Intelligent ore online screening monitoring method and system

Abstract

The invention is suitable for the technical field of screening monitoring, and particularly relates to an intelligent ore online screening monitoring method and system.The method comprises the steps that screening equipment in an ore screening processing area is found out, an image recognition model is created, and ore images in the screening process are collected through video monitoring equipment deployed in the screening equipment in advance; creating an annotated image set, training the image recognition model, and extracting model parameters from the trained image recognition model; an online monitoring platform for ore screening is created, registered users are recognized, and the identity attribute of each registered user is configured. According to the method, the version most suitable for the environment of the subscriber can be selected by determining the target model, the accuracy of the image recognition model is further improved, the target model is deployed in the edge device, real-time monitoring and quick response to the screening device can be achieved, the edge computing power is fully utilized, the data transmission pressure is relieved, and the data transmission efficiency is improved. And the overall calculation cost is reduced.

Description

Technical Field

[0001] This invention relates to the field of screening and monitoring technology, and in particular to an intelligent online screening and monitoring method and system for ores. Background Technology

[0002] Ore screening monitoring refers to the process of real-time collection, analysis, and evaluation of the ore's status during screening, grading, and conveying stages through cameras or online detection devices placed around the screening equipment during ore processing and beneficiation. Combined with deep learning models, it identifies risk characteristics in the ore screening process, including: ore accumulating at the screen edge (approaching clogging), ore containing a large amount of clay (easily sticking to the screen), or the presence of other ores or debris in the ore.

[0003] However, during the transformation towards digitalization and intelligent screening in the mining industry, due to the unique characteristics of ore materials, the ore images, particle characteristics, and operational status data collected from a single mine are far from covering all possible ore types, particle size distributions, and impurities. When training deep learning models, the lack of training samples easily leads to insufficient generalization ability, especially for some small mines. Furthermore, small mines typically have limited computing resources and lack high-performance computing clusters, further increasing the difficulty of training complex neural networks. This prevents the neural networks from undergoing multiple rounds of iterative optimization, significantly limiting the accuracy of model recognition.

[0004] Therefore, "how to address the problem of limited deep learning model training caused by limited sample size and computing resources in small mining farms" is the technical problem that this invention aims to solve. Summary of the Invention

[0005] The purpose of this invention is to provide an intelligent online screening and monitoring method and system for ores, in order to solve the problem of "how to deal with the problem of limited training of deep learning models in small mines due to limited sample size and computing resources" mentioned in the background art.

[0006] To achieve the above objectives, the present invention provides the following technical solution:

[0007] An intelligent online screening and monitoring method for ores, the method comprising:

[0008] Locate the screening equipment in the ore screening and processing area, create an image recognition model, use video monitoring equipment pre-deployed in the screening equipment to collect ore images during the screening process, create a labeled image set, train the image recognition model, and extract model parameters from the trained image recognition model.

[0009] Create an online monitoring platform for ore screening, identify registered users, configure the identity attributes of each registered user, wherein the identity attributes include at least: sharer and subscriber, receive model parameters uploaded by sharers, collect hardware information of screening equipment of each registered user, and cluster registered users into several groups, wherein there is at least one sharer in each group;

[0010] The model parameters of the sharer are distributed in parallel to all subscribers in the same group through the online monitoring platform, and the model parameters are deployed into the image recognition model pre-built by the subscriber to obtain several control models;

[0011] Obtain the labeled image set of the subscriber and train the control model to construct a test image set, input it into the control model, and set the confidence score of each control model according to the output results. Define the control model with the highest confidence score as the target model, where each subscriber corresponds to a target model. Locate the edge equipment in the ore screening and processing area of ​​the subscriber, deploy the target model to the edge equipment, and perform online monitoring of the screening equipment.

[0012] Furthermore, the method also includes:

[0013] The video surveillance equipment collects image data from the subscriber screening device and inputs it into the target model, outputting several image features. These image features are then clustered into a regular group, an uncertain group, and a risk group.

[0014] From the image data, a snapshot containing uncertain groups is extracted, and the snapshot is sent to the image recognition model of the sharer in the same group to output the evaluation result;

[0015] Based on the evaluation results, the image features of the uncertain group are classified into the normal group or the risk group.

[0016] Furthermore, the method also includes:

[0017] Create emergency response rules and establish the correspondence between each risk characteristic in the risk group and the emergency response rules;

[0018] The emergency response rules are written into the edge device, and control permissions for the screening device are granted to the edge device.

[0019] Furthermore, the steps of creating a labeled image set, training the image recognition model, and extracting model parameters from the trained image recognition model include:

[0020] Integrate the sharers and subscribers corresponding to the same set of model parameters and define them as node groups. Based on the node groups, build a federated learning architecture.

[0021] The model parameters of subscribers are dynamically updated through the federated learning architecture.

[0022] Furthermore, the steps of creating an online monitoring platform for ore screening, identifying registered users, and configuring the identity attributes of each registered user include:

[0023] The system receives labeled image sets uploaded by registered users to the online monitoring platform, traces back the data source, and collects the feature information of the data source, wherein the feature information includes at least: ore type and particle size;

[0024] Based on the feature information, the labeled image set is merged.

[0025] Furthermore, the step of distributing the sharer's model parameters in parallel to all subscribers within the same group via the online monitoring platform includes:

[0026] Record the changes in model parameters, generate a version, and upload it to the online monitoring platform for storage;

[0027] A timestamp is embedded into the version and arranged in chronological order to generate a version chain, and a rollback mechanism is integrated.

[0028] Furthermore, the steps of locating the edge devices in the subscriber's ore screening and processing area, deploying the target model to the edge devices, and performing online monitoring of the screening equipment include:

[0029] The confidence score is adjusted according to a preset frequency;

[0030] Establish a mapping between screening equipment and edge devices, build a communication link, and generate an online monitoring network.

[0031] Furthermore, the system includes:

[0032] The training module is used to locate the screening equipment in the ore screening and processing area, create an image recognition model, use video monitoring equipment pre-deployed in the screening equipment to collect ore images during the screening process, create a labeled image set, train the image recognition model, and extract model parameters from the trained image recognition model.

[0033] The clustering module is used to create an online monitoring platform for ore screening, identify registered users, configure the identity attributes of each registered user, wherein the identity attributes include at least: sharer and subscriber, receive model parameters uploaded by sharers, collect hardware information of screening equipment of each registered user, and cluster registered users into several groups, wherein there is at least one sharer in each group;

[0034] The module is used to distribute the model parameters of the sharer to all subscribers in the same group in parallel via the online monitoring platform, and to deploy the model parameters into the image recognition model pre-built by the subscribers to obtain several control models;

[0035] The monitoring module is used to acquire the labeled image set of the subscriber, train the control model, construct a test image set, input it into the control model, and set the confidence score of each control model according to the output results. The control model with the highest confidence score is defined as the target model, where each subscriber corresponds to a target model. The module also locates the edge equipment in the ore screening and processing area of ​​the subscriber, deploys the target model to the edge equipment, and performs online monitoring of the screening equipment.

[0036] Furthermore, the training module includes:

[0037] Integration units are used to integrate the sharers and subscribers corresponding to the same set of model parameters and define them as node groups. Based on the node groups, a federated learning architecture is built.

[0038] The update unit is used to dynamically update the model parameters of the subscribers via the federated learning architecture.

[0039] Furthermore, the clustering module includes:

[0040] The receiving unit is used to receive the set of labeled images uploaded by registered users to the online monitoring platform, trace back the data source, and collect the feature information of the data source, wherein the feature information includes at least: ore type and particle size;

[0041] The merging unit is used to merge the labeled image set based on the feature information.

[0042] Compared with the prior art, the beneficial effects of the present invention are:

[0043] By utilizing video surveillance equipment and image recognition models to monitor screening equipment, issues such as ore condition, screen blockage, and abnormal particle size can be identified in real time, reducing the frequency of manual inspections and the error rate. By dividing registered users into sharers and subscribers, the difficulty of model training caused by insufficient subscriber data can be solved, reducing repeated training and annotation efforts and significantly reducing training time and computational resource consumption. By clustering registered users into groups, model parameters can be shared within the same group, greatly improving the adaptability and recognition accuracy of model parameters and increasing the model deployment efficiency for subscribers. By identifying the target model, the version most suitable for the subscriber's own environment can be selected, further improving the accuracy of the image recognition model. By deploying the target model to edge devices, real-time monitoring and rapid response of screening equipment can be achieved, making full use of edge computing power, reducing data transmission pressure, and lowering overall computing costs. Attached Figure Description

[0044] Figure 1 A flowchart illustrating the intelligent online ore screening and monitoring method provided in this embodiment of the invention;

[0045] Figure 2 This is a first sub-flowchart of the intelligent online ore screening and monitoring method provided in an embodiment of the present invention;

[0046] Figure 3 This is a second sub-flowchart of the intelligent online ore screening and monitoring method provided in an embodiment of the present invention;

[0047] Figure 4 This is a third sub-flowchart of the intelligent online ore screening and monitoring method provided in the embodiments of the present invention;

[0048] Figure 5 This is a fourth sub-flowchart of the intelligent online ore screening and monitoring method provided in this embodiment of the invention;

[0049] Figure 6 This is a block diagram of the intelligent online ore screening and monitoring system provided in an embodiment of the present invention;

[0050] Figure 7 A block diagram of the training module in the intelligent online ore screening and monitoring system provided in this embodiment of the invention;

[0051] Figure 8 A block diagram of the clustering module in the intelligent online ore screening and monitoring system provided in this embodiment of the invention;

[0052] Figure 9 A block diagram of the modules obtained in the intelligent online ore screening and monitoring system provided in the embodiments of the present invention;

[0053] Figure 10This is a block diagram of the monitoring module in the intelligent online ore screening and monitoring system provided in an embodiment of the present invention. Detailed Implementation

[0054] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the invention.

[0055] In Example 1, Figure 1 The implementation flow of the intelligent online ore screening and monitoring method provided in this embodiment of the invention is illustrated below, and is described in detail below:

[0056] S100: Locate the screening equipment in the ore screening and processing area, create an image recognition model, use video monitoring equipment pre-deployed in the screening equipment to collect ore images during the screening process, create a labeled image set, train the image recognition model, and extract model parameters from the trained image recognition model.

[0057] The area for ore screening is designated, which could be a mine or a processing area. Screening equipment, such as vibrating screens or fixed screens, is located through on-site surveys or equipment inventory. A deep learning network architecture (such as a convolutional neural network) suitable for industrial vision scenarios is selected to create an image recognition model. Keyframes of the screening process are captured using video monitoring equipment, and preprocessing operations such as noise reduction, cropping, and viewpoint correction are performed. Then, ore processing managers annotate the keyframes, marking abnormal features in the images. These abnormal features include: ore... Stones accumulate at the edge of the screen (approaching blockage), the ore contains a large amount of clay (easily sticking to the screen), or other ores and miscellaneous stones are present in the ore. A labeled image set is constructed using the labeled keyframes; in other words, the labeled image set is a collection of labeled keyframes. The image recognition model is trained using this labeled image set, enabling it to identify the visual features of the ore under different screening conditions and the operating status of the equipment. Model parameters are extracted from the trained image recognition model, including: convolutional kernel weights, feature extraction layer parameters, fully connected layer weights, and classification thresholds.

[0058] S200: Create an online monitoring platform for ore screening, identify registered users, configure the identity attributes of each registered user, wherein the identity attributes include at least: sharer and subscriber, receive model parameters uploaded by sharers, collect hardware information of screening equipment of each registered user, and cluster registered users into several groups, wherein there is at least one sharer in each group.

[0059] An online monitoring platform is created, primarily for the collection, sharing, and management of model parameters. Users who register with their real identities are defined as registered users. It's important to note that "users" here refers to mining farms. The platform receives registration information uploaded by registered users and, based on their willingness and business capabilities in the ore screening industry, classifies them into sharers and subscribers. Sharers are registered users who voluntarily share the model parameters of their mining farm, while subscribers are registered users who, due to limitations in their own data and computing resources, are unable to train neural networks. Sharers upload the trained and optimized model parameters to the online monitoring platform, making them available to subscribers. This enables the sharing and reuse of model parameters, providing other mining farms with a foundation for directly deployable or further optimized image recognition models. The system receives and stores model parameters uploaded to the online monitoring platform by sharers; it collects hardware information from each registered user, including equipment model, screen specifications, and screen vibration frequency. Based on the similarity of hardware information, registered users are categorized into several groups, with at least one sharer in each group. The model parameters uploaded by this sharer can be accessed by subscribers within the same group. The advantage of this method is that by sharing model parameters from sharers within the same group, the adaptability of model parameters can be further improved, while simultaneously increasing the recognition accuracy of the image recognition model.

[0060] S300: The model parameters of the sharer are distributed in parallel to all subscribers in the same group via the online monitoring platform, and the model parameters are deployed to the image recognition model pre-built by the subscriber to obtain several control models.

[0061] Using an online monitoring platform, the model parameters of the sharer are simultaneously distributed to all subscribers within the same group. Subscribers then deploy the received model parameters into their pre-built image recognition models and perform recognition and analysis on the image data collected by the video surveillance equipment. In actual deployment, when there is only one sharer in the group, each subscriber receives one set of model parameters. Similarly, when there are multiple sharers in the group, each subscriber receives multiple sets of model parameters. Using the same image recognition model architecture and multiple sets of model parameters, multiple control models are generated. It can be seen that each control model carries model parameters from different sources.

[0062] S400: Obtain the labeled image set of the subscriber, train the control model, construct a test image set, input it into the control model, and set the confidence score of each control model according to the output results. Define the control model with the highest confidence score as the target model, where each subscriber corresponds to a target model. Locate the edge equipment in the ore screening and processing area of ​​the subscriber, deploy the target model to the edge equipment, and perform online monitoring of the screening equipment.

[0063] By utilizing video surveillance equipment deployed in subscribers, image data of the screening equipment during processing is collected. Annotated image sets are generated for each subscriber through manual labeling. These sets are used to train a control model, and the model parameters are fine-tuned to adapt to the specific screening environment and ore characteristics of the subscriber's mine. After training, an independent test image set is constructed, still composed of image data from the subscriber's screening equipment. This test image set is input into each control model to obtain output results. Ore processing managers score the control models based on the output results and set corresponding confidence scores. The control model with the highest confidence score is defined as the target model. It can be seen that the number of control models may be one or more, the specific number is not limited, but the number of control models is limited to one.

[0064] Identify edge devices located in the subscriber's ore screening and processing area. These edge devices can be edge computing boxes, smart gateways, etc. Deploy the target model into the edge devices. In the screening equipment, a smart camera with data processing capabilities can be selected, and the target model can be deployed in the smart camera. Using the target model, the image data of the screening equipment collected in real time is processed, thereby enabling online monitoring of the operating status of the screening equipment and the ore screening process.

[0065] In Embodiment 2, unlike Embodiment 1, the method further includes:

[0066] The video surveillance equipment collects image data from the subscriber screening device and inputs it into the target model, outputting several image features. These image features are then clustered into a regular group, an uncertain group, and a risk group.

[0067] From the image data, a snapshot containing uncertain groups is extracted, and the snapshot is sent to the image recognition model of the sharer in the same group to output the evaluation result;

[0068] Based on the evaluation results, the image features of the uncertain group are classified into the normal group or the risk group.

[0069] Video monitoring equipment deployed on subscriber screening devices collects image data in real time during the ore screening process. This image data is then input into a pre-deployed target model. The target model uses a pre-trained feature extraction network to identify and generate several image feature vectors. These feature vectors reflect the ore's packing state, particle size distribution, mud content, impurities, and screen blockage. The feature vectors are compared with a pre-built vector set, and the corresponding image data is categorized into normal, uncertain, and risk groups. The pre-built vector set consists of multiple feature vectors and corresponding risk assessments. The normal group includes image features indicating normal screening and ore characteristics that meet expectations. The uncertain group includes image features with blurred features or other risks that cannot be clearly determined. The uncertain group may be caused by dust obscuring the image data, affecting its accuracy, or by model distortion. The risk group corresponds to image features with obvious abnormalities or risk characteristics, such as screen blockage, severe ore packing, or excessively high impurity ratios.

[0070] From the image data corresponding to the uncertain group, potentially risky features are extracted to obtain a snapshot. This snapshot is then sent to the image recognition model of the sharing user to obtain an evaluation result. Based on this evaluation result, the image features of the uncertain group are classified into the normal group or the risk group. In this application, by sharing model parameters, the image recognition model can be rapidly deployed without relying on a large amount of its own data for independent training. Furthermore, since the image data cannot be transmitted externally, the risk of production data leakage is greatly reduced. By only sending snapshots to the image recognition models of other registered users, the status of the screening equipment can be quickly determined while protecting data privacy.

[0071] In Embodiment 3, unlike Embodiment 1, the method further includes:

[0072] Create emergency response rules and establish the correspondence between each risk characteristic in the risk group and the emergency response rules;

[0073] The emergency response rules are written into the edge device, and control permissions for the screening device are granted to the edge device.

[0074] Set corresponding emergency response rules for each risk characteristic, and write all risk characteristics into the edge device. At the same time, grant control permissions to the screening equipment to the edge device so that the edge device can quickly adjust the screening equipment according to the emergency response rules when a risk characteristic is detected.

[0075] In Example 4, Figure 2The implementation flow of the intelligent online screening and monitoring method for ores provided in this embodiment of the invention is illustrated below. The steps of creating a labeled image set, training the image recognition model, and extracting model parameters from the trained image recognition model are described in detail below:

[0076] S101: Integrate the sharers and subscribers corresponding to the same set of model parameters and define them as node groups. Based on the node groups, build a federated learning architecture.

[0077] S102: The model parameters of the subscribers are dynamically updated via the federated learning architecture.

[0078] By integrating sharers and subscribers using the same set of model parameters to form node groups, a federated learning architecture is constructed. This allows each node group to perform distributed training and iterative optimization of the image recognition model through the sharing and aggregation of model parameters, without sharing the original image data from the screening equipment. The advantage of this approach is that it can maintain the updating and iteration of the image recognition model even when the subscriber's data sample size is small.

[0079] In Example 5, Figure 3 The implementation flow of the intelligent online ore screening and monitoring method provided by an embodiment of the present invention is illustrated below. The steps of creating an online monitoring platform for ore screening, identifying registered users, and configuring the identity attributes of each registered user are described in detail below:

[0080] S201: Receive the set of labeled images uploaded by registered users to the online monitoring platform, trace back the data source, and collect the feature information of the data source, wherein the feature information includes at least: ore type and particle size.

[0081] A community interaction module is built in the online monitoring platform, allowing registered users to share labeled image sets. It should be noted that the labeled image sets are shared voluntarily and are historical data, so there is no risk of data leakage. The data source of each labeled image set is determined, which is the registered user who uploaded and shared the labeled image set. The characteristic information of the data source is determined, including ore type and particle size.

[0082] S202: Based on the feature information, merge the labeled image set.

[0083] By merging labeled image sets with the same or similar features, the generalization ability of the target model can be enhanced while broadening the data sources.

[0084] In Example 6, Figure 4The implementation flow of the intelligent online ore screening and monitoring method provided by an embodiment of the present invention is illustrated. The following details the step of distributing the model parameters of the sharer to all subscribers in the same group in parallel via the online monitoring platform:

[0085] S301: Record the changes in model parameters, generate a version, and upload it to the online monitoring platform for storage.

[0086] After the target model is updated and iterated, the process of changing the model parameters is recorded, including the parameter modification time, modification content, mine and operator information, etc. The process of changing the parameters is recorded in the form of versions, and the resulting versions are uploaded to the online monitoring platform for storage.

[0087] S302: Embed a timestamp into the version, arrange them in chronological order, generate a version chain, and integrate a rollback mechanism.

[0088] Embed timestamps into the versions and integrate all versions in chronological order to generate a version chain. Integrate a rollback mechanism into the version chain. The rollback mechanism means that when the recognition accuracy of the target model decreases, an anomaly is detected, or a deployment failure occurs, it can roll back to any version and re-call the model parameters in that version.

[0089] In Example 7, Figure 5 The implementation flow of the intelligent online ore screening and monitoring method provided by an embodiment of the present invention is illustrated. The following details the steps of locating the edge devices in the subscriber's ore screening and processing area, deploying the target model to the edge devices, and performing online monitoring of the screening devices:

[0090] S401: Adjust the confidence score according to a preset frequency.

[0091] The confidence score is updated at a preset frequency, which can be once a day or once a week.

[0092] S402: Establish a mapping between screening equipment and edge equipment, build a communication link, and generate an online monitoring network.

[0093] Based on the deployment location and usage of screening equipment and edge devices, an association is established between the screening equipment and edge devices, with each screening equipment corresponding to at least one edge device. A communication link is also established to integrate all screening equipment and edge devices and build an online monitoring network. The advantage of this approach is that it enables data sharing and collaborative work between different screening equipment and edge nodes.

[0094] Figure 6This diagram illustrates the structural composition of an intelligent online ore screening and monitoring system provided in an embodiment of the present invention. The intelligent online ore screening and monitoring system 1 includes:

[0095] Training module 11 is used to locate screening equipment in the ore screening and processing area, create an image recognition model, use video monitoring equipment pre-deployed in the screening equipment to collect ore images during the screening process, create a labeled image set, train the image recognition model, and extract model parameters from the trained image recognition model.

[0096] Clustering module 12 is used to create an online monitoring platform for ore screening, identify registered users, configure the identity attributes of each registered user, wherein the identity attributes include at least: sharer and subscriber, receive model parameters uploaded by sharers, collect hardware information of screening equipment of each registered user, and cluster registered users into several groups, wherein there is at least one sharer in each group;

[0097] The module 13 is used to distribute the model parameters of the sharer to all subscribers in the same group in parallel via the online monitoring platform, and to deploy the model parameters into the image recognition model pre-built by the subscribers to obtain several control models;

[0098] The monitoring module 14 is used to acquire the labeled image set of the subscriber, train the control model, construct a test image set, input it into the control model, and set the confidence score of each control model according to the output results. The control model with the highest confidence score is defined as the target model, where each subscriber corresponds to a target model. The module also locates the edge equipment in the ore screening and processing area of ​​the subscriber, deploys the target model to the edge equipment, and performs online monitoring of the screening equipment.

[0099] Figure 7 This diagram illustrates the structural composition of the intelligent online ore screening and monitoring system provided in an embodiment of the present invention. The training module 11 includes:

[0100] Integration unit 111 is used to integrate the sharers and subscribers corresponding to the same set of model parameters and define them as node groups. Based on the node groups, a federated learning architecture is built.

[0101] Update unit 112 is used to dynamically update the model parameters of subscribers via the federated learning architecture.

[0102] Figure 8 This diagram illustrates the structural composition of an intelligent online ore screening and monitoring system provided in an embodiment of the present invention. The clustering module 12 includes:

[0103] The receiving unit 121 is used to receive the set of labeled images uploaded by registered users to the online monitoring platform, trace back the data source, and collect the feature information of the data source, wherein the feature information includes at least: ore type and particle size;

[0104] The merging unit 122 is used to merge the labeled image set based on the feature information.

[0105] Figure 9 This diagram illustrates the structural composition of the intelligent online ore screening and monitoring system provided in an embodiment of the present invention. The obtaining module 13 includes:

[0106] Recording unit 131 is used to record the changes in model parameters, generate a version, and upload it to the online monitoring platform for storage.

[0107] The rollback unit 132 is used to embed timestamps into the version, arrange them in chronological order, generate a version chain, and integrate a rollback mechanism.

[0108] Figure 10 This diagram illustrates the structural composition of an intelligent online ore screening and monitoring system provided in an embodiment of the present invention. The monitoring module 14 includes:

[0109] The adjustment unit 141 is used to adjust the confidence score according to a preset frequency;

[0110] Unit 142 is used to establish a mapping between screening equipment and edge equipment, build a communication link, and generate an online monitoring network.

[0111] The training module 11 is mainly used to complete step S100, the clustering module 12 is mainly used to complete step S200, the acquisition module 13 is mainly used to complete step S300, and the monitoring module 14 is mainly used to complete step S400.

[0112] Integration unit 111 is mainly used to complete step S101, and updating unit 112 is mainly used to complete step S102;

[0113] The receiving unit 121 is mainly used to complete step S201, and the merging unit 122 is mainly used to complete step S202.

[0114] Recording unit 131 is mainly used to complete step S301, and rollback unit 132 is mainly used to complete step S302;

[0115] The adjustment unit 141 is mainly used to complete step S401, and the establishment unit 142 is mainly used to complete step S402.

[0116] The technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0117] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

[0118] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the protection scope of the present invention.

Claims

1. An intelligent on-line ore screening monitoring method, characterized by, The method comprises: Finding out the screening equipment in the ore screening processing area, creating an image recognition model, collecting ore images in the screening process by using the video monitoring equipment pre-deployed in the screening equipment, creating a labeled image set, training the image recognition model, and extracting model parameters from the trained image recognition model; Creating an online monitoring platform for ore screening, identifying registered users, configuring the identity attributes of each registered user, wherein the identity attributes at least include: sharer and subscriber, receiving the model parameters uploaded by the sharer, collecting the hardware information of the screening equipment of each registered user, and clustering the registered users into several groups, wherein at least one sharer exists in each group; Through the online monitoring platform, the model parameters of the sharer are distributed to all subscribers in the same group in parallel, and the model parameters are deployed into the image recognition model pre-built by the subscribers to obtain several comparison models; Obtaining the labeled image set of the subscriber, training the comparison model, constructing a test image set, inputting into the comparison model, and setting the confidence score of each comparison model according to the output result, and defining the comparison model with the highest confidence score as the target model, wherein each subscriber corresponds to a target model, positioning the edge device in the subscriber's ore screening processing area, deploying the target model into the edge device, and performing online monitoring on the screening equipment.

2. The intelligent ore on-line screening monitoring method according to claim 1, characterized in that, The method further comprises: Through the video monitoring equipment, image data of the subscriber's screening equipment is collected and input into the target model, and several image features are outputted, and the image features are clustered into a regular group, an uncertain group and a risk group; From the image data, a snapshot containing the uncertain group is intercepted, and the snapshot is sent to the image recognition model of the sharer in the same group, and an evaluation result is outputted; Through the evaluation result, the image features of the uncertain group are classified into the regular group or the risk group.

3. The intelligent ore on-line screening monitoring method according to claim 2, characterized in that, The method further comprises: Creating an emergency disposal rule, establishing a corresponding relationship between each risk feature in the risk group and the emergency disposal rule; The emergency disposal rule is written into the edge device, and the control authority of the screening equipment is opened to the edge device.

4. The intelligent ore on-line screening monitoring method according to claim 1, characterized in that, The steps of creating a labeled image set, training the image recognition model, and extracting model parameters from the trained image recognition model comprise: Matching the sharer and the subscriber corresponding to the same group of model parameters, and defining as a node group, and constructing a federated learning architecture based on the node group; Through the federated learning architecture, the model parameters of the subscriber are dynamically updated.

5. The intelligent on-line ore screening monitoring method according to claim 1, characterized in that, The steps of creating an online monitoring platform for ore screening, identifying registered users, and configuring the identity attributes of each registered user comprise: Receiving the labeled image set uploaded by the registered user into the online monitoring platform, tracing back the data source, collecting feature information of the data source, wherein the feature information at least includes: ore type and granularity; Based on the feature information, the labeled image set is merged.

6. The intelligent on-line ore screening monitoring method according to claim 4, characterized in that, The step of distributing the model parameters of the sharer to all the subscribers in the same group in parallel via the online monitoring platform comprises: recording the change process of the model parameters, generating a version, and uploading to the online monitoring platform for storage; embedding a timestamp in the version and arranging it in chronological order to generate a version chain and integrate a rollback mechanism.

7. The intelligent on-line ore screening monitoring method according to claim 1, characterized in that, The step of locating the edge device in the subscriber's ore screening processing area, deploying the target model to the edge device, and performing online monitoring on the screening device comprises: adjusting the confidence score according to the preset frequency; establishing a mapping between the screening device and the edge device, building a communication link, and generating an online monitoring network.

8. An intelligent on-line ore screening monitoring system characterized by, The system comprises: a training module for finding the screening device in the ore screening processing area, creating an image recognition model, collecting ore images in the screening process using a video monitoring device pre-deployed in the screening device, creating a labeled image set, training the image recognition model, and extracting model parameters from the trained image recognition model; a clustering module for creating an online monitoring platform for ore screening, identifying registered users, configuring the identity attributes of each registered user, wherein the identity attributes at least include: sharer and subscriber, receiving the model parameters uploaded by the sharer, collecting the hardware information of the screening device of each registered user, and clustering the registered users into several groups, wherein at least one sharer exists in each group; a obtaining module for distributing the model parameters of the sharer to all the subscribers in the same group in parallel via the online monitoring platform, and deploying the model parameters to the image recognition model pre-built by the subscribers to obtain several reference models; a monitoring module for obtaining the labeled image set of the subscribers, training the reference models, building a test image set, inputting it into the reference models, and setting the confidence score of each reference model according to the output result, defining the reference model with the highest confidence score as the target model, wherein each subscriber corresponds to a target model, locating the edge device in the subscriber's ore screening processing area, deploying the target model to the edge device, and performing online monitoring on the screening device.

9. The intelligent ore on-line screening monitoring system according to claim 8, characterized in that, The training module comprises: an integration unit for integrating the sharer and the subscriber corresponding to the same group of model parameters, and defining it as a node group, and building a federated learning architecture based on the node group; an update unit for dynamically updating the model parameters of the subscribers via the federated learning architecture.

10. The intelligent ore on-line screening monitoring system according to claim 8, characterized in that, The clustering module comprises: a receiving unit for receiving the labeled image set uploaded by the registered user to the online monitoring platform, tracing back the data source, collecting the feature information of the data source, wherein the feature information at least includes: ore type and granularity; a merging unit for merging the labeled image set based on the feature information.

Images