Method and system for detecting slag block accumulation in slag well of slag drying machine of slag conveying system and medium

Abstract

The invention provides a method and system for detecting slag block accumulation in a slag well of a slag drying machine of a slag conveying system and a medium. A slag well video is shot through a camera module arranged in a slag well, a slag well image is obtained from the slag well video, the slag well image is sequentially input into a slag block detection model of a target detection module and a slag block detection model of a semantic segmentation module, information of slag blocks in the slag well image is obtained, the target detection module is a model established based on YOLOv7, and the target detection module is a model established based on YOLOv7. The target detection module is used for recognizing a trained model of position information of slag blocks in an input slag well image, the semantic segmentation module is a model established based on SAM, and the semantic segmentation module is used for obtaining information of the slag blocks in the input slag well image according to the position information of the slag blocks in the input slag well image and sending the information of the slag blocks to the target detection module. And according to the information of the slag blocks in the slag well image, the area threshold value and the height threshold value, determining that accumulated slag exists in the current slag well. Various slag blocks can be rapidly detected, the real-time requirement for identifying accumulated slag in the slag well is met, and stable operation of the system is guaranteed.

Description

Technical Field

[0001] The present disclosure relates to the field of image processing technology, and in particular to a method, system and medium for detecting slag block accumulation in a slag pit of a dry slag machine in a slag conveying system. Background Art

[0002] At present, the operation and management of the slag conveying system adopts a combination of indoor video surveillance of important areas, slag conveying system host computer monitoring screen and on-site inspection to ensure the safe and stable operation of the slag conveying system. Among them, whether there is slag accumulation in the slag pit of the steel belt machine of the slag conveying system also needs to be detected by manual video viewing.

[0003] However, the above method places extremely high demands on the operation and maintenance personnel. Once their attention is reduced, they are likely to miss the accumulation of slag blocks in the slag shaft of the steel belt conveyor of the slag conveying system, which poses a hidden danger to the stable operation of the system. Summary of the invention

[0004] The present invention provides a method, system and medium for detecting slag block accumulation in a slag well of a dry slag machine of a slag conveying system, so as to solve the problem that the slag block accumulation in the slag well of a steel belt machine of the slag conveying system cannot be handled in time, resulting in hidden dangers in the stable operation of the slag conveying system.

[0005] In a first aspect, the present disclosure provides a method for detecting slag block accumulation in a slag pit of a dry slag machine of a slag conveying system, the method comprising:

[0006] Shoot the slag well video through the camera module;

[0007] Acquire a slag well image from the slag well video;

[0008] The slag pit image is input into the slag block detection model to obtain the information of the slag blocks in the slag pit image; the slag block detection model includes a target detection module and a semantic segmentation module connected in sequence; the target detection module is a model established based on the target detection YOLOv7; the target detection module is a trained model for identifying the position information of the slag blocks in the input slag pit image; the semantic segmentation module is a model established based on the Segment Everything Model SAM; the semantic segmentation module is used to obtain the information of the slag blocks in the input slag pit image according to the position information of the slag blocks in the input slag pit image; the information of the slag blocks in the input slag pit image is used to indicate the outer edge of the slag blocks in the slag pit image;

[0009] According to the information of the slag blocks in the slag well image, the area threshold and the height threshold, it is determined that there is slag accumulation in the slag well.

[0010] In some embodiments, the target detection module includes a feature extraction submodule Backbone, an intermediate submodule Neck and a prediction submodule Head; the feature extraction submodule is established based on spatial pyramid pooling SPP; the intermediate submodule is established based on feature pyramid network FPN; the prediction submodule is established based on non-maximum suppression algorithm DIOU-NMS.

[0011] In some embodiments, determining that there is slag accumulation in the current slag well according to the information of the slag block in the slag well image, the area threshold and the height threshold includes:

[0012] Determine the total pixel area and the maximum height of the slag blocks according to the information of the slag blocks in the slag well image;

[0013] If the total pixel area of ​​the slag block is greater than the area threshold, and the maximum height of the slag block is greater than the height threshold, it is determined that there is slag accumulation in the slag pit.

[0014] In some embodiments, the camera module includes a first camera and a second camera; the first camera and the second camera have different shooting angles; and the step of shooting a slag well video by the camera module includes:

[0015] Shooting the first slag well video by the first camera;

[0016] Shooting a video of the second slag well by using a second camera;

[0017] Acquiring a slag well image from the slag well video includes:

[0018] Acquire a first slag well image from the first slag well video, and acquire a second slag well image from the second slag well video, wherein the first slag well image and the second slag well image are taken at the same time;

[0019] Inputting the first slag well image and the second slag well image into the slag block detection model respectively, to obtain information about slag blocks in the first slag well image and information about slag blocks in the second slag well image;

[0020] The determining, based on the information of the slag blocks in the slag well image, the area threshold and the height threshold, that there is slag accumulation in the current slag well comprises:

[0021] According to the information of the slag blocks in the first slag well image, the information of the slag blocks in the second slag well image, the area threshold and the height threshold, it is determined that there is slag accumulation in the current slag well.

[0022] In some embodiments, after determining that there is slag accumulation in the slag well according to the information of the slag block in the slag well image, the area threshold and the height threshold, the method further includes:

[0023] Send an alarm message to the power plant control system, wherein the alarm message is used to indicate that there is slag accumulation in the slag pit.

[0024] In some embodiments, the method further comprises:

[0025] Acquire multiple sample slag well images;

[0026] Performing oversampling processing on the multiple sample slag well images, and / or synthesizing a part or all of the multiple sample slag well images to obtain a plurality of processed slag well images;

[0027] The multiple processed slag pit images are input into an initial target detection module for training until the initial target detection module converges to obtain a target detection module.

[0028] In some embodiments, the inputting the plurality of processed slag pit images into an initial target detection module for training until the initial target detection module converges to obtain a target detection module includes:

[0029] Inputting the plurality of processed slag pit images into an initial target detection module to obtain position information of slag blocks in the plurality of processed slag pit images;

[0030] According to the position information of the slag blocks in the processed slag well image, a loss function is calculated until the loss function is less than a loss threshold, thereby obtaining a target detection module.

[0031] In a second aspect, the present disclosure provides a detection system for slag block accumulation in the slag pit of a dry slag machine in a slag conveying system, comprising: a camera module and a server; the camera module is communicatively connected to the server; the camera module is used to capture images in the slag pit; the server is used to implement the method described in the first aspect above.

[0032] In a third aspect, the present disclosure provides a computer-readable storage medium having a computer program stored thereon, which, when executed by a processor, implements the method described in the first aspect.

[0033] In a fourth aspect, the present disclosure provides a computer program product, including a computer program, which implements the steps of the method described in the first aspect when executed by a processor.

[0034] The embodiments of the present disclosure provide a method, system and medium for detecting slag block accumulation in a slag pit of a dry slag machine of a slag conveying system.

[0035] The slag well video is shot by a camera module set in the slag well, and the slag well image is obtained from the slag well video. The slag well image is input into the slag block detection model to obtain the information of the slag blocks in the slag well image, wherein the slag block detection model includes a target detection module and a semantic segmentation module connected in sequence. The target detection module is a model established based on YOLOv7. The target detection module is a trained model for identifying the position information of the slag blocks in the input slag well image. The semantic segmentation module is a model established based on SAM. The semantic segmentation module is used to obtain the information of the slag blocks in the input slag well image according to the position information of the slag blocks in the input slag well image. The information of the slag blocks in the input slag well image is used to indicate the outer edge of the slag blocks in the slag well image. According to the information of the slag blocks in the slag well image, the area threshold and the height threshold, it is determined that there is slag accumulation in the current slag well. By identifying the position information of the slag blocks in the slag well image by the target detection module, various slag blocks in the slag well image can be quickly detected, with high processing accuracy and fast detection speed, which meets the real-time requirements in the detection process of slag block accumulation in the slag well of the dry slag machine of the slag conveying system, and improves the detection efficiency. The slag blocks are segmented from the slag pit image through the semantic segmentation module, and the processing speed is fast. The semantic segmentation model does not need to be trained, which minimizes the need for large-scale retraining or data annotation. The model preparation stage is easy to use, thereby improving the detection efficiency. The method of this embodiment has a fast processing speed and high processing accuracy, meets the real-time requirements of slag accumulation identification in the slag pit, and can detect the accumulation of slag blocks in the slag pit in a timely manner to ensure the stable operation of the system. BRIEF DESCRIPTION OF THE DRAWINGS

[0036] Figure 1 A schematic flow chart of a method for detecting slag block accumulation in a slag pit of a dry slag machine of a slag conveying system provided in an embodiment of the present disclosure;

[0037] Figure 2 A schematic diagram of a flow chart of a method for training a target detection module in a slag detection model provided in an embodiment of the present disclosure;

[0038] Figure 3 A schematic structural diagram of a system for detecting slag block accumulation in a slag pit of a dry slag machine of a slag conveying system provided in an embodiment of the present disclosure. DETAILED DESCRIPTION

[0039] Exemplary embodiments will be described in detail herein, examples of which are shown in the accompanying drawings. When the following description refers to the drawings, the same numbers in different drawings represent the same or similar elements unless otherwise indicated. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with the present disclosure. Instead, they are merely examples of devices and methods consistent with some aspects of the present disclosure as detailed in the appended claims.

[0040] In the present disclosure, "at least one" refers to one or more, and "plurality" refers to two or more. "And / or" describes the association relationship of associated objects, indicating that three relationships may exist. For example, A and / or B can represent: A exists alone, A and B exist at the same time, and B exists alone, where A and B can be singular or plural. The character " / " generally indicates that the objects associated before and after are in an "or" relationship. "At least one of the following" or similar expressions refers to any combination of these items, including any combination of single items or plural items. For example, at least one of a alone, b alone, or c alone can represent: a alone, b alone, c alone, a and b in combination, a and c in combination, b and c in combination, or a, b, and c in combination, where a, b, and c can be single or multiple. In addition, the terms "first" and "second" are used for descriptive purposes only and cannot be understood as indicating or implying relative importance.

[0041] The directions or positional relationships indicated by terms such as "center", "longitudinal", "lateral", "up", "down", "left", "right", "front", and "back" are based on the directions or positional relationships shown in the accompanying drawings and are only for the convenience of describing the present disclosure and simplifying the description. They do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and therefore should not be understood as a limitation on the present disclosure.

[0042] The terms "connected" and "connected" should be understood in a broad sense. For example, the "connected" or "connected" of a circuit structure may refer to not only a physical connection, but also an electrical connection or a signal connection. For example, it may be a direct connection, i.e. a physical connection, or it may be an indirect connection through at least one intermediate element, as long as the circuit is connected. It may also be the internal connection of two elements. In addition to signal connection through a circuit, signal connection may also refer to signal connection through a media medium, such as radio waves. For those of ordinary skill in the art, the specific meanings of the above terms in this disclosure may be understood according to specific circumstances.

[0043] The method for detecting slag block accumulation in the slag pit of a dry slag machine of a slag conveying system provided by the present disclosure can be applied to a detection system for slag block accumulation in the slag pit of a dry slag machine of a slag conveying system. The detection system for slag block accumulation in the slag pit of a dry slag machine of a slag conveying system may include a camera module and a server. The camera module is connected to the server for communication. Exemplarily, the camera module and the server may be connected via a transmission signal cable, for example, the transmission signal cable may be an optical cable.

[0044] The camera module is used to shoot images in the slag pit.

[0045] Exemplarily, the camera module may be located in a slag pit. Since the temperature in the slag pit is relatively high, the camera module has the property of being resistant to high temperatures. Usually, the operating temperature of the camera module is higher than the maximum temperature of the slag pit.

[0046] Furthermore, the server may include a camera module or may not include a camera module. The camera module may be a camera or a webcam.

[0047] Furthermore, the detection system for slag block accumulation in the slag pit of the dry slag machine of the slag conveying system may also include but is not limited to: an on-site sub-control box, a high-temperature cooling protection device, a cooling system, an air purification device, etc. Among them, the high-temperature cooling protection device, the cooling system, the air purification device, etc. are used to cool the camera module so that the temperature of the camera module is within the operating temperature range.

[0048] The server may be one or more servers. The server may include an image processing and analysis program, a graphic storage medium, and an image analysis and processing alarm program. For example, the server may include an image processing and analysis server, a graphic storage server, and an image analysis and processing alarm server.

[0049] In actual applications, the camera module captures the video of the slag pit. The slag block accumulation height, distribution area and other information can be identified in the slag pit video. The camera module transmits the slag pit video to the server, i.e., the video monitoring end, through the optical cable. The large slag blocks in the slag pit are detected in real time through image processing, analysis and recognition technology. The slag accumulation alarm information can be obtained by comparing the proportion of the height of the accumulated slag blocks to the entire slag pit. The server can control the sound and light alarm through the industrial Ethernet serial communication protocol based on the transmission control protocol (ModbusTCP) transmission protocol to remind on-site operators to control the extrusion head in a linked manner to extrude and crush the large pieces of slag.

[0050] In addition, the camera module may include multiple cameras. Since there may be multiple slag pits in the slag conveying system, a camera is set in each slag pit. Furthermore, in order to better detect the accumulation of slag blocks in the slag pit, cameras are set in front and behind the slag pit. The server can process the slag pit images taken by multiple cameras at the same time, realize parallel video access, and improve the detection efficiency.

[0051] The technical solutions and beneficial effects provided by the present disclosure are introduced below with specific embodiments.

[0052] See also Figure 1 , Figure 1 The present invention provides a flow chart of a method for detecting slag block accumulation in a slag pit of a dry slag machine in a slag conveying system. Figure 1As shown, the method provided in this embodiment is executed by a server, and the server may be the server in the above embodiment. The method provided in this embodiment may include the following steps 101 to 104.

[0053] Step 101, shooting a slag pit video by using a camera module.

[0054] Step 102: Acquire a slag well image from the slag well video.

[0055] The slag well video taken by the camera module includes multiple frames. The slag well image can be obtained from the slag well video at a certain sampling period for subsequent processing; the slag well video usually has a key frame, which contains the complete image information of the frame. Therefore, the key frame can also be obtained from the slag well video as the slag well image for subsequent processing.

[0056] Step 103: input the slag well image into the slag block detection model to obtain the information of the slag blocks in the slag well image.

[0057] Among them, the slag block detection model includes a target detection module and a semantic segmentation module connected in sequence.

[0058] The above target detection module is a trained model based on target detection (YOLOv7). The target detection module is a trained model for identifying the position information of slag blocks in the input slag pit image.

[0059] Among them, YOLOv7 improves the network structure based on YOLOv5, making the network more efficient, with target detection algorithm (anchor-based) as the main prediction form, mainly in YOLOv7. The improvements are in the E-ELAN network, convolution re-parameterization (Planned re-parameterized) and label allocation method.

[0060] Among them, the ELAN network is an improved network based on the cross-stage local (CSP) network. The network structure has more branches and fewer convolution kernels. The specific structure of the ELAN network is the same as that of the CSP network, using the gradient truncation method. However, the ELAN network eliminates the 1x1 convolution channel stacking (concat) operation in the network, converts it into a 3x3 convolution direct output, and then performs the concat operation at the end. Intuitively, it retains the residual connection, but the number of convolutions is reduced. This results in fewer and longer gradient paths, making it easier to converge. This improves the structural efficiency of the model and ensures accurate detection capabilities without affecting the processing of details.

[0061] Due to the high processing accuracy and fast detection speed of YOLOv7, it meets the real-time requirements of the detection process of slag block accumulation in the slag pit of the dry slag machine of the slag conveying system, and improves the detection efficiency.

[0062] The semantic segmentation module is a model established based on the segmentation everything model SAM. The semantic segmentation module is used to obtain a model of information of slag blocks in the input slag well image according to the position information of slag blocks in the input slag well image. The information of slag blocks in the input slag well image is used to indicate the outer edge of the slag blocks in the slag well image.

[0063] Among them, the Segment-Anything-Model (SAM) drives computer vision forward by simplifying image segmentation. SAM uses the largest Segment Anything 1-Billion (SA-1B) mask dataset to date to democratize segmentation by reducing dependence on expertise, powerful computing power, and extensive dataset annotation.

[0064] SAM introduces a base model framework that allows tasks to be easily adjusted with simple prompts, mirroring advances in natural language processing. Trained on over a billion different masks, SAM understands the generalized concept of objects, facilitates zero-shot transfer across unfamiliar domains, and enhances its usefulness in a variety of fields including AR / VR, creative arts, and scientific research.

[0065] Since SAM has a fast processing speed and is a model that has been trained using a large amount of data, when it is used in the scenario of detecting slag block accumulation in the slag pit of the dry slag machine in the slag conveying system, SAM does not need to be trained. The model preparation stage is easy to use, thereby improving detection efficiency.

[0066] Step 104: Determine whether there is accumulated slag in the current slag pit according to the information of the slag block in the slag pit image, the area threshold and the height threshold.

[0067] Among them, slag accumulation refers to the accumulation of slag blocks, which may affect the stable operation of the system.

[0068] According to the information of the slag blocks in the slag well image, the ratio of the area of ​​the slag blocks in the slag well image to the area of ​​the slag well image can be determined, that is, the ratio of the number of slag block pixels to the number of pixels in the slag well image, for example, the ratio can be 45%. The ratio is compared with the area threshold to determine whether there is slag accumulation in the current slag well.

[0069] According to the information of the slag block in the slag well image, the ratio of the height of the slag block in the slag well image to the height of the slag well image can be determined, that is, the ratio of the height of the highest pixel point of the slag block to the height of the slag well image, for example, the ratio can be 50%. The ratio is compared with the height threshold to determine whether there is slag accumulation in the current slag well.

[0070] In this embodiment, a slag well video is captured by a camera module arranged in the slag well, and a slag well image is obtained from the slag well video, and the slag well image is input into the slag block detection model to obtain the information of the slag blocks in the slag well image, wherein the slag block detection model includes a target detection module and a semantic segmentation module connected in sequence, the target detection module is a model established based on YOLOv7, and the target detection module is a trained model for identifying the position information of the slag blocks in the input slag well image. The semantic segmentation module is a model established based on SAM, and the semantic segmentation module is used to obtain the information of the slag blocks in the input slag well image according to the position information of the slag blocks in the input slag well image, and the information of the slag blocks in the input slag well image is used to indicate the outer edge of the slag blocks in the slag well image, and the existence of slag accumulation in the current slag well is determined according to the information of the slag blocks in the slag well image, the area threshold and the height threshold. By identifying the position information of the slag blocks in the slag well image by the target detection module, various slag blocks in the slag well image can be quickly detected, the processing accuracy is high, the detection speed is fast, and the real-time requirements in the detection process of slag block accumulation in the slag well of the dry slag machine of the slag conveying system are met, and the detection efficiency is improved. The slag blocks are segmented from the slag pit image through the semantic segmentation module, and the processing speed is fast. The semantic segmentation model does not need to be trained, which minimizes the need for large-scale retraining or data annotation. The model preparation stage is easy to use, thereby improving the detection efficiency. The method of this embodiment has a fast processing speed and high processing accuracy, meets the real-time requirements of slag accumulation identification in the slag pit, and can detect the accumulation of slag blocks in the slag pit in a timely manner to ensure the stable operation of the system.

[0071] In some embodiments, the target detection module in step 103 may include: a feature extraction submodule (Backbone), an intermediate submodule (Neck), and a prediction submodule (Head).

[0072] Furthermore, the feature extraction submodule is established based on Spatial Pyramid Pooling (SPP).

[0073] For a detector, Backbone has three factors that affect performance, namely network depth, downsample rate, and receptive field. The deeper the network, the better the result; the more downsampling times, the more negative impact on small objects; the detection performance at different scales is positively correlated with the dilation rate, also known as the number of holes. Since the slag blocks in the slag well image of the dry slag machine of the slag conveying system are detected, the slag blocks can be regarded as small objects. Therefore, the back part of Backbone in this embodiment can use spatial pyramid pooling to fuse features of different scales and enhance the small target features of slag blocks.

[0074] Furthermore, the intermediate submodule is built based on the feature pyramid network FPN.

[0075] Feature Pyramid Networks (FPN) is a network that uses a single high-level feature (for example, Faster R-CNN uses a convolutional layer that is downsampled four times, Conv4, to perform subsequent object classification and bounding box regression), resulting in small objects having less pixel information, which is easily lost in the downsampling process. Since the slag blocks in the slag well image of the slag well of the slag conveying system are detected, the sizes of the slag blocks vary. In order to handle the detection problem of such objects with obvious size differences, the image pyramid method can be used to perform multi-scale change enhancement, but this will bring a huge amount of calculation. In this embodiment, the network structure of the feature pyramid provided by FPN is used to establish an intermediate submodule based on FPN, so that the multi-scale change problem in object detection can be handled with minimal increase in the amount of calculation.

[0076] Furthermore, the prediction submodule is established based on the non-maximum suppression algorithm DIOU-NMS.

[0077] Since DIOU_nms can distinguish two occluded small objects according to the center distance and detect them both, in the scenario of detecting slag block accumulation in the slag well of the dry slag machine of the slag conveying system disclosed in the present invention, since the slag blocks are usually small, the use of DIOU_nms can accurately identify the slag blocks in the slag well image, thereby improving the detection efficiency of slag block accumulation in the slag well of the dry slag machine of the slag conveying system.

[0078] In some embodiments, step 104 may be implemented by steps 1041 and 1042 as follows.

[0079] Step 1041: Determine the total pixel area and the maximum height of the slag blocks according to the information of the slag blocks in the slag well image.

[0080] Step 1042: Determine whether the total pixel area of ​​the slag block is greater than the area threshold, and whether the maximum height of the slag block is greater than the height threshold.

[0081] If yes, continue to execute step 1043; if no, return to execute step 102.

[0082] Step 1043: Determine whether there is slag accumulation in the current slag pit.

[0083] In this embodiment, when the total pixel area of ​​the slag block is greater than the area threshold and the maximum height of the slag block is greater than the height threshold, it can be determined that there is slag accumulation in the current slag pit.

[0084] In some embodiments, step 104 may be followed by step 105 .

[0085] Step 105: Send an alarm message to the power plant control system.

[0086] Among them, the alarm information is used to indicate that there is slag accumulation in the current slag pit.

[0087] After receiving the alarm information, the power plant control system can give an alarm in the form of sound and / or light.

[0088] In this embodiment, when slag accumulation is detected in the slag pit, an alarm message can be sent to the power plant control system, so that the power plant control system can alarm through sound and light alarms, and remind on-site operators to control the extrusion head to extrude and crush large pieces of slag, thereby quickly processing the slag blocks and ensuring stable operation of the system.

[0089] In some embodiments, the camera module includes a first camera and a second camera; the first camera and the second camera have different shooting angles. Step 101 may be implemented by steps 1011 and 1012. Step 102 may be implemented by steps 1021 and 1022. Step 103 may be implemented by step 1031. Step 104 may be implemented by step 1041.

[0090] Step 1011, capture a first slag well video using a first camera.

[0091] Step 1012: Shoot a second slag pit video using a second camera.

[0092] Step 1021: Acquire a first slag well image from the first slag well video.

[0093] Step 1022: Acquire a second slag well image from the second slag well video.

[0094] The first slag well image and the second slag well image are taken simultaneously.

[0095] Step 1031: input the first slag well image and the second slag well image into the slag block detection model respectively to obtain the information of the slag blocks in the first slag well image and the information of the slag blocks in the second slag well image.

[0096] Step 1041: Determine whether there is accumulated slag in the current slag well according to the information of the slag blocks in the first slag well image, the information of the slag blocks in the second slag well image, the area threshold and the height threshold.

[0097] In this embodiment, two cameras are arranged in the slag pit to capture images of the slag pit at different angles, so that the condition of the slag in the slag pit can be detected more comprehensively, so that the slag accumulation can be detected promptly and quickly, thereby ensuring stable operation of the system.

[0098] In some embodiments, the target detection module needs to be obtained through training. The training process of the target detection model is described in detail below with a specific embodiment.

[0099] See also Figure 2 , Figure 2 A flowchart of a method for training a target detection module in a slag detection model provided in an embodiment of the present disclosure. Figure 2 The embodiments shown can be performed individually or in combination. Figure 1 The embodiment shown is performed. Figure 2 The embodiment shown and Figure 1 The illustrated embodiments are performed in combination, and steps 201 to 203 are performed before step 101 .

[0100] Step 201: Acquire multiple sample slag well images.

[0101] Furthermore, the scales of the sample slag pit images can be multiple. Several sample slag pit images of different scales are set, and a certain number of iterations are separated during training, and a scale is randomly selected for training to enhance robustness. In addition, in practical applications, the size of the slag pit image input to the target detection module can be of multiple different sizes, making the target detection module more widely used.

[0102] Step 202: oversampling the plurality of sample slag well images, and / or synthesizing a part or all of the plurality of sample slag well images to obtain a plurality of processed slag well images.

[0103] Among them, oversampling processing refers to copying small target objects multiple times in the image, increasing the number of times small-sized targets appear in the image without affecting other objects, and improving the probability of being included in the reference frame (anchor). In the scenario of detecting slag block accumulation in the slag pit of the dry slag machine of the slag conveying system, oversampling the sample slag pit image can increase the number of times small target slag blocks often appear in the sample slag pit image, thereby solving the problem of fewer small target slag block images, so that the trained target detection module has higher detection accuracy and higher detection efficiency.

[0104] Among them, the synthesis operation refers to merging two or more sample slag well images into one sample slag well image in a certain way, so as to increase the number of small target slag blocks in the scenario of detecting slag block accumulation in the slag well of the dry slag machine of the slag conveying system, and increase the maximum number of images that the model can process at one time.

[0105] It should be noted that part or all of the multiple sample slag pit images may be obtained for synthesis operation.

[0106] Step 203: input the multiple processed slag pit images into the initial target detection module for training until the initial target detection module converges to obtain a target detection module.

[0107] In this embodiment, the target detection module is trained in combination with the scenario of detecting slag block accumulation in the slag pit of the dry slag machine of the slag conveying system, so as to obtain a target detection module that can accurately detect slag accumulation. The module has a fast processing speed and high processing accuracy, meets the real-time requirements of slag accumulation identification in the slag pit, and can detect the accumulation of slag blocks in the slag pit in a timely manner, thereby ensuring the stable operation of the system.

[0108] Further, step 203 can be implemented in the following manner:

[0109] Step 2031, inputting a plurality of processed slag well images into an initial target detection module to obtain position information of slag blocks in the plurality of processed slag well images;

[0110] Step 2032: Calculate the loss function according to the position information of the slag blocks in the processed slag well image until the loss function is less than the loss threshold, thereby obtaining a target detection module.

[0111] Furthermore, different weights can be given to the losses generated on different feature maps (feature_map), and the losses generated by the small target feature map can be given greater weights. This improves the detection accuracy of small target slag blocks. Thus, a target detection module that accurately detects slag accumulation is obtained, which has a fast processing speed and high processing accuracy, meets the real-time requirements of slag accumulation identification in the slag pit, and can detect the accumulation of slag blocks in the slag pit in a timely manner to ensure stable operation of the system.

[0112] The method provided by the present disclosure is described below with a specific example. It can be understood that this example is used to explain and supplement the above embodiment and does not constitute a limitation of the present disclosure. In this example, it is assumed that there are 4 slag pits, and a high-temperature camera is set on both sides of each of the 4 slag pits, so that a total of 8 high-temperature cameras are set. The method of this example includes the following steps:

[0113] Step 1: Mark the slag in the frame images of the videos taken by 8 high-temperature cameras.

[0114] Step 2: Use the YOLOv7-based target detection module to train the dataset consisting of the frame images marked in step 1.

[0115] Step 3: Get the trained weight model file.

[0116] Step 4: Load the model file to detect slag targets in the image, set the confidence threshold and save the detection results.

[0117] Among them, the main thing is to extract category ID information, bounding box coordinates and confidence score.

[0118] The confidence level refers to the degree of credibility that the detected object is a slag block. If the confidence level is greater than the confidence level threshold, it can be determined that the detected object is a slag block.

[0119] Step 5. Load the Segment-Anything model library and the SAM model weight file, and import the category ID, confidence score, and bounding box coordinates obtained in the previous step into SAM.

[0120] Step 6. Initialize the SAM file with the specified pre-trained weights and select the model type from the SAM model registry to generate the segmentation mask.

[0121] Step 7: Use SAM to segment the loaded image to be detected.

[0122] Step 8. Create an aggregation mask from the segmentation mask to generate a composite image.

[0123] Step 9: Determine whether there is slag accumulation at present based on the synthesized image.

[0124] Among them, the slag block external rectangular frame information (external rectangular frame coordinates, confidence score, area center of gravity, pixel height, etc.) can be obtained through the above steps.

[0125] When multiple slag targets are detected, the ratio of the pixel area to the detection area area and the ratio of the maximum pixel of multiple detection circumscribed rectangular frames to the detection area height can be used to determine whether there is an abnormal situation of excessive slag accumulation. For example, if the total pixel area ratio is greater than 45% and the maximum detection height ratio is greater than 50%, it is determined that there is slag accumulation and an alarm signal needs to be provided.

[0126] In this example, multiple videos in the slag pit are collected, decoded, and displayed in real time on the workstation. Frame images are obtained from the video, and the presence of slag accumulation is determined through the slag block detection model, so an alarm is issued and linkage control is performed. By utilizing the efficient detection function of YOLOv7 and the ability of SAM to segment objects in a zero-sample manner, the need for large-scale retraining or data annotation is minimized, thereby improving the stability of the system.

[0127] See also Figure 3 , Figure 3 The present invention provides a structural schematic diagram of a detection system for slag block accumulation in a slag pit of a slag conveying system dry slag machine, and the detection system for slag block accumulation in a slag pit of a slag conveying system dry slag machine includes: a camera module 301 and a server 302. The camera module 301 is connected to the server 302 for communication; the camera module 301 is used to capture images in the slag pit; the server 302 is used to execute the method of any of the above method embodiments. No further description is given here.

[0128] Based on the detection method of slag block accumulation in the slag pit of the dry slag machine of the slag conveying system described in any of the above embodiments, the embodiment of the present disclosure further provides a computer-readable storage medium, for example, the non-temporary computer-readable storage medium may be a read-only memory (ROM), a random access memory (RAM), a CD-ROM, a magnetic tape, a floppy disk, and an optical data storage device, etc. The storage medium stores computer instructions for executing the detection method of slag block accumulation in the slag pit of the dry slag machine of the slag conveying system described in any of the above embodiments, which will not be described in detail here.

[0129] A person skilled in the art will understand that all or part of the steps to implement the above embodiments may be accomplished by hardware, or by a program to instruct the relevant hardware, and the program may be stored in a computer-readable storage medium, and the above-mentioned storage medium may be a read-only memory, a disk or an optical disk, etc.

[0130] An embodiment of the present disclosure provides a computer program product, including a computer program, which implements the steps of the method in any of the above embodiments when executed by a processor.

[0131] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the disclosure disclosed herein. This application is intended to cover any variations, uses, or adaptations of the present disclosure that follow the general principles of the present disclosure and include common knowledge or customary techniques in the art that are not disclosed in the present disclosure. The description and examples are intended to be exemplary only, and the true scope and spirit of the present disclosure are indicated by the following claims.

Claims

1. A method for detecting slag block accumulation in a slag pit of a dry slag machine of a slag conveying system, characterized in that: The method comprises: Shoot the slag well video through the camera module; Acquire a slag well image from the slag well video; The slag pit image is input into the slag block detection model to obtain the information of the slag blocks in the slag pit image; the slag block detection model includes a target detection module and a semantic segmentation module connected in sequence; the target detection module is a model established based on the target detection YOLOv7; the target detection module is a trained model for identifying the position information of the slag blocks in the input slag pit image; the semantic segmentation module is a model established based on the Segment Everything Model SAM; the semantic segmentation module is used to obtain the information of the slag blocks in the input slag pit image according to the position information of the slag blocks in the input slag pit image; the information of the slag blocks in the input slag pit image is used to indicate the outer edge of the slag blocks in the slag pit image; According to the information of the slag blocks in the slag well image, the area threshold and the height threshold, it is determined that there is slag accumulation in the slag well.

2. The method according to claim 1, characterized in that The target detection module includes a feature extraction submodule Backbone, an intermediate submodule Neck and a prediction submodule Head; the feature extraction submodule is established based on spatial pyramid pooling SPP; the intermediate submodule is established based on feature pyramid network FPN; the prediction submodule is established based on non-maximum suppression algorithm DIOU-NMS.

3. The method according to claim 1, characterized in that The determining, based on the information of the slag blocks in the slag well image, the area threshold and the height threshold, that there is slag accumulation in the current slag well comprises: Determine the total pixel area and the maximum height of the slag blocks according to the information of the slag blocks in the slag well image; If the total pixel area of ​​the slag block is greater than the area threshold, and the maximum height of the slag block is greater than the height threshold, it is determined that there is slag accumulation in the slag pit.

4. The method according to claim 1, characterized in that: The camera module includes a first camera and a second camera; the first camera and the second camera have different shooting angles; The method of shooting the slag pit video by the camera module includes: Shooting the first slag well video by the first camera; Shooting a video of the second slag well by using a second camera; Acquiring a slag well image from the slag well video includes: Acquire a first slag well image from the first slag well video, and acquire a second slag well image from the second slag well video, wherein the first slag well image and the second slag well image are taken at the same time; Inputting the first slag well image and the second slag well image into the slag block detection model respectively, to obtain information about slag blocks in the first slag well image and information about slag blocks in the second slag well image; The determining, based on the information of the slag blocks in the slag well image, the area threshold and the height threshold, that there is slag accumulation in the current slag well comprises: According to the information of the slag blocks in the first slag well image, the information of the slag blocks in the second slag well image, the area threshold and the height threshold, it is determined that there is slag accumulation in the current slag well.

5. The method according to claim 1, characterized in that After determining that there is slag accumulation in the current slag well according to the information of the slag block in the slag well image, the area threshold and the height threshold, the method further includes: Send an alarm message to the power plant control system, wherein the alarm message is used to indicate that there is slag accumulation in the slag pit.

6. The method according to any one of claims 1 to 5, characterized in that: The method further comprises: Acquire multiple sample slag well images; Performing oversampling processing on the multiple sample slag well images, and / or synthesizing a part or all of the multiple sample slag well images to obtain a plurality of processed slag well images; The multiple processed slag pit images are input into an initial target detection module for training until the initial target detection module converges to obtain a target detection module.

7. The method according to claim 6, characterized in that The step of inputting the plurality of processed slag pit images into an initial target detection module for training until the initial target detection module converges to obtain a target detection module comprises: Inputting the plurality of processed slag pit images into an initial target detection module to obtain position information of slag blocks in the plurality of processed slag pit images; According to the position information of the slag blocks in the processed slag well image, a loss function is calculated until the loss function is less than a loss threshold, thereby obtaining a target detection module.

8. A detection system for slag block accumulation in the slag pit of a dry slag machine of a slag conveying system, characterized in that: include: A camera module and a server; the camera module is in communication with the server; the camera module is used to capture images in the slag pit; The server is used to implement the method according to any one of claims 1 to 7.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 7 are implemented.

Images