A machine vision-based plow belt discharger position state monitoring method and device

Abstract

The present application relates to a kind of plough type belt discharger position state monitoring method and device based on machine vision, comprising the following steps: step one image acquisition and annotation;Step two training classification model;Step three target detection model;Step four model deployment;Step five real-time monitoring and control;Step six exception handling;Step seven data storage.The present application can accurately identify the upper and lower limit positions of the discharger through deep learning algorithms, ensuring that the discharger is lifted and lowered into place, reducing material leakage or accumulation caused by inaccurate positioning, and monitoring the discharger status in real time, automatically controlling the discharger to move to the designated position, reducing manual intervention, improving work efficiency and response speed.

Description

Technical Field

[0001] This invention belongs to the field of machine vision technology, and in particular relates to a method and device for monitoring the position and status of a plow-type belt unloader based on machine vision. Background Technology

[0002] As the "switch" at the coal bunker's inlet, the unloader directly determines whether the coal level in the bunker reaches the expected height, thus affecting the completion of the entire process. If the unloader is not raised properly, some of the coal flow on the conveyor belt may be intercepted and spilled onto the ground on either side or into the coal bunker. Conversely, if the unloader is not lowered properly, some coal flow may leak from the bottom of the unloader, affecting production efficiency and safety.

[0003] Currently, traditional methods for monitoring the position of unloaders often rely on manual inspection or simple sensors. These methods have many drawbacks, including low efficiency, susceptibility to human error, slow response time, limited data recording and analysis capabilities, and high cost. Furthermore, they cannot achieve real-time monitoring. With the rapid development of machine vision technology, its application in industrial automation is becoming increasingly widespread. Machine vision technology acquires real-time image information of the target environment through cameras and uses deep learning algorithms to analyze and process the images, thereby achieving functions such as object recognition, positioning, and measurement. Compared to traditional manual inspection and sensor monitoring methods, machine vision technology offers higher accuracy, faster speed, and greater adaptability. Therefore, in the coal conveying process, it is necessary to use machine vision to monitor the position and status of the unloader in real time to improve production efficiency. Summary of the Invention

[0004] (I) Purpose of the Invention

[0005] To overcome the above shortcomings, the present invention aims to provide a machine vision-based method and device for monitoring the position and status of a plow-type belt unloader, so as to solve the above technical problems.

[0006] (II) Technical Solution

[0007] To achieve the above objectives, the technical solution provided in this application is as follows:

[0008] A machine vision-based method for monitoring the position status of a plow-type belt unloader includes the following steps:

[0009] Step 1: Image acquisition and annotation. The image field of view includes the unloader and the rear conveyor belt. The unloader is annotated using the target detection dataset, and the acquired conveyor belt images are categorized using the classification recognition dataset.

[0010] Step 2: Train the classification model. Based on the classification and identification dataset in Step 1, divide the coal presence / absence dataset into a training set and a validation set, and train the coal presence / absence classification model.

[0011] Step 3: Object detection model. Based on the labeled unloader data in Step 1, YOLOv10s is used as the object detection model. The YOLOv10s model structure mainly includes: a backbone network, a feature enhancement network, and a detection head.

[0012] Step four involves deploying the trained model to an AI edge computing server for real-time image processing.

[0013] Step 5: Real-time monitoring and control. The real-time monitored images are transmitted to the deployed model to detect the position of the unloader and the state of the coal flow. Based on the detection results, the unloader is automatically controlled to move to the designated position, and a voice prompt is given.

[0014] Step 6: Abnormal handling. If the unloader does not reach the designated position, the unloader will be automatically controlled to reach the designated position, and evidence images will be sent to the platform. If coal flow is continuously detected, it will be determined that the coal retaining plate at the bottom of the unloader is damaged. Evidence images will be sent and the site will be linked to issue a voice reminder to replace the coal retaining plate.

[0015] Step 7: Data storage. The image identified as a foreign object is stored in an SD card or hard drive with a large storage capacity.

[0016] Preferably, in step one, an intrinsically safe camera is used to acquire images. The camera is installed directly above the belt running direction and captures images of the unloader and the belt downwards. The unloader occupies 40% of the image area and is centered. The image resolution is not less than 1920*1080. The acquired images are divided into images of the unloader at various heights and images of the belt running normally with and without coal.

[0017] Preferably, the target detection data in step one is labeled using the deep learning image labeling software CVAT, with the labeling type being a rectangular box and the labeling category being a unloader.

[0018] Preferably, the classification and recognition dataset in step one includes images of empty conveyor belts and images of conveyor belts containing coal. The collected images are cropped to extract the area of ​​the conveyor belt behind the unloader, and the images are labeled separately to indicate whether they contain coal or not.

[0019] Preferably, the classification model in step two is ResNet18.

[0020] Preferably, the classification model ResNet18 includes the following components:

[0021] The input layer receives images of the belt behind the belt unloader.

[0022] The convolutional layer uses a 7x7 convolutional kernel to perform convolution operations to extract features of coal-free images, with a stride of 2, followed by a max pooling layer.

[0023] Residual blocks: ResNet18 consists of multiple residual blocks, each containing three convolutional layers: 1x1, 3x3, and 1x1. The 1x1 layer is used for dimension transformation, and the 3x3 layer is used for feature extraction.

[0024] Residual connections, where the input of each residual block is directly added to the block's output through a short connection, help solve the gradient vanishing problem;

[0025] Global average pooling layer: Used at the end of the network to reduce the number of parameters in the fully connected layer.

[0026] A fully connected layer is used for coal presence / absence sorting tasks.

[0027] Preferably, the YOLOv10s model specifically includes:

[0028] The backbone network, employing a lightweight convolutional neural network, is responsible for extracting features from the input image. It contains multiple convolutional layers and residual connections to improve the efficiency and accuracy of feature extraction.

[0029] The feature enhancement network, located between the backbone network and the detection head, is responsible for fusing feature maps of different scales.

[0030] The detection head employs the concept of a decoupled head, separating the regression and prediction branches. This makes the network's training and inference more efficient.

[0031] The loss function uses a composite loss function, including classification loss, localization loss, and target loss, which work together on the prediction results during training to optimize model performance.

[0032] Preferably, the target detection model has two red lines, namely the upper line and the lower line. The upper line is used as the judgment line for whether the unloader has risen to the correct position, and the lower line is used as the judgment line for whether the unloader has descended to the correct position. The blue box represents the unloader detection result. The position status of the unloader is determined by the y-coordinate of the lower right corner of the unloader detection box.

[0033] Preferably, step five specifically includes:

[0034] The image is input into the target detection model of the unloader to obtain the unloader detection result. Based on the y-coordinate value of the lower right corner of the detection box, the position of the unloader is determined:

[0035] If the y-value of the unloader detection frame is less than the y-value of the upper line, it means that the unloader has risen to the correct position.

[0036] If the y-value of the unloader detection frame is greater than the upper y-value but less than the lower y-value, it indicates that the unloader is not rising to the correct position.

[0037] If the y-value of the unloader detection frame is greater than the y-value of the lower line, it means that the unloader has descended to the correct position.

[0038] Based on this, the unloader position information is obtained. If the unloader is detected to have descended to the correct position, the presence or absence of coal on the conveyor belt is classified according to step two to obtain the coal presence or absence classification result of the current image. If 10 consecutive frames of images indicate the presence of coal, it indicates that the coal baffle plate of the unloader is damaged. At this time, the relevant images are uploaded to the system platform and the corresponding camera is linked to issue a voice reminder on-site: "Unloader damage detected, please handle it as soon as possible." If the unloader is detected to be not raised to the correct position, the relevant images are uploaded to the system platform and a voice alarm is issued to remind the administrator to handle it as soon as possible. The corresponding camera needs to be linked to issue a voice reminder on-site: "Unloader not raised to the correct position detected, please handle it as soon as possible." At this time, the lifting command is sent through the linked camera to control the unloader to be raised to the designated position.

[0039] A machine vision-based position status monitoring device for a plow-type belt unloader includes an image acquisition device, a communication transmission device, a data processing device, and a data storage device.

[0040] The image acquisition device is an intrinsically safe camera for underground coal mines, capable of capturing high-definition images of the coal conveyor belt during operation. The surface of the intrinsically safe camera is covered with dust-proof material to prevent the adhesion of underground dust and its impact on image quality. The intrinsically safe camera is equipped with a control camera. The DO interface of the control camera control board is connected to the unloader PLC device, enabling voice prompts and control of the unloader. The control camera control board can control the opening and closing of two switches.

[0041] Communication transmission equipment includes equipment required for 5G, industrial Wi-Fi and other wireless transmission methods, as well as equipment required for USB 3.0, Ethernet and other wired transmission methods;

[0042] The edge processing device uses an AI edge computing server, which has the computing power and memory to complete ResNet-based category classification and YOLO-based image object detection tasks;

[0043] The data storage device is an SD card or hard drive with storage capacity, used to store images identified as foreign objects.

[0044] Beneficial effects:

[0045] It has the following beneficial effects:

[0046] 1. Improve monitoring accuracy and speed: By utilizing machine vision technology and deep learning algorithms, high-precision and rapid identification and monitoring of the unloader position and coal flow status are achieved, thereby improving the efficiency and accuracy of monitoring.

[0047] 2. Enhanced system adaptability and real-time monitoring capabilities: The system is unaffected by ambient light and dust, has strong adaptability, and can acquire and process image information in real time to achieve real-time monitoring of the unloader position and coal flow status.

[0048] 3. Reduced costs and improved safety: Compared with traditional monitoring methods, machine vision technology reduces human resource input and equipment maintenance costs, while improving production efficiency and safety through automatic control and voice prompts.

[0049] 4. Intelligent control and anomaly handling: The monitoring results can automatically control the unloader to move to the designated position, and provide voice reminders and upload evidence images when an anomaly is detected, thus realizing intelligent anomaly handling.

[0050] 5. Data recording, analysis and storage: The system can record and analyze monitoring data to support production management and decision-making, and store images identified as foreign objects in a large-capacity storage device for subsequent analysis and processing. Attached Figure Description

[0051] Figure 1 This is a flowchart of the process of the present invention. Detailed Implementation

[0052] To make the objectives, technical solutions, and advantages of this invention clearer, the following detailed embodiments are described in conjunction with the appendix. Figure 1 The present invention will be described in further detail below. It should be understood that these descriptions are merely exemplary and not intended to limit the scope of the invention. Furthermore, descriptions of well-known structures and techniques are omitted in the following description to avoid unnecessarily obscuring the concept of the invention.

[0053] This invention provides a machine vision-based method for monitoring the position status of a plow-type belt unloader, comprising the following steps:

[0054] Step one involves image acquisition and annotation. The image field of view includes the unloader and the rear conveyor belt. In step one, an intrinsically safe camera is used to acquire images. The camera is mounted directly above the conveyor belt in the direction of travel, capturing downward images of both the unloader and the conveyor belt. The unloader occupies 40% of the image area and is centered. The image resolution is no less than 1920*1080. The acquired images are divided into images of the unloader at various heights and images of the conveyor belt operating normally with and without coal. The unloader is annotated using a target detection dataset, and the acquired conveyor belt images are categorized using a classification recognition dataset.

[0055] The target detection data mentioned in step one is labeled using the deep learning image labeling software CVAT, with the label type being a rectangular box and the label category being a unloader;

[0056] The classification and recognition dataset includes images of empty conveyor belts and images of conveyor belts with coal. The collected images are cropped to extract the area of ​​the conveyor belt behind the unloader, and the images are labeled separately to indicate whether they contain coal or not.

[0057] Step 2: Train the classification model. Based on the classification and recognition dataset in Step 1, divide the coal presence / absence dataset into a training set and a validation set, and train the coal presence / absence classification model. The classification model in Step 2 is ResNet18.

[0058] The ResNet18 classification model includes the following components:

[0059] The input layer receives images of the belt behind the belt unloader.

[0060] The convolutional layer uses a 7x7 convolutional kernel to perform convolution operations to extract features of coal-free images, with a stride of 2, followed by a max pooling layer.

[0061] Residual blocks: ResNet18 consists of multiple residual blocks, each containing three convolutional layers: 1x1, 3x3, and 1x1. The 1x1 layer is used for dimension transformation, and the 3x3 layer is used for feature extraction.

[0062] Residual connections, where the input of each residual block is directly added to the block's output through a short connection, help solve the gradient vanishing problem;

[0063] Global average pooling layer: Used at the end of the network to reduce the number of parameters in the fully connected layer.

[0064] A fully connected layer is used for coal presence / absence sorting tasks.

[0065] Step 3: Target detection model. Based on the labeled unloader data in the target detection dataset of Step 1, the target detection model has two red lines: an upper line and a lower line. The upper line is used to determine whether the unloader has risen to the correct position, and the lower line is used to determine whether the unloader has descended to the correct position. The blue box represents the unloader detection result. The position status of the unloader is determined by the y-coordinate of the lower right corner of the unloader detection box.

[0066] YOLOV10s is used as the target detection model. The YOLOV10s model structure mainly includes: a backbone network, a feature enhancement network, and a detection head.

[0067] The YoloV10s model specifically includes:

[0068] The backbone network, employing a lightweight convolutional neural network, is responsible for extracting features from the input image. It contains multiple convolutional layers and residual connections to improve the efficiency and accuracy of feature extraction.

[0069] The feature enhancement network, located between the backbone network and the detection head, is responsible for fusing feature maps of different scales.

[0070] The detection head employs the concept of a decoupled head, separating the regression and prediction branches. This makes the network's training and inference more efficient.

[0071] The loss function uses a composite loss function, including classification loss, localization loss, and target loss, which work together on the prediction results during training to optimize model performance.

[0072] Step four involves deploying the trained model to an AI edge computing server for real-time image processing.

[0073] Step 5: Real-time monitoring and control. The real-time monitored images are transmitted to the deployed model to detect the position of the unloader and the state of the coal flow. Based on the detection results, the unloader is automatically controlled to move to the designated position, and a voice prompt is given.

[0074] Step five specifically includes:

[0075] The image is input into the target detection model of the unloader to obtain the unloader detection result. Based on the y-coordinate value of the lower right corner of the detection box, the position of the unloader is determined:

[0076] If the y-value of the unloader detection frame is less than the y-value of the upper line, it means that the unloader has risen to the correct position.

[0077] If the y-value of the unloader detection frame is greater than the upper y-value but less than the lower y-value, it indicates that the unloader is not rising to the correct position.

[0078] If the y-value of the unloader detection frame is greater than the y-value of the lower line, it means that the unloader has descended to the correct position.

[0079] Based on this, the unloader position information is obtained. If the unloader is detected to have descended to the correct position, the presence or absence of coal on the conveyor belt is classified according to step two to obtain the coal presence or absence classification result of the current image. If 10 consecutive frames of images indicate the presence of coal, it indicates that the coal baffle plate of the unloader is damaged. At this time, the relevant images are uploaded to the system platform and the corresponding camera is linked to issue a voice reminder on-site: "Unloader damage detected, please handle it as soon as possible." If the unloader is detected to be not raised to the correct position, the relevant images are uploaded to the system platform and a voice alarm is issued to remind the administrator to handle it as soon as possible. The corresponding camera needs to be linked to issue a voice reminder on-site: "Unloader not raised to the correct position detected, please handle it as soon as possible." At this time, the lifting command is sent through the linked camera to control the unloader to be raised to the designated position.

[0080] Step 6: Abnormal handling. If the unloader does not reach the designated position, the unloader will be automatically controlled to reach the designated position, and evidence images will be sent to the platform. If coal flow is continuously detected, it will be determined that the coal retaining plate at the bottom of the unloader is damaged. Evidence images will be sent and the site will be linked to issue a voice reminder to replace the coal retaining plate.

[0081] Step 7: Data storage. The image identified as a foreign object is stored in an SD card or hard drive with a large storage capacity.

[0082] A machine vision-based position status monitoring device for a plow-type belt unloader includes an image acquisition device, a communication transmission device, a data processing device, and a data storage device.

[0083] The image acquisition device is an intrinsically safe camera for underground coal mines, capable of capturing high-definition images of the coal conveyor belt during operation. The surface of the intrinsically safe camera is covered with dust-proof material to prevent the adhesion of underground dust and its impact on image quality. The intrinsically safe camera is equipped with a control camera. The DO interface of the control camera control board is connected to the unloader PLC device, enabling voice prompts and control of the unloader. The control camera control board can control the opening and closing of two switches.

[0084] Communication transmission equipment includes equipment required for 5G, industrial Wi-Fi and other wireless transmission methods, as well as equipment required for USB 3.0, Ethernet and other wired transmission methods;

[0085] The edge processing device uses an AI edge computing server, which has the computing power and memory to complete ResNet-based category classification and YOLO-based image object detection tasks;

[0086] The data storage device is an SD card or hard drive with storage capacity, used to store images identified as foreign objects.

[0087] This invention provides a machine vision-based method and device for monitoring the position and status of a plow-type belt unloader. The invention utilizes an intrinsically safe camera to pre-acquire images of the unloader and the coal flow on the belt, establishing an image dataset. Images from this dataset are used to train a unloader position detection model and a coal presence / absence classification model. The unloader detection model includes two types of samples: single-sided and double-sided unloaders. The target detection model is trained using all samples. First, the unloader is detected in the images. The status of the preceding and following frames is used to determine whether the unloader has stopped, and the detection results determine whether the unloader is in a designated position. If the unloader is not in the designated position, it is automatically controlled to move to the designated position, and an evidence image is sent to the platform. If the unloader has descended to the designated position, the image of the belt behind the unloader is cropped, and a classification model is applied to identify the presence or absence of coal. If coal is detected continuously for a period of time, it is determined that the coal retainer plate at the bottom of the unloader is damaged. An evidence image is sent to the platform, and a voice prompt is triggered on-site to promptly replace the coal retainer plate. This invention utilizes deep learning algorithms to enable the system to accurately identify the upper and lower limits of the unloader, ensuring that the unloader is raised and lowered in place. This reduces material leakage or accumulation caused by inaccurate positioning, monitors the unloader's status in real time, and automatically controls the unloader to move to the designated position, reducing manual intervention and improving operational efficiency and response speed.

[0088] This invention utilizes machine vision technology and deep learning algorithms to achieve high-precision and rapid identification and monitoring of the unloader's position and coal flow status, improving monitoring efficiency and accuracy. The system is unaffected by ambient light and dust, exhibiting strong adaptability and capable of real-time acquisition and processing of image information to achieve real-time monitoring of the unloader's position and coal flow status. Compared to traditional monitoring methods, machine vision technology reduces manpower input and equipment maintenance costs, while improving production efficiency and safety through automatic control and voice prompts. Monitoring results can automatically control the unloader to move to a designated position, and provide voice prompts and upload evidence images when an anomaly is detected, achieving intelligent anomaly handling. The system can record and analyze monitoring data, providing support for production management and decision-making, and stores images identified as foreign objects in a large-capacity storage device for subsequent analysis and processing.

[0089] It should be noted that, in this document, relational terms such as "first" and "second" are used only to distinguish one entity or operation from another, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Furthermore, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, article, or apparatus. Without further limitations, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, article, or apparatus that includes said element.

[0090] The above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit it. Although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features. Such modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for monitoring the position status of a plow-type belt unloader based on machine vision, characterized in that, Includes the following steps: Step 1: Image acquisition and annotation. The image field of view includes the unloader and the rear conveyor belt. The unloader is annotated using the target detection dataset, and the acquired conveyor belt images are categorized using the classification recognition dataset. Step 2: Train the classification model. Based on the classification and identification dataset in Step 1, divide the coal presence / absence dataset into a training set and a validation set, and train the coal presence / absence classification model. Step 3: Object detection model. Based on the labeled unloader data in Step 1, YOLOv10s is used as the object detection model. The YOLOv10s model structure mainly includes: a backbone network, a feature enhancement network, and a detection head. The target detection model has two red lines, namely the upper line and the lower line. The upper line is used to judge whether the unloader has risen to the correct position, and the lower line is used to judge whether the unloader has fallen to the correct position. The blue box represents the unloader detection result. The position status of the unloader is determined by the y-coordinate of the lower right corner of the unloader detection box. Step four involves deploying the trained model to an AI edge computing server for real-time image processing. Step 5: Real-time monitoring and control. The real-time monitored images are transmitted to the deployed model to detect the position of the unloader and the state of the coal flow. Based on the detection results, the unloader is automatically controlled to move to the designated position, and a voice prompt is given. Step five specifically includes: The image is input into the target detection model of the unloader to obtain the unloader detection result. Based on the y-coordinate value of the lower right corner of the detection box, the position of the unloader is determined: If the y-value of the unloader detection frame is less than the y-value of the upper line, it means that the unloader has risen to the correct position. If the y-value of the unloader detection frame is greater than the upper y-value but less than the lower y-value, it indicates that the unloader is not rising to the correct position. If the y-value of the unloader detection frame is greater than the y-value of the lower line, it means that the unloader has descended to the correct position. Based on this, the unloader position information is obtained. If the unloader is detected to have descended to the correct position, the presence or absence of coal on the conveyor belt is classified and detected according to step two. The coal presence or absence classification result of the current image is obtained. If 10 consecutive frames of images are judged to contain coal, it indicates that the coal baffle plate of the unloader is damaged. At this time, the relevant images are uploaded to the system platform and the corresponding camera is linked to issue a voice reminder on site: "Unloader damage detected, please handle it as soon as possible." If the unloader is detected to be not raised to the correct position, the relevant images are uploaded to the system platform and a voice alarm is issued to remind the administrator to handle it as soon as possible. The corresponding camera needs to be linked to issue a voice reminder on site: "Unloader not raised to the correct position detected, please handle it as soon as possible." At this time, the lifting command is sent through the linked camera to control the unloader to be raised to the designated position. Step 6: Abnormal handling. If the unloader does not reach the designated position, the unloader will be automatically controlled to reach the designated position, and evidence images will be sent to the platform. If coal flow is continuously detected, it will be determined that the coal retaining plate at the bottom of the unloader is damaged. Evidence images will be sent and the site will be linked to issue a voice reminder to replace the coal retaining plate. Step 7: Data storage. The image identified as a foreign object is stored on an SD card or hard drive.

2. The method for monitoring the position status of a plow-type belt unloader based on machine vision according to claim 1, characterized in that, In step one, an intrinsically safe camera is used to acquire images. The camera is installed directly above the belt running direction and captures images of the unloader and the belt downwards. The unloader occupies 40% of the image area and is centered. The image resolution is not less than 1920*1080. The acquired images are divided into images of the unloader at various heights and images of the belt running normally with and without coal.

3. The method for monitoring the position status of a plow-type belt unloader based on machine vision according to claim 1, characterized in that, The target detection data mentioned in step one is labeled using the deep learning image labeling software CVAT, with the label type being rectangular boxes and the label category being unloader.

4. The method for monitoring the position status of a plow-type belt unloader based on machine vision according to claim 1, characterized in that, The classification and recognition dataset mentioned in step one includes images of empty conveyor belts and images of conveyor belts containing coal. The collected images are cropped to extract the area of ​​the conveyor belt behind the unloader, and the images are labeled separately to indicate whether they contain coal or not.

5. The method for monitoring the position status of a plow-type belt unloader based on machine vision according to claim 1, characterized in that, The classification model in step two is ResNet18.

6. The method for monitoring the position status of a plow-type belt unloader based on machine vision according to claim 5, characterized in that, The ResNet18 classification model includes the following components: The input layer receives images of the belt behind the belt unloader. The convolutional layer uses a 7x7 convolutional kernel to perform convolution operations to extract features of coal-free images, with a stride of 2, followed by a max pooling layer. Residual blocks: ResNet18 consists of multiple residual blocks, each containing three convolutional layers: 1x1, 3x3, and 1x1. The 1x1 layer is used for dimensionality transformation, and the 3x3 layer is used for feature extraction. Residual connections, where the input of each residual block is directly added to the block's output through a short connection, help solve the gradient vanishing problem; Global average pooling layer: Used at the end of the network to reduce the number of parameters in the fully connected layer. A fully connected layer is used for coal presence / absence sorting tasks.

7. The method for monitoring the position status of a plow-type belt unloader based on machine vision according to claim 1, characterized in that, The YoloV10s model specifically includes: The backbone network, employing a lightweight convolutional neural network, is responsible for extracting features from the input image. It contains multiple convolutional layers and residual connections to improve the efficiency and accuracy of feature extraction. The feature enhancement network, located between the backbone network and the detection head, is responsible for fusing feature maps of different scales. The detection head adopts the idea of ​​a decoupled head, which separates the regression branch and the prediction branch, making the training and inference of the network more efficient. The loss function uses a composite loss function, including classification loss, localization loss, and target loss, which work together on the prediction results during training to optimize model performance.

8. A machine vision-based plow-type belt unloader position status monitoring device, characterized in that, A machine vision-based method for monitoring the position status of a plow-type belt unloader as described in any one of claims 1-7, comprising an image acquisition device, a communication transmission device, a data processing device, and a data storage device. The image acquisition device is an intrinsically safe underground camera for coal mines, capable of capturing high-definition images of the coal conveyor belt during operation. The surface of the intrinsically safe camera is covered with dust-proof material to prevent the adhesion of underground dust and its impact on image quality. The intrinsically safe camera is equipped with a control camera, and the DO interface of the control camera control board is connected to the unloader PLC device, enabling voice prompts and control of the unloader. The control camera control board can control the opening and closing of two switches. Communication transmission equipment includes equipment required for 5G, industrial Wi-Fi and other wireless transmission methods, as well as equipment required for USB 3.0, Ethernet and other wired transmission methods; The edge processing device uses an AI edge computing server, which has the computing power and memory to complete ResNet-based category classification and YOLO-based image object detection tasks; The data storage device is an SD card or hard drive with storage capacity, used to store images identified as foreign objects.

Images