, v i is the brightness value of the pixel in the i-th image, w i is the corresponding weight (usually determined by exposure), and n is the number of images.

[0093] The resulting HDR images are visually closer to how the human eye perceives the real world, providing richer color and brightness details, especially in very high-contrast scenes.

[0094] In step S2.3, the purpose of post-processing is to adjust the color balance, contrast and clarity of the synthesized HDR image to suit human eye observation or specific display devices, including color correction, contrast enhancement and sharpening.

[0095] Color balance can be adjusted to make the image color more realistic or in line with a specific visual style by adjusting the color temperature and tint. Contrast can be enhanced by adjusting the grayscale distribution range of the image, making the details from the darkest to the brightest clearer.

[0096] Color Correction Formula:

[0097] C new =C old (1+α)

[0098] Among them, C new is the adjusted color value, C old is the original color value, and α is the correction coefficient.

[0099] Enhanced images are more suitable for further analysis, such as using image recognition algorithms to detect and identify specific environmental pollution indicators. By optimizing image quality, the accuracy and reliability of analysis algorithms can be improved.

[0100] The S4 step specifically includes the following steps:

[0101] S4.1. Select the appropriate drone / robot: Select the appropriate drone or ground robot according to the operation requirements and environment;

[0102] S4.2. Install the camera: Install and fix the camera on the drone or robot;

[0103] S4.3, Programming and Debugging: Program drones or robots to perform scheduled or on-demand filming missions;

[0104] S4.4. Real-time monitoring and control: Set up a control center to receive and process image data sent back by drones or robots in real time.

[0105] Specifically, step S4.1, select the appropriate drone or ground robot based on the specific needs of the construction site (such as the complexity of the terrain, the area to be covered, and the safety of the operation). Drones are suitable for extensive aerial monitoring, especially in areas with complex terrain or difficult to access on foot. Ground robots are suitable for more detailed ground-level observations, especially in environments with narrow spaces or that require continuous monitoring.

[0106] This step ensures that the selected equipment is best suited to perform the required tasks, improving the efficiency and safety of monitoring. Correct equipment selection can maximize the coverage area, ensuring that critical areas are adequately monitored while reducing operational risks.

[0107] In step S4.2, install a dedicated camera on the drone or robot to ensure that the device can shoot stably during movement and obtain high-quality images. Use appropriate fixing devices and anti-vibration systems to reduce the impact of vibration on image quality.

[0108] Ensuring that clear and stable images can be captured even when moving is crucial for subsequent image analysis, especially in dynamic environments where maintaining image quality is key to identifying and analyzing pollution sources.

[0109] Step S4.3 is to program the drone or robot to set its flight route or patrol route, as well as necessary tasks (such as photo taking frequency, stop observation at specific locations, etc.). These programming tasks are usually completed through an advanced programming interface (API) or dedicated software, allowing complex operation sequences to be set.

[0110] Enable drones or robots to automatically perform monitoring tasks, reduce the need for manual operation, and improve the consistency and systematicness of data collection. Automated data capture processes enhance the reliability and efficiency of monitoring.

[0111] Step S4.4, receiving and processing the image data sent back from the drone or robot in real time through the control center, involves the use of wireless communication technology. The control center can monitor the equipment status and collected data in real time, and make remote adjustments or emergency responses when necessary.

[0112] Real-time data transmission and monitoring allows the operation team to quickly respond to changes on site, adjust monitoring strategies or respond to potential problems in a timely manner. Real-time monitoring also increases the transparency and real-time nature of data collection, making decision-making more precise.

[0113] Step S5 specifically includes the following steps:

[0114] S5.1. Establish background model: Capture images of the construction site at different time periods to establish a background model;

[0115] S5.2, Algorithm deployment: Deploy the background subtraction algorithm in the processing system and apply it to the incoming image stream in real time;

[0116] S5.3. Continuous learning and updating: As construction progresses, the background model is continuously updated to adapt to environmental changes.

[0117] Specifically, in step S5.1, the background model is established by capturing images of the construction site at different time periods. A statistical method is used to determine the typical value of each pixel over a period of time. A Gaussian mixture model is used, which is able to handle changes in background values ​​due to factors such as lighting changes and weather changes.

[0118] The value distribution of each pixel is modeled. It is usually assumed that the change of pixels in the time series conforms to the Gaussian distribution, and the mixture of multiple Gaussian distributions represents a variety of background states.

[0119] Building an accurate background model can effectively help the system distinguish between background and foreground (dynamically changing parts such as moving equipment and workers).

[0120] In step S5.2, algorithm deployment refers to the actual application of the background subtraction algorithm to the real-time incoming image stream. This step is the key to automated processing, ensuring that all captured images are processed instantly and only the parts containing important information are retained through background subtraction.

[0121] Background subtraction algorithms identify background pixels and changed pixels (i.e. foreground) by comparing the live image with a background model. They involve direct comparison of pixel values, or more complex statistical methods to determine the deviation of pixels from the background state.

[0122] Performing background subtraction on images in real time can significantly reduce the amount of data and focus only on areas containing potentially important information, such as moving objects or changing scenes, thereby improving the processing efficiency and response speed of the system.

[0123] Gaussian mixture model update formula:

[0124] Where P(x) is the probability of pixel x, ω k is the weight of the kth Gaussian component, μ k and are the mean and variance respectively, and η is the Gaussian function.

[0125] Step S5.3, continuous learning and updating is to cope with the constant changes in the construction site environment, such as the emergence of new structures or the demolition of old structures. The background model needs to be updated regularly to reflect these changes.

[0126] Using online learning techniques, the background model continuously learns and adapts to changes in the environment from the latest captured images. This may involve incremental learning methods, where new data is gradually incorporated into the model and old data that no longer represents the current scene is discarded.

[0127] By continuously adapting the background model to the latest status of the construction site, the system's high sensitivity and correct response to new changes are ensured, avoiding false detections due to environmental changes.

[0128] Step S6 specifically includes the following steps:

[0129] S6.1. Collect and annotate training data: collect images from installed cameras and manually annotate them to train the model;

[0130] S6.2. Model training: Use labeled datasets to train deep learning models, such as convolutional neural networks;

[0131] S6.3. Model optimization and validation: Optimize and validate model performance using test data sets;

[0132] S6.4, Model deployment: Deploy the trained model on edge computing devices for real-time image analysis.

[0133] Specifically, step S6.1, collecting data refers to collecting images from various camera systems at the construction site, which capture various construction activities and possible pollution events, such as dust and smoke emissions.

[0134] Labeling data is the manual classification and labeling of these images, such as identifying pollution sources, construction equipment, and workers in the images. This step provides the model with the "answer" or "truth" for learning.

[0135] Accurate annotation directly affects the quality of model training. By providing high-quality annotations, the model can more accurately learn the specific characteristics and pollution patterns of the construction site.

[0136] In step S6.2, use the labeled dataset to train a deep learning model, such as a convolutional neural network.

[0137] The training process consists of forward propagation, calculation of the loss function, and backpropagation to update the model weights to minimize the prediction error.

[0138] The trained model can automatically identify similar patterns and features in new images, greatly enhancing the intelligence level of the construction monitoring system, enabling it to identify and respond to potential environmental pollution incidents in real time.

[0139] In step S6.3, use a separate test dataset in addition to the training set to verify the performance of the model. This can help detect whether the model is overfitting (i.e., it only performs well on the training data and has poor generalization ability).

[0140] Optimization involves adjusting the model structure (such as the number of layers, nodes), learning rate, or other hyperparameters to find the best model configuration.

[0141] Through testing and optimization, we ensure that the model has good and stable performance in practical applications, improve the model's prediction accuracy for unknown data, and ensure the reliability and effectiveness of the system.

[0142] In step S6.4, the trained and optimized model is deployed on the edge computing device. The edge computing device is usually located at or near the data source (camera), and can process the data when it is generated, reducing transmission delays.

[0143] Deployment includes model loading, performance debugging, and integration with field monitoring systems.

[0144] Running the model on edge devices can achieve near real-time image analysis and quickly respond to environmental changes or pollution events, improving processing speed and timeliness and reducing the computing burden on central servers.

[0145] Step S7 specifically includes the following steps:

[0146] S7.1. Integrate sensor data: Integrate data from different sensors;

[0147] S7.2. Data fusion and analysis: Use data fusion techniques to analyze image and sensor data to identify pollution patterns;

[0148] S7.3. Feature extraction and pollution identification: Extract key features from the fused data for identifying and classifying pollution sources.

[0149] The sensors in step S7.1 include a gas sensor, a temperature sensor and a humidity sensor.

[0150] Specifically, step S7.1, integrating sensor data is the process of integrating data from multiple sensors (gas sensors, temperature sensors, and humidity sensors) from the construction site. Each sensor provides different types of environmental information. For example, gas sensors can detect the concentration of specific harmful gases (such as CO2, NOx, etc.), and temperature and humidity sensors provide environmental condition data.

[0151] Collect and synchronize this data through a unified data platform or system so that its timestamps and geographic information are aligned to facilitate further comprehensive analysis.

[0152] Data integration allows for analysis across sensor and image data sources, providing a more comprehensive view of the monitored environment. This approach can improve the accuracy of monitoring systems because multiple data sources provide the opportunity to verify each other, reducing the impact of single sensor errors.

[0153] In step S7.2, data fusion technology is used to combine image data with data from various sensors and comprehensively analyze these data through advanced algorithms (such as machine learning models) to identify potential pollution patterns and correlations.

[0154] Use statistical methods or artificial intelligence techniques, such as neural networks, to analyze this data for correlations and patterns. For example, regression analysis can be used to understand the relationship between specific gas emissions and temperature changes.

[0155] Comprehensive use of information provided by multiple data sources can more accurately identify pollution sources and their potential causes. Data fusion technology can reveal the hidden relationships between different data and provide decision support for environmental management at construction sites.

[0156] In step S7.3, based on data fusion, feature extraction is the process of identifying the data features that are most informative for classification or prediction. These features are then used to build a model for automatically identifying and classifying pollution sources at the construction site.

[0157] Feature extraction can be manually designed, such as selecting spectral features at specific wavelengths, or through automatic feature learning methods, such as convolutional layers in deep learning that automatically identify image features.

[0158] Correct feature extraction is the key to successful data analysis and pattern recognition. Effective feature extraction not only improves the accuracy of pollution identification, but also optimizes the performance of the model and reduces the demand for computing resources.

[0159] Step S8 specifically includes the following steps:

[0160] S8.1. Select and train a segmentation model: Select a suitable image segmentation model, such as U-Net, and train it with actual data;

[0161] S8.2. Perform image segmentation: Apply the trained segmentation model on real-time data to accurately segment the pollution sources in the image;

[0162] S8.3. Classification and labeling: Classify and label the segmented image areas according to the pollution type.

[0163] Specifically, in step S8.1, a suitable segmentation model is selected, such as U-Net, which is a deep learning network designed for medical image segmentation, but is also widely used in image segmentation tasks in other fields. The architecture of U-Net is particularly suitable for small objects in images and scenes with complex backgrounds, and its symmetrical contraction path and expansion path help the network capture features from coarse to fine.

[0164] Training the models typically involves using large amounts of labeled image data that show different types of pollution sources and non-polluted environments to train the network to recognize and segment the relevant features in the images.

[0165] Accurate image segmentation models can effectively distinguish pollution sources from background in images, which is crucial for subsequent monitoring, analysis and response measures. Through training, the model can automatically process complex image data, improving processing speed and accuracy.

[0166] Step S8.2, performing image segmentation, refers to applying the trained model to the real-time captured data to accurately segment the pollution source from the background. This step uses the model to predict whether each pixel belongs to a specific pollution category.

[0167] Deep learning-based image segmentation techniques, such as U-Net, achieve accurate segmentation of the entire image by analyzing local regions of the image (through convolution operations) and learning the association between these regions and specific categories.

[0168] Real-time and accurate image segmentation enables the system to instantly identify and respond to environmental pollution events, reducing the potential impact of pollution. In addition, accurate segmentation results provide reliable input for subsequent analysis.

[0169] Step S8.3, classification annotation involves classifying the segmented regions of the image according to pre-defined categories. This is usually based on the features learned by the model during the training phase, such as shape, color, and texture.

[0170] The classification process uses the output of the segmentation model (usually a probability map) to determine the pollution class to which each region most likely belongs.

[0171] Providing detailed category information for each segmented pollution source will help more accurate environmental impact assessment and more targeted pollution control. This classification information can help decision makers and environmental scientists understand the specific pollution situation at the construction site and formulate corresponding measures.

[0172] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A construction pollution detection method based on image recognition, It is characterized in that The following steps are involved: S1. Install a multispectral camera to collect image data in multiple spectral bands to obtain comprehensive light information at the construction site; S2. Use high dynamic range imaging computational photography technology to synthesize and process multispectral images to enhance image details and quality; S3. Deploy multiple fixed cameras around the construction site to form a distributed network to capture image data from different angles to cover the entire construction area; S4. Deploy drones or ground robots equipped with cameras, adjust the shooting angle and position regularly or as needed, and obtain image data with more dynamic perspectives; S5, using background subtraction algorithm to learn and remove the unchanged background information according to the dynamic changes of the construction site; S6. Train and apply deep learning adaptive models to analyze collected image data to identify and extract features of potential pollution sources; S7, combining multimodal data to identify pollution sources; S8. Use image segmentation technology to classify and segment pollution sources and identify different types of pollutants; S9. Deploy edge computing devices on site to process image and sensor data in real time and reduce data transmission latency. S10. Use Hadoop and Ceph distributed storage systems to securely store and manage large amounts of image and sensor data.

2. A construction pollution detection method based on image recognition according to claim 1, It is characterized in that The S2 step specifically includes the following steps: S2.1, Image capture: Set the multispectral camera to continuously capture the same scene at different exposure values; S2.2, Image synthesis: using software to synthesize multiple exposure images into a single high dynamic range image; S2.3, Post-processing: Adjust the color balance, contrast and clarity of the composite image.

3. According to the method for detecting pollution level of construction based on image recognition as described in claim 1, It is characterized in that The S4 step specifically includes the following steps: S4.

1. Select the appropriate drone / robot: Select the appropriate drone or ground robot according to the operation requirements and environment; S4.

2. Install the camera: Install and fix the camera on the drone or robot; S4.3, Programming and Debugging: Program drones or robots to perform scheduled or on-demand filming missions; S4.

4. Real-time monitoring and control: Set up a control center to receive and process image data sent back by drones or robots in real time.

4. The method for detecting pollution level of construction based on image recognition according to claim 1, It is characterized in that The step S5 specifically includes the following steps: S5.

1. Establish background model: Capture images of the construction site at different time periods to establish a background model; S5.2, Algorithm deployment: Deploy the background subtraction algorithm in the processing system and apply it to the incoming image stream in real time; S5.

3. Continuous learning and updating: As construction progresses, the background model is continuously updated to adapt to environmental changes.

5. The method for detecting pollution level of construction based on image recognition according to claim 1, It is characterized in that The step S6 specifically includes the following steps: S6.

1. Collect and annotate training data: collect images from installed cameras and manually annotate them to train the model; S6.

2. Model training: Use labeled datasets to train deep learning models, such as convolutional neural networks; S6.

3. Model optimization and validation: Optimize and validate model performance using test data sets; S6.4, Model deployment: Deploy the trained model on edge computing devices for real-time image analysis.

6. A construction pollution detection method based on image recognition according to claim 1, It is characterized in that The step S7 specifically includes the following steps: S7.

1. Integrate sensor data: Integrate data from different sensors; S7.

2. Data fusion and analysis: Use data fusion techniques to analyze image and sensor data to identify pollution patterns; S7.

3. Feature extraction and pollution identification: Extract key features from the fused data for identifying and classifying pollution sources.

7. The method for detecting pollution level of construction based on image recognition according to claim 1, It is characterized in that The step S8 specifically includes the following steps: S8.

1. Select and train a segmentation model: Select a suitable image segmentation model, such as U-Net, and train it with actual data; S8.

2. Perform image segmentation: Apply the trained segmentation model on real-time data to accurately segment the pollution sources in the image; S8.

3. Classification and labeling: Classify and label the segmented image areas according to the pollution type.

8. A method for detecting construction pollution based on image recognition according to claim 6, It is characterized in that The sensors in step S7.1 include a gas sensor, a temperature sensor and a humidity sensor.

Citation Information

Cited By

  • Fog gun eagle eye intelligent linkage control system

    CN119310914A

  • A fog cannon eagle eye intelligent linkage control system

    CN119310914B