Intelligent alarm method for coal blockage and coal overflow of coal conveying system based on image recognition

Abstract

According to the intelligent alarm method for coal blockage and coal overflow of the coal conveying system based on image recognition, visible light and infrared thermal imaging bimodal fusion sensing is adopted, and the recognition accuracy of abnormity such as coal blockage and overheating is improved; local real-time reasoning and event intensity scoring are realized based on edge calculation, and false alarms are effectively inhibited; through a non-intrusive plug-in design, seamless joint with the existing DCS / PLC system is realized, and interlocking shutdown and structured alarm are supported; and an event closed-loop filing mechanism is established, images, logs and model states are automatically stored, and a data basis is provided for self-learning optimization and operation and maintenance traceability. According to the scheme, coal conveying intelligent monitoring which is high in robustness, low in delay, easy to deploy and capable of being evolved is achieved.

Description

Technical Field

[0001] This invention belongs to the field of intelligent monitoring and early warning technology for coal conveying systems, specifically relating to an intelligent alarm method for coal blockage and overflow in coal conveying systems based on image recognition. Background Technology

[0002] Currently, coal conveying systems in thermal power plants primarily rely on rotary paddle switches for monitoring coal blockage and spillage. These devices are typically installed at the coal drop pipe of the conveyor belt to determine if material is accumulating. However, this type of device has several limitations: Susceptible to coal type and environment: Under humid conditions and with easily sticky coal, bridging coal blockage is very likely to occur at the coal chute, causing the rotary switch to fail to turn, refuse to operate, or give false alarms, resulting in the system failing to alarm in time and ultimately leading to serious coal spill.

[0003] High dependence on manual inspection: In order to compensate for the lack of reliability of the equipment itself, on-site personnel often need to conduct manual monitoring through video surveillance in the control room, or arrange inspection personnel to check the coal drop area at regular intervals, which increases the labor intensity and management cost of the operators.

[0004] Delayed response and lack of automatic early warning capabilities: Traditional solutions lack intelligent identification functions, cannot dynamically track the state of coal flow, and do not have the data processing capabilities for automatic recording, statistics, and tracing, resulting in delayed feedback of information after coal spills and missing the opportunity for first-time intervention.

[0005] Unable to identify fire hazards: coal dust accumulation or high equipment temperature can easily cause fires. However, current coal conveying systems are generally not equipped with automatic means to identify abnormal temperatures. They can only make judgments by manually observing external characteristics such as ash accumulation, coal accumulation, and odors, which poses a serious hidden danger.

[0006] In summary, existing coal conveying systems suffer from problems such as slow response, reliance on manual labor, and incomplete identification in monitoring coal blockage, coal spillage, and fire hazards. There is an urgent need for an intelligent identification and early warning system that integrates image recognition, real-time alarm, and data retention to improve the system's operational safety and intelligence level. Summary of the Invention

[0007] The present invention aims to at least partially solve one of the technical problems in the related art.

[0008] Therefore, the first objective of this invention is to propose an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition.

[0009] The second objective of this invention is to propose an intelligent alarm device for coal blockage and overflow in a coal conveying system based on image recognition.

[0010] To achieve the above objectives, a first aspect of the present invention proposes an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition, comprising: S1, deploy visible light spherical cameras and infrared thermal imaging equipment at key locations in the coal conveying system to simultaneously acquire images of the coal pile morphology and temperature field distribution; S2, the edge computing server performs asynchronous parallel processing on multi-source images, extracting the contour features of visible light images and the dynamic hotspot features of infrared images respectively; S3, based on the extracted morphological and temperature features, performs multidimensional fusion analysis to generate event intensity scores and filter false alarms; S4 triggers an adaptive alarm strategy based on the event intensity score, and achieves multi-level linkage response through voice broadcast, switch signal output and text push; S5 automatically archives the identified images, scoring results, and response action data of alarm events, generating event logs containing timestamps and processing suggestions for traceability and analysis.

[0011] In one embodiment of the present invention, S1 includes: S11, a visible light spherical camera is installed 1.5-2.0 meters above the coal drop pipe at the head of the conveyor belt, and an infrared thermal imaging device is installed in the contact area between the side wall of the coal drop pipe and the equipment shell; The S12 is equipped with a visible light camera with a focal length adjustment range of 20-120mm, an infrared device with a temperature detection accuracy of ±0.5℃, and both devices have a sampling frequency synchronized to 10Hz.

[0012] In one embodiment of the present invention, S2 includes: S21 uses a multi-process asynchronous inference engine to perform deep learning-based contour segmentation on visible light images, extracting coal pile height, volume change rate, and occlusion area ratio; S22 executes a hotspot tracking algorithm on infrared images to calculate the area growth rate and temperature gradient of temperature-rising patches, and establishes a dynamic comparison model with historical temperature distribution.

[0013] In one embodiment of the present invention, S3 includes: S31 calculates the event intensity score using a weighted fusion formula, filters false alarms based on the score threshold, and triggers a false alarm flag and records environmental interference characteristics when the score is below the threshold.

[0014] In one embodiment of the present invention, S5 includes: S51 packages and stores the recognition model status snapshot, CPU utilization, camera angle data and alarm event data to form a multi-dimensional event data packet. The S52 displays the event log retrieval function through a web interface, supporting multi-dimensional queries by timestamp, device number, event type, and processing suggestions.

[0015] To achieve the above objectives, a second aspect of the present invention provides an intelligent alarm device for coal blockage and overflow in a coal conveying system based on image recognition, comprising: The multimodal image acquisition module is used to deploy visible light spherical cameras and infrared thermal imaging equipment at key locations in the coal conveying system to simultaneously acquire images of the coal pile morphology and temperature field distribution. The asynchronous feature extraction module is used to run the edge computing server to perform asynchronous parallel processing on multi-source images, extracting the contour features of visible light images and the dynamic hotspot features of infrared images respectively. The multidimensional fusion analysis module is used to perform multidimensional fusion analysis based on extracted morphological and temperature features, generate event intensity scores, and filter false alarms. The alarm strategy triggering module is used to trigger an adaptive alarm strategy based on the event intensity score, and achieve multi-level linkage response through voice broadcast, switch signal output and text push. The event log generation module is used to automatically archive the identification images, scoring results, and response action data of alarm events, and generate event logs containing timestamps and processing suggestions for traceability and analysis.

[0016] This invention provides an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition. By constructing an intelligent identification architecture for the coal conveying system that combines "image recognition + infrared sensing + local intelligent computing + multi-channel linkage," this invention achieves the following technical effects: This invention overcomes the limitations of traditional rotary switch and single-channel camera monitoring through multimodal intelligent fusion recognition. Utilizing simultaneous analysis of visible light and infrared images, it achieves three-dimensional perception of coal flow status, coal accumulation trends, and temperature anomalies, significantly improving the accuracy and real-time performance of identifying coal blockages, overflows, and fire hazards. The system employs a local industrial-grade server for edge computing, enabling on-site data identification, decision-making, and alarms with a response time of less than 3 seconds, avoiding cloud dependence and ensuring high stability. Once an anomaly is detected, it automatically captures and archives images, recognition results, and response actions, supporting post-event traceability, responsibility determination, and statistical analysis, meeting the digital operation and maintenance needs of power plants. The device uses a non-intrusive external deployment, requiring no modification to the existing programmable control system, offering strong compatibility and a short integration cycle, suitable for rapid upgrades and renovations of existing power plants. Simultaneously, intelligent early warning replaces manual operation, effectively reducing labor intensity and the risk of misjudgment, and ensuring reliable operation, especially during vulnerable periods such as night shifts and holidays, comprehensively improving the safety management level of the coal conveying system.

[0017] In summary, this invention not only solves the problems of high false alarm rate, slow response, and heavy reliance on manual labor in existing coal conveying systems, but also provides industry-leading technical support for the intelligent operation and maintenance of fuel systems, with broad application prospects and significant engineering promotion value.

[0018] Additional aspects and advantages of the invention will be set forth in part in the description which follows, and in part will be obvious from the description, or may be learned by practice of the invention. Attached Figure Description

[0019] The above and / or additional aspects and advantages of the present invention will become apparent and readily understood from the following description of the embodiments taken in conjunction with the accompanying drawings, wherein: Figure 1 This is a flowchart of an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition, according to an embodiment of the present invention. Figure 2 This is a schematic diagram of the camera position according to an embodiment of the present invention; Figure 3 This is a schematic diagram of the reflector position according to an embodiment of the present invention; Figure 4 This is a structural diagram of an intelligent alarm device for coal blockage and overflow in a coal conveying system based on image recognition, according to an embodiment of the present invention. Detailed Implementation

[0020] It should be noted that, unless otherwise specified, the embodiments and features described in the present invention can be combined with each other. The present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0021] To enable those skilled in the art to better understand the present invention, the technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0022] The following describes, with reference to the accompanying drawings, an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition, according to an embodiment of the present invention.

[0023] Example 1 Figure 1 This is a flowchart of an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition, according to an embodiment of the present invention. Figure 1 As shown, it includes: S1 deploys visible light spherical cameras and infrared thermal imaging equipment at key locations in the coal conveying system to simultaneously acquire images of the coal pile morphology and temperature field distribution.

[0024] In some implementations, the visible light spherical camera uses an industrial-grade high-definition camera with a resolution of at least 4K, supporting 360° horizontal rotation and a 120° vertical viewing angle. It features intelligent zoom and motion tracking capabilities, and can adapt to image acquisition under complex lighting conditions. The infrared thermal imaging device uses an uncooled infrared detector with a resolution of 640×512 pixels, operating in the wavelength range of 7.5–13 μm, with a frame rate of at least 30 fps, a temperature measurement accuracy of ±1.5 ℃, and supports dynamic temperature spectrum modeling and hotspot tracking. Both devices are connected to a local image recognition server via an industrial Ethernet interface to ensure real-time transmission and synchronous processing of image data.

[0025] Furthermore, the deployment of cameras and infrared equipment must meet the following parameters: installation height of 3–5 meters, horizontal field of view covering the entire coal-falling area, and vertical viewing angle ensuring complete capture of changes in coal pile height; the temperature measurement range of the infrared equipment is set to -20℃ to 150℃ to adapt to temperature changes in the coal pile under different operating conditions. In practical applications, this step is typically deployed at key nodes in the coal conveying system, such as the head of the belt conveyor and the outlet of the coal chute, where coal blockage and spillage are prone to occur, to ensure comprehensive monitoring of the coal flow status.

[0026] Furthermore, S1 includes: S11, a visible light spherical camera is installed 1.5-2.0 meters directly above the coal drop pipe at the head of the conveyor belt, and an infrared thermal imaging device is installed in the contact area between the side wall of the coal drop pipe and the equipment shell.

[0027] Specifically, in the image acquisition unit deployment scheme of this invention, a visible light spherical camera is installed 1.5-2.0 meters directly above the coal drop pipe at the head of the conveyor belt, while an infrared thermal imaging device is installed in the contact area between the side wall of the coal drop pipe and the equipment casing. This step is technically based on the principle of multimodal image fusion recognition, aiming to improve the ability to identify coal flow status, coal accumulation trends, and potential fire hazards through the collaborative analysis of spatial and temperature information.

[0028] In some implementations, the visible light spherical camera uses industrial-grade high-definition video equipment with IP67 protection rating, supporting 360° horizontal rotation and vertical tilt adjustment, with a focal length range of 2.8-12mm, adaptable to the identification needs of different coal flow patterns. Its installation height is 1.5-2.0 meters, ensuring the camera's field of view covers the coal chute outlet, conveyor belt head, and coal pile accumulation area, avoiding obstruction due to excessively low viewing angles or blind spots due to excessively high angles. The infrared thermal imaging equipment uses a non-contact temperature measurement method, installed on the side wall of the coal chute in contact with the equipment casing, ensuring it can capture abnormal heat conduction between the coal and the equipment, such as localized temperature rises and hotspot movement. This equipment has a 640×512 pixel resolution, a temperature measurement range of -20℃ to 300℃, and an accuracy of ±1℃ or ±1%, meeting the identification needs of typical fire hazard scenarios such as coal dust accumulation and frictional heating.

[0029] Specifically, the installation positions of the cameras and infrared devices must meet the recommendations in ISO 17201-1 regarding the installation height and field of view coverage of industrial monitoring equipment to ensure the integrity and continuity of image acquisition. The sampling frequency of the infrared devices is set to 10Hz, synchronized with the 30Hz frame rate of the visible light images, and the alignment and fusion of multi-source data are achieved through the clock synchronization module of the edge server.

[0030] The S12 is equipped with a visible light camera with a focal length adjustment range of 20-120mm, an infrared device with a temperature detection accuracy of ±0.5℃, and both devices have a sampling frequency synchronized to 10Hz.

[0031] In some implementations, visible light cameras employ motorized zoom lenses with a focal length adjustment range of 20-120mm, covering a variety of perspectives from wide-angle to telephoto, suitable for monitoring coal flow conditions at different distances and angles. This focal length range meets the clear imaging requirements of coal chutes, conveyor belt heads, and surrounding areas in coal conveying systems, offering significant advantages, especially in identifying changes in coal pile height and material flow patterns. Infrared thermal imaging equipment uses uncooled infrared detectors with a temperature detection accuracy of ±0.5℃, meeting typical performance standards for industrial-grade thermal imaging equipment (such as ISO 80601-2-61:2012), effectively capturing potential fire hazards such as coal dust accumulation and equipment overheating.

[0032] Specifically, the sampling frequency of both devices was synchronously set to 10Hz, meaning 10 frames of image data were acquired per second. This frequency ensures real-time performance while also accommodating the computational load of data processing, meeting the typical requirements for dynamic event capture in industrial video surveillance systems (such as the sampling recommendations for real-time image processing in the IEC 61850-7-4 standard). The synchronization mechanism is achieved through hardware clock triggering or software timestamp alignment, ensuring strict temporal correspondence between visible light and infrared images, providing a reliable time reference for subsequent multimodal data fusion.

[0033] S2 runs an edge computing server to perform asynchronous parallel processing on multi-source images, extracting contour features from visible light images and dynamic hotspot features from infrared images respectively.

[0034] This step utilizes an industrial-grade edge computing server deployed in the local electronics room of the coal conveying system to process image data from high-definition visible light PTZ cameras and infrared thermal imaging equipment in real time, asynchronously, and in parallel, thereby achieving efficient identification and judgment of coal flow status, coal accumulation trends, and potential fire hazards.

[0035] In some implementations, edge computing servers employ a multi-process asynchronous inference architecture, where each image channel (visible light and infrared) runs its recognition task independently without interference. Visible light images are processed using deep learning-based contour detection models (such as YOLOv5 or U-Net) to extract the edge contours, height variations, and occlusion states of the coal pile. The model input is real-time video frames, and the output is a binarized mask image and key geometric parameters (such as area, perimeter, and centroid coordinates). Infrared images, on the other hand, utilize thermal imaging analysis algorithms to extract the temperature distribution, movement trajectory, and temperature rise rate of hotspot areas, identifying abnormal heat sources or localized overheating phenomena, providing data support for fire hazard early warning.

[0036] Specifically, the visible light image processing frequency is 30 frames / second, with a resolution of 1920×1080, and supports H.265 encoding transmission to reduce bandwidth consumption. The infrared image acquisition frequency is 15 frames / second, with a resolution of 640×480, and a temperature measurement accuracy of ±1℃, supporting dynamic temperature spectrum modeling. The asynchronous processing mechanism is implemented through a multi-threaded scheduler, with each image channel independently occupying a GPU computing unit, ensuring a processing latency of less than 300ms to meet real-time monitoring requirements.

[0037] Furthermore, S2 includes: S21 employs a multi-process asynchronous inference engine to perform deep learning-based contour segmentation on visible light images, extracting coal pile height, volume change rate, and occlusion area ratio.

[0038] Specifically, the present invention employs a multi-process asynchronous inference engine to perform deep learning-based contour segmentation on visible light images in order to extract key parameters such as coal pile height, volume change rate, and occlusion area ratio.

[0039] In some implementations, the system utilizes an industrial-grade graphics computing platform deployed in a local electronics room, running a multi-process asynchronous inference engine to achieve parallel processing of multiple visible light images. Each process independently loads a deep learning model (such as U-Net, Mask R-CNN, etc.) to perform real-time contour segmentation on the input image. The segmentation algorithm is based on a convolutional neural network (CNN) structure, using a pre-trained model to perform pixel-level recognition of the coal pile region, extracting the boundary contours of the coal pile, and combining this with a 3D reconstruction algorithm to estimate the height of the coal pile. Furthermore, the system uses a time-series analysis module to dynamically model the changes in the volume of the coal pile in consecutive frames of images, calculating the volume change rate, as shown in the formula:

[0040] in, This indicates the current volume of the coal pile. This represents the volume at the previous moment, in cubic meters (m³). Simultaneously, the system uses image difference and occlusion detection algorithms to identify invalid areas in the coal pile area caused by factors such as equipment movement and personnel obstruction, and calculates the proportion of obstructed areas using the following formula:

[0041] in, Indicates the area of ​​the obscured region. This represents the total recognizable area of ​​the coal pile, in pixels (px²).

[0042] Specifically, the system supports multiple video inputs (up to 16 channels of 1080P@30fps), with inference latency controlled within 100ms, meeting real-time requirements. In terms of model accuracy, the IoU (Intersection over Union) for contour segmentation reaches over 0.85, the volume estimation error is less than 5%, and the occlusion recognition accuracy is higher than 90%.

[0043] S22 executes a hotspot tracking algorithm on infrared images to calculate the area growth rate and temperature gradient of temperature-rising patches, and establishes a dynamic comparison model with historical temperature distribution.

[0044] Specifically, this step first preprocesses the real-time images acquired by the infrared thermal imaging device, including noise filtering, background correction, and temperature calibration. Then, the system employs a region-growing-based hotspot tracking algorithm to identify hotspots in the image with temperatures exceeding a set threshold. The region is marked and tracked. Between consecutive frames, image registration and motion estimation techniques are used to identify the movement trajectory and morphological changes of the temperature rise patches. Area growth rate. Defined as the area of ​​the temperature rise region in the current frame. Area compared to the previous frame The difference and time interval The ratio, that is:

[0045] Temperature gradient The system calculates the local temperature change rate by using the temperature difference between adjacent pixels in the image, employing the Sobel operator or the finite difference method. This rate is used to determine whether hotspot areas are experiencing rapid temperature increases. In some implementations, a gradient threshold can be set. When the local gradient exceeds this value, the fire hazard warning mechanism is triggered.

[0046] Specifically, the temperature resolution of the infrared thermal imaging equipment should be no less than [specific value missing]. Frame rate not lower than To ensure the accuracy of dynamic tracking. Area growth rate. Calculation cycle Typically set to Adjust according to actual operating conditions. Temperature gradient The calculation window size is or Pixels, to balance sensitivity and noise resistance.

[0047] S3 performs multidimensional fusion analysis based on extracted morphological and temperature features to generate event intensity scores and filter false alarms.

[0048] In some implementations, the system first extracts the contour features of the coal pile using a visible light PTZ camera, including the location of the coal pile, its height variation trend, and the area of ​​obstruction. These features are then quantified and extracted using image segmentation and edge detection algorithms (such as Canny edge detection or U-Net semantic segmentation). Simultaneously, an infrared thermal imaging device acquires temperature distribution maps of the equipment surface and the coal pile area, identifying the movement trajectories of hotspots and the continuous changes in temperature rise patches. Morphological and temperature features are processed separately using independent feature extraction modules, and then fused in an edge server.

[0049] The fusion analysis employs weighted feature concatenation and a multilayer perceptron (MLP) model for feature mapping and score calculation. Event intensity score. It can be represented as:

[0050] in, For morphological feature vectors, This is the temperature feature vector. This is a historical trend feature vector. The weighting coefficients for each feature dimension are determined through optimization using training data. The scoring result is used to determine whether an alarm is triggered, and is further adjusted by setting a threshold (e.g., ...). ) Perform false alarm filtering.

[0051] Specifically, the system sets the morphological feature extraction accuracy to pixel level, and the temperature resolution to be no less than [missing information]. The training dataset for the scoring model contains no fewer than 1,000 samples of coal blockage, coal overflow, and normal state in real-world scenarios. The model accuracy on the test set reaches over 95%, and the false alarm rate is controlled within 5%.

[0052] Furthermore, S3 includes: S31 calculates the event intensity score using a weighted fusion formula, filters false alarms based on the score threshold, and triggers a false alarm flag and records environmental interference characteristics when the score is below the threshold.

[0053] Specifically, this step uses a weighted fusion formula. Calculate event intensity scores. By introducing three key characteristic variables and assigning them preset weights, a quantitative assessment of potential coal blockage, coal spillage, and fire hazards in the coal conveying system is achieved.

[0054] A morphological variation coefficient representing the outline of a coal pile in a visible light image, used to describe the abnormal growth trend of the coal pile's volume or height; It represents the dynamic growth rate of the coal flow area, reflecting the accumulation speed of the coal flow in the coal chute or belt conveyor head area; This represents the gradient change in the temperature field during infrared thermal imaging, used to identify localized temperature rises or hotspot movement trends, thereby assisting in determining the presence of fire hazards. The three variables are extracted independently by the image recognition model and the thermal imaging analysis module, and then fused together on an edge computing server.

[0055] Furthermore, weighting coefficients Normalized parameters set during the system initialization phase based on historical data and expert experience typically satisfy... ,and In practical deployment, the weights can be dynamically adjusted based on the characteristics of different coal types, equipment layout, and historical false alarm rates to optimize identification sensitivity and false alarm suppression capabilities. For example, in operating conditions where coal is prone to sticking and ambient humidity is high, the weights can be appropriately increased. and The weights are adjusted to enhance the response to morphological changes and temperature anomalies.

[0056] In some implementations, based on a scoring threshold False alarm filtering is a key step in the intelligent identification and alarm method of this invention. Its technical implementation is based on the dynamic judgment of real-time analysis results of multimodal image data using an edge computing platform. The core principle of this step is to distinguish between genuine abnormal events and false alarms caused by environmental interference by setting reasonable scoring thresholds and comparing the "event intensity score" output by the identification model.

[0057] The specific operation method is as follows: After the system completes the fusion and recognition of visible light images and infrared thermal images, it will generate a comprehensive score. This score reflects the overall degree of anomaly in the current image, including coal pile morphology, temperature anomalies, and occlusion changes. A preset scoring threshold is used in the edge server. This threshold can be dynamically adjusted based on historical data, coal type characteristics, environmental temperature and humidity, and other factors. When a false alarm occurs, the system determines it to be a false alarm and triggers the false alarm marking mechanism. At the same time, it records the current environmental interference characteristics, such as changes in illumination, reflection interference, and equipment vibration, for subsequent model optimization and false alarm analysis.

[0058] Scoring threshold The value is typically set within the range of [0, 1]. The specific value should be optimized based on the false alarm and false negative rates during system debugging. An initial value of [value missing] is recommended. It also supports remote adjustments via a web management interface. Event Intensity Score The calculation is based on a multi-feature fusion model, and its output precision is floating point, retaining four decimal places to ensure recognition sensitivity.

[0059] S4 triggers an adaptive alarm strategy based on the event intensity score, and achieves multi-level linkage response through voice broadcast, switch signal output and text push.

[0060] Specifically, the system first evaluates the event intensity score in real time using a decision algorithm in the edge recognition server. This score is based on the coal pile morphology changes in visible light images and the temperature field distribution in infrared thermal imaging, employing a weighted fusion strategy. The calculation formula is as follows:

[0061] in, To score the overall intensity of the event, Scoring for visible light image recognition For infrared thermal imaging identification scoring, and These are preset weighting coefficients, typically ranging from [value range missing]. ,and In actual deployment, and It can be dynamically adjusted according to factors such as coal type characteristics and ambient temperature and humidity to optimize identification sensitivity and false alarm rate.

[0062] when ( The alarm threshold is typically set to [value]. to When this occurs (between points), the system will activate a multi-level response mechanism. First, the voice broadcast module will emit a pre-recorded alarm voice through the speaker, broadcasting three times consecutively, with each broadcast spaced apart from the previous one. This ensures that on-site personnel can detect anomalies promptly. Secondly, the system outputs via a USB-to-digital converter module. The switching signals are sent to the DCS system to achieve automatic shutdown or interlock control, with a response time of less than [time missing]. This meets the real-time requirements of industrial control. Finally, the system pushes alarm event types, timestamps, and identified images to the monitoring platform and mobile devices via industrial Ethernet, facilitating rapid problem location and response by remote personnel.

[0063] S5 automatically archives the identified images, scoring results, and response action data of alarm events, generating event logs containing timestamps and processing suggestions for traceability and analysis.

[0064] The system acquires visible light images and infrared thermal imaging data of the current abnormal state through the image acquisition unit, and outputs event scoring results (such as coal blockage degree, coal spill risk level, temperature anomaly index, etc.) by the image recognition model. Response action data includes alarm signal output status, voice broadcast trigger count, DCS / PLC linkage execution status, etc. All data is written to the local database in parallel after the event is triggered, using a unified event structure containing fields such as... (Event timestamp) (Visible light image data) (Infrared image data) (Scoring results) and (Response actions), etc.

[0065] In terms of parameters, the timestamp accuracy is at the millisecond level to ensure the time sequence accuracy of event recording; image data is compressed and stored in JPEG or PNG format with a resolution of no less than 1920×1080 and a frame rate controlled within 15fps to save storage space; the scoring results use a quantitative index of 0-100, combined with preset thresholds (such as a coal blockage threshold of 70 and a coal overflow threshold of 85) to classify events; the response action record includes the signal output status (ON / OFF), execution time and execution module identifier to ensure traceability.

[0066] Furthermore, S5 includes: S51 packages and stores the recognition model status snapshot, CPU utilization, camera angle data and alarm event data to form a multi-dimensional event data packet.

[0067] This step employs a multi-threaded data acquisition and asynchronous packaging mechanism to ensure that the system can quickly and efficiently integrate relevant data after an alarm event is triggered. The model state snapshot includes the current model's inference parameters, activation layer states, input / output feature maps, etc., used to record the model's running state at the time of the event, facilitating analysis of false alarms or missed alarms. CPU utilization is collected in real time through system monitoring interfaces (such as Linux's ` / proc / stat` or Windows performance counters), reflecting the load on the edge server during event processing and providing a basis for system resource scheduling and model optimization. Camera angle data is fed back from the PTZ (Pan-Tilt-Zoom) control interface of the image acquisition unit, including the horizontal rotation angle (…). Vertical pitch angle ( ) and zoom ratio ( This is used to reconstruct the image acquisition perspective and assist in determining the impact of image occlusion or perspective deviation on the recognition results. Alarm event data includes event type (e.g., coal blockage, coal spill, fire hazard), timestamp (...). ), confidence score ( ) and description of the trigger source (such as continuous coal supply, hot spot movement, etc.).

[0068] Specifically, the packaging operation must meet the following standards: the data packet format adopts a common binary or JSON structure, supporting efficient storage and retrieval of database systems (such as MySQL and PostgreSQL); the timestamp accuracy is not less than milliseconds; the model snapshot size is controlled within 10MB to ensure storage efficiency; the camera angle data accuracy is 0.1° to ensure the accuracy of image reconstruction; and the CPU utilization is recorded as a percentage to evaluate the system's operating status.

[0069] The S52 displays the event log retrieval function through a web interface, supporting multi-dimensional queries by timestamp, device number, event type, and processing suggestions.

[0070] Specifically, this function is designed based on a B / S (Browser / Server) architecture. The front end uses HTML5, CSS3, and JavaScript to build a responsive user interface, while the back end interacts with the database through a RESTful API, supporting multi-condition combined queries and paginated display of results. Users can access the system's web console through a browser, enter any combination of query conditions, such as time range, equipment number, event type (e.g., category tags like "coal blockage," "coal overflow," "abnormal temperature," etc.), and handling suggestions (e.g., preset texts like "shutdown and inspection" and "clean coal pile"). The system will then retrieve matching event logs from the local database based on these conditions.

[0071] At the parameter level, the event log data structure includes timestamp, device ID, event type, identified image path, and processing suggestion fields. The query interface supports a maximum of 100 concurrent requests with a response latency controlled within 500ms, meeting real-time requirements. Simultaneously, the system supports both fuzzy and exact matching modes. For example, in the processing suggestion field, users can enter the keyword "clean up" for a fuzzy search, and the system will return all records containing that word.

[0072] This invention provides an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition. This method can effectively reduce the false alarm rate and missed alarm rate of coal blockage, overflow and fire hazards, improve the accuracy and response speed of anomaly identification, realize intelligent, real-time and closed-loop management of the coal conveying system, and significantly enhance the system's operational safety and maintenance efficiency.

[0073] Example 2 The following is a detailed description of an intelligent alarm method for coal blockage and overflow in a coal conveying system based on image recognition, according to an embodiment of the present invention.

[0074] The intelligent identification and alarm system for coal conveying systems described in this invention adopts a modular design architecture, integrating functions such as image acquisition, edge computing, anomaly identification, and linkage response. Specifically, it includes the following components: Multi-source image acquisition unit: such as Figure 2 and Figure 3 As shown, this system installs high-definition visible light PTZ cameras and infrared thermal imaging equipment at key coal conveying points, forming a multimodal image acquisition system. The cameras support variable zoom and intelligent tracking functions, and the infrared equipment has dynamic temperature spectrum modeling capabilities, significantly improving data dimensionality and recognition accuracy.

[0075] Edge intelligent recognition unit: The core recognition server of the system is deployed in the local electronics room. It adopts an industrial-grade graphics computing platform and runs the image recognition model through a multi-process asynchronous inference engine. It supports local parsing of image data and real-time judgment of events, avoiding dependence on the cloud and improving the system's response speed and stability.

[0076] Integrated alarm response module: The identification server determines the alarm level through adaptive strategies and activates multi-level responses, including on-site voice broadcast, local switch signal output, and alarm text content push. It can be flexibly configured to interface with main control systems such as DCS and PLC to achieve automatic shutdown or interlocking.

[0077] Event data archiving and tracing module: Each alarm triggers automatic screenshot and event log storage. Archived information includes timestamp, identified image, preliminary type assessment, and processing suggestions. The data structure is compatible with general database systems and supports multi-dimensional retrieval and trend analysis.

[0078] Non-intrusive integrated communication system: The device adopts an external communication design, which does not require disruption of the original coal conveying control system structure, making installation and deployment convenient. It connects to the existing monitoring network via industrial Ethernet, enabling interconnection between identification data and the monitoring platform.

[0079] The operation process of this invention includes four major steps: system initialization, intelligent identification, proactive response, and event closure. Unlike traditional passive monitoring methods, it achieves intelligent judgment and linkage throughout the entire process.

[0080] System initialization self-verification: After system startup, it automatically completes a self-check of multi-module interconnection, including image channel validity, temperature detection interface, communication status, and alarm links. The server summarizes and displays the self-check results of historical false alarm rate and recognition model accuracy, facilitating quick confirmation by on-site personnel and improving system deployment efficiency and robustness.

[0081] Multimodal fusion recognition: Unlike existing single-image recognition methods, this invention adopts a dual-channel fusion approach of "visible light image + infrared thermal imaging" and uses a deep recognition model to make multidimensional judgments on the morphology and temperature field distribution of the coal pile, achieving the following innovative operation process: real-time extraction of image contour features to determine the location, height trend, and occlusion changes of the coal pile; analysis of hotspot movement and temperature rise patch trends in the infrared spectrum, dynamic comparison with historical data; the fusion recognition results are used to generate an "event intensity score" through an edge decision algorithm to intelligently filter false alarms; all recognition tasks run in parallel processes, improving response speed and accuracy under complex working conditions.

[0082] Proactive linkage response mechanism: When the scoring result exceeds the set threshold, the system not only activates voice alarm and signal output, but also uses the "trigger source analysis" module to determine the cause of the abnormality, such as structural obstruction, continuous coal feeding, image abnormality, etc., and adds a description to the alarm information to prompt the operators to respond quickly.

[0083] Closed-loop event management and self-learning optimization: In addition to images and logs, each alarm event record also includes a snapshot of the identified model status, CPU utilization, and camera angle data, which are then archived for later system self-learning optimization and operational traceability analysis. Users can remotely view the complete event chain through a web interface, improving closed-loop management capabilities.

[0084] Example 3 To achieve the above embodiments, such as Figure 4 As shown, this embodiment also provides an intelligent alarm device 10 for coal blockage and overflow in a coal conveying system based on image recognition. The device 10 includes a multimodal image acquisition module 100, an asynchronous feature extraction module 200, a multidimensional fusion analysis module 300, an alarm strategy triggering module 400, and an event log generation module 500.

[0085] The multimodal image acquisition module 100 is used to deploy a visible light spherical camera and an infrared thermal imaging device at key locations in the coal conveying system to simultaneously acquire images of the coal pile morphology and temperature field distribution. The asynchronous feature extraction module 200 is used to run the edge computing server to perform asynchronous parallel processing on multi-source images, and extract the contour features of visible light images and the dynamic hotspot features of infrared images respectively. The multidimensional fusion analysis module 300 is used to perform multidimensional fusion analysis based on extracted morphological and temperature features, generate event intensity scores, and filter false alarms. The alarm strategy triggering module 400 is used to trigger an adaptive alarm strategy based on the event intensity score, and achieve multi-level linkage response through voice broadcast, switch signal output and text push. The event log generation module 500 is used to automatically archive the identification images, scoring results and response action data of alarm events, and generate event logs containing timestamps and processing suggestions for traceability and analysis.

[0086] Furthermore, the aforementioned multimodal image acquisition module 100 is also used for: The visible light spherical camera is installed 1.5-2.0 meters above the coal drop pipe at the head of the conveyor belt, and the infrared thermal imaging equipment is installed in the contact area between the side wall of the coal drop pipe and the equipment shell. The visible light camera is configured with a focal length adjustment range of 20-120mm, the infrared device has a temperature detection accuracy of ±0.5℃, and the sampling frequency of both devices is synchronized to 10Hz.

[0087] Furthermore, the asynchronous feature extraction module 200 described above is also used for: A multi-process asynchronous inference engine was used to perform deep learning-based contour segmentation on visible light images to extract coal pile height, volume change rate, and occlusion area ratio. A hotspot tracking algorithm is applied to infrared images to calculate the area growth rate and temperature gradient of temperature-rising patches, and a dynamic comparison model with historical temperature distribution is established.

[0088] Furthermore, the aforementioned multidimensional fusion analysis module 300 is also used for: The event intensity score is calculated using a weighted fusion formula, based on a score threshold. False alarm filtering is performed. When the score is lower than the threshold, a false alarm flag is triggered and environmental interference characteristics are recorded.

[0089] Furthermore, the aforementioned event log generation module 500 is also used for: The recognition model status snapshot, CPU utilization, camera angle data and alarm event data are packaged and stored to form a multi-dimensional event data package; The event log retrieval function is displayed through a web interface, supporting multi-dimensional queries by timestamp, device number, event type, and processing suggestions.

[0090] An intelligent alarm device for coal blockage and overflow in a coal conveying system based on image recognition, according to an embodiment of the present invention, can effectively reduce the false alarm rate and missed alarm rate of coal blockage, overflow and fire hazards, improve the accuracy and response speed of anomaly identification, realize intelligent, real-time and closed-loop management of the coal conveying system, and significantly enhance the system's operational safety and maintenance efficiency.

[0091] In the description of this specification, the references to terms such as "one embodiment," "some embodiments," "example," "specific example," or "some examples," etc., refer to specific features, structures, materials, or characteristics described in connection with that embodiment or example, which are included in at least one embodiment or example of the present invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples. Moreover, without contradiction, those skilled in the art can combine and integrate the different embodiments or examples described in this specification, as well as the features of different embodiments or examples.

[0092] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include at least one of that feature. In the description of this invention, "a plurality of" means at least two, such as two, three, etc., unless otherwise explicitly specified.

Claims

1. A method for intelligent alarm of coal blockage and overflow in a coal conveying system based on image recognition, characterized in that, include: S1, deploy visible light spherical cameras and infrared thermal imaging equipment at key locations in the coal conveying system to simultaneously acquire images of the coal pile morphology and temperature field distribution; S2, the edge computing server performs asynchronous parallel processing on multi-source images, extracting the contour features of visible light images and the dynamic hotspot features of infrared images respectively; S3, based on the extracted morphological and temperature features, performs multidimensional fusion analysis to generate event intensity scores and filter false alarms; S4 triggers an adaptive alarm strategy based on the event intensity score, and achieves multi-level linkage response through voice broadcast, switch signal output and text push; S5 automatically archives the identified images, scoring results, and response action data of alarm events, generating event logs containing timestamps and processing suggestions for traceability and analysis.

2. The method as described in claim 1, characterized in that, S1 includes: S11, a visible light spherical camera is installed 1.5-2.0 meters above the coal drop pipe at the head of the conveyor belt, and an infrared thermal imaging device is installed in the contact area between the side wall of the coal drop pipe and the equipment shell; The S12 is equipped with a visible light camera with a focal length adjustment range of 20-120mm, an infrared device with a temperature detection accuracy of ±0.5℃, and both devices have a sampling frequency synchronized to 10Hz.

3. The method as described in claim 1, characterized in that, The S2 includes: S21 uses a multi-process asynchronous inference engine to perform deep learning-based contour segmentation on visible light images, extracting coal pile height, volume change rate, and occlusion area ratio; S22 executes a hotspot tracking algorithm on infrared images to calculate the area growth rate and temperature gradient of temperature-rising patches, and establishes a dynamic comparison model with historical temperature distribution.

4. The method as described in claim 1, characterized in that, The S3 further includes: S31 calculates the event intensity score using a weighted fusion formula, filters false alarms based on the score threshold, and triggers a false alarm flag and records environmental interference characteristics when the score is below the threshold.

5. The method as described in claim 1, characterized in that, The S5 includes: S51 packages and stores the recognition model status snapshot, CPU utilization, camera angle data and alarm event data to form a multi-dimensional event data packet. The S52 displays the event log retrieval function through a web interface, supporting multi-dimensional queries by timestamp, device number, event type, and processing suggestions.

6. A smart alarm device for coal blockage and overflow in a coal conveying system based on image recognition, characterized in that, include: The multimodal image acquisition module is used to deploy visible light spherical cameras and infrared thermal imaging equipment at key locations in the coal conveying system to simultaneously acquire images of the coal pile morphology and temperature field distribution. The asynchronous feature extraction module is used to run the edge computing server to perform asynchronous parallel processing on multi-source images, extracting the contour features of visible light images and the dynamic hotspot features of infrared images respectively. The multidimensional fusion analysis module is used to perform multidimensional fusion analysis based on extracted morphological and temperature features, generate event intensity scores, and filter false alarms. The alarm strategy triggering module is used to trigger an adaptive alarm strategy based on the event intensity score, and achieve multi-level linkage response through voice broadcast, switch signal output and text push. The event log generation module is used to automatically archive the identification images, scoring results, and response action data of alarm events, and generate event logs containing timestamps and processing suggestions for traceability and analysis.

7. The apparatus as claimed in claim 6, characterized in that, The multimodal image acquisition module is also used for: The visible light spherical camera is installed 1.5-2.0 meters above the coal drop pipe at the head of the conveyor belt, and the infrared thermal imaging equipment is installed in the contact area between the side wall of the coal drop pipe and the equipment shell. The visible light camera is configured with a focal length adjustment range of 20-120mm, the infrared device has a temperature detection accuracy of ±0.5℃, and the sampling frequency of both devices is synchronized to 10Hz.

8. The apparatus as claimed in claim 6, characterized in that, The asynchronous feature extraction module is also used for: A multi-process asynchronous inference engine was used to perform deep learning-based contour segmentation on visible light images to extract coal pile height, volume change rate, and occlusion area ratio. A hotspot tracking algorithm is applied to infrared images to calculate the area growth rate and temperature gradient of temperature-rising patches, and a dynamic comparison model with historical temperature distribution is established.

9. The apparatus as claimed in claim 6, characterized in that, The multidimensional fusion analysis module is also used for: The event intensity score is calculated using a weighted fusion formula, based on a score threshold. False alarm filtering is performed. When the score is lower than the threshold, a false alarm flag is triggered and environmental interference characteristics are recorded.

10. The apparatus as claimed in claim 6, characterized in that, The event log generation module is also used for: The recognition model status snapshot, CPU utilization, camera angle data and alarm event data are packaged and stored to form a multi-dimensional event data package; The event log retrieval function is displayed through a web interface, supporting multi-dimensional queries by timestamp, device number, event type, and processing suggestions.

Images