Device and method for dynamically detecting visual concentration of slime water based on neural network algorithm

Abstract

The invention relates to the technical field of coal preparation industry, and particularly discloses a coal slime water visual concentration dynamic detection device and method based on a neural network algorithm, the device comprises a concentration tank, a fixed frame is installed above the concentration tank, the bottom of the fixed frame is immersed in the concentration tank, and an underwater camera and an underwater light source are installed; the underwater lamp source is in electric signal connection with data processing equipment, the data processing equipment further comprises a data processing and output module, the underwater camera is used for collecting real-time videos of slime water in the concentration tank, the underwater lamp source is used for providing illumination to optimize the collection quality of the real-time videos, and the data processing equipment is used for receiving and processing the real-time videos. The slime water concentration is recognized through a neural network model, and the data processing and output module is used for sorting concentration data and outputting the concentration data to a monitoring system. According to the method, the neural network algorithm is effectively combined to analyze the video stream in real time, the slime water concentration can be accurately identified and monitored, the detection efficiency is effectively improved, and the production economic benefits are improved.

Description

Technical Field

[0001] The present invention relates to the technical field of coal preparation industry, and particularly to a visual concentration dynamic detection device and method for coal slime water based on neural network algorithm. Background Art

[0002] With the promotion of green environmental protection and the introduction of intelligent control equipment in the coal mining industry, the dynamic detection of coal slime water concentration is crucial in the coal washing process. Especially the detection of coal slime water concentration in the clarification layer of the thickener directly affects the treatment efficiency of coal slime water and the quality of the final product.

[0003] Currently, traditional detection methods usually rely on physical or chemical means, such as manual experience, turbidimeters, interface meters, etc. The detection process is cumbersome, costly, and it is difficult to achieve real-time online monitoring. With the rapid development of artificial intelligence technology, the application of deep learning algorithms in the field of visual detection provides a new solution for the detection of coal slime water concentration.

[0004] Therefore, to solve such problems, we propose a visual concentration dynamic detection device and method for coal slime water based on neural network algorithm. Summary of the Invention

[0005] The purpose of the present invention is to solve the disadvantages existing in the prior art, and to propose a visual concentration dynamic detection device for coal slime water based on neural network algorithm.

[0006] To achieve the above purpose, the present invention adopts the following technical solutions:

[0007] A visual concentration dynamic detection device for coal slime water based on neural network algorithm, including a thickener. A fixed frame is installed above the thickener, the bottom of the fixed frame is immersed in the thickener, and an underwater camera and an underwater light source are installed. The underwater light source is electrically connected to a data processing device, and the data processing device also includes a data processing and output module;

[0008] The underwater camera is used to collect real-time videos of coal slime water in the thickener;

[0009] The underwater light source is used to provide light to optimize the acquisition quality of real-time videos;

[0010] The data processing device is used to receive and process real-time videos, and identify the coal slime water concentration through a neural network model;

[0011] The data processing and output module is used to sort out the concentration data and output it to the monitoring system.

[0012] Preferably, the fixed frame is fixedly arranged through a support frame. The support frame is vertically fixed with two collar rings, and the two collar rings are respectively sleeved on both sides of the fixed frame. Fixing bolts are threadedly penetrated through both sides of the two collar rings, and the fixing bolts are in abutting fit with the fixed frame.

[0013] Preferably, the underwater camera and the underwater light source are arranged oppositely.

[0014] Preferably, the neural network model integrated in the server of the data processing device is a YOLOv5-based neural network model for video detection and concentration recognition.

[0015] Preferably, the YOLOv5 neural network model is a neural network model optimized based on YOLOv5. It is obtained by introducing an SE attention mechanism module on the basis of the original YOLOv5 framework. The SE attention mechanism module uses the Squeeze operation to compress the spatial dimension of each channel of the feature map into a scalar through global average pooling to obtain global information, uses the Excitation operation to generate the importance weights of each channel through two fully connected layers, and then uses the Scale operation to apply these weights to each channel of the original feature map to enhance important features.

[0016] A method for dynamically detecting the visual concentration of coal slime water based on a neural network algorithm, the method comprising the following steps:

[0017] Step 1: Collect a coal slime water data set, and use an underwater camera to continuously collect real-time videos of coal slime water in the concentration range to be recognized.

[0018] Step 2: Perform label classification, preprocess and label the collected coal slime water data set as the data set for identifying and training the model.

[0019] Step 3: Train the model, and use the data set to train the model in the recognition system program of the data processing device to obtain a trained YOLOv5 neural network model.

[0020] Step 4: Obtain recognition weights. After training the model, adjust the hyperparameters and model structure to obtain the optimal model weights.

[0021] Step 5: Log in to the recognition interface and log in to the detection interface in the data processing device.

[0022] Step 6: Import the weights into the recognition system. After logging in to the detection interface, import the optimal model weights of the model.

[0023] Step 7: Receive the camera data stream. After receiving the video stream of the underwater camera, use the trained YOLOv5 neural network model for real-time processing and analysis, extract the coal slime water concentration characteristics in the video, and obtain the analysis result.

[0024] Step 8: Output the detection result. The analysis result is sorted out by the data processing and output module, and the concentration data of the coal slime water is output to the monitoring system and displayed.

[0025] Compared with the prior art, the beneficial effects of the present invention are as follows:

[0026] 1. By using an underwater camera to monitor the clarification layer of the thickener, the present invention utilizes the characteristics of the light change of the light source passing through coal slime water with different concentrations, and combines the neural network algorithm to perform real-time analysis on the video stream, thereby realizing the accurate identification and monitoring of the concentration of coal slime water, being able to monitor the concentration change of coal slime water in real time, reducing the decline of production economic benefits caused by the delay of manual detection, and providing instant concentration data support.

[0027] 2. By working in cooperation with the data processing device, the present invention can efficiently and accurately complete the detection task of the concentration of coal slime water, can accurately identify the concentration of coal slime water in a complex environment, improve the detection efficiency, and can retrieve the pictures and detection situations in real time through the existing monitoring network in the factory, effectively reducing the instability of manual monitoring.

[0028] 3. The device used in the present invention is applicable to various types of coal slime water treatment scenarios and has a wide application prospect. Description of the Drawings

[0029] In order to more clearly illustrate the specific embodiments of the present invention or the technical solutions in the prior art, the following will briefly introduce the drawings required to be used in the description of the specific embodiments or the prior art. Obviously, the following drawings are some embodiments of the present invention. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0030] Figure 1 It is the overall structure schematic diagram of the present invention;

[0031] Figure 2 It is the functional schematic diagram of the detection interface of the present invention;

[0032] Figure 3 It is the working process schematic diagram of the server of the present invention.

[0033] Figure 4 It is the schematic diagram of the principle of YOLOv5_SE;

[0034] Figure 5 It is the schematic diagram of the principle of the SE attention mechanism module.

[0035] In the figure: 1. Thickener; 2. Fixed frame; 3. Underwater camera; 4. Underwater light source; 5. Data processing device; 6. Fixed bolt; 7. Light source controller. Detailed implementation manners

[0036] The following will clearly and completely describe the technical solutions in the embodiments of the present invention with reference to the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments.

[0037] Refer to Figure 1 、 Figure 2 、 Figure 3 A dynamic visual concentration detection device for coal slime water based on a neural network algorithm includes a thickener 1. A fixed frame 2 is installed above the thickener 1. The bottom of the fixed frame 2 is immersed in the thickener 1 and is provided with an underwater camera 3 and an underwater light source 4. The bottom of the fixed frame 2 is the monitoring area. The underwater camera 3 is installed in the monitoring area of the coal slime water treatment equipment and is responsible for collecting real-time videos during the flow of coal slime water. The underwater camera 3 is an existing waterproof camera with high resolution and good corrosion resistance, and can work stably for a long time in a complex environment. The underwater light source 4 is an LED light source. The underwater light source 4 cooperates with the underwater camera 3 to provide uniform and stable lighting conditions. The underwater light source 4 is electrically connected to a data processing device 5, and the data processing device 5 also includes a data processing and output module.

[0038] The fixed frame 2 can be fixedly arranged through a support frame, and the support frame is arranged on the side of the thickener 1. The height of the fixed frame 2 is 350 cm. The support frame is vertically fixed with two sleeve rings, and the two sleeve rings are respectively sleeved on both sides of the fixed frame 2. Fixing bolts 6 are threadedly penetrated through the sides of the two sleeve rings, and the fixing bolts 6 are in abutting match with the fixed frame 2. By rotating and loosening the fixing bolts 6, the fixed frame 2 can be adjusted up and down.

[0039] The underwater camera 3 is used to collect real-time videos of the coal slime water in the thickener 1;

[0040] The underwater light source 4 is used to provide light to optimize the acquisition quality of real-time videos;

[0041] The data processing device 5 is used to receive and process real-time videos, and identify the concentration of coal slime water through a neural network model;

[0042] The data processing and output module is used to sort out the concentration data and output it to the monitoring system, and can sort out and analyze the detected coal slime water concentration data, and output the results to the monitoring system or control system in real time.

[0043] The underwater camera 3 and the underwater light source 4 are arranged oppositely.

[0044] The underwater light source 4 is electrically connected to a light source controller 7, and the light source controller 7 can control the brightness of the underwater light source 4 to be adjusted according to the change in the concentration of coal slime water, that is, the brightness of the light source of the underwater light source 4 is adjustable and can be adjusted according to the change in the concentration of coal slime water, thereby optimizing the image acquisition effect.

[0045] The neural network model integrated in the server of the data processing device 5 is based on the YOLOv5 neural network model and is used for video detection and concentration recognition.

[0046] A method for dynamically detecting the visual concentration of coal slime water based on a neural network algorithm, the method includes the following steps:

[0047] Step 1: Collect a coal slime water data set, and use the underwater camera 3 to continuously collect real-time videos of coal slime water in the concentration range to be recognized;

[0048] Step 2: Perform label classification, preprocess and label the collected coal slime water data set as a data set for identifying and training the model;

[0049] Step 3: Train the model, use the data set to train the model in the recognition system program of the data processing device 5 to obtain a trained YOLOv5 neural network model;

[0050] Step 4: Obtain recognition weights, and after training the model, obtain the optimal model weights by adjusting hyperparameters and the model structure;

[0051] Step 5: Log in to the recognition interface and log in to the detection interface in the data processing device 5;

[0052] Step 6: Import the weights into the recognition system. After logging in to the detection interface, import the optimal model weights of the model;

[0053] Step 7: Receive the camera data stream. That is, the server of the data processing device 5 sends a SETUP request to the underwater camera 3 through the RTSP protocol to establish a video stream session. After receiving the SETUP request from the client, the underwater camera 3 allocates resources for the video stream and establishes an RTP channel. At this time, the client and the server communicate through the RTSP protocol to ensure the correct transmission of the stream. The RTSP assigns a session ID to the video stream and uses it in subsequent communications. The RTSP protocol is an application layer protocol specifically used for the transmission control of streaming media data. It works in conjunction with RTP (Real-time Transport Protocol) and RTCP (Real-time Control Protocol) to be responsible for the control and synchronization of the video stream. That is, after receiving the video stream from the underwater camera 3, the trained YOLOv5 neural network model is used for real-time processing and analysis to extract the characteristics of the concentration of coal slime water in the video and obtain the analysis results;

[0054] Step 8: Output the detection results. The analysis results are sorted out through the data processing and output module, and the concentration data of the coal slime water is output to the monitoring system and displayed.

[0055] The data processing device 5 is the core processing unit of this device. The installed YOLOv5 neural network model is a neural network model optimized based on YOLOv5 and is obtained by introducing the SE attention mechanism module on the basis of the original YOLOv5 framework. The data processing device 5 receives the video stream from the underwater camera 3, performs real-time analysis on the video through the model, and extracts the change characteristics of the light source passing through coal slime water with different concentrations. The data processing device 5 uses a neural network algorithm to detect the concentration according to the image change characteristics of the light source passing through the coal slime water and generates the corresponding concentration data.

[0056] The YOLOv5 network architecture (as Figure 4 shown) is used as the main detection model. The input image first passes through the backbone network, which contains a series of convolutional layers (P1, P2, P3, P4, P5) and CSPLayer layers. These layers gradually reduce the spatial size of the feature map and increase the number of channels, thereby effectively retaining the key information in the image and extracting important features. During this process, the model extracts more representative features layer by layer to ensure a more comprehensive understanding of the input image.

[0057] Subsequently, the feature map is further processed in the Neck part of the network. The Neck network first reduces the dimension of the feature map through the ReduceLayer operation layer and extracts more abstract features. Subsequently, the FPN (Feature Pyramid Network) structure (composed of Upsample, Concat, and TopDownLayer) is used to extract multi-level features from different levels of the backbone network and fuse high-level semantics and low-level spatial details layer by layer. Then, the PANet structure (composed of DownSample, Concat, and BottomUpLayer) aggregates the features output by the FPN from the bottom up again to ensure that the low-level features contain more high-level semantic information, further improving the accuracy and stability of detection.

[0058] After extracting and aggregating multi-level features, the network enters the Head part for the final detection task. The Head includes a series of Conv2D layers. Through continuous convolution operations, detailed spatial and channel information is extracted from the input feature map, and then the boundaries, positions, and categories of objects are accurately identified. On this basis, the Coupled Head layer combines the shared convolutional feature map and jointly performs classification and bounding box regression to ensure the accuracy and consistency of the output results. Finally, the Head generates output data suitable for the object detection task, including the bounding box coordinates, class labels, and confidence scores of the objects.

[0059] To enhance the feature attention of the model to illumination changes, as Figure 4 shown, in this study, an SE attention mechanism module is added after the BottomUpLayer layer of YOLOv5, an attention mechanism for enhancing feature expression ability. Figure 5 The schematic diagram of the SE attention mechanism module is shown. It can be seen from the figure that the SE attention mechanism module uses the Squeeze operation to compress the spatial dimension of each channel of the feature map into a scalar through global average pooling to obtain global information, uses the Excitation operation to generate the importance weights of each channel through two fully connected layers (FC), and then uses the Scale operation to apply these weights to each channel of the original feature map to enhance important features. Adding the SE attention mechanism module can improve the accuracy of the YOLOv5 model, especially for the object features under illumination changes.

[0060] The working principle is as follows:

[0061] As Figure 1 shown in the overall structure schematic diagram, through orthogonal experiment research, it is found that in the coal slurry water concentration range of 0 - 1000 mg / L, when the distance between the underwater camera 3 and the underwater light source 4 is maintained at about 10 cm, the image effect is the best. Therefore, the distance between the underwater camera 3 and the underwater light source 4 is set to 10 cm.

[0062] Without a recognition model, turn and loosen the two fixing bolts 6, insert the fixing frame 2 into the thickener 1. According to on-site investigation, it is more appropriate that the detection area of the clarification layer of the coal preparation plant thickener 1 is 100 cm below the liquid level. Therefore, when the underwater camera 3 and the underwater light source 4 are about 100 cm below the interface, fix the fixing frame 2;

[0063] Use the light source controller 7 to adjust the illumination intensity of the underwater light source 4 to ensure the visibility of the underwater camera 3 to the light source illumination within the concentration change range, so as to identify the concentration of coal slime water by using the illumination change characteristics of the light source passing through coal slime water with different concentrations;

[0064] Log in to the video acquisition platform in the data processing device 5, use the underwater camera 3 to continuously collect the real-time video of the coal slime water in the concentration range to be identified, preprocess the collected data and label it as the data set for the recognition model training;

[0065] Use the data set to train the model in the recognition system program of the data processing device 5, and obtain the optimal model weights by adjusting the hyperparameters and the model structure;

[0066] Log in to the detection interface in the data processing device 5, import the model recognition weights, and the functions of the detection interface are as shown in the schematic diagram Figure 2 As shown, in the detection interface, you can select the detection model, select the input data method (local file, local camera, rtsp interface data stream), and set some system parameters (IoU, confidence level, frame rate delay, saving of detection results). After receiving the video stream of the underwater camera 3 through the rtsp interface protocol, the detection system uses the trained YOLOv5 neural network model for real-time processing and analysis to extract the concentration characteristics of the coal slime water in the video;

[0067] The data processing and output module sorts out the analysis results and outputs the concentration data of the coal slime water to the monitoring system.

[0068] By using the underwater camera 3 to monitor the clarification layer of the thickener 1, utilizing the light change characteristics of the light source passing through coal slime water with different concentrations, and combining with the neural network algorithm to perform real-time analysis on the video stream, thereby achieving accurate identification and monitoring of the concentration of coal slime water, being able to monitor the concentration change of coal slime water in real time, reducing the decline in production economic benefits caused by the delay of manual detection, and providing instant concentration data support. Through the collaborative work of multiple components, it can efficiently and accurately complete the detection task of the concentration of coal slime water, accurately identify the concentration of coal slime water in a complex environment, improve the detection efficiency, and retrieve the picture and detection situation in real time through the existing monitoring network in the factory. And through the combination of machine vision and neural network, replacing the naked eye with a camera and the human brain with a machine, it reduces the instability of manual monitoring. At the same time, the equipment is applicable to various types of coal slime water treatment scenarios and has a wide application prospect.

[0069] The above is only a preferred specific embodiment of the present invention, but the protection scope of the present invention is not limited thereto. Any person skilled in the art within the technical scope disclosed by the present invention, according to the technical solution and inventive concept of the present invention, makes equivalent substitutions or changes, and should be covered within the protection scope of the present invention.

Claims

1. A device for dynamic detection of coal slurry water visual concentration based on a neural network algorithm, comprising a concentration tank (1), characterized in that: A fixed frame (2) is installed above the concentration tank (1), the bottom of the fixed frame (2) is immersed in the concentration tank (1), and an underwater camera (3) and an underwater light source (4) are installed. The underwater light source (4) is electrically connected to a data processing device (5), and the data processing device (5) also includes a data processing and output module; The underwater camera (3) is used to collect real-time video of the coal sludge water in the concentration tank (1); The underwater light source (4) is used to provide illumination to optimize the acquisition quality of real-time video; The data processing device (5) is used to receive and process real-time video and identify the concentration of coal slime water through a neural network model; The data processing and output module is used to organize the concentration data and output it to the monitoring system.

2. The device for dynamic detection of coal slime water visual concentration based on neural network algorithm according to claim 1 is characterized in that: The fixed frame (2) is fixedly arranged by a support frame, and two collars are vertically fixed to the support frame. The two collars are respectively sleeved on two side edges of the fixed frame (2), and the sides of the two collars are threaded with fixing bolts (6), and the fixing bolts (6) are abutted and matched with the fixed frame (2).

3. The device for dynamic detection of coal slime water visual concentration based on neural network algorithm according to claim 1 is characterized in that: The underwater camera (3) and the underwater light source (4) are arranged opposite to each other.

4. The device for dynamic detection of coal slime water visual concentration based on neural network algorithm according to claim 1 is characterized in that: The underwater light source (4) is electrically connected to a light source controller (7), and the light source controller (7) is capable of controlling the brightness of the underwater light source (4) to be adjusted according to changes in the concentration of coal slurry water.

5. The device for dynamic detection of coal slime water visual concentration based on neural network algorithm according to claim 1 is characterized in that: The neural network model integrated in the server of the data processing device (5) is based on the YOLOv5 neural network model and is used for video detection and concentration recognition.

6. The device for dynamic detection of coal slime water visual concentration based on neural network algorithm according to claim 5 is characterized in that: The YOLOv5 neural network model is a neural network model optimized based on YOLOv5, which is obtained by introducing the SE attention mechanism module on the basis of the original YOLOv5 framework. The SE attention mechanism module uses the Squeeze operation to compress the spatial dimension of each channel of the feature map into a scalar through global average pooling to obtain global information, and uses the Excitation operation to generate the importance weight of each channel through two fully connected layers, and then uses the Scale operation to apply these weights to each channel of the original feature map to enhance important features.

7. A method for dynamic detection of coal slurry water visual concentration based on a neural network algorithm, comprising a device for dynamic detection of coal slurry water visual concentration based on a neural network algorithm as claimed in claim 1, characterized in that: The method comprises the following steps: Step 1: Collect coal slurry water data sets, and use an underwater camera (3) to continuously collect real-time videos of coal slurry water in the concentration range to be identified; Step 2: Perform label classification, preprocess and label the collected coal slime water data set as the data set for the recognition training model; Step 3: training the model, using the data set to train the model in the recognition system program of the data processing device (5) to obtain a trained YOLOv5 neural network model; Step 4: Obtain recognition weights. After training the model, adjust hyperparameters and model structure to obtain the optimal model weights. Step 5: Log in to the identification interface and log in to the detection interface in the data processing device (5); Step 6: Import weights into the recognition system. After logging into the detection interface, import the optimal model weights of the model. Step 7: receiving the camera data stream, after receiving the video stream of the underwater camera (3), using the trained YOLOv5 neural network model to perform real-time processing and analysis, extracting the coal slime water concentration characteristics in the video, and obtaining the analysis results; Step 8: Output the test results, organize the analysis results through the data processing and output module, and output the concentration data of the coal slurry water to the monitoring system for display.

Images