High-speed waste plastic sorting method based on multispectral fusion and element discrimination

Abstract

The invention belongs to the technical field of solid waste recycling and intelligent sorting equipment, and discloses a waste plastic high-speed sorting method based on multispectral fusion and element screening, which is suitable for online separation of general-purpose plastics such as PP (polypropylene), PE (polyethylene), PVC (polyvinyl chloride) and PET (polyethylene terephthalate). Aiming at the problems of difficulty in black plastic identification, insufficient PVC / halogen-containing risk discrimination and wrong sorting caused by industrial drift in traditional near-infrared sorting, a link of visible light positioning, near-infrared main identification, intermediate infrared / medium-wave infrared reinspection, X-ray fluorescence element discrimination, multi-modal fusion decision-gas injection separation is adopted, grading detection is triggered through confidence, and the sorting accuracy is improved. A risk priority rule and an online self-calibration compensation strategy are combined, and a two-stage sorting framework is matched. According to the method, high-risk PVC particles are preferentially removed, black plastic is accurately recognized, the separation purity is improved while high productivity is guaranteed, the industrial operation stability and expandability are enhanced, and the waste plastic recycling benefit is remarkably improved.

Description

Technical Field

[0001] This invention relates to the field of solid waste resource utilization and intelligent sorting equipment technology, specifically a high-speed sorting method for waste plastics based on multispectral fusion and elemental identification. Background Technology

[0002] Waste plastics, as a valuable recyclable resource, play a crucial role in alleviating resource shortages and reducing environmental pollution through recycling. The core prerequisite for efficient resource utilization of waste plastics is achieving high-purity sorting of plastics of different materials, which directly determines the recycling value of plastics and the quality stability of downstream processed products. Among these, the precise separation of general-purpose plastics such as PP, PE, PVC, and PET, which are the main categories of waste plastics recycled, has become a key technological requirement in the industry.

[0003] Currently, near-infrared (NIR) reflectance spectroscopy is the most widely used waste plastic sorting technology in the industrial sector. This technology has become widespread in large-scale waste plastic sorting scenarios due to its advantages such as fast detection speed, high throughput, and good identification of most non-black plastics, enabling preliminary classification of common plastics such as PET, PE, PP, and PVC. However, in actual industrial operating environments, this traditional technology still faces many intractable bottlenecks, severely restricting sorting purity and long-term operational stability, specifically in the following aspects: First, identifying black or dark-colored waste plastics presents a significant challenge. Because these plastics contain carbon black pigments or dark colorants, their near-infrared reflection signals are extremely weak, severely masking their effective spectral characteristics. Near-infrared identification models struggle to extract distinguishable material information, leading to frequent missorting. This not only reduces the sorting purity of the target plastics but also wastes recyclable resources.

[0004] Secondly, there is insufficient ability to identify PVC and other high-risk halogen-containing plastics. If PVC is mixed into high-value plastic streams such as PET, PE, and PP, it will release harmful gases during downstream melting and processing, and severely affect the mechanical properties and processing stability of recycled plastics, bringing significant safety hazards and quality risks. Relying solely on near-infrared spectroscopy can only identify based on the molecular structure characteristics of polymers, which is insufficient to cover the halogen risks in complex pollution scenarios. It cannot accurately identify PVC and other chlorine- or bromine-containing flame-retardant plastics, making it difficult to meet the quality requirements for high-purity recycling.

[0005] Furthermore, industrial environmental factors cause spectral drift and poor long-term operational stability. In industrial sites for waste plastic sorting, problems such as dust pollution, scaling of the detection window, attenuation of light source intensity, and fluctuations in ambient temperature are unavoidable. These factors can cause the collected spectral signals to drift, resulting in a gradual decrease in the accuracy of the originally trained recognition model. This necessitates frequent shutdowns for calibration, which affects production efficiency and increases maintenance costs.

[0006] Finally, traditional technologies struggle to balance the dual demands of high production capacity and high purity. Some improved solutions, in an effort to enhance recognition accuracy, blindly increase the detection process and computational complexity, leading to increased signal processing delays and consequently limiting the conveyor belt's operating speed and material throughput. Conversely, simplifying the detection process to ensure production capacity sacrifices recognition accuracy, making high-purity sorting impossible, thus creating a conflict between production capacity and purity.

[0007] In summary, existing near-infrared sorting technology has significant shortcomings in areas such as black plastic identification, high-risk pollutant identification, industrial environmental adaptability, and balance of production capacity and purity. There is an urgent need for a waste plastic sorting technology and device that can overcome the above-mentioned technical bottlenecks and take into account high sorting accuracy, high operational stability, high processing throughput, and strong risk identification capabilities, so as to promote the high-quality development of the waste plastic recycling industry. Summary of the Invention

[0008] The purpose of this invention is to address the shortcomings of existing technologies by providing a high-speed sorting method for waste plastics based on multispectral fusion and elemental identification, thereby solving the problems mentioned in the background.

[0009] To achieve the above objectives, the present invention provides the following technical solution: a high-speed sorting method for waste plastics based on multispectral fusion and elemental identification, comprising the following steps: S1. After the mixed waste plastic granules are fed and evenly distributed and treated to form a single layer, they form a single layer of dispersed flow on the conveyor belt. The visible light imaging module extracts the geometric size, color / brightness, surface stains and moisture content of the granules to achieve particle positioning and trajectory tracking. S2. Collect near-infrared reflectance spectroscopy or near-infrared hyperspectral data of particles, and output the first determination result and first confidence level of polymer category based on the first identification model; S3. When the particles meet any of the following triggering conditions: black / dark color criterion, first confidence level is lower than preset threshold, surface stains / moisture exceed the standard, or suspected multi-layer / composite structure, collect mid-infrared or mid-wave infrared / hyperspectral data, and output the second judgment result and second confidence level based on the second recognition model. S4. When the first or second determination result is PET and the target is bottle-grade high-purity PET, or PE / PP and the system is in flame-retardant plastic rejection mode, or the risk of PVC is uncertain and mixing will increase the processing risk, X-ray fluorescence detection is performed and a chlorine / bromine element risk label is output. S5. Perform confidence-normalized weighted fusion on the multimodal results. If the chlorine element risk label is true, it is prioritized as high risk and removed. Based on this, a sorting instruction is generated. S6. By combining particle positioning coordinates, conveyor belt speed, flight time model and spray valve response delay model to calculate the trigger time, the multi-channel air spray module is controlled to complete the separation of PP, PE, PVC and PET and the removal of high-risk particles. At the same time, online self-calibration and drift compensation are achieved through periodic calibration of black / gray / white standard parts and temperature and light compensation.

[0010] As a preferred embodiment of the present invention, the near-infrared reflectance spectrum acquisition band in step S2 is 1100-2500nm, and the acquisition device is equipped with a diffuse reflection illumination structure and a dust-resistant window.

[0011] As a preferred technical solution of the present invention, the mid-infrared or mid-wave infrared spectrum acquisition band of step S3 covers 3–14 μm, and the acquisition device is a line-scan hyperspectral camera or an array-type spectral sensor.

[0012] As a preferred embodiment of the present invention, the flight time model parameters in step S6 are corrected in real time through online learning to improve the triggering accuracy of the injection valve.

[0013] As a preferred technical solution of the present invention, the preset threshold in step S3 is determined by sample training and adaptation to industrial scenarios, and supports dynamic adjustment.

[0014] A high-speed waste plastic sorting device based on multispectral fusion and elemental identification, used to implement the above-described method, comprising: The feeding and distribution module is used to uniformly feed mixed plastic granules; A single-layer conveying module is used to disperse particles into a single layer and convey them at a set speed; The visible light imaging module is used to acquire particle images and output their location and state features; The near-infrared recognition module is used to collect near-infrared spectral data and output the first judgment result and confidence level; The mid-infrared / mid-wave infrared recognition module is used to collect corresponding spectral data and output a second judgment result and confidence level when the triggering conditions are met. The X-ray fluorescence element identification module is used to output chlorine / bromine element risk labels when the risk identification conditions are met. The real-time fusion decision module is used to perform multimodal result fusion and risk priority decision-making, and output sorting instructions; The multi-channel air jet separation module uses a multi-jet valve array to achieve particle airflow deflection and separation according to sorting instructions; The online calibration module has a built-in black / gray / white standard reflectance reference and an automatic switching mechanism for periodically calibrating the spectral identification module.

[0015] As a preferred technical solution of the present invention, the real-time fusion decision module includes a confidence assessment unit and a risk priority unit, and prohibits particles from entering the PET collection bin when the chlorine element risk label is true.

[0016] As a preferred embodiment of the present invention, the device has a two-stage sorting structure. The first stage coarsely separates PET, PE / PP, and PVC / halogen-containing hazardous materials, and the second stage finely separates and purifies the PET or PE / PP materials.

[0017] As a preferred technical solution of the present invention, the visible light imaging module and the near-infrared recognition module adopt a coaxial or calibrated extrinsic multi-camera structure to achieve cross-modal alignment.

[0018] As a preferred embodiment of the present invention, the spacing between the spray valves of the spray valve array is matched with the conveying speed, which is suitable for the stable separation of particles of different sizes.

[0019] Compared with the prior art, the beneficial effects of the present invention are: 1. Significantly improved accuracy in identifying black / dark plastics: Addressing the pain points of traditional near-infrared sorting, which suffers from weak reflective signals and difficulty in feature extraction for black / dark plastics, this invention utilizes a confidence-triggered mechanism to initiate mid-infrared / mid-wave infrared re-examination only for black / dark and low-confidence particles. By leveraging the stable polymer fingerprint features in this band, accurate identification of black plastics is achieved, effectively avoiding missorting and improving the identification accuracy by more than 15% compared to traditional solutions.

[0020] 2. Reliable Removal of High-Risk PVC / Halogen-Containing Particles: An innovative X-ray fluorescence elemental identification module is introduced to directly detect chlorine / bromine elements and output risk labels. Combined with risk-priority decision-making rules, PVC and halogen-containing flame-retardant plastics are preferentially removed. This completely solves the shortcomings of traditional spectroscopy, which relies solely on molecular structure characteristics and is difficult to cover complex pollution scenarios. The PVC contamination rate in downstream high-value plastic streams can be reduced to below 500ppm, significantly reducing the risk of melt processing.

[0021] 3. Dual optimization of capacity and purity: Adopting a hierarchical detection architecture of "near-infrared main identification + on-demand re-inspection / risk screening", it performs rapid identification of the vast majority of non-black, high-confidence particles, and only initiates additional detection for special particles. While ensuring sorting accuracy, it significantly reduces the detection and calculation burden. The conveyor belt speed can be maintained at 1.5-3m / s, and the throughput reaches 500-1000kg / h. With the two-stage sorting structure, the final target plastic sorting purity exceeds 98%, achieving a balance between high capacity and high purity.

[0022] 4. Significantly enhanced adaptability and stability to industrial environments: By using built-in black / gray / white standard reflective reference components and an automatic switching mechanism, combined with temperature and light intensity sensors to establish a drift compensation function, the spectral module can achieve online self-calibration and drift compensation, effectively resisting industrial environmental interference such as dust, light source attenuation, temperature drift and window contamination. The recognition accuracy rate decreases by no more than 1% within 72 hours of continuous operation, greatly reducing the frequency of downtime calibration and lowering operation and maintenance costs.

[0023] 5. High scalability and practicality: The device modules adopt a standardized design, which can flexibly adjust the near-infrared / mid-infrared detection parameters, confidence thresholds and risk identification conditions according to different recycling scenarios. It is suitable for sorting various general plastics such as PP, PE, PVC and PET. It is applicable to the sorting of bottle-grade high-purity PET, and can also meet the sorting of mixed hard plastics from household sources with a high proportion of black plastic. It has a wide range of applications and can significantly improve the economic and environmental benefits of waste plastic recycling. Attached Figure Description

[0024] Figure 1 This is the overall system flowchart of the present invention; Figure 2 This is a layout diagram of the sensor and air jet of the present invention; Figure 3 This is the confidence-triggered decision logic diagram of the present invention; Figure 4 This is a diagram of the two-stage sorting scheme of the present invention; Figure 5 This is a schematic diagram of the air jet triggering timing of the present invention. Detailed Implementation

[0025] The preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings, so that the advantages and features of the present invention can be more easily understood by those skilled in the art, thereby providing a clearer and more explicit definition of the scope of protection of the present invention.

[0026] I. Overall Structure and Component Configuration of the Device The core architecture of the device of this invention is as follows: Figure 1 , Figure 2 As shown, the specific configuration and functional implementation of each module are as follows: Feeding and distribution module: It adopts a combination structure of vibrating feeder and spiral distributor to evenly convey mixed waste plastic granules (including PP, PE, PVC, PET and black / dark plastics) with a particle size range of 2-20mm to the subsequent conveying module. The feeding rate is controlled at 500-1000kg / h to ensure that the material does not accumulate or stop flowing.

[0027] Single-layer conveyor module: A stainless steel conveyor belt is used, with a sandblasted surface to increase friction. The conveyor belt speed is adjustable from 1.5-3 m / s. A comb-shaped distributor and airflow distribution device above the conveyor belt disperse the material into a single-layer particle flow, with particle spacing controlled at 5-10 mm to prevent particle overlap from affecting detection accuracy.

[0028] Visible light imaging module: Employs an industrial camera with a resolution of 2048×1536 pixels, paired with an LED surface light source (color temperature 5500K), mounted 1.2m directly above the conveyor belt. The camera frame rate is set to 100fps. A target detection algorithm (YOLOv5) is used to extract the geometric dimensions (length, width, area), color / brightness (RGB mean, grayscale value), and surface dirt and moisture indicators (grayscale variance, percentage of abnormal reflective areas) of each particle. The module outputs a unique ID, positioning coordinates (X, Y), and state feature vector for each particle, enabling full-process particle trajectory tracking.

[0029] Near-infrared recognition module: A near-infrared hyperspectral camera with a working wavelength of 1100-2500nm is used, equipped with a diffuse reflection ring illuminator and a calcium fluoride dust-resistant window, and installed 0.8m below and on the same side as the visible light camera. The camera has a spectral resolution of 10nm, and the acquisition rate is synchronized with the visible light camera. The first recognition model is established using the Partial Least Squares Discriminant Analysis (PLS-DA) algorithm, which outputs the polymer category (PET / PE / PP / PVC / other) and the first confidence level C1 (0-1 interval).

[0030] Mid-infrared / mid-wave infrared recognition module: Utilizes a line-scan mid-infrared hyperspectral camera, operating in the 3-14μm band with a spectral resolution of 4cm². -1 It is installed 0.5m downstream of the near-infrared recognition module. It only starts data acquisition when the trigger condition is met, establishes a second recognition model through the support vector machine (SVM) algorithm, and outputs a second judgment result and a second confidence level C2 (0-1 interval), which focuses on improving the recognition accuracy of black / dark plastic.

[0031] X-ray fluorescence elemental discrimination module: Employs a miniature X-ray fluorescence spectrometer, covering halogens such as Cl and Br, with a detection limit ≤500ppm. It is installed 0.3m downstream of the mid-infrared recognition module. Detection is triggered only when elemental risk discrimination conditions are met, with a detection duration ≤20ms, outputting a chlorine risk label (True / False) and / or a bromine risk label (True / False).

[0032] Real-time fusion decision module: Based on an FPGA+ARM architecture hardware processing platform, the computation latency is ≤5ms. It has a built-in confidence assessment unit and risk priority unit. First, C1 and C2 are normalized (normalization formula: C'=C / (C1+C2)), and then the fusion result is calculated according to the weighted fusion rule (final category confidence = 0.6×C1'+0.4×C2'). If the chlorine element risk label is True, it is directly judged as PVC / chlorine-containing high risk, and its elimination priority is higher than that of polymer category priority. At the same time, the particles are prohibited from entering the PET collection bin.

[0033] Multi-channel air-jet separation module: Includes 6 sets of electromagnetic spray valves (response time ≤ 5ms) and 4 collection bins (PET bin, PE / PP bin, PVC / halogenated risk bin, and other bins). The spray valve spacing is 80mm, matched to the conveyor belt speed (1.5-3m / s). The spray valves are installed 0.4m below the end of the conveyor belt and deflect the airflow to the target particles according to the sorting instructions. The deflection pressure can be adjusted within the range of 0.3-0.6MPa to accommodate the stable separation of particles of different sizes.

[0034] Online calibration module: Built-in black / gray / white standard reflective references (reflectivity 0%, 50%, and 99% respectively) and an electric switching mechanism. The calibration process is automatically triggered every 2 hours of operation or when a light source intensity drift exceeding 10% is detected. Combining the outputs of a temperature sensor (measurement range -10-60℃, accuracy ±0.5℃) and a light intensity sensor (measurement range 0-10000 lux, accuracy ±1%), a drift compensation function (f(λ)=k×I(λ)×T(λ), where k is the calibration coefficient, I(λ) is the light intensity value, and T(λ) is the temperature correction coefficient) is established to standardize and correct the acquired spectrum.

[0035] II. Specific Process of Sorting Method Combination Figure 3 , Figure 4 , Figure 5 Based on the decision-making logic and timing control requirements, the detailed execution steps of the sorting method are as follows: Step 1: Material Pre-processing and Positioning Tracking The feeding and distribution module and the single-layer conveying module are activated. After the mixed waste plastic granules are fed by vibration, spirally distributed, and combed, they form a single-layer dispersed flow on the conveyor belt. The visible light imaging module acquires images of the conveyor belt in real time, detects particle targets using the YOLOv5 algorithm, extracts the geometric size, color / brightness, surface stains, and moisture content indication features of each particle, assigns a unique ID, and records the positioning coordinates (X, Y). Based on the Kalman filter algorithm, particle trajectory tracking is realized, and the trajectory prediction curve is output.

[0036] Step 2: Near-infrared main recognition When a particle moves into the detection area of ​​the near-infrared recognition module, the module simultaneously acquires the near-infrared hyperspectral data of the particle, inputs it into the pre-trained PLS-DA recognition model, outputs the first determination result of the polymer category (such as PET) and the first confidence level C1 (such as 0.85), and stores the result bound to the particle ID.

[0037] Step 3: Triggering and Execution of Mid-Infrared / Mid-Wave Infrared Re-inspection Real-time determination of particle performance based on trigger conditions: ① Particle grayscale value ≤ 30 (black / dark color criterion); ② C1 < 0.7 (preset confidence threshold); ③ Surface stain / moisture indicator characteristics > 0.3 (preset threshold, calculated by weighting grayscale variance and the proportion of reflective abnormal areas); ④ Determined as a multi-layered / composite structure particle based on geometric dimensions and surface features. If any condition is met, the mid-infrared / mid-wave infrared recognition module is triggered. When a particle moves into the detection area of ​​this module, mid-infrared hyperspectral data is collected, input into the SVM recognition model, and a second determination result and a second confidence level C2 (e.g., 0.92) are output, along with the particle ID. Particles that do not meet the trigger conditions skip this step and proceed directly to step 4.

[0038] Step 4: Triggering and Execution of X-ray Fluorescence Element Risk Identification Determine if the particles meet the elemental risk screening criteria: ① The first or second judgment result is PET and the sorting target is bottle-grade high-purity PET (purity requirement ≥99.5%); ② The first or second judgment result is PE / PP and the system is in flame-retardant plastic rejection mode; ③ The fusion decision (based on the initial C1 judgment) outputs an uncertain PVC risk (C1 between 0.5 and 0.7), and PVC contamination will increase the risk of downstream melt processing (e.g., the customer requires PVC content ≤500ppm in the PET stream). If any condition is met, the X-ray fluorescence elemental screening module is triggered. When the particles move to the detection area of ​​this module, elemental detection is initiated, a chlorine / bromine elemental risk label (e.g., chlorine label True) is output, and the particle ID is bound. Particles that do not meet the risk screening criteria skip this step and proceed directly to step 5.

[0039] Step 5: Multimodal fusion decision The real-time fusion decision module calls upon the visible light state characteristics bound to the particle ID, the near-infrared determination result (C1), the mid-infrared / mid-wave infrared determination result (if any, including C2), and the X-ray fluorescence risk label (if any) to perform the fusion decision: If a chlorine risk label exists and is True, the particle is directly identified as PVC / high-risk chlorine content, and a sorting instruction to "remove to the risk warehouse" is generated, with the highest priority. If there is no chlorine element risk label or the label is False, and mid-infrared / mid-wave infrared re-inspection is not triggered, the first judgment result will be used as the final category, and the sorting instruction for the corresponding collection bin will be generated. If there is no chlorine element risk label or the label is False, and mid-infrared / mid-wave infrared re-inspection has been triggered, C1 and C2 are normalized, the final category confidence is calculated according to the weighted fusion rule, the category with the highest confidence is selected as the final result, and the sorting instruction for the corresponding collection bin is generated.

[0040] Step 6: Spray valve timing control and particle separation Based on the particle's current location coordinates, conveyor belt speed, and trajectory prediction curve, combined with the particle flight time model (calculated based on air resistance and gravity, formula: t_flight = √(2h / g), where h is the height difference between the spray valve and the collection chamber inlet, and g is the acceleration due to gravity) and the spray valve response delay model (Δt_delay = 5ms), the spray valve trigger time is calculated as: t_trigger = t_current + (LX) / vt_flight - Δt_delay, where L is the X-coordinate of the spray valve installation position, and v is the conveyor belt speed. An online learning algorithm (adjusting the air resistance coefficient in the flight time model based on actual separation results for every 1000 particles processed) optimizes the trigger time to ensure precise particle entry into the target collection chamber. When a particle reaches the trigger position, the corresponding spray valve opens, and the jet airflow deflects the particle to the designated collection chamber, completing the sorting process.

[0041] Application Scenario Implementation Examples Example 1: Sorting of bottle-grade high-purity PET (target purity ≥ 99.5%) Material composition: PET bottle flakes (90%), PE / PP fragments (6%), PVC granules (3%), black plastic fragments (1%), with a particle size range of 3-10mm.

[0042] Device parameter settings: conveyor belt speed 2m / s, feeding rate 800kg / h, near-infrared confidence threshold 0.7, spray valve air pressure 0.5MPa, X-ray fluorescence detection trigger condition is "first or second judgment result is PET".

[0043] Implementation process: After pretreatment, the mixed bottle flakes form a single-layer dispersed flow. The visible light module completes the positioning and tracking, and identifies the black particles (grayscale value ≤30). The near-infrared module performs primary identification of non-black particles, outputs the PET class judgment result and C1 (e.g., 0.88), and triggers mid-infrared re-inspection for particles with C1 < 0.7; The mid-infrared module re-examines black particles and low-confidence particles, correcting the judgment results (e.g., correcting black PET particles that were originally misjudged as "other" to PET, C2=0.91). X-ray fluorescence detection is triggered for all particles identified as PET, chlorine-containing particles (PVC) are detected and a chlorine label True is output; The fusion decision module prioritizes removing PET particles with a chlorine label of True from the risk warehouse, while the remaining PET particles are assigned to the PET warehouse based on the weighted fusion result. Initiate secondary fine sorting and repeat steps 3-5 above for the PET silo material to further remove residual PVC and other impurities. The final PET sorting purity reaches 99.7%, and the capacity is maintained at 800 kg / h.

[0044] Example 2: Sorting of mixed hard plastics from residential sources (30% black plastic) Material composition: black PE / PP (25%), black PVC (5%), white PET (30%), colored PE / PP (40%), particle size range 5-15mm.

[0045] Device parameter settings: conveyor belt speed 2.5m / s, feeding rate 1000kg / h, near-infrared confidence threshold 0.75, spray valve air pressure 0.45MPa, X-ray fluorescence detection trigger condition is "system in flame retardant plastic rejection mode".

[0046] Implementation process: After pretreatment, the hybrid rigid plastic forms a single-layer dispersed flow, and the visible light module identifies black particles (grayscale value ≤30) and tracks their trajectory; The near-infrared module performs main identification of colored and white particles, and outputs the judgment results such as PE / PP, PET and C1 (e.g., C1=0.90 for colored PE / PP). Mid-infrared re-inspection was triggered for black particles and particles with C1 < 0.75, and black PE / PP (C2 = 0.88) and black PVC (C2 = 0.93) were accurately identified by SVM model. The system activates the flame-retardant plastic rejection mode, triggers X-ray fluorescence detection on particles identified as PE / PP, detects bromine-containing flame-retardant PE / PP and outputs a bromine label True; The integrated decision-making module removes particles labeled True for chlorine / bromine into the risk bin, black PE / PP into the PE / PP bin, white PET into the PET bin, and other materials into the corresponding collection bins according to their categories. Ultimately, the accuracy rate for identifying black plastics reached 98.5%, the rejection rate for PVC / halogenated risk particles reached 99%, and the purity of PE / PP and PET sorting both exceeded 98%, with a production capacity maintained at 1000 kg / h.

[0047] Key technical parameter verification In this embodiment, the core technical parameters of the device have been verified through actual testing as follows: Sorting speed: 1.5-3m / s (conveyor belt speed), throughput: 500-1000kg / h; Recognition accuracy: ≥99% for non-black plastics, ≥98.5% for black plastics; Risk particle rejection rate: PVC / halogen-containing particles ≥99%; Sorting purity: PET bin ≥ 99.5%, PE / PP bin ≥ 98%; Long-term stability: After 72 hours of continuous operation, the recognition accuracy decreases by ≤1%; Applicable particle size: 2-20mm.

[0048] Through the above specific implementation methods, the present invention achieves high-purity and high-speed sorting of general plastics such as PP, PE, PVC, and PET, effectively solving the problems of difficulty in identifying black plastics, insufficient identification of PVC / halogenated risks, and missorting caused by industrial drift in traditional near-infrared sorting. It takes into account both production capacity and purity requirements and can be widely used in waste plastic recycling plants, recycled plastic processing enterprises, and other scenarios.

[0049] The above embodiments merely illustrate implementation methods of the present invention, and while the descriptions are relatively specific and detailed, they should not be construed as limiting the scope of the invention patent. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these all fall within the protection scope of the present invention.

Claims

1. A high-speed sorting method for waste plastics based on multispectral fusion and elemental identification, characterized in that, Includes the following steps: S1. After the mixed waste plastic granules are fed and evenly distributed and treated to form a single layer, they form a single layer of dispersed flow on the conveyor belt. The visible light imaging module extracts the geometric size, color / brightness, surface stains and moisture content of the granules to achieve particle positioning and trajectory tracking. S2. Collect near-infrared reflectance spectroscopy or near-infrared hyperspectral data of particles, and output the first determination result and first confidence level of polymer category based on the first identification model; S3. When the particles meet any of the following triggering conditions: black / dark color criterion, first confidence level is lower than preset threshold, surface stains / moisture exceed the standard, or suspected multi-layer / composite structure, collect mid-infrared or mid-wave infrared / hyperspectral data, and output the second judgment result and second confidence level based on the second recognition model. S4. When the first or second determination result is PET and the target is bottle-grade high-purity PET, or PE / PP and the system is in flame-retardant plastic rejection mode, or the risk of PVC is uncertain and mixing will increase the processing risk, X-ray fluorescence detection is performed and a chlorine / bromine element risk label is output. S5. Perform confidence-normalized weighted fusion on the multimodal results. If the chlorine element risk label is true, it is prioritized as high risk and removed. Based on this, a sorting instruction is generated. S6. By combining particle positioning coordinates, conveyor belt speed, flight time model and spray valve response delay model to calculate the trigger time, the multi-channel air spray module is controlled to complete the separation of PP, PE, PVC and PET and the removal of high-risk particles. At the same time, online self-calibration and drift compensation are achieved through periodic calibration of black / gray / white standard parts and temperature and light compensation.

2. The method according to claim 1, characterized in that, In step S2, the near-infrared reflectance spectrum is collected in the 1100-2500nm band, and the acquisition device is equipped with a diffuse reflection illumination structure and a dust-resistant window.

3. The method according to claim 1, characterized in that, The mid-infrared or mid-wave infrared spectrum acquisition band in step S3 covers 3–14 μm, and the acquisition device is a line-scan hyperspectral camera or an array-type spectral sensor.

4. The method according to claim 1, characterized in that, The flight time model parameters in step S6 are corrected in real time through online learning to improve the triggering accuracy of the injection valve.

5. The method according to claim 1, characterized in that, The preset threshold in step S3 is determined through sample training and adaptation to industrial scenarios, and supports dynamic adjustment.

6. A high-speed waste plastic sorting device based on multispectral fusion and elemental identification, used to implement the method according to any one of claims 1-5, characterized in that, include: The feeding and distribution module is used to uniformly feed mixed plastic granules; A single-layer conveying module is used to disperse particles into a single layer and convey them at a set speed; The visible light imaging module is used to acquire particle images and output their location and state features; The near-infrared recognition module is used to collect near-infrared spectral data and output the first judgment result and confidence level; The mid-infrared / mid-wave infrared recognition module is used to collect corresponding spectral data and output a second judgment result and confidence level when the triggering conditions are met. The X-ray fluorescence element identification module is used to output chlorine / bromine element risk labels when the risk identification conditions are met. The real-time fusion decision module is used to perform multimodal result fusion and risk priority decision-making, and output sorting instructions; The multi-channel air jet separation module uses a multi-jet valve array to achieve particle airflow deflection and separation according to sorting instructions; The online calibration module has a built-in black / gray / white standard reflectance reference and an automatic switching mechanism for periodically calibrating the spectral identification module.

7. The apparatus according to claim 6, characterized in that, The real-time fusion decision-making module includes a confidence assessment unit and a risk priority unit. When the chlorine risk label is true, particles are prohibited from entering the PET collection chamber.

8. The apparatus according to claim 6, characterized in that, The device has a two-stage sorting structure. The first stage coarsely separates PET, PE / PP, and PVC / halogen-containing hazardous materials. The second stage finely separates and purifies either PET or PE / PP.

9. The apparatus according to claim 6, characterized in that, The visible light imaging module and the near-infrared recognition module adopt a coaxial or calibrated extrinsic multi-camera structure to achieve cross-modal alignment.

10. The apparatus according to claim 6, characterized in that, The spacing between the spray valves in the spray valve array is matched with the conveying speed, which is suitable for the stable separation of particles of different sizes.

Images