Dangerous waste landfill leakage source detection method based on instance segmentation
By using the YOLOSeg-EViT network at the bottom of the landfill for instance segmentation, the problem of difficult to identify and detect the leakage source in the landfill in the prior art is solved, and accurate detection of cracks, damage and leakage is achieved, which improves the Detection efficiency and accuracy.
Patent Information
- Application Number
- CN202411556322.4
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-11-04
- Publication Date
- 2025-02-11
AI Technical Summary
The prior art is difficult to effectively identify and detect leakage sources in landfills, especially in complex dark light environments and narrow spaces, and it is impossible to effectively identify leakage risks caused by cracks.
Using a detection method based on instance segmentation, the video collected at the bottom of the landfill is processed through the YOLOSeg-EViT network to generate mask marks and location information of the leakage source to achieve accurate detection of cracks, damage and leakage.
It realizes efficient and accurate detection of landfill leakage sources in complex environments, improves the identification accuracy and detection efficiency of leakage sources, can effectively identify hidden dangers such as cracks, and supports subsequent governance measures.
Smart Images
Figure CN119418122A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of environmental monitoring, and in particular to a method for detecting leakage sources of hazardous waste landfills based on instance segmentation. Background Art
[0002] As an important facility for the disposal of hazardous waste, hazardous waste landfills are mainly used to prevent the migration of harmful components such as heavy metals in the waste to the environment and human body to ensure safe disposal. However, landfill leakage detection is crucial to prevent environmental pollution, ensure the normal operation of facilities and reduce potential safety risks. Landfill leakage mainly comes from liquid leachate generated during the waste landfill process. After these liquids react with harmful substances in the waste, they may cause toxic substances in the leachate. If these substances are discharged directly without treatment, they will cause serious harm to groundwater, soil and ecosystems.
[0003] Although landfills are usually equipped with an impermeable layer, the risk of leakage still exists due to the influence of construction technology, landfill technology and environmental conditions. Therefore, it is necessary to regularly inspect landfills in order to promptly discover the source of leakage and take control measures. Landfill leakage detection faces multiple challenges: first, the bottom of the landfill is 80-90 cm above the ground, and the inspection space is small; second, there is a lack of power and signal at the bottom, and there is a dark environment, which increases the difficulty of detection; finally, the landfill area is large, and the bottom may contain harmful gases.
[0004] In existing research, technicians have proposed a boundary positioning method for emergency detection of landfill leakage. This method arranges electrodes around the landfill and uses signal calculation to obtain the GPS coordinates of the leakage source. However, this method is easily interfered by other liquids and cannot effectively identify leakage hazards caused by cracks.
[0005] In the field of computer vision, image segmentation methods are considered to be a more accurate way to locate targets. Existing studies, such as Gao et al.'s pixel-by-pixel crack detection method based on U-Net and Liu et al.'s improved UNet model, have shown the effectiveness of deep learning in crack detection. However, the leakage source detection method for the special environment of landfills still needs further exploration.
[0006] To this end, those skilled in the art have proposed a method for detecting leakage sources in hazardous waste landfills based on instance segmentation to solve the above problems. Summary of the invention
[0007] In view of the deficiencies of the prior art, the present invention provides a method for detecting leakage sources in hazardous waste landfills based on instance segmentation, which solves the problems raised in the above-mentioned background technology.
[0008] To achieve the above objectives, the present invention is implemented through the following technical solutions: A method for detecting leakage sources in hazardous waste landfills based on instance segmentation, comprising the following steps:
[0009] S1. Using a patrol robot to collect video at the bottom of the landfill, the video collection equipment includes at least two visible light cameras, and the cameras are equipped with infrared fill light devices to adapt to dark light environments;
[0010] S2, collecting forward and backward images through the camera;
[0011] S3. Use the YOLOSeg-EViT network to perform instance segmentation on the collected video to obtain the mask mark and location information of the leakage source;
[0012] S4. Generate a detection report including the leakage source detection image and its corresponding coordinate information.
[0013] Preferably, in the step S3, the YOLOSeg-EViT network includes a backbone network, a neck and a detection head, wherein:
[0014] The backbone network adopts the EfficientViT model to achieve lightweight feature extraction and optimize detection speed and accuracy;
[0015] The neck is used to integrate and process the extracted features, and the detection head is used to output the detection result of the leakage source.
[0016] Preferably, the leakage sources in step S4 include three categories: cracks, damage and leakage, which are classified and detected by feature extraction and segmentation algorithms, respectively. Cracks are narrow strip structures, damage is block-like fragments with irregular shapes, and leakage is an area with darker color and obvious diffusion state.
[0017] Preferably, the data set formed in step S1 is semi-automatically annotated using Segment Anything Model, and the steps include:
[0018] A1. Use SAM to extract features from the image to be tested and generate a mask;
[0019] A2. Select and save the mask of the leakage source, and manually correct the mask using the LabelMe tool to improve the labeling accuracy.
[0020] Preferably, the detection report in step S4 includes a coordinate system of the leakage source location information, wherein the origin position is the starting position of the inspection robot, and the horizontal and vertical coordinates represent the distance in meters.
[0021] Preferably, in the step S3, the training of the YOLOSeg-EViT network is performed on multiple devices, including a PC and an edge computing platform, to achieve effective detection in different environments;
[0022] The PC is a high performance computing platform;
[0023] The edge computing platform is Jetson AGX Orin to adapt to resource-limited environments and enable portable deployment.
[0024] Preferably, the inspection robot in the step S1 can provide real-time feedback of the inspection results during the inspection process of the landfill, support subsequent data analysis and decision-making, and further improve the efficiency and accuracy of landfill leakage source detection.
[0025] Preferably, in step S2, the resolution of the video is 1280×720 pixels, and the frame rate is 6 frames per second.
[0026] Preferably, in step S1, the inspection robot is equipped with an environmental sensor, which can monitor the temperature, humidity and gas concentration of the landfill in real time, and integrate these environmental data with the video acquisition results to improve the detection accuracy and reliability of the leakage source.
[0027] Preferably, the detection report in step S4 also includes risk assessment information on the leakage source, and uses a machine learning model to classify the detected leakage source into risk levels, so that management personnel can take corresponding treatment measures for different risk levels.
[0028] The present invention provides a method for detecting leakage sources in hazardous waste landfills based on instance segmentation. It has the following beneficial effects:
[0029] 1. The present invention adds a lightweight multi-scale attention image segmentation module EfficientViT to the backbone network. When trained using the self-built data set of the present invention, the average accuracy of the YOLOSeg-EViT network reaches 73.9%, and the FPS value of detecting 1280*720 pixel resolution video can reach 43.1. The present invention deploys the network to an edge computing platform with better portability, achieving an average accuracy of 70.5% and an FPS of 32.9. The landfill leakage source detection algorithm in the present invention can effectively detect the leakage source and has certain practical value.
[0030] 2. The present invention uses a lightweight instance segmentation network YOLOSeg-EViT to detect landfill leakage sources. Compared with the target detection method, the instance segmentation method can generate a mask of the leakage source, so as to better understand the geometric characteristics of the leakage source, which is conducive to taking appropriate solutions. The backbone network part of the network of the present invention uses a lightweight EfficientViT module to effectively reduce network parameters and improve detection efficiency. At the same time, lightweight deployment is achieved on the edge computing platform, which improves the applicability of the algorithm. BRIEF DESCRIPTION OF THE DRAWINGS
[0031] Figure 1 It is the overall flow chart of the present invention;
[0032] Figure 2 This is a network structure diagram of YOLOSeg-EViT according to an embodiment of the present invention;
[0033] Figure 3 This is an example diagram of a landfill leakage source of the present invention;
[0034] Figure 4 A schematic diagram of a data set generation process of the present invention;
[0035] Figure 5 This is an example segmentation effect diagram of the present invention;
[0036] Figure 6 The present invention is a flow chart of the method. DETAILED DESCRIPTION
[0037] The following will be combined with the drawings in the specification of the present invention to clearly and completely describe the technical solutions in the embodiments 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.
[0038] Please see attached Figure 1 - Attachment Figure 6 The embodiment of the present invention provides a method for detecting leakage sources in hazardous waste landfills based on instance segmentation, comprising the following steps:
[0039] S1. Use a patrol robot to collect video at the bottom of the landfill. The video collection equipment includes at least two visible light cameras, and both cameras are equipped with infrared fill light devices to adapt to dark light environments;
[0040] Specifically, the inspection robot is designed to move autonomously in complex environments to ensure comprehensive coverage of all areas of the landfill. Its video acquisition equipment includes at least two high-resolution visible light cameras, the layout of which is optimized to capture on-site image information from different angles.
[0041] The visible light cameras used have excellent image acquisition capabilities and can provide clear and detailed images under good lighting conditions. At the same time, in order to adapt to possible dark light environments, each camera is equipped with an infrared fill light device. These fill lights can be automatically turned on under low light conditions to ensure that high-quality video data can be obtained in any environment, so as not to affect the identification and detection of leakage sources.
[0042] During the inspection process, the inspection robot not only performs video acquisition tasks, but also has real-time data processing and feedback capabilities. Through the integrated computing platform, the robot can perform preliminary analysis on the collected video data, identify potential leakage sources in real time and generate detection results. This process uses advanced image processing algorithms to ensure the efficiency and accuracy of detection.
[0043] In step S1, the inspection robot can provide real-time feedback of the inspection results during the inspection process of the landfill, support subsequent data analysis and decision-making, and further improve the efficiency and accuracy of landfill leakage source detection.
[0044] The dataset formed in step S1 is semi-automatically annotated using the Segment Anything Model. The steps include:
[0045] A1. Use SAM to extract features from the image to be tested and generate a mask;
[0046] A2. Select and save the mask of the leakage source, and manually correct the mask using the LabelMe tool to improve labeling accuracy.
[0047] S2, collecting forward and backward images through the camera;
[0048] The resolution of the video in step S2 is 1280×720 pixels and the frame rate is 6 frames per second.
[0049] S3. Use the YOLOSeg-EViT network to perform instance segmentation on the collected video to obtain the mask mark and location information of the leakage source;
[0050] In step S3, the training of the YOLOSeg-EViT network is performed on multiple devices, including PCs and edge computing platforms, to achieve effective detection in different environments;
[0051] The PC is a high-performance computing platform, specifically configured with an Intel i9-11900KF CPU and an NVIDIA A40 GPU;
[0052] The edge computing platform is Jetson AGX Orin to adapt to resource-limited environments and enable portable deployment.
[0053] Specifically, the collected video data will be processed by instance segmentation through the YOLOSeg-EViT network to extract the leakage source information in the landfill. YOLOSeg-EViT combines the efficient target detection capabilities of the YOLO (You Only Look Once) series with the feature extraction advantages of EViT (Efficient Vision Transformer), which can achieve fast and accurate leakage source identification and positioning.
[0054] The main purpose of instance segmentation is to accurately mark each leakage source area in the video and generate a corresponding mask. By learning a large amount of labeled data, the YOLOSeg-EViT network can distinguish different types of leakage sources in complex scenes, such as cracks, damage, and leakage. After processing, the network will output the mask mark of each leakage source and its location information in the video frame, which is convenient for subsequent analysis and report generation.
[0055] In order to ensure the effective detection of the YOLOSeg-EViT network in various environments, the training process will be carried out on multiple devices. These devices include high-performance PCs and specialized edge computing platforms to ensure the flexibility and adaptability of the model.
[0056] PC as a high-performance computing platform: The PC used is equipped with an Intel i9-11900KF CPU and an NVIDIA A40 GPU, which has powerful computing capabilities. This configuration can effectively process large-scale data sets, perform complex model training and parameter tuning, and ensure efficient learning of the YOLOSeg-EViT network.
[0057] Edge computing platform: To adapt to resource-limited environments, especially in field applications, Jetson AGX Orin is used as an edge computing platform. Designed for low-power, high-performance computing, the platform is capable of running trained models in real time on site to enable instant data processing. The advantage of this portable deployment allows inspection robots to perform fast and accurate leakage source detection in various landfills without relying on expensive cloud computing resources.
[0058] In step S3, the YOLOSeg-EViT network includes the backbone network, neck and detection head, where:
[0059] The backbone network adopts the EfficientViT model to achieve lightweight feature extraction and optimize detection speed and accuracy;
[0060] The neck is used to integrate and process the extracted features, and the detection head is used to output the detection results of the leakage source.
[0061] S4. Generate a detection report including the leakage source detection image and its corresponding coordinate information.
[0062] In step S4, the leakage sources include three categories: cracks, damages and leakages, which are classified and detected through feature extraction and segmentation algorithms. Cracks are narrow strip structures, damages are block-like fragments with irregular shapes, and leakages are areas with darker colors and obvious diffusion states.
[0063] The detection report in step S4 includes a coordinate system of the leakage source location information, wherein the origin position is the starting position of the inspection robot, and the horizontal and vertical coordinates represent the distance in meters.
[0064] In step S1, the inspection robot is equipped with environmental sensors that can monitor the temperature, humidity and gas concentration of the landfill in real time, and integrate these environmental data with the video acquisition results to improve the accuracy and reliability of leakage source detection.
[0065] The detection report in step S4 also includes risk assessment information on the leakage source, and uses a machine learning model to classify the risk level of the detected leakage source so that managers can take appropriate treatment measures for different risk levels.
[0066] Experimental results:
[0067] YOLOSeg-EViT Experiment Results
[0068] In order to evaluate the effectiveness of the network in this paper for leakage source detection, the network model is trained using the training set, and finally the pre-trained model is used to detect in the validation set. The validation process is to compare the detection results of the pre-trained model with the true value of the label file. The final iteration results of the PC-side YOLOSeg-EViT network using the self-built data set in this invention are shown in Table 1.
[0069] Table 1 YOLOSeg-EViT network training results
[0070]
[0071] As shown in Table 1, the network has a high detection accuracy for crack and leakage categories, and a low detection accuracy for breakage, which is consistent with the characteristics of the three types of leakage sources. The shapes of crack and leakage categories are relatively simple, so it is easy for the network to learn the characteristics of these leakage sources. The shape and size of breakage have no specific rules, so it is more difficult for the network to learn features. The FPS value of the network is 43.1, which has a fast inference speed.
[0072] Please refer to the instruction manual Figure 5 , Figure 5 This is the experimental segmentation effect diagram.
[0073] It can be seen from the figure that the algorithm in this paper segments the leakage source without being disturbed by the stains on the concrete wall (the black box area in the figure) and the repair die head (the light box area in the figure), and has a good detection effect.
[0074] Comparative experiment
[0075] In this paper, the MASK R-CNN
[11] network is used in a PC environment. The YOLOv8 network is used for comparative testing with the network in this paper. The test data sets are the same. The comparative experimental results are shown in Table 2.
[0076] Table 2 PC side comparison experimental results
[0077]
[0078] Conclusion: As shown in Table 2, when using the dataset of this paper for testing, the network of this paper has higher mAP and FPS than the MASK R-CNN network and the YOLOv8 network. The size of the pre-trained model completed by the network training of this paper is significantly smaller than that of the MASKR-CNN and YOLOv8 networks, indicating the effectiveness of the network of this paper in the application of landfill leakage source detection.
[0079] Edge computing platform deployment results
[0080] The edge computing platform is suitable for various embedded and IoT devices, and can run in resource-limited environments, reducing dependence on high-performance computers. As a high-performance computing platform based on the ARM64 architecture, NVIDIA Jetson AGX Orin has better scene adaptability due to its small size and low power consumption. In order to verify the detection effect on the edge computing platform, the YOLOv8 network, which performed well in the comparative experiment, was used to conduct an edge computing platform comparative experiment with the network in this paper. The experiment used the self-built data set in this paper for training. The experimental results of the network in this paper and the YOLOv8 network are shown in Table 3.
[0081] Table 3. Experimental results of edge computing platform comparison
[0082]
[0083] Conclusion: As shown in Table 3, the average accuracy of the network in this paper is better than that of the YOLOv8 network on the edge computing platform. Although the accuracy of the network in this paper is lower than that of the YOLOv8 network in detecting leakage sources in the crack category, it is better in detecting leakage categories with higher detection priority. This shows that the lightweight network in this paper has good effects on both PC and edge computing platforms. Although the FPS of the edge computing platform is lower than that of the PC environment in this paper, the cost of the PC hardware in this paper is higher than that of the edge computing box, and the PC does not have the portability of the edge computing platform. At the same time, the edge computing platform can be deployed on inspection machines to improve the efficiency of landfill leakage source detection.
[0084] In summary: The present invention uses a YOLOSeg-EViT instance segmentation algorithm to detect landfill leakage sources, and a lightweight multi-scale attention image segmentation module EfficientViT is added to the backbone network of the algorithm. When trained using the self-built data set in this article, the average accuracy of the YOLOSeg-EViT network reached 73.9%, and the FPS value of detecting 1280*720 pixel resolution video can reach 43.1. This article deploys the network to a more portable edge computing platform, achieving an average accuracy of 70.5% and an FPS of 32.9. The landfill leakage source detection algorithm in this article can effectively detect leakage sources and has certain practical value. At the same time, the algorithm can provide reference and inspiration for feature segmentation problems in other fields. In the future, we will deploy the edge computing platform to the inspection robot to realize real-time detection and further improve the detection efficiency of landfill leakage sources.
[0085] 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 method for detecting leakage sources in hazardous waste landfills based on instance segmentation, characterized in that: The following steps are involved: S1. Using a patrol robot to collect video at the bottom of the landfill, the video collection equipment includes at least two visible light cameras, and the cameras are equipped with infrared fill light devices to adapt to dark light environments; S2, collecting forward and backward images through the camera; S3. Use the YOLOSeg-EViT network to perform instance segmentation on the collected video to obtain the mask mark and location information of the leakage source; S4. Generate a detection report including the leakage source detection image and its corresponding coordinate information.
2. According to the method for detecting leakage sources in hazardous waste landfills based on instance segmentation in claim 1, it is characterized in that: In the S3 step, the YOLOSeg-EViT network includes a backbone network, a neck, and a detection head, wherein: The backbone network adopts the EfficientViT model to achieve lightweight feature extraction and optimize detection speed and accuracy; The neck is used to integrate and process the extracted features, and the detection head is used to output the detection result of the leakage source.
3. According to the method for detecting leakage sources in hazardous waste landfills based on instance segmentation in claim 1, it is characterized in that: In the step S4, the leakage sources include three categories: cracks, damages and leakages, which are classified and detected by feature extraction and segmentation algorithms. Cracks are narrow strip structures, damages are block-like fragments with irregular shapes, and leakages are areas with darker colors and obvious diffusion states.
4. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: The data set formed in step S1 is semi-automatically annotated using the Segment Anything Model, and the steps include: A1. Use SAM to extract features from the image to be tested and generate a mask; A2. Select and save the mask of the leakage source, and manually correct the mask using the LabelMe tool to improve the labeling accuracy.
5. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: The detection report in step S4 includes a coordinate system of the leakage source location information, wherein the origin position is the starting position of the inspection robot, and the horizontal and vertical coordinates represent the distance in meters.
6. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: In the S3 step, the training of the YOLOSeg-EViT network is performed on multiple devices, including PCs and edge computing platforms, to achieve effective detection in different environments; The PC is a high performance computing platform; The edge computing platform is Jetson AGX Orin to adapt to resource-limited environments and enable portable deployment.
7. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: In the above-mentioned step S1, the inspection robot can provide real-time feedback of the inspection results during the inspection process of the landfill, support subsequent data analysis and decision-making, and further improve the efficiency and accuracy of landfill leakage source detection.
8. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: The resolution of the video in step S2 is 1280×720 pixels, and the frame rate is 6 frames per second.
9. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: In the step S1, the inspection robot is equipped with an environmental sensor, which can monitor the temperature, humidity and gas concentration of the landfill in real time, and integrate these environmental data with the video acquisition results to improve the detection accuracy and reliability of the leakage source.
10. The method for detecting leakage sources in hazardous waste landfills based on instance segmentation according to claim 1, characterized in that: The detection report in step S4 also includes risk assessment information on the leakage source, and uses a machine learning model to classify the risk level of the detected leakage source so that management personnel can take corresponding treatment measures for different risk levels.