A dynamic feeding strategy-based material level automatic feeding method and system

Abstract

The application provides a kind of based on dynamic feeding strategy's material level automatic feeding method and system, it is related to automatic feeding technical field, its method includes through visual monitoring device and acquisition control system, acquisition multimodal hopper image and pre-processing;Based on Hough circle transformation, color segmentation algorithm, texture analysis algorithm and morphological operation, identify and segment the material level area of multimodal hopper image, obtain material level binary image;Mapping relationship of multimodal hopper image coordinates and material level actual space coordinates is constructed, material level distribution characteristics are extracted and calculated, and material level state analysis result is generated;Based on decision tree algorithm, analyze material level state analysis result and formulate dynamic feeding strategy, adjust the feeding strategy parameter of dynamic feeding strategy in real time by monitoring material level state change, so that the material state in hopper can be monitored in real time, accurate feeding is realized by dynamically adjusting feeding strategy, the accuracy of material level monitoring and the automation level of feeding process are significantly improved.

Description

Technical Field

[0001] This invention relates to the field of automatic feeding technology, and in particular to an automatic feeding method and system based on a dynamic feeding strategy. Background Technology

[0002] In the ongoing development of the materials processing industry, the application of automation and intelligent technologies is becoming increasingly widespread, serving as a crucial driving force for improving production efficiency and ensuring product quality stability. Among these, material level control in the hopper of a plastic extruder is of paramount importance for ensuring a continuous and stable production process.

[0003] In existing technologies, traditional material level control for plastic extruders largely relies on manual observation and simple level sensors to trigger replenishment, which is insufficient to meet the demands for precise, automated, and intelligent material level control in large-scale production. In practical applications, the workshop environment where multiple plastic extruders operate collaboratively is complex and variable. Manual observation is not only labor-intensive but also struggles to capture real-time changes in material level, leading to issues such as untimely replenishment or incorrect judgments. Furthermore, simple level sensors, fixed to the sidewall of the hopper, cannot comprehensively monitor the material distribution inside the hopper, making it difficult to detect anomalies such as material bridging or edge accumulation. Simultaneously, environmental interference such as dust adhering to the hopper's inner wall and temperature fluctuations during extruder barrel operation can reduce the detection accuracy of simple level sensors, hindering precise replenishment.

[0004] Existing technologies mostly employ simple high-low position control strategies, which cannot dynamically adjust the feeding strategy according to the actual operating status of the plastic extruder and the material consumption rate. This leads to problems such as untimely feeding causing material shortages and shutdowns, or excessive feeding causing material waste and hopper blockages. In addition, existing technologies struggle to precisely control and monitor the feeding process in real time, which not only hinders further improvements in production automation but also poses potential risks to product quality consistency.

[0005] Therefore, it is necessary to provide an automatic material replenishment method and system based on a dynamic replenishment strategy to solve the above-mentioned technical problems. Summary of the Invention

[0006] To address the aforementioned technical problems, this invention provides an automatic material replenishment method and system based on a dynamic replenishment strategy. This addresses the shortcomings of existing technologies, such as a lack of precise control and real-time monitoring of the replenishment process, the inability to achieve accurate replenishment, and the difficulty in improving the accuracy of material level monitoring and the automation level of the replenishment process.

[0007] This invention provides an automatic material replenishment method based on a dynamic replenishment strategy, the method comprising:

[0008] The vision acquisition and control device acquires and preprocesses images of the multimodal hopper.

[0009] Based on Hough circle transform, color segmentation algorithm, texture analysis algorithm and morphological operations, the material level region of the preprocessed multimodal hopper image is identified and segmented to obtain a binary image of the material level.

[0010] Construct a mapping relationship between the image coordinates of the multimodal hopper and the actual spatial coordinates of the material level, extract and calculate the material level distribution characteristics of the binary image of the material level, and generate material level status analysis results;

[0011] Based on the decision tree algorithm, the material level status analysis results are analyzed and a dynamic replenishment strategy is formulated. By monitoring changes in the material level status, the replenishment strategy parameters of the dynamic replenishment strategy are adjusted in real time.

[0012] Preferably, the step of acquiring and preprocessing multimodal hopper images via a vision acquisition and control device specifically includes:

[0013] An image acquisition program was developed based on the embedded computing platform of the aforementioned vision acquisition and control device. The program calculates the intrinsic parameter matrix K and distortion coefficients D of the industrial camera using the `calibrateCamera` function, and automatically adjusts the lens parameters of the industrial camera. The corresponding calculation formula is as follows:

[0014] In the formula, , These represent the focal lengths along the x-axis and y-axis, respectively. , These represent the coordinates of the principal point; , , These represent the radial distortion coefficients, respectively. , These represent the tangential distortion coefficients;

[0015] The industrial camera is calibrated using a standard checkerboard calibration board to acquire multi-angle, multi-position images of the multimodal hopper.

[0016] The open-source computer vision library OpenCV is invoked via TCP / IP communication protocol to perform random noise removal processing on the multimodal hopper image based on the Gaussian filtering algorithm.

[0017] A contrast-limited adaptive histogram equalization algorithm is used to enhance the details of the hopper shadow area in the multimodal hopper image;

[0018] The brightness of the highlight areas on the metal surface of the hopper in the multimodal hopper image is limited based on the local contrast suppression algorithm.

[0019] The multimodal hopper image is edge-sharpened using an unsharpened masking technique.

[0020] Preferably, the step of identifying and segmenting the material level region of the preprocessed multimodal hopper image based on Hough circle transform, color segmentation algorithm, texture analysis algorithm, and morphological operations to obtain a binary image of the material level specifically includes:

[0021] Based on the Hough circle transform, the hopper profile is detected, and a trapezoidal ROI region is formed according to the conical geometric features of the hopper.

[0022] A preset pixel value threshold range is used, combined with the trapezoidal ROI region of the hopper, and the preprocessed multimodal hopper image is converted from a three-dimensional image to a two-dimensional image based on the color segmentation algorithm to generate a binary mask.

[0023] Based on the texture analysis algorithm, the binary mask is corrected to identify the material location region;

[0024] The material level region is processed and segmented through morphological operations to obtain the binary image of the material level.

[0025] Preferably, the step of detecting the hopper profile based on the Hough circle transform and forming a trapezoidal ROI region of the hopper according to the conical geometric features of the hopper specifically includes:

[0026] The circular outline of the hopper opening is detected and located using the HoughCircles function, and the center pixel coordinates of the circular outline are extracted. Calculate the left end point of the hopper's upper opening using the radius r. Right end of the hopper opening ;

[0027] Based on the aforementioned cone-shaped geometric parameters of the hopper, the left endpoint of the hopper's lower opening is calculated using geometric relationships. Right end of the hopper lower opening , The angle of inclination of one side wall of the hopper relative to the vertical centerline; h represents the height of the hopper; Represents the vertical coordinate of the hopper's lower opening;

[0028] Connect the left end of the upper opening of the hopper in sequence. The right end point of the upper opening of the hopper The left end point of the lower opening of the hopper and the right end point of the lower opening of the hopper The trapezoidal ROI region of the hopper forms a complete space covering the interior of the hopper from top to bottom.

[0029] Preferably, the preset pixel value threshold range, combined with the trapezoidal ROI region of the hopper, is used to convert the preprocessed multimodal hopper image from a three-dimensional image into a two-dimensional image based on the color segmentation algorithm, generating a binary mask, specifically including:

[0030] Based on the aforementioned color segmentation algorithm, and combined with the trapezoidal ROI region of the hopper, the preprocessed multimodal hopper image is converted from the RGB color space to the HSV color space. According to the material color characteristics, the pixel value threshold ranges for the three independent HSV channels are preset, wherein the pixel value threshold range for the hue channel H is... The pixel value threshold range for saturation channel S is The pixel value threshold range for the luminance channel V is: ;

[0031] The inRange function is used to traverse each pixel of the multimodal hopper image. If the pixel values ​​of the hue channel H, saturation channel S, and luminance channel V corresponding to the pixel are all within the corresponding pixel value threshold range, then the pixel is determined to be a material position area pixel and the pixel value is marked. Otherwise, the pixel is determined to be a hopper background pixel and the pixel value is marked. The binary mask is generated based on the pixel value.

[0032] Preferably, the step of correcting the binary mask based on the texture analysis algorithm to identify the material location region specifically includes:

[0033] Locate the fuzzy trapezoidal ROI region of the material level that the color segmentation algorithm cannot determine, and calculate the local binary pattern feature value based on the texture analysis algorithm. The corresponding calculation formula is as follows:

[0034] In the formula, Indicates the center pixel coordinates of the multimodal hopper image; This represents the grayscale value of the p-th neighboring pixel of the center pixel in the multimodal hopper image; This represents the grayscale value of the center pixel in the multimodal hopper image. Represents a symbolic function;

[0035] The local binary pattern feature values ​​are aggregated into an LBP feature histogram using the LBPH class. The LBP feature histogram is then input into an SVM classifier, which outputs the texture feature classification result.

[0036] Based on the texture feature classification results, the binary mask is supplemented and corrected to generate an optimized binary mask, and the material location area is identified based on the pixel values ​​of the optimized binary mask.

[0037] Preferably, the step of processing and segmenting the material level region through morphological operations to obtain the binary image of the material level specifically includes:

[0038] The morphologyEx function is used with the operation type specified as opening operation. The image of the material level area is processed by opening operation using the preset structuring element to eliminate isolated noise points and small interference areas.

[0039] The morphologyEx function is used with the operation type specified as closing operation to perform closing operation on the material position area image after the opening operation to fill the hole defects in the material position area.

[0040] A preset connected region area threshold is set, and the contour detection function findContours is called to detect and calculate the area of ​​connected regions in the material position region. If the area of ​​the connected region is less than the connected region area threshold, the corresponding connected region is removed, the remaining connected regions are segmented, and the binary image of the material position is obtained.

[0041] Preferably, the step of constructing the mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, extracting and calculating the material level distribution features of the binary material level image, and generating material level status analysis results specifically includes:

[0042] Calculate the homography transformation matrix based on the findHomography function. The image coordinates of the multimodal hopper are mapped to the actual spatial coordinates of the material level, and the corresponding calculation formula is as follows:

[0043] In the formula, Represents the coordinates of the multimodal hopper image; Indicates the actual spatial coordinates of the material level; Indicates a non-zero scaling factor; All represent homography transformation matrices Element;

[0044] The moment function is used to perform order-weighted summation of the pixel coordinates of the binary image of the material level, quantize the pixel coordinates and summarize them into an image moment set, calculate the material level distribution feature parameters based on the image moment set, and extract the material level distribution features;

[0045] The material distribution feature parameters are input into the minAreaRect function. By fitting the outer contour of the material area and performing coordinate projection, the bounding rectangle with the smallest area corresponding to each side of the outer contour is calculated and filtered to obtain the width, height and tilt angle of the bounding rectangle.

[0046] Based on the width, height, and tilt angle of the circumscribed rectangle, the cross-section of the material level region is uniformly divided into multiple vertical material level regions. A weighted average is calculated for the material level height of each vertical material level region to obtain the material level height value of the material level region in the multimodal hopper image coordinates. The corresponding calculation formula is as follows:

[0047] In the formula, n represents the total number of vertical material level zones; This represents the height value of the i-th vertical material level region in the multimodal hopper image coordinates; This represents the weighting coefficient for the i-th vertical material level region;

[0048] Based on the mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, and according to the material level height value... Calculate the actual space height value of the material level The rate of change of material level height was calculated using linear regression. ;

[0049] The preset material level height range is based on the actual spatial height value of the material level. Read the corresponding material level height;

[0050] The standard deviation s of the material level height is calculated based on the material level height of each vertical material level area. The corresponding calculation formula is as follows:

[0051] In the formula, This represents the actual spatial height value of the material level in the i-th vertical material level region; This represents the average actual spatial height of the material level across all vertical material level zones.

[0052] A preset standard deviation threshold for material level height is set. If the standard deviation s of material level height is less than the standard deviation threshold, the material level area is judged to be uniformly distributed; otherwise, the material level area is judged to be unevenly distributed, and a material level distribution uniformity indicator is generated.

[0053] When it is determined that the material level distribution is uneven, a preset threshold for the rate of change of material level height is used. The center height value and the edge height value of the material level area are compared. If the center height value is less than the edge height value, and the rate of change of material level height is... If the material level is less than the threshold value for the rate of change of material level height, then the material level area is marked as a bridge and a bridge flag is generated; otherwise, the material level area is marked as an accumulation and an accumulation flag is generated.

[0054] The material level height level and the material level distribution uniformity indicator are statistically summarized to generate the material level status analysis results.

[0055] Preferably, the step of analyzing the material level status analysis results and formulating a dynamic replenishment strategy based on the decision tree algorithm, and adjusting the replenishment strategy parameters of the dynamic replenishment strategy in real time by monitoring changes in the material level status, specifically includes:

[0056] Based on the material level status analysis results, the decision tree nodes of the decision tree algorithm are called for layer-by-layer analysis;

[0057] Using the material level height level as the root node of the decision tree, check the material level height level. If the material level height level is a low material level level or a critical material level level, trigger the replenishment preparation and determine that the replenishment Boolean value is true. Otherwise, determine that the replenishment Boolean value is false and generate a replenishment flag.

[0058] When the replenishment Boolean value is determined to be true, the bridging flag is checked. If the bridging flag is true, the bridging process is triggered to reacquire the material level status analysis result and re-evaluate the material level height level. If the bridging flag is false, the material level distribution uniformity flag is checked. If the material level distribution uniformity flag is true, the target material level height value is set to the upper limit of the middle material level. If the material level distribution uniformity flag is false, the target material level height value is set to the lower limit of the middle material level.

[0059] The feed rate V is calculated based on the hopper truncated cone volume formula, and the feed rate is based on the material level height change rate. The formula for calculating the feeding rate v is as follows:

[0060] In the formula, Indicates the target material level height; R represents the hopper bottom radius; This indicates the preset base coefficients of the decision tree; Indicates the uniformity correction factor;

[0061] The replenishment control unit verifies the replenishment flag. If the replenishment flag is true, the replenishment time is estimated and the replenishment control process is started to replenish the material. If the replenishment flag is false, the replenishment control process is terminated.

[0062] The material level status changes are monitored according to a preset sampling period. The actual material level distribution uniformity, actual material level accuracy, actual material level rise rate, and actual replenishment time are compared with the dynamic replenishment strategy. The replenishment strategy parameters of the dynamic replenishment strategy are adjusted in real time, and the adjusted replenishment strategy parameters are saved in the replenishment control unit for iterative optimization.

[0063] An automatic material replenishment system based on a dynamic replenishment strategy, the system comprising:

[0064] The image acquisition and preprocessing module is used to acquire and preprocess multimodal hopper images through a vision acquisition control device.

[0065] The material level image segmentation module is used to identify and segment the material level region of the preprocessed multimodal hopper image based on Hough circle transform, color segmentation algorithm, texture analysis algorithm and morphological operations to obtain a binary image of the material level.

[0066] The material level status analysis module is used to construct the mapping relationship between the image coordinates of the multimodal hopper and the actual spatial coordinates of the material level, extract and calculate the material level distribution characteristics of the binary image of the material level, and generate material level status analysis results.

[0067] The material replenishment strategy adjustment module is used to analyze the material level status analysis results based on the decision tree algorithm and formulate a dynamic material replenishment strategy. By monitoring changes in the material level status, the module adjusts the material replenishment strategy parameters of the dynamic material replenishment strategy in real time.

[0068] Compared with related technologies, the automatic material replenishment method and system based on dynamic replenishment strategy provided by the present invention has the following beneficial effects:

[0069] This invention acquires and preprocesses multimodal hopper images using a visual monitoring device and a data acquisition and control system. Based on Hough circle transform, color segmentation algorithm, texture analysis algorithm, and morphological operations, it identifies and segments the material level regions in the multimodal hopper images to obtain binary material level images. It constructs a mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, extracts and calculates material level distribution features, and generates material level status analysis results. Based on a decision tree algorithm, it analyzes the material level status analysis results and formulates a dynamic replenishment strategy. By monitoring changes in material level status, it adjusts the replenishment strategy parameters in real time, thereby enabling real-time monitoring of the material status within the hopper. Through dynamic adjustment of the replenishment strategy, it achieves precise replenishment, significantly improving the accuracy of material level monitoring and the automation level of the replenishment process.

[0070] This invention deeply integrates multimodal hopper images, using Hough circle transform to detect trapezoidal ROI regions, color segmentation to generate binary masks, texture analysis algorithms to identify material level regions, and morphological operations to segment binary material level images. This ensures accurate positioning and pixel-level segmentation of material level regions under complex conditions such as light fluctuations, dust adhesion, and material color differences, improving the accuracy of material level region identification and segmentation. Furthermore, this invention constructs a mapping relationship between multimodal hopper image coordinates and actual material level spatial coordinates, extracts and calculates material level distribution features, and generates material level status analysis results. This ensures accurate conversion of two-dimensional image coordinates into three-dimensional actual material level coordinates, enabling comprehensive quantitative analysis of multi-dimensional material level status. Finally, this invention uses a decision tree algorithm to formulate a dynamic replenishment strategy, responding in real-time to dynamic changes in material level and adaptively adjusting replenishment strategy parameters. This solves the problems of traditional simple high / low level control strategies and the lack of fine control and monitoring capabilities for the replenishment process, effectively avoiding production interruptions caused by inaccurate replenishment and ensuring the continuity and stability of the replenishment process. Attached Figure Description

[0071] Figure 1 A flowchart of an automatic material replenishment method and system based on a dynamic replenishment strategy provided in this embodiment of the invention;

[0072] Figure 2 A system block diagram of an automatic material replenishment system based on a dynamic replenishment strategy provided in an embodiment of the present invention;

[0073] Figure 3 This is a schematic diagram of the hardware structure of an electronic device provided in an embodiment of the present invention. Detailed Implementation

[0074] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0075] like Figure 1 The diagram shown is a flowchart of an automatic material replenishment method based on a dynamic replenishment strategy provided by an embodiment of the present invention. Figure 1The execution entity of the method shown can be a software and / or hardware device. The execution entity of this application can include, but is not limited to, at least one of the following: user equipment, network equipment, etc. User equipment can include, but is not limited to, computers, smartphones, personal digital assistants (PDAs), and the aforementioned electronic devices. Network equipment can include, but is not limited to, a single network server, a server group consisting of multiple network servers, or a cloud based on cloud computing consisting of a large number of computers or network servers. Cloud computing is a type of distributed computing, consisting of a super virtual computer composed of a group of loosely coupled computers. This embodiment does not limit this. Steps S1 to S4 are detailed as follows:

[0076] S1, through a vision acquisition and control device, acquires and preprocesses images of the multimodal hopper;

[0077] The visual acquisition and control device refers to an image acquisition device and an image control system used to acquire multimodal images of the plastic extruder hopper and perform preprocessing. The multimodal hopper images are a collection of images of the hopper interior, covering visual information under different lighting conditions and material distribution states, preserving complete details such as the hopper inner wall, material particles, and material level boundaries.

[0078] In practical applications, the image acquisition device is deployed by mounting an industrial camera equipped with a CMOS sensor onto the frame of a plastic extruder using an L-shaped metal bracket. Shock-absorbing pads are installed at the bottom of the frame to effectively isolate vibrations generated during extruder operation and prevent image blurring. The industrial camera is installed at an appropriate distance above the plastic extruder hopper, ensuring a complete field of view covering the hopper's interior and sufficient resolution to clearly display material distribution details. A ring of LED lights is installed around the camera, using a coaxial mounting method to reduce shadows within the hopper and provide uniform and stable illumination. The power cables for both the camera and LED lights are fitted with high-temperature resistant sheaths to prevent damage from high temperatures. The industrial camera, ring of LED lights, shock-absorbing mounting bracket, and high-temperature resistant cables enable stable acquisition of multi-dimensional image information of the material within the hopper in complex industrial production environments. The image control system is used to control the industrial camera lens parameters, transmit image data, and execute preprocessing algorithms. Specifically, it includes an embedded computing platform, an Ethernet communication module, and the open-source computer vision library OpenCV. The image acquisition device and image control system work together to stably output high-quality preprocessed images in complex industrial environments, providing a data foundation for subsequent material level analysis.

[0079] The process of acquiring and preprocessing multimodal hopper images via a vision acquisition and control device specifically includes:

[0080] An image acquisition program was developed based on the embedded computing platform of the aforementioned vision acquisition and control device. The program calculates the intrinsic parameter matrix K and distortion coefficients D of the industrial camera using the `calibrateCamera` function, and automatically adjusts the lens parameters of the industrial camera. The corresponding calculation formula is as follows:

[0081] In the formula, , These represent the focal lengths along the x-axis and y-axis, respectively. , These represent the coordinates of the principal point; , , These represent the radial distortion coefficients, respectively. , These represent the tangential distortion coefficients;

[0082] The industrial camera is calibrated using a standard checkerboard calibration board to acquire multi-angle, multi-position images of the multimodal hopper.

[0083] The open-source computer vision library OpenCV is invoked via TCP / IP communication protocol to perform random noise removal processing on the multimodal hopper image based on the Gaussian filtering algorithm.

[0084] A contrast-limited adaptive histogram equalization algorithm is used to enhance the details of the hopper shadow area in the multimodal hopper image;

[0085] The brightness of the highlight areas on the metal surface of the hopper in the multimodal hopper image is limited based on the local contrast suppression algorithm.

[0086] The multimodal hopper image is edge-sharpened using an unsharpened masking technique.

[0087] Preprocessing, based on an embedded computing platform and the open-source computer vision library OpenCV, addresses issues such as noise interference, uneven lighting, specular reflection, and edge blurring in multimodal hopper images. The intrinsic parameter matrix K is a parameter matrix containing information such as focal length and coordinates, ensuring the consistency between the multimodal hopper image size and the actual spatial size. The distortion coefficients D are a set of optical distortion parameters from the industrial camera lens, used to correct image distortion caused by lens aberration. The standard checkerboard calibration board is a standard reference object with a black and white alternating rectangular grid array of known size printed on its surface, used to ensure that the industrial camera covers all visual angles for acquisition.

[0088] In practical applications, the image acquisition program automatically adjusts the lens parameters of the industrial camera by calculating the intrinsic parameter matrix K and distortion coefficient D. This includes dynamically adapting the exposure time based on ambient light intensity, setting the optimal gain value for material imaging, and performing white balance calibration for the internal environment of the hopper. This ensures that the images acquired by the industrial camera have true color and uniform brightness, reducing errors caused by distortion and ambient lighting at the source. A standard checkerboard calibration plate adapted to the internal space of the hopper is selected and fixed on a bracket with adjustable angle and height. Multimodal images of the hopper at different heights and angles are then captured sequentially by the industrial camera.

[0089] Furthermore, the acquired multimodal hopper images are transmitted to the embedded computing platform via the TCP / IP communication protocol, and the open-source computer vision library OpenCV is called to perform standardized preprocessing operations.

[0090] It should be noted that the Gaussian filtering algorithm, by constructing a Gaussian kernel function, performs a weighted average of the grayscale values ​​of each pixel in the multimodal hopper image, effectively removing random noise and eliminating image artifacts caused by thermal noise from industrial camera sensors and equipment vibration, thus avoiding interference from noise in material area identification. A contrast-limited adaptive histogram equalization algorithm is used to convert the RGB format multimodal hopper image to the HSV color space, enhancing only the luminance channel before converting it back to the RGB color space. This enhances details in shadow areas, highlighting the material outlines in low-light areas of the hopper and compensating for detail loss due to uneven lighting. Based on a local contrast suppression algorithm, the mean and standard deviation of the luminance in the multimodal hopper image are calculated, and a reasonable threshold is set to limit the brightness of highlight areas. This suppresses overly bright areas caused by reflections from the hopper's metal surface, preventing highlights from obscuring material information. An unsharpened masking technique is employed: the multimodal hopper image is first Gaussian blurred, and then the difference between the multimodal hopper image and the Gaussian blurred image is superimposed back onto the multimodal hopper image, enhancing the clarity of material level boundaries and particle edges.

[0091] S2, based on Hough circle transform, color segmentation algorithm, texture analysis algorithm and morphological operations, identify and segment the material level region of the preprocessed multimodal hopper image to obtain a binary image of the material level;

[0092] The process of identifying and segmenting the material level region of the preprocessed multimodal hopper image based on Hough circle transform, color segmentation algorithm, texture analysis algorithm, and morphological operations to obtain a binary image of the material level specifically includes:

[0093] Based on the Hough circle transform, the hopper profile is detected, and a trapezoidal ROI region is formed according to the conical geometric features of the hopper.

[0094] A preset pixel value threshold range is used, combined with the trapezoidal ROI region of the hopper, and the preprocessed multimodal hopper image is converted from a three-dimensional image to a two-dimensional image based on the color segmentation algorithm to generate a binary mask.

[0095] Based on the texture analysis algorithm, the binary mask is corrected to identify the material location region;

[0096] The material level region is processed and segmented through morphological operations to obtain the binary image of the material level.

[0097] It is understandable that the material level area refers to the actual spatial area occupied by the material in the hopper of a plastic extruder. The binary image of the material level refers to the image of the material level area segmented after Hough circle transform, color segmentation, texture analysis, and morphological operations.

[0098] The process of detecting the hopper profile based on the Hough circle transform and forming a trapezoidal ROI region for the hopper according to the conical geometric features of the hopper specifically includes:

[0099] The circular outline of the hopper opening is detected and located using the HoughCircles function, and the center pixel coordinates of the circular outline are extracted. Calculate the left end point of the hopper's upper opening using the radius r. Right end of the hopper opening ;

[0100] Based on the aforementioned cone-shaped geometric parameters of the hopper, the left endpoint of the hopper's lower opening is calculated using geometric relationships. Right end of the hopper lower opening , The angle of inclination of one side wall of the hopper relative to the vertical centerline; h represents the height of the hopper; Represents the vertical coordinate of the hopper's lower opening;

[0101] Connect the left end of the upper opening of the hopper in sequence. The right end point of the upper opening of the hopper The left end point of the lower opening of the hopper and the right end point of the lower opening of the hopper The trapezoidal ROI region of the hopper forms a complete space covering the interior of the hopper from top to bottom.

[0102] The Hough Circle Transform is an algorithm in the open-source computer vision library OpenCV used for accurate detection of circular contours in images. It maps the pixels of circular contours in multimodal hopper images to a parameter space, and uses an accumulator voting mechanism to select parameter combinations that match circular features, thereby locating and extracting the circular contour. The trapezoidal ROI region of the hopper refers to the trapezoidal region of interest defined by combining the circular contour of the hopper's upper opening detected by the Hough Circle Transform with the conical geometry of the hopper. The upper boundary is the circular contour of the hopper's upper opening, and the lower boundary is the estimated contour of the hopper's bottom, which helps to accurately narrow down the range for subsequent calculations.

[0103] Specifically, the HoughCircles function can perform circle detection on preprocessed multimodal hopper images. By setting reasonable accumulator thresholds, center-distance thresholds, and minimum and maximum radii based on the actual size of the hopper, it accurately locates the circular outline of the hopper's top opening. This is achieved by calculating the left endpoint of the hopper's top opening. Right end of the upper opening of the hopper Define the upper boundary of the trapezoidal ROI area of ​​the hopper to ensure complete coverage of the hopper's upper opening. Obtain the hopper's conical geometric parameters through actual measurements, and calculate the left endpoint of the hopper's lower opening based on these parameters. Right end of the hopper lower opening This design is adaptable to the narrow top and wide bottom of the hopper, preventing the omission of the trapezoidal ROI area due to structural compatibility issues. Connect the left end point of the hopper's top opening sequentially. Right end of the upper opening of the hopper Left end of the hopper lower opening and the right end of the hopper lower opening This forms a complete and closed trapezoidal ROI area in the hopper, ensuring that all materials from the top to the bottom of the hopper are included in the analysis scope, and avoiding omissions of material level areas due to improper definition of the trapezoidal area.

[0104] In practical applications, the HoughCircles function is not affected by uneven lighting in industrial environments or localized reflections on the metal surface of the hopper. Even under complex lighting conditions, it can still accurately locate the center and radius of the upper opening, ensuring that the upper boundary of the trapezoidal ROI area of ​​the hopper is without deviation.

[0105] The above method effectively achieves precise focusing of the trapezoidal ROI area of ​​the hopper and rapid filtering of invalid backgrounds, solving the problems of inaccurate boundary positioning, poor structural adaptability, and severe background interference of traditional methods.

[0106] The preset pixel value threshold range, combined with the trapezoidal ROI region of the hopper, is used to convert the preprocessed multimodal hopper image from a three-dimensional image to a two-dimensional image based on the color segmentation algorithm, generating a binary mask, specifically including:

[0107] Based on the aforementioned color segmentation algorithm, and combined with the trapezoidal ROI region of the hopper, the preprocessed multimodal hopper image is converted from the RGB color space to the HSV color space. According to the material color characteristics, the pixel value threshold ranges for the three independent HSV channels are preset, wherein the pixel value threshold range for the hue channel H is... The pixel value threshold range for saturation channel S is The pixel value threshold range for the luminance channel V is: ;

[0108] The inRange function is used to traverse each pixel of the multimodal hopper image. If the pixel values ​​of the hue channel H, saturation channel S, and luminance channel V corresponding to the pixel are all within the corresponding pixel value threshold range, then the pixel is determined to be a material position area pixel and the pixel value is marked. Otherwise, the pixel is determined to be a hopper background pixel and the pixel value is marked. The binary mask is generated based on the pixel value.

[0109] It should be noted that the HSV color space refers to a color display system composed of three independent channels: hue, saturation, and lightness. The hue channel (H) represents the color type, with a pixel value threshold range between 0 and 180; different values ​​correspond to different color tones. The saturation channel (S) represents the vividness of a color, with a pixel value threshold range between 0 and 255; higher values ​​indicate more vivid colors, while lower values ​​indicate colors closer to gray. The lightness channel (V) represents the brightness of a color, with a pixel value threshold range between 0 and 255; higher values ​​indicate brighter colors, while lower values ​​indicate colors closer to black. A binary mask is a two-dimensional image containing only pixel values ​​of 0 and 255, used to distinguish between material areas marked as 255 and the hopper background marked as 0.

[0110] The pixel value threshold ranges for the three independent channels of HSV need to be dynamically adjusted according to the material color characteristics. Specifically, the preprocessed RGB format multimodal hopper image is converted to the HSV color space. Based on the material color characteristics, a reasonable pixel value threshold range for the three independent channels of HSV is determined through preliminary sample testing. The `inRange` function is called to traverse only the pixels of the multimodal hopper image within the trapezoidal ROI area of ​​the hopper, eliminating external background interference. If the H, S, and V values ​​of a pixel are all within the corresponding pixel value threshold range, it is determined to be a pixel in the material location area and its pixel value is marked as 255 (white); otherwise, it is determined to be a pixel in the hopper background, including the inner wall, reflective areas, etc., and its pixel value is marked as 0 (black). After traversal, a binary mask containing only pixel values ​​of 0 and 255 is generated, initially separating the material location area from the hopper background. Compared with the color segmentation of the traditional RGB color space, the threshold division of the HSV color space is more targeted and can effectively prevent interference caused by uneven lighting and slight reflections in industrial environments.

[0111] By using the above method, based on the range limitation of the trapezoidal ROI area of ​​the hopper and the automatic traversal of the inRange function, the material location area and the background can be effectively separated quickly and accurately.

[0112] The step of correcting the binary mask based on the texture analysis algorithm to identify the material location region specifically includes:

[0113] Locate the fuzzy trapezoidal ROI region of the material level that the color segmentation algorithm cannot determine, and calculate the local binary pattern feature value based on the texture analysis algorithm. The corresponding calculation formula is as follows:

[0114] In the formula, Indicates the center pixel coordinates of the multimodal hopper image; This represents the grayscale value of the p-th neighboring pixel of the center pixel in the multimodal hopper image; This represents the grayscale value of the center pixel in the multimodal hopper image. Represents a symbolic function;

[0115] The local binary pattern feature values ​​are aggregated into an LBP feature histogram using the LBPH class. The LBP feature histogram is then input into an SVM classifier, which outputs the texture feature classification result.

[0116] Based on the texture feature classification results, the binary mask is supplemented and corrected to generate an optimized binary mask, and the material location area is identified based on the pixel values ​​of the optimized binary mask.

[0117] Among them, the fuzzy trapezoidal ROI region refers to the area where the material level is not accurately identified by the color segmentation algorithm due to factors such as the similarity of the material level area to the background color of the hopper, the transparency or semi-transparency of the material leading to insignificant color features, or local color distortion caused by light reflection or shadows in the industrial environment. Local binary mode feature values ​​refer to the binary encoded values ​​that quantify the local texture features of a multimodal hopper image, reflecting both granular and smooth textures. The LBPH class refers to the functional class that extracts LBP feature histograms from the open-source computer vision library OpenCV, used to statistically analyze the distribution of local binary mode feature values. The LBP feature histogram reflects the statistical distribution of local binary mode feature values ​​within a preset grayscale range, used to aggregate scattered local binary mode feature values ​​into standardized texture features. The SVM classifier is a binary classification model based on statistical learning theory, which accurately distinguishes two different types of features by finding the optimal separating hyperplane.

[0118] Furthermore, by traversing the binary mask and combining it with the trapezoidal ROI region of the hopper, the ambiguous trapezoidal ROI region of the material level is located using grayscale fluctuation detection and pixel confidence determination. The multimodal hopper image corresponding to the ambiguous trapezoidal ROI region is converted into a grayscale image. The local binary mode feature value is calculated using the above formula to achieve quantization encoding of texture features. The LBPH class of the open-source computer vision library OpenCV is called to statistically analyze the local binary mode feature value according to the preset grayscale interval, generating a fixed-dimensional LBP feature histogram. The LBP feature histogram is normalized to eliminate the influence of pixel quantity differences. The LBP feature histogram is input into a pre-trained SVM classifier, and the optimal separating hyperplane is used for determination. The texture of the material level region is marked as 1, and the texture of the hopper background is marked as 0. If the texture feature classification result is a material location area texture, but the binary mask misses its pixel value, then the missing pixel value is filled with 255; if the texture feature classification result is a hopper background texture, but the binary mask misjudges its pixel value as 255, then the pixel value is corrected to 0. The final pixel values ​​are counted, and an optimized binary mask containing only two pixel values, 0 and 255, is accurately generated. The optimized binary mask is traversed, and all areas with a pixel value of 255 are the finally identified material location areas.

[0119] Understandably, when the color segmentation algorithm cannot accurately identify the material location area, the texture analysis algorithm is used to accurately distinguish and identify it. Finally, by filling in the gaps and correcting the binary mask, the omissions are filled in and the misjudgments are eliminated, ensuring the integrity and purity of the material location area identification.

[0120] The step of processing and segmenting the material level region through morphological operations to obtain the binary image of the material level specifically includes:

[0121] The morphologyEx function is used with the operation type specified as opening operation. The image of the material level area is processed by opening operation using the preset structuring element to eliminate isolated noise points and small interference areas.

[0122] The morphologyEx function is used with the operation type specified as closing operation to perform closing operation on the material position area image after the opening operation to fill the hole defects in the material position area.

[0123] A preset connected region area threshold is set, and the contour detection function findContours is called to detect and calculate the area of ​​connected regions in the material position region. If the area of ​​the connected region is less than the connected region area threshold, the corresponding connected region is removed, the remaining connected regions are segmented, and the binary image of the material position is obtained.

[0124] Among them, the preset structural element refers to a fixed-shape pixel matrix used for morphological operations, the size and shape of which are dynamically adjusted according to the particle size of the material and the degree of noise interference. The connected region area threshold refers to the critical area value for screening the effective material location area, used to distinguish the effective material location area from the residual small interference area.

[0125] It should be noted that calling the `morphologyEx` function and specifying the operation type as opening first reduces small noise points in the identified material location area image through erosion, and then restores the original contour of the material location area through dilation, ultimately eliminating isolated noise points, dust interference, and other small invalid areas, clearing interference for subsequent processing. Calling the `morphologyEx` function again and specifying the operation type as closing first fills the holes and gaps inside the material location area image after the opening operation through dilation, and then restores the true size of the material location area through erosion, making the contour of the material location area continuous and complete, avoiding the deviation in material location area calculation caused by holes. A preset connected region area threshold is set, and the findContours function for contour detection is called. The outermost contour detection retrieval mode and the compressed contour point approximation method are used to detect all connected regions in the material level area image. The area of ​​each connected region is calculated by the contourArea function. Connected regions with an area smaller than the connected region area threshold are identified as residual interference and removed. Connected regions with an area greater than or equal to the connected region area threshold are retained, and their pixel values ​​are uniformly marked as 255. Finally, a material level binary image with clear contours, no interference, and no holes is obtained.

[0126] Through the above methods, morphological operations are used to purify and optimize the material location regions identified by the texture analysis algorithm. The combination of opening and closing operations achieves a logical closed loop of first denoising and then completing. The screening of connected regions further eliminates residual minor interference, ensuring that the final binary image of the material location retains only the effective material location regions, and outputs a binary image of the material location with clear contours and complete information.

[0127] S3, construct the mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, extract and calculate the material level distribution characteristics of the binary image of the material level, and generate the material level status analysis results;

[0128] The process of constructing a mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, extracting and calculating the material level distribution features of the binary material level image, and generating material level status analysis results specifically includes:

[0129] Calculate the homography transformation matrix based on the findHomography function. The image coordinates of the multimodal hopper are mapped to the actual spatial coordinates of the material level, and the corresponding calculation formula is as follows:

[0130] In the formula, Represents the coordinates of the multimodal hopper image; Indicates the actual spatial coordinates of the material level; Indicates a non-zero scaling factor; All represent homography transformation matrices Element;

[0131] The moment function is used to perform order-weighted summation of the pixel coordinates of the binary image of the material level, quantize the pixel coordinates and summarize them into an image moment set, calculate the material level distribution feature parameters based on the image moment set, and extract the material level distribution features;

[0132] The material distribution feature parameters are input into the minAreaRect function. By fitting the outer contour of the material area and performing coordinate projection, the bounding rectangle with the smallest area corresponding to each side of the outer contour is calculated and filtered to obtain the width, height and tilt angle of the bounding rectangle.

[0133] Based on the width, height, and tilt angle of the circumscribed rectangle, the cross-section of the material level region is uniformly divided into multiple vertical material level regions. A weighted average is calculated for the material level height of each vertical material level region to obtain the material level height value of the material level region in the multimodal hopper image coordinates. The corresponding calculation formula is as follows:

[0134] In the formula, n represents the total number of vertical material level zones; This represents the height value of the i-th vertical material level region in the multimodal hopper image coordinates; This represents the weighting coefficient for the i-th vertical material level region;

[0135] Based on the mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, and according to the material level height value... Calculate the actual space height value of the material level The rate of change of material level height was calculated using linear regression. ;

[0136] The preset material level height range is based on the actual spatial height value of the material level. Read the corresponding material level height;

[0137] The standard deviation s of the material level height is calculated based on the material level height of each vertical material level area. The corresponding calculation formula is as follows:

[0138] In the formula, This represents the actual spatial height value of the material level in the i-th vertical material level region; This represents the average actual spatial height of the material level across all vertical material level zones.

[0139] A preset standard deviation threshold for material level height is set. If the standard deviation s of material level height is less than the standard deviation threshold, the material level area is judged to be uniformly distributed; otherwise, the material level area is judged to be unevenly distributed, and a material level distribution uniformity indicator is generated.

[0140] When it is determined that the material level distribution is uneven, a preset threshold for the rate of change of material level height is used. The center height value and the edge height value of the material level area are compared. If the center height value is less than the edge height value, and the rate of change of material level height is... If the material level is less than the threshold value for the rate of change of material level height, then the material level area is marked as a bridge and a bridge flag is generated; otherwise, the material level area is marked as an accumulation and an accumulation flag is generated.

[0141] The material level height level and the material level distribution uniformity indicator are statistically summarized to generate the material level status analysis results.

[0142] First, the homography transformation matrix is ​​calculated using a homography transformation matrix solver function, accurately mapping the two-dimensional coordinates of the multimodal hopper image to the three-dimensional actual spatial coordinates of the material level. During the mapping process, a non-zero scale factor is introduced to adapt and calibrate the coordinate transformation relationship, ensuring the consistency of the ratio between the image coordinates and the actual spatial coordinates. The elements of the homography transformation matrix collectively construct the mathematical relationship of the coordinate transformation, guaranteeing the accuracy of the mapping process and achieving a one-to-one correspondence between each pixel in the image and its actual spatial position within the hopper.

[0143] Then, the image moment calculation function is used to perform an order-weighted summation operation on the pixel coordinates of the binary image of the material level. This operation quantizes the pixel coordinates by assigning weights to pixel coordinates of different orders and accumulating them, and then summarizes the quantized pixel coordinates to form an image moment set. The image moment set can comprehensively reflect the distribution pattern of pixels in the image. Based on this set, material level distribution characteristic parameters, such as the area of ​​the material level region, the centroid position, and the distribution dispersion, can be further calculated, ultimately completing the extraction of material level distribution features.

[0144] Furthermore, the extracted material distribution feature parameters are input into the minimum area bounding rectangle calculation function. This function first fits the outer contour of the material location area, then maps the three-dimensional contour information to a two-dimensional plane using coordinate projection technology. Subsequently, it calculates the bounding rectangle corresponding to each side of the outer contour of the material location area and selects the bounding rectangle with the smallest area, i.e., the minimum area bounding rectangle. Through this process, three key geometric parameters of the minimum area bounding rectangle can be obtained: the width of the bounding rectangle, the height of the bounding rectangle, and the tilt angle of the bounding rectangle. These parameters can accurately characterize the overall morphological features of the material location area.

[0145] Based on the width, height, and tilt angle of the aforementioned circumscribed rectangle, the cross-section of the material level area is divided into multiple vertical material level areas according to a uniform division principle. For each vertical material level area, a weighted average method is used to calculate its material level height. That is, each vertical material level area is assigned a corresponding weight coefficient according to its importance in the overall material level distribution, and then the weighted sum and mean value of each area are calculated together with its own material level height value to obtain the overall material level height value of the area in the multimodal hopper image coordinates. The total number of vertical material level areas determines the level of precision in the height calculation, and the setting of the weight coefficients needs to be combined with the structural characteristics of the hopper. For example, the central area has a greater impact on material level judgment, and its weight can be appropriately increased.

[0146] Specifically, based on the previously established mapping relationship between the multimodal hopper image coordinates and the actual spatial coordinates of the material level, the material level height value under the image coordinates is converted into the actual spatial height value of the material level, realizing the transformation from image data to physical space data. At the same time, the linear regression method is used to analyze the actual spatial height value of the material level over a continuous period of time, and calculate the material level height change rate. This change rate can intuitively reflect the dynamic change trend of the material level over time, such as the rate of material level descent and rise.

[0147] It should be noted that a pre-defined range of material level height grades, such as low material level, medium material level, high material level, critical material level, etc., is used to compare the calculated actual spatial height value of the material level with the preset grade range to determine the material level height to which the height value belongs, thus completing the preliminary classification of the static state of the material level.

[0148] Based on the actual spatial height of the material level in each vertical material level area, the average actual spatial height of all vertical material level areas is first calculated. Then, the deviation of each vertical area's height value from the average value is calculated to obtain the standard deviation of the material level height. A threshold for the standard deviation of the material level height is preset. If the calculated standard deviation of the material level height is less than the threshold, it indicates that the difference in material level height between areas is small, and the material level area is judged to be uniformly distributed. If the standard deviation of the material level height is greater than or equal to the threshold, the material level area is judged to be unevenly distributed, and a corresponding material level distribution uniformity indicator, such as a uniform or uneven indicator, is generated.

[0149] When it is determined that the material level distribution is uneven, a threshold for the rate of change of material level height is preset, and the center height value and the edge height value of the material level area are compared: if the center height value is less than the edge height value and the rate of change of material level height is less than the preset rate of change threshold, it indicates that the material level has almost no dynamic change, which meets the static characteristics of bridging phenomenon. Then, the material level area is marked as bridging state and a bridging mark is generated; if the above conditions are not met, the material level area is marked as accumulation state and an accumulation mark is generated.

[0150] Finally, the material level height grade and material level distribution uniformity indicators are comprehensively statistically summarized, and key information such as the static grade, distribution uniformity and abnormal state of the material level are integrated to form a complete material level status analysis result.

[0151] S4. Based on the decision tree algorithm, analyze the material level status analysis results and formulate a dynamic replenishment strategy. By monitoring changes in the material level status, adjust the replenishment strategy parameters of the dynamic replenishment strategy in real time.

[0152] The process of analyzing the material level status based on the decision tree algorithm and formulating a dynamic replenishment strategy involves monitoring changes in the material level status and adjusting the replenishment strategy parameters in real time. Specifically, this includes:

[0153] Based on the material level status analysis results, the decision tree nodes of the decision tree algorithm are called for layer-by-layer analysis;

[0154] Using the material level height level as the root node of the decision tree, check the material level height level. If the material level height level is a low material level level or a critical material level level, trigger the replenishment preparation and determine that the replenishment Boolean value is true. Otherwise, determine that the replenishment Boolean value is false and generate a replenishment flag.

[0155] When the replenishment Boolean value is determined to be true, the bridging flag is checked. If the bridging flag is true, the bridging process is triggered to reacquire the material level status analysis result and re-evaluate the material level height level. If the bridging flag is false, the material level distribution uniformity flag is checked. If the material level distribution uniformity flag is true, the target material level height value is set to the upper limit of the middle material level. If the material level distribution uniformity flag is false, the target material level height value is set to the lower limit of the middle material level.

[0156] The feed rate V is calculated based on the hopper truncated cone volume formula, and the feed rate is based on the material level height change rate. The formula for calculating the feeding rate v is as follows:

[0157] In the formula, Indicates the target material level height; R represents the hopper bottom radius; This indicates the preset base coefficients of the decision tree; Indicates the uniformity correction factor;

[0158] The replenishment control unit verifies the replenishment flag. If the replenishment flag is true, the replenishment time is estimated and the replenishment control process is started to replenish the material. If the replenishment flag is false, the replenishment control process is terminated.

[0159] The material level status changes are monitored according to a preset sampling period. The actual material level distribution uniformity, actual material level accuracy, actual material level rise rate, and actual replenishment time are compared with the dynamic replenishment strategy. The replenishment strategy parameters of the dynamic replenishment strategy are adjusted in real time, and the adjusted replenishment strategy parameters are saved in the replenishment control unit for iterative optimization.

[0160] First, using the material level status analysis results as input, the decision tree algorithm is invoked to perform a layer-by-layer traversal analysis. The decision tree algorithm, through hierarchical node judgment logic, progressively breaks down the key information in the material level status, providing a structured analysis path for formulating replenishment strategies. This ensures that every decision is based on actual material level data, avoiding subjective judgment bias.

[0161] Secondly, the material level is set as the root node of the decision tree, prioritizing the checking of material level levels. If the current material level is a low level or a critical level, both indicating that the material reserve is close to or has reached the production guarantee limit and needs to be replenished promptly, the replenishment preparation process is immediately triggered, and the replenishment Boolean value is determined to be true. If the material level is a high level or a medium level, and the material reserve meets the current production needs, the replenishment Boolean value is determined to be false. Based on the determination result of the replenishment Boolean value, a corresponding replenishment flag is generated, providing a clear instruction basis for subsequent replenishment control.

[0162] When the replenishment Boolean value is true, check the bridging flag. If the bridging flag is true, it indicates that there is material bridging in the hopper. At this time, the material level detection may be inaccurate, and the actual usable material amount is lower than the detected value. In this case, the bridging processing procedure is triggered immediately, such as starting the hopper vibrator to break the bridging structure. After the bridging processing is completed, the hopper image is re-acquired and the material level status analysis is performed to reassess the material level level and avoid incorrect or insufficient replenishment due to bridging. If the bridging flag is false and there is no bridging, then further check the material level distribution uniformity flag. If the material level distribution uniformity flag is true, the material distribution in the hopper is flat, with no local accumulation or gaps. In this case, the target material level height value for replenishment is set to the upper limit of the middle material level, thus meeting subsequent production needs while avoiding material waste due to over-replenishment. If the material level distribution uniformity flag is false, the material distribution is uneven, and there is a risk of local accumulation. In this case, the target material level height value is set to the lower limit of the middle material level, thereby reducing the replenishment amount and reducing the possibility of aggravated accumulation.

[0163] After determining the target material level height, the replenishment quantity and replenishment rate are calculated separately. The replenishment quantity is calculated based on the hopper truncated cone volume formula. This formula requires combining parameters such as the difference between the target material level height and the actual material level height, and the hopper lower radius. By calculating the truncated cone volume difference, the volume of material to be replenished is obtained, ensuring that the replenishment quantity matches the remaining space in the hopper. The replenishment rate is calculated based on the material level height change rate. First, the base coefficients pre-set in the decision tree are obtained, and then the uniformity correction coefficient corresponding to the uniformity of material level distribution is combined. If the uniformity is good, the correction coefficient approaches 1; if the uniformity is poor, it is appropriately lowered to reduce the rate, thus obtaining a replenishment rate suitable for the current working conditions, thereby avoiding material level fluctuations caused by replenishment that is too fast or too slow.

[0164] The replenishment control unit must first perform a secondary verification of the replenishment flag to ensure the accuracy of the replenishment command and avoid invalid replenishment triggered by signal interference or misjudgment. If the replenishment flag is still true after verification, the estimated replenishment time is calculated based on the replenishment amount and rate, and then the replenishment control process is initiated, such as controlling the vacuum feeding system to start the vacuum pump and adjusting the screw conveyor speed. If the replenishment flag is false after verification, the replenishment control process is terminated directly to avoid energy consumption due to equipment idling.

[0165] Finally, following a preset sampling period (typically ranging from a few seconds to tens of seconds) that is set in conjunction with the production rhythm and material level change rate, the material level status is continuously monitored. During monitoring, key data such as the uniformity of actual material level distribution, actual material level accuracy (i.e., the deviation between the actual and target material level heights), actual material level rise rate (i.e., the actual speed at which the material level rises during replenishment), and actual replenishment time are collected in real time. This actual data is then compared and analyzed with the theoretical parameters in the current dynamic replenishment strategy. If deviations are found between the actual data and theoretical parameters, such as an actual material level rise rate lower than the theoretical value or an actual replenishment time exceeding the expected time, the replenishment strategy parameters are adjusted in real time. This includes correcting the replenishment rate coefficient and optimizing the target material level height value. The adjusted parameters are then saved to the storage module of the replenishment control unit.

[0166] The above methods can be used to iteratively optimize the replenishment strategy parameters, ensuring that the replenishment process is always highly adapted to the actual production conditions, thereby improving the accuracy of replenishment and production stability.

[0167] like Figure 2 The diagram shown is a system block diagram of an automatic material replenishment system based on a dynamic replenishment strategy provided in an embodiment of the present invention. The system includes:

[0168] The image acquisition and preprocessing module is used to acquire and preprocess multimodal hopper images through a vision acquisition control device.

[0169] The material level image segmentation module is used to identify and segment the material level region of the preprocessed multimodal hopper image based on Hough circle transform, color segmentation algorithm, texture analysis algorithm and morphological operations to obtain a binary image of the material level.

[0170] The material level status analysis module is used to construct the mapping relationship between the image coordinates of the multimodal hopper and the actual spatial coordinates of the material level, extract and calculate the material level distribution characteristics of the binary image of the material level, and generate material level status analysis results.

[0171] The material replenishment strategy adjustment module is used to analyze the material level status analysis results based on the decision tree algorithm and formulate a dynamic material replenishment strategy. By monitoring changes in the material level status, the module adjusts the material replenishment strategy parameters of the dynamic material replenishment strategy in real time.

[0172] Figure 2 The system of the illustrated embodiment can be used to perform corresponding operations. Figure 1 The steps in the method embodiments shown are implemented in a similar manner and have similar technical effects, and will not be repeated here.

[0173] An electronic device includes a memory and a processor, wherein the memory stores a computer program, and when the processor runs the computer program stored in the memory, the processor executes the steps of an automatic material level replenishment method based on a dynamic replenishment strategy as described above.

[0174] like Figure 3 The diagram shown is a hardware structure schematic of an electronic device according to an embodiment of the present invention. The electronic device 30 includes: a processor 31, a memory 32, and a computer program; wherein...

[0175] The memory 32 is used to store the computer program, and the memory may also be flash memory. The computer program is, for example, an application program or functional module that implements the above method.

[0176] Processor 31 is configured to execute the computer program stored in the memory to implement the various steps performed by the device in the above method. For details, please refer to the relevant descriptions in the preceding method embodiments.

[0177] Alternatively, the memory 32 can be either standalone or integrated with the processor 31.

[0178] When the memory 32 is a device independent of the processor 31, the device may further include:

[0179] Bus 33 is used to connect the memory 32 and the processor 31.

[0180] A readable storage medium storing a computer program, which, when executed by a processor, is used to implement the steps of an automatic material level replenishment method based on a dynamic replenishment strategy as described above.

[0181] The readable storage medium can be a computer storage medium or a communication medium. A communication medium includes any medium that facilitates the transfer of computer programs from one location to another. A computer storage medium can be any available medium accessible to a general-purpose or special-purpose computer. For example, a readable storage medium is coupled to a processor, enabling the processor to read information from and write information to the readable storage medium. Of course, the readable storage medium can also be a component of the processor. The processor and the readable storage medium can reside in an Application-Specific Integrated Circuit (ASIC). Alternatively, the ASIC can be located in a user equipment. Of course, the processor and the readable storage medium can also exist as discrete components in a communication device. The readable storage medium can be a read-only memory (ROM), random access memory (RAM), CD-ROM, magnetic tape, floppy disk, and optical data storage device, etc.

[0182] The present invention also provides a program product including executable instructions stored in a readable storage medium. At least one processor of the device can read the executable instructions from the readable storage medium, and the at least one processor executes the executable instructions to cause the device to implement the methods provided in the various embodiments described above.

[0183] In the embodiments of the above-described device, it should be understood that the processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in this invention can be directly manifested as execution by a hardware processor, or execution by a combination of hardware and software modules within the processor.

[0184] Through the above embodiments, this invention acquires and preprocesses multimodal hopper images using a visual monitoring device and a data acquisition and control system; based on Hough circle transform, color segmentation algorithm, texture analysis algorithm, and morphological operations, it identifies and segments the material level region of the multimodal hopper image to obtain a binary image of the material level; it constructs a mapping relationship between the coordinates of the multimodal hopper image and the actual spatial coordinates of the material level, extracts and calculates the material level distribution characteristics, and generates material level status analysis results; based on a decision tree algorithm, it analyzes the material level status analysis results and formulates a dynamic replenishment strategy. By monitoring changes in the material level status, it adjusts the replenishment strategy parameters of the dynamic replenishment strategy in real time, thereby enabling real-time monitoring of the material status in the hopper. By dynamically adjusting the replenishment strategy, it achieves precise replenishment, significantly improving the accuracy of material level monitoring and the automation level of the replenishment process.

[0185] This invention deeply integrates multimodal hopper images, using Hough circle transform to detect trapezoidal ROI regions, color segmentation to generate binary masks, texture analysis algorithms to identify material level regions, and morphological operations to segment binary material level images. This ensures accurate positioning and pixel-level segmentation of material level regions under complex conditions such as light fluctuations, dust adhesion, and material color differences, improving the accuracy of material level region identification and segmentation. Furthermore, this invention constructs a mapping relationship between multimodal hopper image coordinates and actual material level spatial coordinates, extracts and calculates material level distribution features, and generates material level status analysis results. This ensures accurate conversion of two-dimensional image coordinates into three-dimensional actual material level coordinates, enabling comprehensive quantitative analysis of multi-dimensional material level status. Finally, this invention uses a decision tree algorithm to formulate a dynamic replenishment strategy, responding in real-time to dynamic changes in material level and adaptively adjusting replenishment strategy parameters. This solves the problems of traditional simple high / low level control strategies and the lack of fine control and monitoring capabilities for the replenishment process, effectively avoiding production interruptions caused by inaccurate replenishment and ensuring the continuity and stability of the replenishment process.

[0186] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some or all of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A dynamic feed strategy based automatic feed method for material level, characterized in that, The method comprises: acquiring and preprocessing a multi-modal hopper image by a visual acquisition control device; identifying and segmenting a material level area of the preprocessed multi-modal hopper image based on a Hough circle transformation, a color segmentation algorithm, a texture analysis algorithm and morphological operations to obtain a material level binary image; the identification and segmentation of the material level area of the preprocessed multi-modal hopper image based on the Hough circle transformation, the color segmentation algorithm, the texture analysis algorithm and the morphological operations to obtain the material level binary image specifically comprises: detecting a hopper contour based on the Hough circle transformation, and forming a hopper trapezoidal ROI area according to the geometric characteristics of the hopper cone; presetting a pixel value threshold range, combining the hopper trapezoidal ROI area, converting the preprocessed multi-modal hopper image from a three-dimensional image to a two-dimensional image based on the color segmentation algorithm, and generating a binary mask; correcting the binary mask based on the texture analysis algorithm to identify the material level area; processing and segmenting the material level area by morphological operations to obtain the material level binary image; constructing a mapping relationship between a multi-modal hopper image coordinate and a material level actual space coordinate, extracting and calculating a material level distribution feature of the material level binary image, and generating a material level state analysis result; the construction of the mapping relationship between the multi-modal hopper image coordinate and the material level actual space coordinate, the extraction and calculation of the material level distribution feature of the material level binary image, and the generation of the material level state analysis result specifically comprise: Compute homography transformation matrix based on findHomography function The multi-modal hopper image coordinates are mapped to the actual space coordinates of the material level, and the corresponding calculation formula is as follows: wherein, denotes a multi-modal hopper image coordinate; denotes a level actual space coordinate; denotes a non-zero scale factor; denote elements of a homography transformation matrix denote elements of a homography transformation matrix using a moments function to perform order weighting summation on pixel coordinates of the material level binary image, quantizing the pixel coordinates and summarizing them into an image moment set, calculating a material level distribution feature parameter based on the image moment set, and extracting the material level distribution feature; inputting the material level distribution feature parameter into a minAreaRect function, fitting the outer contour of the material level area and performing coordinate projection, calculating and screening the minimum area circumscribed rectangle corresponding to each side of the outer contour, and obtaining the circumscribed rectangle width, the circumscribed rectangle height and the circumscribed rectangle inclination angle; According to the circumscribed rectangle width, the circumscribed rectangle height and the circumscribed rectangle inclination angle, the cross section of the material level area is evenly divided into a plurality of vertical material level areas, and the material level height of each vertical material level area is weighted and averaged to obtain a material level height value of the material level area under the multi-modal hopper image coordinates The corresponding calculation formula is as follows: where n represents the total number of vertical material level regions; represents the material level height value of the i-th vertical material level region in the multi-modal hopper image coordinate; represents the weight coefficient of the i-th vertical material level region; Based on the mapping relationship between the multi-modal hopper image coordinates and the actual space coordinates of the material level, the material level height value is calculated according to the material level height value The actual space height value of the material level is calculated The material level height change rate is calculated by linear regression method ; a preset level height grade range, according to the actual space height value of the level read the corresponding level height grade; calculating a material level height standard deviation s according to the material level height of each vertical material level area, and the corresponding calculation formula is as follows: wherein, represents the actual space height value of the level of the i-th vertical level area; represents the average actual space height value of all vertical level areas; presetting a material level height standard deviation threshold, if the material level height standard deviation s is less than the material level height standard deviation threshold, judging that the material level area is uniformly distributed, otherwise judging that the material level area is not uniformly distributed, and generating a material level distribution uniformity flag; When it is determined that the material level distribution is uneven, a preset threshold for the rate of change of material level height is used. The center height value and the edge height value of the material level area are compared. If the center height value is less than the edge height value, and the rate of change of material level height is... If the material level is less than the threshold value for the rate of change of material level height, then the material level area is marked as a bridge and a bridge flag is generated; otherwise, the material level area is marked as an accumulation and an accumulation flag is generated. statistically summarizing the material level height grade and the material level distribution uniformity flag to generate the material level state analysis result; analyzing the material level state analysis result based on a decision tree algorithm, formulating a dynamic replenishment strategy, and adjusting replenishment strategy parameters of the dynamic replenishment strategy in real time by monitoring the material level state change; the analysis of the material level state analysis result based on the decision tree algorithm, the formulation of the dynamic replenishment strategy, and the real-time adjustment of the replenishment strategy parameters of the dynamic replenishment strategy by monitoring the material level state change specifically comprise: calling decision tree nodes of the decision tree algorithm for layer-by-layer analysis based on the material level state analysis result; The material level height level is taken as a root node of a decision tree, the material level height level is checked, if the material level height level is a low material level or a critical material level, a replenishment preparation is triggered and a replenishment Boolean value is determined to be true, otherwise the replenishment Boolean value is determined to be false, and a replenishment flag is generated; When the replenishment Boolean value is determined to be true, the bridge flag is checked, if the bridge flag is true, a bridge processing flow is triggered, the material level state analysis result is re-acquired and the material level height level is re-evaluated, if the bridge flag is false, the material level distribution uniformity flag is checked, if the material level distribution uniformity flag is true, a target material level height value is set to an upper limit of a medium material level, if the material level distribution uniformity flag is false, the target material level height value is set to a lower limit of the medium material level; The feed rate V is calculated based on the hopper truncated cone volume formula, and the feed rate is based on the material level height change rate. The formula for calculating the feeding rate v is as follows: wherein, represents the target level height value; R represents the radius of the lower opening of the hopper; represents the preset base coefficient of the decision tree; represents the uniformity correction coefficient; The replenishment control unit checks the replenishment flag, if the replenishment flag is true, a replenishment time is predicted and a replenishment control flow is started for replenishment, if the replenishment flag is false, the replenishment control flow is ended; The material level state change is monitored according to a preset sampling period, the actual material level distribution uniformity, the actual material level precision, the actual material level rising rate and the actual replenishment time acquired are compared with the dynamic replenishment strategy, the replenishment strategy parameters of the dynamic replenishment strategy are adjusted in real time, and the adjusted replenishment strategy parameters are saved in the replenishment control unit for iterative optimization.

2. The method according to claim 1, wherein, The multi-modal hopper image is acquired and preprocessed through the visual acquisition control device, specifically including: An image acquisition program is developed based on an embedded computing platform of the visual acquisition control device, an industrial camera intrinsic matrix K and a distortion coefficient D are calculated through a calibrateCamera function, industrial camera lens parameters are automatically adjusted, and the corresponding calculation formula is as follows: wherein, , respectively represent focal lengths in x-axis, y-axis directions; , respectively represent principal point coordinates; , , respectively represent radial distortion coefficients; , respectively represent tangential distortion coefficients; The industrial camera is calibrated through a standard chessboard calibration plate, and the multi-modal hopper image is acquired at multiple angles and multiple positions; The multi-modal hopper image is processed for random noise removal based on a Gaussian filtering algorithm through a TCP / IP communication protocol and an open computer vision library OpenCV; A contrast-limited adaptive histogram equalization algorithm is used for detail enhancement processing of the hopper shadow area of the multi-modal hopper image; Based on a local contrast suppression algorithm, the highlight area of the hopper metal surface in the multi-modal hopper image is subjected to brightness limiting processing; An unsharpening masking technique is used for edge sharpening processing of the multi-modal hopper image.

3. The method of claim 1, wherein, The hopper profile is detected based on the Hough circle transformation, and a hopper trapezoidal ROI area is formed according to the hopper conical geometric characteristics, specifically including: Detect and locate the circular profile of the hopper upper opening by HoughCircles function, extract the center pixel coordinates of the circular profile With radius r, calculate the left end point of the hopper upper opening With radius r, calculate the right end point of the hopper upper opening ; According to the hopper cone geometry parameters, the left end point of the hopper lower opening is converted by geometric relationship and the right end point of the hopper lower opening , represents the inclination angle of the single-sided wall of the hopper to the vertical center line direction; h represents the height of the hopper; represents the vertical coordinate of the hopper lower opening; sequentially connecting the left end point of the upper opening of the hopper the right end point of the upper opening of the hopper the left end point of the lower opening of the hopper and the right end point of the lower opening of the hopper to form the trapezoidal ROI region of the hopper covering the complete space inside the hopper from the upper opening to the lower opening.

4. The method of claim 1, wherein, The preset pixel value threshold range is combined with the hopper trapezoidal ROI area, the preprocessed multi-modal hopper image is converted from a three-dimensional image to a two-dimensional image based on the color segmentation algorithm, a binary mask is generated, specifically including: Based on the color segmentation algorithm, combined with the hopper trapezoidal ROI region, the pre-processed multi-modal hopper image is converted from the RGB color space to the HSV color space, and the pixel value threshold range of the three independent channels of HSV is pre-set according to the material color characteristics, wherein the pixel value threshold range of the hue channel H is , the pixel value threshold range of the saturation channel S is ; and the pixel value threshold range of the brightness channel V is ; The inRange function is used to traverse each pixel of the multi-modal hopper image, if the pixel values of hue channel H, saturation channel S and brightness channel V corresponding to the pixel are all within the corresponding pixel value threshold range, it is determined that the pixel is a material level region pixel and the pixel value is marked, otherwise, it is determined that the pixel is a hopper background pixel and the pixel value is marked, and the binary mask is generated according to the pixel value.

5. The method of claim 1, wherein, The binary mask is corrected based on the texture analysis algorithm to identify the material level region, specifically including: Lock the color segmentation algorithm can not determine the ambiguous level trapezoidal ROI region, based on the texture analysis algorithm to calculate the local binary pattern characteristic value The corresponding calculation formula is as follows: wherein, denotes the center pixel coordinate of the multimodal hopper image; denotes the gray value of the p-th neighborhood pixel of the center pixel of the multimodal hopper image; denotes the gray value of the center pixel of the multimodal hopper image; denotes the sign function; The LBP feature histogram is input into the SVM classifier, and the texture feature classification result is output; According to the texture feature classification result, the binary mask is supplemented and corrected to generate an optimized binary mask, and the material level region is identified based on the pixel value of the optimized binary mask.

6. The method of claim 1, wherein, The material level region is processed and segmented through morphological operation to obtain the material level binary image, specifically including: The morphological morphologyEx function is used and the operation type is specified as opening operation, and the opening operation is performed on the material level region image using a preset structure element to eliminate isolated noise points and small interference regions; The morphological morphologyEx function is used and the operation type is specified as closing operation, and the closing operation is performed on the material level region image after the opening operation to fill the hole defects of the material level region; A preset connected region area threshold is set, the findContours function is called to detect and calculate the connected region area in the material level region, if the connected region area is less than the connected region area threshold, the corresponding connected region is removed, and the remaining connected region is segmented to obtain the material level binary image.

7. A dynamic feeding strategy based stock level automatic feeding system applied to the dynamic feeding strategy based stock level automatic feeding method according to any one of claims 1-6, characterized in that, The system comprises: An image acquisition and preprocessing module for acquiring a multi-modal hopper image and performing preprocessing through a visual acquisition control device; A material level image segmentation module for identifying and segmenting the material level region of the preprocessed multi-modal hopper image based on Hough circle transformation, color segmentation algorithm, texture analysis algorithm and morphological operation to obtain a material level binary image; The material level region of the preprocessed multi-modal hopper image is identified and segmented based on Hough circle transformation, color segmentation algorithm, texture analysis algorithm and morphological operation to obtain a material level binary image, specifically including: The hopper profile is detected based on the Hough circle transformation, and the hopper trapezoidal ROI region is formed according to the hopper conical geometric feature; A preset pixel value threshold range is set, and the preprocessed multi-modal hopper image is converted from a three-dimensional image to a two-dimensional image based on the color segmentation algorithm to generate a binary mask in combination with the hopper trapezoidal ROI region; The binary mask is corrected based on the texture analysis algorithm to identify the material level region; The material level region is processed and segmented through morphological operation to obtain the material level binary image; The material level state analysis module is configured to construct a mapping relationship between multi-modal hopper image coordinates and actual material level spatial coordinates, extract and calculate material level distribution features of the material level binary image, and generate a material level state analysis result. The material level state analysis module is configured to construct a mapping relationship between multi-modal hopper image coordinates and actual material level spatial coordinates, extract and calculate material level distribution features of the material level binary image, and generate a material level state analysis result. Compute homography transformation matrix based on findHomography function The multi-modal hopper image coordinates are mapped to the actual space coordinates of the material level, and the corresponding calculation formula is as follows: wherein, denotes a multi-modal hopper image coordinate; denotes a level actual space coordinate; denotes a non-zero scale factor; denote elements of a homography transformation matrix denote elements of a homography transformation matrix The moments function is used to perform order weighting summation on pixel coordinates of the material level binary image, quantize the pixel coordinates, and aggregate the pixel coordinates into an image moment set, and based on the image moment set, material level distribution feature parameters are calculated to extract the material level distribution features. The material level distribution feature parameters are input into the minAreaRect function, the outer contour of the material level region is fitted and projected, the minimum area circumscribed rectangle corresponding to each side of the outer contour is calculated and selected, and the circumscribed rectangle width, height and inclination angle are obtained. According to the circumscribed rectangle width, the circumscribed rectangle height and the circumscribed rectangle inclination angle, the cross section of the material level area is evenly divided into a plurality of vertical material level areas, and the material level height of each vertical material level area is weighted and averaged to obtain a material level height value of the material level area under the multi-modal hopper image coordinates The corresponding calculation formula is as follows: where n represents the total number of vertical stock level regions; represents the stock level height value of the i-th vertical stock level region in the multi-modal hopper image coordinates; represents the weight coefficient of the i-th vertical stock level region; Based on the mapping relationship between the multi-modal hopper image coordinates and the actual space coordinates of the material level, the material level height value is calculated according to the material level height value The actual space height value of the material level is calculated The material level height change rate is calculated by linear regression method ; a preset level height grade range, according to the level actual space height value reading the corresponding level height grade; The material level height standard deviation s is calculated according to the material level height of each vertical material level region, and the corresponding calculation formula is as follows: wherein, represents the actual space height value of the level of the i-th vertical level area; represents the average actual space height value of all vertical level areas; A preset material level height standard deviation threshold is set, and if the material level height standard deviation s is less than the material level height standard deviation threshold, it is determined that the material level region is uniformly distributed, otherwise it is determined that the material level region is not uniformly distributed, and a material level distribution uniformity flag is generated. When it is determined that the material level distribution is uneven, a preset threshold for the rate of change of material level height is used. The center height value and the edge height value of the material level area are compared. If the center height value is less than the edge height value, and the rate of change of material level height is... If the material level is less than the threshold value for the rate of change of material level height, then the material level area is marked as a bridge and a bridge flag is generated; otherwise, the material level area is marked as an accumulation and an accumulation flag is generated. The material level height grade and the material level distribution uniformity flag are statistically aggregated to generate the material level state analysis result. The material level state analysis module is configured to construct a mapping relationship between multi-modal hopper image coordinates and actual material level spatial coordinates, extract and calculate material level distribution features of the material level binary image, and generate a material level state analysis result. The material level state analysis module is configured to construct a mapping relationship between multi-modal hopper image coordinates and actual material level spatial coordinates, extract and calculate material level distribution features of the material level binary image, and generate a material level state analysis result. Based on the material level state analysis result, the decision tree nodes of the decision tree algorithm are called for layer-by-layer analysis. The material level height grade is used as a decision tree root node, and the material level height grade is checked. If the material level height grade is a low material level grade or a critical material level grade, a material replenishment preparation is triggered and a material replenishment Boolean value is determined to be true, otherwise the material replenishment Boolean value is determined to be false, and a material replenishment flag is generated. When the material replenishment Boolean value is determined to be true, the bridge flag is checked. If the bridge flag is true, a bridge processing flow is triggered, the material level state analysis result is reacquired, and the material level height grade is reevaluated. If the bridge flag is false, the material level distribution uniformity flag is checked. If the material level distribution uniformity flag is true, a target material level height value is set to an upper limit of a medium material level. If the material level distribution uniformity flag is false, the target material level height value is set to a lower limit of the medium material level. The feed rate V is calculated based on the hopper truncated cone volume formula, and the feed rate is based on the material level height change rate. The formula for calculating the feeding rate v is as follows: wherein, represents the target level height value; R represents the radius of the lower opening of the hopper; represents the preset base coefficient of the decision tree; represents the uniformity correction coefficient; The material replenishment control unit checks the material replenishment flag. If the material replenishment flag is true, a material replenishment time is predicted and a material replenishment control flow is started for material replenishment. If the material replenishment flag is false, the material replenishment control flow is ended. According to a preset sampling period, the material level state change is monitored, the obtained actual material level distribution uniformity, actual material level precision, actual material level rising rate and actual replenishment time are compared with the dynamic replenishment strategy, the replenishment strategy parameters of the dynamic replenishment strategy are adjusted in real time, and the adjusted replenishment strategy parameters are saved in the replenishment control unit for iterative optimization.

Images