Intelligent monitoring method and system for port coal dust
By setting up sampling and monitoring points in enclosed storage areas and using image recognition and convolutional neural networks for real-time monitoring, the safety threats and management deficiencies in coal storage in enclosed silos have been addressed, enabling intelligent monitoring and management of coal dust in ports and improving data interaction and ecological management levels.
Patent Information
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- ZHENJIANG PORT GRP CO LTD
- Filing Date
- 2026-01-26
- Publication Date
- 2026-06-05
AI Technical Summary
Coal storage in enclosed silos presents challenges such as spontaneous combustion and fire risks, explosion risks due to excessive dust concentration, and insufficient operational management experience, particularly lacking effective means for monitoring coal dust in ports.
A smart monitoring method based on image recognition and convolutional neural networks is adopted. By setting up sampling monitoring points in a closed storage area, the dust concentration is monitored in real time. Data analysis is performed using image recognition models and cloud databases to determine the dust status and trigger alarms. Image classification and prediction are performed by combining infrared monitoring sensors and convolutional neural networks.
It enables timely detection and elimination of safety threats and hidden dangers in the coal dust storage process, improves the accuracy and reliability of monitoring data, supports the overall improvement of port ecological management, and realizes data interconnection and intelligent management.
Smart Images
Figure CN122156716A_ABST
Abstract
Description
Technical Field
[0001] This invention relates to the field of port ecological monitoring technology, and more specifically, to a smart monitoring method and system for coal dust in ports. Background Technology
[0002] In recent years, with increasingly stringent environmental protection requirements in my country, port bulk cargo terminals, especially newly built and expanded terminals, have gradually begun to adopt enclosed silos for material storage. Compared with traditional open-air storage yards, enclosed silos have certain advantages in terms of environmental protection and material quality assurance, but they also have disadvantages in terms of safety, economy, fire protection, and management.
[0003] Currently, the operation of closed storage facilities containing dust particles, such as coal, faces the following problems: 1. The spontaneous combustion characteristics and fire risk of stored coal; 2. The explosion risk after the coal dust concentration exceeds the limit; 3. Lack of experience in the operation and management of large closed coal yards at coal ports.
[0004] Against this backdrop, we propose a smart monitoring method and system for coal dust in ports to address the aforementioned issues. Summary of the Invention
[0005] To address the aforementioned problems in existing technologies, the present invention aims to provide a smart monitoring method and system for coal dust in ports. This system can perform data analysis based on on-site monitoring data and numerical simulations, monitor the concentration of dust in storage facilities in real time, and provide a scientific basis for decision-making on environmental protection construction plans for coal storage yards.
[0006] To solve the above problems, the technical solution adopted by the present invention is as follows: A smart monitoring method for coal dust in ports includes the following steps: S1. Based on the closed storage area of coal powder at the dock, sampling and monitoring points are set up in the surrounding areas of the dust-prone environment and the upwind area to collect images. S2. The sampled image data is transmitted to the system in real time, and an image recognition model is built within the system to judge the image data. The judgment steps include: The uploaded image data is forward propagated to generate corresponding pixel information. The pixel information is preprocessed to generate boundary points. The image recognition model predicts the information of adjacent pixels to amplify the data. Finally, the optimal boundary point is suppressed and the image information within the optimal boundary point is used as the judgment criterion for cloud upload. S3. Build a cloud database on the system side to classify coal powder storage status; S4. The image recognition model extracts features from the image information within the corresponding best boundary point in each image data, and outputs the extracted feature vector in a single channel. This feature vector is then matched with the feature vector of the coal powder storage status image in S3 using cosine similarity. The system determines whether to trigger an alarm based on whether the similarity exceeds the threshold.
[0007] Furthermore, in step S2, the specific steps for generating the optimal boundary points are as follows: Collect all pixels used to define the boundary, and each pixel has a corresponding feature vector; Determine the starting point of the boundary; the starting point is the point with the smallest absolute value of the feature vector among all pixels. The remaining boundary pixels are sorted according to the polar angle between the starting pixel and each remaining boundary pixel. All optimal boundary points are constructed by traversing the Graham scan algorithm.
[0008] Furthermore, the polar angle is the angle between the coordinate features of each pixel and the horizontal direction of the starting point, which is counterclockwise.
[0009] Furthermore, based on the sorted boundary pixels, the initially generated boundary points are amplified, and the amplification is based on the difference in vector coordinates between the current sequence pixels and the initial boundary points.
[0010] Furthermore, in step S3, the system performs in-depth learning on the sampled image data through a convolutional neural network model to classify the image data into multiple storage state image groups.
[0011] Furthermore, the cosine similarity matching calculation formula is as follows:
[0012] Where x is the feature vector of the boundary pixel information, and y is the feature vector of the coal powder storage status image in the system.
[0013] Furthermore, the similarity threshold is not less than 0.5.
[0014] The system constructed based on the aforementioned intelligent monitoring method for coal dust in ports includes: The data acquisition module uses an infrared monitoring sensor to sample images of the coal powder status at each monitoring point; On the system side, an image recognition model and a convolutional neural network model are deployed internally. The convolutional neural network model is used to identify and classify the sampled images, and to cluster comparison images of different dust states. The image recognition model is used to perform forward propagation calculation on the image data and generate corresponding pixel information. The pixel information is preprocessed to generate boundary points, and after amplification, the optimal boundary points are extracted. The image within the optimal boundary points is used as the judgment image. The alarm module determines whether to trigger an alarm based on the cosine similarity matching value between the judged image and the comparison image group.
[0015] Compared with the prior art, the beneficial effects of the present invention are as follows: (1) This invention can detect safety threats in the coal dust storage process in a timely manner by monitoring data in a closed storage environment, and eliminate hidden dangers by replacing human intervention with intelligent monitoring; (2) By amplifying, predicting and suppressing the monitoring image data, the present invention can ensure the accuracy of the judgment data image and indirectly improve the accuracy of the subsequent deep learning simulation recognition, which is more reliable than traditional image recognition. (3) The monitoring system of the present invention can be connected to an external port ecological platform after expansion to conduct data interaction and further improve the overall management level of port ecology. Attached Figure Description
[0016] Figure 1 This is a schematic diagram of the monitoring method of the present invention; Figure 2 This is a schematic diagram illustrating the boundary construction using the Graham algorithm in this invention. Figure 3 This is a schematic diagram of the convolutional model of the convolutional neural network model in this invention; Figure 4 This is a state diagram of different coal dust particles identified by the convolutional neural network model in this invention. Detailed Implementation
[0017] The present invention will be further described below with reference to specific embodiments.
[0018] like Figures 1-4 As shown, A smart monitoring method for coal dust in ports includes the following steps: S1. Based on the closed storage area of coal powder at the dock, sampling and monitoring points are set up in the surrounding areas of the dust-prone environment and the upwind area to collect images. S2. The sampled image data is transmitted to the system in real time, and an image recognition model is built within the system to judge the image data. The judgment steps include: The uploaded image data is forward propagated to generate corresponding pixel information. The pixel information is preprocessed to generate boundary points. The image recognition model predicts the information of adjacent pixels to amplify the data. Finally, the optimal boundary point is suppressed and the image information within the optimal boundary point is used as the judgment criterion for cloud upload. S3. Build a cloud database on the system side to classify coal powder storage status; S4. The image recognition model extracts features from the image information within the corresponding best boundary point in each image data, and outputs the extracted feature vector in a single channel. This feature vector is then matched with the feature vector of the coal powder storage status image in S3 using cosine similarity. The system determines whether to trigger an alarm based on whether the similarity exceeds the threshold.
[0019] The above method flow will be further explained below with reference to specific embodiments: For step S2, the preprocessing of the data image can be achieved by performing histogram equalization on the acquired data image, which equalizes and redistributes pixel intensity values to expand the dynamic range and increase the contrast between the coal dust-affected area and the surrounding background. After histogram equalization, an image segmentation algorithm is used to segment the image to separate specific coal dust areas. The image segmentation algorithm will not be described in detail here. The boundaries of the corresponding segmented areas can be regarded as subsequent boundary points.
[0020] The specific steps for generating the corresponding optimal boundary points are as follows: Collect all pixels used to define the boundary, and each pixel has a corresponding feature vector; Determine the starting point of the boundary; the starting point is the point with the smallest absolute value of the feature vector among all pixels. The remaining boundary pixels are sorted according to the polar angle between the starting pixel and each remaining boundary pixel. All optimal boundary points are constructed by traversing the Graham scan algorithm.
[0021] The polar angle is the counterclockwise angle of the coordinate features of each pixel relative to the horizontal direction of the starting point. Based on the sorted boundary pixels, the initially generated boundary points are expanded, with the expansion based on the difference in vector coordinates between the current sequence pixels and the initial boundary points. This expansion of each boundary point allows adjacent pixels to be accommodated within the corresponding boundary range, facilitating subsequent determination of image boundary sharpness and thus meeting the accuracy requirements of subsequent cosine similarity matching.
[0022] Similarly, in step S3, the system performs in-depth learning on the sampled image data through a convolutional neural network model to classify the image data into multiple storage state image groups.
[0023] The specific steps for constructing the neural network structure are as follows: The first layer of the neural network, layer zero, is a convolutional layer (conv2d_1): the kernel size is 3×3, the number of kernels is 32, and the stride is 2. An input image with a size of 416×416×3 is processed by convolution to output an image with a size of 208×208×32.
[0024] The second layer of the neural network is constructed, which is a convolutional layer (conv2d_2): the kernel size is 3×3, the number of kernels is 64, and the stride is 2. The input image with size 208×208×32 is processed by convolution and the output image size is 208×208×64.
[0025] The third layer of the neural network is constructed, which is an inverse residual convolutional layer (bneck_1): the inverse residual convolution consists of two 1×1 convolutions and one 3×3 convolution, with a BN layer and a ReLU activation function added after each convolutional layer. The 208×208×64 feature map is output as a 208×208×64 feature map after the inverse residual convolution, and then output to bneck_2; The fourth layer of the neural network is constructed, which is an inverse residual convolutional layer (bneck_2): the inverse residual convolution consists of two 1×1 convolutions and one 3×3 convolution, with a BN layer and a ReLU activation function added after each convolutional layer. The 208×208×64 feature map is output as a 104×104×128 feature map after the inverse residual convolution, and then output to bneck_3; The fifth layer of the neural network is constructed. The fifth layer is an inverse residual convolutional layer (bneck_3): the feature map with a size of 104×104×128 is output as a feature map with a size of 52×52×256 after inverse residual convolution, and then output to samBneck_1.
[0026] The sixth layer of the neural network is constructed. The sixth layer is a SAM inverse residual convolution (samBneck_1): the feature map of size 52×52×256 is output as a feature map of size 52×52×256 after passing through the SAM inverse residual convolution, and then output to bneck_4.
[0027] The seventh layer of the neural network is constructed. The seventh layer is an inverse residual convolutional layer (bneck_4): the feature map with a size of 52×52×256 is output as a feature map with a size of 26×26×512 after inverse residual convolution, and then output to samBneck_2.
[0028] The eighth layer of the neural network is constructed. The eighth layer is a SAM inverse residual convolution (samBneck_2): the feature map of size 26×26×512 is output as a feature map of size 26×26×512 after the SAM inverse residual convolution, and the output is bneck_5.
[0029] The ninth layer of the neural network is constructed. The ninth layer is an inverse residual convolutional layer (bneck_5): the feature map with a size of 26×26×512 is output as a feature map of 13×13×1024 after inverse residual convolution, and the output is samBneck_3.
[0030] Construct the tenth layer of the neural network. The tenth layer is SAM inverse residual convolution 1 (SAMBneck_3): the feature map of size 13×13×1024 is output as a feature map of size 13×13×1024 after SAM inverse residual convolution, and the output is bneck_6.
[0031] In the entire 10-layer convolutional neural network, layers 3, 4, 5, 7, and 9 are inverse residual convolutional layers. The inverse residual structures in these layers are as follows: Figure 3 As shown, residual connections are performed if and only if the input and output have the same number of channels.
[0032] In the entire 10-layer convolutional neural network, layers 6, 8, and 10 are inverse residual convolutional layers that introduce a spatial attention module. In these three layers, we added a spatial attention module to the inverse residual convolutional layer. The spatial attention module is as follows: Figure 3 As shown, this module takes the 3D features extracted by the feature extraction network as input and generates a 2D vector representing the importance of each region. Considering that the weight information of local features cannot only be based on the features of the current region, but also on contextual information, this network does not directly use 1×1 convolution. Instead, it first uses a 2D convolution pair with a kernel size of 3 to reduce the dimensionality, making the output channels 1 / r of the original, until the output channels are less than r.
[0033] The formula for calculating cosine similarity matching is as follows.
[0034] Where x is the feature vector of the boundary pixel information, and y is the feature vector of the coal powder storage status image within the system. The similarity threshold is generally not less than 0.5, and in this embodiment, it can be set to 0.7.
[0035] Similarly, the system constructed based on the aforementioned intelligent monitoring method for coal dust in ports is also within the scope of protection of this invention. This system mainly includes: The data acquisition module uses an infrared monitoring sensor to sample images of the coal powder status at each monitoring point; On the system side, an image recognition model and a convolutional neural network model are deployed internally. The convolutional neural network model is used to identify and classify the sampled images, and to cluster comparison images of different dust states. The image recognition model is used to perform forward propagation calculation on the image data and generate corresponding pixel information. The pixel information is preprocessed to generate boundary points, and after amplification, the optimal boundary points are extracted. The image within the optimal boundary points is used as the judgment image. The alarm module determines whether to trigger an alarm based on the cosine similarity matching value between the judged image and the comparison image group.
[0036] The aforementioned system, after expansion, can integrate with existing port management systems, dust monitoring systems, equipment and facility integrated management systems, and other third-party systems to complete interface data docking and joint debugging, enabling real-time interaction of various environmental and treatment facility data. System integration can be carried out using interface protocols such as Web Service, MQTT, and API, or by establishing an intermediate library for docking.
[0037] This forms a holistic intelligent ecological environment monitoring platform, providing a data exchange system to support internal and external data interaction, thereby creating a data exchange and transmission system that is interconnected and shared across all levels. Standardized interface protocols are used to define the specifications for application and service interaction, including data transmission formats and protocols. The data exchange interface should adhere to the principle of technology neutrality, selecting mainstream technologies such as XML, JCA, Web Service, and XPDL.
[0038] Furthermore, the aforementioned methods, steps, and systems can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of this invention, or the part that contributes to the prior art, or a portion of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of this invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.
[0039] The technical scope of this invention is not limited to the content described above. Those skilled in the art can make various modifications and variations to the above embodiments without departing from the technical concept of this invention, and all such modifications and variations should fall within the protection scope of this invention.
Claims
1. A smart monitoring method for coal dust in ports, characterized in that, Includes the following steps: S1. Based on the closed storage area of coal powder at the dock, sampling and monitoring points are set up in the surrounding areas of the dust-prone environment and the upwind area to collect images. S2. The sampled image data is transmitted to the system in real time, and an image recognition model is built within the system to judge the image data. The judgment steps include: The uploaded image data is forward propagated to generate corresponding pixel information. The pixel information is preprocessed to generate boundary points. The image recognition model predicts the information of adjacent pixels to amplify the data. Finally, the optimal boundary point is suppressed and the image information within the optimal boundary point is used as the judgment criterion for cloud upload. S3. Build a cloud database on the system side to classify coal powder storage status; S4. The image recognition model extracts features from the image information within the corresponding best boundary point in each image data, and outputs the extracted feature vector in a single channel. This feature vector is then matched with the feature vector of the coal powder storage status image in S3 using cosine similarity. The system determines whether to trigger an alarm based on whether the similarity exceeds the threshold.
2. The intelligent monitoring method for coal dust in ports according to claim 1, characterized in that: In step S2, the specific steps for generating the optimal boundary points are as follows: Collect all pixels used to define the boundary, and each pixel has a corresponding feature vector; Determine the starting point of this boundary; the starting point is the point with the smallest absolute value of the feature vector among all pixels. The remaining boundary pixels are sorted according to the polar angle between the starting pixel and each remaining boundary pixel. All optimal boundary points are constructed by traversing the Graham scan algorithm.
3. The intelligent monitoring method for coal dust in ports according to claim 2, characterized in that: The polar angle is the angle between the coordinate features of each pixel and the horizontal direction of the starting point, which is counterclockwise.
4. The intelligent monitoring method for coal dust in ports according to claim 2, characterized in that: Based on the sorted boundary pixels, the initially generated boundary points are amplified, and the amplification is based on the difference between the vector coordinates of the current sequence pixels and the initial boundary points.
5. The intelligent monitoring method for coal dust in ports according to claim 1, characterized in that: In step S3, the system performs in-depth learning on the sampled image data through a convolutional neural network model to classify the image data into multiple storage state image groups.
6. The intelligent monitoring method for coal dust in ports according to claim 1, characterized in that: The formula for calculating cosine similarity matching is as follows. Where x is the feature vector of the boundary pixel information, and y is the feature vector of the coal powder storage status image in the system.
7. The intelligent monitoring method for coal dust in ports according to claim 6, characterized in that: The similarity threshold is not less than 0.
5.
8. The system constructed according to the intelligent monitoring method for coal dust in ports as described in any one of claims 1-7, characterized in that, include: The data acquisition module uses an infrared monitoring sensor to sample images of the coal powder status at each monitoring point; On the system side, an image recognition model and a convolutional neural network model are deployed internally. The convolutional neural network model is used to identify and classify the sampled images, and to cluster comparison images of different dust states. The image recognition model is used to perform forward propagation calculation on the image data and generate corresponding pixel information. The pixel information is preprocessed to generate boundary points, and after amplification, the optimal boundary points are extracted. The image within the optimal boundary points is used as the judgment image. The alarm module determines whether to trigger an alarm based on the cosine similarity matching value between the judged image and the comparison image group.
9. An electronic device comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, characterized in that... When the processor executes the program, it implements the steps of the intelligent monitoring method for coal dust in ports as described in any one of claims 1 to 7.
10. A non-transitory computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by the processor, it is used to implement the steps of the intelligent monitoring method for coal dust in ports as described in any one of claims 1 to 7.