Full-automatic breaking and detecting method and system for grain particle material

Abstract

The application discloses a kind of grain granular material full-automatic crushing, detection method and system, it is related to crushing detection technical field.The present application includes the following steps, step 1: whether first raw grain is rice, if first raw grain is not rice, next step is carried out, otherwise, if first raw grain is rice, jump to step 3, step 2: first raw grain is treated with constant volume, jump to step 4, step 3: rice is treated with constant volume, huller module is husked to obtain brown rice, huller module automatically adjusts gap, and different types of rice are husked, step 4: whether the moisture content of brown rice after husking exceeds 15%, if moisture content does not exceed 15%, jump to step 7, if moisture content exceeds 15%, next step is carried out, the present application can realize sample from crushing, quantitative weighing, liquid addition, oscillation extraction, separation, pipetting, dilution, incubation to physical and chemical index detection full-process automation and unmanned.

Description

Technical Field

[0001] This invention relates to the field of crushing and detection technology, specifically to a fully automated crushing and detection method and system for grain pellets. Background Technology

[0002] The grain storage process requires testing for mycotoxins and heavy metals. There are national and industry standards for the testing of mycotoxins and heavy metals. Both require that the raw grain be crushed and mixed first, then weighed to a specified weight, extracted with extractant, diluted with diluent, and then checked, incubated, and tested. Currently, this series of steps is mainly done manually.

[0003] Existing crushing methods cannot achieve fully automated adjustment and detection throughout the entire process, requiring significant human intervention in the crushing and detection process, especially in the identification and sieving of grain types, which relies entirely on manual selection. This results in inefficient detection processes and the susceptibility of human detection to subjective factors, thus affecting the test results. Therefore, utilizing machine vision to improve the fully automated crushing and detection process for grain types is a technical problem that needs to be solved by those skilled in the art. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a fully automated method and system for crushing and testing grain pellets. Given the current reliance on manual operation for testing the physicochemical indicators of grain during procurement, this invention utilizes a third-party mycotoxin and heavy metal detection module, supplemented by a robotic arm module, successfully achieving fully automated operation from raw grain reception to result output, thus resolving the problems mentioned in the background section.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a fully automated method for crushing and detecting grain pellets, comprising the following steps:

[0006] Step 1: Distribute the grain type, sample number, test items and moisture content of the first raw grain, and determine whether the first raw grain is rice. If the first raw grain is not rice, proceed to the next step 2. Otherwise, if the first raw grain is rice, skip to step 3.

[0007] Step 2: Perform volume adjustment on the first raw grain. Use a fixed-volume container to adjust the volume, discard any excess first raw grain, and proceed to Step 4.

[0008] Step 3: The paddy rice is subjected to volume control. The rice hulling module removes the hull from the paddy rice to obtain brown rice. The rice hulling module uses an image recognition algorithm to determine the hulling efficiency. The rice hulling module automatically adjusts the gap to adapt to the hulling of different types of paddy rice.

[0009] Step 4: Determine whether the moisture content of the hulled brown rice exceeds 15%. If the moisture content does not exceed 15%, proceed to step 7. If the moisture content exceeds 15%, proceed to step 5.

[0010] Step 5: Microwave heating and drying the first raw grain or brown rice to obtain the second raw grain. Specifically, the first raw grain or brown rice is transferred to the microwave heating stage using a robotic arm module for heating and dehydration. The microwave heating and moisture measurement module is used to detect the moisture content of the first raw grain or brown rice. If the moisture content exceeds the set moisture threshold, it is transferred to the microwave heating unit for microwave heating and drying using the robotic arm module. The moisture content of the second raw grain after heating and drying is lower than the set moisture threshold. Then, it is transferred to the crushing module using the robotic arm module.

[0011] Step 6: Determine again whether the moisture content of the second raw grain exceeds 15%. If the moisture content does not exceed 15%, proceed to the next step 7. If the moisture content exceeds 15%, re-mark the second raw grain as the first raw grain and jump to step 5.

[0012] Step 7: The robotic arm module pours brown rice or the second raw grain into the crushing module, which is a crusher. The crushing module crushes the second raw grain or brown rice to obtain a powder sample. The brown rice or the second raw grain is graded and crushed and the powder sample is screened by airflow. After the powder sample is transferred from the crushing module to the weighing and feeding module by the robotic arm module, the robotic arm module uses its mounted nozzle to clean the top cover and the cavity. The crushing module is used to crush the brown rice or the second raw grain. After crushing, the crushed powder sample is transferred to the weighing and feeding module by the robotic arm module.

[0013] Step 8: The robotic arm module transfers approximately 15 grams to the weighing and feeding module for weighing and feeding. This precisely controls the quantity of powdered sample entering subsequent operations. The weighing and feeding module is used to weigh the crushed powdered sample and then feed it to the feeding station. The robotic arm module takes the large centrifuge tube after the bottle cap is opened and moves it to the weighing and feeding area. The weighing and feeding module accurately weighs the corresponding amount of powder into the large centrifuge tube. After weighing is completed, the residue in the weighing tube is cleaned.

[0014] Step 9: The robotic arm module picks up the large centrifuge tube consumables and moves them to the bottle cap opening station. The robotic arm module performs the bottle cap opening operation and then moves the large centrifuge tube with the bottle cap open to the weighing and feeding module. The large centrifuge tube is weighed and then picked up by the robotic arm module and moved to the bottle cap opening station. The bottle cap opening and closing module performs the bottle cap closing operation, and then the robotic arm module picks up the large centrifuge tube with the bottle cap closed and moves it to the detection module.

[0015] Step 10: Accurately weigh 5 grams into the large centrifuge tube, weigh and load the material, recalibrate to ensure weighing accuracy, and clean the weighing tubing residue after weighing.

[0016] Step 11: The robotic arm module picks up the weighed large centrifuge tube and moves it to the bottle cap opening station. After weighing, it performs the bottle cap closing operation.

[0017] Step 12: The robotic arm module picks up the large centrifuge tube with the bottle cap closed and transfers it to the detection module to begin detection;

[0018] Step 13: Configure the testing items, including setting the powder sample, test number and test item content, querying and reporting the results. The testing module is used to test the corresponding hygiene indicators of the test samples in the large centrifuge tube.

[0019] Furthermore, the rice hulling machine module uses an image recognition algorithm to determine the dehulling efficiency. The specific steps are as follows:

[0020] Step 31: The rice hulling module completes the initial hulling of the paddy rice to obtain brown rice. The hulled brown rice enters the screening tray through the discharge port of the rice hulling module.

[0021] Step 32: The processing unit sends instructions to the vibration module and the airflow module respectively. The airflow module generates airflow that blows towards the screening tray. Unhulled rice grains will carry the rice husks into the screening tray. Since the weight of the rice husks is much less than the weight of the rice grains themselves, when the airflow module blows air onto the surface of the vibrating screening tray, the rice husks will shift under the influence of the airflow and be inconsistent with the vibration direction, making it easier for the vision module to detect.

[0022] Step 33: Several image sensors are mounted above the screening tray. Each image sensor corresponds to a different recognition angle and focal length. The intersection of the several image sensors is located on the inner surface of the screening tray. By setting multiple image sensors with different recognition angles and focal lengths to form a sensor array, high-definition image acquisition of the screening tray can be achieved using image fusion technology, providing image data support for subsequent machine vision algorithms and helping to improve the accuracy of image recognition algorithms. The processing unit sets the acquisition frequency.

[0023] Step 34: The image sensor acquires image information according to the acquisition frequency and transmits it to the processing unit. The processing unit executes the image fusion program to obtain frame images.

[0024] Step 35: The processing unit compares the frame images before and after the previous frame images in chronological order using optical flow. The grayscale value of the same pixel in adjacent frame images remains unchanged, according to the formula... Taylor expansion and ignoring higher-order terms yields x represents the initial x-direction in the frame image, y represents the initial y-direction in the frame image, and t represents the initial time in the frame image. Here, the initial x-direction, initial y-direction, and initial time are all located in the frame preceding the frame being compared. , u and v are the velocity components of the same pixel in the x and y directions, respectively; Δx is the displacement distance of the pixel in the x direction between two consecutive frames; Δy is the displacement distance of the pixel in the y direction between two consecutive frames; and Δt is the time interval between two consecutive frames. s, Ix are the x-direction values ​​in the spatial gradient of the frame image, Iy are the y-direction values ​​in the spatial gradient of the frame image, and It is the value of the temporal gradient of the frame image. The processing unit preprocesses the frame image, performs Gaussian blur noise reduction on the frame image to reduce the interference of high-frequency noise on gradient calculation, and normalizes the gray values ​​of all pixels in the frame image to ensure consistency under different lighting conditions.

[0025] Step 36: The processing unit calculates the spatial gradient (Ix, Iy) and the temporal gradient It for the frame image respectively;

[0026] Step 37: For each pixel (x, y) in the frame image, extract all points within the adjacent region window of the pixel and construct a system of equations:

[0027] ,in , Matrix A represents the set of motion constraints for all pixels within a neighboring region, with each row representing the contribution of a pixel to the optical flow equation. Vector b contains rows where each row represents the inverted temporal gradient of each pixel within the neighboring region. The velocity vector is then solved using the least squares method. ATA is the covariance matrix of the spatial gradients of adjacent regions, reflecting the gradient distribution; ATb is the correlation between the temporal gradient and time variation, determining the optical flow direction. The processing unit analyzes the displacement features of each pixel in the frame image according to the formula... Calculate the velocity amplitude Vmag of the pixel using the formula. Calculate the motion direction θ of the pixel using the formula. The trajectory length L of pixel i in the frame image is calculated, where N is the number of frames. When the optical flow method calculates the displacement of a pixel in a single frame image, it is easily affected by noise, changes in illumination, etc., resulting in large fluctuations in the displacement data of a single frame. By accumulating the displacement of N consecutive frames, the influence of short-term noise can be smoothed and more stable motion features can be extracted.

[0028] Step 38: The processing unit sets a velocity threshold Vth, a direction threshold θth, and a length threshold Lth. The processing unit sets trigger conditions one, two, and three. Trigger condition one is that the velocity amplitude of a pixel in the frame image is greater than the velocity threshold Vth. Trigger condition two is that the motion direction θ of a pixel in the frame image is greater than the direction threshold θth. Trigger condition three is that the trajectory length L of a pixel in the frame image is less than the length threshold Lth. When any two of trigger conditions one, two, and three are satisfied, the processing unit marks the corresponding pixel as a rice husk.

[0029] Step 39: The processing unit presets an adjustment threshold and calculates the proportion of rice husks in the rice grains in the frame image in real time. When the calculated proportion is greater than the adjustment threshold, the processing unit controls the rice hulling machine module to increase the squeezing friction by a fixed proportion. The squeezing friction of the rice hulling machine module increases to its maximum and then stops increasing. The processing unit presets an adjustment period. If the proportion is never greater than the adjustment threshold within the adjustment period, the processing unit controls the rice hulling machine module to reduce the squeezing friction by half of the fixed proportion. After rice hulling, brown rice is obtained, which is used to improve the hulling efficiency of the rice hulling machine and avoid excessive rice husks affecting subsequent processing steps.

[0030] Furthermore, the image fusion process specifically includes the following steps:

[0031] Step 341: The processing unit fuses image information from multiple recognition angles at the same focal length, identifies feature points in each image based on the SIFT recognition algorithm, and fuses the image information with the help of the recognition angle. The image information at the same focal length is fused to obtain a focal length map.

[0032] Step 342: The processing unit repeats step 341 until all image information is fused into focal length maps of different focal lengths. The focal length maps are arranged in ascending order of focal length to obtain a focal length queue. The processing unit counts the number p of the focal length queue.

[0033] Step 343: The processing unit divides each focal length image into equal parts. The process of obtaining a patch involves dividing the length and width of the rectangular focal length image into p equal parts, converting the patch into a grayscale image, calculating and accumulating the grayscale value difference between adjacent pixels in the patch, and arranging the patches in descending order of the difference value to obtain the difference queue of the focal length image.

[0034] Step 344: The processing unit extracts the first values ​​from the difference queue of each focal length image. The processing unit places the extracted image tiles according to their positions in the focal length image to obtain a stitched image;

[0035] Step 345: The processing unit scans the stitched image. The extracted patches may not be able to completely stitch together the frame image. Patches at the same position in the focal length image may overlap or be missing. If patches overlap at the same position in the stitched image, the processing unit compares the difference between the overlapping patches and retains the patch with the largest difference. If there are missing patches in the stitched image, the processing unit extracts the patch with the largest difference from the corresponding position in all focal length images and adds it to the stitched image.

[0036] Step 346: The processing unit repeats step 345 until there are no overlapping or missing tiles in the stitched image. Then, the processing unit marks the complete stitched image as a frame image.

[0037] Furthermore, the calculation of the spatial gradient (Ix, Iy) of the frame image specifically includes the following steps:

[0038] The spatial gradient (Ix, Iy) is calculated using either the Sobel operator or the Scharr operator. Both the Sobel and Scharr operators are discrete differential operators that calculate the spatial gradient of an image through convolution. The Sobel operator has a 3x3 kernel and is combined with Gaussian smoothing, making it robust to noise. The Scharr operator also has a 3x3 kernel but with larger coefficients, making it more sensitive to low-contrast edges.

[0039] The specific steps for calculating the Sobel operator are as follows:

[0040] Step 361:

[0041] The x-direction kernel of the transverse gradient. The central column has a higher weight, highlighting the horizontal edges;

[0042] The y-direction kernel of the longitudinal gradient. The central row has a higher weight, highlighting the vertical edges;

[0043] Step 362: Fill the edges of the frame image with blank or mirrored pixels to increase the frame image size and avoid shrinking the size after convolution;

[0044] Step 363: For each pixel (g, k) in the frame image, multiply the resulting 3x3 neighboring regions and sum them. The specific formula is as follows:

[0045] ,

[0046] Where Igray is the gray value of the corresponding pixel when the frame image is converted into a grayscale image;

[0047] Step 364: Normalize Ix and Iy to [0, 255], using the formula: , The convolution result may be negative, so normalization is required.

[0048] Furthermore, the Scharr operator computation steps are as follows:

[0049] Lateral gradient kernel in the x-direction: Vertical gradient kernel in the y-direction: It has a larger central coefficient and is more sensitive to weak edges;

[0050] The remaining computation steps of the Scharr operator are the same as those of the Sobel operator, only the kernel is different.

[0051] Furthermore, the calculation of the temporal gradient It for the frame image specifically includes the following steps:

[0052] Step 365: Calculate the time gradient It using the finite difference method, as shown in the following formula:

[0053] For each pixel (e, o) in the frame image, subtract the gray value of the corresponding pixel in the current frame image from the gray value of the corresponding pixel in the next frame image.

[0054] Step 366: Smooth the obtained temporal gradient It using Gaussian filtering. Due to image noise and the discontinuity of object motion, the directly calculated temporal gradient may contain noise and outliers. To make the calculation results more stable and reliable, the calculated temporal gradient can be smoothed.

[0055] Furthermore, the graded crushing process employs image recognition algorithms to determine the crushing effect. The specific steps are as follows:

[0056] Step 71: The crushing process is divided into three stages: coarse crushing, medium crushing, and fine crushing. The crushing range for the coarse crushing stage of brown rice or the second raw grain is set to 10-20 mm, the crushing range for the medium crushing stage is 3-5 mm, and the crushing threshold for the fine crushing stage is 20 mesh.

[0057] Step 72: The processing unit controls the crushing module to crush the brown rice or the second raw grain. The processing unit obtains image information from the vision module and identifies whether the diameter of 95% of the crushed brown rice or the second raw grain meets the range set for the corresponding stage. If it meets the range, it enters the next stage; otherwise, if it does not meet the set range, it repeats the current stage. When 98% of the brown rice or the second raw grain is smaller than 20 mesh in the fine crushing stage, the grading crushing is completed.

[0058] Step 73: After the brown rice or the second raw grain has completed the fine crushing stage, a powder sample is obtained. Airflow sieving is used, with a 20-mesh sieve set up. The sieve is perpendicular to the airflow direction of the airflow module. The processing unit sends a command to the airflow module, and the airflow module blows airflow onto the crushed powder sample. The processing unit adjusts the airflow volume and wind speed of the airflow module. The airflow volume is controlled at 500-1000 cubic meters / hour, and the wind speed is controlled at 10-15 meters / second, so that the airflow can lift the powder sample and sieve it.

[0059] Furthermore, the vision module determines the crushing efficiency of the crushing module by including the following steps:

[0060] Step 76: The processing unit acquires the focal length f, pixel size s, and mounting height H of each image sensor in the vision module. The processing unit acquires the image information of each image sensor, randomly selects one of the powder samples, and measures its image size. Image size Specifically, it refers to the number of pixels occupied by the particles in the powder sample;

[0061] Step 77: The processing unit processes the data according to the formula. Calculate and record the size of the selected powder sample particles;

[0062] Step 78: Repeat step 77. The processing unit continuously counts the percentage of powdered samples that meet the range. Each time step 77 is repeated, the percentage is recalculated.

[0063] An automated grain pellet crushing and testing system is disclosed, comprising a volume control module, a rice huller module, a vision module, an airflow module, a vibration module, a microwave heating and moisture measurement module, a crushing module, a large centrifuge tube management module, a weighing and feeding module, a bottle cap opening and closing module, a testing module, and a robotic arm module. The outputs of the vision module, weighing and feeding module, bottle cap opening and closing module, and testing module are all connected to the input of a processing unit. The output of the processing unit is connected to the inputs of the volume control module, microwave heating and moisture detection module, airflow module, crushing module, large centrifuge tube management module, robotic arm module, vibration module, and rice huller module, respectively.

[0064] The volume control module is used to receive the first raw grain or rice sample from the upper level and perform volume control. The rice hulling module is used to dehull the rice. The microwave heating and moisture measurement module is used to dry the first raw grain or brown rice with excessive moisture. The crushing module is used to crush the second raw grain through a 20-mesh sieve. The large centrifuge tube management module is used to manage the consumables of the large centrifuge tube. The weighing and feeding module is used to accurately weigh the powdered sample and transfer it to the large centrifuge tube to be tested. The bottle cap opening and closing module is used for opening and closing the bottle cap before weighing. The detection module is used to detect specified hygiene indicators. The robotic arm module is used to transfer the powdered sample between different modules.

[0065] Furthermore, the processing unit integrates a GPU for image processing. The vibration module is located directly below the screening tray of the rice huller module. The bottom of the screening tray is a slide rail, and the direction of the slide rail is consistent with the vibration direction of the vibration module, which can realize the reciprocating vibration of the screening tray. The vibration module is specifically a vibration motor, which is used to drive the entire screening tray to vibrate and turn the brown rice. The airflow module is specifically two sets of blowers with adjustable speed. One set of blowers has its air outlet facing perpendicular to the vibration direction of the vibration module. The blower is located on one side of the rice huller's discharge port and is used to screen rice husks. The other set of blowers is located on one side of the crushing module's discharge port and is used to screen powder samples of different sizes after crushing.

[0066] The vision module consists of several image sensors with different focal lengths and different recognition angles. The vision module is located above the rice hulling module and the crushing module, respectively. It is used to detect the dehulling efficiency of the rice hulling module and the crushing efficiency of the crushing module. Through the image recognition algorithm of the vision module, the size of rice husks and powder samples can be dynamically identified during the dehulling and crushing process.

[0067] The present invention has the following beneficial effects:

[0068] 1. It can realize the full automation and unmanned operation of the sample from crushing, quantitative weighing, liquid addition, shaking extraction, separation, transfer, dilution, incubation to physicochemical index detection;

[0069] 2. By working together with the vibration module and the airflow module, the vision module can quickly identify rice husks, while the airflow module can simultaneously sieve the crushed powdery samples, realizing a fully automated crushing and detection process without human intervention and improving the efficiency of crushing and detection.

[0070] 3. Through the image recognition algorithm of the vision module, the size of rice husks and powder samples can be dynamically identified during the dehulling and crushing process. Compared with manual sieving, machine vision has higher sieving accuracy and lower error rate, which is beneficial to the reliability of subsequent detection data.

[0071] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0072] To more clearly illustrate the technical solutions of the embodiments of the present invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0073] Figure 1 This is a flowchart of a fully automated method for crushing and detecting grain pellets according to the present invention.

[0074] Figure 2 This is a block diagram of a fully automatic grain pellet material crushing and detection system according to the present invention. Detailed Implementation

[0075] 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.

[0076] Please see Figure 1-2 This invention provides a technical solution: a fully automated method for crushing and detecting grain pellets, such as... Figure 1 As shown, it includes the following steps:

[0077] Step 1: Distribute the grain type, sample number, test items and moisture content of the first raw grain, and determine whether the first raw grain is rice. If the first raw grain is not rice, proceed to the next step; otherwise, if the first raw grain is rice, skip to step 3.

[0078] Step 2: Perform volume adjustment on the first raw grain. Use a fixed-volume container to adjust the volume, discard any excess first raw grain, and proceed to Step 4.

[0079] Step 3: The paddy rice is subjected to volume control. The rice hulling module removes the hull from the paddy rice to obtain brown rice. The rice hulling module automatically adjusts the gap to adapt to different types of paddy rice.

[0080] Step 4: Determine if the moisture content of the hulled brown rice exceeds 15%. If the moisture content does not exceed 15%, proceed to step 7. If the moisture content exceeds 15%, proceed to the next step.

[0081] Step 5: Microwave the first raw grain or brown rice to dry it and obtain the second raw grain;

[0082] Step 6: Determine again whether the moisture content of the second raw grain exceeds 15%. If the moisture content does not exceed 15%, proceed to the next step 7. If the moisture content exceeds 15%, re-mark the second raw grain as the first raw grain and jump to step 5.

[0083] Step 7: The second raw grain or brown rice is crushed to obtain a powder sample. The brown rice or second raw grain is graded, crushed, and air-flow sieve is used to separate the powder sample.

[0084] Step 8: Weigh and load the powdered sample to precisely control the quantity of powdered sample entering subsequent operations;

[0085] Step 9: Weigh the centrifuge tubes;

[0086] Step 10: Weigh and load the material, recalibrate, and ensure weighing accuracy;

[0087] Step 11: Weigh the bottle and then close the cap;

[0088] Step 12: Transfer to the detection module to begin detection;

[0089] Step 13: Query and report the results.

[0090] The rice huller module uses an image recognition algorithm to determine the dehulling efficiency. The specific steps are as follows:

[0091] Step 31: The rice hulling module completes the initial hulling of the paddy rice to obtain brown rice. The hulled brown rice enters the screening tray through the discharge port of the rice hulling module.

[0092] Step 32: The processing unit sends instructions to the vibration module and the airflow module respectively. The airflow module generates airflow that blows towards the screening tray. Unhulled rice grains will carry the rice husks into the screening tray. Since the weight of the rice husks is much less than the weight of the rice grains themselves, when the airflow module blows air onto the surface of the vibrating screening tray, the rice husks will shift under the influence of the airflow and be inconsistent with the vibration direction, making it easier for the vision module to detect.

[0093] Step 33: Several image sensors are mounted above the screening tray. Each image sensor has a different recognition angle and focal length. The intersection of the orientations of several image sensors is located on the inner surface of the screening tray. For example, the first image sensor has a recognition angle of 15 degrees vertically downward and eastward and a focal length of 1 meter. The default orientation of the rice huller's discharge port is east. The second image sensor has a recognition angle of 10 degrees vertically downward and southward and a focal length of 0.9 meters. By setting multiple image sensors with different recognition angles and focal lengths to form a sensor array, high-definition image acquisition of the screening tray can be achieved using image fusion technology. This provides image data support for subsequent machine vision algorithms and helps improve the accuracy of image recognition algorithms. The processing unit is set to acquire images at a frequency of 24 Hz, meaning that the image sensors acquire image information 24 times per second.

[0094] Step 34: The image sensor acquires image information according to the acquisition frequency and transmits it to the processing unit. The processing unit executes the image fusion program to obtain frame images.

[0095] Step 35: The processing unit compares the frame images before and after the previous frame images in chronological order using optical flow. The grayscale value of the same pixel in adjacent frame images remains unchanged, according to the formula... Taylor expansion and ignoring higher-order terms yields x represents the initial x-direction in the frame image, y represents the initial y-direction in the frame image, and t represents the initial time in the frame image. Here, the initial x-direction, initial y-direction, and initial time are all located in the frame preceding the frame being compared. , u and v are the velocity components of the same pixel in the x and y directions, respectively; Δx is the displacement distance of the pixel in the x direction between two consecutive frames; Δy is the displacement distance of the pixel in the y direction between two consecutive frames; and Δt is the time interval between two consecutive frames. s, Ix are the x-direction values ​​in the spatial gradient of the frame image, Iy are the y-direction values ​​in the spatial gradient of the frame image, and It is the value of the temporal gradient of the frame image. The processing unit preprocesses the frame image, performs Gaussian blur noise reduction on the frame image to reduce the interference of high-frequency noise on gradient calculation, and normalizes the gray values ​​of all pixels in the frame image to ensure consistency under different lighting conditions.

[0096] Step 36: The processing unit calculates the spatial gradient (Ix, Iy) and the temporal gradient It for the frame image respectively;

[0097] Step 37: For each pixel (x, y) in the frame image, extract all points within the neighboring region window of the pixel. For example, for 8 neighboring pixels within a 3x3 window, construct a system of equations:

[0098] ,in , Matrix A represents the set of motion constraints for all pixels within a neighboring region, with each row representing the contribution of a pixel to the optical flow equation. Vector b contains rows where each row represents the inverted temporal gradient of each pixel within the neighboring region. The velocity vector is then solved using the least squares method. ATA is the covariance matrix of the spatial gradients of adjacent regions, reflecting the gradient distribution; ATb is the correlation between the temporal gradient and time variation, determining the optical flow direction. The processing unit analyzes the displacement features of each pixel in the frame image according to the formula... Calculate the velocity amplitude Vmag of the pixel using the formula. Calculate the motion direction θ of the pixel using the formula. The trajectory length L of pixel i in the frame image is calculated, where N is the number of frames. When the optical flow method calculates the displacement of a pixel in a single frame image, it is easily affected by noise, changes in illumination, etc., resulting in large fluctuations in the displacement data of a single frame. By accumulating the displacement of N consecutive frames, the influence of short-term noise can be smoothed and more stable motion features can be extracted.

[0099] Step 38: The processing unit sets a velocity threshold Vth, a direction threshold θth, and a length threshold Lth. The velocity threshold Vth is 2.8 pixels / frame, the direction threshold θth is 5 degrees, and the length threshold Lth is 0.3 meters. The processing unit sets trigger conditions one, two, and three. Trigger condition one is that the velocity amplitude of a pixel in the frame image is greater than the velocity threshold Vth. Trigger condition two is that the motion direction θ of a pixel in the frame image is greater than the direction threshold θth. Trigger condition three is that the trajectory length L of a pixel in the frame image is less than the length threshold Lth. When any two of trigger conditions one, two, and three are satisfied, the processing unit marks the corresponding pixel as a rice husk.

[0100] Step 39: The processing unit presets the adjustment threshold to 1.5%. The processing unit calculates the proportion of rice husks in the rice in the frame image in real time. The proportion is specifically obtained by calculating the proportion of the number of pixels marked as rice husks to the total number of pixels in the frame image. Since the frame image contains not only rice but also the screening tray and the corresponding pixels of the environment, the adjustment threshold needs to be set lower to reserve sufficient data redundancy. When the calculated proportion is greater than the adjustment threshold, the processing unit controls the rice hulling module to increase the squeezing friction by a fixed proportion of 1%. The squeezing friction of the rice hulling module will not increase further when it reaches its maximum. The processing unit presets the adjustment cycle to 5 minutes. If the proportion is never greater than the adjustment threshold within the adjustment cycle, the processing unit controls the rice hulling module to reduce the squeezing friction by half of the fixed proportion of 0.5% to improve the hulling efficiency of the rice hulling machine and avoid excessive rice husks affecting subsequent processing steps.

[0101] It should be noted that the rice huller module adjusts the extrusion friction by adjusting the speed difference between the two extrusion rollers inside the rice huller. Rice hullers equipped with speed-adjustable motors on the market can meet the requirements, and no specific model is limited here.

[0102] The image fusion process specifically includes the following steps:

[0103] Step 341: The processing unit fuses image information from multiple recognition angles at the same focal length, identifies feature points in each image based on the SIFT recognition algorithm, and fuses the image information according to the recognition angle. For example, the first image sensor recognizes the image at a vertical downward angle of 15 degrees east, with the direction of the rice huller's discharge port as east. Before executing the recognition algorithm, the image information corresponding to the first image sensor is arranged with an emphasis on the east direction. After executing the SIFT recognition algorithm, the image information at the same focal length is fused to obtain a focal length image.

[0104] Step 342: The processing unit repeats step 341 until all image information is fused into focal length maps of different focal lengths. The focal length maps are arranged in ascending order of focal length to obtain a focal length queue. The processing unit counts the number p of the focal length queue.

[0105] Step 343: The processing unit divides each focal length image into equal parts. The process involves dividing the length and width of the rectangular focal length image into p equal parts, converting the image into a grayscale image with a grayscale value range of [0, 255], calculating and accumulating the grayscale value difference between adjacent pixels in the image, and arranging the images in descending order of the difference value to obtain the focal length image difference queue.

[0106] Step 344: The processing unit extracts the first values ​​from the difference queue of each focal length image. The image patches with higher differences indicate that the clarity of the patch is among the highest in the focal length image, which facilitates the subsequent identification of powdery samples and rice husks. The processing unit places the extracted patches according to their positions in the focal length image to obtain a stitched image.

[0107] Step 345: The processing unit scans the stitched image. The extracted patches may not be able to completely stitch together the frame image. Patches at the same position in the focal length image may overlap or be missing. If patches overlap at the same position in the stitched image, the processing unit compares the difference between the overlapping patches and retains the patch with the largest difference. Theoretically, the larger the difference, the clearer the pixels covered by the patch. If there are missing patches in the stitched image, the processing unit extracts the patch with the largest difference from the corresponding position in all focal length images and adds it to the stitched image.

[0108] Step 346: The processing unit repeats step 345 until there are no overlapping or missing tiles in the stitched image. Then, the processing unit marks the complete stitched image as a frame image.

[0109] The calculation of spatial gradient (Ix, Iy) for a frame image specifically includes the following steps:

[0110] The spatial gradient (Ix, Iy) is calculated using either the Sobel operator or the Scharr operator. Both the Sobel and Scharr operators are discrete differential operators that calculate the spatial gradient of an image through convolution. The Sobel operator has a 3x3 kernel and is combined with Gaussian smoothing, making it robust to noise. The Scharr operator also has a 3x3 kernel but with larger coefficients, making it more sensitive to low-contrast edges.

[0111] The specific steps for calculating the Sobel operator are as follows:

[0112] Step 361:

[0113] The x-direction kernel of the transverse gradient. The central column has a higher weight, highlighting the horizontal edges;

[0114] The y-direction kernel of the longitudinal gradient. The central row has a higher weight, highlighting the vertical edges;

[0115] Step 362: Fill the edges of the frame image with blank or mirrored pixels to increase the frame image size and avoid shrinking the size after convolution;

[0116] Step 363: For each pixel (g, k) in the frame image, multiply the resulting 3x3 neighboring regions and sum them. The specific formula is as follows:

[0117] ,

[0118] Where Igray is the gray value of the corresponding pixel when the frame image is converted into a grayscale image;

[0119] Step 364: Normalize Ix and Iy to [0, 255], using the formula: , The convolution result may be negative, so normalization is required.

[0120] The Scharr operator is calculated as follows:

[0121] Lateral gradient kernel in the x-direction: Vertical gradient kernel in the y-direction: It has a larger central coefficient and is more sensitive to weak edges;

[0122] The remaining computation steps of the Scharr operator are the same as those of the Sobel operator, only the kernel is different.

[0123] The calculation of the temporal gradient It for a frame image specifically includes the following steps:

[0124] Step 365: Calculate the time gradient It using the finite difference method, as shown in the following formula:

[0125] For each pixel (e, o) in the frame image, the gray value of the corresponding pixel in the frame image of the current frame is subtracted from the gray value of the corresponding pixel in the frame image of the next frame. The gray value range is [0, 255].

[0126] Step 366: Smooth the obtained temporal gradient It using Gaussian filtering. Due to image noise and the discontinuity of object motion, the directly calculated temporal gradient may contain noise and outliers. To make the calculation results more stable and reliable, the calculated temporal gradient can be smoothed.

[0127] Among them, the graded crushing process uses an image recognition algorithm to determine the crushing effect. The specific steps are as follows:

[0128] Step 71: The crushing process is divided into three stages: coarse crushing, medium crushing, and fine crushing. The crushing range for the coarse crushing stage of brown rice or the second raw grain is set to 10-20 mm, the crushing range for the medium crushing stage is 3-5 mm, and the crushing threshold for the fine crushing stage is 20 mesh.

[0129] Step 72: The processing unit controls the crushing module to crush the brown rice or the second raw grain. The processing unit obtains image information from the vision module and identifies whether the diameter of 95% of the crushed brown rice or the second raw grain meets the range set for the corresponding stage. If it meets the range, it enters the next stage; otherwise, if it does not meet the set range, it repeats the current stage. When 98% of the brown rice or the second raw grain is smaller than 20 mesh in the fine crushing stage, the grading crushing is completed.

[0130] Step 73: After the brown rice or the second raw grain has completed the fine crushing stage, a powder sample is obtained. Airflow sieving is used, with a 20-mesh sieve set up. The sieve is perpendicular to the airflow direction of the airflow module. The processing unit sends a command to the airflow module, and the airflow module blows airflow onto the crushed powder sample. The processing unit adjusts the airflow volume and wind speed of the airflow module. The airflow volume is controlled at 500-1000 cubic meters / hour, and the wind speed is controlled at 10-15 meters / second, so that the airflow can lift the powder sample and sieve it.

[0131] The vision module determines the crushing efficiency of the crushing module by including the following steps:

[0132] Step 76: The processing unit acquires the focal length f, pixel size s, and mounting height H of each image sensor in the vision module. The processing unit acquires the image information of each image sensor, randomly selects one of the powder samples, and measures its image size. Image size Specifically, it refers to the number of pixels occupied by the particles in the powder sample;

[0133] Step 77: The processing unit processes the data according to the formula. Calculate and record the size of the selected powder sample particles;

[0134] Step 78: Repeat step 77. The processing unit continuously counts the percentage of powdered samples that meet the range. Each time step 77 is repeated, the percentage is recalculated.

[0135] A fully automated grain pellet crushing and detection system, such as Figure 2 As shown, the system includes a volume control module, a rice huller module, a vision module, an airflow module, a vibration module, a microwave heating and moisture measurement module, a crushing module, a large centrifuge tube management module, a weighing and feeding module, a bottle cap opening and closing module, a detection module, and a robotic arm module. The output terminals of the vision module, the weighing and feeding module, the bottle cap opening and closing module, and the detection module are all connected to the input terminal of the processing unit. The output terminal of the processing unit is connected to the input terminals of the volume control module, the microwave heating and moisture detection module, the airflow module, the crushing module, the large centrifuge tube management module, the robotic arm module, the vibration module, and the rice huller module, respectively.

[0136] The volume control module is used to receive the first raw grain or rice sample from the upper level and perform volume control. The rice hulling module is used to dehull the rice. The microwave heating and moisture measurement module is used to dry the first raw grain or brown rice with excessive moisture. The crushing module is used to crush the second raw grain to achieve a crushing rate of over 99% and pass through a 20-mesh sieve. The large centrifuge tube management module is used to manage the consumables of the large centrifuge tube. The weighing and feeding module is used to accurately weigh the powdered sample and transfer it to the large centrifuge tube to be tested. The bottle cap opening and closing module is used for opening and closing the bottle cap before weighing. The detection module is used to test the specified hygiene indicators. The robotic arm module is used to transfer the powdered sample between different modules.

[0137] The processing unit integrates a GPU for image processing, i.e., a graphics acceleration computing unit. Processing frame images, focal length images, and tiles requires a large amount of parallel data computation. The image acceleration unit can alleviate the computational pressure of image information processing in the processing unit and improve the overall response speed. The vibration module is located directly below the screening tray of the rice huller module. The bottom of the screening tray is a slide rail, and the direction of the slide rail is consistent with the vibration direction of the vibration module, which can realize the reciprocating vibration of the screening tray. The vibration module is specifically a vibration motor, which is used to drive the entire screening tray to vibrate and turn the brown rice. The vibration frequency of the vibration module is 100-120 times / minute. The airflow module consists of two sets of adjustable speed blowers. One set of blowers has its air outlet facing perpendicular to the vibration direction of the vibration module. The blower is located on the side of the rice huller's discharge port and is used to screen rice husks. The other set of blowers is located on the side of the crushing module's discharge port and is used to screen powder samples of different sizes after crushing.

[0138] The vision module consists of several image sensors with different focal lengths and different recognition angles. The vision module is located above the rice hulling module and the crushing module, respectively, and is used to detect the dehulling efficiency of the rice hulling module and the crushing efficiency of the crushing module.

[0139] The above description is only a preferred embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A fully automated method for crushing and detecting grain pellets, characterized in that: Includes the following steps: Step 1: Determine if the first raw grain is rice. If the first raw grain is not rice, proceed to the next step. Otherwise, if the first raw grain is rice, skip to step 3. Step 2: Perform volume adjustment on the first raw grain, then proceed to Step 4; Step 3: The paddy rice is subjected to volume control. The rice hulling module removes the hull from the paddy rice to obtain brown rice. The rice hulling module uses an image recognition algorithm to determine the hulling efficiency. The rice hulling module automatically adjusts the gap to adapt to the hulling of different types of paddy rice. Step 4: Determine if the moisture content of the hulled brown rice exceeds 15%. If the moisture content does not exceed 15%, proceed to step 7. If the moisture content exceeds 15%, proceed to the next step. Step 5: Microwave the first raw grain or brown rice to dry it and obtain the second raw grain; Step 6: Determine again whether the moisture content of the second raw grain exceeds 15%. If the moisture content does not exceed 15%, proceed to the next step 7. If the moisture content exceeds 15%, re-mark the second raw grain as the first raw grain and jump to step 5. Step 7: The second raw grain or brown rice is crushed to obtain a powder sample, which is then subjected to graded crushing and airflow sieving. Step 8: Weigh and load the powdered sample to precisely control the quantity of powdered sample entering subsequent operations; Step 9: Weigh the centrifuge tubes; Step 10: Weigh and load the material, then recalibrate; Step 11: Weigh the bottle and then close the cap; Step 12: Transfer to the detection module to begin detection; Step 13: Query and report the results; The graded crushing process uses image recognition algorithms to determine the crushing effect. The specific steps are as follows: Step 71: The crushing process is divided into three stages: coarse crushing, medium crushing, and fine crushing. The crushing range for the coarse crushing stage of brown rice or the second raw grain is set to 10-20 mm, the crushing range for the medium crushing stage is 3-5 mm, and the crushing threshold for the fine crushing stage is 20 mesh. Step 72: The processing unit controls the crushing module to crush the brown rice or the second raw grain. It obtains image information from the vision module and identifies whether the diameter of 95% of the crushed brown rice or the second raw grain meets the range set for the corresponding stage. If it meets the range, it enters the next stage; otherwise, if it does not meet the set range, it repeats the current stage. When 98% of the brown rice or the second raw grain is smaller than 20 mesh in the fine crushing stage, the grading crushing is completed. Step 73: After the brown rice or the second raw grain has completed the fine crushing stage, a powder sample is obtained. Airflow sieving is used, with a 20-mesh sieve set up. The sieve is perpendicular to the blowing direction of the airflow module. The processing unit sends a command to the airflow module, and the airflow module blows airflow onto the crushed powder sample. The airflow volume and wind speed of the airflow module are adjusted to perform sieving. The vision module determines the crushing efficiency of the crushing module through the following steps: Step 76: Obtain the focal length f, pixel size s, and mounting height H of each image sensor in the vision module; acquire the image information of each image sensor; randomly select one of the powder samples and measure its image size. ; Step 77: According to the formula Calculate and record the size of the selected powder sample; Step 78: Repeat step 77 to continuously count the percentage of powdered samples that fall within the range. Each time step 77 is repeated, the percentage is recalculated.

2. The fully automated crushing and detection method for grain pellets according to claim 1, characterized in that, The rice huller module uses an image recognition algorithm to determine the dehulling efficiency. The specific steps are as follows: Step 31: The rice hulling module completes the initial hulling of the paddy rice to obtain brown rice. The hulled brown rice enters the screening tray through the discharge port of the rice hulling module. Step 32: The processing unit sends instructions to the vibration module and the airflow module respectively. The airflow module works to generate airflow that blows towards the screening tray. Step 33: Several image sensors are mounted above the screening tray. Each image sensor corresponds to a different recognition angle and focal length. The intersection of the several image sensors is located on the inner surface of the screening tray. The acquisition frequency is set. Step 34: The image sensor acquires image information according to the acquisition frequency and transmits it to the processing unit. The processing unit executes the image fusion program to obtain frame images. Step 35: Compare the frames in chronological order using optical flow. The grayscale value of the same pixel remains unchanged in adjacent frames. According to the formula... Taylor expansion and ignoring higher-order terms yields x represents the initial x-direction in the frame image, y represents the initial y-direction in the frame image, and t represents the initial time in the frame image. , u and v are the velocity components of the same pixel in the x and y directions, respectively; Δx is the displacement distance of the pixel in the x direction between two consecutive frames; Δy is the displacement distance of the pixel in the y direction between two consecutive frames; and Δt is the time interval between two consecutive frames. s, Ix are the x-direction values ​​in the spatial gradient of the frame image, Iy are the y-direction values ​​in the spatial gradient of the frame image, It is the value of the temporal gradient of the frame image. The frame image is preprocessed, Gaussian blur is applied to the frame image for noise reduction, and the gray values ​​of all pixels in the frame image are normalized. Step 36: Calculate the spatial gradient (Ix, Iy) and temporal gradient It for each frame image; Step 37: For each pixel (x, y) in the frame image, extract all points within the adjacent region window of the pixel and construct a system of equations: Where A is a matrix and b is a vector. , The velocity vector is solved using the least squares method. ATA is the covariance matrix of the spatial gradients of adjacent regions, and ATb is the correlation between the temporal gradient and time variation. The displacement characteristics of each pixel in the frame image are analyzed according to the formula... Calculate the velocity amplitude Vmag of the pixel using the formula. Calculate the motion direction θ of the pixel using the formula. Calculate the length L of the trajectory of pixel i in the frame image, where N is the number of frames; Step 38: Set the velocity threshold Vth, direction threshold θth, and length threshold Lth. Set trigger conditions one, two, and three. Trigger condition one is that the velocity amplitude of a pixel in the frame image is greater than the velocity threshold Vth. Trigger condition two is that the motion direction θ of a pixel in the frame image is greater than the direction threshold θth. Trigger condition three is that the trajectory length L of a pixel in the frame image is less than the length threshold Lth. When any two of trigger conditions one, two, and three are satisfied, the corresponding pixel is marked as rice husk. Step 39: Set an adjustment threshold and calculate the proportion of rice husks in the rice grains in the frame image in real time. When the calculated proportion is greater than the adjustment threshold, the processing unit controls the rice hulling machine module to increase the squeezing friction by a fixed proportion. When the squeezing friction of the rice hulling machine module reaches its maximum, it will not increase further. Set an adjustment period. If the proportion is never greater than the adjustment threshold within the adjustment period, the processing unit controls the rice hulling machine module to reduce the squeezing friction by half of the fixed proportion. After rice hulling, brown rice is obtained.

3. The fully automated method for crushing and detecting grain pellets according to claim 2, characterized in that, The image fusion process specifically includes the following steps: Step 341: Fuse image information from multiple recognition angles at the same focal length, identify feature points in each image based on the recognition algorithm, and fuse the image information with the help of the recognition angle. The image information at the same focal length is fused to obtain a focal length map. Step 342: Repeat step 341 until all image information is fused into focal length maps of different focal lengths. Arrange the focal length maps in ascending order of focal length to obtain a focal length queue. Count the number of focal length queues p. Step 343: Divide each focal length image into p equal parts. 2 The process involves obtaining a patch, converting the patch into a grayscale image, calculating the grayscale difference between adjacent pixels in the patch and accumulating the difference values, and arranging the patches in descending order of the difference values ​​to obtain a difference queue. Step 344: Extract the first values ​​from the difference queue. The extracted image tiles are placed according to their positions in the focal length image to obtain a stitched image; Step 345: Scan the stitched image. If there are overlapping patches at the same position in the stitched image, compare the difference between the overlapping patches and keep the patch with the largest difference. If there are missing patches in the stitched image, extract the patch with the largest difference from the corresponding position in all focal length images and add it to the stitched image. Step 346: Repeat step 345 until there are no overlapping or missing tiles in the stitched image, then mark the complete stitched image as a frame image.

4. The fully automated crushing and detection method for grain pellets according to claim 2, characterized in that, The calculation of spatial gradient (Ix, Iy) for a frame image specifically includes the following steps: The spatial gradient (Ix, Iy) is calculated using either the Sobel operator or the Scharr operator. Both the Sobel and Scharr operators are discrete differential operators, and the spatial gradient of the image is calculated through convolution. The specific steps for calculating the Sobel operator are as follows: Step 361: The x-direction kernel of the transverse gradient. ; The y-direction kernel of the longitudinal gradient. ; Step 362: Fill the edges of the frame image with blank or mirrored pixels; Step 363: For each pixel (g, k) in the frame image, multiply the resulting 3x3 neighboring regions and sum them. The specific formula is as follows: ； Where Igray is the gray value of the corresponding pixel when the frame image is converted into a grayscale image; Step 364: Normalize Ix and Iy to [0, 255].

5. The fully automated method for crushing and detecting grain pellets according to claim 4, characterized in that, The steps for calculating the Scharr operator are as follows: Lateral gradient kernel in the x-direction: Vertical gradient kernel in the y-direction: ; The remaining computation steps of the Scharr operator are the same as those of the Sobel operator, only the kernel is different.

6. The fully automated method for crushing and detecting grain pellets according to claim 2, characterized in that, The calculation of the temporal gradient It for a frame image specifically includes the following steps: Step 365: Calculate the time gradient It using the finite difference method, as shown in the following formula: For each pixel (e, o) in the frame image, subtract the gray value of the corresponding pixel in the current frame from the gray value of the corresponding pixel in the next frame. Step 366: Smooth the obtained time gradient It using Gaussian filtering.

7. A fully automated crushing and detection system for grain pellets, characterized in that... The fully automated crushing and testing method for grain pellets as described in claim 1 includes a volume control module, a rice huller module, a vision module, an airflow module, a vibration module, a microwave heating and moisture measurement module, a crushing module, a large centrifuge tube management module, a weighing and feeding module, a bottle cap opening and closing module, a testing module, and a robotic arm module. The outputs of the vision module, the weighing and feeding module, the bottle cap opening and closing module, and the testing module are all connected to the input of the processing unit. The output of the processing unit is connected to the inputs of the volume control module, the microwave heating and moisture detection module, the airflow module, the crushing module, the large centrifuge tube management module, the robotic arm module, the vibration module, and the rice huller module, respectively. The volume control module is used to receive the first raw grain or rice sample from the upper level and perform volume control. The rice hulling module is used to dehull the rice. The microwave heating and moisture measurement module is used to dry the first raw grain or brown rice with excessive moisture. The crushing module is used to crush the second raw grain through a 20-mesh sieve. The large centrifuge tube management module is used to manage the consumables of the large centrifuge tube. The weighing and feeding module is used to accurately weigh the powdered sample and transfer it to the large centrifuge tube to be tested. The bottle cap opening and closing module is used to open or close the bottle cap before weighing. The detection module is used to detect specified hygiene indicators. The robotic arm module is used to transfer the powdered sample between different modules.

8. The fully automatic grain pellet crushing and detection system according to claim 7, characterized in that, The processing unit integrates a GPU for image processing. The vibration module is located directly below the rice huller module. The vibration module is used to drive the brown rice to turn over. The airflow module is used to sieve the rice husks and sieve the powdered samples of different sizes after crushing. The vision module consists of several image sensors with different focal lengths and different recognition angles. The vision module is located above the rice hulling module and the crushing module, respectively, and is used to detect the dehulling efficiency of the rice hulling module and the crushing efficiency of the crushing module.

Images