System and device for real-time on-line detection of coarse separation of coal preparation plant

Abstract

The invention discloses a system and device for real-time on-line detection of separation coarse particles of a coal preparation plant, and the system comprises a rack, and also comprises a flattening assembly used for avoiding mutual stacking of coal briquettes; the anti-stacking assembly is used for layering the coal briquettes; the flattening assembly comprises a lead screw, a sliding sleeve, a connecting plate, a guide rod and a sliding block, the lead screw is rotationally connected with the rack, the sliding sleeve is in threaded connection with the lead screw, the sliding sleeve is in sliding connection with the rack, the connecting plate is fixedly connected with the sliding sleeve, the guide rod is fixedly connected with the connecting plate, the sliding block is in sliding connection with the guide rod, and the sliding block is in sliding connection with the guide rod. A pressing cylinder is rotationally connected to the sliding block and used for rolling on coal briquettes, and the screen is connected with a transmission assembly; according to the invention, the precision of rough recognition can be effectively improved.

Description

Technical Field

[0001] The invention belongs to the technical field of coal block sorting, and in particular relates to a system and a device for real-time online detection of coarse coal sorting in a coal preparation plant. Background Art

[0002] Coarse coal refers to the phenomenon that in the process of coal processing, the unqualified coarse particle size of the coal lump exceeds the specified range during the classification process. This phenomenon usually occurs in coal processing plants or coal preparation plants. Specifically, the unqualified coarse particle size in the graded product exceeds the specified range. Usually, an industrial camera is used to collect data and an algorithm model is used to identify and warn the coarse coal. However, during sampling, the coal blocks are mixed together, resulting in inaccurate identification. A structure that can avoid the mixed and stacked coal blocks is proposed. Summary of the invention

[0003] In view of the deficiencies in the prior art, the present invention provides a system and device for real-time online detection of coarse coal in coal preparation plants, which solves the above-mentioned problems.

[0004] To achieve the above objectives, the present invention is implemented through the following technical solutions: a system and device for real-time online detection of coarse coal in coal preparation plants, including a frame, and also including: a flattening component for preventing coal blocks from stacking on each other; an anti-stacking component for layering the coal blocks; the flattening component includes a screw rod, a sleeve, a connecting plate, a guide rod and a slider, the screw rod is rotatably connected to the frame, the sleeve is threadedly connected to the screw rod, the sleeve is slidably connected to the frame, the connecting plate is fixedly connected to the sleeve, the guide rod is fixedly connected to the connecting plate, the slider is slidably connected to the guide rod, the slider is rotatably connected to a pressure cylinder for rolling on the coal blocks, and the screen is connected to the transmission component.

[0005] On the basis of the above technical solution, the present invention also provides the following optional technical solution:

[0006] Further technical solution: a spring is sleeved on the guide rod, one end of the spring is fixedly connected to the slider, the other end of the spring is fixedly connected to the connecting plate, a frame is slidably connected to the frame, limiting rods are symmetrically fixedly connected on both sides of the frame, and a screen is slidably connected to the limiting rods.

[0007] Further technical solution: The transmission assembly includes a waterproof motor, a cam and a screw. Cams are provided on both sides of the screen. The two cams cyclically push the screen to slide. The cams are fixedly connected to the output shaft of the waterproof motor. The waterproof motor is fixedly connected to the frame, and the frame is threadedly connected to the screw.

[0008] Further technical solution: one end of the screw is rotatably connected to the frame, the other end of the screw is fixedly connected to the output shaft of motor A, the motor A is fixedly connected to the frame, one end of the screw is fixedly connected to the output shaft of motor B, and the motor B is fixedly connected to the frame.

[0009] Further technical solution: The anti-stacking assembly includes a shaft, a coal block rack, a gear A and a rack A. A shaft is rotatably connected between the two sliding sleeves. A damping rubber ring is provided at the connection between the shaft and the sliding sleeve. The coal block rack is fixedly connected to the shaft, the gear A is fixedly connected to the shaft, the rack A is fixedly connected to the frame, and the rack A is above the rack frame. The gear A is meshed with the rack A and is used to rotate the front end of the coal block rack downward to contact the coal blocks on the frame when the front end of the coal block rack is on the frame.

[0010] Further technical solution: The coal block rack is fixedly connected with an axis rod, and a push plate is fixedly connected with the axis rod. Both ends of the push plate for contacting the coal blocks are provided with soft rubber pads to avoid damaging the coal blocks. The axis rod is fixedly connected with a gear B, and the gear B is meshed and connected with the rack B after the coal block rack rotates downward, and the rack B is fixedly connected to the frame.

[0011] Beneficial Effects

[0012] The present invention provides a system and device for real-time online detection of coarse coal in coal preparation plants, which has the following beneficial effects compared with the prior art:

[0013] 1. When it is necessary to identify the particle size of coal blocks, the user starts motor A. At this time, the screw fixedly connected to the output shaft of motor A starts to rotate, and drives the frame threaded thereon, so that the frame starts to slide upward along the connection between it and the frame, so that the frame rises and leaves the tailings pond. During this process, the screen can intercept some coal blocks on it. When the screen rises above the water level of the tailings pond, the user starts to start the waterproof motor, so that the waterproof motor starts to drive the cam fixedly connected on its output shaft to rotate at a constant speed. At this time, the two cams start to contact the screen in turn, and in the process of contacting them, the screen is pushed to start sliding linearly along the limit rod, and contact with another cam under the action of inertia, so that the screen starts to perform circular linear motion on the screen under the cooperation of the two cams, so that the coal blocks on it slide back and forth under the action of inertia, and then the moisture on it is thrown off to avoid excessive moisture on it, which causes the coal blocks to stick to each other and affect the recognition accuracy of the industrial camera. At the same time, the coal blocks can also be preliminarily spread on the screen. When the frame rises to When the pressing cylinder is in contact with the screen, since the pressing cylinder is cylindrical, the sliding sleeve can make the pressing cylinder start to slide upward, so that the sliding blocks fixedly connected on both sides start to slide upward along the connection between the pressing cylinder and the guide rod, and compress the spring fixedly connected thereto, so that the pressing cylinder can slide into the screen, and after the pressing cylinder slides into the screen, the frame of the screen stops limiting the pressing cylinder, and the spring starts to rebound and reset, and pushes the pressing cylinder to press against the coal layer on the screen, and the screw rod keeps rotating in the process, so as to push the pressing cylinder to roll on the screw rod, and then flatten the coal blocks stacked on the screen, so that they can be evenly spread on the screen, so as to increase the recognition accuracy of the industrial camera for the coal block particle size;

[0014] 2. During the sliding of the sleeve, the coal rack between the two sleeves starts to slide synchronously and gradually moves to the top of the screen. In the initial state, the coal rack is parallel to the screen. When the front end of the coal rack slides to the top of the screen, the gear A moves synchronously to the rack A, so that the gear A starts to roll on it with the cooperation of the rack A meshing with it, and drives the shaft fixed at its axis to rotate synchronously. At this time, the coal rack starts to rotate downward and tilt so that it is in the screen. After the coal rack rotates downward, the gear A slides off the rack A, and the coal rack stops rotating. At the same time, since a damping rubber ring is set at the connection between the shaft and the sleeve, it can effectively increase the damping of its rotation to avoid it without gear A. It rotates automatically in cooperation with rack A. At this time, the front end of the shaft is in the screen and contacts the coal layer above it. At the same time, when the coal block rack rotates downward, gear B meshes with rack B synchronously, so that when the coal block rack slides, gear B rolls on rack B meshed with it, and drives the shaft fixed at its axis to start sliding, so that gear B fixed on the shaft starts to rotate. When gear B contacts the stacked coal blocks, the upper coal blocks can be swept into the coal block rack for storage, further avoiding interference with the industrial camera recognition. At the same time, after the coal block rack is reset to the initial position, the coal blocks on it can be identified by the identification industrial camera, avoiding a reduction in the number of samples and affecting the results.

[0015] 3. When gear A rotates in cooperation with rack A meshing with it, since the contact parts between it and the pad are arc-shaped, it can push the shaft downward, so that the pad starts to move linearly at the connection with the guide rod to compress the return spring fixedly connected with it, so that after gear A stops rotating, it can be re-engaged in the teeth of gear A through the positioning block, thereby limiting the position of gear A and preventing the coal rack from rotating automatically due to the increase of weight caused by the coal placed on it during the sliding process of the coal rack; BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 It is a schematic diagram of the three-dimensional structure of the present invention.

[0017] Figure 2 It is an enlarged schematic diagram of the transmission structure of the present invention.

[0018] Figure 3 It is a side view structural schematic diagram of the present invention.

[0019] Figure 4 It is an enlarged schematic diagram of the gear structure of the present invention.

[0020] Figure 5 Pave the structural schematic diagram of the present invention.

[0021] Figure 6 It is an enlarged schematic diagram of the positioning structure of the present invention.

[0022] Figure 7 This is a framework diagram of the attention mechanism of a system and device for real-time online detection of coarse separation in a coal preparation plant described in the present invention.

[0023] Figure 8 The present invention is a schematic diagram of a system and device for real-time online detection of coarse coal separation in a coal preparation plant.

[0024] Fig. 9 It is a structural schematic diagram of the middle layer of the improved YOLO network model structure of a system and device for real-time online detection of coarse separation in a coal preparation plant described in the present invention.

[0025] Fig.10 It is a schematic diagram of the SE attention mechanism of a system and device for real-time online detection of coarse separation in a coal preparation plant described in the present invention.

[0026] Fig.11 A flow chart of the image processing method of a system and device for real-time online detection of coarse coal in a coal preparation plant according to the present invention.

[0027] Notes on figure marks: frame 101, flattening component 2, transmission component 3, anti-stacking component 4, screw 201, sleeve 202, connecting plate 203, guide rod 204, slider 205, spring 206, pressing cylinder 207, frame 208, limit rod 209, screen 2001, waterproof motor 301, cam 302, screw 303, motor A304, motor B305, shaft 401, coal block rack 402, gear A403, rack A404, push plate 405, shaft 406, gear B407, rack B408, positioning block 409, pad 4001, guide rod 4002, return spring 4003. DETAILED DESCRIPTION

[0028] In order to make the purpose, technical solution and advantages of the present invention more clearly understood, the present invention is further described in detail below in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.

[0029] The specific implementation of the present invention is described in detail below in conjunction with specific embodiments.

[0030] Embodiment 1:

[0031] See also Figures 1 to 6 , provided in one embodiment of the present invention, is a system and device for real-time online detection of coarse separation in a coal preparation plant, including a frame 101, and also including:

[0032] A flattening assembly 2 is used to prevent coal blocks from stacking on each other;

[0033] An anti-stacking component 4, used for layering coal blocks;

[0034] The flattening assembly 2 includes a screw rod 201, a sleeve 202, a connecting plate 203, a guide rod 204 and a slider 205. The screw rod 201 is rotatably connected to the frame 101, the sleeve 202 is threadedly connected to the screw rod 201, the sleeve 202 is slidably connected to the frame 101, the connecting plate 203 is fixedly connected to the sleeve 202, the guide rod 204 is fixedly connected to the connecting plate 203, the slider 205 is slidably connected to the guide rod 204, and a pressure cylinder 207 is rotatably connected to the slider 205 for rolling on the coal blocks. The screen 2001 is connected to the transmission assembly 3.

[0035] Specifically, a spring 206 is sleeved on the guide rod 204, one end of the spring 206 is fixedly connected to the slider 205, and the other end of the spring 206 is fixedly connected to the connecting plate 203. A frame 208 is slidably connected to the frame 101, and limiting rods 209 are symmetrically fixedly connected on both sides of the frame 208. A screen 2001 is slidably connected to the limiting rods 209, and a blackout cloth for avoiding interference from external light sources is provided on the screen 2001, and a light source for fill light is provided in the blackout cloth.

[0036] Specifically, the transmission assembly 3 includes a waterproof motor 301, a cam 302 and a screw 303. Cams 302 are respectively provided on both sides of the screen 2001. The two cams 302 cyclically push the screen 2001 to slide. The cam 302 is fixedly connected to the output shaft of the waterproof motor 301. The waterproof motor 301 is fixedly connected to the frame 208, and the frame 208 is threadedly connected to the screw 303.

[0037] Specifically, one end of the screw rod 303 is rotatably connected to the frame 101, the other end of the screw rod 303 is fixedly connected to the output shaft of the motor A304, the motor A304 is fixedly connected to the frame 101, one end of the screw rod 201 is fixedly connected to the output shaft of the motor B305, and the motor B305 is fixedly connected to the frame 101.

[0038] In the embodiment of the present invention, when it is necessary to remove the coal particles in the tailings pond water through module 4 (screen), the user starts the motor A304, and the screw 303 fixedly connected to the output shaft of the motor A304 starts to rotate, and drives the frame 208 threadedly connected thereon, so that the frame 208 starts to slide upward along the connection between it and the frame 101, so that the frame 208 rises and leaves the tailings pond, and the screen 2001 can intercept some coal blocks on it during this process. When the screen 2001 rises above the water level of the tailings pond, the user starts to start the waterproof motor 301, so that the waterproof motor 301 starts to drive the screw 303 fixedly connected on its output shaft. The cam 302 rotates at a constant speed, and then the two cams 302 begin to contact the screen 2001 in turn, and in the process of contacting with it, push the screen 2001 to start sliding linearly along the limit rod 209, and contact with another cam 302 under the action of inertia, so that the screen 2001 starts to perform circular linear motion on the screen 2001 under the cooperation of the two cams 302, so that the coal blocks on it slide back and forth under the action of inertia, and then the moisture on it is thrown off, so as to avoid excessive moisture on it, causing the coal blocks to stick to each other and affect the recognition accuracy of the industrial camera. At the same time, the coal blocks can also be initially spread on the screen 2001. When the frame 208 rises to the rack 101 When the pressing cylinder 207 is in contact with the screen 2001, since the pressing cylinder 207 is cylindrical, the pressing cylinder 207 can start to slide upward when the sliding sleeve 202 continues to move linearly, so that the sliders 205 fixedly connected on both sides of the pressing cylinder 207 start to move along the connection between the pressing cylinder 207 and the guide rod 2001. 4, and compresses the spring 206 fixedly connected thereto, so that the pressing cylinder 207 can slide into the screen 2001. After the pressing cylinder 207 slides into the screen 2001, the frame of the screen 2001 stops limiting the pressing cylinder 207. At this time, the spring 206 starts to rebound and reset, and pushes the pressing cylinder 207 to press against the coal layer on the screen 2001. At the same time, the screw rod 201 keeps rotating during this process, thereby pushing the pressing cylinder 207 to roll on the screw rod 201, and then flattening the coal blocks stacked on the screen 2001, so that they can be evenly spread on the screen 2001, so as to increase the recognition accuracy of the industrial camera for the coal block particle size.

[0039] Specifically, the anti-stacking component 4 includes a shaft 401, a coal block rack 402, a gear A403 and a rack A404. The shaft 401 is rotatably connected between the two sleeves 202. A damping rubber ring is provided at the connection between the shaft 401 and the sleeve 202. The coal block rack 402 is fixedly connected to the shaft 401. The gear A403 is fixedly connected to the shaft 401. The rack A404 is fixedly connected to the frame 101, and the rack A404 is above the rack frame 208. The gear A403 is meshed with the rack A404, and is used to rotate the front end of the coal block rack 402 downward to contact the coal blocks on the frame 208 when the front end of the coal block rack 402 is on the frame 208.

[0040] Specifically, the coal block rack 402 is fixedly connected with an axis rod 406, and the axis rod 406 is fixedly connected with a push plate 405. Both ends of the push plate 405 for contacting the coal blocks are provided with soft rubber pads to avoid damaging the coal blocks. The axis rod 406 is fixedly connected with a gear B407, and the gear B407 is meshed and connected with a rack B408 after the coal block rack 402 rotates downward, and the rack B408 is fixedly connected to the frame 208.

[0041] In the embodiment of the present invention, during the sliding of the sleeve 202, the coal block rack 402 between the two sleeves 202 begins to slide synchronously and gradually moves to above the screen 2001. In the initial state, the coal block rack 402 is parallel to the screen 2001. When the front end of the coal block rack 402 slides to above the screen 2001, the gear A403 moves synchronously to the rack A404, so that the gear A403 begins to roll on it with the cooperation of the rack A404 meshing with it, and drives the shaft 401 fixedly connected at its axis to rotate synchronously. At this time, the coal block rack 402 begins to rotate and tilt downward so that it is in the screen 2001. After the coal block rack 402 rotates downward by 10°, the gear A403 slides off the rack A404. At this time, the coal block rack 402 stops rotating. At the same time, since a damping rubber ring is provided at the connection between the shaft 401 and the sleeve 202, its rotation can be effectively increased. The damping of the movement is prevented from rotating on its own without the cooperation of gear A403 and rack A404. At this time, the front end of shaft 401 is in screen 2001 and contacts the coal layer above it. At the same time, when coal block rack 402 rotates downward, gear B407 synchronously meshes with rack B408, so that when coal block rack 402 slides, gear B407 rolls on rack B408 meshing with it, and drives shaft 401 fixedly connected at its axis to start sliding, so that gear B407 fixedly connected to shaft 401 starts to rotate, so that when gear B407 contacts the stacked coal blocks, the upper layer of coal blocks can be swept into coal block rack 402 for storage, further avoiding interference with industrial camera recognition. At the same time, after coal block rack 402 is reset to its initial position, the coal blocks on it can be identified by identifying the industrial camera, avoiding a reduction in the number of samples and affecting the results.

[0042] Specifically, a positioning block 409 is provided under any of the gears A403, and the positioning block 409 is used to limit the gear A403. The positioning block 409 is fixedly connected to the pad 4001, and the two sides of the pad 4001 are respectively slidably connected to the guide rod 4002, and the guide rod 4002 is fixedly connected to the pad 4001. A return spring 4003 is sleeved on the guide rod 4002, and one end of the return spring 4003 is fixedly connected to the pad 4001, and the other end of the pad 4001 is fixedly connected to the coal block rack 402.

[0043] In the embodiment of the present invention, when the gear A403 rotates in cooperation with the rack A404 meshing with it, since the contact parts between it and the pad 4001 are arc-shaped, the shaft 401 can be pushed downward, so that the pad 4001 starts to move linearly at the connection between it and the guide rod 4002, so as to compress the reset spring 4003 fixedly connected thereto, so that after the gear A403 stops rotating, it can be re-engaged in the teeth of the gear A403 through the positioning block 409, thereby limiting the gear A403 and preventing the coal block rack 402 from rotating on its own due to the increase in weight caused by the coal blocks placed on it during the sliding process of the coal block rack 402.

[0044] Embodiment 2:

[0045] A system and device for real-time online detection of coarse coal in coal preparation plant, comprising: a data processing method of an image processing model, an algorithm framework of the image processing model, an image processing method and an image processing device

[0046] A data processing method for an image processing model, the method comprising:

[0047] Based on the model to be trained, data enhancement processing is performed on the initial image data, wherein the data enhancement methods include improved Gamma change data enhancement, filtering data enhancement, scaling data enhancement, flipping data enhancement and arbitrary angle rotation data enhancement to obtain enhanced image data. The purpose of the data enhancement processing is to expand the data and improve the generalization ability of the model.

[0048] Preferably, the fixed range (such as 0.5-1.5) of the dynamic / adaptive gamma range in the gamma change data enhancement may not be applicable to all scenes, and the gamma value can be dynamically adjusted based on the image brightness distribution, and different gamma values ​​can be applied to different RGB channels.

[0049] Preferably, filtering type data enhancement is used. Since the data set collected by the photographic equipment has a lot of background interference, the present application adjusts the filtering intensity according to the image content, uses the sigma of Gaussian blur, and introduces complex filtering such as motion blur and fogging.

[0050] Preferably, scaling data augmentation uses padding instead of stretching after scaling to avoid image deformation.

[0051] Preferably, arbitrary angle rotation data enhancement is changed to use GAN to generate filling content instead of fixed color, so as to avoid the situation where the enhanced image data is not applicable, and to generate rotated rectangular annotations for small target detection tasks.

[0052] The main problem of small sample learning is that the sample size is too small, resulting in insufficient sample diversity to characterize the complete sample distribution. Sample enhancement can be used to improve sample diversity. The method based on data enhancement is to use auxiliary data sets or auxiliary information to enhance the target data set for data expansion or feature enhancement so that the model can fit better. Data expansion can be unlabeled or synthetic labeled data; feature enhancement is to add features that are easy to classify in the feature space of the original sample to increase feature diversity; according to the image data enhancement, not only can the number of training samples be expanded, but also the diversity of training samples can be increased, which can avoid overfitting on the one hand and improve model performance on the other.

[0053] According to the data generated by the data enhancement, the model to be trained is adjusted to obtain a suitable training algorithm.

[0054] In a second aspect, the present application also provides an algorithm framework of an image processing model, the framework comprising:

[0055] Preferably, the FPN (Feature Pyramid Networks) network structure uses an FPN network combined with a conventional YOLOv5 native intermediate layer network PAN.

[0056] Preferably, the preliminary coal particle accumulation detection model constructs a feature extraction backbone network through a CSPDarknet53 network structure. The CSPDarknet53 network structure constructs a feature extraction backbone network for extracting features from an input image, and the feature extraction backbone network outputs four feature maps of different sizes: C1 (160x160), C2 (80x80), C3 (40x40), and C4 (20x20).

[0057] Preferably, the introduction of the SE attention mechanism effectively improves the feature expression capability and enhances the model's focus on important features, thereby improving the accuracy of the network, especially when processing complex scenes or coal particles.

[0058] Preferably, the EIoU Loss loss function is introduced, in which the Focal Loss function is added to solve the sample imbalance problem in BBox regression, which is mainly used to solve the category imbalance problem in target detection. It is used to increase the model's attention to difficult-to-classify samples (such as coal particles of different particle sizes), thereby improving detection accuracy. The formula of Focal Loss is as follows:

[0059]

[0060] Among them: pt is the predicted category probability, for positive samples it is the probability of being predicted as a positive category, for negative samples it is the probability of being predicted as a negative category.

[0061] Preferably, the preliminary coal particle accumulation detection model, the coal particle accumulation detection model and the complete coal particle accumulation detection model are arranged on a server on a network.

[0062] In a third aspect, the present application provides an image processing method, comprising:

[0063] Acquire image data to be enhanced;

[0064] The model to be trained is an algorithm model that needs to be trained. The model to be trained can be an initial algorithm model before training, or an algorithm model adjusted during the training process. The algorithm model to be trained is a neural network model, and the algorithm model to be trained can perform data enhancement processing based on the image data input to itself to adjust the parameters of the image data. Optionally, the algorithm model to be trained is an improved neural network model that can identify fine particles on the sieve in a complex background environment based on the image data input thereto.

[0065] A series of data enhancements are performed on the image data to be enhanced by training the algorithm model to obtain target image data. The training algorithm model is an improved algorithm based on the Yolov5 algorithm.

[0066] In a fourth aspect, the present application provides an image processing device, the device comprising:

[0067] like Figure 8 As shown in the overall structural diagram, module 7 (light source) provides uniform illumination for the device. It should be noted that the device is in a shade cloth, which is not shown in the diagram. The shade cloth is used to avoid interference from external light sources. When module 5 (flotation machine) is started, the tailings begin to be discharged, and some particles that do not float with the foam will be discharged from the tailings. These particles are often larger coal particles that cannot pass through the screen. When the coal particles accumulate on the screen, it will cause blockage. When the screen is lifted out of the surface of the coal slurry water, the distance between the industrial camera and the screen must be clear.

[0068] In the absence of an identification model, module 4 (screen) in the tailings treatment pool is used to block larger coal particles. When it is necessary to detect whether flotation is coarse, module 4 (screen) is slid up to above the water surface of the tailings pool. At the same time, module 2 (industrial camera) starts working, collecting real-time video data on the screen and transmitting it to module 1 (data processing equipment). Module 3 (gravity sensor) transmits real-time gravity data to module 1 (data processing equipment) to calculate the unit coarseness.

[0069] Log in to the video acquisition platform in module 1 (data processing equipment), use module 2 (industrial camera) to continuously collect real-time video of the accumulation of coal particles on the screen to be identified, perform image enhancement on the photos, manually annotate the coal particles in the enhanced photos, perform the above-mentioned improved image data enhancement on the annotated photos, and generate a training set. For all photos in the training set, this embodiment divides them into a test set and a validation set at a ratio of 8:2. The test set is used to input into the neural network model for model training, and the validation set is used to verify the output results of the training.

[0070] In the recognition system program of module 1 (data processing equipment), the model is trained using the data set, and the optimal model weight is obtained by adjusting the hyperparameters and model structure, and the image feature points are extracted for matching and restoring the camera shooting position and angle;

[0071] Log in to the detection interface in module 1 (data processing equipment) and import the model recognition weight. In the detection interface, you can select the detection model, select the data input method (local file, local camera, rtsp interface data stream), and select some system parameter settings (IoU, confidence, frame rate delay, and detection result preservation). After the detection system receives the industrial camera video stream through the rtsp interface protocol, it uses the trained neural network model for real-time processing and analysis to extract the data features of the coal particles on the screen in the video;

[0072] The data processing module organizes the analysis results and outputs the characteristic data of the coal particles on the screen to the monitoring system. If the system determines that the coal is coarse, it will transmit a signal to module 6 (alarm) to alert the workers to stop sorting or readjust the feed.

[0073] The training method of the image processing model provided in the embodiment of the present application can be applied to Figure 8 In the application environment shown in (device diagram), module 2 (industrial camera) communicates with module 1 (data processing device) through a network, and the data storage system can store data that the server (module 1) needs to process. The data storage system can be integrated on the server (module 1).

[0074] Based on the algorithm model to be trained, a series of data enhancement processes are performed on the initial image data to obtain diversified image data in order to deal with the monotony of small sample data training.

[0075] Preferably, image data enhancement is performed on the original image captured by the camera to increase the diversity of small sample data and enhance the neural network's ability to recognize targets of different scales.

[0076] Preferably, the image data enhancement can select a data enhancement method based on the actual image, and can select one or more data enhancement methods, and then select an existing small sample data set to generate enhanced images and labels, thereby increasing the diversity of the data set.

[0077] Preferably, the loss function is optimized: the EIoU (Efficient IoU) loss function is adopted, which optimizes the regression loss calculation method based on the traditional IoU, improves the accuracy of bounding box regression, and combines the Focal Loss function to solve the imbalance problem of positive and negative samples and enhance the model's detection ability for small targets.

[0078] Preferably, replacing the loss function is to replace the bbox_iou function in the metric.py section in the Yolov5 directory with the function required by EIoU, and reconfigure the loss.py function in the utils directory.

[0079] Preferably, the SE attention mechanism is introduced: the Squeeze-and-Excitation (SE) attention mechanism is added to the model structure to improve the network's ability to extract key information by adaptively adjusting the weights of feature channels.

[0080] Preferably, if Fig.10 As shown in the figure, the SE attention mechanism compresses the spatial dimension through the squeeze operation. Simply put, it performs global pooling on each feature map and averages it into a real value. Then the excitaton operation is performed. Since the network outputs a feature map of size 1*1*C after the squeeze operation, the attention mechanism uses weights to learn the direct correlation of C channels. In actual applications, some frameworks use full connection, and some frameworks use 1*1 convolution. In this process, the SE attention mechanism first reduces the dimension of C channels and then expands back to C channels. In the last operation, the output of exciation is regarded as the importance of each channel after feature selection, and is multiplied by the previous feature in a weighted multiplication manner to achieve the function of enhancing important features and suppressing unimportant features.

[0081] Preferably, if you want to introduce the SE attention mechanism, you need to configure a SE yaml file in the yolov5 directory, add the SE attention code to the end of the common.py program in the yolov5 directory, and add the SE class in the yolo.py program.

[0082] Preferably, the neural network structure is optimized: the FPN (feature pyramid network) + PAN (path aggregation network) structure is adopted to improve the multi-scale feature fusion capability so that the model has better robustness when detecting targets of different sizes.

[0083] Preferably, the model training and optimization uses the enhanced data to train the improved YOLOv5 algorithm, adopts an adaptive learning rate adjustment strategy, and verifies it in combination with the data set to ensure the generalization ability of the model in different scenarios.

[0084] Preferably, in actual applications, users can input real-time video streams or static images into the trained model, and the system can automatically output the recognition results of the target, and can calculate the particle size distribution and proportion of coal particles in combination with the post-processing module.

[0085] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0086] The fixed connection referred to in this application refers to a connection in which a part or component is fixed without any relative movement, which can be divided into detachable connection and non-detachable connection.

[0087] (1) Removable connection: The parts are fixed together by screws, splines, wedge pins, etc. This connection method can be disassembled for maintenance without damaging the parts. However, the specifications of the connectors used must be correct (such as the length of the bolts, keys, and wedge pins) and they must be properly tightened.

[0088] (2) Non-detachable connection: mainly refers to welding, riveting and tenoning. Since forging, sawing or oxygen cutting are required for disassembly during maintenance or replacement, spare parts generally cannot be reused. At the same time, attention should be paid to process quality, technical inspection and remedial measures (such as calibration, polishing, etc.) during connection.

[0089] The sliding connection referred to in the present application means that the component can slide along a linear trajectory, and the hinged connection referred to in the present application means that the component can rotate along an axial constraint.

[0090] In some cases, the sliding connection and hinge referred to in the present application may also be damped so that the components have the ability to maintain a desired position.

[0091] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. A real-time online detection device for coarse separation in a coal preparation plant, comprising a frame (101), characterized in that: Also includes: A flattening assembly (2) for preventing coal blocks from stacking on top of each other; An anti-stacking assembly (4) for layering coal blocks; The flattening assembly (2) comprises a screw rod (201), a sleeve (202), a connecting plate (203), a guide rod (204) and a slider (205); the screw rod (201) is rotatably connected to the frame (101); the sleeve (202) is threadedly connected to the screw rod (201); the sleeve (202) is slidably connected to the frame (101); the connecting plate (203) is fixedly connected to the sleeve (202); the guide rod (204) is fixedly connected to the connecting plate (203); the slider (205) is slidably connected to the guide rod (204); a pressing cylinder (207) is rotatably connected to the slider (205) for flattening coal. The screen (2001) is connected to the transmission assembly (3), the guide rod (204) is sleeved with a spring (206), one end of the spring (206) is fixedly connected to the slider (205), and the other end of the spring (206) is fixedly connected to the connecting plate (203), the frame (208) is slidably connected to the frame (101), the two sides of the frame (208) are symmetrically fixedly connected to limit rods (209), the limit rods (209) are slidably connected to the screen (2001), and the screen (2001) is provided with a shading cloth for avoiding interference from external light sources, and a light source for supplementary light is provided in the shading cloth.

2. The real-time online detection device for coarse separation in a coal preparation plant according to claim 1 is characterized in that: The transmission assembly (3) comprises a waterproof motor (301), a cam (302) and a screw (303). Cams (302) are respectively provided on both sides of the screen (2001). The two cams (302) cyclically push the screen (2001) to slide. The cams (302) are fixedly connected to the output shaft of the waterproof motor (301). The waterproof motor (301) is fixedly connected to the frame (208), and the frame (208) is threadedly connected to the screw (303).

3. The real-time online detection device for coarse separation in a coal preparation plant according to claim 1 is characterized in that: One end of the screw rod (303) is rotatably connected to the frame (101), the other end of the screw rod (303) is fixedly connected to the output shaft of the motor A (304), the motor A (304) is fixedly connected to the frame (101), one end of the screw rod (201) is fixedly connected to the output shaft of the motor B (305), the motor B (305) is fixedly connected to the frame (101).

4. The real-time online detection device for coarse separation in a coal preparation plant according to claim 1 is characterized in that: The anti-stacking assembly (4) comprises a shaft (401), a coal block frame (402), a gear A (403) and a rack A (404); the shaft (401) is rotatably connected between the two sliding sleeves (202); a damping rubber ring is provided at the connection between the shaft (401) and the sliding sleeve (202); the coal block frame (402) is fixedly connected to the shaft (401); the gear A (403) is fixedly connected to the shaft (401); the rack A (404) is fixedly connected to the frame (101); and the rack A (404) is located above the rack frame (208); the gear A (403) is meshedly connected to the rack A (404) and is used for rotating the front end of the coal block frame (402) downward to contact the coal blocks on the frame (208) when the front end of the coal block frame (402) is located on the frame (208).

5. The real-time online detection device for coarse separation in a coal preparation plant according to claim 4 is characterized in that: The coal block rack (402) is fixedly connected with an axle rod (406), and a push plate (405) is fixedly connected with the axle rod (406). Both ends of the push plate (405) for contacting the coal blocks are provided with soft rubber pads to avoid damaging the coal blocks. The axle rod (406) is fixedly connected with a gear B (407), and the gear B (407) is meshed and connected with a rack B (408) after the coal block rack (402) rotates downward. The rack B (408) is fixedly connected with the frame (208).

6. The real-time online detection device for coarse separation in a coal preparation plant according to claim 1 is characterized in that: A positioning block (409) is provided below any of the gears A (403), and the positioning block (409) is used to limit the gear A (403). The positioning block (409) is fixedly connected to the pad (4001), and the two sides of the pad (4001) are respectively slidably connected to the guide rod (4002), and the guide rod (4002) is fixedly connected to the pad (4001). A return spring (4003) is sleeved on the guide rod (4002), and one end of the return spring (4003) is fixedly connected to the pad (4001), and the other end of the pad (4001) is fixedly connected to the coal block rack (402).

7. A real-time online detection system for coarse coal separation in a coal preparation plant, characterized in that: The invention comprises a data processing method of an image processing model, an algorithm framework of the image processing model and an image processing method. The data processing method of the image processing model comprises performing data enhancement processing on initial image data based on a model to be trained to obtain enhanced image data.

8. The real-time online detection system for coarse separation in coal preparation plants according to claim 7 is characterized in that: The filtering data enhancement uses the sigma of Gaussian blur, the scaling data enhancement uses padding after scaling, and the arbitrary angle rotation data enhancement uses GAN to generate filling content to avoid the inapplicability of the enhanced image data, and generates rotated rectangular annotations for small target detection tasks.

9. The system for real-time online detection of coarse separation in coal preparation plants according to claim 7 is characterized in that: The algorithm framework of the image processing model includes an FPN network structure, the FPN network is combined with the conventional YOLOv5 native intermediate layer network PAN, and the preliminary coal particle accumulation detection model constructs a feature extraction backbone network through the CSPDarknet53 network structure. The CSPDarknet53 network structure constructs a feature extraction backbone network for extracting features from the input image. The feature extraction backbone network outputs four feature maps of different sizes, and introduces the SE attention mechanism to improve the feature expression capability.

10. The system for real-time online detection of coarse separation in coal preparation plants according to claim 7, characterized in that: The image processing method comprises acquiring image data to be enhanced, performing a series of data enhancements on the image data to be enhanced by training an algorithm model, and obtaining target image data. The training algorithm model is an improved algorithm based on the Yolov5 algorithm.

Images