Method and apparatus for identifying abnormal polyethylene particles

Abstract

The application discloses an abnormal polyethylene particle identification method and device, and the method comprises the following steps: A, collecting a plurality of polyethylene particle image data in a rolling state at the corresponding position of a polyethylene chute; B, target detection is performed on the plurality of polyethylene particles to obtain a target detection frame containing position information and size information; C, a large-size image containing a plurality of polyethylene particle images is segmented into a plurality of rectangular image blocks through a clustering method, and the rectangular image blocks are associated with the position information of the polyethylene particles; D, the rectangular image blocks are formed into an image Batch for parallel processing, and the chute background, qualified polyethylene particles and abnormal polyethylene particles are distinguished through instance segmentation; E, the identified abnormal polyethylene particles are tracked as targets, the target trajectories are predicted, and the defective products are automatically screened out through the position information and rolling time information of the abnormal polyethylene particles. The application can effectively avoid missed detection of polyethylene particles with black spots.

Description

Technical Field

[0001] This invention relates to the field of chemical product quality testing technology, and in particular to a method and apparatus for identifying abnormal polyethylene particles. Background Technology

[0002] Polyethylene is a thermoplastic resin obtained by polymerizing ethylene. It has excellent low-temperature resistance, good chemical stability, and is resistant to most acids and alkalis. Polyethylene production processes can be divided into four main categories based on reaction conditions: gas-phase processes, slurry processes, solution processes, and high-pressure processes. Among these, the gas-phase process mainly uses the Unipol process. The Unipol polyethylene process is a low-pressure gas-phase fluidized bed method for producing ethylene (co)polymers. This process can use titanium-based catalysts, solid chromium-based catalysts, metallocene catalysts, and bimodal catalysts to produce resin products with different properties, such as HDPE, LLDPE, and VLDPE. Typical product densities range from 0.916 to 0.961 g / cm³, melt indexes from 0.1 to 200, and relative molecular masses from 30,000 to 250,000. Narrow or wide molecular weight distributions can be adjusted depending on the catalyst type. The reactor used in this process is a vertical gas-phase fluidized bed, with a typical reaction pressure of 2.4 MPa and a reaction temperature of 80–110℃.

[0003] In polyethylene (PE) plants, extrusion granulation is a crucial step. Due to prolonged operation and factors such as catalyst coking, abnormally colored or even black granules may appear in the finished PE product. The presence of such granules during product quality testing will result in the product being deemed substandard. High-density polyethylene (HDPE) granules are typically 3mm in diameter, milky white in color, and those with black spots are visually noticeable. These black spots can be randomly distributed across the surface of the white granules, or even inside the granules themselves.

[0004] Chinese patent application CN112837311A discloses a deep learning-based polyethylene particle defect detection and identification system and method. The system includes a control unit, a feeding unit, an identification unit, a screening unit, and a collection component. The feeding unit causes the polyethylene particles to fall in a waterfall-like manner. The identification unit includes several multimodal cameras and an AI processor connected to each camera. Each multimodal camera is located on one side of the polyethylene particle's falling path and captures images of each particle, sending them to the AI ​​processor. The AI ​​processor is connected to the control unit and identifies defective polyethylene particles, sending these images to the control unit. This deep learning-based polyethylene particle defect detection and identification system can achieve high throughput detection. However, this system and method suffer from a high false negative rate when acquiring and analyzing images at high polyethylene particle speeds, and it can only identify particles with obvious black spots on the surface, failing to distinguish defective products with black spots inside the particles.

[0005] Therefore, there is an urgent need for a vision-based method and device for identifying abnormal polyethylene particles without changing the existing process. This method can not only minimize the missed detection of particles with black spots on the surface, but also effectively identify unqualified products with internal black spots.

[0006] The information disclosed in this background section is intended only to enhance the understanding of the overall background of the invention and should not be construed as an admission or in any way implying that the information constitutes prior art known to those skilled in the art. Summary of the Invention

[0007] The purpose of this invention is to provide a vision-based method and apparatus for identifying abnormal polyethylene particles without changing the existing process, which can effectively avoid missing detection of polyethylene particles with black spots.

[0008] To achieve the above objectives, according to a first aspect of the present invention, the present invention provides a method for identifying abnormal polyethylene particles, comprising the following steps: A. acquiring image data of multiple polyethylene particles in a rolling state at corresponding positions on a polyethylene chute; B. performing target detection on the multiple polyethylene particles to obtain target detection boxes containing position and size information; C. segmenting a large image containing multiple polyethylene particle images into several rectangular image blocks using a clustering method, and associating the rectangular image blocks with the position information of the polyethylene particles; D. forming image batches from the rectangular image blocks for parallel processing, and distinguishing between the chute background, qualified polyethylene particles, and abnormal polyethylene particles through instance segmentation; E. tracking the identified abnormal polyethylene particles as targets, predicting the target trajectory, and automatically screening out defective products using the position and rolling time information of the abnormal polyethylene particles.

[0009] Furthermore, in the above technical solution, the abnormal polyethylene particles may include particles with black spots on the surface and particles with black spots inside.

[0010] Furthermore, in the above technical solution, the clustering method in step C can be a fast adaptive clustering method, specifically including: for n coordinate points, k iterations of bisector random sampling are used to calculate the cluster center coordinates and cluster metric, and the number of cluster centers with the minimum cluster metric obtained is the number of initial rectangular image patches; let the number of cluster centers be n. k n k Cluster centers determine n k From the initial rectangular image patches, new rectangular image patches are generated for the coordinate points of particles not included in the initial patches, ultimately obtaining n. k +s coordinate frames.

[0011] Furthermore, in the above technical solution, step D may include: D1, placing nk +s coordinate boxes are arranged into a slice of batch in order from left to right and from top to bottom, and all slices are processed in parallel; D2, perform instance segmentation of the image in batch parallel mode, and determine whether there are surface black spots on the polyethylene particles in the rectangular image block. If surface black spots are present, they are identified as abnormal polyethylene particles.

[0012] Furthermore, in the above technical solution, step D may also include: D3, comparing the image of polyethylene particles without black spots on the surface with a preset fingerprint database of abnormal polyethylene particles; through similarity comparison, if it is close to the threshold, it is determined that the polyethylene particles have internal black spots, and the polyethylene particles with internal black spots are also identified as abnormal polyethylene particles.

[0013] Furthermore, in the above technical solution, the fingerprint information in the fingerprint database may include the minimum, average, and maximum values ​​of image pixels within a closed area, and the fingerprint information may be information after normalization processing.

[0014] Furthermore, in the above technical solution, step E may include: E1, calculating the coordinates of the original large-size image by using the center coordinates of the identified abnormal polyethylene particles and the coordinate information of the corresponding rectangular image block; E2, in multi-target tracking, predicting the position and time required for the abnormal polyethylene particles to pass through the nozzle using a trajectory prediction algorithm; E3, based on the position and time required to pass through the nozzle, controlling the nozzle through an air valve to blow off the abnormal polyethylene particles, thereby completing the automatic screening of defective products.

[0015] According to a second aspect of the present invention, an abnormal polyethylene particle identification device is provided, comprising: an image acquisition unit for acquiring image data of multiple polyethylene particles in a rolling state at corresponding positions of a polyethylene chute; a target detection unit for performing target detection on the multiple polyethylene particles and obtaining a target detection box containing position information and size information; a clustering unit for segmenting a large-size image containing multiple polyethylene particle images into several rectangular image blocks using a clustering method, and associating the rectangular image blocks with the position information of the polyethylene particles; an instance segmentation unit for forming image batches from the rectangular image blocks for parallel processing, and distinguishing between the chute background, qualified polyethylene particles, and abnormal polyethylene particles through instance segmentation; and a tracking and prediction processing unit for tracking the identified abnormal polyethylene particles as targets, predicting the target trajectory, and automatically screening out defective products using the position information and rolling time information of the abnormal polyethylene particles.

[0016] Furthermore, in the above technical solution, the image acquisition unit may specifically include: a polyethylene slide rail, which is used to carry the polyethylene particles to be identified, and the slide rail is a rough surface slide rail with a certain tilt angle; a high-speed camera, which has explosion-proof performance and is arranged above the slide rail, for capturing the polyethylene particles rolling down the slide rail; and an illumination device, which provides supplementary lighting for the polyethylene particles to be captured, and the light source of the illumination device is white light with an illuminance of at least 60,000 Lux.

[0017] Furthermore, in the above technical solution, the polyethylene slide can be made of metal, the surface of the slide is made into a long and narrow chute and sprayed with green; the lighting equipment and high-speed camera are integrated.

[0018] Furthermore, in the above technical solution, the polyethylene slide can also be made of a transparent material, with one side of the transparent material being frosted and the other side being smooth. The frosted surface serves as the contact surface of the slide, and the smooth surface serves as the supplementary lighting surface.

[0019] Furthermore, in the above technical solution, the polyethylene slide is a structure of multiple parallel guide channels. The guide channel has guide strips with an isosceles right triangle cross-section on both sides. The guide strips are made of light-transmitting material and are bonded to the frosted surface of the slide. When the supplementary light shines on the smooth surface of the slide, refraction and reflection are formed at the inclined surface of the guide strips towards the guide channel. The refracted light shines laterally inward into the polyethylene particles in the guide channel, and the reflected light shines laterally outward into the polyethylene particles of the adjacent guide channel.

[0020] Furthermore, in the above technical solution, the transparent material of the polyethylene slide can be glass or acrylic.

[0021] Compared with the prior art, the present invention has the following beneficial effects:

[0022] 1) The method of the present invention acquires images of polyethylene particles through image acquisition, performs target detection of polyethylene particles through a target detection unit, forms image batches through a clustering unit, distinguishes qualified polyethylene particles from abnormal polyethylene particles through an instance segmentation unit, and finally tracks and predicts the multi-target trajectory of polyethylene particles through a tracking and prediction processing unit, and filters out polyethylene particles identified as "abnormal" and blows them off. The whole process is fully automated and does not require operator intervention. It can perform visual screening of massive amounts of polyethylene particles and select unqualified products.

[0023] 2) This invention employs a fast adaptive clustering method, which can effectively reduce the amount of computation while ensuring the required analytical accuracy;

[0024] 3) This invention forms image batches from rectangular image blocks for parallel processing, which can maximize computational efficiency;

[0025] 4) This invention can identify abnormal polyethylene particles with black spots on the surface by segmentation of examples, and can also identify abnormal polyethylene particles with black spots inside by comparing fingerprint information.

[0026] 5) The slide in the device of the present invention is made of transparent material and the design of the guide strip can provide supplementary lighting at the bottom of the slide. The supplementary lighting effect of the particles in the guide groove and the adjacent guide groove is enhanced by refraction and reflection, which creates better conditions for subsequent image analysis and calculation and improves the accuracy.

[0027] The above description is merely an overview of the technical solution of the present invention. In order to better understand the technical means of the present invention and to implement it according to the contents of the specification, and to make the above and other objects, technical features and advantages of the present invention easier to understand, one or more preferred embodiments are listed below and described in detail with reference to the accompanying drawings. Attached Figure Description

[0028] Figure 1 This is a flowchart illustrating the abnormal polyethylene particle identification method of the present invention.

[0029] Figure 2 This is a schematic diagram of embodiment 1 of the image acquisition unit in this invention (the slide is made of metal).

[0030] Figure 3 This is a schematic diagram of embodiment 2 of the image acquisition unit in this invention (the slide is made of transparent material; the arrow in the figure indicates the direction of the supplementary light).

[0031] Figure 4 This is another schematic diagram of embodiment 2 of the image acquisition unit in this invention (the slide is made of transparent material; multiple guide channels are shown arranged side by side).

[0032] Figure 5 This is a schematic diagram of a rectangular image block in the clustering unit of the present invention (showing polyethylene particles and target detection boxes).

[0033] Figure 6 This is a schematic diagram of rectangular image blocks forming a batch slice in the clustering unit of this invention.

[0034] Figure 7 This is a schematic diagram of the image recognition and tracking process in the abnormal polyethylene particle identification method of the present invention.

[0035] Explanation of key figure labels:

[0036] 1-Polyethylene granules, 1A-Abnormal polyethylene granules, 2-Guide channel, 20-Guide strip, 21-Frosted surface, 22-Smooth surface, 3-High-speed camera, 4-Rectangular image block, 5-Target detection box. Detailed Implementation

[0037] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings, but it should be understood that the scope of protection of the present invention is not limited to the specific embodiments.

[0038] Unless otherwise expressly stated, throughout the specification and claims, the term "comprising" or its variations such as "including" or "comprises" shall be understood to include the stated elements or components without excluding other elements or other components.

[0039] In this document, for ease of description, spatial relative terms such as “below,” “under,” “down,” “above,” “above,” “upper,” etc., are used to describe the relationship of one element or feature to another element or feature in the accompanying drawings. It should be understood that spatial relative terms are intended to encompass different orientations of an object in use or operation, in addition to those depicted in the figures. For example, if an object in the figure is flipped, an element described as “below” or “under” another element or feature would be oriented “above” that element or feature. Thus, the exemplary term “below” can encompass both the downward and upward orientations. An object may also have other orientations (rotated 90 degrees or other orientations), and the spatial relative terms used herein should be interpreted accordingly.

[0040] In this document, the terms "first," "second," etc., are used to distinguish two different elements or parts, and are not used to define specific positions or relative relationships. In other words, in some embodiments, the terms "first," "second," etc., can also be used interchangeably.

[0041] like Figure 1 As shown, the present invention provides a method for identifying abnormal polyethylene particles, the method comprising the following steps:

[0042] Step S101: Collect image data of multiple polyethylene particles in a rolling state at the corresponding positions of the polyethylene slide.

[0043] Specifically, this invention acquires images through an image acquisition unit. The image acquisition unit includes a polyethylene chute, a high-speed camera, and lighting equipment. The polyethylene chute, which carries the polyethylene particles to be identified, is a rough-surfaced chute with a certain angle of inclination. The high-speed camera, which is explosion-proof and positioned above the chute, is used to capture images of the polyethylene particles rolling down the chute. The lighting equipment provides supplemental lighting for the polyethylene particles to be photographed; the light source of this lighting equipment is white light with an illuminance of at least 60,000 Lux. Furthermore, image transmission can utilize a 10 Gigabit Ethernet card, and the lighting equipment can utilize industrial supplemental lighting.

[0044] Further as Figures 2 to 4As shown, after extrusion granulation, polyethylene granules are divided into several streams by a diverting device. Each stream passes through a polyethylene granule chute with a certain inclined angle, causing the polyethylene granules to roll. The chute is a rough chute with a certain coefficient of friction (the coefficient of friction can be 0.4 < μ < 0.7). The polyethylene granules roll and fall on the chute, and a high-speed camera captures images of the rolling polyethylene granules at a certain angle. The captured images are transmitted to a back-end computer for image recognition and target tracking.

[0045] For the image acquisition unit, the technical solution of Embodiment 1 can be adopted.

[0046] Example 1

[0047] like Figure 2 As shown, the high-speed camera 3 meets the safety standards of a polyethylene chemical plant, possessing explosion-proof properties. The 10 Gigabit Ethernet card used for transmission meets the requirements for large-scale data transmission. The lighting equipment is selected with a power of 300W, and the light source is white light, deployed above the chute. The polyethylene particle chute (i.e., the guide channel 2) can be made of metal, with the surface made into a long, narrow chute and painted green. The lighting equipment and high-speed camera can be integrated. The high-speed camera 3 is deployed above the guide channel 2, 20cm away from the guide channel, with a shooting angle of 30 degrees to the longitudinal direction of the guide channel 2. Two industrial high-speed cameras 3 are installed above each guide channel 2, and each camera is equipped with two industrial lights (not shown in the figure), with the lights having a 30-degree angle to the longitudinal direction of the guide channel 2. In this embodiment, the camera and lights are combined into an integrated unit, equipped with a pan-tilt unit, and the image acquisition size (width * height) is W * H. The guide channel 2 can be designed with three ( Figure 2 (Only one is shown in the image). Each stream has a flow rate of 4 tons, which meets the flow rate requirements of polyethylene granules. It can achieve a transmission capacity of 7-12 tons per hour. The granule chute is made of stainless steel. The light source is located above the chute. The metal chute needs to be surface treated and coated with paint. The surface treatment involves making it into a long and narrow chute. The paint color is green to ensure the best imaging effect and meet the granule rolling effect.

[0048] For the image acquisition unit, the technical solution of Embodiment 2 can also be adopted.

[0049] Example 2

[0050] like Figure 3 , 4As shown, unlike Example 1, the polyethylene particle chute (i.e., guide channel 2) in this example uses a transparent material. When the material is transparent, such as glass or acrylic, single-sided frosted glass or single-sided frosted acrylic is chosen. The frosted surface 21 serves as the contact surface of the chute, and the smooth surface 22 serves as the illumination surface of the supplementary light. The main purpose is to allow the polyethylene particles to penetrate through the light. The chute can be configured as multiple parallel guide channels 2. The guide channel 2 can divide the polyethylene particles into multiple sub-streams. On the other hand, the protruding parts on both sides of the guide channel (i.e., guide strips 20) are made into isosceles triangles with a cross-section. The guide strips 20 are made of a material with good light transmittance and are bonded to the frosted surface 21 of the guide channel 2 using a high-performance adhesive. At this time, the frosted surface 21 bonded to the guide strips 20 becomes transparent. When the supplementary light shines on the smooth surface 22 of the guide channel from the bottom of the guide channel 2, refraction and reflection are formed at the inclined surface of the guide strips 20 towards the guide channel. Figure 3 , 4 As shown, refracted light (i.e. Figure 3 The light rays (pointing to the right) irradiate laterally inward onto the polyethylene particles inside the guide channel 2, reflecting the light ( Figure 3 The light rays (pointing to the left) illuminate the polyethylene particles in the adjacent guide channels laterally and outward. Thus, by forming a light path in the area of ​​the guide strip 20, the light passes through the transparent material and is refracted and reflected at the inclined surface. The refracted light then illuminates the polyethylene particles. When multiple guide channels 2 are arranged side-by-side, the reflected light enhances the illumination intensity of adjacent guide channels. Therefore, the design using a transparent material and guide strip 20 effectively improves the lateral illumination conditions of the polyethylene particles.

[0051] Step S102: Target detection is performed on multiple polyethylene particles in the image acquired in step S101 to obtain target detection boxes containing location and size information.

[0052] refer to Figure 7 This invention acquires images of polyethylene particles through image acquisition. First, the target detection unit detects the polyethylene particles. Then, the clustering unit forms an image batch. Next, the instance segmentation unit distinguishes between qualified and abnormal polyethylene particles. Finally, the tracking and prediction processing unit tracks and predicts the multi-target trajectories of the polyethylene particles, and filters out the polyethylene particles identified as "abnormal" for blow-off.

[0053] Specifically, in step S102 (i.e., the target detection step), target detection involves identifying polyethylene particles in the acquired image and obtaining the polyethylene particle target detection box 5 (see reference). Figure 5 The polyethylene particle target detection frame 5 contains the location information and width and height information of the polyethylene particles.

[0054] Step S103: The large image containing multiple polyethylene particle images is divided into several rectangular image blocks by a clustering method, and the rectangular image blocks are associated with the location information of the polyethylene particles.

[0055] refer to Figure 5 , 6 In step S102, the location information is used to segment the large image into several rectangular image blocks 4 using a clustering method. The rectangular image blocks 4 are associated with the location information of the polyethylene particles. Several rectangular image blocks form an image batch, which is then transmitted to the image instance segmentation unit for subsequent recognition processing.

[0056] Furthermore, the preferred clustering method is the fast adaptive clustering method, with the following algorithm flow: For n coordinate points, k iterations of bisector random sampling are performed. In the first iteration, n / 2 coordinate points are selected from the n coordinate points as initial cluster centers. After m iterations, the coordinates of the cluster centers and the cluster metric are calculated. The cluster metric is the sum of the distances D1 from the cluster centers to all points in the cluster. In the second iteration, n / 4 coordinate points are selected from the n coordinate points as initial cluster centers. After another m iterations, the cluster metric D2 is calculated. This process continues until the number of cluster centers with the minimum cluster metric is obtained, which is the number of initial rectangular image patches. This clustering method effectively reduces the computational load while maintaining the required analytical accuracy.

[0057] The final number of cluster centers is n k n k Cluster centers determine n k Four rectangular image patches, each with a scale of w*h, are generated. If the coordinates of any points not yet included are further subdivided into separate rectangular image patches, then n = n. k +s coordinate frames (reference) Figure 6 Each rectangular image block 4 is arranged in a left-to-right, top-to-bottom order to form a slice of the Batch. The absolute coordinates of the slices and the coordinates of the top-left corner of each slice are stored in the same order as the original image size in dictionary form.

[0058] Step S104: The rectangular image block 4 is formed into an image batch for parallel processing. The slide background, qualified polyethylene particles 1 and abnormal polyethylene particles 1A are distinguished by instance segmentation.

[0059] Specifically, refer to Figure 6Rectangular image block 4 contains both the slide background and qualified polyethylene particles and abnormal polyethylene particles (i.e., particles with obvious black spots on the surface and particles with black spots inside). Rectangular image block 4 is fed into the instance segmentation unit in batch form. The instance segmentation unit identifies whether there are black spots on the surface of the polyethylene particles in the rectangular image block; if black spots are present, they are judged as defective products. Instance segmentation obtains the closed contour information and center point coordinates of each polyethylene particle, and the instance segmentation categories are normal polyethylene particles and polyethylene particles with black spots on the surface (i.e., abnormal particles).

[0060] Furthermore, preferably but not limitingly, for normal polyethylene particles, there may be cases where black spots are encased inside the polyethylene particle. In this case, it is necessary to extract the "normal surface particles" and compare them with a pre-established fingerprint database of internally abnormal polyethylene particles. The fingerprint information includes the minimum, average, and maximum values ​​of image pixels within the closed area, and the fingerprint information will be normalized. Through similarity comparison, if the value is close to a threshold, it is determined that there are internal spots, and the "normal surface" polyethylene particle is still judged as a defective product (i.e., an abnormal polyethylene particle). Therefore, the defective products of this invention include two cases: those with black spots on the surface and those with black spots inside; the rest are judged as qualified polyethylene particles.

[0061] Step S105: Track the identified abnormal polyethylene particles as targets and predict their trajectories. Automatically screen out defective products using the location and rolling time information of the abnormal polyethylene particles.

[0062] Specifically, this step includes the following sub-steps: First, the coordinates of the original large-size image are calculated by using the center coordinates of the identified abnormal polyethylene particles and the coordinate information of the corresponding rectangular image blocks; then, in multi-target tracking, the position and time required for the abnormal polyethylene particles to pass through the nozzle are predicted by a trajectory prediction algorithm; finally, based on the position and time required to pass through the nozzle, the nozzle is controlled by an air valve to blow away the abnormal polyethylene particles, thus completing the automatic screening of defective products. This invention controls the nozzle's airflow through an air valve. By transmitting the control signal obtained from the aforementioned analysis (i.e., the position and time information required to pass through the nozzle) to the air valve, the air valve controls the nozzle to spray air, screening and blowing away the abnormal polyethylene particles. The control signal includes the position information of the abnormal polyethylene particles and the time it takes for the particles to roll to the air valve. The position information is interpreted as a specific air valve responsible for the response, and the time information is interpreted as delay information, i.e., the time delay of the control signal.

[0063] This invention acquires images of polyethylene particles through image acquisition, performs target detection of polyethylene particles using a target detection unit, forms image batches using a clustering unit, distinguishes between qualified and abnormal polyethylene particles using an instance segmentation unit, and finally tracks and predicts the multi-target trajectories of polyethylene particles using a tracking and prediction processing unit, filtering out and blowing away the polyethylene particles identified as "abnormal". The entire process is fully automated, requiring no operator intervention, and can visually screen massive amounts of polyethylene particles to select unqualified products. This invention employs a fast adaptive clustering method, which effectively reduces computational load while ensuring the required analytical accuracy. This invention forms image batches from rectangular image blocks for parallel processing, maximizing computational efficiency. This invention can identify abnormal polyethylene particles with black spots on the surface through instance segmentation, and can also identify abnormal polyethylene particles with black spots inside by comparing fingerprint information. The slide in this invention uses a transparent material, and the design of the guide strip allows for supplementary lighting at the bottom of the slide. The reflection enhances the supplementary lighting effect on particles in the guide channel and adjacent guide channels, creating better conditions for subsequent analysis and calculation.

[0064] The foregoing description of specific exemplary embodiments of the present invention is for illustrative and explanatory purposes. These descriptions are not intended to limit the invention to the precise forms disclosed, and it will be apparent that many changes and variations can be made in accordance with the foregoing teachings. The exemplary embodiments were chosen and described in order to explain the specific principles of the invention and its practical application, thereby enabling those skilled in the art to implement and utilize various different exemplary embodiments of the invention, as well as various different choices and variations. Any simple modifications, equivalent changes, and alterations made to the foregoing exemplary embodiments should fall within the scope of protection of the present invention.

Claims

1. A method for identifying abnormal polyethylene particles, characterized in that, The abnormal polyethylene particles include particles with black spots on the surface and particles with black spots inside, and the process includes the following steps: A. Collect image data of multiple polyethylene particles in a rolling state at the corresponding positions of the polyethylene slide; B. Perform target detection on the multiple polyethylene particles to obtain a target detection box containing position and size information; C. Using a clustering method, the large image containing the multiple polyethylene particle images is divided into several rectangular image blocks, and the rectangular image blocks are associated with the location information of the polyethylene particles; The clustering method in step C is a fast adaptive clustering method, specifically including: for n coordinate points, k iterations of split-half random sampling are used to calculate the cluster center coordinates and cluster metric; the number of cluster centers with the minimum cluster metric obtained is the initial number of rectangular image patches; let the number of cluster centers be... indivual, Cluster centers determined For each of the initial rectangular image blocks, new particle coordinate points that were not included are generated individually. s A rectangular image patch is finally obtained. One coordinate frame; D. Form the rectangular image blocks into image batches for parallel processing, and distinguish the slide background, qualified polyethylene particles, and abnormal polyethylene particles through instance segmentation; step D includes: D1 .... The coordinate frames are arranged in a left-to-right, top-to-bottom order to form a slice of a batch. All slices are processed in parallel. D2. The image is segmented in a batch-parallel manner to determine whether there are surface black spots on the polyethylene particles in the rectangular image block. If surface black spots are present, they are identified as abnormal polyethylene particles. D3. The image of polyethylene particles without surface black spots is compared with a preset internal abnormal polyethylene particle fingerprint database. Through similarity comparison, if it is close to the threshold, it is determined that the polyethylene particle has internal black spots, and the polyethylene particles with internal black spots are also identified as abnormal polyethylene particles. E. Track the identified abnormal polyethylene particles as targets and predict their trajectories. Automatically screen out defective products using the location and rolling time information of the abnormal polyethylene particles.

2. The method for identifying abnormal polyethylene particles according to claim 1, characterized in that, The fingerprint information in the fingerprint database includes the minimum, average, and maximum values ​​of image pixels within a closed area. The fingerprint information is information that has undergone normalization processing.

3. The method for identifying abnormal polyethylene particles according to claim 1, characterized in that, Step E includes: E1. The coordinates of the original large-size image are calculated by using the center coordinates of the abnormal polyethylene particles and the coordinate information of the corresponding rectangular image blocks. E2. In multi-target tracking, the position and time required for the abnormal polyethylene particle to pass through the nozzle are predicted by a trajectory prediction algorithm. E3. Based on the required position and time when passing through the nozzle, the nozzle is controlled by an air valve to blow off the abnormal polyethylene particles, thereby completing the automatic screening of defective products.

4. An abnormal polyethylene particle identification device, characterized in that, The method described in any one of claims 1 to 3 includes: An image acquisition unit is used to acquire image data of multiple polyethylene particles in a rolling state at corresponding positions on the polyethylene chute. The target detection unit performs target detection on the plurality of polyethylene particles and obtains a target detection box containing position information and size information. The clustering unit divides a large image containing multiple polyethylene particle images into several rectangular image blocks using a clustering method, and associates the rectangular image blocks with the location information of the polyethylene particles. The instance segmentation unit forms image batches from the rectangular image blocks for parallel processing, and distinguishes the slide background, qualified polyethylene particles and abnormal polyethylene particles through instance segmentation. The tracking and prediction processing unit tracks the identified abnormal polyethylene particles as targets and predicts their trajectories. It then automatically screens out defective products based on the location and rolling time information of the abnormal polyethylene particles.

5. The abnormal polyethylene particle identification device according to claim 4, characterized in that, The image acquisition unit specifically includes: A polyethylene chute, used to carry polyethylene particles to be identified, is a rough-surfaced chute with a certain angle of inclination. A high-speed camera, which is explosion-proof and positioned above the slide, is used to capture polyethylene particles rolling down the slide. A lighting device for supplementing light to the polyethylene particles to be photographed, wherein the light source of the lighting device is white light and the illuminance is at least 60,000 Lux.

6. The abnormal polyethylene particle identification device according to claim 5, characterized in that, The polyethylene slide is made of metal, and the surface of the slide is made into a long and narrow chute and painted green; the lighting equipment and the high-speed camera are integrated.

7. The abnormal polyethylene particle identification device according to claim 5, characterized in that, The polyethylene slide is made of transparent material, with one side being frosted and the other side being smooth. The frosted side serves as the contact surface of the slide, and the smooth side serves as the supplementary lighting surface.

8. The abnormal polyethylene particle identification device according to claim 7, characterized in that, The polyethylene slide is a structure of multiple parallel guide channels. Each guide channel has a guide strip with an isosceles right triangle cross-section on both sides. The guide strip is made of a light-transmitting material and is bonded to the frosted surface of the slide. When the supplementary light shines on the smooth surface of the slide, refraction and reflection occur at the inclined surface of the guide strip towards the guide channel. The refracted light shines laterally inward into the polyethylene particles in the guide channel, and the reflected light shines laterally outward into the polyethylene particles in the adjacent guide channel.

9. The abnormal polyethylene particle identification device according to claim 7, characterized in that, The transparent material of the polyethylene slide is glass or acrylic.

Images