Gravity separation and AI recognition combined broken plastic separation device and method

Abstract

The invention discloses a broken plastic sorting device and method combining gravity sorting and AI recognition, the broken plastic sorting device comprises a gravity sorting system and an AI intelligent sorting system which are connected in sequence, and the gravity sorting system is used for sorting out part of relatively light plastic in broken plastic and conveying sorted relatively heavy intermediate materials to the AI intelligent sorting system; and the AI intelligent sorting is used for sorting target electric wires and fine copper wires through color sorting and visual identification. Gravity separation and intelligent separation are combined together, about 80%-90% or above of light plastic can be removed through gravity separation, only less than 10%-20% of materials need to be subjected to subsequent separation, and very high economic value is achieved.

Description

Technical Field

[0001] The present invention relates to the technical field of waste household appliance recycling, and in particular to a broken plastic sorting device and method combining gravity sorting and AI recognition. Background Art

[0002] The crushing and sorting of waste refrigerators for resource utilization has been implemented in my country and around the world for many years, and most plastics, metals, and hazardous wastes are recycled in different categories. Studies have found that among the plastics recycled from refrigerators at this stage, some thread ends are still in the plastics and have not been fully sorted out. The reason is that the refrigerator processing line has a large processing capacity, the materials are superimposed, and the eddy current sorting cannot sort out the thread ends and they are mixed in the plastic. Although the plastic conveyor belt is equipped with personnel to sort the thread ends, it is impossible to sort out all the thread ends due to the large output and thick material layer. There are still a considerable number of thread ends and thin copper wires in the plastic. This thread end and thin copper wire account for 1.5%-5% of the plastic. If the thread ends are not sorted out, it will not only waste resources, but also affect the recycling of plastics.

[0003] At present, magnetic separation is usually used to select iron-containing materials for the recycling and sorting of scrapped refrigerators; eddy current sorting is used to separate non-ferrous metals (copper, aluminum and zinc alloys) and plastics, but it has limited separation effect on metals with small volume and light weight, such as wires and thin copper wires or small-sized aluminum sheets, because eddy current sorting relies on the volume and density of the metal to generate sufficient separation force; wires and thin copper wires are currently sorted by manpower to separate thin copper wires, wires, etc. There are also methods of sorting thin copper wires, wires, aluminum sheets, etc. by gravity, but when the specific gravity of the waste materials varies greatly, the effect of gravity sorting cannot meet the requirements and manual sorting is still required. Even if the sorting is combined with image recognition and robot arm gripping, there are still certain disadvantages. For example: 1. In the screening process of some scrap refrigerators, the proportion of target objects that need to be screened out is small, such as wires in refrigerator screening. If the existing visual recognition algorithm is simply used in combination with robot arm gripping for sorting, it will result in low efficiency and high energy consumption. 2. Image recognition is easily affected by environmental interference or material aging, which leads to unstable color parameters in the extracted feature information, thereby reducing recognition accuracy. 3. Target recognition is usually performed on the entire image, and the number of comparison and analysis of the extracted feature information of each target object is fixed, which will greatly waste computing power and prolong the recognition time, resulting in low efficiency. Therefore, at this stage, a more efficient and intelligent material sorting management technology solution is needed to solve the above problems and realize fully automatic sorting. Summary of the invention

[0004] In the existing refrigerator recycling process, refrigerator plastics are often mixed with wires and thin copper wires. These fine metals are difficult to completely separate through traditional methods, resulting in a decrease in the quality of subsequent plastic recycling and affecting the reuse rate. Existing sorting methods such as magnetic separation are only effective for ferromagnetic metals. Gravity sorting has a low separation efficiency for thinner copper wires and plastic mixtures, and is difficult to automate. Therefore, there is an urgent need for a sorting method that combines gravity sorting and intelligent AI color sorting to improve the separation efficiency of wires and thin copper wires in refrigerator crushed plastics.

[0005] The sorting method of the present invention is achieved by a combination of gravity and AI intelligent color sorting, including enriching heavy plastics, wire ends and thin copper wires through gravity vibration, bouncing and centrifugal sorting and sorting out pure plastic sheets. Then, the enriched heavy plastics, wires and thin copper wires are separated by a double-layer AI intelligent color sorter to achieve full automatic separation.

[0006] The purpose of the present invention is to provide a crushed plastic sorting device and method combining gravity sorting and AI recognition to solve the problems raised in the above background technology.

[0007] In order to solve the above technical problems, the present invention provides a crushed plastic sorting device combining gravity sorting and AI recognition: comprising a gravity sorting system and an AI intelligent sorting system 6 connected in sequence, wherein the gravity sorting system is used to sort out some relatively light plastics in the crushed plastics, and transport the relatively heavy intermediate materials after sorting to the AI ​​intelligent sorting system; the AI ​​intelligent sorting sorts out the target wires through visual recognition.

[0008] Preferably, the aforementioned device for separating and crushing wires in plastic by gravity sorting and AI intelligent sorting, wherein the gravity sorting system comprises a bouncing gravity sorting device (2) and a centrifugal gravity sorting device (3); the bouncing gravity sorting device (2) comprises a bouncing gravity sorting channel (30), a bouncing gravity sorting inlet (20) arranged at the upper end of the bouncing gravity sorting channel (30), and a bouncing gravity sorting heavy material outlet (40) arranged at the lower end of the bouncing gravity sorting channel (30); the centrifugal gravity sorting device (3) comprises a centrifugal gravity sorting fan (80) and a centrifugal gravity sorting cavity (60); the lower end of the bouncing gravity sorting channel (30) is connected to the centrifugal gravity sorting fan (80) through a bouncing gravity sorting air inlet (90); and the lower end of the centrifugal gravity sorting cavity (60) is also provided with a centrifugal gravity sorting heavy material outlet (70).

[0009] The present invention also discloses a broken plastic sorting method combining gravity sorting and AI recognition, wherein the AI ​​intelligent sorting comprises the following steps:

[0010] S100, collecting a material comparison table and a material attribute table, as well as batch information of raw materials after gravity sorting, and collecting image information of the raw materials through a high-definition camera in the sorting area.

[0011] S200, analyzing the raw material information and establishing a color comparison library, dividing the recognition area on the raw material image through a machine vision algorithm, establishing a rule library for each recognition area in combination with the color comparison library and calculating a priority coefficient.

[0012] S300, defining a detection order according to the priority coefficients of the identification zones, using a target recognition algorithm to perform detection and dynamically adjusting a rule base and a priority coefficient of each identification zone until the proportion of the detected raw materials reaches a preset value.

[0013] S400, the operation interface displays the image information and detection marks of the raw materials in real time, feeds back the detection results of each time to the staff in real time, and automatically updates the material attribute table.

[0014] In S100, the raw material refers to the mixture obtained after the scrapped refrigerator is disassembled and initially gravity-sorted, and the material refers to the electric wires and thin copper wires mixed in the raw materials. The material comparison table includes sorting records of scrapped refrigerators of different models, and each model of scrapped refrigerator corresponds to a sorting record, including an equipment code and a sorting material set. The sorting material set includes parameter records of different types of materials, and each material corresponds to a parameter record, including a material identifier and length. The material attribute table includes attribute information of different types of materials, including a material identifier, an RGB color code set, a mass per unit length, and a wire diameter. Batch information refers to the equipment code and quantity corresponding to all models of scrapped refrigerators from which this batch of raw materials comes. The sorting area refers to the operating area used to sort raw materials.

[0015] Gravity sorting is a process that uses gravity differences to separate materials of different densities. This sorting technology is widely used in the field of waste treatment. The basic principle is to remove most of the lightweight plastics in the raw materials through gravity sorting to solve the capacity problem of subsequent sorting and improve the accuracy of sorting. Gravity sorting can remove about 80%-90% of lightweight plastics, and only less than 10%-20% of the materials need to be sorted later. This difference can be used to separate the plastic pieces from the scrapped refrigerators and materials such as metal wire.

[0016] Each model of scrapped refrigerator includes multiple sortable materials, and each material includes multiple color codes. The equipment code is used to distinguish different models of scrapped refrigerators, and the material identifier is used to distinguish different types of materials. The batch information contains at least one or more models of scrapped refrigerators, and the number of scrapped refrigerators of the same model may be one or more.

[0017] In S200, the specific steps are as follows:

[0018] S201. Obtain all equipment codes in the batch information P of the current raw material, retrieve all material identifiers in the sorting material set corresponding to these equipment codes in the material comparison table and mark them. Establish a color comparison library, and put all elements in the RGB color code set corresponding to the marked material identifier in the material attribute table into the color comparison library.

[0019] The material comparison table is used to display the length information of various materials that can be sorted out from the raw materials after the scrapped refrigerators of different models are crushed. The material attribute table is used to display the RGB color code set, unit length mass and wire diameter of each material.

[0020] S202, setting the number of grids n, evenly dividing the raw material image of the current batch into n grid areas of the same size, extracting the pixel value of each grid area and calculating the RGB color code, and using the Euclidean distance algorithm to calculate the deviation value between the RGB color code of each grid area and each RGB color code in the color comparison library.

[0021] S203, set a deviation threshold U, and use the RGB color codes in the color comparison library with deviation values ​​less than U as similar color codes for the corresponding grid area. Search the material attribute table for corresponding material identifiers containing similar color codes in the RGB color code set, and use these material identifiers as similar material identifiers for the corresponding grid area.

[0022] A similar color code refers to an RGB color code that exists in the color comparison library and has a deviation value less than U from the RGB color code of the grid area. A similar material identifier refers to the material identifier corresponding to each RGB color code set containing the similar color code. Each grid area has multiple similar color codes and multiple similar material identifiers.

[0023] S204, there is at least one grid area with a deviation value less than U, and the adjacent marked grid areas in the raw material image together constitute the identification area. A rule base is established for each identification area, and similar material identifiers of all grid areas in the identification area are obtained. The line diameter and mass information per unit length of each similar material identifier are used as judgment rules, and all judgment rules of the grid area are placed in the rule base of the corresponding identification area.

[0024] There is only one determination rule for the same or similar material identifier in the same rule base. There are multiple determination rules for the same or similar material identifier in different rule bases.

[0025] S205. Count the number of all grid areas j in the identification area and the minimum deviation value of each grid. Calculate the deviation value of the RGB color code between the grids in the identification area, and classify the grids with a deviation value less than U into the same category. Count the number of grids in different categories in each identification area and calculate the standard deviation BZ, and substitute it into the formula to calculate the priority coefficient of each identification area:

[0026]

[0027] In the formula, YX is the priority coefficient, j ave is the average number of grids in all recognition areas, α is a constant, CT is the number of judgment rules in the rule base of the recognition area, is the minimum deviation value of the i-th grid in the identification area.

[0028] The minimum deviation value refers to the minimum value among the deviation values ​​of all RGB color codes in the corresponding grid and the color comparison library. The smaller the deviation value, the more similar the colors of the two are.

[0029] In S300, the specific steps are as follows:

[0030] S301. Obtain the number h of each device code in the batch information P and the length of the material identifier in the material comparison table. Multiply the length of each material identifier by the number h to obtain the estimated length, and add the estimated length to the judgment rule of the corresponding similar material identifier. Arrange the identification areas in descending order according to the priority coefficient, and obtain the corresponding rule base for detection according to the arrangement order of the identification areas.

[0031] S302. Use the target recognition algorithm to detect all target objects in the recognition area, collect the wire diameter and length Q of each target object, arrange the target objects in descending order of length, and perform an initial comparison with each judgment rule in the rule library in the order of arrangement of the target objects to determine whether there is a judgment rule that the wire diameter is equal and the length Q is less than the expected length. If so, control the robot arm to clamp the target object. If not, continue to compare other target objects in the recognition area.

[0032] S303, after the robot arm grasps the target object, it collects its mass R, divides the mass R by the length Q to obtain the mass per unit length T, and compares it again in the corresponding rule library to determine whether there is a judgment rule that the mass per unit length is equal to T. If so, the target object is marked and transferred to the material accumulation area, and the estimated length is subtracted from Q to obtain a new estimated length and the judgment rule is updated synchronously; if not, the robot arm is controlled to put down the target object and continue to compare and identify other target objects in the area.

[0033] S304. When the estimated length of the marked material identifier F is zero, delete all the judgment rules corresponding to F in the rule base and recalculate the priority coefficient of each identification area. Set the preset value B. When all identification areas have completed the detection, calculate the ratio Y of the sum of the masses of all marked target objects and the sum of the masses of the materials corresponding to the marked material identifier. Determine whether the ratio Y is greater than the preset value B. If the result is yes, enter the detection of the next batch of raw materials. If the result is no, increase the deviation threshold U and detect again until the ratio Y is greater than the preset value B.

[0034] The sum of the masses of all marked target objects refers to the cumulative sum of the masses of all materials sorted into the material accumulation area. The sum of the masses of materials corresponding to the marked material identifiers refers to the theoretical mass obtained by multiplying the number of scrapped refrigerators of various models in the batch information P and the mass of all materials that can be sorted out of each model of scrapped refrigerators. By calculating the ratio between the two, the sorting and recycling efficiency can be better determined.

[0035] In S400, the image information of the current batch of raw materials and the specific location of each identification area, as well as the judgment result after each robot arm grasps the target, are displayed through the operation interface, and real-time feedback is given to the staff, and the RGB color code of the marked target object is automatically added to the RGB color code set of the corresponding material in the material attribute table.

[0036] The broken plastic sorting system combines gravity sorting with AI recognition. The system includes a data acquisition module, a sorting planning module, a dynamic adjustment module and a visualization module.

[0037] The data acquisition module is used to collect material comparison tables and material attribute tables, as well as batch information of raw materials after gravity sorting, and collect image information of raw materials through cameras; the sorting planning module establishes a color comparison library based on the raw material information, divides the recognition area on the raw material image through machine vision algorithms, and establishes a rule library for each recognition area in combination with the color comparison library and calculates the priority coefficient. The dynamic adjustment module defines the detection order according to the priority coefficient of the recognition area, uses the target recognition algorithm for detection, and dynamically adjusts the rule library and priority coefficient of each recognition area until the proportion of detected raw materials reaches the preset value. The visualization module operation interface is used to display the image information and detection marks of the raw materials and feedback to the staff.

[0038] The data acquisition module includes an image information acquisition unit, a sorting information acquisition unit and a material table acquisition unit.

[0039] The image information acquisition unit acquires image information of the raw materials through a high-definition camera in the sorting area. The sorting area refers to an operation area for sorting raw materials.

[0040] The sorting information collection unit is used to collect the batch information of the raw materials obtained after gravity sorting. The batch information refers to the equipment codes and quantities corresponding to all models of scrapped refrigerators from which the raw materials in this batch come.

[0041] Heavy and dense thin copper wires, wire ends and other metal parts are separated from plastics by gravity sorting.

[0042] The material table collection unit is used to collect material comparison tables and material attribute tables. The material comparison table includes sorting records of scrapped refrigerators of different models, including equipment codes and sorting material sets; the sorting material sets include parameter records of different types of materials, including material identifiers and lengths; the material attribute table includes attribute information of different types of materials, including material identifiers, RGB color code sets, mass per unit length, and wire diameter.

[0043] The sorting planning module includes a regional planning unit and a coefficient calculation unit.

[0044] The regional planning unit is used to establish a color comparison library and divide the recognition area.

[0045] First, mark the material identifier corresponding to the equipment code in the batch information in the material comparison table, establish a color comparison library, and put all RGB color codes corresponding to the material identifier marked in the material attribute table into the color comparison library.

[0046] Secondly, the number of grids n is set, and n grid areas of the same size are evenly divided on the raw image. The RGB color code of each grid area is calculated, and the Euclidean distance algorithm is used to calculate the deviation value between the RGB color code of each grid area and each RGB color code in the color comparison library.

[0047] Finally, the deviation threshold U is set, and the RGB color code with a deviation value less than U in the color comparison library is used as the similar color code of the corresponding grid area. The corresponding material identifiers containing similar color codes in the RGB color code set are retrieved in the material attribute table, and these material identifiers are used as the similar material identifiers of the corresponding grid area. There is at least one grid area with a deviation value less than U in the mark, and the adjacent marked grid areas in the raw material image together constitute the recognition area.

[0048] The coefficient calculation unit is used to establish a rule base for each identification area and calculate a priority coefficient.

[0049] First, a rule base is established in each identification area to obtain similar material identifiers of all grid areas in the identification area. The wire diameter of the similar material identifier of each grid area in the identification area and the mass information per unit length are used as judgment rules. All judgment rules of the grid area are placed in the rule base of the identification area to which it belongs.

[0050] Secondly, count the number of grid areas j and the minimum deviation value of each grid in each recognition area. Calculate the deviation value of the RGB color code between each grid in the recognition area, and classify the grids with deviation values ​​less than U into the same category. Count the number of grids in different categories in each recognition area and calculate the standard deviation BZ.

[0051] Finally, the average number of grids j in all recognition areas is calculated ave , substitute into the formula: In , the priority coefficient of each identification area is calculated respectively. Among them, α is a constant, CT is the number of judgment rules in the rule base of the identification area, is the minimum deviation value of the i-th grid in the identification area.

[0052] An AI intelligent color sorter is used to optically sort the broken materials remaining after gravity sorting. Through near-infrared imaging and visible light imaging technology combined with AI algorithms of photo graphics, plastics and thin copper wires and electrical wire ends can be distinguished.

[0053] The dynamic adjustment module includes a machine sorting unit and an adjustment optimization unit.

[0054] The machine sorting unit is used to define the inspection sequence and control the robot arm sorting.

[0055] First, obtain the number h of each device code in the batch information P and the length of the marked material identifier. Multiply the length of each marked material identifier by the number h to obtain the expected length, and add the expected length to the judgment rule of the corresponding similar material identifier.

[0056] Secondly, the recognition areas are arranged in descending order according to the priority coefficients, and the corresponding rule base is obtained in sequence for detection according to the arrangement order of the recognition areas. The target recognition algorithm is used to detect all target objects in the recognition area, and the target objects are arranged in descending order according to the length Q, and the robot arm is controlled to grasp them in sequence according to the arrangement order of the target objects.

[0057] Finally, the arrangement order of the target objects is compared with the judgment rules in the rule base to determine whether there is a judgment rule that the wire diameter is equal and the length Q is less than the expected length. If so, the robot arm is controlled to clamp the target object. If not, other target objects in the recognition area are compared.

[0058] The adjustment and optimization unit is used to dynamically adjust the rule base and priority coefficient of each identification area.

[0059] After the robot grabs the target object, collect its mass R, divide the mass R by the length Q to get the mass per unit length T and compare it again in the corresponding rule library to determine whether there is a judgment rule that the mass per unit length is equal to T. If so, mark the target object and transfer it to the material accumulation area, subtract Q from the expected length to get a new expected length and update the judgment rule synchronously. If not, control the robot to put down the target object and continue to compare other target objects in the identification area. When the expected length of the marked material identifier F is zero, delete the judgment rules corresponding to F in all rule libraries and recalculate the priority coefficients of each identification area. Set the preset value B. When all identification areas have completed the detection, calculate the ratio Y of the sum of the masses of all marked target objects to the sum of the masses of the materials corresponding to the marked material identifier. Determine whether the ratio Y is greater than the preset value B. If the result is yes, enter the detection of the next batch of raw materials; if the result is not, increase the deviation threshold U and detect again until the ratio Y is greater than the preset value B.

[0060] The visualization module displays the image information of the raw materials and the specific location of each identification area through the operation interface and provides real-time feedback to the staff, obtains the RGB color code of the marked target object, and automatically updates the RGB color code set of the corresponding material in the material attribute table.

[0061] The AI ​​color sorter uses multispectral imaging technology, with a spectral range of visible light (400-700nm) and near infrared (700-1000nm) to accurately identify the difference in reflection between plastic and copper wire. Photographic image extraction technology extracts the shapes of thin copper wires and wires of different shapes for machine learning to form a picture library. The resolution of identifying plastic, copper wire and wire is continuously improved by comparing the similarity of the images in the picture library with the image extraction technology.

[0062] Compared with the prior art, the beneficial effects achieved by the present invention are:

[0063] 1. The present invention combines gravity sorting and intelligent sorting together. About 80%-90% of lightweight plastics can be removed by gravity sorting, and only less than 10%-20% of the materials need to undergo subsequent sorting, which has very high economic value.

[0064] 2. Efficient target recognition: This application extracts the color features of each grid area in the raw image and divides the identification area, and establishes a rule base for each identification area. By analyzing the number of grids in the identification area, the number of judgment rules in the rule base, and the minimum deviation value of each grid to calculate the priority coefficient, the number of judgment rules in the rule base of each identification area is reduced as soon as possible, thereby improving the detection efficiency. Compared with the traditional method of target recognition for global images, it is more efficient.

[0065] 3. Dynamic sorting and detection: In this application, the target objects are first sorted by image sorting to identify the target objects that meet the length and wire diameter requirements at the same time, and then the mass sorting is used in the process of the robot arm gripping to identify the target objects whose mass meets the quality requirements per unit length. The double comparison sorting method can improve the modifiability and accuracy of the results, which is more intelligent and accurate than the prior art in which the robot arm only acts as an execution unit and the results cannot be modified.

[0066] This solution combines gravity sorting with AI image sorting to meet production capacity needs and sort out thread ends. If only AI image sorting is used, the goal can be achieved, but the equipment requires a large space, and the existing production line cannot reserve enough space to place large equipment. First, most of the lighter plastics are removed through gravity sorting. Only the remaining thread ends and a small part of the impurities less than 10% of the total material need to be sorted by AI. This can greatly improve the sorting accuracy and reduce the production capacity requirements for AI sorting.

[0067] In summary, compared with traditional technologies, the present invention has the advantages of efficient target recognition and dynamic sorting detection, and can improve the efficiency of material sorting management. BRIEF DESCRIPTION OF THE DRAWINGS

[0068] The accompanying drawings are used to provide a further understanding of the present invention and constitute a part of the specification. Together with the embodiments of the present invention, they are used to explain the present invention and do not constitute a limitation of the present invention. In the accompanying drawings:

[0069] Figure 1 It is a three-dimensional structural schematic diagram of the sorting device combining gravity sorting with AI intelligent color sorter of the present invention.

[0070] Figure 2 It is a schematic diagram of the cross-sectional structure of the AI ​​intelligent color sorter of the present invention.

[0071] Figure 3 It is a schematic diagram of the cross-sectional structure of the gravity separation device of the present invention.

[0072] Figure 4 It is a schematic diagram of the overall process operation of the present invention.

[0073] Figure 5 It is a schematic flow chart of the crushed plastic sorting method combining gravity sorting and AI recognition of the present invention.

[0074] Figure 6 It is a structural schematic diagram of the crushed plastic sorting system of the present invention which combines gravity sorting and AI recognition.

[0075] In the figure: 1- elevator; 2- bouncing gravity sorting device; 3- centrifugal gravity sorting device; 4- scraper elevator; 5- intermediate material return elevator; 6- AI intelligent color sorter; 10- incoming material elevator outlet; 20- bouncing gravity sorting entrance; 30- bouncing gravity sorting; 40- bouncing gravity sorting heavy material outlet; 50- bouncing gravity sorting light material outlet; 60- centrifugal gravity sorting; 70- centrifugal gravity sorting heavy material outlet; 80- centrifugal gravity sorting fan; 90- bouncing gravity sorting air inlet; 100- material to be selected; 200- light source; 300- upper multi-spectral camera; 400- lower multi-spectral camera; 500- high-speed nozzle group; 600- plastic outlet; 700- outlet for wire ends and thin copper wires. DETAILED DESCRIPTION

[0076] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0077] like Figures 1 to 3 As shown: The present invention discloses a crushed plastic sorting device combining gravity sorting and AI recognition, including a gravity sorting system and an AI intelligent sorting system connected in sequence. The gravity sorting system is used to sort out some relatively light plastics in the crushed plastics, and convey the relatively heavy intermediate materials after sorting to the AI ​​intelligent sorting system; the AI ​​intelligent sorting sorts out the target wires and thin copper wires through visual recognition. Among them, the gravity sorting system includes an elevator 1, a bouncing gravity sorting device 2, and a centrifugal gravity sorting device 3 connected in sequence; the centrifugal gravity sorting device 3 conveys the intermediate materials to the AI ​​intelligent sorting system 6 through the intermediate material return elevator 5.

[0078] Among them, the gravity sorting system includes a bouncing gravity sorting device 2 and a centrifugal gravity sorting device 3; the bouncing gravity sorting device 2 includes a bouncing gravity sorting channel 30, a bouncing gravity sorting inlet 20 arranged at the upper end of the bouncing gravity sorting channel 30, and a bouncing gravity sorting heavy material outlet 40 arranged at the lower end of the bouncing gravity sorting channel 30; the centrifugal gravity sorting device 3 includes a centrifugal gravity sorting fan 80 and a centrifugal gravity sorting cavity 60, and the lower end of the bouncing gravity sorting channel 30 is connected to the centrifugal gravity sorting fan 80 through the bouncing gravity sorting air inlet 90; the lower end of the centrifugal gravity sorting cavity 60 is also provided with a centrifugal gravity sorting heavy material outlet 70.

[0079] Specifically, the mixed materials such as plastics, wire ends, and thin copper wires coming out of the non-metallic discharge port of the refrigerator eddy current sorting are lifted by the elevator to the gravity sorting unit, and the outlet 10 of the incoming material elevator is used to pour the crushed materials into the bouncing gravity sorting inlet 20. After the sorted materials are sorted by the bouncing gravity sorting device 30 in turn, a part of the relatively heavy intermediate materials (including wire ends, thin copper wires and high-density plastic blocks) enter the AI ​​intelligent sorting system 6 through the bouncing gravity sorting heavy material outlet 40; furthermore, the relatively light plastics (mostly plastic sheets) enter the bouncing gravity sorting light material outlet 50, and then through. The fan 80 of the centrifugal gravity sorting provides the wind / airflow for gravity sorting to the bouncing gravity sorting device 30 through the air inlet 90 of the bouncing gravity sorting. The light material outlet 50 of the bouncing gravity sorting is connected to the air inlet end of the fan 80 of the centrifugal gravity sorting, so that the relatively light plastic sorted out will enter the inside of the fan 80 of the centrifugal gravity sorting for centrifugal sorting. After the plastic sheets are centrifugally sorted by the centrifugal gravity sorting cavity 60, the lighter dust, stickers, etc. inside will enter the air inlet 90 of the bouncing gravity sorting again, and the plastic sheets will be discharged through the heavy material outlet 70 of the centrifugal gravity sorting.

[0080] See also Figure 4 and Figure 5 The crushed plastic sorting method combining gravity sorting and AI recognition provided by the present invention comprises the following steps:

[0081] S100, collecting a material comparison table and a material attribute table, as well as batch information of raw materials after gravity sorting, and collecting image information of the raw materials through a high-definition camera in the sorting area.

[0082] In practical applications, the gravity sorting in this step is usually wind flotation. In the field of refrigerator sorting, wind flotation can usually screen out about 80%-90% of the waste, so that the material to be sorted that flows into the intelligent sorting link of machine vision is only 10%-20% of the raw material, which can greatly save the energy consumption required for sorting and improve sorting efficiency.

[0083] S200, analyzing the raw material information and establishing a color comparison library, dividing the recognition area on the raw material image through a machine vision algorithm, establishing a rule library for each recognition area in combination with the color comparison library and calculating a priority coefficient.

[0084] S300, defining a detection order according to the priority coefficients of the identification zones, using a target recognition algorithm to perform detection and dynamically adjusting a rule base and a priority coefficient of each identification zone until the proportion of the detected raw materials reaches a preset value.

[0085] S400, the operation interface displays the image information and detection marks of the raw materials in real time, feeds back the detection results of each time to the staff in real time, and automatically updates the material attribute table.

[0086] In S100, the raw material refers to the mixture obtained after the scrapped refrigerator is disassembled and initially gravity-sorted, and the material refers to the electric wires and thin copper wires mixed in the raw materials. The material comparison table includes sorting records of scrapped refrigerators of different models, and each model of scrapped refrigerator corresponds to a sorting record, including an equipment code and a sorting material set. The sorting material set includes parameter records of different types of materials, and each material corresponds to a parameter record, including a material identifier and length. The material attribute table includes attribute information of different types of materials, including a material identifier, an RGB color code set, a mass per unit length, and a wire diameter. Batch information refers to the equipment code and quantity corresponding to all models of scrapped refrigerators from which this batch of raw materials comes. The sorting area refers to the operating area used to sort raw materials.

[0087] Gravity sorting is a process that uses gravity differences to separate materials of different densities. This sorting technology is widely used in the field of waste treatment. The basic principle is to remove most of the lightweight plastics in the raw materials through gravity sorting to solve the capacity problem of subsequent sorting and improve the accuracy of sorting. Gravity sorting can remove about 80%-90% of lightweight plastics, and only less than 10%-20% of the materials need to be sorted later. This difference can be used to separate the plastic pieces from the scrapped refrigerators and materials such as metal wire.

[0088] There are many methods of gravity sorting, including brine sorting, airflow sorting and vibrating screen sorting. Among them, brine sorting equipment occupies a large area and has sewage treatment problems, and vibrating screen sorting equipment has a small processing capacity. Therefore, in this scheme, the bouncing airflow sorting equipment in the airflow sorting is preferred. After the bouncing sorting, the material collides and disperses with the bouncing structure inside, and the sorting effect of light and heavy materials is better.

[0089] Each model of scrapped refrigerator includes multiple sortable materials, and each material includes multiple color codes. The equipment code is used to distinguish different models of scrapped refrigerators, and the material identifier is used to distinguish different types of materials. The batch information contains at least one or more models of scrapped refrigerators, and the number of scrapped refrigerators of the same model may be one or more.

[0090] In S200, the specific steps are as follows:

[0091] S201. Obtain all equipment codes in the batch information P of the current raw material, retrieve all material identifiers in the sorting material set corresponding to these equipment codes in the material comparison table and mark them. Establish a color comparison library, and put all elements in the RGB color code set corresponding to the marked material identifier in the material attribute table into the color comparison library.

[0092] The material comparison table is used to display the length information of various materials that can be sorted out from the raw materials after the scrapped refrigerators of different models are crushed. The material attribute table is used to display the RGB color code set, unit length mass and wire diameter of each material.

[0093] S202, setting the number of grids n, evenly dividing the raw material image of the current batch into n grid areas of the same size, extracting the pixel value of each grid area and calculating the RGB color code, and using the Euclidean distance algorithm to calculate the deviation value between the RGB color code of each grid area and each RGB color code in the color comparison library.

[0094] S203, set a deviation threshold U, and use the RGB color codes in the color comparison library with deviation values ​​less than U as similar color codes for the corresponding grid area. Search the material attribute table for corresponding material identifiers containing similar color codes in the RGB color code set, and use these material identifiers as similar material identifiers for the corresponding grid area.

[0095] A similar color code refers to an RGB color code that exists in the color comparison library and has a deviation value less than U from the RGB color code of the grid area. A similar material identifier refers to the material identifier corresponding to each RGB color code set containing the similar color code. Each grid area has multiple similar color codes and multiple similar material identifiers.

[0096] S204, there is at least one grid area with a deviation value less than U, and the adjacent marked grid areas in the raw material image together constitute the identification area. A rule base is established for each identification area, and similar material identifiers of all grid areas in the identification area are obtained. The line diameter and mass information per unit length of each similar material identifier are used as judgment rules, and all judgment rules of the grid area are placed in the rule base of the corresponding identification area.

[0097] There is only one determination rule for the same or similar material identifier in the same rule base. There are multiple determination rules for the same or similar material identifier in different rule bases.

[0098] S205. Count the number of all grid areas j in the identification area and the minimum deviation value of each grid. Calculate the deviation value of the RGB color code between the grids in the identification area, and classify the grids with a deviation value less than U into the same category. Count the number of grids in different categories in each identification area and calculate the standard deviation BZ, and substitute it into the formula to calculate the priority coefficient of each identification area:

[0099]

[0100] In the formula, YX is the priority coefficient, j ave is the average number of grids in all recognition areas, α is a constant, CT is the number of judgment rules in the rule base of the recognition area, is the minimum deviation value of the i-th grid in the identification area.

[0101] The minimum deviation value refers to the minimum value among the deviation values ​​of all RGB color codes in the corresponding grid and the color comparison library. The smaller the deviation value, the more similar the colors of the two are.

[0102] In S300, the specific steps are as follows:

[0103] S301. Obtain the number h of each device code in the batch information P and the length of the material identifier in the material comparison table. Multiply the length of each material identifier by the number h to obtain the estimated length, and add the estimated length to the judgment rule of the corresponding similar material identifier. Arrange the identification areas in descending order according to the priority coefficient, and obtain the corresponding rule base for detection according to the arrangement order of the identification areas.

[0104] S302. Use the target recognition algorithm to detect all target objects in the recognition area, collect the wire diameter and length Q of each target object, arrange the target objects in descending order of length, and perform an initial comparison with each judgment rule in the rule library in the order of arrangement of the target objects to determine whether there is a judgment rule that the wire diameter is equal and the length Q is less than the expected length. If so, control the robot arm to clamp the target object. If not, continue to compare other target objects in the recognition area.

[0105] S303, after the robot arm grasps the target object, it collects its mass R, divides the mass R by the length Q to obtain the mass per unit length T, and compares it again in the corresponding rule library to determine whether there is a judgment rule that the mass per unit length is equal to T. If so, the target object is marked and transferred to the material accumulation area, and the estimated length is subtracted from Q to obtain a new estimated length and the judgment rule is updated synchronously; if not, the robot arm is controlled to put down the target object and continue to compare and identify other target objects in the area.

[0106] S304. When the estimated length of the marked material identifier F is zero, delete all the judgment rules corresponding to F in the rule base and recalculate the priority coefficient of each identification area. Set the preset value B. When all identification areas have completed the detection, calculate the ratio Y of the sum of the masses of all marked target objects and the sum of the masses of the materials corresponding to the marked material identifier. Determine whether the ratio Y is greater than the preset value B. If the result is yes, enter the detection of the next batch of raw materials. If the result is no, increase the deviation threshold U and detect again until the ratio Y is greater than the preset value B.

[0107] The sum of the masses of all marked target objects refers to the cumulative sum of the masses of all materials sorted into the material accumulation area. The sum of the masses of materials corresponding to the marked material identifiers refers to the theoretical mass obtained by multiplying the number of scrapped refrigerators of various models in the batch information P and the mass of all materials that can be sorted out of each model of scrapped refrigerators. By calculating the ratio between the two, the sorting and recycling efficiency can be better determined.

[0108] In S400, the image information of the current batch of raw materials and the specific location of each identification area, as well as the judgment result after each robot arm grasps the target, are displayed through the operation interface, and real-time feedback is given to the staff, and the RGB color code of the marked target object is automatically added to the RGB color code set of the corresponding material in the material attribute table.

[0109] See also Figure 6 The present invention provides a broken plastic sorting system combining gravity sorting and AI recognition, the system comprising a data acquisition module, a sorting planning module, a dynamic adjustment module and a visualization module.

[0110] The data acquisition module is used to collect material comparison tables and material attribute tables, as well as batch information of raw materials after gravity sorting, and collect image information of raw materials through cameras; the sorting planning module establishes a color comparison library based on the raw material information, divides the recognition area on the raw material image through machine vision algorithms, and establishes a rule library for each recognition area in combination with the color comparison library and calculates the priority coefficient. The dynamic adjustment module defines the detection order according to the priority coefficient of the recognition area, uses the target recognition algorithm for detection, and dynamically adjusts the rule library and priority coefficient of each recognition area until the proportion of detected raw materials reaches the preset value. The visualization module operation interface is used to display the image information and detection marks of the raw materials and feedback to the staff.

[0111] The data acquisition module includes an image information acquisition unit, a sorting information acquisition unit and a material table acquisition unit.

[0112] The image information acquisition unit acquires image information of the raw materials through a high-definition camera in the sorting area. The sorting area refers to an operation area for sorting raw materials.

[0113] The sorting information collection unit is used to collect the batch information of the raw materials obtained after gravity sorting. The batch information refers to the equipment codes and quantities corresponding to all models of scrapped refrigerators from which the raw materials in this batch come.

[0114] Gravity sorting is used to separate heavier, denser metal parts such as thin copper wires and wire ends from lighter plastics.

[0115] The material table collection unit is used to collect material comparison tables and material attribute tables. The material comparison table includes sorting records of scrapped refrigerators of different models, including equipment codes and sorting material sets; the sorting material sets include parameter records of different types of materials, including material identifiers and lengths; the material attribute table includes attribute information of different types of materials, including material identifiers, RGB color code sets, mass per unit length, and wire diameter.

[0116] The sorting planning module includes a regional planning unit and a coefficient calculation unit.

[0117] The regional planning unit is used to establish a color comparison library and divide the recognition area.

[0118] First, mark the material identifier corresponding to the equipment code in the batch information in the material comparison table, establish a color comparison library, and put all RGB color codes corresponding to the material identifier marked in the material attribute table into the color comparison library.

[0119] Secondly, the number of grids n is set, and n grid areas of the same size are evenly divided on the raw image. The RGB color code of each grid area is calculated, and the Euclidean distance algorithm is used to calculate the deviation value between the RGB color code of each grid area and each RGB color code in the color comparison library.

[0120] Finally, the deviation threshold U is set, and the RGB color code with a deviation value less than U in the color comparison library is used as the similar color code of the corresponding grid area. The corresponding material identifiers containing similar color codes in the RGB color code set are retrieved in the material attribute table, and these material identifiers are used as the similar material identifiers of the corresponding grid area. There is at least one grid area with a deviation value less than U in the mark, and the adjacent marked grid areas in the raw material image together constitute the recognition area.

[0121] The coefficient calculation unit is used to establish a rule base for each identification area and calculate a priority coefficient.

[0122] First, a rule base is established in each identification area to obtain similar material identifiers of all grid areas in the identification area. The wire diameter of the similar material identifier of each grid area in the identification area and the mass information per unit length are used as judgment rules. All judgment rules of the grid area are placed in the rule base of the identification area to which it belongs.

[0123] Secondly, count the number of grid areas j and the minimum deviation value of each grid in each recognition area. Calculate the deviation value of the RGB color code between each grid in the recognition area, and classify the grids with deviation values ​​less than U into the same category. Count the number of grids in different categories in each recognition area and calculate the standard deviation BZ.

[0124] Finally, the average number of grids j in all recognition areas is calculated ave , substitute into the formula: In , the priority coefficient of each identification area is calculated respectively. Among them, α is a constant, CT is the number of judgment rules in the rule base of the identification area, is the minimum deviation value of the i-th grid in the identification area.

[0125] An AI intelligent color sorter is used to optically sort the broken materials remaining after gravity sorting. Through near-infrared imaging and visible light imaging technology combined with AI algorithms of photo graphics, plastics and thin copper wires and electrical wire ends can be distinguished.

[0126] The dynamic adjustment module includes a machine sorting unit and an adjustment optimization unit.

[0127] The machine sorting unit is used to define the inspection sequence and control the robot arm sorting.

[0128] First, obtain the number h of each device code in the batch information P and the length of the marked material identifier. Multiply the length of each marked material identifier by the number h to obtain the expected length, and add the expected length to the judgment rule of the corresponding similar material identifier.

[0129] Secondly, the recognition areas are arranged in descending order according to the priority coefficients, and the corresponding rule base is obtained in sequence for detection according to the arrangement order of the recognition areas. The target recognition algorithm is used to detect all target objects in the recognition area, and the target objects are arranged in descending order according to the length Q, and the robot arm is controlled to grasp them in sequence according to the arrangement order of the target objects.

[0130] Finally, the arrangement order of the target objects is compared with the judgment rules in the rule base to determine whether there is a judgment rule that the wire diameter is equal and the length Q is less than the expected length. If so, the robot arm is controlled to clamp the target object. If not, other target objects in the recognition area are compared.

[0131] The adjustment and optimization unit is used to dynamically adjust the rule base and priority coefficient of each identification area.

[0132] After the robot grabs the target object, collect its mass R, divide the mass R by the length Q to get the mass per unit length T and compare it again in the corresponding rule library to determine whether there is a judgment rule that the mass per unit length is equal to T. If so, mark the target object and transfer it to the material accumulation area, subtract Q from the expected length to get a new expected length and update the judgment rule synchronously. If not, control the robot to put down the target object and continue to compare other target objects in the identification area. When the expected length of the marked material identifier F is zero, delete the judgment rules corresponding to F in all rule libraries and recalculate the priority coefficients of each identification area. Set the preset value B. When all identification areas have completed the detection, calculate the ratio Y of the sum of the masses of all marked target objects to the sum of the masses of the materials corresponding to the marked material identifier. Determine whether the ratio Y is greater than the preset value B. If the result is yes, enter the detection of the next batch of raw materials; if the result is not, increase the deviation threshold U and detect again until the ratio Y is greater than the preset value B.

[0133] The visualization module displays the image information of the raw materials and the specific location of each identification area through the operation interface and provides real-time feedback to the staff, obtains the RGB color code of the marked target object, and automatically updates the RGB color code set of the corresponding material in the material attribute table.

[0134] The AI ​​color sorter uses multispectral imaging technology, with a spectral range of visible light (400-700nm) and near infrared (700-1000nm) to accurately identify the difference in reflection between plastic and copper wire. Photographic image extraction technology extracts the shapes of thin copper wires and wires of different shapes for machine learning to form a picture library. The resolution of identifying plastic, copper wire and wire is continuously improved by comparing the similarity of the images in the picture library with the image extraction technology.

[0135] Embodiment 1:

[0136] Assume that there are five grid areas A1, A2, A3, A4 and A5 in the identification area, the number of judgment rules in the rule base of the identification area is 16, the standard deviation of the number of grids in different categories is 0.94, the average number of grids in all identification areas is 8, the constant α is 1, and the minimum deviation values ​​of grid areas A1, A2, A3, A4 and A5 are 2, 3, 5, 2, 3 respectively; substitute into the formula to calculate the priority coefficient of the identification area:

[0137]

[0138] Then the priority coefficient of this identification area is 0.02526.

[0139] Embodiment 2:

[0140] This embodiment also discloses a sorting system, which includes gravity sorting and the aforementioned intelligent sorting data management system connected in sequence. The sequential connection here means that during operation, gravity sorting and intelligent sorting are performed in sequence.

[0141] Compared with the existing technology, if traditional gravity sorting is used, then the materials to be sorted after gravity sorting account for about 10%, and additional manual labor is needed to further sort the wires and thin copper wires for this 10% of the materials to be sorted. Then, at least 1-2 personnel are needed for sorting on a sorting line. The cost of manual sorting is high, and the level and work quality of the sorting personnel are uneven, resulting in the sorting results being difficult to control.

[0142] Assuming that the price of wires and thin copper wires is 20,000 yuan / ton, and the price of plastics is 4,000 yuan / ton, when 1.8% of the wires and copper wires are not manually detected, then according to the mixed weight of a certain model of refrigerators 8.51kg / unit, 100 units are dismantled per hour, and 10 hours of work per day, the value of the undetected wires and the value of the plastics minus the wires are shown in Table 1 below:

[0143] Table 1

[0144]

[0145] The loss of profit is 2450.88 yuan per day. If machine sorting is adopted, the labor cost can be reduced from 4 people to 1 person. The added cost is shown in Table 2 based on 300 working days per year:

[0146] Table 2

[0147]

[0148] If visual recognition sorting is performed directly without gravity sorting, a large number of robots will be needed to achieve the same production capacity, which is very inefficient.

[0149] The present invention combines gravity sorting and intelligent sorting together. About 80%-90% of lightweight plastics can be removed by gravity sorting, and only less than 10%-20% of the materials need to undergo subsequent sorting, which has very high economic value. For example, the gravity sorting capacity designed by this unit is 1200KG-2000KG of plastic wire mixture per hour, corresponding to the processing of waste plastics generated by about 120-190 refrigerators per hour; the capacity of the AI ​​intelligent sorting equipment mentioned above can reach 600KG / hour, and the actual processing is about 130~200KG. It consumes about 10-13 kWh of electricity per hour, and through AI intelligent learning, the wire end sorting rate can reach about 95% or more.

[0150] It should be noted that, in this article, relational terms such as first and second, etc. are only used to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any such actual relationship or order between these entities or operations. Moreover, the terms "include", "comprise" or any other variants thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device.

[0151] Finally, it should be noted that the above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art can still modify the technical solutions described in the aforementioned embodiments or replace some of the technical features therein by equivalents. Any modification, equivalent replacement, improvement, etc. made within the spirit and principle of the present invention shall be included in the protection scope of the present invention.

Claims

1. A crushed plastic sorting device combining gravity sorting and AI recognition, characterized by: It comprises a gravity sorting system and an AI intelligent sorting system (6) connected in sequence, wherein the gravity sorting system is used to sort out some relatively light plastics from the crushed plastics, and to convey the relatively heavy intermediate materials after sorting to the AI ​​intelligent sorting system; the AI ​​intelligent sorting system sorts out target electric wires and thin copper wires through color sorting and visual recognition.

2. The crushed plastic sorting device combining gravity sorting and AI recognition according to claim 1 is characterized in that: The gravity separation system comprises a bouncing gravity separation device (2) and a centrifugal gravity separation device (3); the bouncing gravity separation device (2) comprises a bouncing gravity separation channel (30), a bouncing gravity separation inlet (20) arranged at the upper end of the bouncing gravity separation channel (30), and a bouncing gravity separation heavy material outlet (40) arranged at the lower end of the bouncing gravity separation channel (30); the centrifugal gravity separation device (3) comprises a centrifugal gravity separation fan (80) and a centrifugal gravity separation chamber (60); the lower end of the bouncing gravity separation channel (30) is connected to the centrifugal gravity separation fan (80) through a bouncing gravity separation air inlet (90); and the lower end of the centrifugal gravity separation chamber (60) is also provided with a centrifugal gravity separation heavy material outlet (70).

3. A broken plastic sorting method combining gravity sorting and AI recognition, characterized by: Gravity sorting adopts the crushed plastic sorting device combining gravity sorting and AI recognition as described in claim 1, and AI intelligent sorting includes the following steps: S100, collecting a material comparison table and a material attribute table, as well as batch information of raw materials after gravity sorting, and collecting image information of the raw materials through a high-definition camera in the sorting area; S200, analyzing raw material information and establishing a color comparison library, dividing identification areas on the raw material image through a machine vision algorithm, establishing a rule library for each identification area in combination with the color comparison library and calculating a priority coefficient; S300, defining the detection order according to the priority coefficient of the identification zone, using the target recognition algorithm to perform detection and dynamically adjusting the rule base and priority coefficient of each identification zone until the proportion of the detected raw materials reaches a preset value; S400, the operation interface displays the image information and detection marks of the raw materials in real time, feeds back the detection results of each time to the staff in real time, and automatically updates the material attribute table.

4. The broken plastic sorting method combining gravity sorting and AI recognition according to claim 3 is characterized in that: In S100, the raw material refers to the mixture obtained after the scrapped refrigerator is disassembled and initially gravity-sorted, and the material refers to the electric wires and thin copper wires mixed in the raw materials; the material comparison table includes sorting records of scrapped refrigerators of different models, each model of scrapped refrigerator corresponds to a sorting record, including an equipment code and a sorted material set; the sorted material set includes parameter records of different types of materials, each material corresponds to a parameter record, including a material identifier and a length; the material attribute table includes attribute information of different types of materials, including a material identifier, an RGB color code set, a mass per unit length, and a wire diameter; the batch information refers to the equipment code and quantity corresponding to all models of scrapped refrigerators from which the raw materials of this batch are sourced; the sorting area refers to an operation area used to sort raw materials.

5. The broken plastic sorting method combining gravity sorting and AI recognition according to claim 4 is characterized in that: In S200, the specific steps are as follows: S201. Obtain all equipment codes in the batch information P of the current raw material, retrieve all material identifiers in the sorting material set corresponding to these equipment codes in the material comparison table and mark them; establish a color comparison library, and put all elements in the RGB color code set corresponding to the marked material identifier in the material attribute table into the color comparison library; S202, setting the number of grids n, evenly dividing the raw material image of the current batch into n grid areas of the same size, extracting the pixel value of each grid area and calculating the RGB color code, and using the Euclidean distance algorithm to calculate the deviation value between the RGB color code of each grid area and each RGB color code in the color comparison library; S203, setting a deviation threshold U, and using the RGB color code with a deviation value less than U in the color comparison library as the similar color code of the corresponding grid area; searching the corresponding material identifiers containing similar color codes in the RGB color code set in the material attribute table, and using these material identifiers as the similar material identifiers of the corresponding grid area; S204, marking at least one grid area with a deviation value less than U, and adjacent marked grid areas in the raw material image together constitute an identification area; establishing a rule base for each identification area, obtaining similar material identifiers of all grid areas in the identification area, using the line diameter and mass information per unit length of each similar material identifier as a judgment rule, and placing all judgment rules of the grid area into the rule base of the corresponding identification area; S205, counting the number j of all grid areas in the identification area and the minimum deviation value of each grid; calculating the deviation value of the RGB color code between the grids in the identification area, and classifying the grids with a deviation value less than U into the same category; counting the number of grids in different categories in each identification area and calculating the standard deviation BZ, and substituting it into the formula to calculate the priority coefficient of each identification area respectively: In the formula, YX is the priority coefficient, j ave is the average number of grids in all recognition areas, α is a constant, CT is the number of judgment rules in the rule base of the recognition area, is the minimum deviation value of the i-th grid in the identification area.

6. The broken plastic sorting method combining gravity sorting and AI recognition according to claim 5 is characterized in that: In S300, the specific steps are as follows: S301, obtaining the number h of each device code in the batch information P and the length of the material identifier marked in the material comparison table, multiplying the length of each marked material identifier by the number h to obtain an estimated length, and adding the estimated length to the determination rule of the corresponding similar material identifier; arranging the identification areas in descending order according to the priority coefficients, and obtaining the corresponding rule base for detection in sequence according to the arrangement order of the identification areas; S302, using a target recognition algorithm to detect all target objects in the recognition area, collecting the wire diameter and length Q of each target object, arranging the target objects in descending order of length, and performing an initial comparison with each judgment rule in the rule base in turn according to the arrangement order of the target objects, to determine whether there is a judgment rule that the wire diameter is equal and the length Q is less than the expected length, if so, control the robot arm to clamp the target object, if not, continue to compare with other target objects in the recognition area; S303, after the robot arm grasps the target object, it collects its mass R, divides the mass R by the length Q to obtain the mass per unit length T, and compares it again in the corresponding rule library to determine whether there is a judgment rule that the mass per unit length is equal to T. If so, the target object is marked and transferred to the material accumulation area, and the estimated length is subtracted from Q to obtain a new estimated length and the judgment rule is updated synchronously; if not, the robot arm is controlled to put down the target object and continue to compare other target objects in the identification area; S304, when the estimated length of the marked material identifier F is zero, delete all the judgment rules corresponding to F in the rule base and recalculate the priority coefficient of each identification area; Set a preset value B. When all identification areas have completed the detection, calculate the ratio Y of the total mass of all marked target objects to the total mass of the materials corresponding to the marked material identifiers; determine whether the ratio Y is greater than the preset value B. If the result is yes, proceed to the detection of the next batch of raw materials; if the result is no, increase the deviation threshold U and detect again until the ratio Y is greater than the preset value B.

7. The broken plastic sorting method combining gravity sorting and AI recognition according to claim 6 is characterized in that: In S400, the image information of the current batch of raw materials and the specific location of each identification area, as well as the judgment result after each robot arm grasps the target, are displayed through the operation interface, and real-time feedback is given to the staff, and the RGB color code of the marked target object is automatically added to the RGB color code set of the corresponding material in the material attribute table.

8. A crushed plastic sorting system combining gravity sorting and AI recognition, applied to the crushed plastic sorting method combining gravity sorting and AI recognition as claimed in claim 3, characterized in that: The system includes a data acquisition module, a sorting planning module, a dynamic adjustment module and a visualization module; The data acquisition module is used to collect material comparison tables and material attribute tables, as well as batch information of raw materials after gravity sorting, and collect image information of raw materials through cameras; the sorting planning module establishes a color comparison library based on the raw material information, divides the recognition area on the raw material image through machine vision algorithms, and establishes a rule library for each recognition area in combination with the color comparison library and calculates the priority coefficient; the dynamic adjustment module defines the detection order according to the priority coefficient of the recognition area, uses the target recognition algorithm for detection and dynamically adjusts the rule library and priority coefficient of each recognition area until the proportion of detected raw materials reaches the preset value; the visualization module operation interface is used to display the image information and detection marks of the raw materials and feedback to the staff.

9. The crushed plastic sorting system combining gravity sorting and AI recognition according to claim 8 is characterized in that: The sorting planning module includes a regional planning unit and a coefficient calculation unit; The regional planning unit is used to establish a color comparison library and divide the identification area; First, mark the material identifier corresponding to the equipment code in the batch information in the material comparison table, establish a color comparison library, and put all the RGB color codes corresponding to the material identifier marked in the material attribute table into the color comparison library; Secondly, the number of grids n is set, and n grid areas of the same size are evenly divided on the raw image. The RGB color code of each grid area is calculated, and the deviation value between the RGB color code of each grid area and each RGB color code in the color comparison library is calculated using the Euclidean distance algorithm. Finally, a deviation threshold U is set, and the RGB color code with a deviation value less than U in the color comparison library is used as the similar color code of the corresponding grid area; the corresponding material identifiers containing similar color codes in the RGB color code set are retrieved in the material attribute table, and these material identifiers are used as the similar material identifiers of the corresponding grid area; there is at least one grid area with a deviation value less than U, and the adjacent marked grid areas in the raw material image together constitute the identification area; The coefficient calculation unit is used to establish a rule base for each identification area and calculate the priority coefficient; First, a rule base is established for each identification area, and similar material identifiers of all grid areas in the identification area are obtained. The wire diameter of the similar material identifier of each grid area in the identification area and the mass information per unit length are used as the judgment rules, and all the judgment rules of the grid area are placed in the rule base of the corresponding identification area; Secondly, the number of grid areas j and the minimum deviation value of each grid in each recognition area are counted respectively; the deviation value of the RGB color code between the grids in the recognition area is calculated, and the grids with deviation values ​​less than U are classified into the same category; the number of grids in different categories in each recognition area is counted and the standard deviation BZ is calculated; Finally, calculate the average number of grids j in all recognition areas ave , substitute into the formula: In the above example, the priority coefficient of each identification area is calculated respectively; where α is a constant, CT is the number of judgment rules in the rule base of the identification area, is the minimum deviation value of the i-th grid in the identification area.

10. The broken plastic sorting system combining gravity sorting and AI recognition according to claim 9, characterized in that: The dynamic adjustment module includes a machine sorting unit and an adjustment optimization unit; The machine sorting unit is used to define the detection sequence and control the robot arm sorting; First, the number h of each device code and the length of the marking material identifier in the batch information P are obtained, and the length of each marking material identifier is multiplied by the number h to obtain the expected length, and the expected length is added to the judgment rule of the corresponding similar material identifier; Secondly, the recognition areas are arranged in descending order according to the priority coefficients, and the corresponding rule base is obtained in sequence for detection according to the arrangement order of the recognition areas; the target recognition algorithm is used to detect all target objects in the recognition area, and the target objects are arranged in descending order according to the length Q, and the robot arm is controlled to grasp them in sequence according to the arrangement order of the target objects; Finally, the target objects are initially compared with the judgment rules in the rule base in the order of their arrangement to determine whether there is a judgment rule that the wire diameter is equal and the length Q is less than the expected length. If so, the robot arm is controlled to clamp the target object. If not, other target objects in the recognition area are compared. The adjustment and optimization unit is used to dynamically adjust the rule base and priority coefficient of each identification area; After the robot grasps the target object, its mass R is collected, and the mass R is divided by the length Q to obtain the mass per unit length T, which is then compared again in the corresponding rule library to determine whether there is a judgment rule that the mass per unit length is equal to T. If so, the target object is marked and transferred to the material accumulation area, and the estimated length is subtracted from Q to obtain a new estimated length and the judgment rule is updated synchronously; if not, the robot arm is controlled to put down the target object and continue to compare other target objects in the identification area; when the estimated length of the marked material identifier F is zero, delete all judgment rules corresponding to F in the rule library and recalculate the priority coefficients of each identification area; set a preset value B, and when all identification areas have completed the detection, calculate the ratio Y of the sum of the masses of all marked target objects to the sum of the masses of the materials corresponding to the marked material identifiers; determine whether the ratio Y is greater than the preset value B, and if so, enter the detection of the next batch of raw materials; if not, increase the deviation threshold U and detect again until the ratio Y is greater than the preset value B.

Images