Material particle size distribution and volume flow online detection device and detection method

Abstract

The invention belongs to the technical field of industrial detection, and particularly discloses a material particle size distribution and volume flow online detection device and method.The detection device comprises a door-shaped frame, and the door-shaped frame stretches across a conveying belt used for conveying materials to be detected; the detection box is detachably installed on the door-shaped frame and located over the conveying belt, a closed cavity is formed in the detection box, an industrial camera and a binocular structure light depth camera are installed in the detection box, and a perspective window is formed in the bottom of the detection box; and the image processing and analyzing system is connected with the industrial camera and the binocular structure light depth camera, and analyzes and obtains material particle size distribution and volume flow based on images obtained by the industrial camera and the binocular structure light depth camera at the same time. Through the structural design of the detection device, the industrial camera and the binocular structure light depth camera can monitor material particle size distribution and volume flow in real time, detection errors are reduced, and meanwhile detection efficiency and environmental adaptability are improved.

Description

Technical Field

[0001] This application belongs to the field of industrial testing technology, and more specifically, relates to an online detection device and method for material particle size distribution and volumetric flow rate. Background Technology

[0002] In industrial production such as coal-fired power plants, coal chemical plants, and mines, real-time and accurate acquisition of the particle size distribution and volumetric flow rate of materials on conveyor belts is a core element for achieving intelligent and digital management. Taking coal as an example, particle size distribution directly determines combustion efficiency, stable boiler operation, and overall energy consumption. Excessively large particle sizes or uneven distribution can lead to incomplete combustion, increased equipment wear, and higher dust emission risks. Meanwhile, accurate measurement of volumetric flow rate is related to the precise control of coal input to the furnace, optimization of transportation scheduling, coal quality control, and accuracy of trade settlement.

[0003] In related technologies, traditional particle size detection mainly relies on two methods: manual sieving and sedimentation analysis. Manual sieving requires operators to take samples periodically for laboratory analysis, which is time-consuming and labor-intensive. Furthermore, due to the limited sampling frequency, it is difficult to accurately reflect the dynamic changes of materials during continuous production. While sedimentation analysis can provide relatively accurate particle size distribution data, the detection cycle is too long, and the measurement accuracy for ultrafine particles is easily affected by factors such as ambient temperature and liquid viscosity. Regarding flow rate detection, weighing devices are currently the primary method. This method assumes a constant material density, but in actual production, material density fluctuates with factors such as moisture content and particle composition, leading to systematic errors in the measurement results. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this application provides an online detection device and method for material particle size distribution and volumetric flow rate, aiming to solve the problems of long detection time, poor real-time performance, and susceptibility to interference in traditional particle size detection methods; and systematic errors in flow rate detection caused by fluctuations in material density.

[0005] This application provides an online detection device for material particle size distribution and volumetric flow rate, specifically comprising: A portal frame spanning a conveyor belt used for conveying materials to be tested; The detection box is detachably installed on the portal frame and located directly above the conveyor belt. The interior of the detection box is a sealed cavity and is equipped with an industrial camera for detecting the particle size distribution of materials and a binocular structured light depth camera for detecting the flow rate and volume of materials. A viewing window is provided at the bottom of the detection box. An image processing and analysis system is connected to an industrial camera and a binocular structured light depth camera, and simultaneously analyzes the material particle size distribution and volumetric flow rate based on the images acquired by the industrial camera and the binocular structured light depth camera.

[0006] Compared with the prior art, the technical solution conceived in this application, through the integration of an industrial camera, a binocular structured light depth camera, a data transmission module, and an image processing and analysis system, enables the analysis of information from two-dimensional particle size distribution to three-dimensional volume, thereby more accurately detecting the particle size distribution and volumetric flow rate of materials. Simultaneously, the real-time detection by the industrial camera and the binocular structured light depth camera makes the detection more efficient. Furthermore, the binocular structured light depth camera directly measures the volumetric flow rate, independent of density, reducing systematic errors. Both the industrial camera and the binocular structured light depth camera are housed in a detection chamber, isolating them from dust and vibration, thus improving anti-interference capabilities. Therefore, this detection device achieves the beneficial effects of high efficiency, real-time performance, anti-interference capabilities, and more accurate detection results.

[0007] As a further preferred embodiment, a transparent baffle is provided at the viewing window, and a cleaning component for cleaning the outer surface of the transparent baffle is provided on the detection box.

[0008] As a further preferred embodiment, the cleaning assembly includes a motor, a connecting rod, and wiper blades. The motor is vertically fixed inside the testing box. One end of the connecting rod is coaxially fixedly connected to the output end of the motor, and the other end extends to the bottom of the testing box. The wiper blades are all connected to the connecting rod and located outside the testing box. The connecting rod drives the wiper blades to fit against the outer surface of the transparent baffle.

[0009] As a further preferred embodiment, the portal frame includes a square crossbeam and two columns. The two columns are vertically arranged and located on both sides of the conveyor belt. The square crossbeam is horizontally arranged and its two ends are connected to the two columns respectively. The detection box is installed on the square crossbeam. The square crossbeam is movably arranged in the vertical direction and its position is fixed by a locking device.

[0010] This application provides a method for online detection of material particle size distribution and volumetric flow rate, comprising the following steps: S1: An industrial camera and a binocular structured light depth camera respectively acquire images of the material to be inspected on the conveyor belt, and transmit the images to the image processing and analysis system; S2: The image processing and analysis system performs material particle size distribution detection based on the images acquired by the industrial camera; at the same time, it performs online material volumetric flow rate detection based on the images acquired by the binocular structured light depth camera.

[0011] As a further preferred embodiment, the image processing and analysis system performs material particle size distribution detection based on the image acquired by the industrial camera, including the following sub-steps: S21: Reduce the resolution of the image acquired by the industrial camera and crop it; S22: Extract the region of interest from the cropped image and generate a mask; S23: Based on the SAM model, the mask is segmented to generate multiple sub-masks, and invalid sub-masks are filtered out; S24: Perform edge detection on the remaining sub-mask after filtering to generate an edge image; S25: Divide the particle size of the material in the edge image into multiple intervals, calculate the proportion of each interval, and thus complete the particle size distribution detection.

[0012] As a further preferred embodiment, in sub-step S23, the method for filtering invalid sub-masks includes: removing sub-masks with mask overlap, and removing sub-masks with pixel areas smaller than 1000 pixels. The submask and pixel area are greater than Sub-mask.

[0013] As a further preferred embodiment, in sub-step S25, the method for dividing the material particle size into multiple intervals includes: dividing the material particle size into multiple intervals based on multiple preset pixel area thresholds, wherein the multiple pixel area thresholds are set according to a gradient.

[0014] As a further preferred embodiment, the image processing and analysis system performs online detection of material volumetric flow rate based on images acquired by the binocular structured light depth camera, including the following sub-steps: S31: Align the depth image acquired by the binocular structured light depth camera with the color image to generate point cloud data and filter out invalid point cloud data; S32: Calculate the material volume based on the filtered point cloud data using a gridded calculation method; S33: Save the filtered point cloud data, depth image and material volume data to complete the material volume flow rate detection.

[0015] As a further preferred embodiment, in sub-step S31, the method for filtering invalid point cloud data includes: based on the generated point cloud data, preset X-axis thresholds and Y-axis thresholds, and sequentially removing point cloud data whose X-axis coordinates are outside the preset X-axis threshold, whose Y-axis coordinates are outside the preset Y-axis threshold, and whose Z-axis coordinates are outside the minimum to maximum depth of the binocular structured light depth camera.

[0016] In summary, compared with the prior art, the technical solutions conceived in this application have the following main technical advantages: 1. The detection device of this application, through the design of an industrial camera and a binocular structured light depth camera, can monitor volume data and particle size distribution in real time, meeting the high-frequency monitoring needs of dynamic materials in industrial settings. The automated image acquisition, data processing, and result visualization process replaces the inefficient traditional manual sampling and laboratory analysis, reducing human error and improving detection efficiency.

[0017] 2. The testing device of this application achieves regular automatic cleaning through the equipped cleaning components, ensuring that the viewing window remains clear for a long time and guaranteeing the stable operation of the device in harsh environments such as dust and humidity.

[0018] 3. The detection device of this application utilizes the SAM model for automatic image segmentation, which can adapt to different materials and particle morphologies, accurately extracting particle regions and greatly improving the versatility and intelligence of the algorithm. The data processing module supports dynamic correction of mass ratio, enhancing the reliability of the results. In addition, by providing intuitive particle size distribution bar charts, volume change line charts, and depth image displays, the device allows users to flexibly adjust warning thresholds, exposure parameters, and correction values ​​to adapt to different working conditions, improving the interactivity and operability of the device.

[0019] 4. The detection device of this application adopts a portal frame structure. Through the combination design of square beams and two columns, it achieves height adjustment and stable support, adapting to different belt conveyor heights and site layouts. The detection box integrates an industrial camera, a binocular structured light depth camera, and communication equipment. A viewing window is provided at the bottom, meeting explosion-proof sealing requirements. The lighting enhances imaging quality in complex coal yard environments such as dust and humidity. This ensures the long-term stable operation of the detection device in harsh industrial environments, extending the equipment's service life. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the overall structure of the online material particle size distribution and volumetric flow rate detection device provided in the embodiments of this application; Figure 2 This is a schematic diagram of the internal structure of the testing box provided in an embodiment of this application; Figure 3 This is a schematic diagram of the overall structure of the cleaning component provided in an embodiment of this application; Figure 4 This is a flowchart of a material particle size distribution detection method based on images acquired by an industrial camera, provided in an embodiment of this application. Figure 5 This is a flowchart of a method for online detection of material volumetric flow rate based on images acquired by a binocular structured light depth camera, provided in an embodiment of this application. Figure 6 This is a schematic diagram of the image capture interface provided in the embodiments of this application using an industrial camera and a binocular structured light depth camera; Figure 7 This is a schematic diagram of the interface displaying the material particle size distribution ratio and volumetric flow rate change trend provided in the embodiments of this application.

[0021] In all the accompanying drawings, the same reference numerals are used to denote the same elements or structures, wherein: 1. Conveyor belt; 2. Portal frame; 201. Square beam; 202. Column; 3. Lighting lamp; 4. Lifting ring; 5. Inspection box; 6. Lighting power cord; 7. Power supply bus; 8. Optical cable; 9. Computer; 10. U-shaped channel; 11. Fiber optic transceiver; 12. Switch; 13. Terminal block; 14. Relay; 15. Perspective window; 16. Industrial camera; 17. Binocular structured light depth camera; 18. Digital-to-analog socket; 19. Air switch; 20. Switching power supply; 21. Motor; 22. First support; 23. Second support; 24. Connecting rod; 25. Wiper blade. Detailed Implementation

[0022] To make the objectives, technical solutions, and advantages of this application clearer, the following detailed description is provided in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative and not intended to limit the scope of this application.

[0023] This application discloses an online detection device and method for material particle size distribution and volumetric flow rate. It addresses the problems of low detection efficiency, low accuracy, poor real-time performance, and insufficient automation in existing material particle size distribution and volumetric flow rate detection technologies. The device is designed to be suitable for dynamic monitoring of materials conveyed by conveyor belts. It can realize synchronous, real-time, and automated detection of material particle size distribution and volumetric flow rate. It can be widely used in coal-fired power plants, coal chemical plants, mines, and other scenarios to support the intelligent and digital upgrade of material production and transportation processes.

[0024] Reference Figures 1-3An online detection device for material particle size distribution and volumetric flow rate includes a portal frame 2, a detection box 5, and an image processing and analysis system. In practice, the material is transported via a conveyor belt 1. The portal frame 2 provides rigid support for the entire device and spans across the conveyor belt 1. The detection box 5 is detachably installed on the portal frame 2 and is located directly above the conveyor belt 1 containing the material to be detected. The interior of the detection box 5 is a sealed cavity to reduce dust interference. Inside the detection box 5, an industrial camera 16 for detecting the particle size distribution of the material and a binocular structured light depth camera 17 for detecting the flow rate and volume of the material are installed. The industrial camera 16 is fixedly installed via a first bracket 22, and the binocular structured light depth camera 17 is fixedly installed via a second bracket 23. A viewing window 15 is provided at the bottom of the detection box 5 to ensure that the industrial camera 16 and the binocular structured light depth camera 17 have good vertical field of view at the same time, for collecting particle size and pile shape image information of the material on the conveyor belt 1. In addition, a data transmission module is fixedly installed inside the detection box 5 and is connected to the industrial camera 16 and the binocular structured light depth camera 17 respectively. The image processing and analysis system is located outside the detection box and is connected to the data transmission module. The image processing and analysis system is connected to the industrial camera 16 and the binocular structured light depth camera 17 through the data transmission module. The image processing and analysis system can simultaneously analyze the material particle size distribution and volumetric flow rate based on the images acquired by the industrial camera 16 and the binocular structured light depth camera 17. The image processing and analysis system includes a computer 9. The computer 9 can analyze the particle size and pile shape image information of the material on the conveyor belt 1 acquired by the industrial camera 16 and the binocular structured light depth camera 17 in real time and output particle size distribution and volumetric flow rate data.

[0025] In this example, the portal frame 2 includes a square beam 201 and two columns 202. The two columns 202 are located on both sides of the conveyor belt 1 and are vertically arranged. The bottom end of the column 202 is fixed to the ground by locking bolts. The square beam 201 is horizontally arranged and its two ends are connected to the two columns 202 respectively. The detection box 5 is installed on the square beam 201. The square beam 201 is movably arranged in the vertical direction and its position is fixed by locking parts to adapt to different conveyor belt heights. The uprights 202 are external square steel structures, and the two ends of the square beam 201 are internal square steel structures. The locking element is a positioning rod. Openings are provided at the top of both uprights 202 to form guide grooves. The two ends of the square beam 201 are respectively inserted into the openings of the two uprights 202 to achieve vertical sliding. First positioning holes are provided on the side walls of the uprights 202, and several second positioning holes are provided at both ends of the square beam 201 along the vertical direction. The positioning rods pass through the first and second positioning holes in sequence to achieve rapid locking of the square beam 201 at any height. By changing different hole positions, the square beam 201 can be raised and lowered, thereby enabling the detection box 5 to adapt to different conveyor belt heights. In another feasible embodiment, linear motors 21 are vertically fixedly installed on both uprights 202. The two ends of the square beam 201 are respectively fixedly connected to the output shafts of the two linear motors 21, allowing automatic adjustment of the height of the square beam 201 through the linear motors 21.

[0026] In this embodiment, the detection box 5 adopts an explosion-proof enclosure, and its top cover is fixedly installed below the square crossbeam 201 by high-strength fixing bolts. The viewing window 15 is set to allow the industrial camera 16 and the binocular structured light depth camera 17 to acquire material images in real time, ensuring imaging quality while meeting explosion-proof sealing requirements. To ensure stable power and signal transmission, the detection box 5 is externally fixed with a power aviation plug for power input and a fiber optic aviation plug for fiber optic communication.

[0027] Specifically, a transparent baffle is fixedly installed at the viewing window 15. The inspection box 5 is equipped with a cleaning component for cleaning the outer surface of the transparent baffle. The cleaning component includes a motor 21, a connecting rod 24, and a wiper blade 25. The motor 21 is vertically fixed inside the inspection box 5 with its output shaft facing downwards. One end of the connecting rod 24 is coaxially fixedly connected to the output end of the motor 21, and the other end extends to the bottom of the inspection box 5. The wiper blade 25 is connected to the connecting rod 24 and is located outside the inspection box 5. The wiper blade 25 can fit against the outer surface of the transparent baffle under the drive of the connecting rod 24. The rotation radius of the wiper blade 25 is slightly larger than the diagonal of the viewing window 15 to ensure full coverage when it rotates.

[0028] The testing box 5 also houses a fiber optic transceiver 11, a switch 12, a time-delay relay 14, a digital-to-analog socket 18, an air switch 19, and a switching power supply 20 to achieve data transmission and power distribution functions. All components are installed via U-shaped slots 10. This application uses a power supply bus 7 connected to an aviation power plug to supply power to the testing box 5. The air switch 19 is connected to the aviation power plug for overall power control. The input terminal of the digital-to-analog socket 18 is connected to the air switch 19, and the output terminal of the digital-to-analog socket 18 is connected to the various components inside the testing box 5 via terminal blocks 13 to provide power. The data transmission module includes a fiber optic transceiver 11 and a switch 12. The switch 12 is a PoE switch. The input of the switch 12 is connected to both the industrial camera 16 and the binocular structured light depth camera 17 for data transmission. The output of the switch 12 is connected to the fiber optic transceiver 11, converting the RGB image of the material acquired by the industrial camera 16 and the 3D point cloud data generated by the binocular structured light depth camera 17 into optical signals. The fiber optic transceiver 11 is connected to a fiber optic aviation connector, which in turn connects to a computer in the remote monitoring room. This enables data transmission and visualization, supports the saving of detection data, and allows exporting data to Excel format by time period for subsequent analysis and traceability. A switching power supply 20 and a time-delay relay 14 are both connected to a motor 21. The switching power supply 20 controls the start and stop of the motor 21, while the time-delay relay 14 controls the motor 21 to drive the wiper blades 25 at a set cycle, automatically cleaning the viewing window 15 to remove dust and debris, ensuring camera imaging quality.

[0029] To improve the shooting effect of the industrial camera 16 and the binocular structured light depth camera 17, the device also includes two illumination lamps 3, which are connected to the power supply bus 7 via illumination power lines 6. The built-in drivers of the illumination lamps 3 convert AC power to DC power. Both illumination lamps 3 are mounted on the square beam 201 and are symmetrically arranged on both sides of the detection box 5, with the optical axis facing the conveyor belt 1. This provides constant, symmetrical, and sufficient illumination for the industrial camera 16 and the binocular structured light depth camera 17, ensuring consistent and clear image brightness even under extremely short exposures. Specifically, sliders are slidably mounted at both ends of the square beam 201 in the vertical direction. The two illumination lamps 3 are mounted on the two sliders via rotating shafts. The illumination lamps 3 can adjust their illumination angle. Linear drive components for driving the slider movement are also provided at both ends of the square beam 201. The linear drive components are electric push rods or lead screw stepper motors 21. By adjusting the positions of the sliders and illumination lamps 3, precise adjustment of the supplementary lighting height can be achieved.

[0030] Furthermore, both sides of the testing box 5 are fixedly connected with lifting rings 4, which are connected to the square crossbeam 201 by iron chains. This structure can effectively prevent the testing box 5 from falling onto the conveyor belt 1 in the event of accidental shaking or loosening of the fixing bolts, thereby avoiding damage to the conveyor belt 1, ensuring the safety of equipment and personnel, and achieving double safety protection.

[0031] Furthermore, the external of the testing box 5 is detachably equipped with heat dissipation fins, which can effectively improve the heat dissipation efficiency of the testing box 5. The heat dissipation fins are connected to the testing box 5 by magnetic attraction, adhesive bonding or screws.

[0032] Reference Figures 4-7 Due to the significant differences in particle size of materials (such as a mixture of sand and stones), and the presence of oil stains and metallic reflections on conveyor belt 1, traditional segmentation algorithms (such as threshold segmentation and edge detection) may miss small particles due to the large particle size range, or mistakenly identify shadows as materials due to complex backgrounds. To improve the segmentation and recognition accuracy of material particle size, the core algorithm of the detection device in this application integrates the Segment Anything Model (SAM). SAM combines the local feature extraction of Convolutional Neural Network (CNN) with the global context modeling of Transformer, possessing cross-task adaptation and zero-shot segmentation capabilities. Cross-task adaptation allows the same model to simultaneously handle segmentation tasks (distinguishing between materials and background) and recognition tasks (marking different particle size regions). Zero-shot segmentation capability allows the model to adapt to new materials directly through prompts (such as switching from segmenting ore to segmenting plastic granules) without retraining for each type of material. Its architecture consists of three parts: an image encoder, a cue encoder, and a mask decoder. The image encoder extracts image features based on the Vision Transformer, dividing the entire image into several small blocks (e.g., 16×16 pixels), and captures global relationships (e.g., "this shadow belongs to the background, not the material") through the Transformer's self-attention mechanism. The cue encoder converts points, boxes, and other cues into embedding vectors, allowing manual or automated equipment to input minimalist cues (e.g., clicking a point in the center of the material or drawing a rough box), converting these into "search instructions" that the algorithm understands. The mask decoder fuses global image features with cue instructions to generate a pixel-level accurate mask (a transparent layer of the material outline). SAM effectively addresses industrial scenarios with large particle size differences and complex backgrounds, automatically generating mask images that reflect the material outline and distribution characteristics. Compared to traditional segmentation methods, SAM significantly improves the accuracy and robustness of image segmentation, providing a reliable image foundation for subsequent granular analysis.

[0033] This application also discloses an online detection method for material particle size distribution and volumetric flow rate, which uses the above-mentioned detection device and includes the following steps: S1: Industrial camera 16 and binocular structured light depth camera 17 acquire images of the material to be inspected on conveyor belt 1 and transmit the images to the image processing and analysis system. Specifically, the industrial camera 16 is initialized and images are captured, including a complete process of setting up the camera, capturing, and preprocessing the images. First, parameters such as camera IP address, model type, region of interest (ROI) coordinates, and area threshold are read, and the exposure time is obtained. Then, the industrial camera 16 is connected using the specified IP address, and its serial number is verified to ensure authorization. If the camera is not found or the serial number does not match, an error is thrown. Next, the camera is configured to continuous acquisition mode, the trigger signal is turned off, and parameters such as exposure time, gain, and single-shot white balance are set to optimize image quality. The camera begins acquisition and converts the acquired image to RGB format; if this fails, an error is reported. Simultaneously, the binocular structured light depth camera 17 and related configurations are initialized to prepare tools and environment for subsequent volume detection. First, create three storage directories: a depth map directory, a point cloud directory, and a color image directory. Next, define key constants: a reference height as the volume calculation baseline, minimum and maximum depths for filtering the point cloud, a JSON file path for storing data, and the ESC key for exiting the program. Then, initialize the stereo structured light depth camera 17, configure the depth and color streams, enable frame synchronization to ensure alignment of the color and depth images, initialize a temporal smoothing filter to reduce noise, and an alignment filter and point cloud filter for generating the point cloud. Finally, start the camera using the configuration parameters to begin capturing frame data; if initialization fails, exit.

[0034] S2: The image processing and analysis system completes the material particle size distribution detection based on the image acquired by the industrial camera 16; at the same time, it completes the online detection of material volumetric flow rate based on the image acquired by the binocular structured light depth camera 17.

[0035] Furthermore, the image processing and analysis system, based on the images acquired by the industrial camera 16, completes the particle size distribution detection of materials, including the following sub-steps: S21: Reduce the resolution of the image acquired by the industrial camera 16 and crop it; the captured image is preprocessed: adjust the brightness and contrast, reduce the resolution to reduce the amount of computation, and then crop a 2048×1536 pixel area from the center. Finally, the processed image is saved as a JPG file.

[0036] S22: Extract the Region of Interest (ROI) from the cropped image and generate a mask. Specifically, extracting the ROI from the captured image lays the foundation for subsequent image segmentation. First, read the coordinates of the four vertices of the ROI, for example, [(x1,y1),(x2,y2),(x3,y3),(x4,y4)]. These coordinates define the rectangular region to be analyzed. Next, generate a binary mask of the same size as the original image based on these coordinates, where the ROI region is white (pixel value 255) and the remaining regions are black (pixel value 0). Then, combine the original image with the mask, retaining only the pixels of the ROI region and setting the other regions to black. Simultaneously, calculate the total area (number of pixels) of the ROI region for subsequent verification. Finally, output the cropped ROI image and its area, providing input for the next step of image segmentation processing.

[0037] S23: Based on the SAM model, the mask is segmented to generate multiple sub-masks, and invalid sub-masks are filtered out. The methods for filtering invalid sub-masks include: removing sub-masks with overlapping masks, and removing sub-masks with pixel areas smaller than a certain value. The submask and pixel area are greater than The process involves using a sub-mask. Specifically, the SAM model is used to segment the ROI image, generating multiple masks which are then filtered and sorted. First, a pre-trained SAM model is loaded based on the model type and path, and then loaded onto a GPU (CUDA device) to accelerate processing, while segmentation parameters (such as point sampling density and IOU threshold) are set. Next, the SAM model automatically segments the ROI image, generating multiple masks. Each mask represents a detected object (such as a particle), containing the segmented region (the pixel area of ​​the object), area (number of pixels), and center point coordinates (the reference point of the object). Subsequently, mask filtering is performed: area filtering removes masks that are too small or too large to eliminate noise or invalid regions; logical AND operations are used to check for overlap between masks, eliminating overlapping masks and retaining only independent masks, outputting a filtered mask list, an area list, and a center point coordinate list; finally, the masks are sorted in descending order of area size, returning the sorted mask list and area list to prepare for subsequent processing.

[0038] S24: Perform edge detection on the remaining sub-masks after filtering to generate edge images. Specifically, perform edge detection on the segmented masks, visualize the segmentation results and edges, and finally save them as images. First, perform edge detection on each mask: use the Canny algorithm to convert the mask into a binary image (pixel value 0 or 255), then dilate the edges to enhance their visibility, and output the edge image (Boolean array, edge pixels are True). Next, create the visualization results: initialize a fully transparent RGBA image, assign random RGB colors to each mask (transparency 0.35), and mark the edges as black (transparency 1.0). Overlay the masks and edges onto the ROI image to generate an intuitive visualization effect. Finally, generate a save path based on the current date and time, save the visualization image as a PNG format, containing the segmented regions and edges, for easy subsequent inspection and analysis.

[0039] S25: Divide the particle size of the material in the edge image into multiple intervals, calculate the proportion of each interval, and thus complete the particle size distribution detection. The method for dividing the particle size into multiple intervals includes: dividing the particle size into multiple intervals based on pre-set pixel area thresholds, with the pixel area thresholds set in a gradient. Before detection, particle size images are captured experimentally. Pixel area thresholds can be manually set based on the particle size in different images, such as preset pixel area thresholds [500, 1200, 2000]. Particle sizes with a pixel area ≤ 500 are set as the first interval, those with 500 < pixel area ≤ 1200 as the second interval, those with 1200 < pixel area ≤ 2000 as the third interval, and those with a pixel area > 2000 as the fourth interval. Particles are divided into different intervals according to the area thresholds, the proportion of each interval is calculated, and the results are corrected to generate the final particle size distribution data. Calculate the percentage of particles in each interval relative to the total number of particles, accurate to two decimal places. Then, read the correction value, add the percentage of each interval to the corresponding correction value, and generate the corrected percentage. If the number of correction values ​​does not match the number of intervals, an error signal is generated.

[0040] S26: Save all processing results (including timestamps, granularity distribution percentages, and image paths) as a JSON file and clean up old files to manage storage space. First, create a results dictionary containing: timestamps, corrected granularity distribution percentages, mask image paths, and original image paths. Then, generate a save path based on the current date and write the results to the JSON file in append mode, storing one record per line. Simultaneously, write the results to the running status file. Finally, check the original image folder and the mask image folder. If the number of images exceeds 500, sort them by creation time, delete the oldest files, and keep the newest 500 to avoid excessive storage space consumption.

[0041] Furthermore, the image processing and analysis system, based on images acquired by the binocular structured light depth camera 17, performs online detection of material volumetric flow rate, including the following sub-steps: S31: Align the depth image acquired by the binocular structured light depth camera 17 with the color image, generate point cloud data, and filter invalid point cloud data. First, acquire a set of frame data (including color image and depth image). If no frame is acquired, skip the current loop and continue waiting for the next frame. Next, extract the color image from the frame data and convert it to BGR format color image. If no color image is available, skip the loop. Then, extract the depth image, acquire the depth data, convert the depth value to meters using the scaling factor of the depth image, and filter the depth values, keeping the data between the minimum and maximum depths and setting those outside the range to 0. Subsequently, use a time smoothing filter to smooth the depth data to reduce noise. Finally, normalize the depth data to the range of 0-255, use OpenCV to generate a color depth map (blue for near objects, red for distant objects), crop it to the specified area (rows a:b, columns c:d), and adjust the resolution to prepare for subsequent point cloud generation and volume calculation. The color and depth images are processed to ensure they are spatially aligned so that each pixel corresponds to the same physical location. Next, the aligned frame data is converted into a point cloud, a set of three-dimensional coordinate points, each containing X, Y, and Z coordinates (in meters). The point cloud is then filtered, retaining points that meet the following conditions: X-axis coordinates between e and f, Y-axis coordinates between g and h, and Z-axis coordinates between the minimum and maximum depths of the binocular structured light depth camera 17. The filtered point cloud data is used for subsequent volume calculations, ensuring that only points within the valid region are included. The volume of objects located above a reference height in the point cloud is then calculated. The method for filtering invalid point cloud data includes: setting preset X-axis thresholds [e,f] and Y-axis thresholds [g,h] based on the point cloud data in the generated point cloud image to ensure that the material image can be clearly displayed in the plane of X-axis thresholds [e,f] and Y-axis thresholds [g,h], such as X-axis thresholds [-0.1m, 1m] and Y-axis thresholds [-0.1m, 1.2m]. This involves removing point cloud data whose X-axis coordinates are outside the preset X-axis thresholds [e,f], removing point cloud data whose Y-axis coordinates are outside the preset Y-axis thresholds [g,h], and removing point cloud data whose Z-axis coordinates are outside the range of the minimum depth of the binocular structured light depth camera 17 from 0.1m to the maximum depth of 0.72m.

[0042] S32: Based on the filtered point cloud data, the material volume is calculated using a gridded calculation method. First, check if the point cloud data is empty. If it is empty (no valid points), the volume is set to 0 and the calculation is skipped. Next, Open3D is used to remove outliers (such as noise) from the point cloud by checking the 20 neighboring points of each point and setting a standard deviation multiple to eliminate outliers. Then, points with Z coordinates less than the reference height are selected from the point cloud. These points are considered objects located above the reference plane. If no points meet the condition, the volume is set to 0. Next, gridded calculation is performed: the X and Y coordinate ranges of the object points are obtained, a grid is created, and the point cloud is assigned to the grid according to the X and Y coordinates. The maximum Z value (representing the object height) of each grid is calculated. For each grid, its height difference is calculated and multiplied by the grid area to obtain the volume of that grid. The volumes of all grids are summed (taking a negative value because the Z value is less than the reference height) to obtain the total volume (unit: m). 3 Finally, the volume value is rounded to two decimal places as the final result.

[0043] S33: Saves the filtered point cloud data, depth image, and material volume data to complete the material volumetric flow rate detection. The volumetric flow rate of the material can be intuitively seen based on the point cloud data, depth image, and material volume data. Specifically, when the estimated volume is greater than 0.3m³... 3 During the process, point cloud, depth map, and volume data are saved to specified files to record the detection results. First, the point cloud data is saved as a PLY file, with the file path based on the depth image's timestamp; the X, Y, and Z coordinates are saved using Open3D. Next, the point cloud directory is checked; if the number of PLY files exceeds 100, they are sorted by creation time, the oldest files are deleted, and the newest 100 are kept to manage storage space. Then, the depth image is saved as a PNG file, with the file path also based on the timestamp, and the image directory is checked; if the number of PNG files exceeds 100, the oldest files are deleted, and the newest 100 are kept. Finally, JSON entries are created, containing the volume value, point cloud file path, depth image path, and timestamp; if the JSON file already exists, the existing data is loaded; otherwise, a new list is created; new entries are appended to the list, the newest 100 records are kept, and saved to the JSON file.

[0044] The particle size distribution and volumetric flow rate of coal were detected using the detection device described in this application, as detailed below: After the device is started, the binocular structured light depth camera 17 begins to monitor the coal flow on the conveyor belt 1 in real time, continuously acquiring the three-dimensional structural information of the coal flow and calculating its volume; when the detected coal flow volume exceeds the set threshold (0.3m), the camera will proceed with the detection. 3When the device automatically triggers the industrial camera 16 and the binocular structured light depth camera 17 to work together, they acquire visible light images and depth images respectively. The acquired image data is transmitted to the fiber optic transceiver 11 via the switch 12. The fiber optic transceiver 11 converts the optical signal into an electrical signal and then transmits the data stably to the computer 9 via the optical cable 8. After receiving the image and point cloud data, the computer 9 automatically executes algorithms such as image preprocessing, target segmentation, and volume calculation to extract key parameters such as the particle size distribution and volumetric flow rate of the coal flow. The analysis results are then displayed in real time through a graphical interface, providing accurate and visualized data support for the backend control or management system. Figure 6 and Figure 7 As shown, Figure 6 An interface for capturing images for industrial camera 16 and binocular structured light depth camera 17. Figure 7 The interface displays the proportion of material particle size distribution and the trend of volumetric flow rate change. If the detected coal flow volume does not exceed the set threshold, the device enters a low-power monitoring mode, the industrial camera 16 remains in standby mode, and only the binocular structured light depth camera 17 continues to monitor to ensure continuous acquisition of basic volume information.

[0045] When detecting coal particle size distribution, the system operates as follows: Before formal testing, the industrial camera 16 needs to be initialized, and the camera's IP address and serial number need to be verified to ensure device authorization and prevent IP conflicts and unauthorized device access. The camera's acquisition mode is set to continuous acquisition, the trigger mode is rising edge trigger, the exposure time is set to 200μs, the white balance mode is set to single white balance, and the gain value is set to 20 to ensure that the image is clear and stable under different lighting conditions.

[0046] The image was captured using an industrial camera (16), and the brightness and contrast were adjusted (default brightness 30, contrast 30, can be adjusted according to lighting conditions). The resolution was reduced to 2048×1536, cropped to the specified area, and saved as a JPG file.

[0047] Load the SAM model type (e.g., high-performance model) and the path to the pre-trained weight file, and set the segmentation parameters (set the number of grid points to 32, the predicted crossover ratio threshold to 0.86, and the segmentation stability threshold to 0.92) to control the accuracy and stability of particle segmentation.

[0048] Based on the coordinates of four points (136, 30), (136, 1987), (2423, 1987), and (2423, 30) in the configuration file, a binary polygon mask is generated (ROI area is 255, non-ROI area is 0), and the ROI image of the material on the conveyor belt is extracted; the SAM model is used to automatically segment the particles in the ROI, generating multiple masks (each mask corresponds to one particle); particles with too small an area (< ) or too large (> The mask is designed to avoid noise or excessively large area interference, eliminate overlapping masks, and sort them by area from largest to smallest; edge detection is performed on each mask using the Canny algorithm to generate an edge image; a transparent background image is created, each mask is assigned a random color, the transparency is set to 0.35, and the edges are set to black to visualize the particle boundaries.

[0049] Based on the three thresholds (500, 1200, 2000 pixels) in the configuration file, the particles are divided into five intervals: interval 1 (representing special particles), interval 2 (area ≤ 500 pixels), interval 3 (500 < area ≤ 1200 pixels), interval 4 (1200 < area ≤ 2000 pixels), and interval 5 (area > 2000 pixels). The number of particles in each interval is counted, the total number of particles is calculated, and the percentage of each interval is obtained. Correction values ​​are loaded from JSON to adjust the percentage of each interval, compensate for segmentation errors introduced by particle overlap or lighting effects, and ensure that the particle size distribution results are more accurate.

[0050] The particle size distribution data, timestamps, and image paths are saved to a JSON file. The JSON data supports time synchronization with the particle size detection program, facilitating correlation analysis. It can also be converted to Excel format via a script and exported by time period for easy statistics and traceability. During runtime, the system checks the original image and mask image folders. If there are more than 500 images, the oldest file is deleted to ensure efficient use of storage space.

[0051] When detecting the volumetric flow rate of coal, the system operates as follows: Initialize the binocular structured light depth camera 17 and configure the data stream. Set the camera to capture depth and color images at a resolution of 1280×800, ensuring clear images are captured. The depth image records the distance to the material, and the color image records the color.

[0052] Real-time acquisition of depth and color images. First, the depth data undergoes temporal smoothing (smoothing coefficient 0.5): the depth value of the current frame is weighted and fused with the previous frame to reduce random noise and improve data stability. Then, distance filtering is performed, retaining only valid pixels with depth values ​​between 0.02m and 0.72m, and removing invalid points caused by being too close (background interference) or too far (range measurement failure). Subsequently, the processed depth data is mapped to a pseudo-color image (color depth map) to enhance visual readability and facilitate human-machine interface display. To focus on key detection areas, the color depth map is cropped, retaining only the image region within the range [240:560, 448:960], further improving image display efficiency and interface response speed.

[0053] Align the depth and color images to generate a 3D point cloud containing spatial coordinates (x, y, z) and optional color information (RGB). Range filtering is applied to the point cloud, limiting it to a specified region (x: -0.1 to 1m, y: -0.1 to 1.2m, z: 0.1 to 0.72m). Statistical outlier removal is performed using the Open3D library to further improve data quality.

[0054] The volume of coal is estimated based on the processed point cloud data. The reference plane height is set to the surface of conveyor belt 1 when it is unloaded (elevation set to 0.72m), and volume calculations are performed only on the portion of the coal pile above this plane. First, all points with Z-coordinates less than 0.001m (i.e., points below the reference plane by 0.001m) are selected from the point cloud to construct the volume estimation region. A gridded calculation method is used to divide the point cloud space into a two-dimensional grid of size 0.05m × 0.05m. For each grid cell, the maximum z-value (i.e., the deepest point) is found. The difference between the maximum depth of each square and the reference height (0.72m) is calculated, multiplied by the square area to obtain the volume of that grid cell. Then, the volume values ​​of all grid cells are summed to obtain the total volume of the coal pile in the current frame.

[0055] When the volume is greater than The program saves point clouds as PLY files (3D models) and depth images as PNG files, and records the volume, file path, and timestamp in a JSON file. The timestamp in the JSON file is synchronized with the granularity distribution detection program, ensuring temporal consistency of volume and particle size data for easy comprehensive analysis. Data can be converted to Excel format via scripts and exported by time period for convenient statistics and traceability. To manage storage, the program automatically cleans up more than 100 PLY files and PNG images, retaining only the latest 100 sets of data to ensure efficient use of storage space.

[0056] It should be understood that expressions such as "comprising" and "may include" as used in this application indicate the existence of the disclosed functions, operations, or constituent elements, and do not limit one or more additional functions, operations, and constituent elements. In this application, terms such as "comprising" and / or "having" may be interpreted as indicating a specific characteristic, number, operation, constituent element, component, or combination thereof, but should not be interpreted as excluding the existence or possibility of adding one or more other characteristics, numbers, operations, constituent elements, components, or combinations thereof.

[0057] It should be understood that the terms “center,” “upper,” “lower,” “front,” “rear,” “left,” “right,” “vertical,” “horizontal,” “inner,” “outer,” “clockwise,” “counterclockwise,” “axial,” “radial,” and “circumferential” indicate the orientation or positional relationship based on the orientation or positional relationship shown in the accompanying drawings. They are used only for the convenience of describing this application and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific orientation, or be constructed and operated in a specific orientation. Therefore, they should not be construed as limitations on this application.

[0058] Furthermore, the terms "first" and "second" are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of technical features indicated. Thus, a feature defined as "first" or "second" may explicitly or implicitly include one or more of that feature. In the description of this application, "multiple" means two or more, unless otherwise explicitly specified.

[0059] In this application, unless otherwise expressly specified and limited, the terms "installation," "connection," "linking," and "fixing," etc., should be interpreted broadly. For example, they can refer to a fixed connection, a detachable connection, or an integral connection; they can refer to a mechanical connection or an electrical connection; they can refer to a direct connection or an indirect connection through an intermediate medium; and they can refer to the internal connection between two components. Those skilled in the art can understand the specific meaning of the above terms in this application according to the specific circumstances.

[0060] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of this application and is not intended to limit this application. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of this application should be included within the protection scope of this application.

Claims

1. An online detection device for material particle size distribution and volumetric flow rate, characterized in that, include: A portal frame (2) spans across a conveyor belt (1) used for conveying materials to be tested; The detection box (5) is detachably installed on the portal frame (2) and located directly above the conveyor belt (1). The interior of the detection box (5) is a sealed cavity and is equipped with an industrial camera (16) for detecting the particle size distribution of materials and a binocular structured light depth camera (17) for detecting the flow rate and volume of materials. A viewing window (15) is provided at the bottom of the detection box (5). The image processing and analysis system is connected to an industrial camera (16) and a binocular structured light depth camera (17), and simultaneously analyzes the material particle size distribution and volumetric flow rate based on the images acquired by the industrial camera (16) and the binocular structured light depth camera (17).

2. The online detection device for material particle size distribution and volumetric flow rate as described in claim 1, characterized in that, A transparent baffle is provided at the viewing window (15), and a cleaning component for cleaning the outer surface of the transparent baffle is provided on the detection box (5).

3. The online detection device for material particle size distribution and volumetric flow rate as described in claim 2, characterized in that, The cleaning assembly includes a motor (21), a connecting rod (24), and a wiper blade (25). The motor (21) is vertically fixed inside the detection box (5). One end of the connecting rod (24) is coaxially fixed to the output end of the motor (21), and the other end extends to the bottom of the detection box (5). The wiper blade (25) is connected to the connecting rod (24) and located outside the detection box (5). The connecting rod (24) drives the wiper blade (25) to fit against the outer surface of the transparent baffle.

4. The online detection device for material particle size distribution and volumetric flow rate as described in claim 1, characterized in that, The portal frame (2) includes a square beam (201) and two columns (202). The two columns (202) are both vertically arranged and located on both sides of the conveyor belt (1). The square beam (201) is horizontally arranged and its two ends are connected to the two columns (202) respectively. The detection box (5) is installed on the square beam (201). The square beam (201) is movably arranged in the vertical direction and its position is fixed by locking components.

5. A method for online detection of material particle size distribution and volumetric flow rate using the detection device as described in any one of claims 1-4, characterized in that, Includes the following steps: S1: The industrial camera (16) and the binocular structured light depth camera (17) respectively acquire images of the material to be detected on the conveyor belt (1) and transmit the images to the image processing and analysis system; S2: The image processing and analysis system completes the material particle size distribution detection based on the image acquired by the industrial camera (16); at the same time, it completes the online detection of material volume flow rate based on the image acquired by the binocular structured light depth camera (17).

6. The method for online detection of material particle size distribution and volumetric flow rate as described in claim 5, characterized in that, The image processing and analysis system performs material particle size distribution detection based on the images acquired by the industrial camera (16), including the following sub-steps: S21: Reduce the resolution of the image acquired by the industrial camera (16) and crop it; S22: Extract the region of interest from the cropped image and generate a mask; S23: Based on the SAM model, the mask is segmented to generate multiple sub-masks, and invalid sub-masks are filtered out; S24: Perform edge detection on the remaining sub-mask after filtering to generate an edge image; S25: Divide the particle size of the material in the edge image into multiple intervals, calculate the proportion of each interval, and thus complete the particle size distribution detection.

7. The method for online detection of material particle size distribution and volumetric flow rate as described in claim 6, characterized in that, In sub-step S23, the method for filtering invalid sub-masks includes: removing sub-masks with overlapping masks, and removing sub-masks with pixel areas smaller than 100 pixels. The submask and pixel area are greater than Sub-mask.

8. The method for online detection of material particle size distribution and volumetric flow rate as described in claim 6, characterized in that, In sub-step S25, the method for dividing the material particle size into multiple intervals includes: dividing the material particle size into multiple intervals based on multiple preset pixel area thresholds, wherein the multiple pixel area thresholds are set according to a gradient.

9. The method for online detection of material particle size distribution and volumetric flow rate as described in claim 5, characterized in that, The image processing and analysis system performs online detection of material volumetric flow rate based on the images acquired by the binocular structured light depth camera (17), including the following sub-steps: S31: Align the depth image acquired by the binocular structured light depth camera (17) with the color image, generate point cloud data and filter out invalid point cloud data; S32: Calculate the material volume based on the filtered point cloud data using a gridded calculation method; S33: Save the filtered point cloud data, depth image and material volume data to complete the material volume flow rate detection.

10. The method for online detection of material volumetric flow rate as described in claim 9, characterized in that, In the sub-step S31, the method for filtering invalid point cloud data includes: based on the generated point cloud data, preset X-axis threshold and Y-axis threshold, and sequentially remove point cloud data whose X-axis coordinate is outside the preset X-axis threshold, whose Y-axis coordinate is outside the preset Y-axis threshold, and whose Z-axis coordinate is outside the range from the minimum depth to the maximum depth of the binocular structured light depth camera (17).

Images