Particle size detection method, device and equipment of material particle image, medium and product

Abstract

The invention discloses a particle size detection method, device and equipment of a material particle image, a medium and a product, and relates to the field of blast furnace ironmaking. Performing coarse segmentation on the material particle image, marking a material particle area and a background area of the material particle image, and determining an image after coarse segmentation; processing the roughly segmented image by using an image segmentation model, and determining a material segmentation image and a depth estimation image; performing joint processing on the material segmentation map and the depth estimation map to generate a visual image containing material contour, region division and depth information; determining material particle information according to the visual image; the material particle information comprises a material particle size parameter and a particle size distribution condition. According to the method, the large-scale particle detection precision in an industrial scene can be improved.

Description

Technical Field

[0001] This application relates to the field of blast furnace ironmaking, and particularly to a method, device, equipment, medium and product for detecting the particle size of material particle images. Background Art

[0002] In the blast furnace ironmaking process, the particle size of materials has a crucial impact on the gas permeability, heat and mass transfer processes, and chemical reaction efficiency in the furnace. Precise control of the material particle size can improve the production efficiency of the blast furnace, reduce energy consumption, extend the lining life, and improve the quality of hot metal. However, the existing material particle size detection methods cannot accurately detect large-scale particles in industrial scenarios and cannot meet the increasing requirements for refined production in blast furnace ironmaking. Summary of the Invention

[0003] The purpose of this application is to provide a method, device, equipment, medium and product for detecting the particle size of material particle images to solve the problem of low detection accuracy of large-scale particles in industrial scenarios by the existing material particle size detection methods.

[0004] To achieve the above purpose, this application provides the following solutions:

[0005] In the first aspect, this application provides a method for detecting the particle size of material particle images, including:

[0006] Collecting material particle images from multiple perspectives;

[0007] Performing rough segmentation on the material particle images, marking the material particle regions and background regions of the material particle images, and determining the images after rough segmentation;

[0008] Processing the images after rough segmentation by using an image segmentation model to determine a material segmentation map and a depth estimation map;

[0009] Performing joint processing on the material segmentation map and the depth estimation map to generate a visual image including material contours, region division, and depth information;

[0010] Determining material particle information according to the visual image; the material particle information includes material particle size parameters and particle size distribution.

[0011] In the second aspect, this application provides a device for detecting the particle size of material particle images, including:

[0012] A visual detection device for collecting material particle images from multiple perspectives;

[0013] A rough segmentation module for performing rough segmentation on the material particle images, marking the material particle regions and background regions of the material particle images, and determining the images after rough segmentation;

[0014] An image segmentation model for processing the roughly segmented image to determine a material segmentation map and a depth estimation map;

[0015] An image integration and statistics module for jointly processing the material segmentation map and the depth estimation map to generate a visual image containing material contours, area division, and depth information;

[0016] A material particle information determination module for determining material particle information according to the visual image; the material particle information includes material particle size parameters and particle size distribution.

[0017] In a third aspect, the present application provides a computer device, including: a memory, a processor, and a computer program stored on the memory and executable on the processor, where the processor executes the computer program to implement the particle size detection method for the material particle image described in any one of the above.

[0018] In a fourth aspect, the present application provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it implements the particle size detection method for the material particle image described in any one of the above.

[0019] In a fifth aspect, the present application provides a computer program product, including a computer program, and when the computer program is executed by a processor, it implements the particle size detection method for the material particle image described in any one of the above.

[0020] According to the specific embodiments provided by the present application, the following technical effects are disclosed:

[0021] First, the present application obtains a material particle image by constructing a multi-view depth image acquisition system. The material particle image is comprehensive image information of the material, including surface texture information and depth information at different angles. The material particle image is roughly segmented, the particle area and the background area of the material particle image are marked, and the roughly segmented image is determined, that is, advanced image preprocessing technology is used to remove noise and highlight material features, improving the accuracy of material particle detection.

[0022] Secondly, an image segmentation model based on deep learning (FastSAM) is adopted to combine with depth information to achieve precise segmentation and size correction of the material area. Then, through feature extraction, matching, alignment, and fusion of multi-view images, a complete and accurate material segmentation map and depth estimation map are obtained. Finally, edge detection and particle size calculation are performed in combination with depth information to determine the particle size information of the material, thereby realizing high-precision, high-adaptability, and high-efficiency detection of the particle size of blast furnace ironmaking materials, providing reliable data support for material control in the blast furnace ironmaking process, solving the problems of insufficient accuracy, poor adaptability, and low efficiency existing in the prior art in material particle size detection, and helping the blast furnace ironmaking process to develop towards a more intelligent and refined direction. Brief Description of the Drawings

[0023] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the drawings required in the embodiments. Obviously, the drawings in the following description are only some embodiments of the present application. For those of ordinary skill in the art, other drawings can be obtained based on these drawings without creative efforts.

[0024] Figure 1 Flowchart of the method for detecting the particle size of the material particle image provided by the present application;

[0025] Figure 2 Operation flowchart of each step in the method for detecting the particle size of the material particle image provided by the present application;

[0026] Figure 3 Network structure diagram of FastSAM provided by the present application. Detailed Description of the Embodiments

[0027] The following will clearly and completely describe the technical solutions in the embodiments of the present application in conjunction with the drawings in the embodiments of the present application. Obviously, the described embodiments are only some embodiments of the present application, rather than all embodiments. Based on the embodiments in the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0028] To make the above objects, features, and advantages of the present application more obvious and understandable, the present application will be further described in detail below in conjunction with the drawings and specific embodiments.

[0029] The embodiments of the present application provide a method for detecting the particle size of a material particle image. This method is executed by a computer device, which can be specifically executed by a computer device such as a terminal or a server alone, or jointly executed by a terminal and a server. In the embodiments of the present application, as Figures 1 - 2 shown, this method includes the following steps.

[0030] S1: Collect material particle images from multiple perspectives.

[0031] S2: Coarsely segment the material particle images, mark the material particle regions and background regions of the material particle images, and determine the images after coarse segmentation.

[0032] S3: Process the images after coarse segmentation using an image segmentation model to determine a material segmentation map and a depth estimation map.

[0033] S4: Jointly process the material segmentation map and the depth estimation map to generate a visualization image containing material contours, region divisions, and depth information.

[0034] S5: Determine material particle information based on the visualization image; the material particle information includes material particle size parameters and particle size distribution.

[0035] In an exemplary embodiment, before S1, it further includes:

[0036] S11: Determine the number of cameras according to the geometric shape and size of the material transport channel.

[0037] S12: Determine the focal length of the camera using the principle of similar triangles based on the minimum detection size of the material and the size of the desired imaging size on the image sensor.

[0038] S13: Determine the resolution of the image sensor according to the detection accuracy requirements.

[0039] S14: Arrange the lighting system within the set distance range of the camera in a circular or strip distribution manner.

[0040] In practical applications, when using a vision detection device to collect material particle images, first build a multi - perspective image acquisition system to construct a set of multi - perspective image acquisition devices, including multiple high - definition cameras evenly distributed around the material transport channel to ensure shooting the material from different directions. According to the size, shape of the material and the detection accuracy requirements, reasonably determine parameters such as the focal length, resolution, and shooting angle of the camera. For smaller and irregularly shaped materials, use high - resolution, short - focal - length lenses and appropriately increase the number of cameras to obtain more comprehensive perspective information. At the same time, install a stable and brightness - adjustable lighting system to ensure that the material images have clear contrast and brightness uniformity, and avoid interference of shadows on image segmentation and size measurement.

[0041] (1) Camera layout and parameter determination:

[0042] According to the geometric shape and size of the material transport channel, evenly distribute multiple high - definition cameras. Let the number of cameras be n, and the distribution angle interval be θ = 360° / n.

[0043] Determination of the focal length f: According to the minimum detectable size d of the material min and the desired imaging size D img on the image sensor, using the principle of similar triangles, where is the working distance from the camera to the material.

[0044] Resolution R: Determined according to the detection accuracy requirements. If it is necessary to detect details of m×m millimeters on the surface of the material, the actual size corresponding to each pixel on the image sensor should be less than m millimeters, which can be calculated based on the sensor size and resolution.

[0045] (2) Installation of the lighting system: Adopt a ring-shaped or strip-shaped lighting layout around the camera. The lighting intensity I is adjustable and is feedback through a light intensity sensor to control the uniformity error of the illuminance on the material surface within a certain range, ΔI / I≤5%, where ΔI is the difference in illuminance at different positions on the material surface.

[0046] In an exemplary embodiment, S2 can be replaced by the following steps.

[0047] S21: Perform grayscale processing on the material particle image to determine a grayscale image.

[0048] S22: Normalize the pixel values of the grayscale image and use the local binary pattern to calculate the relationship between the neighborhood pixel values and the central pixel value in the local window corresponding to each pixel point to determine the local binary pattern feature map of the image.

[0049] S23: Use the connected region algorithm to perform region labeling on the local binary pattern feature map of the image, label the material particle region and the background region, and determine the image after rough segmentation.

[0050] In practical applications, rough segmentation is a key preprocessing step in material particle size detection. Its goal is to quickly distinguish the material particle region from the background region and provide high-quality preliminary results for subsequent fine segmentation and depth estimation. As Figure 2 shown, for image preprocessing, after collecting multi-view images, first perform grayscale processing to convert the color image into a grayscale image to reduce the data volume and highlight the grayscale difference between the material and the background.

[0051] Use the method of adaptive threshold to determine the threshold of grayscale processing to ensure effective separation of the material and the background under different lighting conditions.

[0052] Then apply the median filtering algorithm to denoise the grayscale image. Median filtering can effectively remove common noises such as salt-and-pepper noise while better preserving the edge information of the material.

[0053] (1) Normalize the pixel values ​​of the grayscale image. In order to capture the texture information of the material particles, the local binary pattern (LBP) is used to calculate the relationship between the neighborhood pixel value and the central pixel value in the local window around the pixel (x, y). The binary pattern is calculated using the following formula:

[0054]

[0055] Among them, p i is the value of the i-th neighborhood pixel in the window; s(x) is the sign function, defined as: P is the total number of neighborhood pixels. The result is a binary number that represents the local texture pattern of the window. The LBP values ​​of all pixels are summed up to form a local binary pattern feature map of the image.

[0056] (2) Region division: Connected region marking: marking the regions of the denoised grayscale image, using the connected region algorithm to mark the material particle regions and background; setting a minimum region threshold to exclude small region noise; treating regions with a marked area smaller than the minimum region threshold as noise and eliminating them; outputting the region division results, outputting the marked image, where each material particle region is represented by a unique marking value, providing input for the subsequent detection model.

[0057] Coarse segmentation transforms the complex whole-image segmentation problem into the processing of local areas through image preprocessing, which significantly reduces the amount of calculation and improves the overall processing efficiency, while having strong environmental adaptability.

[0058] Coarse segmentation breaks down the complex full-image processing task into preliminary area division based on texture features, greatly reducing the amount of input data for subsequent detection models and reducing computational complexity, thereby significantly improving overall efficiency. In addition, the combination of highly adaptable adaptive thresholds and texture feature extraction methods makes the algorithm more robust to complex industrial scenes (such as uneven lighting and high noise).

[0059] In an exemplary embodiment, S3 may be replaced by the following steps.

[0060] S31: training the image segmentation model according to conventional image samples, conventional image sample annotation data, and a material image dataset with dimension annotations; the image segmentation model includes an image segmentation stage and a depth estimation stage.

[0061] S32: In the image segmentation stage, extracting a feature map of the roughly segmented image, and performing multi-level convolution and pooling operations on the feature map using a trained image segmentation model to determine feature maps of different scales.

[0062] S33: Based on feature maps of different sizes, calculate the size ratio relationship according to the material contour features, texture features, and multi-view set relationships, and combine the pre-learned size prior knowledge to generate a material segmentation image.

[0063] S34: In the depth estimation stage, perform multi-scale feature extraction on the material segmentation image to determine multi-scale feature maps.

[0064] S35: Use a feature pyramid network to fuse the multi-scale feature maps, and perform depth prediction and optimization on the fused feature maps to determine a depth estimation map.

[0065] In practical applications, the image segmentation model is an intelligent segmentation network. Using the intelligent segmentation network for image segmentation and depth estimation specifically includes the following steps.

[0066] (1) In the image segmentation stage, the result after rough segmentation received is processed using the image segmentation model (FastSAM). In the network training stage, in addition to using conventional image samples and annotation data, a material image dataset with accurate size annotations is introduced. These size annotation information serves as additional supervision information during the network training process, enabling the network to learn the internal relationship between image segmentation and size correction. Figure 3 This is the network structure diagram of FastSAM provided by this application. The full instance segmentation is as Figure 3 shown. FPN is used in computer vision tasks such as object detection and instance segmentation. By fusing feature information at different levels, it enhances the model's detection ability for targets of different scales. P3 - P5 are feature maps of different levels output by the Feature Pyramid Network (FPN). The numbers in P3 - P5 represent different levels. The higher the level (such as P5), the lower the resolution of the feature map, but the richer the semantic information; the lower the level (such as P3), the higher the resolution of the feature map and the richer the detailed information.

[0067] Based on the network training of FastSAM: The size perception module in the network structure, through multi-level analysis of image features, directly calculates the pixel - actual size ratio relationship corresponding to each material area while segmenting the material area. Using the contour features, texture features, and geometric relationships of the material from multiple perspectives, combined with the pre-learned size prior knowledge, realizes the preliminary estimation of the material size. In the design of the network's loss function, in addition to the loss term for segmentation accuracy, a loss term for size correction error is added, and the segmentation performance and size correction accuracy of the network are optimized simultaneously through the backpropagation algorithm.

[0068] After the network is trained, when segmenting actual images, it can output the binary image of the segmented materials and the size correction coefficients corresponding to each material region. These coefficients will be directly used for subsequent particle size calculation, avoiding the introduction of additional errors in the process from the segmentation result to the size calculation in the traditional method.

[0069] Let the feature map obtained after image feature extraction be F. Through multi-level convolution and pooling operations on F, feature representations at different scales are obtained. After n layers of convolution, the size of the feature map is H n ×W n ×C n , where, H n 、W n are the height and width, and C n is the number of channels).

[0070] Using the contour feature C (gradient information of edge pixels), texture feature T (gray-level co-occurrence matrix feature) of the material, and multi-view geometric relationship G (corresponding relationship of the material contour under different views), combined with the pre-learned size prior knowledge K (standard size range of common materials), calculate the size ratio r: r = f(C, T, G, K), where f is a complex function model learned by the network, which fuses these feature information to estimate the size ratio.

[0071] (2) In the depth estimation stage, apply DispNet depth estimation to the segmented material region image. First, use a parallel multi-scale convolution structure, and at the same time, use convolution kernels of different scales such as 3x3, 5x5, and 7x7 to perform convolution operations on the image to extract multi-scale feature maps. Each scale of convolution kernel can capture different levels of image details and context information. Small-scale convolution kernels focus on local details, while large-scale convolution kernels can obtain more macroscopic structural information.

[0072] Then, use FPN to fuse feature maps at different scales. FPN fuses high-level feature maps with rich semantic information but low resolution and low-level feature maps with strong resolution but weak semantic information through top-down and bottom-up paths, so that the finally obtained feature map contains both rich detail information and strong semantic expression ability, providing strong support for accurate depth estimation.

[0073] Establish a material shape database to store the three-dimensional shape models of common materials and their discretized two-dimensional view information. For the segmented material region, use an image feature matching algorithm to find the most similar material shape template in the shape database. According to the matched shape template, combined with the principle of perspective transformation, use the known geometric relationships and size information in the template to estimate the depth of the material.

[0074] Calculation of the combined loss function: During the model training process, based on the segmentation result and the true segmentation label, the segmentation loss \(L\) based on the Dice coefficient is calculated. seg , and the formula is:

[0075] where, is the true segmentation label of the \(i\)-th pixel (taking values of 0 or 1), is the probability that the \(i\)-th pixel predicted by the model belongs to the material, \(n\) is the total number of pixels in the image, and \(\epsilon\) is a small constant (used to prevent the denominator from being zero).

[0076] Meanwhile, the computer calculates the improved mean squared error depth loss \(L\) based on the depth estimation result and the true depth value. depth , and the formula is:

[0077] where, and are the true depth value and the predicted depth value of the \(i\)-th pixel respectively, represents the gradient operator, and \(\lambda\) is a weight parameter (determined by experiments as \(\lambda = 0.1\)) used to balance the importance of the depth value error and the depth gradient error. Then, the computer obtains the combined loss function \(L\) through weighted combination, and the formula is \(L=\alpha L\) seg +(1 - \(\alpha\))\(L\) depth , where \(\alpha\) is a weight parameter (determined by experiments as \(\alpha = 0.6\)) used to balance the importance of the segmentation loss and the depth loss during the training process.

[0078] During the backpropagation process, based on the combined loss function, the shared layers (the first two convolutional layers) of FastSAM and DispNet and the specific layer parameters of each module are synchronously updated. Using the Adam optimizer, combined with the learning rate decay strategy (such as every 10 epochs, the learning rate decays to 0.9 of the original), the parameters are updated according to a certain learning rate, so that the two modules can promote each other and jointly improve the overall performance of the model. Finally, the model outputs the material segmentation map and the depth estimation map.

[0079] The image segmentation model provided in this application is a cascaded detection model (image segmentation + depth estimation). Through the cascaded method, the detection model realizes the organic combination of the segmentation and depth estimation tasks, significantly improving the overall detection accuracy and robustness.

[0080] Optimized design in the image segmentation stage. The improved FastSAM architecture adopts a lightweight convolutional neural network and an attention mechanism, enabling the segmentation module to significantly improve speed while maintaining high accuracy. The segmentation module focuses on the precise extraction of the material contour and area, providing high-quality input data for the depth estimation stage, thereby avoiding depth estimation bias caused by segmentation errors. Collaborative optimization in the depth estimation stage, based on DispNet for depth estimation, converts the two-dimensional information in the segmentation stage into high-precision three-dimensional depth information through multi-scale feature extraction and feature fusion. The design of the joint loss function (a weighted combination of segmentation loss and depth loss) ensures the coordination of the segmentation and depth estimation tasks, avoiding inconsistent problems that may occur in independently trained models.

[0081] The cascaded design allows the segmentation information to be directly used as the input for depth estimation, fundamentally reducing the error propagation problem between different stages and ensuring the precise positioning of the material area. At the same time, the collaborative optimization and feature sharing strategies further enhance the overall performance of the model, enabling it to quickly segment materials and generate high-precision depth information in complex scenarios, thus effectively meeting the high requirements of industrial production.

[0082] In an exemplary embodiment, S4 can be replaced by the following steps.

[0083] S41: Mark the foreground area in the material segmentation map as the region of interest, extract the depth values corresponding to the region of interest, and perform filling or interpolation processing on the depth values of the non-material areas in the depth estimation image to generate an integrated image containing the material contour and depth information.

[0084] S42: Based on the integrated image, extract the regional depth features of each material particle, and combine the regional depth features with the particle contour information to generate a visualization image containing the material contour, regional division, and depth information.

[0085] This application generates high-precision material detection results by jointly processing the segmentation image and the depth estimation image through the image integration step, ensuring the reliability and usability of the final output data. Pixel-level and regional-level information fusion. Pixel-level fusion extracts depth information according to the segmentation mask and combines the depth image with the segmentation image, improving the integrity of the particle area. Regional-level fusion extracts the local depth distribution features of the material particles and generates a clear particle model in combination with the segmentation contour, facilitating subsequent analysis. Image integration eliminates the accuracy loss that may be caused by single-image errors through the joint processing of multi-source data, thereby significantly improving the accuracy and stability of the final detection results.

[0086] In an exemplary embodiment, S5 can be replaced by the following steps.

[0087] S51: Detect the edges of the particles in the visualization image to determine the set of edge contour pixel points.

[0088] S52: Based on the set of edge contour pixel points, calculate the particle size parameters of the material, divide the particle size regions, count the proportion of the material quantity in each particle size region, and determine the particle size distribution.

[0089] In practical applications, image integration and statistics include:

[0090] (1) Jointly process the segmented image and the depth estimation image to generate an integrated image containing the material contour and depth information. Mark the foreground region in the segmented image as the region of interest (ROI), and extract its corresponding depth value. Fill or interpolate the depth values of the non-material regions in the depth image to enhance the integrity of the integration result.

[0091] Where S(x, y) is the binary mask of the segmented image and D(x, y) is the depth image.

[0092] According to the segmentation result, divide the depth information into regions, extract the regional depth features of each material particle, combine the depth features with the particle contour information, overlay the segmented image and the corrected depth image, and generate a visualization image containing the material contour, regional division, and depth value. Use pseudo-color to represent the depth information to visually display the shape and distance distribution of the particles.

[0093] (2) Particle size calculation and evaluation:

[0094] Perform edge detection (Canny edge detection) on the fused image to obtain the set of edge contour pixel points E = {e 1 , e 2 , … e k}. Convert the edge contour to a three-dimensional space curve in combination with the depth information. Assume that the edge point e i = (x i , y i ), and the corresponding depth value is D i , then the three-dimensional space curve point is (x i , y i , D i ).

[0095] Calculate the particle size parameters of the material. For spherical materials, the equivalent diameter can be calculated according to the three-dimensional contour where V is the volume estimated according to the three-dimensional contour (which can be calculated by the integral method).

[0096] Statistical particle size distribution. Assume the particle size interval is [d 1 , d 2 , … d m, count the proportion P of the material quantity in each interval i , evaluate whether the material meets the production process requirements, and judge whether it meets the expected mean μ of the particle size distribution d and the standard deviation σ d Requirements:

[0097] If (T p is the deviation threshold allowed by the process, T p = 2), it is considered that the material meets the requirements, otherwise corresponding measures (such as screening, reprocessing, etc.) are taken.

[0098] This application takes the end-to-end integrated design as the core. Through rough segmentation information acquisition, cascade detection model optimization, and the integration of segmentation and depth estimation, it realizes high efficiency, significantly reduces data redundancy and processing time. High precision, significantly optimizes the accuracy of material area segmentation, size, and depth estimation. Comprehensiveness and intuitiveness, generates integrated visualization results, which is convenient for subsequent analysis and application. These advantages make this solution show strong competitiveness in the fields of industrial material detection, quality control, and particle size evaluation, and it is an important improvement and enhancement to the existing technology.

[0099] In practical applications, since this model considers various factors such as the shape, size, material of different materials, as well as light and working environment, by using a diverse dataset and introducing prior knowledge in network training, it can adapt to complex and changing industrial production scenarios. Whether it is detecting small particle materials, medium-sized block materials, or large industrial components, whether it is under strong light, weak light, or natural light conditions, this model can work stably and maintain high detection accuracy. In the process of image acquisition and processing, the anti-interference ability is improved through various technical means. It has significant advantages compared with other material detection methods in terms of accuracy, efficiency, adaptability, and stability.

[0100] 1. Detection method based on acoustic imaging.

[0101] Principle: Utilize the reflection, scattering, etc. characteristics of sound waves on the surface and inside of the material, collect sound signals through multiple acoustic sensors, and construct an acoustic image of the material through signal processing and imaging algorithms. The structure and particle size information of the material can be analyzed from the acoustic image. In the detection of blast furnace materials, an acoustic sensor array can be arranged around the material container.

[0102] Advantages: It has a certain detection ability for the internal structure of the material. In addition to the surface particle size, it can also provide information about the internal particle aggregation state of the material. It is not affected by light conditions and can work normally in dark or strong light environments.

[0103] Disadvantages: The resolution of acoustic imaging is relatively low, and the image quality is relatively poor, which may cause large errors in precise particle size calculation. It is vulnerable to ambient noise interference. In an industrial environment such as a blast furnace, noises generated by mechanical operation may affect the acquisition and processing of acoustic signals.

[0104] 2. Laser scanning technology.

[0105] Principle: The laser scanning system obtains three-dimensional point cloud data of the material surface by emitting laser beams and receiving reflected light. These point cloud data can accurately represent the shape and position information of the material. By processing the point cloud data, such as operations like segmentation and fitting, the particle size of the material can be calculated. In the material detection of blast furnace ironmaking, the laser scanning device can be installed above the material conveying device to scan the passing materials and obtain a three-dimensional model of the material pile.

[0106] Advantages: It can provide high-precision three-dimensional information, and can well obtain the geometric shape of materials with complex shapes and stacking states, so as to accurately calculate the particle size. It can quickly scan a large area of materials, with high detection efficiency, and is suitable for scenarios with continuous feeding such as blast furnace charging.

[0107] Disadvantages: The equipment cost is relatively high, including laser emission and receiving devices, high-precision scanning mechanisms, and complex data processing software. It is sensitive to ambient light and the reflection characteristics of the material surface. In strong light environments or when the material surface has strong reflectivity or strong light absorption, etc., it may affect the reception of laser reflection signals, resulting in inaccurate data.

[0108] Based on the same inventive concept, the embodiments of the present application also provide a particle size detection device for a material particle image for implementing the particle size detection method of the material particle image involved above. The implementation solutions provided by this device to solve problems are similar to the implementation solutions recorded in the above method. Therefore, the specific limitations in one or more embodiments of the particle size detection device for a material particle image provided below can refer to the limitations on the particle size detection method of the material particle image in the above text, and will not be elaborated here.

[0109] In an exemplary embodiment, a particle size detection device for a material particle image is provided, including:

[0110] A visual detection device for collecting material particle images from multiple perspectives.

[0111] A rough segmentation module for roughly segmenting the material particle image, marking the material particle area and the background area of the material particle image, and determining the roughly segmented image.

[0112] An image segmentation model for processing the roughly segmented image to determine a material segmentation map and a depth estimation map.

[0113] An image integration and statistics module, which is used to jointly process the material segmentation map and the depth estimation map to generate a visual image including material contours, area division, and depth information.

[0114] A material particle information determination module, which is used to determine material particle information according to the visual image; the material particle information includes material particle size parameters and particle size distribution.

[0115] In an exemplary embodiment, a computer device is provided. The computer device can be a server or a terminal. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store particle size detection data of material particle images. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with an external terminal through a network connection. When the computer program is executed by the processor, it implements a method for detecting the particle size of a material particle image.

[0116] In an exemplary embodiment, a computer device is provided, including a memory and a processor. A computer program is stored in the memory, and when the processor executes the computer program, the above method is implemented.

[0117] In an exemplary embodiment, a computer-readable storage medium is provided, storing a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0118] In an exemplary embodiment, a computer program product is provided, including a computer program, and when the computer program is executed by a processor, the above method is implemented.

[0119] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in the present application can include at least one of non-volatile and volatile memories. Non-volatile memories can include read-only memory (ROM), magnetic tapes, floppy disks, flash memories, optical memories, high-density embedded non-volatile memories, resistive random-access memories (ReRAM), magnetoresistive random-access memories (MRAM), ferroelectric random-access memories (FRAM), phase change memories (PCM), graphene memories, etc. Volatile memories can include random-access memory (RAM) or external cache memories, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random-access memory (SRAM) or dynamic random-access memory (DRAM), etc.

[0120] In this application, all actions of obtaining signals, information, or data are carried out on the premise of complying with the corresponding data protection regulations and policies of the country where the location is located and obtaining authorization from the owner of the corresponding device.

[0121] The databases involved in the embodiments provided in the present application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in the present application can be general-purpose processors, central processors, graphics processors, digital signal processors, programmable logics, data processing logics based on quantum computing, etc., without limitation.

[0122] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this specification.

[0123] In this text, specific examples are used to elaborate on the principles and implementation manners of this application. The description of the above embodiments is only used to help understand the method and its core idea of this application; at the same time, for those of ordinary skill in the art, according to the idea of this application, there will be changes in the specific implementation manners and application scopes. In summary, the content of this specification should not be construed as a limitation to this application.

Claims

1. A particle size detection method for a material particle image, characterized in that: The particle size detection method of the material particle image comprises: Collect material particle images from multiple perspectives; Roughly segmenting the material particle image, marking the material particle area and the background area of ​​the material particle image, and determining the image after rough segmentation; Processing the roughly segmented image using an image segmentation model to determine a material segmentation map and a depth estimation map; The material segmentation map and the depth estimation map are jointly processed to generate a visual image including material contours, area divisions and depth information; The material particle information is determined according to the visual image; the material particle information includes material particle size parameters and particle size distribution.

2. The particle size detection method of material particle image according to claim 1, characterized in that: Multi-angle acquisition of material particle images, previously also included: Determine the number of cameras based on the geometry and size of the material transmission channel; According to the minimum detection size of the material and the expected imaging size on the image sensor, the focal length of the camera is determined using the principle of similar triangles; Determine the resolution of the image sensor based on the detection accuracy requirements; The lighting system is arranged in a ring or strip distribution manner within the set distance range of the camera.

3. The particle size detection method of material particle image according to claim 1, characterized in that: Roughly segmenting the material particle image, marking the material particle area and the background area of ​​the material particle image, and determining the image after rough segmentation, specifically includes: grayscale the material particle image to determine a grayscale image; Normalizing the pixel values ​​of the grayscale image, and using the local binary pattern to calculate the relationship between the neighborhood pixel value and the central pixel value in the local window corresponding to each pixel point, to determine the local binary pattern feature map of the image; A connected region algorithm is used to perform region marking on the local binary pattern feature map of the image, marking the material particle region and the background region, and determining the image after rough segmentation.

4. The particle size detection method of material particle image according to claim 1, characterized in that: The image after rough segmentation is processed by using an image segmentation model to determine a material segmentation map and a depth estimation map, specifically including: The image segmentation model is trained according to conventional image samples, conventional image sample annotation data, and a material image dataset with dimension annotations; the image segmentation model includes an image segmentation stage and a depth estimation stage; In the image segmentation stage, a feature map of the roughly segmented image is extracted, and a multi-level convolution and pooling operation is performed on the feature map using a trained image segmentation model to determine feature maps of different scales; Based on feature maps of different sizes, according to the material contour features, texture features and multi-view set relationships, combined with pre-learned size priors, the size ratio relationship is simply calculated to generate a material segmentation image; In the depth estimation stage, multi-scale feature extraction is performed on the material segmentation image to determine a multi-scale feature map; The multi-scale feature maps are fused using a feature pyramid network, and depth prediction and optimization are performed on the fused feature maps to determine a depth estimation map.

5. The particle size detection method of material particle image according to claim 1, characterized in that: The material segmentation map and the depth estimation map are jointly processed to generate a visual image containing material contours, area divisions and depth information, specifically including: Marking the foreground area in the material segmentation image as an area of ​​interest, extracting the depth value corresponding to the area of ​​interest, and filling or interpolating the depth value of the non-material area in the depth estimation image to generate an integrated image containing material contour and depth information; Based on the integrated image, the regional depth feature of each material particle is extracted, and the regional depth feature is combined with the particle contour information to generate a visual image including material contour, regional division and depth information.

6. The particle size detection method of material particle image according to claim 1, characterized in that: Determining material particle information according to the visual image specifically includes: Performing particle edge detection on the visualized image to determine an edge contour pixel point set; Based on the edge contour pixel point set, the particle size parameters of the material are calculated, and the particle size areas are divided, and the proportion of the material quantity in each particle size area is counted to determine the particle size distribution.

7. A particle size detection device for material particle images, characterized in that: The particle size detection device of the material particle image comprises: Visual inspection device, used to collect material particle images from multiple perspectives; A coarse segmentation module, used to perform coarse segmentation on the material particle image, mark the material particle area and the background area of ​​the material particle image, and determine the image after coarse segmentation; An image segmentation model is used to process the roughly segmented image to determine a material segmentation map and a depth estimation map; An image integration and statistics module, used for jointly processing the material segmentation map and the depth estimation map to generate a visual image containing material contours, area divisions and depth information; The material particle information determination module is used to determine the material particle information according to the visual image; the material particle information includes material particle size parameters and particle size distribution.

8. A computer device comprising: A memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor executes the computer program to implement the particle size detection method for material particle images described in any one of claims 1 to 6.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the method for detecting the particle size of a material particle image according to any one of claims 1 to 6 is implemented.

10. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the method for detecting the particle size of a material particle image according to any one of claims 1 to 6 is implemented.

Images