Intelligent table concentrator diversion ore receiving sorting system and sorting method and software center platform

Abstract

The invention relates to an intelligent table concentrator diversion ore-receiving sorting system, a sorting method and a software central platform, and the system comprises a high-performance server, a touch display screen, a data switch, a central control cabinet, a gigabit Ethernet line, a table concentrator group, a diversion ore-receiving sorting device, a table concentrator water supply control group and a table concentrator ore-feeding control group, the shaking table group comprises a shaking table group upstream, a shaking table group midstream and a shaking table group downstream, each shaking table is used for mineral separation and is provided with a diversion ore receiving separation device, the diversion ore receiving separation device is connected with the data switch through a first gigabit Ethernet line and transmits image data to the high-performance server, and the high-performance server is connected with the data switch through a second gigabit Ethernet line. And a control instruction is transmitted to the high-performance server through the second gigabit Ethernet line. Through the one-control-multiple intelligent control center, a plurality of shakers are controlled at the same time, and the functions of accurately identifying the demarcation point of the ore belt and dynamically adjusting the flow guide ore receiving plate are achieved, so that the labor intensity is reduced, and the separation precision and efficiency are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of mineral processing, and in particular relates to an intelligent shaking table diversion and ore sorting system, a sorting method and a software central platform. Background Art

[0002] In the mineral processing process of non-coal mines, the gravity separation process plays a vital role, especially in the separation of rare and precious metal ores such as tungsten, tin, tantalum, niobium, and gold. As the core equipment in this process, the mineral processing shaking table loosens and layers the minerals on the bed surface through its unique water flow push and asymmetric reciprocating motion, forming fan-shaped mineral belt distribution such as enriched concentrate belt, intermediate ore belt and tailing belt. However, although the shaking table plays an irreplaceable role in the field of mineral processing, its operation still mainly relies on manual labor, especially in the dynamic positioning and diversion sorting and collection of the concentrate belt. The mineral processing plant is usually equipped with dozens or even hundreds of shaking tables. Workers need to continuously monitor the changes in the ore belt and manually adjust the diversion ore receiving plate based on experience to adapt to the real-time changes in the position of the ore belt. This manual operation method is not only labor-intensive, but also difficult to ensure the consistency of sorting accuracy and efficiency due to differences in workers' subjective judgment and experience. It is also difficult to avoid concentrate loss caused by adjustment lag.

[0003] Although some automation technologies have appeared in the market to try to replace or assist manual operations, they have shown obvious limitations in practical applications. Existing automation technologies usually have several problems: First, automatic control equipment can often only control a single process section or only control a single process section, and it is difficult to achieve centralized control of the shaking table in multiple process sections, which not only leads to high investment costs, but also makes overall management and control inconvenient; secondly, these devices have deficiencies in the accuracy of the ore receiving position, adjustment frequency and response speed, and the processing performance cannot meet production requirements; finally, the existing automation system is greatly affected by the operating environment, and is highly sensitive to the feeding, water supply, vibration parameters and ambient light of the shaking table, and lacks closed-loop feedback control, which makes it difficult for the system to achieve stable operation. These drawbacks not only fail to effectively reduce the labor intensity of workers, but may also lead to concentrate loss due to technical errors, increase production costs, and reduce mineral processing efficiency. Summary of the invention

[0004] In view of the shortcomings of the prior art, the present invention provides an intelligent shaking table diversion and ore sorting system, a sorting method and a software central platform to realize the precise identification of the boundary points of the ore belts and the dynamic adjustment of the diversion and ore receiving plates, thereby reducing labor intensity, improving sorting accuracy and efficiency, and overcoming the limitations of existing automation equipment in terms of accuracy, response speed and environmental adaptability.

[0005] To achieve the above object, the technical solution of the present invention is as follows: An intelligent shaking table diversion and ore sorting system includes a high-performance server, a touch screen, a data switch, a central control cabinet, a first Gigabit Ethernet line, a second Gigabit Ethernet line, a shaking table group, a diversion and ore sorting device, a shaking table water supply control group and a shaking table ore supply control group. The high-performance server is used to process the collected image data and generate control commands. The touch screen is directly connected to the high-performance server to provide a user operation and monitoring interface. The data switch is connected to the high-performance server to aggregate image data from multiple shaking tables. The central control cabinet is used to integrate and install high-performance servers, touch screens and data switches. The shaking table group includes multiple shaking tables, and the shaking table group includes an upstream shaking table group, a midstream shaking table group, and a shaking table group. The downstream of the shaking table group is the next ore dressing process stage upstream of the shaking table group, and the downstream of the shaking table group is the next ore dressing process stage upstream of the shaking table group, wherein each shaking table is used to handle mineral sorting, and each shaking table is equipped with a diversion and ore sorting device, which is connected to the data switch via a first Gigabit Ethernet line to transmit image data to a high-performance server, and transmits control instructions to the high-performance server via a second Gigabit Ethernet line. The high-performance server controls the shaking table water supply control group and the shaking table ore feeding control group respectively through the signal bus according to the sorting situation of the diversion and ore sorting device, so as to realize closed-loop automatic control of automatic ore receiving, automatic water supply and automatic ore feeding of the shaking tables in each ore dressing process section.

[0006] Preferably, each shaking table in the shaking table group comprises a shaking table surface located at the top, a shaking table chute located at the edge of the shaking table surface, and a shaking table fence located at the edge of the shaking table chute.

[0007] Preferably, the diversion and ore sorting device is installed on the shaking table fence, includes a network camera located directly above the shaking table surface, connected to the device support frame through an extendable camera bracket, and is used to collect image data; the drive and controller are located inside the diversion and ore sorting device, and are connected to the stepper motor to receive control instructions from a high-performance server and drive the stepper motor to rotate; the control button panel is connected to the drive and controller, and is used to manually control the rotation of the stepper motor; the stepper motor is connected to the linear belt module to drive the linear module slider to move; the front end of the calibration plate is connected to the linear module slider, and the rear end is connected to the diversion and ore receiving plate, so as to realize the left and right linear movement of the diversion and ore receiving plate under the control of the control button panel or the high-performance server.

[0008] On the other hand, the present invention also discloses an intelligent shaking table diversion and ore sorting method based on the above-mentioned intelligent shaking table diversion and ore sorting system, comprising: Step S1, image preprocessing, compressing the image of the shaking table collected by the network camera to keep the image aspect ratio consistent, and filling the edge area to form a standardized image size; Step S2: constructing a deep learning algorithm model, including the collection and labeling of image samples, model design, and model training; Step S3, model prediction and recognition, inputting the image data of non-training samples into the trained model, using the model weight parameters to identify the boundary points of the mineral belt, generating a segmentation mask to distinguish the various areas of the mineral belt, and the prediction output includes a category label, a bounding box and a pixel-level mask; Step S4, post-processing of ore belt position identification, accurately identifying the boundary point between the concentrate belt and the medium ore belt by analyzing the segmentation mask output by the model, and determining the boundary point as the target position of the guide receiving plate; Step S5, dynamically adjust the guide receiving plate, generate a control command based on the real-time identified ore belt boundary point information, and dynamically adjust the position of the guide receiving plate to align it with the boundary point to ensure accurate guide and sorting of the minerals.

[0009] Preferably, step S1 comprises: S11, input original image; S12, determining the target size; S13, calculating the proportional scaling factor and comparing the sizes; S14, applying the scaling factor to scale the source image in equal proportion; S15, calculating the scaled image size; S16, copying the scaled image to the center position of the target image array to form a new image; S17, filling the edge of the new image array; S18. Output the filled image.

[0010] Preferably, step S2 is completed using a high-performance server, including: S21. Image sample collection and labeling: for a single shaker, image sample data is collected periodically, with the number of samples being no less than 1,000, and regional labeling is performed according to the distribution of ore zones in the samples. The labeling areas are defined as background, concentrate area, intermediate ore area, and tailings area, to form a labeled image data set; S22. Model design, using the improved YOLOv7-Segmentation model, based on multi-scale feature fusion, path aggregation network and fast spatial pyramid pooling technology, combined with the EIoU loss function, to improve the detection and segmentation speed and accuracy of large targets in mining belts and small targets at the marking position of moving devices; S23, model training, input the labeled image sample data set into the designed algorithm model, so that the model can learn the classification and segmentation features of each area. The trained model can generate the final prediction results with category labels, bounding boxes and pixel-level masks, thereby realizing the accurate identification of the boundary points of the mineral belts.

[0011] Preferably, in step S22, the improved YOLOv7-Segmentation model is that the input stage is used to receive an image with a size of 512x512x3, and includes a backbone network C3 module, the C3 module includes: a first convolution layer, configured with 32 filters, a convolution kernel size of 3x3, a step size of 1, a padding of 1, and an output size of 512x512x32 after batch normalization and ReLU activation function processing; a second convolution layer, configured with 64 filters, a convolution kernel size of 3x3, a step size of 1, a padding of 1, and an output size of 512x512x64 after batch normalization and ReLU activation function processing; and a c3k parameter is set to False in the shallow network to form a C2f structure similar to that in YOLOv8.

[0012] Preferably, in step S22, the improved YOLOv7-Segmentation model includes a feature fusion part, wherein the C3 module outputs and generates four feature maps of different scales, with sizes of 512x512x64, 256x256x128, 128x128x256 and 64x64x512 respectively, and further fuses the feature maps of different scales through the feature pyramid network FPN and the path aggregation network PAN, and the output size remains consistent; the multi-head attention mechanism C2PSA is embedded in the C2 mechanism to enhance the spatial perception ability of the feature map; the multi-scale feature map is detected in the P2 detection layer, The 1x1 convolution dimensionality reduction is achieved through the fast spatial pyramid pooling SPPF module, and MaxPool2d with different convolution kernel sizes is used for multiple pooling. The results are connected and the output feature map sizes are 512x512x64, 256x256x128, 128x128x256 and 64x64x512. The detection head part includes a depthwise separable convolution DWConv, an adaptive anchor box mechanism is configured to optimize the anchor box configuration, and an EIoU loss function is introduced to improve the prediction accuracy by considering the overlapping area, aspect ratio and center point offset of the predicted box and the true box. Finally, the prediction results of the category label, bounding box and pixel-level mask are output.

[0013] On the other hand, the present invention also discloses a software central platform for executing the above-mentioned intelligent shaking table diversion and ore sorting method. The tasks of the software central platform are divided into a main thread, an image acquisition thread, a real-time display thread and a logic control thread. The main thread runs the main interface of the software central platform, and the main interface includes a parameter setting module, a monitoring module, a user operation module and a communication module. The main thread starts the other three threads through the user operation module; the image acquisition thread is set to read the image data collected by the network camera at a fixed frequency, and transmits the read image to the data queue; the real-time display thread reads the image data from the data queue at a fixed frequency, and on the one hand displays the image in the main interface in the form of video, and on the other hand sends the latest image data to the logic control thread; the logic control thread includes a mineral belt boundary point recognition module, a logic control algorithm core and a command generation module. The logic control thread generates control commands based on the boundary point information in the mineral belt image, and continuously monitors the changes in the mineral belt boundary point to dynamically adjust the system control parameters.

[0014] Preferably, the software central platform also includes configuration and monitoring functions, which supports the operator to complete the setting of the target offset value of the diversion and receiving plate position in the parameter setting window, and stop the operation of the diversion and ore sorting device at any time through the emergency stop button in the control button panel. After operation, the position of the diversion and receiving plate is automatically adjusted to align with the boundary point of the ore belt, and the changes of the boundary point are continuously monitored. The operator can monitor the system status in real time and view the operation log.

[0015] Compared with the prior art, the advantages of the present invention are: The present invention realizes the automation and intelligence of mineral processing operation by providing an intelligent shaking table diversion and ore sorting system, a sorting method and a software central platform, and significantly improves the sorting accuracy and efficiency. The system uses a deep learning algorithm, and through an improved YOLOv7-segmentation model, while maintaining a high segmentation accuracy, it significantly improves the detection and segmentation speed and accuracy of large targets such as mineral belts and small targets such as motion device flags, and has important practical application value. In addition, by setting the "one control multiple" mode, the system can simultaneously control the operation of multiple shaking tables, simplify the operation process, reduce labor intensity, and enable the system to maintain stable operation in the face of different operating environments, without being affected by environmental factors such as light, and achieve a higher level of automation. The software central platform integrates modules such as parameter setting, real-time monitoring, and user operation, and realizes dynamic monitoring and real-time adjustment of mineral belt demarcation points through logical control threads, further optimizing the sorting process. The overall design of the system helps to improve the accuracy of mineral belt identification, ensure the continuity and accuracy of the sorting process, and thus improve the stability and efficiency of mineral sorting operations, providing reliable technical support for mineral processing operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 This is a schematic diagram of the structure of the multiple intelligent shaking table diversion and ore sorting system of the present invention; Figure 2 It is a top view of the intelligent shaking table diversion and ore sorting system of the present invention; Figure 3 It is a schematic structural diagram of the diversion and ore separation device of the present invention; Figure 4 This is a design framework diagram of a single intelligent shaking table diversion and ore sorting system of the present invention; Figure 5 This is a design framework diagram of the multiple intelligent shaking table diversion and ore sorting system of the present invention; Figure 6 It is a schematic diagram of the flow chart of the intelligent shaking table diversion and ore separation method of the present invention; Figure 7 This is a prediction effect diagram of the deep learning algorithm model of the present invention; Figure 8 It is a schematic diagram of the location of the mineral belt boundary point of the deep learning algorithm model of the present invention; Fig. 9 This is a flow chart of the image preprocessing method of the present invention; Fig.10 A flowchart of building a deep learning algorithm model for the present invention; Fig.11 The original image collected by the image sample of the present invention; Fig.12 is a labeling diagram of an image sample of the present invention; Fig.13 It is the overall framework diagram of the software process of the present invention; Fig.14 It is the overall system architecture diagram of the present invention; Figure markings: 1-high-performance server; 2-touch screen; 3-data switch; 4-central control cabinet; 5-first Gigabit Ethernet line; 6-second Gigabit Ethernet line; 7-shaking table group; 8-diversion and ore sorting device; 9-shaking table water supply control group; 10-shaking table ore feeding control group; 701-upstream of shaking table group; 702-midstream of shaking table group; 703-downstream of shaking table group; 71-shaking table surface; 72-shaking table chute; 73-shaking table fence; 81-network camera; 82-extendable camera bracket; 83-device support frame; 84-drive and controller; 85-control button panel; 86-stepping motor; 87-linear belt module; 88-linear module slider; 89-calibration plate; 810-diversion and ore receiving plate. DETAILED DESCRIPTION

[0017] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the drawings in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by ordinary technicians in this field based on the embodiments of the present invention belong to the present invention.

[0018] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the present application.

[0019] like Figure 1-3 As shown, this embodiment discloses an intelligent shaking table diversion and ore sorting system, including a high-performance server 1, a touch screen 2, a data switch 3, a central control cabinet 4, a first Gigabit Ethernet line 5, a second Gigabit Ethernet line 6, a shaking table group 7, a diversion and ore sorting device 8, a shaking table water supply control group 9 and a shaking table ore supply control group 10. The high-performance server 1 is used to process the collected image data and generate control commands. The touch screen 2 is directly connected to the high-performance server 1 to provide a user operation and monitoring interface, which is convenient for operators to monitor the sorting of the ore belt in real time. The data switch 3 is connected to the high-performance server 1 to aggregate image data from multiple shaking tables, optimize the transmission speed of image data, and ensure the rapid response of the sorting system as a whole. The central control cabinet 4 is used to integrate and install the high-performance server 1, the touch screen 2 and the data switch 3, so as to provide centralized control and security protection of the system.

[0020] The shaking table group 7 includes multiple shaking tables for completing the mineral sorting task. Specifically, the shaking table group 7 includes the upstream shaking table group 701, the midstream shaking table group 702, and the downstream shaking table group 703, forming a segmented process flow. The midstream shaking table group 702 is the next beneficiation process stage of the upstream shaking table group 701, and the downstream shaking table group 703 is the next beneficiation process stage of the midstream shaking table group 702. This structural design divides the beneficiation process into different stages, which is convenient for more accurate segmented control and sorting. A diversion and ore sorting device 8 is installed on each shaking table to divert and connect the minerals. The diversion and ore sorting device 8 is connected to the data switch 3 through the first Gigabit Ethernet line 5 to realize the real-time transmission of image data to the high-performance server 1; at the same time, the control instructions are transmitted to the high-performance server 1 through the second Gigabit Ethernet line 6. Based on this architecture, a master-substation communication mode is formed between the high-performance server 1 and the diversion and ore sorting device 8 to ensure the rapid transmission and execution of control instructions.

[0021] In addition, the high-performance server 1 controls the shaking table water supply control group 9 and the shaking table ore feeding control group 10 respectively through the signal bus according to the sorting situation of the diversion ore sorting device 8, so as to realize the closed-loop automatic control of automatic ore receiving, automatic water supply and automatic ore feeding of the shaking table in each ore dressing process section. This closed-loop control mode effectively improves the overall sorting efficiency while achieving sorting accuracy, reduces the intervention of manual operation, and realizes intelligent ore dressing automation management.

[0022] Each shaking table in the shaking table group 7 includes a shaking table surface 71 at the top, a shaking table chute 72 at the edge of the shaking table surface 71, and a shaking table fence 73 at the edge of the shaking table chute 72. The shaking table surface 71 is a working surface for mineral sorting, and mineral separation is achieved through water flow and reciprocating motion. The shaking table chute 72 is used to guide the flow direction of the sorted minerals, while the shaking table fence 73 ensures the area of ​​mineral flow is limited to prevent mineral overflow.

[0023] The diversion and ore sorting device 8 is installed on the shaking table fence 73, and the structural design helps to accurately control and flexibly adjust its position. The diversion and ore sorting device 8 includes a network camera 81 located directly above the shaking table surface 71, which is connected to the device support frame 83 through an extendable camera bracket 82, and is used to collect image data of the ore belt. The real-time image transmission of the network camera 81 can provide the high-performance server 1 with a basis for identifying the boundary point of the ore belt, ensuring that the position adjustment of the ore receiving plate can accurately align with the boundary point of the ore belt. The drive and controller 84 is located inside the diversion and ore sorting device 8 and is connected to the stepper motor 86 to receive control instructions from the high-performance server 1 and drive the stepper motor 86 to rotate. The control button panel 85 is connected to the drive and controller 84 for manually controlling the rotation of the stepper motor 86 so that manual operation can be performed when necessary. The rotation of the stepper motor 86 drives the linear belt module 87 to drive the linear module slider 88 to move. The front end of the calibration plate 89 is connected to the linear module slider 88, and the rear end is connected to the guide receiving plate 810. The left and right linear movement of the guide receiving plate 810 is realized by the movement of the linear module slider 88, so that the left and right linear movement of the guide receiving plate 810 is realized under the control of the control button panel 85 or the high-performance server 1, ensuring that the guide receiving plate 810 can be accurately aligned with the boundary point of the ore belt to achieve accurate mineral diversion reception.

[0024] The first Gigabit Ethernet line 5 adopts the TCP / IP protocol to transmit the image data collected by the diversion and ore sorting device 8 to the high-performance server 1. Through the TCP / IP protocol, the data transmission has high compatibility and reliability, ensuring that the image data from multiple shaking tables can be stably gathered to the high-performance server 1 for centralized processing, thereby supporting the real-time data input of multiple diversion and ore sorting devices 8.

[0025] The second Gigabit Ethernet line 6 uses the Ethercat protocol to transmit control instructions between the high-performance server 1 and each diversion and ore sorting device 8. The Ethercat protocol has fast and accurate communication characteristics, which can significantly reduce the delay of instruction transmission, so that the high-performance server 1 can achieve real-time control and rapid response to the action of the diversion and ore sorting device 8, ensuring the accuracy of the position adjustment of the diversion and ore receiving plate 810. This network configuration, through the combination of TCP / IP and Ethercat protocols, not only ensures the compatibility of data transmission, but also improves the real-time response performance of the system, ensuring the efficient and stable operation of the intelligent shaking table diversion and ore sorting system.

[0026] like Figure 4As shown in the figure, the design framework of a single intelligent shaking table diversion and ore sorting system mainly includes three components: a camera-based image acquisition module, a motion control and drive module that controls the movement of the diversion and ore receiving plate, and a high-performance server (computer) responsible for image data processing and system control. The system obtains the operating status image of the shaking table surface in real time through the image acquisition module and transmits it to the main control unit via Ethernet. The main control unit uses a preprocessing algorithm and a deep learning model to process the acquired image data, generates control instructions and transmits them to the motion controller and drive module, thereby driving the diversion and ore receiving plate to the target position and completing the mineral diversion and sorting operation. This constitutes a closed-loop intelligent automatic control system of image acquisition-main control unit-ore receiving plate control, which effectively realizes the intelligent sorting of a single shaking table.

[0027] like Figure 5 As shown in the figure, the design framework of the multi-intelligent shaking table diversion and ore sorting system is based on the design of the above single system. Through the distributed layout scheme, the image data of multiple shaking tables are aggregated and transmitted to the high-performance server 1 through the data switch to realize "one-to-many" processing. The processor transmits the processing result instructions to the slave stations (drive and control modules) of each corresponding point through the Ethercat communication protocol, thereby driving the diversion and ore receiving plates of each shaking table to the target position respectively, realizing accurate mineral diversion and sorting, and forming a multi-shaking table intelligent automatic control system controlled by image acquisition-data switch-main control unit-drive and controller-ore receiving plate.

[0028] like Figure 6 As shown, based on a system for guiding and connecting ore for sorting using multiple intelligent shaking tables, an embodiment of the present application also discloses an intelligent shaking table guiding and connecting ore for sorting, including: Step S1, image preprocessing, compressing the image of the shaking table collected by the network camera 81, maintaining the image aspect ratio, and filling the edge area to form a standardized image size. By compressing the image size and filling the edge, the consistency of images of different sizes in subsequent processing is ensured, and the adaptability of the algorithm model to the input image is improved.

[0029] Step S2, building a deep learning algorithm model, including the collection and labeling of image samples, model design, and model training. This step collects sample data and labels it to form a clearly classified training set. The model design uses the YOLOv7-Segmentation model integrated with the CBAM (Convolutional Block Attention Module) module to improve recognition accuracy. Finally, through model training, the system can learn the characteristics of each area and accurately distinguish between the concentrate zone, the middling zone, and the tailings zone.

[0030] Step S3, model prediction and recognition, input the image data of non-training samples into the trained model, use the model weight parameters to identify the boundary points of the mineral belt, generate a segmentation mask to distinguish the various areas of the mineral belt, and the predicted output includes category labels, bounding boxes and pixel-level masks. By using the trained model parameters, the system can quickly analyze and segment new images to achieve accurate identification of various areas of the mineral belt. After learning a large amount of labeled sample data, the model will eventually output a model file with new weight parameters. The file is loaded and the image data in the non-training sample is read to perform model prediction and recognition. The effect is as follows: Figure 7 shown.

[0031] Step S4, post-processing of the ore belt position identification, by analyzing the segmentation mask output by the model, accurately identify the boundary point between the concentrate belt and the medium ore belt, and determine the boundary point as the target position of the diversion and ore receiving plate 810. After the model algorithm is processed, it can be seen that the ore belt has been clearly segmented. As long as the boundary point adjacent to the bottom concentrate belt (yellow area) and the medium ore belt (green area) is found, it can be used as the target position of the diversion and ore receiving plate. This solution obtains the boundary point through the image processing algorithm. Figure 8 Shown in blue circle.

[0032] Step S5, the diversion receiving plate is dynamically adjusted, and a control command is generated according to the real-time identified ore belt demarcation point information, and the position of the diversion receiving plate 810 is dynamically adjusted to align it with the demarcation point to ensure accurate diversion and sorting of the minerals. The dynamic adjustment realizes automatic operation by controlling the movement of the diversion receiving plate 810 to align with the ore belt demarcation point, ensuring that the system can accurately divert and sort, and improve the utilization efficiency of mineral resources.

[0033] like Fig. 9 As shown, the image preprocessing in step S1 includes: S11, input original image; S12, determining the target size; S13, calculating the proportional scaling factor and comparing the sizes; S14, applying the scaling factor to scale the source image in equal proportion; S15, calculating the scaled image size; S16, copying the scaled image to the center position of the target image array to form a new image; S17, filling the edge of the new image array; S18. Output the filled image.

[0034] Among them, the original image size is: 2560×1920; the compressed target size (algorithm model input size): 512×512; the calculation formula involved in the process is as follows: 1) Target and original image size: 2) Calculate the scaling factor: 3) Choose the minimum scaling factor to preserve the aspect ratio: 4) Calculate the scaled image size: 5) Calculate the size that needs to be filled: 6) Scaling the image proportionally: For each pixel In the source image, the new position = , perform zero padding: 7) Output the preprocessed image: The output padded image size is .

[0035] like Fig.10 As shown, step S2 is completed using the high-performance server 1, and specifically includes: S21, image sample collection and labeling, for a single shaker, image sample data is collected periodically, with the number of samples not less than 1,000, and regional labeling is performed according to the distribution of mineral belts in the samples. The labeled areas are defined as background, concentrate area, intermediate mining area and tailings area to form a labeled image data set. In this process, the collection of a large number of image samples ensures the adequacy and generalization ability of model training. Regional labeling enables the model to learn the characteristics of different mineral belts, laying the foundation for subsequent accurate identification. Fig.11 is the original image from which the image samples are collected. Fig.12 is the label map of the image samples.

[0036] S22, model design, adopts the improved YOLOv7-Segmentation model, based on multi-scale feature fusion, path aggregation network and fast spatial pyramid pooling technology, combined with EIoU loss function, to improve the detection and segmentation speed and accuracy of large targets in mining belts and small targets in the marking position of moving devices.

[0037] S23, model training, inputs the labeled image sample data set into the designed algorithm model, so that the model can learn the classification and segmentation features of each area. The trained model can generate the final prediction results with category labels, bounding boxes and pixel-level masks, thereby realizing the accurate identification of the boundary points of the ore belt. Through model training, the algorithm adjusts the internal parameters to adapt to the data characteristics. Finally, the model can accurately parse the newly input shaking table images and provide reliable data support for the automatic control of the system.

[0038] In this embodiment, an improved YOLOv7-Segmentation model is proposed, which significantly improves the performance of object detection and instance segmentation by integrating the CBAM attention mechanism. The model uses the channel and spatial attention capabilities of CBAM to dynamically enhance the key information in the feature map while suppressing irrelevant features. This improvement enables the model to more accurately identify and locate targets and generate high-quality segmentation masks when processing high-resolution input images of 512x512 pixels. Through a carefully designed loss function and optimization strategy, the model further improves the detection accuracy and segmentation quality while maintaining the original fast detection advantage of YOLOv7.

[0039] In this embodiment, the input stage of the model receives an image of size 512x512x3, in the format of width x height x number of channels, to meet the computational requirements of the subsequent network. The backbone network of the model uses the C3 module as the basic structure, in which the first layer of convolution is configured with 32 filters, the convolution kernel size is 3x3, the step size is 1, and the padding is 1 (same padding). After this layer is processed, the output feature map size is 512x512x32. Then, the feature map is processed by batch normalization (BatchNorm) and ReLU activation function to further improve the training stability of the network, and the output size is maintained at 512x512x32.

[0040] In the second convolution layer of the backbone network, the number of filters increases to 64, the convolution kernel size is still 3x3, the stride is 1, and the padding is 1. After the same batch normalization and ReLU activation processing, the output size is updated to 512x512x64. To optimize shallow feature learning, the C3 module uses the C3k2 mechanism, and sets the c3k parameter to False in the shallow network, forming a C2f structure similar to YOLOv8, which enhances the model's performance in processing small target features.

[0041] In the feature fusion part, the C3 module outputs four feature maps of different scales, with sizes of 512x512x64, 256x256x128, 128x128x256, and 64x64x512. These feature maps are further fused through the Feature Pyramid Network (FPN) to extract rich feature information from different scales and ensure consistent output sizes. At the same time, the Path Aggregation Network (PAN) is used to further optimize the feature fusion process so that the final output feature map sizes are 512x512x64, 256x256x128, 128x128x256, and 64x64x512 to improve the detection of large-scale and small targets.

[0042] The C2PSA multi-head attention mechanism is embedded in the C2 mechanism, and the spatial perception ability of the feature map is improved through pyramid spatial attention (PSA), thereby enhancing the network's capture and recognition of targets. The P2 detection layer is set after the first output of the C3 module to increase feature output for multi-scale feature map detection. The output feature map sizes are 512x512x64, 256x256x128, 128x128x256, and 64x64x512.

[0043] The fast spatial pyramid pooling module (SPPF) achieves dimensionality reduction through 1x1 convolution and uses MaxPool2d with different convolution kernel sizes for multiple pooling to obtain the fusion output of features of different scales. The output feature map sizes are 512x512x64, 256x256x128, 128x128x256 and 64x64x512, providing rich multi-scale information for the model detection stage.

[0044] The detection head uses a depthwise separable convolution (DWConv) structure. By adding two DWConv convolution kernels, the model calculation and parameter amount are reduced, and the overall efficiency is improved. The output size of the detection head is determined by the specific parameters and settings of DWConv.

[0045] In addition, to adapt to the characteristics of different data sets, the model introduces an adaptive anchor box mechanism to automatically optimize the anchor box configuration on different data sets, thereby improving the accuracy of target detection. At the same time, the EIoU (Extended IoU) loss function is introduced in the model. When calculating the loss, the overlapping area, aspect ratio, and center point offset of the predicted box and the real box are considered, which further improves the prediction accuracy of the model. In the output stage, the model finally generates prediction results including category labels, bounding boxes, and pixel-level masks, achieving high-precision detection and segmentation of targets.

[0046] like Fig.13As shown, the embodiment of the present application also discloses a software central platform for executing the above-mentioned intelligent shaking table diversion and ore sorting method. The tasks of the software central platform are divided into a main thread, an image acquisition thread, a real-time display thread and a logic control thread. The main thread runs the main interface of the software central platform, and the main interface includes a parameter setting module, a monitoring module, a user operation module and a communication module. Through the user operation module, the main thread can start the other three threads, so that the entire system has good interactivity and flexibility in operation. The image acquisition thread is set to read the image data collected by the network camera 81 at a fixed frequency, and transmits the read image to the data queue to ensure that the image data can be stably collected and transmitted for subsequent processing. The real-time display thread reads the image data from the data queue at a fixed frequency. On the one hand, the image is displayed in the main interface in the form of a video, providing a visual interface for real-time monitoring; on the other hand, the latest image data is sent to the logic control thread to ensure that the control decision is based on the latest image information. The logic control thread includes a ore belt demarcation point recognition module, a logic control algorithm core and a command generation module. The logic control thread generates control commands based on the demarcation point information in the ore belt image, and continuously monitors the changes in the ore belt demarcation points to dynamically adjust the system control parameters, thereby ensuring the accurate positioning and real-time response of the diversion and receiving plate.

[0047] Furthermore, the software hub platform also includes configuration and monitoring functions, which supports the operator to complete the target offset value setting of the diversion receiving plate 810 position in the parameter setting window, so that the operator can flexibly adjust the position of the receiving plate to adapt to different ore belt demarcation point requirements. Through the emergency stop button in the control button panel 85, the operator can stop the operation of the diversion receiving and sorting device 8 at any time, so as to quickly control the equipment status in an emergency and ensure safety. After restarting, the system automatically adjusts the position of the diversion receiving plate 810 to align with the ore belt demarcation point, ensuring that the diversion receiving plate 810 is always in the correct position to achieve efficient mineral diversion sorting. In addition, the system also continuously monitors the changes in the ore belt demarcation point, so that the diversion receiving plate 810 can dynamically adapt to the fluctuation of the ore belt position, further improving the sorting accuracy. The operator can monitor the system status in real time through the software hub platform, view the operation log, track and analyze the historical operation of the system, and provide important reference data for subsequent operation and maintenance.

[0048] This application builds a general framework of the software process according to the functional requirements of the intelligent shaking table diversion and ore sorting system. The general framework design of the software process integrates the functions required by the intelligent shaking table diversion and ore sorting system, covering the control, monitoring and operation functions of the system. In practical applications, the hardware facilities are first installed and perfected to ensure that each component is in normal working condition. Then, the debugging tool is used to debug the communication function between the motion control system to verify whether the data transmission and response between the devices are normal, and the diversion and ore receiving plate 810 is calibrated at the same time to ensure that the diversion and ore receiving plate 810 can accurately align with the ore belt boundary point when the system starts. After completing the hardware debugging, the intelligent shaking table diversion and ore sorting system is debugged as a whole, and the appropriate target offset value is set and confirmed. The offset value is used to affect the final position reached by the diversion and ore receiving plate 810, thereby adjusting the ore receiving grade of the shaking table. This offset value is set in the parameter setting window to ensure that the various configurations of the system meet the sorting requirements. After the configuration is completed, the operator turns on the intelligent shaking table diversion and ore sorting system through the software hub platform, and the system enters the automatic operation state. An emergency stop button is also provided on the platform interface, which allows operators to stop the operation of the automatic ore receiving system at any time in an emergency, thereby improving the safety and reliability of the system. After the system is started, the diversion receiving plate 810 automatically moves to the ore belt demarcation point and starts normal mineral diversion and sorting work. During operation, the system continuously monitors the changes in the ore belt demarcation point, and the diversion receiving plate 810 will automatically adjust according to the latest identified demarcation point position to maintain sorting accuracy and stability. In addition, operators can monitor the operating status of the system in real time through the software hub platform, view the dynamic performance of the system at any time, and access and view the operation log after the system is turned on, providing detailed data support for daily operations and subsequent maintenance.

[0049] like Fig.14 As shown in the figure, the operation process of the intelligent shaking table diversion and ore sorting system is as follows: First, the system enters from "Start" and performs system initialization operations. The initialization process includes checking the connection status of the equipment camera, driver, and ore / water flow valve to ensure that all equipment is connected normally before entering the next step of the operation process. When the camera is connected well, the system executes the data acquisition module to collect shaking table image data from the camera and perform data annotation. Through the training and evaluation of the deep learning model of the annotated data, the system generates a deep learning model suitable for the current operation. Next, the system reads the image information of each shaking table group, and uses the deep learning model to determine the ore sorting position of each shaking table to ensure that the diversion and ore sorting device is positioned in the optimal sorting area.

[0050] During the operation of each shaking table ore receiving and sorting device, the system will continuously monitor the equipment status and the range of the ore belt. When it is detected that the position of the sorting device exceeds the set ore receiving range (that is, an abnormal situation occurs), the system will immediately adjust the ore and water flow rates to adjust the ore flow and water flow to ensure the stability of the ore belt distribution, thereby maintaining the sorting accuracy and efficiency. If the ore receiving and sorting device does not exceed the range, the system will operate normally and continue to move the sorting device to achieve dynamic sorting. Finally, when the system determines that the exit conditions are met or receives an exit instruction, the program will end the operation and exit the process. Through this closed-loop automatic control, the system effectively improves the intelligence level of ore dressing operations and reduces the burden of manual operations.

[0051] In summary, the present application discloses an intelligent shaking table diversion and ore sorting system, a sorting method and a software hub platform. The system is composed of a high-performance server 1, a touch screen 2, a data switch 3, a central control cabinet 4, a network camera 81 and a diversion and ore receiving plate 810, etc., to build a highly integrated intelligent sorting control system. The system divides the operation of the intelligent shaking table into a preparation layer, a control layer and an application layer, and realizes the full process control from image acquisition, data processing, dynamic recognition to the position adjustment of the receiving plate through the software hub platform. The whole system uses a deep learning model to identify the boundary point of the ore belt and automatically adjust the diversion and ore receiving plate 810, which improves the sorting accuracy and automation level of the system. The system improves the recognition accuracy of the ore belt image and ensures the accuracy of sorting through the improved YOLOv7-Segmentation model and CBAM attention mechanism; through the "one control multiple" intelligent control center, the system can control the work of multiple shaking tables at the same time, realize efficient management of operations, and improve response speed; the designed image preprocessing and compression algorithm significantly improves the algorithm's reasoning speed while maintaining recognition accuracy. The real-time ore belt demarcation point recognition algorithm based on deep learning can dynamically identify and adjust the ore belt demarcation point to achieve accurate mineral diversion and sorting effects. The system adopts a distributed architecture to centralize the image data of multiple shaking tables to the high-performance server 1 through the data switch 3 for processing, thereby achieving efficient data integration and processing. The automated diversion receiving plate 810 automatically adjusts its position according to the instructions of the control layer, making the diversion and sorting of the ore belt more accurate. The software hub platform provides a comprehensive software process framework, integrates all necessary functions, provides an intuitive user interface and real-time monitoring functions, and operators can view the system status and record operation logs at any time, which is convenient for troubleshooting and system optimization. The present invention effectively improves the efficiency and accuracy of mineral sorting, while reducing dependence on manual operation, and provides innovative solutions for the intelligence and automation of mine sorting technology.

[0052] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Under the concept of the present invention, the technical features in the above embodiments or different embodiments may also be combined, the steps may be implemented in any order, and there are many other changes in different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity. Although the present invention has been described in detail with reference to the above embodiments, a person of ordinary skill in the art should understand that the technical solutions described in the above embodiments may still be modified, or some of the technical features may be replaced by equivalents. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. An intelligent shaking table diversion and ore sorting system, characterized in that: The invention comprises a high-performance server (1), a touch screen (2), a data switch (3), a central control cabinet (4), a first Gigabit Ethernet line (5), a second Gigabit Ethernet line (6), a shaking table group (7), a diversion and ore separation device (8), a shaking table water supply control group (9) and a shaking table ore supply control group (10), wherein the high-performance server (1) is used to process the collected image data and generate control commands, the touch screen (2) is directly connected to the high-performance server (1) to provide a user operation and monitoring interface, the data switch (3) is connected to the high-performance server (1) to aggregate image data from multiple shaking tables, the central control cabinet (4) is used to integrate and install the high-performance server (1), the touch screen (2) and the data switch (3), the shaking table group (7) comprises multiple shaking tables, and the shaking table group (7) comprises an upstream shaking table group (701), a midstream shaking table group (702) and a downstream shaking table group (703). ), the downstream of the shaking table group (703), the midstream of the shaking table group (702) is the next ore dressing process stage of the upstream of the shaking table group (701), and the downstream of the shaking table group (703) is the next ore dressing process stage of the midstream of the shaking table group (702), wherein each shaking table is used for processing mineral sorting, and each shaking table is equipped with a diversion and ore sorting device (8), the diversion and ore sorting device (8) is connected to the data switch (3) through a first Gigabit Ethernet line (5) to transmit image data to the high-performance server (1), and transmits control instructions to the high-performance server (1) through a second Gigabit Ethernet line (6), and the high-performance server (1) controls the shaking table water supply control group (9) and the shaking table ore feeding control group (10) respectively through a signal bus according to the sorting situation of the diversion and ore sorting device (8), so as to realize closed-loop automatic control of automatic ore receiving, automatic water supply and automatic ore feeding of the shaking tables in each ore dressing process section.

2. The intelligent shaking table diversion and ore sorting system according to claim 1 is characterized in that: Each shaking table in the shaking table group (7) comprises a shaking table surface located at the top, a shaking table ore chute (72) located at the edge of the shaking table surface (71), and a shaking table fence (73) located at the edge of the shaking table ore chute (72).

3. The intelligent shaking table diversion and ore separation system according to claim 2 is characterized in that: The diversion and ore sorting device (8) is installed on the shaking table fence (73), and includes a network camera (81) located directly above the shaking table surface (71), connected to the device support frame (83) through an extendable camera bracket (82), and used to collect image data; the drive and controller (84) is located inside the diversion and ore sorting device (8), and is connected to the stepper motor (86) to receive control instructions from the high-performance server (1) and drive the stepper motor (86) to rotate; the control button panel (85) is connected to the drive and controller (84) and is used to manually control the rotation of the stepper motor (86); the stepper motor (86) is connected to the linear belt module (87) to drive the linear module slider (88) to move; the front end of the calibration plate (89) is connected to the linear module slider (88), and the rear end is connected to the diversion and ore receiving plate (810), so as to realize the left and right linear movement of the diversion and ore receiving plate (810) under the control of the control button panel (85) or the high-performance server (1).

4. An intelligent shaking table diversion and ore sorting method based on the intelligent shaking table diversion and ore sorting system according to any one of claims 1 to 3, characterized in that: include: Step S1, image preprocessing, compressing the image of the shaking table collected by the network camera (81) to keep the image aspect ratio consistent, and filling the edge area to form a standardized image size; Step S2: constructing a deep learning algorithm model, including the collection and labeling of image samples, model design, and model training; Step S3, model prediction and recognition, inputting the image data of non-training samples into the trained model, using the model weight parameters to identify the boundary points of the mineral belt, generating a segmentation mask to distinguish the various areas of the mineral belt, and the prediction output includes a category label, a bounding box and a pixel-level mask; Step S4, post-processing of ore belt position identification, accurately identifying the boundary point between the concentrate belt and the medium ore belt by analyzing the segmentation mask output by the model, and determining the boundary point as the target position of the guide receiving plate (810); Step S5, dynamically adjusting the guide receiving plate (810), generating a control command according to the real-time identified ore zone boundary point information, and dynamically adjusting the position of the guide receiving plate (810) to align it with the boundary point, so as to ensure accurate guide and separation of the minerals.

5. The intelligent shaking table diversion and ore separation method according to claim 4 is characterized in that: Step S1 includes: S11, input original image; S12, determining the target size; S13, calculating the proportional scaling factor and comparing the sizes; S14, applying the scaling factor to scale the source image in equal proportion; S15, calculating the scaled image size; S16, copying the scaled image to the center position of the target image array to form a new image; S17, filling the edge of the new image array; S18. Output the filled image.

6. The intelligent shaking table diversion and ore separation method according to claim 4 is characterized in that: Step S2 is completed using a high-performance server (1), and includes: S21. Image sample collection and labeling: for a single shaker, image sample data is collected periodically, with the number of samples being no less than 1,000, and regional labeling is performed according to the distribution of ore zones in the samples. The labeling areas are defined as background, concentrate area, intermediate ore area, and tailings area, to form a labeled image data set; S22. Model design, using the improved YOLOv7-Segmentation model, based on multi-scale feature fusion, path aggregation network and fast spatial pyramid pooling technology, combined with the EIoU loss function, to improve the detection and segmentation speed and accuracy of large targets in mining belts and small targets at the marking position of moving devices; S23, model training, input the labeled image sample data set into the designed algorithm model, so that the model can learn the classification and segmentation features of each area. The trained model can generate the final prediction results with category labels, bounding boxes and pixel-level masks, thereby realizing the accurate identification of the boundary points of the mineral belts.

7. The intelligent shaking table diversion and ore separation method according to claim 6 is characterized in that: In step S22, the improved YOLOv7-Segmentation model is that the input stage is used to receive an image with a size of 512x512x3, and includes a backbone network C3 module, which includes: a first convolution layer, configured with 32 filters, a convolution kernel size of 3x3, a step size of 1, a padding of 1, and an output size of 512x512x32 after batch normalization and ReLU activation function processing; a second convolution layer, configured with 64 filters, a convolution kernel size of 3x3, a step size of 1, a padding of 1, and an output size of 512x512x64 after batch normalization and ReLU activation function processing; and a c3k parameter is set to False in the shallow network to form a C2f structure similar to that in YOLOv8.

8. The intelligent shaking table diversion and ore separation method according to claim 7 is characterized in that: In step S22, the improved YOLOv7-Segmentation model includes a feature fusion part, in which the C3 module outputs four feature maps of different scales, with sizes of 512x512x64, 256x256x128, 128x128x256 and 64x64x512 respectively, and further fuses the feature maps of different scales through the feature pyramid network FPN and the path aggregation network PAN, and the output size remains consistent; the multi-head attention mechanism C2PSA is embedded in the C2 mechanism to enhance the spatial perception ability of the feature map; the multi-scale feature map is detected in the P2 detection layer, and the The 1x1 convolution dimension reduction is achieved through the fast spatial pyramid pooling SPPF module, and MaxPool2d with different convolution kernel sizes is used for multiple pooling. The output feature map sizes are 512x512x64, 256x256x128, 128x128x256 and 64x64x512 after the results are connected. The detection head part includes the depthwise separable convolution DWConv, and an adaptive anchor box mechanism is configured to optimize the anchor box configuration. The EIoU loss function is introduced to improve the prediction accuracy by considering the overlapping area, aspect ratio and center point offset of the predicted box and the true box, and finally outputs the prediction results of the category label, bounding box and pixel-level mask.

9. A software central platform for executing the intelligent shaking table diversion and ore separation method according to any one of claims 4 to 8, characterized in that: The tasks of the software central platform are divided into a main thread, an image acquisition thread, a real-time display thread and a logic control thread, wherein the main thread runs the main interface of the software central platform, the main interface includes a parameter setting module, a monitoring module, a user operation module and a communication module, and the main thread starts the other three threads through the user operation module; the image acquisition thread is set to read the image data collected by the network camera (81) at a fixed frequency, and transmit the read image to the data queue; The real-time display thread reads image data from the data queue at a fixed frequency, and displays the image in the main interface in the form of video on the one hand, and sends the latest image data to the logic control thread on the other hand; the logic control thread includes a mineral belt boundary point identification module, a logic control algorithm core and a command generation module. The logic control thread generates control commands based on the boundary point information in the mineral belt image, and continuously monitors the changes in the mineral belt boundary points to dynamically adjust the system control parameters.

10. The software hub platform according to claim 9, characterized in that: The software central platform also includes configuration and monitoring functions, which support the operator to complete the setting of the target offset value of the diversion receiving plate (810) position in the parameter setting window, and to stop the operation of the diversion receiving and sorting device (8) at any time through the emergency stop button in the control button panel (85). After operation, the position of the diversion receiving plate (810) is automatically adjusted to align with the boundary point of the ore belt, and the changes of the boundary point are continuously monitored. The operator can monitor the system status in real time and view the operation log.

Images