Automated stacking and reclaiming method and system for belt-conveyed materials based on image recognition

Abstract

The present application provides an automated stacking and reclaiming method and system for belt-conveyed materials based on image recognition, which relates to the technical field of image processing. The method includes: constructing a material feature level model; collecting the first material stacking information; performing feature recognition on the first material image information to obtain the humidity level of the material; calculating the first reclaiming value of the material to obtain the first reclaiming sequence; the reclaiming machine reclaims materials according to the sorting of the first reclaiming sequence; collecting the second material stacking information; obtaining the volume level of the material; calculating the second reclaiming value of the material to obtain the second reclaiming sequence; the reclaiming machine performs secondary reclaiming according to the sorting of the second reclaiming sequence; the reclaiming machines set at the final sorting ports respectively reclaim materials to obtain a plurality of uniform material piles. The technical problems of low efficiency, high safety hazards and easy classification errors existing in the prior art are solved, and the technical effects of enhancing efficiency, reducing risks, accurately classifying materials and thus obtaining uniform material piles are achieved.

Description

Technical Field

[0001] This application relates to the field of image processing technology, and particularly to an automated stacking and reclaiming method and system for belt-conveyed materials based on image recognition. Background Art

[0002] In a cement plant, since it is necessary to ensure that the materials meet the corresponding humidity and volume standards during the mixing process to avoid affecting the grindability of the materials, the output of the mill, as well as the strength and storage performance of the finally produced cement, each type of material needs to be classified by humidity and volume of the lumps before mixing, so as to facilitate subsequent drying and crushing operations to obtain standard materials. To solve this problem, traditional classification methods set multiple discharge points on the belt conveyor, and manually unload materials with different humidities and volumes, and then classify and stack them. However, this method has significant drawbacks. First, the efficiency of manual operation is low and cannot meet the needs of large-scale raw material classification. Second, there are safety hazards in manual operation. For example, working in a high-temperature and high-dust environment is likely to cause occupational diseases. Most importantly, manual operation inevitably brings subjectivity and instability during the material classification process, which is likely to lead to classification errors and is difficult to ensure the clarity and accuracy of material classification. The mixed material piles formed in this case have uneven humidity and particle size, and cannot be uniformly processed during subsequent drying and crushing operations, nor can they be mixed into a uniformly composed material pile, which will directly affect the quality of the final product.

[0003] In summary, in the prior art, there are technical problems such as low efficiency of manual operation, high safety hazards, easy occurrence of classification errors, difficulty in ensuring the clarity and accuracy of material classification, inability to achieve the goal of uniform mixing, and direct impact on the quality of the final product. Therefore, a method is needed to solve the above problems. Summary of the Invention

[0004] The present disclosure provides an automated stacking and reclaiming method and system for belt-conveyed materials based on image recognition, which is used to solve the technical problems in the prior art, such as low efficiency of manual operation, high safety hazards, easy occurrence of classification errors, difficulty in ensuring the clarity and accuracy of material classification, inability to achieve the goal of uniform mixing, and direct impact on the quality of the final product.

[0005] According to the first aspect of the present disclosure, there is provided an automated stacking and reclaiming method for belt-conveyed materials based on image recognition, including:

[0006] Construct a material feature level model according to the historical production record data of cement plant materials, and the material feature level model includes a material humidity level table and a material volume level table;

[0007] Collect the first material accumulation information on the target belt through the first vision sensor. Among them, the first vision sensor is set above the stacker. The first material accumulation information includes the first material position information and the first material image information. The first material position information includes M positioning identifiers;

[0008] The reclaimer set at the first sorting port obtains the first material accumulation information, performs feature recognition on the first material image information, and combines the material feature level model to obtain the humidity levels of the materials with M positioning identifiers;

[0009] Calculate the first reclaiming values of the materials with M positioning identifiers according to the material humidity levels and position information. The first reclaiming value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity level, and the first time value is determined according to the time when the reclaimer moves to the positioning identifier. The order of magnitude of the humidity value is 100 times that of the first time value. Integrate the M first reclaiming values, arrange the first reclaiming values from largest to smallest, and obtain the first reclaiming sequence;

[0010] The reclaimer reclaims materials according to the sorting order of the first reclaiming sequence, respectively takes materials with different humidities to Q different belts, and transports them to the second sorting port through the Q belts;

[0011] Collect the second material accumulation information on the Q belts through Q second vision sensors. The second vision sensors are located above the Q belts at the first sorting port. The second material accumulation information includes the second material position information and the second material image information. The second material position information includes N positioning identifiers;

[0012] The reclaimer set at the second sorting port obtains the second material accumulation information, calculates the material volume in combination with the volume information in the feature level model, and obtains the volume levels of the materials with N positioning identifiers;

[0013] Calculate the second reclaiming values of the materials with N positioning identifiers according to the material volume levels and position information. The second reclaiming value is the sum of the volume value and the second time value. The volume value is determined according to the volume level, and the acquisition method of the second time value is the same as that of the first time value. Integrate the N second reclaiming values, arrange the second reclaiming values from largest to smallest, and obtain the second reclaiming sequence;

[0014] The reclaimer performs secondary reclaiming according to the sorting order of the second reclaiming sequence, respectively takes materials with different volumes to P different belts, and transports them to the final sorting port through the P belts;

[0015] The P reclaimers arranged at the final sorting port respectively take materials to obtain multiple uniform material piles.

[0016] According to a second aspect of the present disclosure, there is provided an automated stacking and reclaiming system for belt-conveyed materials based on image recognition, including:

[0017] A material feature level model construction module, which is used to construct a material feature level model according to the historical production record data of cement plant materials. The material feature level model includes a material humidity level table and a material volume level table;

[0018] A first material stacking information acquisition module, which is used to acquire the first material stacking information on the target belt through a first vision sensor. Among them, the first vision sensor is arranged above the stacker. The first material stacking information includes first material position information and first material image information. The first material position information includes M positioning identifiers;

[0019] A humidity level acquisition module, which is used to obtain the first material stacking information by the reclaimer set at the first sorting port, perform feature recognition on the first material image information, and combine the material feature level model to obtain the humidity levels of the materials at the M positioning identifiers;

[0020] A first reclaiming sequence acquisition module, which is used to calculate the first reclaiming values of the materials at the M positioning identifiers according to the material humidity levels and position information. The first reclaiming value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity level, and the first time value is determined according to the time when the reclaimer moves to the positioning identifier. The order of magnitude of the humidity value is 100 times that of the first time value. Integrate the M first reclaiming values, arrange the first reclaiming values from large to small, and obtain the first reclaiming sequence;

[0021] A primary reclaiming module, which is used for the reclaimer to reclaim materials according to the sorting of the first reclaiming sequence, respectively take materials with different humidities to Q different belts, and transmit them to the second sorting port through the Q belts;

[0022] A second material stacking information acquisition module, which is used to acquire the second material stacking information on the Q belts through Q second vision sensors. The second vision sensors are located above the Q belts at the first sorting port. The second material stacking information includes second material position information and second material image information. The second material position information includes N positioning identifiers;

[0023] A volume level acquisition module, which is used to obtain the second material stacking information by the reclaimer set at the second sorting port, calculate the material volume in combination with the volume information in the feature level model, and obtain the volume levels of the materials at the N positioning identifiers;

[0024] The second material fetching sequence obtaining module is used to calculate the second material fetching values of N positioned marked materials according to the material volume grade and position information. The second material fetching value is the sum of a volume value and a second time value. The volume value is determined according to the volume grade, and the obtaining method of the second time value is the same as that of the first time value. Integrate the N second material fetching values, arrange the second material fetching values from large to small, and obtain the second material fetching sequence;

[0025] The intermediate material fetching module is used for the material fetching machine to perform secondary material fetching according to the sorting of the second material fetching sequence, fetch materials of different volumes to P different belts respectively, and transport them to the final sorting port through the P belts;

[0026] The final material fetching module is used for the P material fetching machines arranged at the final sorting port to fetch materials respectively to obtain a plurality of uniform material piles.

[0027] One or more technical solutions provided in the present disclosure have at least the following technical effects or advantages:

[0028] Construct a material characteristic grade model based on the historical production record data of materials in a cement plant. The material characteristic grade model includes a material humidity grade table and a material volume grade table. Collect the first material accumulation information on the target conveyor belt through a first vision sensor. The first vision sensor is set above the stacker. The first material accumulation information includes first material position information and first material image information. The first material position information includes M positioning identifiers. The reclaimer set at the first sorting port obtains the first material accumulation information, performs feature recognition on the first material image information, and combines the material characteristic grade model to obtain the humidity grades of the materials with M positioning identifiers. Calculate the first reclaiming value of the materials with M positioning identifiers according to the material humidity grade and position information. The first reclaiming value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity grade, and the first time value is determined according to the time when the reclaimer moves to the positioning identifier. The order of magnitude of the humidity value is a hundred times that of the first time value. Integrate the M first reclaiming values, arrange the first reclaiming values from large to small to obtain the first reclaiming sequence. The reclaimer reclaims materials according to the sorting order of the first reclaiming sequence, separately takes materials with different humidities to Q different conveyor belts, and transports them to the second sorting port through the Q conveyor belts. Collect the second material accumulation information on the Q conveyor belts through Q second vision sensors. The second vision sensors are located above the Q conveyor belts at the first sorting port. The second material accumulation information includes second material position information and second material image information. The second material position information includes N positioning identifiers. The reclaimer set at the second sorting port obtains the second material accumulation information, calculates the material volume in combination with the volume information in the characteristic grade model to obtain the volume grades of the materials with N positioning identifiers. Calculate the second reclaiming value of the materials with N positioning identifiers according to the material volume grade and position information. The second reclaiming value is the sum of the volume value and the second time value. The volume value is determined according to the volume grade, and the obtaining method of the second time value is the same as that of the first time value. Integrate the N second reclaiming values, arrange the second reclaiming values from large to small to obtain the second reclaiming sequence. The reclaimer performs secondary reclaiming according to the sorting order of the second reclaiming sequence, separately takes materials with different volumes to P different conveyor belts, and transports them to the final sorting port through the P conveyor belts. The P reclaimers arranged at the final sorting port respectively take materials to obtain a plurality of uniform material piles. It solves the technical problems in the prior art that manual operation has low efficiency, high safety hazards, is prone to classification errors, is difficult to ensure the clarity and accuracy of material classification, cannot achieve the goal of uniform mixing, and directly affects the quality of the final product. It achieves the technical effects of enhancing the sorting efficiency, effectively reducing the safety risk, accurately classifying materials, and thus obtaining uniform material piles.

[0029] The above description is only an overview of the technical solution of this application. In order to understand the technical means of this application more clearly, it can be implemented according to the content of the specification. And in order to make the above and other purposes, features and advantages of this application more obvious and understandable, the specific implementation manners of this application are given below. BRIEF DESCRIPTION OF THE DRAWINGS

[0030] In order to more clearly illustrate the technical solutions in the present disclosure or the prior art, the following will briefly introduce the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings in the following description are only exemplary. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained according to the provided drawings.

[0031] Figure 1 It is a schematic flowchart of the automatic stacking and reclaiming method for belt-conveyed materials based on image recognition provided by an embodiment of this application;

[0032] Figure 2 It is a schematic structural diagram of the automatic stacking and reclaiming system for belt-conveyed materials based on image recognition provided by an embodiment of this application.

[0033] Description of reference numerals: Material feature level model construction module 11, First material stacking information acquisition module 12, Humidity level acquisition module 13, First reclaiming sequence acquisition module 14, Initial reclaiming module 15, Second material stacking information acquisition module 16, Volume level acquisition module 17, Second reclaiming sequence acquisition module 18, Intermediate reclaiming module 19, Final reclaiming module 20. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0034] The following describes exemplary embodiments of the present disclosure in conjunction with the drawings. Various details of the embodiments of the present disclosure are included to assist in understanding, and they should be considered merely exemplary. Therefore, those of ordinary skill in the art should recognize that various changes and modifications can be made to the embodiments described herein without departing from the scope and spirit of the present disclosure. Similarly, for the sake of clarity and conciseness, the description below omits the description of well-known functions and structures.

[0035] Embodiment 1. The automatic stacking and reclaiming method for belt-conveyed materials based on image recognition provided by an embodiment of the present disclosure is hereinafter referred to as Figure 1 for description. The method includes:

[0036] S1: Construct a material feature level model according to the historical production record data of the materials in the cement plant. The material feature level model includes a material humidity level table and a material volume level table;

[0037] Further, constructing the material feature level model includes:

[0038] Extract the type, humidity, and volume data from the historical production records of the materials to obtain the basic material feature set, and preprocess the basic material feature set. The preprocessing includes supplementing missing data values and removing abnormal data values.

[0039] Traverse the basic material feature set, use the material type as the row index, and the corresponding humidity data and volume data of the material as columns to generate a material humidity statistical table and a material volume statistical table.

[0040] Based on the K-means clustering algorithm, classify the material humidity statistical table and the material volume statistical table to obtain a material humidity grade table and a material volume grade table.

[0041] Construct a material feature grade model according to the material humidity grade table and the material volume grade table.

[0042] Specifically, obtain the historical production records of the materials in the cement plant and collect the key information related to the materials. Conduct stratified sampling on the materials of different batches, different production workshops, and different types to ensure that the samples can cover all production situations. Use appropriate measuring tools to measure the humidity and volume of the samples. For example, use a humidity sensor or the drying and weighing method to measure the humidity of the samples, and use measuring tools or the drainage method to measure the volume of the samples. Accurately record the measured humidity and volume data and transmit them to the basic material feature set. Organize the data and check for missing values in the humidity and volume data. If there are missing humidity or volume data for a material, re-detect the material data for supplementation. Check for abnormal values in the data. Determine a reasonable range based on the statistical analysis of historical data. For example, by calculating the mean μ and standard deviation σ of the historical data, judge the data outside the range of μ ± 3σ as abnormal values. If the data of a certain material significantly deviates from the normal range, further check whether there are errors in the measurement process. If it is a measurement error, re-measure. If it is due to special reasons, consider correcting it or excluding the abnormal value during data analysis to ensure the accuracy of subsequent statistical analysis.

[0043] Traverse the basic material feature set, use the material type as the row index, and use its corresponding humidity data and volume data as columns respectively. For example, use material types such as limestone and clay as rows respectively, the humidity data obtained from the sampling measurement of each material constitutes a column, and the volume data constitutes another column, thus generating a material humidity statistical table and a material volume statistical table. The statistical table will provide the basic data structure for subsequent grade classification.

[0044] The humidity level table and volume level table are obtained using the K-means clustering algorithm. Since the measurement values of different material types may have different dimensions and value ranges, in order to make the clustering results more accurate, the data in the statistical table can be standardized. For example, standard deviation standardization is used, and the formula is z = (x - μ) / σ, where z is the standard score, representing the dimensionless value after data standardization, reflecting its deviation from the mean, x is the original data, μ is the mean, and σ is the standard deviation. The appropriate number of clusters k is determined by the elbow method. The elbow method is to plot the sum of squared errors curve for different k values and find the inflection point of the curve. The k value corresponding to this inflection point is the more appropriate number of clusters. Randomly select k measurement values as the initial centroids, calculate the distances from each measurement value to the k centroids, and assign it to the cluster where the nearest centroid is located. For each cluster, recalculate its centroid, which is the average of the measurement values of all material types within the cluster. Iteratively optimize by repeating the steps of assigning clusters and updating centroids until the centroids no longer change or the preset number of iterations is reached. According to the clustering results, the material types are divided into k levels to obtain the humidity level table and volume level table.

[0045] Backtrack the samples according to the humidity level table and volume level table, select the material samples corresponding to the median data at this level, take pictures to obtain image information, and perform feature extraction on the images to obtain the texture features, color features, and volume features of the materials at each level. Link the texture features and color features to the humidity level table, and link the texture features and volume features to the volume level table to construct a material feature level model;

[0046] Specifically, taking limestone materials as an example, the humidity grade table includes three levels. Among them, the first level represents low humidity, the second level represents medium humidity, and the third level represents high humidity. The humidity value of the first-level humidity is mainly between 0 - 0.5%, which is the ideal humidity for the materials. Materials within this range are not prone to caking during storage and can maintain good fluidity during transportation, which is beneficial for subsequent processes such as grinding and calcination. The humidity value of the second-level humidity is mainly between 0.5% - 1.5%. Within this humidity range, although the materials may be affected by moisture to a certain extent, they can still be normally processed in the production process through some auxiliary means. The humidity value of the third-level humidity is greater than 1.5%. At this time, the materials may have obvious caking phenomena and are prone to accumulation during storage, which is not conducive to subsequent production processes. Such materials must be dried, otherwise, processes such as grinding and calcination cannot be carried out normally. Similarly, the volume grade table of limestone materials also includes three levels. Among them, the first level represents small volume, the second level represents medium volume, and the third level represents large volume. Small-volume materials can be quickly and evenly dispersed when mixed with other materials. Medium-volume materials can be mixed evenly with other materials relatively quickly. They are not as difficult to disperse as large-volume materials, nor are they as easily over-dispersed as small-volume materials, which helps to improve the mixing effect in the cement production process. The feeding uniformity of large-volume materials is difficult to control. For example, when using crushers or mills, if the material volume is too large, it may lead to uneven feeding, affecting the crushing or grinding efficiency of the equipment. Large-volume materials may block the feeding port or cause material accumulation inside the equipment, making the equipment unable to work properly. Therefore, subsequent crushing operations are required.

[0047] S2: Collect the first material accumulation information on the target belt through the first vision sensor. Among them, the first vision sensor is set above the stacker. The first material accumulation information includes the first material position information and the first material image information. The first material position information includes M positioning marks;

[0048] Further, collecting the first material accumulation information on the target belt includes:

[0049] Obtain the image information of the material through the first vision sensor;

[0050] Preprocess the obtained image. The preprocessing includes grayscale conversion, filtering, enhancement, and histogram equalization;

[0051] Based on the SSD target detection algorithm, identify and locate the material positions in the image, and generate M positioning marks;

[0052] Divide the image into M regional images according to the M positioning marks, and summarize, compress, and store them in the first material accumulation information.

[0053] Specifically, a first sensor is arranged above the stacker to ensure that the sensor can completely cover the target belt area and avoid visual blind spots. The stacker stacks materials of the same type onto the conveyor belt to obtain multiple evenly spread material stacks. Set the triggering device of the first sensor, and trigger the sensor regularly according to the belt running speed to collect images at fixed time intervals.

[0054] Obtain image data, and convert the RGB image to a grayscale image using the weighted average method. The formula is: Gray = 0.299R + 0.587G + 0.114B, where Gray represents the grayscale value, and R, G, and B represent the values of the red, green, and blue channels respectively.

[0055] Perform Gaussian filtering on the image data based on the Gaussian function. Generate a Gaussian kernel according to the required filter size and standard deviation, and perform boundary processing on the image. Add extra pixel rows and columns at the image edges to achieve zero-padding operation. Align each pixel in the image with the Gaussian kernel, perform weighted summation according to the weights of the filter, and perform convolution calculation to generate new pixel values. According to the above convolution operation, starting from the top-left pixel of the image, process each pixel in the image row by row and column by column until the bottom-right pixel to complete the traversal of the image and generate a smoothed image.

[0056] Perform enhancement processing on the generated smoothed image again, including contrast stretching and sharpening. Contrast stretching is mainly achieved by adjusting the grayscale value range of the image. For example, expand the grayscale value range of the original image from a relatively narrow interval to the entire grayscale value range. Assume that the grayscale values of the original image are concentrated between 50 - 150. Through contrast stretching, map the pixel with a grayscale value of 50 to 0, the pixel with a grayscale value of 150 to 255, and reassign the grayscale values of the intermediate pixels according to a linear or non-linear mapping relationship to enhance the contrast of the image. Sharpening processing requires calculating the gradient of the image and superimposing the gradient value on the original image to make the edges in the image more obvious, facilitating the subsequent distinction of the boundaries between materials and the boundaries between materials and the background.

[0057] Obtain the filtered and enhanced picture and perform histogram equalization. Calculate the histogram of the original image, count the number of times each grayscale value appears in the image, and calculate the cumulative distribution function according to the histogram. The cumulative distribution function represents the proportion of pixels with grayscale values less than or equal to a certain grayscale value in the image. According to the cumulative distribution function, map the grayscale value of each pixel in the original image to a new grayscale value, making the histogram of the new image close to a uniform distribution and clarifying the details of the dark and bright parts in the image.

[0058] Based on the SSD object detection algorithm, perform position recognition on the preprocessed image data to obtain M positioning identifiers. The specific steps include: generating multi-layer feature maps through multiple convolutional layers and pooling layers of the VGG16 convolutional neural network to obtain features of different scales; for the pixel points in each feature map, set multiple anchor boxes of different sizes and aspect ratios for coverage; predict the target appearance area by obtaining the target category probability and the position offset of the anchor box relative to the true target box; perform non-maximum suppression on the prediction results to remove redundant detection boxes to obtain the actual position of the material; according to the actual position of the material, set the positioning identifier, and number the positions of each material pile from 1 to M, where M < 100.

[0059] According to the distance relationship between the identification numbers, use a clustering algorithm to determine the range of each area, thereby dividing the image into M regional images, arranging the regional images in the order of the identification numbers, and storing them in the first material stacking information.

[0060] S3: The material taking machine set at the first sorting port obtains the first material stacking information, performs feature recognition on the first material image information, and combines with the material feature level model to obtain the humidity levels of the M positioning identifier materials;

[0061] Further, obtaining the humidity levels of the M positioning identifier materials includes:

[0062] Obtain the first material stacking information, and perform key feature extraction on the M regional images respectively. The key features mainly include texture features and color features;

[0063] Input the texture features of the M regional images into the material feature level model to determine the types of the materials transmitted this time and the corresponding humidity level tables for the types;

[0064] Compare the color information of the M regional images with the humidity level tables to obtain the humidity levels of the M positioning identifier materials.

[0065] Specifically, the material taking machine set at the first sorting port acquires M regional images in the first material accumulation information. The regional images are converted into grayscale images by the weighted average method, and feature extraction is performed on the grayscale images, including texture feature extraction and color feature extraction. The texture feature extraction is obtained according to the gray-level co-occurrence matrix (GLCM). Set the GLCM parameters, including the distance d and the direction θ. Based on each pixel point in the image, construct the gray-level co-occurrence matrix. Among them, the distance d is set to 1, representing adjacent pixel points, and the direction θ is set to 0°, 45°, 90°, and 135°, representing the adjacent pixel points in the 0°, 45°, 90°, and 135° directions of the selected pixel point. For each pixel point in the image, according to the set distance and direction, set the gray values of it and its adjacent pixel points as the row and column indexes of the matrix respectively, and construct the co-occurrence matrix by increasing the count at this position in the matrix. For example, if the gray value of a pixel is i and the gray value of its adjacent pixel is j, then the count in the i-th row and j-th column of the GLCM is increased by 1. Divide the value of each element by the total number of pixel pairs to convert the count in the co-occurrence matrix into probability, and realize the conversion of the value of GLCM to probability distribution, which is convenient for subsequent feature extraction.

[0066] Extract the GLCM texture feature parameters, and the calculation formulas are as follows:

[0067] Contrast:

[0068] Where P(i, j) represents the value of the element (i, j) in the GLCM, and n represents the number of gray levels. The contrast reflects the clarity and local variation degree of the texture in the image. A high contrast indicates that the texture changes violently.

[0069] Energy:

[0070] Energy represents the uniformity of the image texture. The higher the energy value, the more uniform the texture.

[0071] Entropy:

[0072] Entropy reflects the randomness of the image texture. The higher the entropy value, the more complex and random the texture.

[0073] Correlation:

[0074] Where μ i and μ j represent the means of the row and column respectively, and σ i and σ j represent the standard deviations of the row and column respectively. The correlation reflects the linear relationship between pixel points of different gray levels in the image.

[0075] Combine the texture feature parameters such as contrast, energy, entropy, and correlation calculated above into a texture feature vector V = [Contrast, Energy, Entropy, Correlation] for comparison with the texture features in the material feature level model.

[0076] Calculate the mean values of the image in the R, G, and B channels. The formula is:

[0077] Mean value of the R channel:

[0078] where I represents the image, (x, y) are the pixel coordinates, M and N are the height and width of the image respectively, and I R (x, y) is the R channel value of the pixel at coordinate (x, y). Similarly, the mean values of the G and B channels can be calculated.

[0079] Calculate the variance of each channel to reflect the degree of color dispersion. The formula is:

[0080] Variance of the R channel:

[0081] where μ R is the mean value of the R channel, and the other symbols are the same as those described above. Similarly, the variances of the G and B channels can be calculated.

[0082] Divide the value ranges of the R, G, and B channels into several intervals respectively. For example, if each channel is divided into 16 intervals, there will be a total of 16×16×16 = 4096 intervals. Count the number of pixels in each interval to form a color histogram. This histogram can reflect the distribution of different colors in the image. Construct a feature vector V RGB = [μ R , σ R , μ G , σ G , μ B , σ B , h R , h G , h B , where h R , h G , h B are certain statistics of the color histograms of the R, G, and B channels, such as the peak value of the histogram. This color feature vector is used for comparison with the color features in the humidity level table.

[0083] Obtain the material feature level model and retrieve the texture feature data of different types of materials in the model. Based on the Euclidean distance, compare the texture features. The Euclidean distance formula is Where xi is the i-th element in the regional image texture feature vector, yi is the i-th element in the texture feature vectors of each material in the material feature level model, and n is the dimension of the feature vector. In this formula, the value of n is 4, corresponding to the four features of contrast, energy, entropy, and correlation mentioned above. For each material in the material feature model, calculate the Euclidean distance between the two texture feature vectors, determine the minimum distance, and judge the material type. According to the determined type, call the humidity level table corresponding to the material type to obtain the color features corresponding to each humidity level. Similarly, use the method of calculating the Euclidean distance to compare the color features of the regional image to obtain the humidity levels of the M identification points.

[0084] S4: Calculate the first material taking value of the M positioning identification materials according to the material humidity level and position information. The first material taking value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity level, and the first time value is determined according to the time when the material taking machine moves to the positioning identification. The order of magnitude of the humidity value is a hundred times that of the first time value. Integrate the M first material taking values, arrange the first material taking values from large to small, and obtain the first material taking sequence.

[0085] Further, obtaining the first material taking sequence includes:

[0086] Set the humidity value H. The formula for the humidity value is H = humidity level × 100.

[0087] Obtain the M positioning identifications, calculate the time required for the material taking machine to move to each positioning identification respectively, obtain the M material taking times, sort the material taking times, and assign the values 1 to M in order to generate the material taking time sequence.

[0088] Set the first time value T1. The formula for the first time value is T1 = material taking time sequence × 1.

[0089] Calculate the first material taking value. The first material taking value is the sum of the humidity value H and the first time value T1. Integrate the M first material taking values for sorting, trace back the positioning identifications corresponding to each value, and obtain the first material taking sequence.

[0090] Specifically, obtain M positioning identifiers and the humidity levels of the materials at each position. Set a humidity value H, which is determined according to the material humidity level. The higher the level, the larger the value. The formula is H = humidity level × 100. For example, if the humidity level of the material under a positioning identifier is 3, then its humidity value is 300. According to the material positions, the initial position of the reclaimer, and the moving speed of the reclaimer, calculate the time required for the reclaimer to move to each material position, and obtain M reclaiming times. Sort the M reclaiming times in descending order to generate a reclaiming time sequence table, and assign values from 1 to M in order, covering the original reclaiming time values. The reclaiming time sequence table describes the position relationship of each material with respect to the reclaimer from far to near. Calculate the first time value T1, which reflects the reclaiming time. The shorter the time, the larger the value. The formula is T1 = reclaiming time sequence × 1.

[0091] Calculate the first reclaiming value and obtain the first reclaiming sequence. By adding the humidity values of the materials at each positioning identifier and the first time value, obtain M first reclaiming values, such as specific values like 301, 106, etc. Re-sort these M first reclaiming values in descending fractional order to obtain the first reclaiming sequence. Given that the order of magnitude of the humidity value is a hundred times that of the first time value, the leading indicator for sorting is the material humidity level, and the subsequent indicator is the position relationship of the materials. In this way, it can ensure that materials with different humidities are divided and the materials with each humidity are grabbed in order from near to far.

[0092] S5: The reclaimer reclaims materials according to the sorting of the first reclaiming sequence, takes materials with different humidities to Q different belts respectively, and transports them to the second sorting port through the Q belts;

[0093] Specifically, the reclaimer classifies and reclaims materials according to the first reclaiming order of the materials, takes materials with the same humidity level from near to far to the same belt, and obtains evenly spread materials. For example, detect the first reclaiming value, gather materials with a reclaiming value above 300 from near to far onto the same conveyor belt. When all materials with a reclaiming value above 300 are sorted, the reclaimer changes the belt carrying the materials and reclaims the materials with a reclaiming value above 200 again, and so on until all reclaiming is completed. The materials are divided into Q belts for transportation.

[0094] S6: Collect the second material stacking information on the Q belts through Q second vision sensors. The second vision sensors are located above the Q belts at the first sorting port. The second material stacking information includes second material position information and second material image information. The second material position information includes N positioning identifiers;

[0095] Specifically, a second sensor is set above the picker at the first sorting port to collect image information. The setting requirements and triggering device of the second sensor are the same as those of the first sensor. Since the materials have been sorted onto Q belts for transmission at this time, the number of second sensors is also Q. The subsequent volume classification of the materials also occurs on Q belts. Here, only one example is given to illustrate the process.

[0096] Obtain the image data collected by the second sensor, and perform preprocessing operations on the image such as grayscale conversion, filtering, enhancement, and histogram equalization to obtain clear and smooth image information. Apply the SSD object detection algorithm to perform position recognition on the preprocessed image data to obtain N positioning identifiers, and set the value of N to be less than 100. Based on the clustering algorithm, divide the image into N regional images, arrange the regional images in the order of the identifier numbers, and store them in the second material stacking information.

[0097] S7: The picker set at the second sorting port obtains the second material stacking information, and combines the volume information in the feature level model to calculate the volume of the materials to obtain the volume levels of the materials with N positioning identifiers;

[0098] Furthermore, obtaining the volume levels of the materials with N positioning identifiers includes:

[0099] Obtain the second material stacking information, and perform key feature extraction on each of the N regional images. The key features mainly include texture features and shape features;

[0100] Combine the texture features of the N regional images and the material feature level model to determine the type of the materials transmitted this time and the corresponding volume level table for this type;

[0101] Estimate the size of the object by calculating the size of the bounding box, and obtain the size information of the N regional images;

[0102] Compare the size information of the M regional images with the volume level table to obtain the volume levels of the materials with N positioning identifiers.

[0103] Specifically, obtain the N regional images in the second material stacking information, and perform feature extraction on each image, including texture feature extraction and shape feature extraction. The texture feature extraction and comparison are the same as the texture feature extraction and comparison in step S3, including obtaining the texture feature vector according to the gray-level co-occurrence matrix GLCM, calculating the Euclidean distance between this feature vector and the texture feature data in the material feature level model to determine the minimum distance, and judging to obtain the material type. The difference is that this comparison obtains the corresponding volume level table for this material type according to the material type, rather than the humidity level table.

[0104] Based on the Canny edge detection algorithm, the shape and contour of the object are determined. The gradient amplitude and direction of the image are calculated by using the first-order partial derivative finite difference to obtain a preliminary edge response map. Then non-maximum suppression is performed, and the pixel points with the local maximum gradient amplitude are retained as edge points, so as to obtain more accurate edges. Finally, double threshold detection and regional connectivity are used to distinguish strong edges, weak edges and non-edges, and weak edges are connected with strong edges to obtain complete and accurate edges.

[0105] For the extracted complex contour, a polygonal approximation method is used to simplify it into a polygon. For example, the Douglas-Peucker algorithm is used to merge the points on the contour that are close to each other by setting a distance threshold, so as to obtain a polygon that can approximate the original contour. The geometric shape that the object may approximate is judged based on the number of sides, side length ratio, angle and other features of the polygon. If the object is judged to be approximately a cuboid, the pixel lengths lp and wp in the length and width directions of the object are obtained, and at the same time, according to the known image scale s, the actual length l = lp × s and the width w = wp × s are calculated. For the height h, since the shape of the material conveyed by the belt is roughly circular or block-shaped, its simplified three-dimensional shape is mostly a cube or a cuboid. In order to reduce the amount of calculation, the height of each material block is directly set to the mean of the length and width of the material block to avoid excessive errors in the final volume calculation. The volume V = l × w × h of the cuboid is calculated, and the size information of the N regional images is finally obtained. The volume level table is retrieved, the difference between the image size information and each volume level is calculated, the minimum difference is obtained, and the volume level of the N identification points is obtained.

[0106] S8: Calculate the second material fetching values of N positioning marked materials according to the material volume level and position information, the second material fetching value is the sum of the volume value and the second time value, the volume value is determined according to the volume level, the second time value is obtained in the same way as the first time value, integrate the N second material fetching values, arrange the second material fetching values from large to small, and obtain the second material fetching sequence;

[0107] Further, a second material extraction sequence is obtained, including:

[0108] Set a volume value V, the volume value formula is V=volume level×100;

[0109] Get N positioning marks, calculate the time required for each mark to fetch materials, sort and assign values to generate the fetching time sequence, and then obtain the second time value T2, the formula is T2 = fetching time sequence × 1;

[0110] Calculate the second material fetching value, integrate N second material fetching values and sort them, trace back the location identifier corresponding to each value, and obtain the second material fetching sequence.

[0111] Specifically, obtain N positioning identifiers and the corresponding volume grades of the materials. Set the volume value V, which is determined according to the material volume grade. The higher the grade, the larger the value. The formula is V = volume grade × 100. For example, if the material volume grade under this positioning identifier is 2, then its volume value is 200. According to the material positions, the initial position of the reclaimer, and the moving speed of the reclaimer, calculate the time required for the reclaimer to move to each material position, and obtain N reclaiming times. Sort the N reclaiming times in descending order, generate a reclaiming time sequence table, and assign values from 1 to N in order to overwrite the original reclaiming time values. Calculate the second time value T2, and the formula is T2 = reclaiming time sequence × 1.

[0112] Calculate the second reclaiming value, which is the sum of the volume value and the second time value. Sort the second reclaiming values in descending order to obtain the second reclaiming sequence. Divide the materials of different volumes, and grab the materials of each volume in the order from near to far.

[0113] S9: The reclaimer performs secondary reclaiming according to the sorting of the second reclaiming sequence, takes the materials of different volumes to P different belts respectively, and transports them to the final sorting port through the P belts;

[0114] Specifically, the reclaimer classifies and reclaims according to the second reclaiming order of the materials, and takes the materials of the same volume from near to far to the same belt. Since at the first sorting port, the materials have been classified by humidity and transported by Q belts, this reclaiming is a secondary classification of the materials transported by the Q belts. Assume that each belt will be divided into R belts for transportation after volume classification, then after this sorting, there will be P belts transporting materials in the end, where P = Q × R.

[0115] S10: The P reclaimers arranged at the final sorting port reclaim respectively to obtain multiple uniform material piles.

[0116] Through the P reclaimers arranged at the final sorting port, reclaim the materials on the P belts, and finally obtain P material piles with consistent humidity and volume, complete the classification and division of the materials, and ensure the consistency of the properties of the material piles.

[0117] Embodiment 2, based on the same inventive concept as the automated stacking and reclaiming method for belt-conveyed materials based on image recognition in the foregoing embodiment, the present application also provides an automated stacking and reclaiming system for belt-conveyed materials based on image recognition. Please refer to the appendix Figure 2 The system includes:

[0118] A material feature grade model construction module 11, which is used to construct a material feature grade model according to the historical production record data of the materials in the cement plant. The material feature grade model includes a material humidity grade table and a material volume grade table;

[0119] The first material accumulation information acquisition module 12 is configured to acquire the first material accumulation information on the target belt through the first vision sensor. The first vision sensor is arranged above the stacker. The first material accumulation information includes the first material position information and the first material image information. The first material position information includes M positioning identifiers;

[0120] The humidity level acquisition module 13 is configured to acquire the first material accumulation information by the reclaimer arranged at the first sorting port, perform feature recognition on the first material image information, and combine with the material feature level model to acquire the humidity levels of the materials with M positioning identifiers;

[0121] The first material retrieval sequence acquisition module 14 is configured to calculate the first material retrieval values of the materials with M positioning identifiers according to the material humidity levels and position information. The first material retrieval value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity level, and the first time value is determined according to the time when the reclaimer moves to the positioning identifier. The order of magnitude of the humidity value is a hundred times that of the first time value. Integrate the M first material retrieval values, arrange the first material retrieval values from largest to smallest to obtain the first material retrieval sequence;

[0122] The initial material retrieval module 15 is configured to retrieve materials by the reclaimer in the order of the first material retrieval sequence, retrieve materials with different humidities to Q different belts respectively, and transport them to the second sorting port through the Q belts;

[0123] The second material accumulation information acquisition module 16 is configured to acquire the second material accumulation information on the Q belts through Q second vision sensors. The second vision sensors are located above the Q belts at the first sorting port. The second material accumulation information includes the second material position information and the second material image information. The second material position information includes N positioning identifiers;

[0124] The volume level acquisition module 17 is configured to acquire the second material accumulation information by the reclaimer arranged at the second sorting port, calculate the material volume by combining with the volume information in the feature level model, and acquire the volume levels of the materials with N positioning identifiers;

[0125] The second material fetching sequence obtaining module 18 is configured to calculate the second fetching values of N positioned and identified materials according to the material volume grade and position information. The second fetching value is the sum of a volume value and a second time value. The volume value is determined according to the volume grade, and the obtaining method of the second time value is the same as that of the first time value. Integrate the N second fetching values, arrange the second fetching values from largest to smallest, and obtain the second fetching sequence;

[0126] The intermediate material fetching module 19 is configured to enable the material fetching machine to perform secondary material fetching according to the sorting of the second fetching sequence, fetch materials of different volumes to P different belts respectively, and transport them to the final sorting port through the P belts;

[0127] The final material fetching module 20 is configured to fetch materials respectively by P material fetching machines arranged at the final sorting port to obtain multiple uniform material piles.

[0128] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise 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.

[0129] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be obvious to those skilled in the art, and the general principles defined herein can be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will be accorded the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An automatic belt conveyor material stacking and reclaiming method based on image recognition, characterized in that The method includes: Constructing a material characteristic level model based on the historical production record data of materials in a cement plant, where the material characteristic level model includes a material humidity level table and a material volume level table; Collecting the first material accumulation information on the target conveyor belt through a first vision sensor. Among them, the first vision sensor is arranged above the stacker, and the first material accumulation information includes the first material position information and the first material image information, and the first material position information includes M positioning identifiers; The reclaimer set at the first sorting port obtains the first material accumulation information, performs feature recognition on the first material image information, and combines the material characteristic level model to obtain the humidity levels of the materials with M positioning identifiers; Calculating the first reclaiming value of the materials with M positioning identifiers according to the material humidity level and the position information. The first reclaiming value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity level, and the first time value is determined according to the time when the reclaimer moves to the positioning identifier. The order of magnitude of the humidity value is a hundred times that of the first time value. Integrate the M first reclaiming values, arrange the first reclaiming values from largest to smallest, and obtain the first reclaiming sequence; The reclaimer reclaims materials according to the sorting order of the first reclaiming sequence, separately reclaims materials with different humidities onto Q different conveyor belts, and transports them to the second sorting port through the Q conveyor belts; Collecting the second material accumulation information on the Q conveyor belts through Q second vision sensors. The second vision sensors are located above the Q conveyor belts at the first sorting port. The second material accumulation information includes the second material position information and the second material image information, and the second material position information includes N positioning identifiers; The reclaimer set at the second sorting port obtains the second material accumulation information, calculates the material volume in combination with the volume information in the characteristic level model, and obtains the volume levels of the materials with N positioning identifiers; Calculating the second reclaiming value of the materials with N positioning identifiers according to the material volume level and the position information. The second reclaiming value is the sum of the volume value and the second time value. The volume value is determined according to the volume level, and the acquisition method of the second time value is the same as that of the first time value. Integrate the N second reclaiming values, arrange the second reclaiming values from largest to smallest, and obtain the second reclaiming sequence; The reclaimer performs secondary reclaiming according to the sorting order of the second reclaiming sequence, separately reclaims materials with different volumes onto P different conveyor belts, and transports them to the final sorting port through the P conveyor belts; The P reclaimers arranged at the final sorting port reclaim materials respectively to obtain multiple uniform material piles.

2. The automated stacking and reclaiming method of belt-conveyed materials based on image recognition according to claim 1, characterized in that Constructing a material characteristic level model includes: Extracting the type, humidity, and volume data from the historical production records of materials to obtain a material basic feature set, and preprocessing the material basic feature set. The preprocessing includes supplementing missing data values and removing abnormal data values; Traversing the material basic feature set, using the material type as the row index, and the corresponding humidity data and volume data of the material as the columns, to generate a material humidity statistical table and a material volume statistical table; Based on the K-means clustering algorithm, performing level division on the material humidity statistical table and the material volume statistical table to obtain a material humidity level table and a material volume level table; Construct a material characteristic level model according to the material humidity level table and the material volume level table.

3. The automatic stacking and reclaiming method of belt-conveyed materials based on image recognition according to claim 1, characterized in that, Collect the first material accumulation information on the target conveyor belt, including: Obtain the image information of the material through the first vision sensor; Preprocess the acquired image, and the preprocessing includes grayscale conversion, filtering, enhancement, and histogram equalization; Based on the SSD object detection algorithm, identify and locate the positions of the materials in the image, and generate M positioning identifiers; Segment the image into M regional images according to the M positioning identifiers, and summarize and compress them for storage in the first material accumulation information.

4. The automatic stacking and reclaiming method of belt-conveyed materials based on image recognition according to claim 1, wherein, Obtain the humidity levels of the materials with M positioning identifiers, including: Obtain the first material accumulation information, and perform key feature extraction on the M regional images respectively. The key features mainly include texture features and color features; Input the texture features of the M regional images into the material characteristic level model to determine the type of the material transmitted this time and the corresponding humidity level table of the type; Compare the color information of the M regional images with the humidity level table to obtain the humidity levels of the materials with M positioning identifiers.

5. The automatic stacking and reclaiming method of belt-conveyed materials based on image recognition according to claim 1, characterized in that, Obtain the first material taking sequence, including: Set the humidity value H, and the humidity value formula is H = humidity level × 100; Obtain the M positioning identifiers, calculate the time required for the material taking machine to move to each positioning identifier respectively, obtain M material taking times, sort the material taking times, and assign numerical values 1 to M in order to generate the material taking time sequence; Set the first time value T1, and the first time value formula is T1 = material taking time sequence × 1; Calculate the first material taking value, which is the sum of the humidity value H and the first time value T1. Integrate the M first material taking values for sorting, and trace back the positioning identifiers corresponding to each value to obtain the first material taking sequence.

6. The automatic stacking and reclaiming method of belt-conveyed materials based on image recognition according to claim 1, characterized in that, Obtain the volume levels of the materials with N positioning identifiers, including: Obtain the second material accumulation information, and perform key feature extraction on the N regional images respectively. The key features mainly include texture features and shape features; Combine the texture features of the N regional images and the material characteristic level model to determine the type of the material transmitted this time and the corresponding volume level table of the type; Estimate the size of the object by calculating the size of the bounding box, and obtain the size information of the N regional images; Compare the size information of the M regional images with the volume level table to obtain the volume levels of the materials with N positioning identifiers.

7. The automatic stacking and reclaiming method of belt-conveyed materials based on image recognition according to claim 1, characterized in that Obtain the second material taking sequence, including: Set the volume value V, and the volume value formula is V = volume level × 100; Obtain the N positioning identifiers, calculate the time required for material taking for each identifier, sort and assign values to generate the material taking time sequence, and then obtain the second time value T2. The formula is T2 = material taking time sequence × 1; Calculate the second material taking value, integrate the N second material taking values for sorting, and trace back the positioning identifiers corresponding to each value to obtain the second material taking sequence.

8. The automatic stacking and reclaiming system for belt-conveyed materials based on image recognition is characterized in that, The system includes: A material characteristic level model construction module, which is used to construct a material characteristic level model according to the historical production record data of cement plant materials. The material characteristic level model includes a material humidity level table and a material volume level table; The first material accumulation information acquisition module is used to acquire the first material accumulation information on the target belt through the first vision sensor. Among them, the first vision sensor is arranged above the stacker. The first material accumulation information includes the first material position information and the first material image information. The first material position information includes M positioning identifiers; The humidity level acquisition module is used to acquire the first material accumulation information by the reclaimer set at the first sorting port, perform feature recognition on the first material image information, and combine the material feature level model to acquire the humidity levels of the materials with M positioning identifiers; The first material retrieval sequence acquisition module is used to calculate the first material retrieval values of the materials with M positioning identifiers according to the material humidity levels and position information. The first material retrieval value is the sum of the humidity value and the first time value. The humidity value is determined according to the humidity level, and the first time value is determined according to the time when the reclaimer moves to the positioning identifier. The order of magnitude of the humidity value is 100 times that of the first time value. Integrate M first material retrieval values, arrange the first material retrieval values from largest to smallest, and obtain the first material retrieval sequence; The initial material retrieval module is used for the reclaimer to retrieve materials according to the sorting order of the first material retrieval sequence, separately retrieve materials with different humidities onto Q different belts, and transport them to the second sorting port through the Q belts; The second material accumulation information acquisition module is used to acquire the second material accumulation information on the Q belts through Q second vision sensors. The second vision sensors are located above the Q belts at the first sorting port. The second material accumulation information includes the second material position information and the second material image information. The second material position information includes N positioning identifiers; The volume level acquisition module is used to acquire the second material accumulation information by the reclaimer set at the second sorting port, calculate the material volume by combining the volume information in the feature level model, and obtain the volume levels of the materials with N positioning identifiers; The second material retrieval sequence acquisition module is used to calculate the second material retrieval values of the materials with N positioning identifiers according to the material volume levels and position information. The second material retrieval value is the sum of the volume value and the second time value. The volume value is determined according to the volume level, and the acquisition method of the second time value is the same as that of the first time value. Integrate N second material retrieval values, arrange the second material retrieval values from largest to smallest, and obtain the second material retrieval sequence; The intermediate material retrieval module is used for the reclaimer to perform secondary material retrieval according to the sorting order of the second material retrieval sequence, separately retrieve materials with different volumes onto P different belts, and transport them to the final sorting port through the P belts; The final material retrieval module is used for the P reclaimers arranged at the final sorting port to retrieve materials respectively to obtain multiple uniform material piles.

Images