An interpretable plastic material identification method and system

By constructing a material knowledge base that combines spectral features and physical auxiliary features, and using rule-based reasoning to generate an interpretable chain of evidence, the problem of uninterpretable plastic identification results in existing technologies is solved, and a highly accurate and traceable identification process is achieved.

CN123066005APending Publication Date: 2026-11-10GUANGZHOU CITY POLYTECHNIC
View PDF 4 Cites 0 Cited by

Patent Information

Application Number
CN202611212125.X
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-08-11
Publication Date
2026-11-10

AI Technical Summary

Technical Problem

Existing plastic identification technologies struggle to provide interpretable identification results and lack sufficient accuracy and consistency in non-laboratory environments, failing to meet the requirements of modern industry for process traceability and verifiable results.

Method used

By constructing a structured materials knowledge base, combining standardized spectral features with quantifiable physical auxiliary features, rule-based reasoning is used to generate an interpretable chain of evidence, outputting identification results and providing evidence support.

Benefits of technology

It achieves transparency in the plastic identification process and verifiability of conclusions, improves the accuracy and robustness of identification, reduces the risk of misjudgment, supports ease of use for non-professionals, and ensures the traceability of results.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123066005A_ABST
    Figure CN123066005A_ABST
Patent Text Reader

Abstract

The application provides an interpretable plastic material identification method and system, which specifically comprises the following steps: S1, obtaining the standard wavelength absorbance characteristics of a plastic sample to be measured; S2, obtaining the auxiliary identification characteristics of the plastic sample to be measured; S3, reading a material knowledge base; S4, calculating the optical matching score, the auxiliary characteristic matching score and the characteristic conflict penalty term for each candidate plastic material; S5, sorting the candidate material probabilities; S6, outputting the identification result, the confidence, the unknown material prompt or the re-inspection suggestion according to the highest probability, the difference between the highest probability and the second highest probability, the distance outside the interval and the characteristic conflict degree; and S7, generating an interpretable evidence chain. The application fuses the spectral characteristics and quantifiable physical auxiliary characteristics, and constructs a three-dimensional knowledge base rule network, thereby improving the accuracy and robustness of identification, realizing the interpretability and traceability of the identification process, and greatly enhancing the credibility of the result.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This invention relates to the field of materials testing and information processing technology, specifically to a method and system for identifying plastic materials that combines spectroscopic features and physical property features, and more particularly to a method and system for identifying plastic materials that can output an interpretable chain of evidence. Background Technology

[0002] Plastic products are widely used in packaging, daily chemicals, electronics, electrical appliances, automobiles, and medical fields. Different plastic materials require significantly different treatment methods in recycling, reprocessing, and quality control. For example, polyethylene (PE), polypropylene (PP), polystyrene (PS), and polyethylene terephthalate (PET) differ in melting point, toughness, transparency, acoustic response, and spectral absorption characteristics. Failure to accurately identify the plastic material can easily lead to mixing, decreased performance of recycled materials, increased sorting costs, and reduced recycling value.

[0003] Existing plastic identification technologies are mainly divided into two categories: methods based on instrumental analysis and simple identification methods based on sensory experience.

[0004] Instrument-based analysis methods, particularly spectroscopic methods, have become a research hotspot due to their non-destructive and rapid nature. For example, Chinese patents CN121612832A ("Method for Identifying Plastic Materials Based on Near-Infrared Spectral Wavelength Ratio Reconstruction Features") and CN119880846A ("Method for Identifying Plastic Materials Based on the Slope of Near-Infrared Spectral Valleys") both focus on using specific mathematical features of near-infrared spectra (such as ratios and slopes) to distinguish plastics. While these methods improve machine recognition efficiency, their decision-making process is like a "black box," making it difficult for users to understand why the model makes a certain judgment and lacking interpretable evidence to support the recognition results. If the model misjudges, it not only affects the sorting accuracy but also makes it difficult to quickly pinpoint the cause of the problem.

[0005] On the other hand, in non-laboratory environments, operators often rely on traditional sensory experience methods (such as observing transparency, testing hardness, and listening to the sound when tapped) to roughly identify plastics. These methods are intuitive but highly dependent on personal experience, subjective, inconsistent, and unable to form digital, structured judgment records, making it difficult to meet the requirements of modern industry for process traceability and verifiable results. Chinese patents CN121457586A, entitled "Method, System, and Medium for Managing Material Knowledge Base Based on Autonomous Memory Evolution," and CN121641281A, entitled "Method, Apparatus, Equipment, System, Medium, and Product for Generating Material Knowledge Base," demonstrate advanced directions in material data management, but do not address how to combine this knowledge with simple, efficient on-site identification methods to generate easily understandable identification evidence.

[0006] Therefore, there is a need for a plastic identification scheme that balances low cost, interpretability, maintainability, and scalability. This scheme should integrate quantifiable standard wavelength absorbance with easily accessible sensory / mechanical / acoustic features, output probabilistic results through knowledge base and rule-based reasoning, and explain the identification conclusions with a clear logical chain, thereby improving identification stability and verification efficiency. Summary of the Invention

[0007] This invention aims to address the problems existing in the prior art by providing an interpretable method and system for identifying plastic materials. This method constructs a structured material knowledge base, combining standardized spectral features (such as absorbance at a specific wavelength) with a series of quantifiable or describable physical auxiliary features (such as transparency and hardness), and assigning rules and weights to different features to achieve rule-based reasoning. Ultimately, it not only outputs the identification result but also simultaneously generates an interpretable chain of evidence containing hit rules, conflict features, and confidence analysis, making the identification process transparent and the conclusions verifiable.

[0008] To achieve the above technical solution, the present invention provides an interpretable method for identifying plastic materials, specifically including the following steps: S1. Obtain the standard wavelength absorbance characteristics of the plastic sample to be tested; S2. Obtain auxiliary identification features of the plastic sample to be tested; S3. Read the material knowledge base, which is implemented using JSON, database tables or embedded configuration files. The material knowledge base includes at least candidate plastic material codes, standard wavelength absorbance range, absorbance center value, auxiliary feature rules, evidence text, feature weights and version information. S4. For each candidate plastic material, calculate the optical matching score, auxiliary feature matching score, and feature conflict penalty term. S5. Based on the number of input features and feature reliability, perform missing feature normalization and probability calibration on the scores of each candidate material to obtain the probability ranking of the candidate materials. S6. Based on the highest probability, the difference between the highest and second-highest probabilities, the distance outside the interval, and the degree of feature conflict, output the identification result, confidence level, unknown material hints, or re-examination suggestions. S7. Generate an interpretable chain of evidence from the hit rules, the missed rules, conflicting features, and recommended supplementary detection items.

[0009] Preferably, in step S1, the standard wavelength absorbance feature includes the absorbance at 550 nm, or a combination of the absorbance at 550 nm and the absorbance at one or more extended wavelengths. In practical applications, taking the absorbance at the visible light wavelength of 550 nm as the core or basic feature reduces the hardware cost and the dependence on complex spectral instruments, making the method easier to deploy and apply on site or in environments with limited conditions.

[0010] Preferably, in step S2, the auxiliary identification feature includes at least one of transparency level, hand feel hardness, bending toughness, bending white mark, fracture mode, tapping sound, surface gloss, sample thickness or color. By systematically defining a variety of intuitive and easily obtainable physical auxiliary features, the sensory experience of non-professionals can be digitally and structurally collected and utilized, which greatly expands the applicable scenarios and ease of use of the method.

[0011] Preferably, in step S4, the calculation process of the optical matching score is as follows: for the candidate material m, let its standard wavelength absorbance range be [L m , U m , and the central value be C m , the absorbance of the test sample is A, if L m ≤A≤U m , the optical matching score S opt (m)=1; if A<Lm or A>Um, attenuation is performed according to the distance from A to the range boundary, S opt (m)=exp(-d m / τ m ), where d m is the distance from A to the nearest boundary of the absorbance range of the material, and τ m is the attenuation coefficient of the material. In practical applications, the distance-based exponential attenuation function is used to calculate the optical matching score, which not only supports exact matching, but also allows "fuzzy matching" under multiple principles. Even if the optical characteristics of the sample are slightly offset, a reasonable score can be given, which improves the fault tolerance and generalization ability of the method.

[0012] Preferably, in step S4, the calculation formula of the auxiliary feature matching score is: S aux (m)=Σw i ·match(f i , r m,i ) / Σw i , where f i is the i-th input feature, r m,i is the corresponding rule of the candidate material m, and w i is the feature weight. The auxiliary feature matching score adopts a weighted average calculation method, which allows the knowledge base to dynamically adjust the weight according to the importance and reliability of the feature, making the identification logic more flexible and intelligent.

[0013] Preferably, in step S4, the feature conflict penalty term is determined based on the matching relationship between the auxiliary identification features of the plastic sample to be tested and the rejection rules of the candidate plastic materials. In practical applications, by introducing a feature conflict penalty term, when the input feature clearly violates the inherent properties of a certain material, the score of that material will be significantly reduced. This effectively prevents misjudgments that violate basic common sense and reflects the constraint and guidance role of physical knowledge on the data-driven model.

[0014] Preferably, in step S5, the specific process of comprehensive scoring and probability ranking is as follows: S51. Calculate the comprehensive score S(m) of candidate material m: The comprehensive score S(m) consists of optical matching score, auxiliary feature matching score, conflict penalty term, and reliability coefficient. The specific calculation formula is as follows: S(m) = α·S opt (m)+β·S aux (m)-γ·P conflict (m)+δ·R quality Among them, α, β, γ, and δ are weights; P conflict (m) represents the feature conflict penalty term; R quality This is the data quality coefficient, used to indicate whether the absorbance measurement is stable, whether the input features are complete, and whether the sample is contaminated or has color interference. S52. Normalize the scores of all candidate materials to obtain a probability ranking: P(m) = S(m) / ΣS(k), or use the Softmax function for probability calibration. The output includes top results. material The criteria are: confidence, results, and evidence. Results is a list of candidate materials, and evidence is the rules and conflict descriptions for each candidate material.

[0015] In the above scheme, the comprehensive scoring mechanism integrates multiple scoring factors and data quality coefficients, and outputs a probability distribution through normalization or a softmax function. This not only provides an intuitive confidence index, but its structured output (such as results and evidence lists) also provides a convenient data interface for subsequent system integration and result display.

[0016] Preferably, in step S6, when the highest probability is lower than a first threshold, the difference between the highest and second-highest probabilities is lower than a second threshold, or the distance between the absorbance feature and the absorbance range of all candidate plastic materials exceeds a third threshold, a prompt indicating unknown material, suspected composite material, or need for supplementary testing is output. In practical applications, by setting multi-dimensional rejection judgment logic (such as probability threshold, difference threshold, and distance threshold), it is possible to proactively prompt "unknown material" or "re-inspection required" when the confidence level is insufficient, avoiding the drawbacks of traditional methods that forcibly give incorrect results, and improving the reliability and practicality of the system.

[0017] Preferably, in step S6, when the probabilities of candidate materials are close, the next detection feature is automatically recommended based on the rule with the highest discriminative power among the candidate materials. The recommendation strategy is based on information gain or entropy reduction calculation. Specifically, let the current candidate probability distribution be P, and the expected entropy reduction brought by a certain supplementary feature g be IG(g). Select the feature with the largest IG(g) as the next detection suggestion. When the recognition result is ambiguous, the system can proactively recommend the next detection feature with the highest information gain, guiding the user to perform supplementary detection. This achieves interactive recognition through human-machine collaboration, which can effectively reduce uncertainty and improve the final recognition efficiency.

[0018] This invention also provides an interpretable plastic material identification system for implementing the above-mentioned identification method, specifically including: The front-end input module is used to acquire the standard wavelength absorbance characteristics and auxiliary identification characteristics of the plastic sample to be tested. The materials knowledge base module is used to store the absorbance range, auxiliary feature rules, feature weights, and evidence text of candidate plastic materials; Calculation module: used to calculate the optical matching score, auxiliary feature matching score, and feature conflict penalty term for candidate plastic materials respectively; The rule-based reasoning module is used to calculate the comprehensive score of each candidate plastic material based on the absorbance characteristics and auxiliary identification characteristics. The probability ranking module is used to generate a probability ranking and confidence level of candidate plastic materials based on the comprehensive score; The interpretation generation module is used to generate the judgment criteria corresponding to the recognition results; The output module is used to output candidate plastic materials, confidence levels, judgment criteria, and review recommendations. The feedback calibration module is used to receive manual review results or laboratory test results, and to update the material knowledge base based on the manual review results or laboratory test results. The Unknown Material Rejection Module is used to output prompts indicating unknown material, suspected composite material, or material requiring supplementary testing when the probability of a candidate material does not meet the preset confidence conditions.

[0019] The beneficial effects of the explanatory plastic material identification method and system provided by this invention are as follows: (1) By integrating spectral features with quantifiable physical auxiliary features and constructing a three-dimensional knowledge base rule network, this invention not only improves the accuracy and robustness of identification, but also achieves interpretability and traceability of the identification process. Through the interpretable evidence chain generated in step S7, the abstract matching score is transformed into a specific rule hit and conflict description, enabling operators to clearly understand the logical basis of the identification conclusion, which greatly enhances the credibility and auditability of the results.

[0020] (2) This invention uses quantifiable optical data such as 550nm absorbance as strong evidence and artificially collectable features such as transparency, hardness, bending, and sound as auxiliary evidence to achieve low-cost identification; it avoids rigid misjudgment caused by simple table lookup by attenuation of distance outside the interval and probability ranking; it allows users to obtain usable probability results when features are incomplete by normalizing missing features; it makes the identification process interpretable by the evidence chain output, which is convenient for teaching, quality inspection and verification and result traceability; it reduces the risk of materials not covered by the knowledge base being misjudged as known materials by the unknown material rejection mechanism; and it guides users to improve confidence with less supplementary detection by the next detection feature recommendation mechanism. Attached Figure Description

[0021] Figure 1 This is a flowchart of the method of the present invention.

[0022] Figure 2 This is a system module diagram of the present invention. Detailed Implementation

[0023] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art without creative effort are within the protection scope of the present invention.

[0024] Example 1: An interpretable plastic material identification system.

[0025] Reference Figure 2 As shown, an interpretable plastic material identification system specifically includes: Front-end input module 1 is used to acquire the standard wavelength absorbance characteristics and auxiliary identification characteristics of the plastic sample to be tested; Material knowledge base module 2 is used to store the absorbance range, auxiliary feature rules, feature weights and evidence text of candidate plastic materials; Calculation module 3: Used to calculate the optical matching score, auxiliary feature matching score, and feature conflict penalty term for candidate plastic materials respectively; Rule reasoning module 4 is used to calculate the comprehensive score of each candidate plastic material based on the absorbance characteristics and auxiliary identification characteristics; Probability ranking module 5 is used to generate a probability ranking and confidence level of candidate plastic materials based on the comprehensive score; The interpretation and generation module 6 is used to generate the judgment criteria corresponding to the recognition results; Output module 7 is used to output candidate plastic materials, confidence levels, judgment criteria, and review recommendations; Feedback calibration module 8 is used to receive manual review results or laboratory test results, and to update the material knowledge base based on the manual review results or laboratory test results. The Unknown Material Rejection Module 9 is used to output a prompt indicating that the candidate material is unknown, a suspected composite material, or requires supplementary testing when the probability of the candidate material does not meet the preset confidence conditions.

[0026] This system is deployed on local computers, intranet servers, cloud servers, or edge terminals to run web services. The feedback calibration module 7 enables continuous evolution and version management of system knowledge, while the unknown material rejection module 8 ensures the system's robustness. The entire system forms a closed loop integrating data acquisition, knowledge reasoning, result interpretation, and self-optimization, greatly satisfying the requirements of industrial applications for environmental adaptability and long-term reliability.

[0027] Example 2: An interpretable method for identifying plastic materials.

[0028] Reference Figure 1 and Figure 2 As shown, an interpretable plastic material identification method, based on the identification system in Example 1, specifically includes the following steps: S1. Front-end input module 1 acquires the standard wavelength absorbance characteristics of the plastic sample to be tested; the standard wavelength absorbance characteristics include 550nm absorbance, or a combination of 550nm absorbance and one or more extended wavelength absorbances. In practical applications, using 550nm visible light wavelength absorbance as the core or basic characteristic reduces hardware costs and dependence on complex spectroscopic instruments, making the method easier to deploy and apply in field or under limited conditions.

[0029] S2. Front-end input module 1 acquires auxiliary identification features of the plastic sample to be tested; auxiliary identification features include at least one of transparency level, hand hardness, bending toughness, bending white marks, fracture mode, tapping sound, surface gloss, sample thickness, or color. By systematically defining a variety of intuitive and easily obtainable physical auxiliary features, the sensory experience of non-professionals can be digitally and structurally collected and utilized, greatly expanding the applicable scenarios and ease of use of this method.

[0030] S3. Read the material knowledge base from module 2. The material knowledge base can be implemented using JSON, database tables, or embedded configuration files. Taking PE, PP, PS, and PET as examples, the knowledge base can include the following standard data:

[0031] The knowledge base can also configure evidence text for each rule, such as "550nm absorbance falls within the PET standard range" and "The sample does not break after bending and produces a deep, solid sound, which meets the auxiliary characteristics of PET." When the material range or new materials need to be adjusted, only the knowledge base needs to be updated; the scoring program does not need to be rewritten.

[0032] The knowledge base also includes version number, data source, applicable sample thickness, applicable measuring instrument, calibration date, and administrator signature to ensure the traceability of identification results.

[0033] S4. Calculation module 3 calculates the optical matching score, auxiliary feature matching score, and feature conflict penalty for each candidate plastic material.

[0034] The calculation process for the optical matching score is as follows: For candidate material m, let its standard wavelength absorbance range be [L]. m U m The center value is C. m The absorbance of the sample to be tested is A. When A is located in [L... m U m Within the time frame, the optical matching score S opt (m) represents a high score; when A is outside the range, the score is reduced based on the distance from A to the range boundary, rather than being directly determined as 0 points. S opt (m)=1, if L m ≤A≤U m ; S opt (m)=exp(-d m / τ m If A <L m or A>U m .

[0035] Where d m τ is the distance from A to the nearest boundary of the material's absorbance range. m This represents the attenuation coefficient of the material. Distance attenuation can accommodate measurement errors, sample thickness variations, surface contamination, and batch fluctuations.

[0036] When the system uses multi-wavelength extension, the 550nm absorbance can be combined with the absorbance of near-infrared, short-wave infrared or other visible light bands to form a vector, and the optical matching score can be calculated using weighted distance, cosine similarity, spectral angle mapping or lightweight neural network.

[0037] In practical applications, the optical matching score is calculated using a distance-based exponential decay function. This not only supports precise matching but also allows for "fuzzy matching" under various principles. Even if the optical properties of the sample are slightly off, a reasonable score can still be given, which improves the method's fault tolerance and generalization ability.

[0038] Auxiliary feature scoring is used to supplement the insufficient information from a single absorbance value. The system converts inputs such as transparency, hardness, flexural toughness, and sound into standardized enumeration values ​​or grade values, and matches them with rules in the candidate material knowledge base. For example, transparency can be categorized as transparent, relatively transparent, translucent, and opaque; hardness can be categorized as soft, medium, stiff, and brittle; flexural toughness can be categorized as easily bent, bendable but difficult, breaks easily when bent, and does not break after repeated bending; sound can be categorized as deep, mid-range, crisp, and metallic. Each feature that matches a candidate material rule adds a corresponding evidence score, while obvious conflicts are penalized.

[0039] The auxiliary feature matching score can be expressed as: S aux (m)=Σw i ·match(f i , r m,i ) / Σw i , Where f i For the i-th input feature, r m,i For candidate material m, the corresponding rule is w i The weights are used for features. Features not input are not included in the denominator calculation, thus supporting incomplete feature input. The auxiliary feature matching score uses a weighted average calculation method, allowing the knowledge base to dynamically adjust the weights based on the importance and reliability of the features, making the recognition logic more flexible and intelligent.

[0040] The feature conflict penalty term is determined based on the matching relationship between the auxiliary identification features of the plastic sample to be tested and the rejection rules of the candidate plastic materials. In practical applications, by introducing the feature conflict penalty term, when the input features clearly violate the inherent properties of a certain material, the score of that material will be significantly reduced. This effectively prevents misjudgments that violate basic common sense, reflecting the constraint and guidance role of physical knowledge on the data-driven model.

[0041] S5, rule reasoning module 4, and probability ranking module 5 perform missing feature normalization and probability calibration on the scores of each candidate material based on the number of input features and feature reliability, and obtain the probability ranking of the candidate materials.

[0042] The specific process of comprehensive scoring and probability ranking is as follows: S51. Calculate the comprehensive score S(m) of candidate material m: The comprehensive score S(m) consists of optical matching score, auxiliary feature matching score, conflict penalty term, and reliability coefficient. The specific calculation formula is as follows: S(m) = α·S opt (m)+β·S aux (m)-γ·P conflict (m)+δ·R quality Among them, α, β, γ, and δ are weights; P conflict (m) represents the feature conflict penalty term; R quality This is the data quality coefficient, used to indicate whether the absorbance measurement is stable, whether the input features are complete, and whether the sample is contaminated or has color interference. S52. Normalize the scores of all candidate materials to obtain a probability ranking: P(m) = S(m) / ΣS(k), or use the Softmax function for probability calibration. The output includes top results. material The criteria are: confidence, results, and evidence. Results is a list of candidate materials, and evidence is the rules and conflict descriptions for each candidate material.

[0043] The comprehensive scoring mechanism integrates multiple scoring factors and data quality coefficients, and outputs a probability distribution through normalization or a softmax function. This not only provides an intuitive confidence index, but its structured output (such as results and evidence lists) also provides a convenient data interface for subsequent system integration and result presentation.

[0044] S6 and output module 7 output the recognition result, confidence level, unknown material hints or re-inspection suggestions based on the highest probability, the difference between the highest and second highest probabilities, the distance outside the interval, and the degree of feature conflict.

[0045] Specifically, when the highest probability is below a first threshold, the difference between the highest and second-highest probabilities is below a second threshold, or the distance between the absorbance feature and the absorbance range of all candidate plastic materials exceeds a third threshold, an output indicating unknown material, suspected composite material, or requiring supplementary testing is provided. This mechanism avoids the problem of traditional classification systems issuing high-confidence erroneous conclusions when the knowledge base does not cover PVC, ABS, PC, PA, composite films, filled modified plastics, or severely contaminated samples.

[0046] In practical applications, by setting multi-dimensional rejection judgment logic (such as probability threshold, difference threshold, and distance threshold), it can proactively prompt "unknown material" or "re-inspection required" when the confidence level of recognition is insufficient, avoiding the drawbacks of traditional methods that forcibly give incorrect results and improving the reliability and practicality of the system.

[0047] When the probabilities of candidate materials are similar, the system automatically recommends the next detection feature based on the rule that best distinguishes between the candidate materials. For example, when the probabilities of PE and PP are similar, it recommends adding features such as white marks from bending, hardness, or density; when the probabilities of PET and PS are similar, it recommends adding features such as bending fracture mode and tapping sound; when all candidate materials do not match the 550nm absorbance, it recommends adding near-infrared or multi-wavelength absorbance detection.

[0048] Recommendation strategies can be calculated based on information gain or entropy reduction. Let the current candidate probability distribution be P, and the expected entropy reduction that a certain supplementary feature g may bring be IG(g). The system selects the feature with the largest IG(g) as the next detection suggestion, thereby improving the recognition confidence with the least amount of supplementation operations.

[0049] When the recognition result is ambiguous, the system can proactively recommend the next detection feature with the highest information gain, guiding the user to perform supplementary detection. This achieves interactive recognition through human-machine collaboration, which can effectively reduce uncertainty and improve the final recognition efficiency.

[0050] In actual operation, the system receives manual review results, laboratory test results, or subsequent production verification results, and uses them as feedback samples. For confirmed samples, the system can update the material's mean absorbance, range, auxiliary feature weights, and conflict rules, but retains historical versions before updating to avoid data overlap between different batches. The system maintains calibration parameters separately according to material code, sample thickness, color category, and measuring instrument model. For example, for dark plastic or surface-contaminated samples, the system can reduce the weights of color and transparency features, and increase the weights of bending, sound, or multi-wavelength spectral features.

[0051] S7. The explanation generation module 6 generates an interpretable chain of evidence from the hit rules, the missed rules, conflicting features, and recommended supplementary detection items.

[0052] To better explain the present invention, the following example illustrates its implementation process in practical applications: (1) Identification of PE, PP, PS and PET based on 550nm absorbance Users input the 550nm absorbance, transparency level, tactile hardness, flexural toughness, and sound characteristics of the sample to be tested on the web page. The system reads the materials knowledge base in materials.json and calculates a comprehensive score for PE, PP, PS, and PET respectively.

[0053] If the absorbance of a sample at 550nm is 0.045, the system determines that it falls within the PE absorbance range of 0.042 to 0.050 and assigns a high optical score to PE. At the same time, if the user inputs "semi-transparent", "soft", "white marks after multiple bends", and "deep sound", all of the above auxiliary features match the PE rules, the system outputs that PE is the Top 1 material and displays the evidence that "the 550nm absorbance falls within the PE standard range", "the white marks after bending match the PE characteristics", and "the deep sound matches the PE characteristics".

[0054] (2) Interval distance attenuation identification If the sample's absorbance at 550 nm is 0.052, slightly higher than the upper limit of the PE standard range of 0.050, the system does not directly exclude PE. Instead, it calculates the distance from the absorbance to the PE range boundary and applies attenuation. If the auxiliary characteristics still clearly conform to PE, the system can output "PE, high confidence or medium confidence" and prompt "Absorbance slightly higher than the standard range, it is recommended to retest or check sample thickness / contamination." This method can accommodate measurement errors and differences in sample condition.

[0055] (3) Rejection of unknown materials If the sample's 550nm absorbance is significantly different from the ranges of PE, PP, PS, and PET, and the auxiliary features conflict with the rules for each candidate material, the system outputs "Unknown material or material not covered by the knowledge base," and recommends supplementing with near-infrared multi-wavelength detection, density detection, or manual verification. This approach avoids the system forcing incorrect judgments in order to arrive at a conclusion.

[0056] (4) Recommendation of the next detection feature If the scores for PE and PP are similar, the system analyzes the differences in their rules and finds that bending marks and hardness contribute significantly to distinguishing them. The system then prompts the user to provide additional information on "whether multiple bends produce white marks" and "the degree of hardness felt." After the user provides this information, the system recalculates the probabilities and updates the chain of evidence.

[0057] (5) Multiwavelength extension In the low-cost version, the system uses 550nm absorbance as the primary quantization input; in the enhanced version, the system can connect to multi-wavelength LEDs, near-infrared sensors, or short-wave infrared sensors to form an absorbance vector. The system retains its knowledge base and interpretable rule structure, fusing multi-wavelength optical matching results with auxiliary feature scoring, making it suitable for more complex recycling and sorting scenarios.

[0058] (6) Experimental design and evaluation indicators It can collect samples of four types: PE, PP, PS, and PET, with no fewer than 100 samples of each type, covering transparent, semi-transparent, different thicknesses, different colors, and slightly contaminated states. For each sample, the absorbance at 550nm, transparency, hardness, bending toughness, and sound characteristics are recorded, and the authenticity of the material is confirmed by manual verification or laboratory methods.

[0059]

[0060] The above indicators are example indicators in the application documents. They can be replaced with real data based on the actual sample test results before formal submission.

[0061] This invention, by fusing spectral features with quantifiable physical auxiliary features and constructing a three-dimensional knowledge base rule network, not only improves the accuracy and robustness of identification but also achieves interpretability and traceability of the identification process. The interpretable evidence chain generated in step S7 transforms abstract matching scores into concrete rule hit and conflict descriptions, enabling operators to clearly understand the logical basis of the identification conclusions and greatly enhancing the credibility and auditability of the results. Furthermore, this invention uses quantifiable optical data such as 550nm absorbance as strong evidence and manually collectable features such as transparency, hardness, bending, and sound as auxiliary evidence, achieving low-cost identification. Through inter-interval distance attenuation and probability ranking, it avoids rigid misjudgments caused by simple table lookups. By normalizing missing features, it allows users to obtain usable probability results even when features are incomplete. The evidence chain output makes the identification process interpretable, facilitating teaching, quality inspection review, and result traceability. The unknown material rejection mechanism reduces the risk of materials not covered by the knowledge base being misclassified as known materials. The next detection feature recommendation mechanism guides users to improve confidence with fewer supplementary detections.

[0062] The above description is only a preferred embodiment of the present invention, but the present invention should not be limited to the content disclosed in the embodiments and drawings. Therefore, any equivalent or modified embodiments made without departing from the spirit of the present invention shall fall within the protection scope of the present invention.

Claims

1. A method for interpretable plastic material identification, characterized in that... Specifically, the steps include the following: S1. Obtain the standard wavelength absorbance characteristics of the plastic sample to be tested; S2. Obtain auxiliary identification features of the plastic sample to be tested; S3. Read the material knowledge base, which is implemented using JSON, database tables or embedded configuration files. The material knowledge base includes at least candidate plastic material codes, standard wavelength absorbance range, absorbance center value, auxiliary feature rules, evidence text, feature weights and version information. S4. For each candidate plastic material, calculate the optical matching score, auxiliary feature matching score, and feature conflict penalty term. S5. Based on the number of input features and feature reliability, perform missing feature normalization and probability calibration on the scores of each candidate material to obtain the probability ranking of the candidate materials. S6. Based on the highest probability, the difference between the highest and second-highest probabilities, the distance outside the interval, and the degree of feature conflict, output the identification result, confidence level, unknown material hints, or re-examination suggestions. S7. Generate an interpretable chain of evidence from the hit rules, the missed rules, conflicting features, and recommended supplementary detection items.

2. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S1, the standard wavelength absorbance characteristic includes 550 nm absorbance, or a combination of 550 nm absorbance and one or more extended wavelength absorbances.

3. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S2, the auxiliary identification features include at least one of the following: transparency level, tactile hardness, bending toughness, bending white marks, fracture mode, tapping sound, surface gloss, sample thickness, or color.

4. The interpretable plastic material identification method as described in claim 1, characterized in that, In the described step S4, the calculation process of the optical matching score is as follows: For candidate material m, let its standard wavelength absorbance range be [L m , U m , and the center value is C m , the absorbance of the test sample is A. If L m ≤A≤U m , the optical matching score S opt (m)=1; if A<Lm or A>Um, attenuation is performed according to the distance from A to the range boundary, S opt (m)=exp(-d m / τ m ), where d m is the distance from A to the nearest boundary of the material absorbance range, and τ m is the attenuation coefficient of the material.

5. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S4, the formula for calculating the auxiliary feature matching score is: S aux (m)=Σw i ·match(f i , r m,i ) / Σw i , where f i For the i-th input feature, r m,i For candidate material m, the corresponding rule is w i These are the feature weights.

6. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S4, the feature conflict penalty term is determined based on the matching relationship between the auxiliary identification features of the plastic sample to be tested and the rejection rules of the candidate plastic materials.

7. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S5, the specific process of comprehensive scoring and probability ranking is as follows: S51. Calculate the comprehensive score S(m) of candidate material m: The comprehensive score S(m) consists of optical matching score, auxiliary feature matching score, conflict penalty term, and reliability coefficient. The specific calculation formula is as follows: S(m)=α·S opt (m)+β·S aux (m)-γ·P conflict (m)+δ·R quality Among them, α, β, γ, and δ are weights; P conflict (m) represents the feature conflict penalty term; R quality This is the data quality coefficient, used to indicate whether the absorbance measurement is stable, whether the input features are complete, and whether the sample is contaminated or has color interference. S52. Normalize the scores of all candidate materials to obtain a probability ranking: P(m) = S(m) / ΣS(k), or use the Softmax function for probability calibration. The output includes top results. material The criteria are: confidence, results, and evidence. Results is a list of candidate materials, and evidence is the rules and conflict descriptions for each candidate material.

8. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S6, when the highest probability is lower than the first threshold, the difference between the highest probability and the second highest probability is lower than the second threshold, or the distance between the absorbance feature and the absorbance range of all candidate plastic materials exceeds the third threshold, an output prompt indicating unknown material, suspected composite material, or need for supplementary testing is provided.

9. The interpretable plastic material identification method as described in claim 1, characterized in that, In step S6, when the probabilities of candidate materials are close, the next detection feature is automatically recommended according to the rule with the highest distinguishability among the candidate materials. The recommendation strategy is based on information gain or entropy reduction calculation. The specific process is as follows: Let the current candidate probability distribution be P, and the expected entropy reduction brought by a certain feature to be supplemented g be IG(g). Select the feature with the largest IG(g) as the next detection suggestion.

10. An interpretable plastic material identification system, characterized in that, To implement the identification method according to any one of claims 1-9, specifically comprising: The front-end input module is used to acquire the standard wavelength absorbance characteristics and auxiliary identification characteristics of the plastic sample to be tested. The materials knowledge base module is used to store the absorbance range, auxiliary feature rules, feature weights, and evidence text of candidate plastic materials; Calculation module: used to calculate the optical matching score, auxiliary feature matching score, and feature conflict penalty term for candidate plastic materials respectively; The rule-based reasoning module is used to calculate the comprehensive score of each candidate plastic material based on the absorbance characteristics and auxiliary identification characteristics. The probability ranking module is used to generate a probability ranking and confidence level of candidate plastic materials based on the comprehensive score; The interpretation generation module is used to generate the judgment criteria corresponding to the recognition results; The output module is used to output candidate plastic materials, confidence levels, judgment criteria, and review recommendations. The feedback calibration module is used to receive manual review results or laboratory test results, and to update the material knowledge base based on the manual review results or laboratory test results. The Unknown Material Rejection Module is used to output prompts indicating unknown material, suspected composite material, or material requiring supplementary testing when the probability of a candidate material does not meet the preset confidence conditions.

Citation Information

Patent Citations

  • Plastic material identification method based on slopes of two sides of wave trough of near infrared spectrum

    CN119880846A

  • Material knowledge base management method and system based on autonomous memory evolution and medium

    CN121457586A

  • Plastic material identification method based on near infrared spectrum wavelength ratio reconstruction characteristics

    CN121612832A

  • Method, device, equipment and system for generating material knowledge base, medium and product

    CN121641281A