A method and device for quantitatively detecting recycled plastic based on intelligent sensing, and a medium

Abstract

The application discloses a kind of based on intelligent sensing's recycled plastic quantitative detection method, equipment and medium, it is related to plastic material quantitative detection field, comprising: obtaining the sample information and detection environment information of the plastic sample to be detected, generates this batch threshold group and mixed proportion response relationship;The plastic sample to be detected is divided into detection area, forms area binding record;Progressive state determination is carried out to area binding record based on threshold group, forms area state combination;According to area state combination, spectrum dominant calculation, thermal response compensation calculation, thickness linkage correction calculation or surface anomaly review calculation are selected, and initial detection value of sample recycled plastic content is generated;Initial detection value is checked by area dispersion, spectrum and thermal response direction verification, state label back check and calibration spline back calculation, and recycled plastic content quantitative detection result is generated.The application realizes the associated detection of area state determination, branch quantitative calculation and reverse check.

Description

Technical Field

[0001] This invention relates to the field of quantitative detection of plastic materials, and in particular to a method, equipment and medium for quantitative detection of recycled plastics based on intelligent sensing. Background Technology

[0002] The detection of recycled plastic content falls under the technical field of plastic material composition analysis and quality evaluation, typically involving plastic type identification, material aging characteristic detection, reprocessing trace analysis, and blending ratio estimation. In practical testing, common methods include near-infrared spectroscopy, thermal analysis, image recognition, and density or ash content-assisted detection. By collecting spectral absorption, thermal response, color state, appearance characteristics, or physical properties of plastic samples, the origin, material composition, and blending of recycled materials can be determined. With the increasing application of recycled plastics in packaging, building materials, automotive parts, and daily consumer goods, the testing process is gradually evolving from single-indicator judgment to comprehensive analysis of multi-source testing data to meet the quantitative testing needs of plastic samples with different forms, colors, and processing histories.

[0003] However, existing methods still have two limitations: First, some detection methods mainly rely on a single spectrum, a single thermal response, or manually selected empirical features. When faced with dark, mixed-color, transparent, unevenly thick, or surface-contaminated samples, the detection signal is easily affected by color, thickness, and surface condition, resulting in insufficient stability of the quantitative results of recycled plastic content. Second, multi-source detection data are usually collected in parallel and simply fused, lacking a linkage mechanism that triggers compensation, correction, and reverse verification based on the regional state, making it difficult to effectively distinguish abnormal detection areas and sensor source conflicts. Summary of the Invention

[0004] In view of the aforementioned existing problems, the present invention is proposed.

[0005] Therefore, this invention provides a method for quantitative detection of recycled plastics based on intelligent sensing to solve the problem of unstable quantitative results of recycled plastic content caused by interference from color, thickness and surface abnormalities in the multi-source detection results of existing technologies.

[0006] To solve the above-mentioned technical problems, the present invention provides the following technical solution: In a first aspect, the present invention provides a method for quantitative detection of recycled plastics based on intelligent sensing, comprising: acquiring sample information and detection environment information of the plastic sample to be tested; confirming the type of plastic in the plastic sample to be tested; generating a threshold group and a blending ratio response relationship corresponding to the current batch of tests based on a virgin plastic reference sample of the same type of plastic, a calibration sample with a known blending ratio of recycled plastics, an unloaded sensing response, and repeated detection results; dividing the plastic sample to be tested into detection areas; collecting spectral response data, image state data, thermal response data, and geometric state data of each detection area; binding the data according to the detection area number to form an area binding record; performing progressive state determination on the area binding record based on the threshold group to form an area state combination; selecting the corresponding quantitative calculation path according to the area state combination; performing spectral-dominant calculation, thermal response compensation calculation, thickness linkage correction calculation, or surface anomaly verification calculation on the effective detection area; outputting a predicted value of the recycled plastic content in the area; generating an initial detection value of the recycled plastic content of the sample based on the predicted value of the area after area consistency screening; performing reverse verification on the initial detection value of the recycled plastic content of the sample; and generating a quantitative detection result of the recycled plastic content and a corresponding detection record based on the verification result.

[0007] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics described in this invention, the generation of the threshold group and blending ratio response relationship corresponding to the current batch of detection includes: confirming the type of plastic in the plastic sample to be detected based on the sample source record and the near-infrared spectral resin characteristic band; selecting or preparing a virgin plastic reference strip and a known recycled plastic blending ratio calibration strip belonging to the same type of plastic as the plastic sample to be detected; repeatedly detecting the virgin plastic reference strip, the known recycled plastic blending ratio calibration strip, and the no-load detection state to obtain the no-load noise response, the virgin plastic repeated response, the calibration response for different blending ratios, and the sensor repeated detection error; according to The permissible detection error is determined by the target accuracy of the detection method, the calibration strip preparation error, and the sensor repeatability error, and a threshold group is generated based on the permissible detection error. A blending ratio response relationship is established based on the detection characteristics of the calibration strips according to known recycled plastic blending ratios and the corresponding known recycled plastic blending ratios. When a single detection feature exhibits monotonic variation under each known recycled plastic blending ratio, a piecewise interpolation relationship is used to determine the blending ratio response relationship. When at least two detection features participate in the quantitative calculation, a partial least squares regression relationship is used to determine the blending ratio response relationship. In the piecewise interpolation relationship, a local linear fitting relationship is used between adjacent calibration points to determine the response change within the corresponding ratio interval.

[0008] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics according to the present invention, the threshold group includes: a coverage ratio threshold, a spectral signal-to-noise ratio threshold, a color dispersion threshold, a thickness adaptation threshold, a surface anomaly area threshold, an anomaly rejection upper limit, a thermal response compensation trigger condition, a region dispersion threshold, a spectral and thermal response direction conflict threshold, and a calibration strip back-calculation allowable error; the coverage ratio threshold is determined based on the sensor's effective field of view area, the spectral sampling spot area, and the spectral response stability corresponding to changes in sample coverage area; the spectral signal-to-noise ratio threshold is determined based on the no-load noise response, the fluctuation of repeated detection of the virgin plastic reference strip, and the minimum spectral response difference between adjacent blending ratio calibration strips; The color dispersion threshold is determined based on the spectral quantitative deviation caused by calibration strips with the same recycled plastic blending ratio but different color states; the thickness adaptation threshold is determined based on the spectral and thermal response changes of calibration strips under different thickness conditions; the surface anomaly area threshold is determined based on the spectral, image, and thermal response deviations caused by calibration strips of the same material with known surface anomalies; the anomaly rejection upper limit is determined based on the proportion of anomaly area when surface anomalies cause simultaneous distortion of spectral and thermal responses and the calibration strip back-calculation deviation exceeds the allowable error of calibration strip back-calculation, and the anomaly rejection upper limit is greater than the surface anomaly area threshold but not greater than 1; the allowable error of calibration strip back-calculation is set correspondingly to the allowable error of detection.

[0009] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics described in this invention, the step of binding the detection area according to the detection area number to form an area binding record includes: dividing the plastic sample to be tested into detection areas according to the effective detection area of ​​the sample, the field of view area of ​​a single sensor, and the minimum number of detection areas; collecting spectral response data of the detection areas through a near-infrared spectral sensing unit, collecting image state data of the detection areas through a visible light image sensing unit, collecting thermal response data of the detection areas through a thermal response sensing unit, and collecting geometric state data of the detection areas through a thickness detection unit; binding the spectral response data, image state data, thermal response data, and geometric state data of the same detection area according to the detection area number; the minimum number of detection areas is determined by the results of repeated detection of multiple areas of the calibration strip. When the number of detection areas is increased, if the change in the predicted value of recycled plastic content of the calibration strip does not exceed the allowable error of the calibration strip back calculation, the corresponding number of detection areas is taken as the minimum number of detection areas.

[0010] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics described in this invention, the progressive state determination of the region-bound records based on threshold groups includes: calculating the sample coverage ratio of the detection area based on image state data, comparing the sample coverage ratio with a coverage ratio threshold to determine an effective coverage area or an insufficient coverage area; for an effective coverage area, calculating the spectral signal-to-noise ratio based on spectral response data, comparing the spectral signal-to-noise ratio with a spectral signal-to-noise ratio threshold to determine a spectrally usable area or a spectrally low confidence area; calculating the color discrete value based on image state data, comparing the color discrete value with a color discrete threshold to determine a color interference area; calculating the thickness deviation value based on geometric state data, comparing the thickness deviation value with a thickness adaptation threshold to determine a thickness correction area or a thickness rejection area; calculating the surface abnormal area ratio based on image state data, comparing the surface abnormal area ratio with a surface abnormal area threshold and an abnormal rejection upper limit; when the surface abnormal area ratio reaches the surface abnormal area threshold but does not exceed the abnormal rejection upper limit, it is determined as a surface abnormality verification area; when the surface abnormal area ratio exceeds the abnormal rejection upper limit, it is determined as a surface rejection area.

[0011] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics described in this invention, the progressive state determination of the region-bound records based on threshold groups includes: the color discrete value is obtained by normalizing and weighting the mean difference of color channels, the standard deviation of brightness, and the discrete term of the main color category within the detection area; the main color category is determined by clustering the hue, saturation, and brightness distribution of pixels in the HSV color space of the detection area; the discrete term of the main color category is one minus the area ratio of the main color category; the thickness deviation value is determined according to the degree of deviation of the thickness of the detection area from the effective thickness range of similar calibration strips; when the thickness of the detection area is within the effective thickness range, the thickness deviation value is zero; when the thickness of the detection area is lower than the lower limit of the effective thickness range or higher than the upper limit of the effective thickness range, the thickness deviation value is the difference between the corresponding boundary and the thickness of the detection area.

[0012] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics described in this invention, the step of generating the initial detection value of the recycled plastic content of the sample based on the predicted value of the region after regional consistency screening includes: when the detection region is determined to be a spectrally usable region and not a color interference region, the spectral-dominant calculation is entered, and the predicted value of the recycled plastic content of the region is calculated based on spectral attenuation characteristics, oxidative aging characteristics, and additive residue characteristics; when the detection region is determined to be a color interference region or a spectrally low confidence region, it is determined whether the detection region meets the thermal response compensation trigger condition. If it does, the thermal response compensation calculation is entered, and the color-corrected spectral features and thermal response features are jointly input into the mixing ratio response relationship to calculate the predicted value of the recycled plastic content of the region; when the detection region is determined to be a thickness correction region, before entering the spectral-dominant calculation or thermal response compensation calculation, thickness linkage correction is performed on the spectral response and thermal response; when the detection region is determined to be a surface anomaly verification region, the sampling points corresponding to the abnormal pixels are removed, and the spectral signal-to-noise ratio, color discrete value, and surface anomaly area ratio are recalculated, and the spectral-dominant calculation or thermal response compensation calculation is entered based on the recalculation results.

[0013] As a preferred embodiment of the intelligent sensing-based quantitative detection method for recycled plastics described in this invention, the reverse verification of the initial detection value of the recycled plastic content in the sample includes: calculating a regional dispersion index based on the predicted value of the recycled plastic content in the region involved in the calculation of the initial detection value of the recycled plastic content in the sample, and comparing the regional dispersion index with a regional dispersion threshold; determining outlier candidate regions based on the natural dispersion range formed by repeated detection of multiple regions of the calibration strip, and determining the initial detection value of the recycled plastic content in the sample according to the proportion of outlier candidate regions; and performing spectral and thermal response direction verification on the detection region that simultaneously possesses spectral features and thermal response features, and generating a sensor source conflict record when the deviation between the spectral feature change direction and the thermal response feature change direction exceeds the spectral and thermal response direction conflict threshold. Based on the combination of regional states of discrete contribution areas or conflicting sensor source areas, a status label backcheck is performed. Thermal response compensation calculation, thickness linkage correction calculation, or surface anomaly verification calculation are retried based on the regional state combination. The number of status label backchecks does not exceed the preset backcheck limit. If discrete contribution areas or conflicting sensor source areas still exist after reaching the backcheck limit, the corresponding detection area is marked as a low-confidence area or removed, and is not included in the calculation of the initial detection value of the recycled plastic content in the sample. A calibration strip with a known recycled plastic blending ratio is input into the same detection process for calibration strip backcheck. When the calibration strip backcheck deviation does not exceed the allowable error of the calibration strip backcheck, the quantitative detection result of the recycled plastic content and the corresponding detection record are output. When the calibration strip backcheck deviation exceeds the allowable error of the calibration strip backcheck, a detection failure record for this batch is generated, and a recalibration or re-detection prompt is output; the quantitative detection result of the recycled plastic content is not output.

[0014] In a second aspect, the present invention provides a computer device including a memory and a processor, wherein the memory stores a computer program, wherein when the computer program is executed by the processor, it implements any step of the method for quantitative detection of recycled plastics based on intelligent sensing as described in the first aspect of the present invention.

[0015] Thirdly, the present invention provides a computer-readable storage medium having a computer program stored thereon, wherein the computer program, when executed by a processor, implements any step of the method for quantitative detection of recycled plastics based on intelligent sensing as described in the first aspect of the present invention.

[0016] The beneficial effects of this invention are as follows: By acquiring the plastic sample to be tested and constructing the calibration strip and threshold group for this batch, comprehensive collection and recording of the plastic type, morphology, color, thickness, surface condition, and testing environment of the plastic sample to be tested are achieved. Combined with the virgin plastic reference strip, the calibration strip with the known blending ratio of recycled plastic, the no-load response, and the repeated test results, a systematic threshold group and blending ratio response relationship are formed for subsequent multi-source data analysis and quantitative calculation. This ensures that the characteristic parameters of each region during the testing process are traceable, quantifiable, and comparable, providing a reliable basis for accurately determining the recycled plastic content. By performing a progressive state judgment step based on the threshold group to bind the records of the regions, a layered judgment of the coverage effectiveness, spectral effectiveness, color interference, thickness deviation, and surface anomalies of the detection region is achieved. Through interactive mechanisms such as triggering thermal response compensation and thickness linkage correction, abnormal regions are targeted, thereby ensuring the reliability and consistency of the quantitative results of each region. At the same time, the accuracy and stability of sample-level recycled plastic content detection are improved, ensuring the effectiveness and scientific nature of the test results under complex sample conditions. Attached Figure Description

[0017] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0018] Figure 1 This is a flowchart of a method for quantitative detection of recycled plastics based on intelligent sensing.

[0019] Figure 2 A flowchart for constructing calibration splines and threshold sets is provided.

[0020] Figure 3 This is a flowchart of the progressive state determination process for batch threshold groups.

[0021] Figure 4A flowchart for multi-branch quantitative calculation and reverse verification driven by regional state. Detailed Implementation

[0022] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, the specific embodiments of the present invention will be described in detail below with reference to the accompanying drawings.

[0023] Many specific details are set forth in the following description in order to provide a full understanding of the invention. However, the invention may also be practiced in other ways different from those described herein, and those skilled in the art can make similar extensions without departing from the spirit of the invention. Therefore, the invention is not limited to the specific embodiments disclosed below.

[0024] Secondly, the term "one embodiment" or "embodiment" as used herein refers to a specific feature, structure, or characteristic that may be included in at least one implementation of the present invention. The phrase "in one embodiment" appearing in different places in this specification does not necessarily refer to the same embodiment, nor is it a single or selective embodiment that is mutually exclusive with other embodiments.

[0025] Reference Figures 1-4 This is one embodiment of the present invention, which provides a method for quantitative detection of recycled plastics based on intelligent sensing, including the following steps: S1. Obtain the sample information and testing environment information of the plastic sample to be tested, confirm the type of plastic in the plastic sample to be tested, and generate the threshold group and blending ratio response relationship corresponding to this batch of tests based on the reference sample of virgin plastic of the same type, the calibration sample of known recycled plastic blending ratio, the no-load sensor response and repeated test results.

[0026] Furthermore, information such as the sample number, batch number, type of plastic, sample shape, color, thickness, surface condition, and testing environment of the plastic sample to be tested is obtained to form a sample testing record.

[0027] The recycled plastic content is defined as the percentage of the total mass of recycled plastic in the plastic sample being tested. For granular samples, sheet samples, film samples, block samples, or product slices, the recycled plastic content is expressed as a percentage by mass. When the plastic sample being tested is a granular sample, record the granule spreading area, granule spreading thickness, granule coverage ratio, and granule gap ratio; when the plastic sample being tested is a sheet, film, or product slice, record the effective detection area, nominal thickness, thickness measurement location, and surface orientation. The sample testing records, subsequent data collection, calibration strip data, threshold group data, and test results are all assigned the same batch number.

[0028] If the plastic type in the sample source record does not match the near-infrared spectral resin characteristics, a plastic type conflict record is generated. This record includes the sample number, the plastic type in the source record, the spectrally identified plastic type, the resin characteristic band where the conflict occurred, and the time of the conflict. After generating the conflict record, the selection of calibration strips for this batch based on the plastic type in the source record, threshold generation, and the establishment of blending ratio response relationships are paused, and a plastic type verification process is performed. This verification process involves re-acquiring near-infrared spectral resin characteristic band responses from at least two detection regions of the plastic sample to be tested, and matching them with the resin characteristic band range corresponding to the candidate plastic type and the native plastic reference response, respectively.

[0029] When multiple re-collected detection areas show consistent spectral identification of the plastic type, and the matching deviation with the corresponding virgin plastic reference response does not exceed the allowable deviation for resin characteristic matching, the spectral identification of the plastic type is used as the plastic type for this batch's calibration sample selection, threshold generation, and quantitative calculation of recycled plastic content, and the plastic type verification result is written into the sample detection record. If the re-collected results still show different plastic type identifications, or if the matching deviation between the spectral identification result and the virgin plastic reference response exceeds the allowable deviation for resin characteristic matching, the plastic sample to be tested is marked as a sample requiring plastic type verification. No threshold group or blending ratio response relationship is generated for this batch, and a manual verification prompt or resampling prompt is output.

[0030] The allowable deviation for resin characteristic matching is determined by repeated detection fluctuations of virgin plastic reference specimens and calibration specimens with known recycled plastic blending ratios in the resin characteristic band. The value is expressed in the same unit as the response intensity of the corresponding resin characteristic band, or in the form of normalized deviation.

[0031] After confirming the type of plastic, select or prepare a reference strip of virgin plastic belonging to the same type as the plastic sample to be tested, and a calibration strip with a known blending ratio of recycled plastic. The virgin plastic reference strip is prepared using the same type of virgin resin without any recycled plastic added. The calibration strip with a known blending ratio of recycled plastic is prepared by mixing, melting, tableting, extruding, or granulating virgin and recycled plastics of the same type in a predetermined mass ratio. The main resin type, detection morphology, and detection thickness range of the calibration strip are consistent with those of the plastic sample to be tested.

[0032] After generating a reference material mismatch record, the use of the threshold group and blending ratio response relationship corresponding to the current recycled plastic reference material is paused, and reference material verification processing is performed. If a suitable recycled plastic reference material can be reselected, the threshold group and blending ratio response relationship for this batch are regenerated based on the reselected reference material. If the deviation characteristics can be corrected through drying, surface cleaning, particle size uniformity, thickness adjustment, or homogenization, the basic characteristics are re-collected after performing the same pretreatment on the plastic sample to be tested and the reference material. If the compatibility range is met, the test continues. If a suitable reference material cannot be reselected and the compatibility range is still exceeded after pretreatment, the plastic sample to be tested is marked as a reference material mismatch sample, no quantitative detection result of recycled plastic content is output, and a resampling prompt, a supplementary reference material prompt, or a manual verification prompt is generated.

[0033] The calibration data for this batch is generated by repeatedly testing virgin plastic reference samples, calibration samples with known recycled plastic blending ratios, and under no-load testing conditions. The number of repeated tests is determined based on the stability of the sensor's repeated testing error, and the repeated tests must cover at least repeated acquisition of the same detection area and repeated acquisition of different detection areas. Under no-load testing conditions, the no-load response of near-infrared spectrum, image no-load response, thermal response, and thickness detection is collected to form the no-load noise response. For the virgin plastic reference samples, the virgin plastic spectral response, virgin plastic image state, virgin plastic thermal response, and virgin plastic thickness response are collected to form the virgin plastic repeated response. For each known recycled plastic blending ratio calibration sample, the spectral response, image state, thermal response, and thickness response are collected to form calibration responses for different blending ratios.

[0034] Spectral response includes target band reflectance intensity, target band absorption changes, spectral attenuation characteristics, oxidation aging characteristics, and additive residue characteristics. Image status includes color channel mean, brightness distribution, area ratio of main color category, sample coverage area, surface anomalous area, and transparency status. Thermal response includes heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time. Thickness response includes spline thickness, packing thickness, or areal density. During repeated detection, fluctuations in light source output, thermal excitation output, thickness detection repeatability deviation, and image acquisition repeatability deviation are recorded simultaneously.

[0035] The permissible error for testing is jointly determined by the target accuracy of the testing method, the calibration strip preparation error, and the sensor repeatability error. The target accuracy of the testing method is determined by the resolution of the recycled plastic content required in the test report. The calibration strip preparation error is determined by the weighing errors of virgin and recycled plastics, the mixing uniformity error, and the sample thickness error. The sensor repeatability error is determined by the dispersion of the predicted content values ​​obtained from repeated testing of the same calibration strip under the same testing conditions. The permissible error is the maximum value among the target accuracy of the testing method, the calibration strip preparation error, and the sensor repeatability error, serving as the common error basis for threshold generation, branch triggering, and back-calculation verification for this batch. The permissible error record includes the target accuracy value, weighing error value, mixing uniformity error value, sample thickness error value, sensor repeatability error value, and the finally adopted permissible error value.

[0036] It should be noted that the threshold set for this batch is generated based on the no-load noise response, the repeated test results of the calibration strip, and the allowable error of the test. This batch threshold set includes the coverage ratio threshold, spectral signal-to-noise ratio threshold, color dispersion threshold, thickness adaptation threshold, surface anomaly area threshold, thermal response compensation trigger condition, regional dispersion threshold, spectral-thermal response direction conflict threshold, and allowable error of calibration strip back-calculation. The threshold set is linked and saved with the sample test records, calibration strip records, recycled plastic reference material adaptation records, and no-load test records.

[0037] The coverage ratio threshold is the percentage of the actual area covered by the plastic sample within the detection area to the total area of ​​the detection area, and its value is greater than 0 and not greater than 1. The surface anomaly area threshold is the percentage of the surface anomaly area to the plastic coverage area within the detection area, and its value is greater than 0 and not greater than 1. The color dispersion threshold uses the same units as the color dispersion value. The color dispersion value is obtained by weighting the normalized difference of the color channel mean, the standard deviation of brightness, and the dispersion term of the main color category, and its value is greater than or equal to 0 and not greater than 1. The main color category dispersion term is 1 minus the main color category area ratio, which is the ratio of the number of main color category pixels to the number of plastic coverage pixels in the detection area. The main color category dispersion term value is greater than or equal to 0 and less than 1, and the corresponding color dispersion threshold is greater than 0 and not greater than 1. The spectral signal-to-noise ratio threshold is the ratio between the effective response intensity of the target detection band and the standard deviation of the no-load noise, and its value is positive. The thickness adaptation threshold uses the same unit of thickness as the thickness deviation value, including millimeters, micrometers, or units consistent with the thickness conversion of areal density, and its value is positive. The thermal response stability threshold and the repeatability deviation of the corresponding thermal response feature use the same unit. The minimum distinguishable difference in thermal response and the difference in thermal response features between adjacent blending ratio calibration samples also use the same unit, and both are positive numbers. The regional dispersion threshold and the regional dispersion index use the same expression method; when the regional dispersion index is the range or standard deviation, the regional dispersion threshold uses the percentage of recycled plastic content, and the value is greater than or equal to 0; when the regional dispersion index is the coefficient of variation, the regional dispersion threshold is a proportional value, and the value is greater than 0 and not greater than 1. The spectral-thermal response direction conflict threshold is determined according to the expression method of the direction deviation amplitude; when the direction deviation amplitude is expressed as the difference between the normalized change of the spectral feature and the normalized change of the thermal response feature, the direction conflict threshold is the normalized difference threshold, and the value is greater than 0 and not greater than 1; when the direction deviation amplitude is expressed as the proportion of the number of opposite-direction features to the total number of features participating in the direction verification, the direction conflict threshold is the proportion threshold of the number of opposite-direction features, and the value is greater than 0 and not greater than 1. The allowable error for calibration sample back-calculation and the allowable error for testing use the same unit, both expressed as the deviation of the percentage of recycled plastic mass, and the value is greater than or equal to 0.

[0038] The coverage ratio threshold is determined based on the correspondence between the sensor's effective field of view, the spectral sampling spot area, and the sample coverage area. Within the same sensing field of view, the sample coverage area is gradually changed, and the stability of the spectral response intensity as a function of the coverage area is detected. When the sample coverage area continues to increase without the change in spectral response intensity exceeding the allowable detection error, the corresponding coverage ratio is determined as the coverage ratio threshold. The coverage ratio threshold record includes the sensor's effective field of view, the spectral sampling spot area, the sequence of sample coverage area changes, the spectral response intensity corresponding to each coverage area, and the determined coverage ratio threshold.

[0039] The spectral signal-to-noise ratio (SNR) threshold is determined based on the no-load noise response, the repeated detection fluctuation of the virgin plastic reference sample, and the minimum spectral response difference between adjacent blending ratio calibration samples. The spectral SNR is expressed as the ratio of the effective response intensity of the target detection band to the standard deviation of the no-load noise. The effective response intensity of the target detection band is the response intensity of the detection area in the target band minus the no-load background response. The standard deviation of the no-load noise is calculated from multiple response data of the same target band under no-load detection conditions. To determine the spectral SNR threshold, the background noise level is first determined based on the no-load noise response, then the equipment repeatability fluctuation is determined based on the repeated detection results of the virgin plastic reference sample, and finally, the target band response difference between calibration samples with different blending ratios is compared. The minimum SNR that allows adjacent blending ratio calibration samples to be distinguished is determined as the spectral SNR threshold. The spectral SNR threshold record includes the no-load noise level, the standard deviation of the no-load noise, the repeated fluctuation of the virgin plastic, the target band response difference between adjacent ratio calibration samples, the effective response intensity of the target detection band, and the finally determined spectral SNR threshold.

[0040] The color dispersion threshold is determined based on the detection results of calibration samples with the same blending ratio of recycled plastic but different color states. Image acquisition and spectral detection are performed on calibration samples with the same blending ratio but different colors to calculate the color channel mean difference, luminance standard deviation, and the area ratio of the main color category. The color channel mean difference is determined based on the color channel mean of the plastic-covered pixels in the detection area. When using the RGB color space, the mean values ​​of the red, green, and blue channels are calculated separately, and each channel mean is normalized. The difference between the maximum and minimum values ​​of the three normalized color channel means is taken as the color channel mean difference, which is greater than or equal to 0 and not greater than 1. The luminance standard deviation is calculated based on the normalized luminance value of the plastic-covered pixels in the luminance channel of the detection area, and its value is greater than or equal to 0 and not greater than 1.

[0041] The primary color category is obtained by clustering the hue, saturation, and brightness distribution of pixels in the detection area within the HSV color space. The number of clusters is determined by the number of hue distribution peaks within the detection area, or by dividing color categories according to preset hue, saturation, and brightness intervals. The color category with the highest proportion is designated as the primary color category, and its area proportion is the ratio of the number of pixels in that category to the number of pixels covered by plastic in the detection area, with a value greater than 0 and not greater than 1. The color channel mean difference, brightness standard deviation, and primary color category area proportion are normalized and then weighted and summed to obtain the color discrete value. Each weight is determined by regression analysis of the spectral prediction deviation caused by different color calibration splines. When color differences cause the spectral quantitative result to deviate from the known mixing ratio and reach the allowable detection error, the corresponding color discrete value is determined as the color discrete threshold. The color discrete threshold record includes the color channel mean difference, brightness standard deviation, primary color category area proportion, normalization method, weight source, color discrete value, and the finally determined color discrete threshold.

[0042] The thickness adaptation threshold is determined based on the effective detection depth of the near-infrared spectral sensing unit, the effective thermal diffusion depth of the thermal response sensing unit, and the repeated detection results of calibration samples under different thickness conditions. Different thicknesses or different stacking thicknesses are set for calibration samples with the same blending ratio, and spectral and thermal responses are collected respectively. The predicted deviation of recycled plastic content caused by thickness changes is calculated. When the predicted deviation of recycled plastic content caused by thickness changes reaches the allowable detection error, the corresponding thickness deviation range is determined as the thickness adaptation threshold. The thickness deviation value is determined according to the degree of deviation of the detection area thickness from the effective thickness range of similar calibration samples; when the detection area thickness is within the effective thickness range, the thickness deviation value is recorded as zero; when the detection area thickness is lower than the lower limit of the effective thickness range, the thickness deviation value is the difference between the lower limit of the effective thickness range and the detection area thickness; when the detection area thickness is higher than the upper limit of the effective thickness range, the thickness deviation value is the difference between the detection area thickness and the upper limit of the effective thickness range. The thickness adaptation threshold record includes the effective detection depth, effective thermal diffusion depth, calibration strip thickness change sequence, spectral response under each thickness condition, thermal response under each thickness condition, effective thickness range, thickness deviation calculation results, and the finally determined thickness adaptation threshold.

[0043] The surface anomaly area threshold is determined based on repeated testing results of samples of the same material with known areas of scratches, contamination, foreign matter spots, or charring marks. Scratches, contamination, foreign matter spots, and charring marks are recorded as corresponding anomaly types. When multiple anomalies coexist, the area ratio of each anomaly type is calculated separately, and the overall anomaly area ratio is calculated. The overall anomaly area ratio is obtained by weighted summation of the area ratios of each anomaly type according to the influence weights of the corresponding anomaly type on the spectral and thermal responses. The influence weights are determined by the content prediction deviation caused by the calibration sample of the corresponding anomaly type. The surface anomaly area ratio is gradually increased, and the spectral response, image response, and thermal response deviations of the anomaly area and the normal area are compared. When the area ratio of any anomaly type or the overall anomaly area ratio causes the predicted deviation of recycled plastic content to reach the detection allowable error, the corresponding anomaly area ratio is determined as the surface anomaly area threshold. The surface anomaly area threshold record includes the anomaly type, the anomaly area ratio change sequence, the overall anomaly area ratio, the spectral response deviation, image response deviation, thermal response deviation corresponding to each anomaly area ratio, the influence weight of the anomaly type, and the finally determined surface anomaly area threshold.

[0044] The thermal response compensation triggering conditions include color triggering conditions, spectral triggering conditions, thermal response stability conditions, and thermal response distinguishability conditions. The color triggering condition is when the color dispersion value reaches the color dispersion threshold; the spectral triggering condition is when the spectral signal-to-noise ratio is lower than the spectral signal-to-noise ratio threshold; the thermal response stability condition is when the thermal response repetition detection deviation does not exceed the thermal response stability threshold; and the thermal response distinguishability condition is when the difference in thermal response characteristics between adjacent blending ratio calibration samples is greater than the minimum distinguishable difference in thermal response. The thermal response stability threshold is determined based on the dispersion of the heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time obtained from repeated thermal excitation detection of the same calibration sample. The minimum distinguishable difference in thermal response is determined based on the difference in thermal response characteristics between adjacent blending ratio calibration samples. Thermal response compensation triggering conditions are considered satisfied when at least one of the color and spectral triggering conditions is met, and both the thermal response stability and thermal response distinguishability conditions are met simultaneously. Thermal response compensation is not triggered when neither the color nor spectral triggering conditions are met; thermal response compensation is not triggered when either the thermal response stability or thermal response distinguishability condition is not met. The thermal response compensation trigger condition record includes color trigger condition, spectral trigger condition, thermal response stability threshold, minimum distinguishable difference in thermal response, heating rate range, steady-state temperature difference range, cooling slope range, thermal diffusion time range, and trigger judgment result.

[0045] The regional dispersion threshold is determined based on the results of repeated testing of the same calibration strip in multiple regions. Multiple testing regions are divided for the calibration strip with the same mixing ratio, and the range, standard deviation, and coefficient of variation of the predicted values ​​for each region are calculated. The selection rules for the regional dispersion index are as follows: when the units of the predicted values ​​in each testing region are consistent and the sample size is less than the preset number of regions, the range is used as the regional dispersion index; when the units of the predicted values ​​in each testing region are consistent and the sample size reaches the preset number of regions, the standard deviation is used as the regional dispersion index; when it is necessary to compare the relative dispersion between calibration strips with different mixing ratios, the coefficient of variation is used as the regional dispersion index. The preset number of regions is determined by the stability of the regional predicted values ​​in the repeated testing of the calibration strip. The regional dispersion range under normal sample preparation and normal testing conditions is determined as the regional dispersion threshold. The regional dispersion threshold record includes the calibration strip number, mixing ratio, number of testing regions, predicted value for each region, range, standard deviation, coefficient of variation, selection result of the regional dispersion index, and the finally determined regional dispersion threshold.

[0046] The spectral-thermal response direction conflict threshold is determined based on the correspondence between the directions of change of spectral features and thermal response features in calibration samples with different blending ratios. A direction code is established for each target spectral feature and each target thermal response feature. If the target feature increases with the increase of the recycled plastic blending ratio in the calibration sample, the direction code for the increase of the target feature's content is positive; if the target feature decreases with the increase of the recycled plastic blending ratio in the calibration sample, the direction code for the increase of the target feature's content is negative. When the direction of change of the target feature relative to the virgin plastic reference response in the detection area is consistent with the corresponding direction code, it is recorded as consistent direction; when it is opposite to the corresponding direction code, it is recorded as opposite direction. The spectral feature direction code results are compared with the thermal response feature direction code results, and the deviation magnitude is calculated. The deviation magnitude is determined based on the difference between the normalized change of the spectral feature and the normalized change of the thermal response feature, or the proportion of opposite-direction features. When the deviation magnitude exceeds the allowable normal deviation range in the calibration sample, the spectral-thermal response direction conflict threshold is considered to have been reached. The spectral-thermal response direction conflict threshold record includes the spectral characteristic change direction, thermal response characteristic change direction, direction encoding, normalized change amount, normal deviation range, and the finally determined direction conflict threshold for each mixing ratio.

[0047] The allowable error for calibration sample back-calculation is consistent with the allowable error for detection. After inputting calibration samples with known recycled plastic blending ratios into the same detection process, the absolute difference between the back-calculated recycled plastic content and the corresponding known blending ratio is calculated and used as the back-calculation deviation. The back-calculation deviation is compared with the allowable error for calibration sample back-calculation. The record of the allowable error for calibration sample back-calculation includes the allowable error for detection, the calibration sample number involved in the back-calculation, the known blending ratio, the back-calculated recycled plastic content, and the back-calculation deviation record.

[0048] S2. Divide the plastic sample to be tested into testing areas, collect spectral response data, image status data, thermal response data and geometric status data of each testing area, and bind them according to the testing area number to form an area binding record.

[0049] Furthermore, the plastic sample to be tested is placed on a multifunctional intelligent sensing detection platform, and divided into several detection areas based on the sample's shape, area, and thickness. The size and number of each detection area are determined based on the effective detection area of ​​the sample, the area of ​​a single sensor's field of view, the regional dispersion results of repeated detection in multiple areas of the calibration strip, and the minimum number of detection areas. The minimum number of detection areas is calculated by comparing the range, standard deviation, and coefficient of variation of repeated detection in multiple areas of the calibration strip. When increasing the number of detection areas, the predicted value of the recycled plastic content of the calibration strip does not exceed the allowable error of the calibration strip's back-calculation; this number is the minimum number of detection areas. For granular samples, the number of detection areas also needs to consider the particle spreading area, coverage ratio, and particle gap ratio; for sheets, films, or slices of products, the number of detection areas is divided based on the ratio of the effective area of ​​the sample to the area of ​​a single sensor's field of view.

[0050] Within each detection area, near-infrared spectroscopy, visible light images, thermal response, and thickness data are collected to form multi-source data. The near-infrared spectroscopy sensing unit acquires the reflectance spectrum of the detection area in a preset wavelength band, generating spectral response data, including target wavelength band reflectance intensity, spectral attenuation characteristics, oxidation aging characteristics, and additive residue characteristics. The visible light image sensing unit acquires color distribution, brightness distribution, area proportion of the dominant color category, coverage area, transparency status, and surface anomaly information. The dominant color category is determined through HSV color space clustering, with the category having the highest proportion as the dominant color category, and its area proportion is the ratio of the number of pixels in that category to the total number of pixels covered by the detection area. The thermal response sensing unit acquires data on heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time under low-power thermal excitation conditions. The thickness detection unit measures and records the thickness or accumulated thickness of the detection area.

[0051] Multi-source data from each detection area are bound together according to the area number to form an area binding record. Each record includes the area number, spectral response data, image status data, thermal response data, thickness data, acquisition time, sensor number, and detection environment parameters. The area binding record serves as the data object for subsequent progressive status determination and multi-branch quantitative calculation.

[0052] During the generation of region binding records, the coverage ratio threshold, spectral signal-to-noise ratio threshold, color dispersion threshold, thickness adaptation threshold, and surface anomaly area threshold in this batch of threshold groups are recorded. The coverage ratio threshold is used to determine whether the effective coverage area of ​​the detection region meets the spectral and thermal response detection conditions; the spectral signal-to-noise ratio threshold is used to determine whether the spectral data has reached the minimum distinguishable level; the color dispersion threshold is used to determine whether the image color state interferes with the spectral quantitative path; the thickness adaptation threshold is used to determine the thickness deviation of the detection region to decide whether thickness linkage correction is needed; the surface anomaly area threshold is used to determine the degree of influence of scratches, contamination, foreign objects, or charring marks on the detection region. When the surface anomaly area ratio reaches the surface anomaly area threshold, the detection region is marked as a surface anomaly verification region, and surface anomaly verification processing is performed in subsequent quantitative calculations.

[0053] S3. Based on the threshold group, perform progressive state determination on the region-bound records to form a region state combination.

[0054] Furthermore, based on the threshold group of this batch, a progressive state determination is performed on each region-binding record to form a region state combination corresponding to each detection region. The region state combination includes one or more of the following: insufficient coverage region, spectrally usable region, spectrally low confidence region, color interference region, thermal response compensation candidate region, thickness correction region, thickness rejection region, surface anomaly verification region, and surface rejection region. The progressive state determination is performed in the order of coverage validity determination, spectral validity determination, color interference determination, thickness deviation determination, and surface anomaly determination, and the result of each determination is written into the region state record of the corresponding detection region.

[0055] When determining the coverage validity of region-bound records, the sample coverage ratio is first calculated based on the region image state data. The sample coverage ratio is the ratio of the pixel area actually covered by the plastic sample within the detection region to the total pixel area of ​​the detection region. The pixel area actually covered by the plastic sample is determined based on the segmentation results of the plastic region and the background region in the visible light image; the background region includes the exposed area of ​​the sample stage, the interparticle gap area, and the field of view area not covered by the sample. The segmentation criteria for the plastic region and the background region include brightness difference, edge contour, transparency status, and background recording of the unloaded image. For transparent or semi-transparent samples, the plastic-covered pixels are determined by combining the background response of the unloaded image and the weak response area of ​​the near-infrared spectrum, avoiding misjudging transparent samples as background regions.

[0056] The coverage ratio threshold is adopted from the threshold group of this batch. This threshold is determined by experiments on the effective field of view of the sensor, the area of ​​the spectral sampling spot, and the change in sample coverage area. When the sample coverage area continues to increase while the change in spectral response intensity does not exceed the allowable detection error, the corresponding coverage ratio is used as the coverage ratio threshold. The sample coverage ratio of the current detection area is compared with the coverage ratio threshold. When the sample coverage ratio is less than the coverage ratio threshold, the detection area is marked as an insufficiently covered area, and the coverage ratio, coverage ratio threshold, uncovered pixel area, and reason for insufficient coverage are recorded. Insufficiently covered areas are not included in the spectral validity determination, color interference determination, and thermal response compensation determination. When the sample coverage ratio is greater than or equal to the coverage ratio threshold, the detection area is written into the coverage validity record and included in the spectral validity determination.

[0057] After determining the coverage effectiveness, the spectral effectiveness of the effective coverage area is then assessed. The spectral signal-to-noise ratio (SNR) is calculated based on the regional spectral response data. The SNR is expressed as the ratio of the effective response intensity of the target detection band to the standard deviation of the no-load noise. The effective response intensity of the target detection band is the response intensity of the detection area in the target band minus the no-load background response; the no-load noise standard deviation is calculated from multiple response data of the same target band under no-load detection conditions in this batch. The target detection band is determined based on bands with distinguishable response differences between virgin plastic reference samples of the same plastic type and calibration samples with different blending ratios of recycled plastics.

[0058] The spectral signal-to-noise ratio (SNR) threshold is adopted from the threshold group of this batch. This threshold is determined by the no-load noise response, the fluctuation of repeated detection of the virgin plastic reference sample, and the minimum spectral response difference between adjacent blending ratio calibration samples. The spectral SNR of the current detection area is compared with the spectral SNR threshold. When the spectral SNR is less than the spectral SNR threshold, the detection area is marked as a low-confidence region and written into the low-confidence record. The low-confidence record includes the target detection band, effective response intensity, no-load noise standard deviation, spectral SNR, spectral SNR threshold, and reason for low confidence. The low-confidence region does not directly enter the spectral-dominated quantitative path, but continues to call the regional thermal response data to determine whether it has the candidate conditions for thermal response compensation. When the spectral SNR is greater than or equal to the spectral SNR threshold, the detection area is marked as a spectrally usable region and enters the color interference judgment.

[0059] Color interference is assessed in the effective spectral region or the detection region with low spectral confidence but effective coverage. Color discrete values ​​are calculated based on the region's image state data. These values ​​are determined by the color channel mean difference, brightness standard deviation, and the main color category discrete term. The color channel mean difference represents the deviation between different color channel responses within the detection region; the brightness standard deviation represents the unevenness of the brightness distribution within the detection region; and the main color category discrete term represents the inverse change in color concentration within the detection region. The main color category is obtained by clustering the hue, saturation, and brightness distribution of pixels in the detection region within the HSV color space. The number of clusters is determined by the number of hue distribution peaks within the detection region, or by dividing color categories according to preset hue, saturation, and brightness intervals. The color category with the highest percentage is designated as the main color category. The area percentage of the main color category is the ratio of the number of pixels in that color category to the number of pixels covered by plastic in the detection region. The main color category discrete term is one minus the area percentage of the main color category.

[0060] After normalizing the color channel mean difference, brightness standard deviation, and main color category discrete term, the values ​​are weighted and summed according to the weights obtained from the regression of the color calibration strips in this batch to obtain the color discrete value of the current detection area. The color discrete threshold is the color discrete threshold in the threshold group of this batch. This threshold is determined by the detection results of calibration strips with the same recycled plastic blending ratio but different color states; when the color difference causes the spectral quantitative result to deviate from the known blending ratio and reaches the detection allowable error, the corresponding color discrete value is used as the color discrete threshold. The color discrete value of the current detection area is compared with the color discrete threshold. When the color discrete value is less than the color discrete threshold, the detection area is marked as a spectrally usable area, and the color discrete value, color discrete threshold, main color category, and color state are recorded. When the color discrete value is greater than or equal to the color discrete threshold, the detection area is marked as a color interference area, and a color interference record is written. The color interference record includes the color channel mean difference, brightness standard deviation, main color category discrete term, color discrete value, color discrete threshold, and color interference source.

[0061] For detection areas marked as color interference regions, the process continues to determine whether a thermal response compensation candidate region is formed. The thermal response compensation candidate determination invokes the thermal response compensation trigger conditions from the current batch threshold group. These trigger conditions include color trigger conditions, spectral trigger conditions, thermal response stability conditions, and thermal response distinguishability conditions. A color trigger condition is when the color dispersion value reaches a color dispersion threshold; a spectral trigger condition is when the spectral signal-to-noise ratio is lower than a spectral signal-to-noise ratio threshold; a thermal response stability condition is when the thermal response repetition detection deviation does not exceed a thermal response stability threshold; and a thermal response distinguishability condition is when the difference in thermal response characteristics between adjacent mixing ratio calibration samples is greater than the minimum distinguishable difference in thermal response. When at least one of the color trigger condition and the spectral trigger condition is met, and both the thermal response stability condition and the thermal response distinguishability condition are met simultaneously, the corresponding detection area is marked as a thermal response compensation candidate region. The thermal response compensation candidate record includes the color trigger result, spectral trigger result, thermal response stability determination result, thermal response distinguishability determination result, heating rate, steady-state temperature difference, cooling slope, thermal diffusion time, and trigger determination result.

[0062] The thickness conversion based on unit area areal density is used to characterize the thickness of granular or irregularly spread samples. Unit area areal density is the ratio of the mass of plastic particles within the detection area to the area covered by plastic. The thickness conversion based on unit area areal density is expressed as follows: ; in, Indicates the thickness converted from the surface density of the unit detection area; Indicates the surface density of the unit detection area. ,in, Indicates the mass of plastic particles within the detection area. Indicates the area covered by plastic within the detection zone; This indicates the material density corresponding to the type of plastic in the plastic sample being tested. This represents the particle gap correction coefficient, with a value greater than 0 and not greater than 1.

[0063] The particle gap correction coefficient is determined based on the particle coverage area, particle gap area, and particle stacking state in the image state data of the detection area. When the particles are continuously spread and the proportion of the particle gap area is low, the particle gap correction coefficient is close to 1; when the proportion of the particle gap area increases or there are stacking gaps, the particle gap correction coefficient decreases based on the particle spreading test results of the calibration strip.

[0064] The particle gap correction factor is determined through repeated testing data of calibration strips of the same type of plastic particles. Specifically, a calibration strip with known mass, material density, and spread area is laid in the test area, image status data is collected, the particle coverage area ratio and particle gap area ratio are calculated, and the corresponding actual packing thickness is measured; the areal density, material density, and actual packing thickness of the unit test area are substituted into the following formula to calculate the particle gap correction factor: ; in, This represents the particle gap correction factor; Indicates the surface density of the unit detection area; This indicates the material density of the plastic type corresponding to the particle calibration strip; This indicates the actual packing thickness of the particle calibration specimen under the corresponding spreading condition.

[0065] After repeated testing of particle calibration strips with different particle sizes, spreading areas, and stacking states, a particle gap correction coefficient table is generated. When determining the thickness deviation of the particle sample to be tested, the corresponding particle gap correction coefficient is matched according to its image state data, and the thickness converted to the unit detection area surface density is calculated. If the mass of plastic particles in the detection area cannot be obtained for the plastic sample to be tested, the thickness converted to the unit detection area surface density is determined by the "coverage area ratio - stacking thickness" correspondence formed by the particle spreading test of the calibration strip; at this time, the particle coverage area ratio and particle gap area ratio are obtained according to the image state data of the detection area, the converted thickness is matched in the correspondence, and the matched converted thickness is used as the basis for calculating the thickness deviation value.

[0066] The current thickness of the detection area can be the thickness of a sheet, film, product slice, particle spread and stacking, or the thickness converted from the areal density of the unit detection area. The thickness deviation is determined by the degree of deviation of the current detection area thickness from the effective thickness range of similar calibration strips. When the current detection area thickness is within the effective thickness range, the thickness deviation is recorded as zero; when the current detection area thickness is below the lower limit of the effective thickness range, the thickness deviation is the difference between the lower limit of the effective thickness range and the current detection area thickness; when the current detection area thickness is above the upper limit of the effective thickness range, the thickness deviation is the difference between the current detection area thickness and the upper limit of the effective thickness range.

[0067] The thickness adaptation threshold is adopted from the threshold group of this batch. This threshold is determined by the effective detection depth of the near-infrared spectral sensing unit, the effective thermal diffusion depth of the thermal response sensing unit, and the repeated detection results of the calibration sample under different thickness conditions. When the predicted deviation of recycled plastic content caused by thickness change reaches the allowable detection error, the corresponding thickness deviation range is used as the thickness adaptation threshold. The upper limit of thickness correction is determined by the critical value of the predicted deviation of recycled plastic content caused by thickness change exceeding the allowable error of the calibration sample back calculation. The thickness deviation value is compared with the thickness adaptation threshold and the upper limit of thickness correction. When the thickness deviation value is less than the thickness adaptation threshold, no thickness correction label is generated. When the thickness deviation value is greater than or equal to the thickness adaptation threshold and less than or equal to the upper limit of thickness correction, the detection area is marked as a thickness correction area. When the thickness deviation value is greater than the upper limit of thickness correction, the detection area is marked as a thickness rejection area and written into the thickness rejection record. The thickness deviation judgment record includes the current detection area thickness, effective thickness range, thickness deviation value, thickness adaptation threshold, upper limit of thickness correction, and judgment label.

[0068] After thickness deviation determination, surface anomaly assessment is performed on the detection areas with effective coverage that were not rejected by thickness criteria. Scratches, contaminant adhesions, foreign matter points, and charring marks are identified based on the area image status data, and the area ratio of each anomaly type is calculated. The anomaly type area ratio is the ratio of the pixel area of ​​the corresponding anomaly type to the pixel area of ​​the plastic coverage in the detection area. When multiple anomaly types exist simultaneously, a comprehensive anomaly area ratio is calculated. The comprehensive anomaly area ratio is obtained by weighted summation of the area ratios of each anomaly type according to the influence weights of the corresponding anomaly type on the spectral and thermal responses. The influence weights are determined by the content prediction deviation caused by the calibration strips of the corresponding anomaly type.

[0069] The surface anomaly area threshold is adopted from the threshold group of this batch. This threshold is determined by repeated testing results of the same material sample with known area scratches, contamination, foreign matter spots, or charring marks; when the area ratio of any anomaly type or the combined area ratio of anomalies causes the predicted deviation of recycled plastic content to reach the allowable error of the test, the corresponding anomaly area ratio is used as the surface anomaly area threshold.

[0070] The upper limit for anomaly rejection is adopted from the threshold group of this batch. The upper limit is determined by the proportion of the anomaly area when surface anomalies cause simultaneous distortion of both the spectral and thermal responses, and the back-calculation error of the calibration sample exceeds the allowable error of the calibration sample back-calculation. The upper limit is an area proportion threshold, with a value greater than the surface anomaly area threshold but not greater than 1. When different anomaly types correspond to different upper limits, the upper limits for scratch rejection, contamination rejection, foreign object rejection, and charring rejection are recorded separately. When multiple anomaly types exist simultaneously, the area proportion of each anomaly type and the overall anomaly area proportion are calculated simultaneously and compared with the corresponding upper limit.

[0071] Specifically, if the area ratio of any anomaly type reaches the corresponding anomaly removal limit, or the overall anomaly area ratio reaches the overall anomaly removal limit, the detected area is marked as a surface removal area; if the area ratio of any anomaly type or the overall anomaly area reaches the surface anomaly area threshold, but does not reach the corresponding anomaly removal limit, the detected area is marked as a surface anomaly verification area; if the area ratios of each anomaly type and the overall anomaly area do not reach the surface anomaly area threshold, the detected area is retained as a valid area.

[0072] The overall anomaly area ratio is obtained by weighting the area ratio of each anomaly type and the influence weight of the corresponding anomaly type on the spectral response and thermal response. The overall anomaly removal limit is determined by the repeated detection results of the same material sample where multiple anomaly types coexist. When no separate overall anomaly removal limit is established, the overall anomaly removal limit is the minimum value among the individual anomaly removal limits.

[0073] When the area ratio of each anomaly type and the overall anomaly area ratio are both less than the surface anomaly area threshold, the detected area is retained as a valid area. When the area ratio of any anomaly type or the overall anomaly area ratio is greater than or equal to the surface anomaly area threshold, but less than or equal to the anomaly removal upper limit, the detected area is marked as a surface anomaly verification area and written into the surface anomaly verification record. The surface anomaly verification area enters the surface anomaly verification processing in subsequent quantitative calculations. The surface anomaly verification processing includes removing the spectral sampling points and image pixels corresponding to the abnormal pixels, recalculating the spectral signal-to-noise ratio, color discrete value, and surface anomaly area ratio, and entering the spectral-dominant quantitative branch or thermal response compensation branch based on the recalculation results. When the area ratio of any anomaly type or the overall anomaly area ratio is greater than the anomaly removal upper limit, the detected area is marked as a surface removal area and written into the surface removal record. The surface anomaly determination record includes the anomaly type, the anomaly type area ratio, the overall anomaly area ratio, the anomaly type influence weight, the surface anomaly area threshold, the anomaly removal upper limit, and the surface anomaly verification label or the surface removal label.

[0074] After determining coverage validity, spectral validity, color interference, thickness deviation, and surface anomaly, a region state combination is generated for each detection area. The region state combination is indexed by the detection area number and records the coverage state, spectral state, color state, thermal response compensation candidate state, thickness state, and surface anomaly state. Areas with insufficient coverage, thickness rejection, and surface rejection are added to the rejection area set; spectrally usable areas are added to the spectral dominant candidate set; spectrally low confidence areas, color interference areas, and thermal response compensation candidate areas are added to the thermal response compensation candidate set; thickness correction areas are added to the thickness linkage correction set; and surface anomaly verification areas are added to the surface anomaly verification set.

[0075] The regional status combination is bound to the corresponding regional record, and the threshold group and judgment record for this batch are bound and saved. The judgment record for each detection area includes the original sensor data, the calculated coverage ratio, spectral signal-to-noise ratio, color dispersion value, thickness deviation value, surface anomaly area ratio, the thresholds used for comparison, the threshold source number, and the judgment label. The progressive status judgment is executed in the following order: coverage effectiveness restricts subsequent detection paths, spectral effectiveness determines whether to call thermal response data, color interference judgment triggers thermal response compensation candidates, thickness deviation judgment triggers spectral response and thermal response linkage correction, and surface anomaly judgment triggers surface anomaly review or rejection.

[0076] S4. Select the corresponding quantitative calculation path according to the combination of regional states, perform spectral-dominant calculation, thermal response compensation calculation, thickness linkage correction calculation or surface anomaly verification calculation on the effective detection area, output the predicted value of recycled plastic content in the region, and generate the initial detection value of recycled plastic content in the sample based on the predicted value of the region after regional consistency screening.

[0077] Furthermore, multi-branch quantitative calculations are performed on the region binding records that have completed the progressive state determination. The calculation objects are detection areas with effective coverage, not rejected by thickness, and not rejected by surface. Each detection area enters the corresponding quantitative calculation path according to the region state combination, which includes spectrally usable areas, spectrally low confidence areas, color interference areas, thermal response compensation candidate areas, thickness correction areas, and surface anomaly verification areas. During the calculation process, no fusion weight is set for each detection area. Instead, a regional recycled plastic content prediction value is first generated for each effective detection area, and then regional prediction values ​​that can participate in sample-level calculations are selected based on regional consistency and sensor source consistency.

[0078] Before performing regional quantitative calculations, a blending ratio response relationship is established based on the calibration sample data of this batch. The blending ratio response relationship is established by combining the effective detection features of the calibration sample with known recycled plastic blending ratios with the corresponding known blending ratios. Effective detection features include spectral attenuation features, oxidative aging features, additive residue features, thermal response attenuation features, and thickness-corrected response features. When only a single dominant feature is used and this feature exhibits monotonical variation at each known blending ratio, a piecewise interpolation relationship is employed; when at least two effective detection features participate in the quantitative calculation, a partial least squares regression relationship is used; linear fitting relationships are used as a local fitting method between adjacent calibration points in piecewise interpolation. After the blending ratio response relationship is established, it is bound and saved with the batch threshold group, calibration sample number, known blending ratio of the calibration sample, detection features involved in the modeling, and response relationship type.

[0079] For detection areas marked as spectrally usable regions and not marked as color interference regions, the spectral-dominant quantitative branch is entered. This branch extracts target band reflectance intensity, target band absorption changes, spectral attenuation characteristics, oxidative aging characteristics, and additive residue characteristics from the regional spectral response data. The extracted results are then input into the blending ratio response relationship to obtain the predicted value of the recycled plastic content for that detection area. The target band is defined as the band where there is a distinguishable response difference between the virgin plastic reference sample and calibration samples with different recycled plastic blending ratios. The distinguishable response difference is determined based on the spectral signal-to-noise ratio threshold for this batch. When the response difference between adjacent blending ratio calibration samples in the target band exceeds the combined fluctuation range formed by the no-load noise response and sensor repeatability error, that band is included in the target band set. The spectral-dominant prediction record includes the detection area number, target band number, spectral characteristic value, corresponding blending ratio response relationship, predicted value of the regional recycled plastic content, and the calibration sample number involved in the calculation.

[0080] For detection areas marked as color interference regions or low-confidence spectral regions, the spectral-dominated quantitative branch is not directly used; instead, the thermal response compensation branch is entered. The thermal response compensation branch first determines whether the detection area meets the thermal response compensation triggering conditions. These conditions include color triggering, spectral triggering, thermal response stability, and thermal response distinguishability. The color triggering condition is when the color dispersion value reaches the color dispersion threshold; the spectral triggering condition is when the spectral signal-to-noise ratio is lower than the spectral signal-to-noise ratio threshold; the thermal response stability condition is when the thermal response repeatability deviation does not exceed the thermal response stability threshold; and the thermal response distinguishability condition is when the difference in thermal response characteristics between adjacent mixing ratio calibration samples is greater than the minimum distinguishable difference in thermal response. When at least one of the color and spectral triggering conditions is met, and both the thermal response stability and distinguishability conditions are met simultaneously, the detection area is confirmed to enter the thermal response compensation branch. When either the thermal response stability or distinguishability condition is not met, the detection area is marked as a low-confidence reference region and is not included in the sample-level content calculation.

[0081] Upon entering the detection area of ​​the thermal response compensation branch, color correction is first performed on the regional spectral response based on the color discrete values. The color correction parameters are formed as follows: images and spectroscopic detection are performed on calibration samples with the same recycled plastic blending ratio but different colors. The mean difference of color channels, standard deviation of brightness, and discrete term of the main color category for each calibration sample are calculated, and regression analysis is performed with the corresponding target band spectral deviation to obtain the spectral response regression coefficients corresponding to the mean difference of color channels, standard deviation of brightness, and discrete term of the main color category, respectively.

[0082] The primary color category discrete term is 1 minus the primary color category area ratio, which is the ratio of the number of primary color category pixels to the number of pixels covered by plastic in the detection area. By using the primary color category discrete term, the color correction parameter and the color discrete value are calculated in the same direction; that is, the more concentrated the primary color categories are, the smaller the primary color category discrete term is; the more dispersed the primary color categories are, the larger the primary color category discrete term is.

[0083] The mean difference of the color channels, the standard deviation of brightness, and the discrete term of the main color category in the test area are multiplied by the corresponding spectral response regression coefficients and summed to obtain the color correction parameters. These parameters correct the shift in color state to the spectral response of the target band. After color correction, the heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time of the test area are extracted. The color-corrected spectral features and thermal response features are then input into the blending ratio response relationship to obtain the predicted value of the recycled plastic content in the test area.

[0084] The color correction parameter record includes the detection area number, color channel mean difference, brightness standard deviation, main color category area ratio, main color category discrete term, spectral response regression coefficient corresponding to each color feature, target band spectral deviation, color correction parameters, and color-corrected spectral features.

[0085] The spectral correction parameters are formed as follows: spectral response data are repeatedly collected for virgin plastic reference samples and calibrated samples of recycled plastic at various proportions. The average deviation and standard deviation of each band are calculated to obtain the band correction coefficient. The spectral response of the area to be measured is multiplied by the correction coefficient to complete the spectral correction.

[0086] The thermal response correction parameters are formed as follows: Thermal response tests are repeatedly performed on the same batch of calibration samples under different thicknesses and colors. Heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time are collected. The difference between the average value of each characteristic's repeated tests and the standard conditions is calculated to obtain the thermal response correction coefficient. The thermal response characteristics of the test area are subtracted from the corresponding correction coefficient, and the blending ratio response relationship is then input to calculate the recycled plastic content.

[0087] The thickness-linked correction parameters are formed as follows: The calibration sample is repeatedly tested under different thickness conditions, and spectral and thermal response characteristics are collected. The difference between the characteristics under each thickness condition and the standard thickness condition is calculated to obtain the thickness correction coefficient. After the thickness deviation value of the area to be tested is obtained, the corresponding thickness correction coefficient is applied to the spectral and thermal response characteristics to achieve thickness-linked correction.

[0088] For the detection area marked as a thickness correction region, thickness linkage correction is performed before entering the spectral-dominant quantitative branch or the thermal response compensation branch. Thickness linkage correction affects both the spectral and thermal responses. The thickness linkage correction parameters are formed by repeated detection data of the same recycled plastic blending ratio calibration strip under different thickness conditions. Specifically, multiple thickness or stacking thickness conditions are set for the same blending ratio calibration strip, and the spectral response, heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time of the target band are repeatedly collected under each thickness condition. The detection results of the calibration strip within the effective thickness range are used as the standard thickness response, and the spectral response offset and thermal response offset under each thickness condition are calculated. The thickness deviation value is then correlated with the spectral response offset and thermal response offset to form the spectral thickness correction parameters and the thermal response thickness correction parameters.

[0089] When the thickness deviation of the current detection area reaches the thickness adaptation threshold but does not exceed the thickness correction upper limit, the spectral thickness correction parameters and thermal response thickness correction parameters corresponding to the thickness deviation value are invoked to normalize and correct the region's spectral response intensity, heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time. If the thickness deviation is below the thickness adaptation threshold, no thickness linkage correction is performed; if the thickness deviation exceeds the thickness correction upper limit, the detection area is not included in the quantitative calculation of this step. The thickness linkage correction record includes the detection area number, detection area thickness, effective thickness range, thickness deviation value, thickness adaptation threshold, thickness correction upper limit, spectral thickness correction parameters, thermal response thickness correction parameters, sensing characteristics before correction, and sensing characteristics after correction.

[0090] For detection areas that simultaneously have thickness correction labels and color interference labels, thickness linkage correction is performed first, and then color verification and thermal response compensation are performed based on the corrected image state and spectral response.

[0091] For detection areas simultaneously labeled with thickness correction and color interference, processing is performed in the following order: thickness-linked correction, color influence verification, and thermal response compensation calculation. First, thickness-linked correction is used to eliminate the combined effect of thickness changes on the spectral and thermal response amplitudes. Then, the color discrete value is recalculated based on the corrected image state and spectral response. If the recalculated color discrete value still reaches the color discrete threshold, the thermal response compensation branch is initiated; if the recalculated color discrete value is lower than the color discrete threshold and the spectral signal-to-noise ratio reaches the spectral signal-to-noise ratio threshold, the spectral-dominant quantitative branch is initiated. This sequence distinguishes between spectral intensity shifts caused by thickness changes and spectral response shifts caused by color state.

[0092] For detection areas marked as surface anomaly verification regions, surface anomaly verification processing is initiated. This processing is performed based on the anomaly type area ratio, the overall anomaly area ratio, the surface anomaly area threshold, and the anomaly removal upper limit. When a surface anomaly region overlaps with a color interference region, the detection region is preferentially entered into the thermal response compensation branch, and the spectral sampling points and image pixels corresponding to the anomaly pixels are removed during the calculation. When a surface anomaly region does not overlap with a color interference region, and the overall anomaly area ratio does not exceed the anomaly removal upper limit, the non-anomaly sampling points of the detection region are retained and entered into the spectral-dominant quantitative branch. When the number of remaining effective sampling points after removing anomaly pixels is lower than the minimum number of effective sampling points for this batch, the detection region is marked as an invalid surface verification region and is not included in the sample-level content calculation. The minimum number of effective sampling points is determined by the prediction deviation of the calibration strip under different sampling point numbers. When the number of sampling points continues to increase while the change in the region's predicted value does not exceed the allowable error of the calibration strip's backcalculation, the corresponding number of sampling points is taken as the minimum number of effective sampling points.

[0093] After surface anomaly verification, the region calculation regenerates spectral response data and image state data based on the retained sampling points. For the spectral response data formed by the retained sampling points, the spectral signal-to-noise ratio (SNR) is recalculated; for the image state data formed by the retained sampling points, the color discrete value and the surface anomaly area ratio are recalculated. When the recalculated spectral SNR reaches the spectral SNR threshold and the color discrete value is lower than the color discrete value threshold, it enters the spectral-dominated quantitative branch; when the recalculated spectral SNR is lower than the spectral SNR threshold or the color discrete value reaches the color discrete value threshold, and the thermal response compensation trigger condition is met, it enters the thermal response compensation branch; when the recalculated data still does not meet the spectral dominance condition and the thermal response compensation condition, the detection area is marked as a low-confidence area for surface verification and is not included in the sample-level content calculation. The surface anomaly verification record includes the detection area number, anomaly type, number of anomaly pixels, location of anomaly pixels, number of effective sampling points after removal, minimum number of effective sampling points, recalculated spectral SNR, recalculated color discrete value, recalculated surface anomaly area ratio, and branch selection result.

[0094] After completing the branch calculations for each detection region, multiple predicted values ​​for the recycled plastic content of each region are obtained. The initial detection value of the recycled plastic content of the sample is not generated through regional weight fusion, but is determined based on the regional prediction values ​​after consistency screening. First, regional dispersion verification is performed on all regional prediction values ​​participating in the sample-level calculation. The regional dispersion verification uses the regional dispersion threshold of this batch, and the regional dispersion index is selected from the range, standard deviation, or coefficient of variation based on the number of regions and detection requirements. When the regional dispersion index does not exceed the regional dispersion threshold, the regional prediction values ​​participating in the sample-level calculation are calculated by arithmetic mean or median to obtain the initial detection value of the recycled plastic content of the sample; when the regional dispersion index exceeds the regional dispersion threshold, the detection region with the largest dispersion contribution is identified and added to the set of regions to be reviewed, and is not included in the calculation of the initial detection value of the recycled plastic content of the current sample.

[0095] After completing the branch calculations for each detection region, multiple predicted values ​​for the recycled plastic content of each region are obtained. The initial detected value of the recycled plastic content of the sample is not generated through regional weight fusion, but is determined based on the regional predicted values ​​after consistency screening. First, regional discreteness verification is performed on all regional predicted values ​​participating in the sample-level calculation. The regional discreteness verification uses the regional discreteness threshold from the current batch threshold group. The regional discreteness threshold is formed by the range, standard deviation, or coefficient of variation of the predicted values ​​of each region in the repeated testing of the calibration strip in multiple regions. The maximum value covering the 95% confidence interval in the repeated testing of different batches is taken as the threshold of this batch, ensuring that the threshold can cover the natural discrete fluctuations of most normal detection regions.

[0096] Outlier identification is based on the natural dispersion range formed by repeated testing of calibration strips across multiple regions. The portion of the deviation between the predicted regional value and the median of the predicted regional values ​​that exceeds the natural dispersion range is marked as an outlier candidate. When the number of outlier candidate values ​​participating in the sample-level calculation does not exceed 20% of the total number of regions, the deviation is considered minor, and the median is used to calculate the initial detection value of the recycled plastic content in the sample. When there are no outlier candidate values, or when the number of outlier candidate values ​​exceeds 20% but the overall regional dispersion index still does not exceed the regional dispersion threshold, the arithmetic mean is used to calculate the initial detection value of the recycled plastic content in the sample. The 20% outlier candidate value threshold is derived from the statistical analysis of repeated testing data of calibration strips: by calculating the maximum deviation ratio among the predicted values ​​of repeated regions of different calibration strips, the deviation ratio at 99% coverage is taken as the upper limit. Statistical results show that the outlier ratio in most sample regions does not exceed 20%, therefore this ratio is used as an feasible judgment threshold to ensure that the detection method is repeatable and stable under actual sample conditions.

[0097] The initial sample-level detection record includes the detection area number involved in the calculation, the state combination of each area, the quantitative branch used for each area, the predicted value of recycled plastic content for each area, the excluded area number, the reason for exclusion, the area dispersion index, the area dispersion threshold, the calculation method of the initial sample-level detection value, and the initial detection value of the sample's recycled plastic content. For detection areas written into the set of areas to be reviewed, the corresponding processing path will be re-triggered in subsequent reverse verification based on color interference labels, thickness correction labels, surface anomaly verification labels, and sensor source conflict situations.

[0098] S5. Perform reverse verification on the initial detection value of recycled plastic content in the sample, and generate quantitative detection results of recycled plastic content and corresponding detection records based on the verification results.

[0099] Furthermore, reverse verification is performed on the initial detection values ​​of recycled plastic content in the samples. Reverse verification includes regional discrete verification, spectral-thermal response direction verification, status label back-check verification, calibration strip back-calculation verification, and result generation processing. Reverse verification uses region binding records, region status combinations, branch calculation records, batch threshold groups, and sample-level initial detection records as data objects. Verification objects include the detection areas involved in the calculation of sample-level initial detection values, excluded areas, the set of areas to be back-checked, and calibration strip back-calculation objects. A reverse verification record is generated during the reverse verification process. This record includes the sample number, detection area number, region status combination, region predicted value, sample-level initial detection value, thresholds used for comparison, threshold source number, verification result, and processing path.

[0100] When performing regional dispersion verification, the predicted values ​​of recycled plastic content in each detection area used in the initial sample-level detection value calculation are read, and the regional dispersion index is calculated. The regional dispersion index includes range, standard deviation, or coefficient of variation. The selection of the regional dispersion index is consistent with the regional dispersion threshold in the current batch threshold group: when the number of areas involved in the calculation is less than the preset number of areas, the range is used as the regional dispersion index; when the number of areas involved in the calculation reaches the preset number of areas, the standard deviation is used as the regional dispersion index; when it is necessary to compare the relative dispersion between different blending ratio ranges or different sample morphologies, the coefficient of variation is used as the regional dispersion index. The preset number of areas is determined by the stability of the predicted values ​​in the repeated detection of multiple areas of the calibration strip. The regional dispersion threshold is determined by the results of repeated detection of multiple areas of the same calibration strip under normal sample preparation and normal detection conditions, and records the calibration strip number, blending ratio, number of detection areas, predicted value of each area, range, standard deviation, coefficient of variation, and the finally adopted regional dispersion threshold.

[0101] When the regional dispersion index is less than or equal to the regional dispersion threshold, the regional dispersion verification result is recorded as passed, and the process proceeds to the spectral-thermal response direction verification. When the regional dispersion index is greater than the regional dispersion threshold, the discrete contribution region is determined based on the deviation between the predicted value of each region and the initial detection value at the sample level. The discrete contribution region is the detection region where the regional predicted value deviates the most from the initial detection value at the sample level. When multiple detection regions with the same deviation exist, the detection regions with color interference labels, thickness correction labels, or surface anomaly labels are preferentially selected for inclusion in the set of regions to be reviewed. The regional dispersion anomaly record includes the regional dispersion index, regional dispersion threshold, predicted value of each region, initial detection value at the sample level, discrete contribution region number, and discrete contribution type.

[0102] Perform a status label lookup for discrete contribution regions. If a discrete contribution region has a color interference label or a low-confidence spectral label, reread the thermal response compensation trigger condition record for that region to determine whether the color trigger condition, spectral trigger condition, thermal response stability condition, and thermal response distinguishability condition are met. The color trigger condition is that the color discrete value reaches the color discrete threshold; the spectral trigger condition is that the spectral signal-to-noise ratio is lower than the spectral signal-to-noise ratio threshold; the thermal response stability condition is that the thermal response repetition detection deviation does not exceed the thermal response stability threshold; the thermal response distinguishability condition is that the difference in thermal response characteristics between adjacent mixing ratio calibration samples is greater than the minimum distinguishable difference in thermal response. When at least one of the color trigger condition and the spectral trigger condition is met, and the thermal response stability condition and the thermal response distinguishability condition are both met, re-execute the thermal response compensation calculation and generate a thermal response compensation lookup record. When the thermal response stability condition or the thermal response distinguishability condition is not met, mark the region as an invalid thermal response compensation region and do not include it in the recalculation.

[0103] If the discrete contribution region has a thickness correction label, reread the thickness deviation value, thickness adaptation threshold, thickness correction upper limit, and thickness linkage correction record. The thickness deviation value is determined by the degree of deviation of the thickness of the detection region from the effective thickness range of similar calibration strips; the thickness adaptation threshold is determined by the repeated detection results of the calibration strip under different thickness conditions; the thickness correction upper limit is determined by the critical value at which the thickness change causes the predicted deviation of recycled plastic content to exceed the allowable error of the calibration strip back-calculation. When the thickness deviation value is still between the thickness adaptation threshold and the thickness correction upper limit, re-execute the thickness linkage correction, and re-enter the corrected spectral response and thermal response into the corresponding quantitative branch. When the thickness deviation value exceeds the thickness correction upper limit, mark the region as a thickness back-check exclusion region and do not include it in the recalculation. The thickness back-check record includes the detection region thickness, effective thickness range, thickness deviation value, thickness adaptation threshold, thickness correction upper limit, predicted region value before correction, and predicted region value after correction.

[0104] If the discrete contribution region has a surface weighting region label or surface anomaly record, the anomaly type area ratio, comprehensive anomaly area ratio, surface anomaly area threshold, anomaly removal upper limit, and anomaly pixel location are reread. The surface anomaly area threshold is determined by repeated detection results of the same material sample with known area scratches, contamination adhesion, foreign object points, or charring marks; the anomaly removal upper limit is determined by the anomaly area ratio where surface anomalies cause simultaneous distortion of spectral and thermal responses and the back-calculation error exceeds the allowable back-calculation error of the calibration sample. When the anomaly type area ratio or comprehensive anomaly area ratio is greater than the anomaly removal upper limit, the region is marked as a surface back-check removal region. When the anomaly type area ratio or comprehensive anomaly area ratio does not exceed the anomaly removal upper limit, the spectral sampling points and image pixels corresponding to the anomaly pixels are removed, the spectral signal-to-noise ratio, color discrete value, and surface anomaly area ratio are recalculated, and the region is entered into the spectral-dominant quantitative branch or thermal response compensation branch based on the recalculation results. The surface retrospective record includes anomaly type, anomaly area ratio, overall anomaly area ratio, anomaly pixel location, number of remaining valid sampling points, minimum number of valid sampling points, recalculated spectral signal-to-noise ratio, recalculated color discrete value, and retrospective branch results.

[0105] After processing the areas to be reviewed, a new set of areas for sample-level calculations is determined. This new set excludes areas with insufficient coverage, areas removed due to thickness review, areas removed due to surface review, areas with invalid thermal response compensation, and areas with invalid surface verification. The regional dispersion index is recalculated for this new set and compared to the regional dispersion threshold. If the recalculated regional dispersion index is less than or equal to the regional dispersion threshold, the corrected sample recycled plastic content detection value is calculated based on the recalculated regional predicted value. If the recalculated regional dispersion index is still greater than the regional dispersion threshold, a sample regional non-uniformity record is generated. This record includes the detection area number participating in the recalculation, the regional predicted value, the regional state combination, the regional dispersion index, the regional dispersion threshold, and the reason for failure.

[0106] After the regional discrete verification passes or the regional discrete back-lookup process is completed, a spectral-thermal response direction verification is performed. The spectral-thermal response direction verification reads the region records in the detection region participating in the sample-level calculation that simultaneously possess spectral and thermal response characteristics. The direction determination is based on calibration strips with known recycled plastic blending ratios. If the target spectral characteristic increases with the increase of the recycled plastic blending ratio in the calibration strip, the direction of increase of the target spectral characteristic is encoded as positive; if the target spectral characteristic decreases with the increase of the recycled plastic blending ratio in the calibration strip, the direction of increase of the target spectral characteristic is encoded as negative. The thermal response characteristics are coded in the same way. The target thermal response characteristics include at least one of the following: heating rate, steady-state temperature difference, cooling slope, and thermal diffusion time.

[0107] It should be noted that the direction of change of the spectral features of the area to be calibrated relative to the virgin plastic reference response is compared with the spectral direction code, and the direction of change of the thermal response features relative to the virgin plastic reference response is compared with the thermal response direction code. When both the spectral feature direction and the thermal response feature direction are consistent with their corresponding direction codes, it is considered that the sensing source direction is consistent. When any spectral feature direction is opposite to any thermal response feature direction, the directional deviation magnitude is calculated. The directional deviation magnitude is determined based on the difference between the normalized change of the spectral feature and the normalized change of the thermal response feature, or based on the proportion of the number of features with opposite directions to the total number of features participating in the directional calibration. The normalized change is based on the response difference between the virgin plastic reference response and the calibration response of adjacent blending ratios. The spectral-thermal response direction conflict threshold is determined by the allowable normal directional deviation range in the calibration sample, and records the spectral feature change direction, thermal response feature change direction, direction code, normalized change, and normal deviation range for each blending ratio.

[0108] When the directional deviation is less than or equal to the spectral-thermal response directional conflict threshold, the spectral-thermal response directional verification result is recorded as passed. When the directional deviation is greater than the spectral-thermal response directional conflict threshold, a sensor source conflict record is generated, and the conflict area is written to the status tag lookup set. The sensor source conflict record includes the detection area number, target spectral characteristics, target thermal response characteristics, spectral direction code, thermal response direction code, spectral normalization change, thermal response normalization change, directional deviation magnitude, directional conflict threshold, and conflict type.

[0109] A status label lookup verification is performed on the conflict region of the sensor source. If the conflict region has a color interference label, the region prediction value formed by the thermal response compensation branch is used first, and the region prediction value formed by the dominant spectral branch is saved as the conflict reference value. If the conflict region has a thickness correction label, the thickness linkage correction is re-executed, and the orientation verification is re-performed using the corrected spectral and thermal responses. If the conflict region has a surface anomaly label, the sampling points corresponding to the abnormal pixels are removed, the spectral and thermal response features are recalculated, and the orientation verification is re-performed. If the conflict region does not have a color interference label, thickness correction label, or surface anomaly label, the region is marked as a sensor source conflict region and is not included in the sample-level quantitative result calculation. A sensor source conflict region refers to a detection region that does not have a color interference label, thickness correction label, or surface anomaly label, but whose spectral feature direction conflicts with the thermal response feature direction and reaches the orientation conflict threshold.

[0110] After completing the status label back-check verification, a new set of valid result regions is formed based on the detection regions that have passed the regional discrepancy verification and spectral-thermal response direction verification. The predicted values ​​of the regions in the valid result region set are used to generate the verified sample recycled plastic content detection value. When there are no outliers in the predicted values ​​of the regions in the valid result region set, the arithmetic mean is used to generate the verified sample recycled plastic content detection value. When there are one or a few outlier candidate values, but the regional discrepancy index of the valid result region set still does not exceed the regional discrepancy threshold, the median is used to generate the verified sample recycled plastic content detection value. Outlier candidate values ​​are determined based on the natural discrepancy range formed by repeated detection of multiple regions of the calibration strip. A deviation between the predicted regional value and the median of the predicted regional values ​​exceeding the natural discrepancy range is recorded as an outlier candidate value.

[0111] When performing calibration strip back-calculation verification, the calibration strips with known recycled plastic blending ratios for this batch are input into the same testing process as the plastic samples to be tested. Multi-source sensor data acquisition, region state determination, branch quantitative calculation, and reverse verification are completed sequentially to obtain the back-calculated content of the calibration strips. The calibration strip back-calculation deviation is the absolute difference between the back-calculated content of the calibration strips and the corresponding known recycled plastic blending ratio. The allowable error for calibration strip back-calculation is consistent with the allowable error for testing, determined by the maximum value among the target accuracy of the testing method, the calibration strip preparation error, and the sensor repeatability error. The calibration strip back-calculation deviation is compared with the allowable error for calibration strip back-calculation. When the back-calculation deviations of all calibration strips are less than or equal to the allowable error for calibration strip back-calculation, the testing process for this batch is validly recorded. When the back-calculation deviation of any calibration strip is greater than the allowable error for calibration strip back-calculation, a testing failure record for this batch is generated.

[0112] This batch of test failure records includes the calibration sample number, known recycled plastic blending ratio, calculated content of the calibration sample, back-calculation deviation, allowable error of calibration sample back-calculation, threshold group number involved in the back-calculation, and reason for failure. Reasons for failure include at least one of the following: light source output fluctuation exceeding the stable range of light source output; thermal excitation output fluctuation exceeding the thermal response stability threshold; thickness detection repeatability deviation exceeding the allowable range of thickness detection; mismatch of color dispersion threshold; abnormal back-calculation deviation of blending ratio response relationship; and failure to resolve sensor source direction conflict. After this batch of test failure records is generated, the verified content detection value of the plastic sample to be tested is marked as the failure result of this batch, and a recalibration prompt or a resample preparation prompt is generated.

[0113] When the regional discreteness verification, spectral-thermal response direction verification, status label back-check verification, and calibration strip back-calculation verification are all passed, the quantitative detection result of recycled plastic content is output. The quantitative detection result of recycled plastic content includes the sample number, plastic type, sample form, source batch, verified recycled plastic content detection value, unit of measurement for the detection result, number of regions involved in the calculation, number of regions removed, number of regions reviewed, set of valid result regions, status combination of each region, quantitative branch of each region, predicted value of each region, regional discreteness index, regional discreteness threshold, spectral-thermal response direction verification result, calibration strip back-calculation verification result, and threshold group number for this batch. The unit of measurement for the detection result is mass percentage.

[0114] When the regional discreteness verification fails and a valid result set cannot be formed after review, a low-confidence detection record is output. The low-confidence detection record includes the sample number, regional discreteness index, regional discreteness threshold, set of regions to be reviewed, reason for failure, and re-detection prompt. When the spectral-thermal response direction verification fails and a sensor source conflict region still exists after reviewing the status label, a sensor source conflict record and a re-sample preparation prompt are output. When the calibration strip back-calculation verification fails, a batch-wide detection failure record and a recalibration prompt are output. The re-detection prompt, re-sample preparation prompt, and recalibration prompt are each linked to and saved with their corresponding exception records.

[0115] The detection reliability is generated based on the reverse verification results. The detection reliability is jointly determined by the number of valid result areas, the regional discrete verification results, the spectral-thermal response direction verification results, the status label back-check results, and the calibration spline back-calculation verification results. The detection reliability record is valid when the number of valid result areas reaches the minimum number of detection areas, the regional discrete index is less than or equal to the regional discrete threshold, the spectral-thermal response direction verification passes, there are no unprocessed conflict areas after the status label back-check, and the calibration spline back-calculation deviation is less than or equal to the allowable error of the calibration spline back-calculation. If any condition is not met, a corresponding reliability description is generated based on the unmet item. The detection reliability record includes the number of valid result areas, the minimum number of detection areas, the regional discrete verification results, the direction verification results, the back-check processing results, the back-calculation verification results, and the reliability description. This ultimately forms the detection report data.

[0116] This embodiment also provides a computer device applicable to the quantitative detection method of recycled plastics based on intelligent sensing, including: a memory and a processor; the memory is used to store computer-executable instructions, and the processor is used to execute the computer-executable instructions to realize the quantitative detection method of recycled plastics based on intelligent sensing as proposed in the above embodiment.

[0117] The computer device can be a terminal, comprising a processor, memory, communication interface, display screen, and input devices connected via a system bus. The processor provides computing and control capabilities. The memory includes non-volatile storage media and internal memory. The non-volatile storage media stores the operating system and computer programs. The internal memory provides an environment for the operation of the operating system and computer programs stored in the non-volatile storage media. The communication interface is used for wired or wireless communication with external terminals; wireless communication can be achieved through Wi-Fi, carrier networks, NFC (Near Field Communication), or other technologies. The display screen can be an LCD screen or an e-ink screen. The input devices can be a touch layer covering the display screen, buttons, a trackball, or a touchpad on the computer device's casing, or an external keyboard, touchpad, or mouse.

[0118] This embodiment also provides a storage medium storing a computer program, which, when executed by a processor, implements the method for quantitative detection of recycled plastics based on intelligent sensing as proposed in the above embodiments. The storage medium can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0119] In summary, this invention, through the steps of acquiring the plastic sample to be tested and constructing the calibration strip and threshold group for this batch, achieves comprehensive collection and recording of the plastic type, morphology, color, thickness, surface condition, and testing environment of the plastic sample to be tested. It combines virgin plastic reference strips, calibration strips with known recycled plastic blending ratios, no-load response, and repeated test results to form a systematic threshold group and blending ratio response relationship for subsequent multi-source data analysis and quantitative calculation. This ensures that the characteristic parameters of each region during the testing process are traceable, quantifiable, and comparable, providing a reliable foundation for accurately determining the recycled plastic content. Through a progressive state judgment step based on the threshold group and binding records of the regions, it achieves stratified judgment of the detection area coverage effectiveness, spectral effectiveness, color interference, thickness deviation, and surface anomalies. Furthermore, through interactive mechanisms such as triggered thermal response compensation and thickness linkage correction, it provides targeted processing for abnormal regions, thereby ensuring the reliability and consistency of quantitative results in each region. Simultaneously, it improves the accuracy and stability of sample-level recycled plastic content detection, ensuring the effectiveness and scientific validity of the test results under complex sample conditions.

[0120] It should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and are not intended to limit it. Although the present invention has been described in detail with reference to preferred embodiments, those skilled in the art should understand that modifications or equivalent substitutions can be made to the technical solutions of the present invention without departing from the spirit and scope of the technical solutions of the present invention, and all such modifications or substitutions should be covered within the scope of the claims of the present invention.

Claims

1. A method for quantitative detection of recycled plastics based on intelligent sensing, characterized in that, include: Obtain sample information and testing environment information of the plastic sample to be tested, confirm the type of plastic in the plastic sample to be tested, and generate threshold groups and blending ratio response relationships corresponding to this batch of tests based on the reference sample of virgin plastic of the same type, the calibration sample of known recycled plastic blending ratio, no-load sensor response and repeated test results. The plastic sample to be tested is divided into testing areas. Spectral response data, image status data, thermal response data, and geometric status data of each testing area are collected and bound according to the testing area number to form an area binding record. Based on threshold groups, progressive state determination is performed on region-bound records to form region state combinations; Select the corresponding quantitative calculation path according to the combination of regional states, perform spectral-dominated calculation, thermal response compensation calculation, thickness linkage correction calculation or surface anomaly verification calculation on the effective detection area, output the predicted value of the recycled plastic content in the area, and generate the initial detection value of the recycled plastic content of the sample based on the predicted value of the area after regional consistency screening. The initial detection value of recycled plastic content in the sample is reverse-verified, and the quantitative detection result of recycled plastic content and the corresponding detection record are generated based on the verification result.

2. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 1, characterized in that, The generation of the threshold group and blending ratio response relationship corresponding to this batch of detections includes: The type of plastic in the plastic sample to be tested was identified based on the sample source record and the characteristic bands of the near-infrared spectroscopy resin. Select or prepare reference specimens of virgin plastics that are the same type of plastic as the plastic sample to be tested and calibration specimens with known blending ratios of recycled plastics; Repeated tests were performed on virgin plastic reference specimens, known recycled plastic blending ratio calibration specimens, and no-load test conditions to obtain no-load noise response, virgin plastic repeated response, calibration response for different blending ratios, and sensor repeated test error. The allowable detection error is determined based on the target accuracy of the detection method, the calibration strip preparation error, and the sensor repeatability error, and a threshold group is generated based on the allowable detection error; Based on the detection characteristics of the calibration strips with known recycled plastic blending ratios and the corresponding known recycled plastic blending ratios, a blending ratio response relationship is established. When a single detection feature exhibits monotonic variation under various known recycled plastic blending ratios, a piecewise interpolation relationship is used to determine the blending ratio response relationship. When at least two detection features are involved in quantitative calculation, partial least squares regression is used to determine the mixing ratio response relationship. In piecewise interpolation, a local linear fitting relationship is used between adjacent calibration points to determine the response change within the corresponding proportional interval.

3. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 2, characterized in that, The threshold group includes: Coverage ratio threshold, spectral signal-to-noise ratio threshold, color discretization threshold, thickness adaptation threshold, surface anomaly area threshold, anomaly removal upper limit, thermal response compensation trigger condition, region discretization threshold, spectral and thermal response direction conflict threshold, and calibration spline back calculation allowable error; The coverage ratio threshold is determined based on the spectral response stability corresponding to changes in the sensor's effective field of view, the spectral sampling spot area, and the sample coverage area. The spectral signal-to-noise ratio threshold is determined based on the no-load noise response, the fluctuation of repeated detection of the virgin plastic reference sample, and the minimum spectral response difference between adjacent blending ratio calibration samples. The color discrete threshold is determined based on the spectral quantitative deviation caused by calibration samples with the same recycled plastic blending ratio but different color states; The thickness adaptation threshold is determined based on the changes in spectral and thermal response of the calibration strip under different thickness conditions; The surface anomaly area threshold is determined based on the deviations in spectral response, image response, and thermal response caused by a spline of the same material with a known area surface anomaly. The upper limit for anomaly removal is determined based on the proportion of the anomaly area when the surface anomaly causes simultaneous distortion of the spectral and thermal responses and the back-calculation deviation of the calibration sample exceeds the allowable error of the back-calculation of the calibration sample. The upper limit for anomaly removal is greater than the surface anomaly area threshold and not greater than 1. The allowable error for calibration spline back calculation is set to correspond to the allowable error for detection.

4. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 3, characterized in that, The process of binding region binding records according to detection region numbers includes: The plastic sample to be tested is divided into testing areas according to the effective detection area of ​​the sample, the field of view of a single sensor, and the minimum number of detection areas. The spectral response data of the detection area is collected by the near-infrared spectral sensing unit, the image state data of the detection area is collected by the visible light image sensing unit, the thermal response data of the detection area is collected by the thermal response sensing unit, and the geometric state data of the detection area is collected by the thickness detection unit. The spectral response data, image state data, thermal response data, and geometric state data within the same detection area are bound according to the detection area number; The minimum number of detection areas is determined by the results of repeated testing of multiple areas of the calibration sample. When the number of detection areas is increased, if the change in the predicted value of recycled plastic content of the calibration sample does not exceed the allowable error of the calibration sample back calculation, the corresponding number of detection areas will be used as the minimum number of detection areas.

5. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 1, 3, or 4, characterized in that, The progressive state determination of region-bound records based on threshold groups includes: The sample coverage ratio of the detection area is calculated based on the image status data. The sample coverage ratio is compared with the coverage ratio threshold to determine the effective coverage area or the insufficient coverage area. For the effective coverage area, the spectral signal-to-noise ratio is calculated based on the spectral response data. The spectral signal-to-noise ratio is then compared with the spectral signal-to-noise ratio threshold to determine the spectral usable area or the spectral low confidence area. The color discrete value is calculated based on the image state data, and the color discrete value is compared with the color discrete threshold to determine the color interference area. The thickness deviation value is calculated based on the geometric state data, and the thickness deviation value is compared with the thickness adaptation threshold to determine the thickness correction area or the thickness rejection area. The surface anomaly area ratio is calculated based on the image state data, and then compared with the surface anomaly area threshold and the anomaly removal upper limit. When the proportion of surface abnormal area reaches the surface abnormal area threshold but does not exceed the upper limit of abnormal removal, it is identified as a surface abnormality review area. When the proportion of abnormal surface area exceeds the upper limit of abnormal rejection, it is identified as a surface rejection area.

6. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 5, characterized in that, The progressive state determination of region-bound records based on threshold groups includes: The color discrete value is obtained by normalizing and weighting the color channel mean difference, brightness standard deviation and main color category discrete term within the detection area. The main color category is determined by clustering the hue, saturation and brightness distribution of pixels in the detection area in the HSV color space. The main color category discrete term is one minus the area ratio of the main color category. The thickness deviation value is determined according to the degree of deviation of the thickness of the detection area from the effective thickness range of the same type of calibration strip. When the thickness of the detection area is within the effective thickness range, the thickness deviation value is zero. When the thickness of the detection area is lower than the lower limit of the effective thickness range or higher than the upper limit of the effective thickness range, the thickness deviation value is the difference between the corresponding boundary and the thickness of the detection area.

7. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 6, characterized in that, The process of generating the initial detection value of recycled plastic content in the sample based on the regional prediction value after regional consistency screening includes: When the detection area is determined to be a spectrally usable area and not a color interference area, the spectral-driven calculation is entered, and the predicted value of the recycled plastic content in the area is calculated based on the spectral attenuation characteristics, oxidative aging characteristics and additive residue characteristics. When the detection area is determined to be a color interference area or a low-confidence spectral area, it is determined whether the detection area meets the thermal response compensation trigger condition. If it does, the thermal response compensation calculation is entered. The color-corrected spectral characteristics and thermal response characteristics are jointly input into the blending ratio response relationship to calculate the predicted value of the recycled plastic content in the area. When the detection area is determined to be a thickness correction area, thickness linkage correction is performed on the spectral response and thermal response before entering the spectral-dominant calculation or thermal response compensation calculation. When the detection area is determined to be a surface anomaly verification area, the sampling points corresponding to the abnormal pixels are removed, and the spectral signal-to-noise ratio, color discrete value, and surface anomaly area ratio are recalculated. Based on the recalculation results, the calculation is then performed either spectral-dominant calculation or thermal response compensation calculation.

8. The method for quantitative detection of recycled plastics based on intelligent sensing as described in claim 7, characterized in that, The reverse verification of the initial detection value of the recycled plastic content in the sample includes: Based on the initial detection value of recycled plastic content in the participating samples, the predicted value of regional recycled plastic content is calculated, the regional dispersion index is calculated, and the regional dispersion index is compared with the regional dispersion threshold. Outlier candidate regions are determined based on the natural discrete range formed by repeated detection of multiple regions of calibration strips, and the initial detection value of recycled plastic content in the sample is determined based on the proportion of outlier candidate regions. For detection areas that simultaneously possess spectral and thermal response characteristics, spectral and thermal response directions are verified. When the deviation between the spectral characteristic change direction and the thermal response characteristic change direction exceeds the spectral and thermal response direction conflict threshold, a sensor source conflict record is generated. Based on the combination of regional states of discrete contribution areas or conflicting sensor source areas, perform status label back-check verification, and re-trigger thermal response compensation calculation, thickness linkage correction calculation, or surface anomaly verification calculation based on the combination of regional states. If the number of status tag re-checks does not exceed the preset maximum number of re-checks, and there are still discrete contribution areas or sensor source conflict areas after the maximum number of re-checks is reached, the corresponding detection area will be marked as a low confidence area or a rejection area, and will not be included in the calculation of the initial detection value of the sample's recycled plastic content. Input the calibration sample with the known blending ratio of recycled plastics into the same test process for calibration sample back-calculation verification. When the back-calculation deviation of the calibration sample does not exceed the allowable error of the back-calculation of the calibration sample, output the quantitative test result of the recycled plastic content and the corresponding test record. When the back-calculation deviation of the calibration sample exceeds the allowable error of the calibration sample back-calculation, a failure record for this batch of tests is generated, and a prompt for recalibration or retesting is output. The quantitative test result of the recycled plastic content is not output.

9. A computer device comprising a memory and a processor, wherein the memory stores a computer program, characterized in that, When the processor executes the computer program, it implements the steps of the method for quantitative detection of recycled plastics based on intelligent sensing as described in any one of claims 1 to 8.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by the processor, it implements the steps of the method for quantitative detection of recycled plastics based on intelligent sensing as described in any one of claims 1 to 8.

Images