An image processing-based automatic identification and classification method for recycled material

Abstract

The present application relates to the technical field of image processing, in particular to a kind of recycling material quality automatic identification and classification method based on image processing, by detecting old household appliance panel lap joint boundary, extract the mean difference of yellow and blue channel of exposure side and shielding side, construct net yellowing amount difference signal, and take shielding side color as original color reference datum;Combined with the anisotropic characteristics of texture, ABS main body and PP ventilation structure area segmentation is realized, the net yellowing amount of each material and the partition confidence are obtained;Further calculate the cross-material chroma shift direction angle difference to construct degradation path characteristics, form two-dimensional feature vector with net yellowing amount and map to pre-built age characteristic space to complete manufacturing age classification;Through the sufficiency of data determined by coverage index, when the condition is met, output deterministic classification result, otherwise output interval age range and retake guide information.The method improves the accuracy and reliability of identification and classification in multi-material mixed scene.

Description

Technical Field

[0001] This invention relates to the field of image processing technology, specifically to an automatic identification and classification method for recyclable materials based on image processing. Background Technology

[0002] With the advancement of digital management and control of old home appliance recycling platforms, the use of image processing technology for automated analysis of recyclables has become an important means to improve management efficiency.

[0003] In the field of image processing, existing methods for extracting and analyzing the color features of target objects typically use color statistics of global or local regions of an image as feature descriptors, and achieve target state classification and recognition through color space transformation and regional feature modeling. However, in the scenario of image inference of the manufacturing age of old home appliances, the color value of the outer shell surface is composed of the sum of the initial color design of the product and the amount of yellowing due to aging during use. Existing color feature extraction methods cannot separate the two from the image observation of a single region, resulting in the extracted color features failing to accurately encode age information. Although structurally occluded areas on the same device retain color information close to the initial state in the image, theoretically they can be used as a reference for cross-regional difference calculation to eliminate the interference of the initial color, existing image segmentation and region detection methods are designed for target regions with clear boundaries and sufficient area. For occluded regions, which present characteristics such as extremely small area in the image, semi-occluded gradient transitions at the boundaries, and generally being in a low-resolution out-of-focus state, existing methods cannot reliably locate and effectively extract color information for such regions, making it difficult to effectively implement age inference methods based on cross-regional color differences at the image processing level.

[0004] To address this, an automatic identification and classification method for recyclable materials based on image processing is proposed. Summary of the Invention

[0005] The purpose of this invention is to provide an automatic identification and classification method for recycled materials based on image processing. By extracting multi-material features through the difference in net yellowing at overlapping boundaries and texture anisotropy segmentation, and combining cross-material degradation paths and coverage indices, the method can achieve the output of manufacturing age grading and material classification.

[0006] To achieve the above objectives, the present invention provides the following technical solution: An automatic identification and classification method for recyclable materials based on image processing includes: Acquire images of old home appliances, detect panel overlap boundaries, extract the mean difference in yellow and blue channels between the exposed and shaded areas of the same material at the overlap boundary, obtain the net yellowing difference signal, and use the mean color of the shaded side as the original color reference benchmark of the material. Texture anisotropic features are extracted from local regions of the image. Material segmentation is performed on the PP ventilation structure region with periodic grid texture and the ABS main body region with isotropic random texture. Combined with the original color reference benchmark, the net yellowing amount and partition confidence of each material region are extracted. Using the original color reference as the starting point, the offset direction angle of the current color of ABS and PP relative to their respective references in the chromaticity plane is calculated to obtain the cross-material degradation path characteristics. A two-dimensional feature vector is constructed based on the net yellowing differential signal and cross-material degradation path characteristics, and mapped to a pre-built age feature space to complete the manufacturing age classification; the coverage index is obtained based on the number of effective overlap boundaries and the confidence of the partition; when the coverage index is insufficient, the age range of the interval is output and supplementary shooting guidance is generated; the iteration terminates when the deterministic classification conditions are met, and the material classification results of the recycled materials are output.

[0007] Preferably, the process of acquiring the net yellowing difference signal includes: extracting the cross-boundary pixel sequence along the direction perpendicular to the overlap boundary; using the stability of the brightness channel value in each side pixel region as a criterion, defining stable pixel bands on both the exposure side and the occlusion side of the boundary, and excluding mixed pixels with continuously changing brightness in the boundary transition region; after removing outliers from the yellow and blue channel values ​​of all pixels in each stable pixel band, taking the average value to obtain the average yellow and blue channel values ​​on the exposure side and the occlusion side, and using the average yellow and blue channel value on the occlusion side as the original color reference benchmark for the material; subtracting the average yellow and blue channel values ​​on the exposure side and the occlusion side to obtain the single net yellowing difference at the overlap boundary; taking the median of the single net yellowing differences of all effective overlap boundaries in the image, and outputting the net yellowing difference signal.

[0008] Preferably, the material segmentation process includes: dividing the image into several local region blocks; statistically analyzing the gray-level difference intensity of adjacent pixels in multiple directions for each region block; calculating the normalized difference ratio using the sum of the maximum and minimum gray-level difference intensities in each direction as the denominator and the difference between the maximum and minimum values ​​as the numerator, which is used as the texture anisotropy feature value of the region block; and recording the direction corresponding to the maximum gray-level difference intensity as the main texture direction of the region block; marking the set of region blocks whose anisotropy feature values ​​exceed the discrimination boundary and whose main texture directions of spatially adjacent region blocks are highly consistent as ventilation structure candidate regions; and marking the set of region blocks whose anisotropy feature values ​​are below the discrimination boundary and have no stable main texture direction as main shell candidate regions; performing connectivity analysis on each candidate region, merging spatially adjacent region blocks of the same type, and removing isolated fragment regions with too small an area to obtain the ventilation structure material segmentation region and the main shell material segmentation region; and using the consistency ratio of the texture classification judgment results of all region blocks within each material segmentation region as the partition confidence of the segmentation region.

[0009] Preferably, the process of obtaining the net yellowing amount and zoning confidence of each material region includes: for the segmented region of the main body shell material, extracting the yellow and blue channel values ​​of all pixels and taking the average value, using the original color reference benchmark of the main body shell material as the starting point, calculating the difference between the current average value of the yellow and blue channels and the original color reference benchmark to obtain the net yellowing amount of the main body shell material; performing the same process for the segmented region of the ventilation structure material: when there is a corresponding effective overlapping boundary in the segmented region, using the average value of the yellow and blue channels on the boundary occlusion side as the original color reference benchmark of the ventilation structure material, calculating the net yellowing amount of the ventilation structure material; when there is no corresponding effective overlapping boundary in the segmented region of the ventilation structure material, marking the net yellowing amount of the ventilation structure material as missing, and setting the zoning confidence of the region to the lowest value, triggering supplementary shooting guidance by the coverage index mechanism.

[0010] Preferably, the process of obtaining cross-material degradation path features includes: using the coordinates of the original color reference of the main shell material in the chromaticity plane as the starting point of the color offset of the main shell material, and using the average value of the current chromaticity plane coordinates of the segmented region of the main shell material as the offset endpoint, calculating the offset direction angle of the offset vector relative to the positive yellow-blue axis in the chromaticity plane to obtain the offset direction angle of the main shell material; using the coordinates of the original color reference of the ventilation structure material in the chromaticity plane as the starting point of the offset, and using the average value of the current chromaticity plane coordinates of the segmented region of the ventilation structure material as the offset endpoint, calculating the offset direction angle of the ventilation structure material in the same way; calculating the difference between the offset direction angle of the main shell material and the offset direction angle of the ventilation structure material to obtain the cross-material offset direction angle difference value, which is used as the cross-material degradation path feature; when there are multiple sets of valid material pairs in the image, taking the average value of the cross-material offset direction angle difference values ​​of each set, and using the dispersion of each set of differences as the reliability index of the feature.

[0011] Preferably, the process of obtaining the coverage index includes: taking the ratio of the number of effective overlapping boundaries in the image from which the net yellowing differential signal is successfully extracted to the number of boundaries required for sufficient coverage as the overlapping coverage component; taking the smaller value of the confidence scores of the segmented areas of the main body shell material and the segmented areas of the ventilation structure material as the segmented area component; taking the smaller value of the overlapping coverage component and the segmented coverage component as the coverage index; when the coverage index reaches the sufficient coverage determination threshold, determining that the current image data meets the deterministic grading conditions; when the coverage index is lower than the sufficient coverage determination threshold, statistically analyzing the image orientation distribution of the missing effective overlapping boundaries in the image, and generating supplementary shooting guidance information output based on the image orientation description.

[0012] Preferably, the method for obtaining the material classification results of recyclables includes: constructing a two-dimensional feature vector by combining the net yellowing difference signal and cross-material degradation path features; mapping the two-dimensional feature vector to a pre-built age feature space, which is constructed based on the two-dimensional feature vectors corresponding to old home appliance samples with known manufacturing years, and dividing the space into several age interval clusters by the time nodes corresponding to major updates of material stabilizer formulas in different historical stages of the home appliance manufacturing industry; calculating the distance between the current two-dimensional feature vector and the cluster centers of each age interval, and outputting the age interval with the smallest distance as the manufacturing year classification result when the coverage index meets the deterministic classification conditions; outputting the age range result of the closest adjacent age intervals as the age range result when the coverage index still does not meet the deterministic classification conditions after the number of supplementary shooting guidance times reaches the upper limit; and performing a probabilistic mapping between the manufacturing year classification result and the corresponding hazardous substance content level of each age interval to output the material classification result of recyclables and the corresponding confidence level.

[0013] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention detects the overlap boundary of old home appliance panels and extracts the mean difference of the yellow and blue channels in the same material area on the exposed and shaded sides. Using the shaded side as the original color reference, a net yellowing difference signal is constructed. This method utilizes the naturally existing "light-shielded" contrast relationship within the same component, avoiding reliance on external standard color cards or historical databases, and achieving self-calibration analysis of single samples. This invention can effectively suppress interference caused by differences in shooting lighting, white balance drift, and batch color differences, making the extraction of aging features based on the relative changes of the material itself, significantly improving the stability and repeatability of manufacturing age grading and material identification.

[0014] 2. This invention extracts texture anisotropy features by calculating the normalized difference ratio of local grayscale difference intensity and combining it with connectivity analysis based on the consistency of the main texture direction. This enables automatic segmentation of PP ventilation structures with periodic grid textures and ABS main shells with isotropic random textures. This solution can reliably distinguish materials based on structural texture differences even when colors are similar or aging synchronously, avoiding misclassification due to similar yellowing levels. Furthermore, the segmentation reliability is quantified through partitioned confidence metrics, providing a reliable basis for subsequent age determination and result output, ensuring strong robustness even under complex backgrounds and partial occlusion conditions.

[0015] 3. This invention constructs a two-dimensional feature vector by combining the net yellowing difference signal with the cross-material offset direction angle difference, and maps it to a pre-built age feature space for grading judgment. Simultaneously, it constructs a coverage index by combining the overlap coverage component and the partition coverage component to determine data sufficiency. When information is insufficient, it outputs the age range of the interval and generates supplementary image guidance, rather than forcibly providing a deterministic result. This invention establishes a "feature sufficiency—result determinism" linkage mechanism, which not only improves the grading accuracy but also allows for controlled downgrading output in uncertain situations, enhancing the reliability and engineering applicability of the system in practical recycling and sorting applications. Attached Figure Description

[0016] Figure 1 A schematic diagram of the process for an automatic identification and classification method for recyclable materials based on image processing provided by the present invention; Figure 2 A schematic diagram of the material segmentation logic flow provided by the present invention; Figure 3 This is a schematic diagram of the cross-material degradation path feature extraction process provided by the present invention. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of this invention clearer, the invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are for illustrative purposes only and are not intended to limit the invention.

[0018] Example 1: Over long-term use, the casing material of old home appliances undergoes photo-oxidative degradation due to continuous exposure to ultraviolet radiation, manifesting as a yellowing of the casing color. The degree of yellowing is determined by both the material's inherent ultraviolet sensitivity and the cumulative ultraviolet radiation exposure. These two factors are superimposed in color observations of a single surface and cannot be separated by single-point color readings. The overlapping and shielded areas on the casing, formed by the stacking of upper panels, receive significantly less ultraviolet radiation than the fully exposed outer areas, retaining color information close to their original factory condition. Since the exposed and shielded sides are made of the same material from the same batch of injection molding, their initial colors are identical. Under typical mass production conditions, the initial color difference between different parts of the same batch of injection molded parts is far less than the color change caused by long-term ultraviolet aging. Therefore, by subtracting the average colors of the exposed and shielded sides, the initial color deviation can be offset within an acceptable error range. The resulting difference value only encodes the net yellowing amount of the material over the entire service life of the device, excluding interference from differences in product color design.

[0019] Please see Figure 1This invention provides an automatic identification and classification method for recyclable materials based on image processing. The technical solution is as follows: Acquire images of old home appliances; detect panel overlap boundaries; extract the mean difference in yellow and blue channels between the exposed and shaded sides of the overlap boundary for the same material area, obtaining a net yellowing difference signal; use the mean color of the shaded side as the original color reference benchmark for the material; extract texture anisotropy features from local image regions; segment the material for the PP ventilation structure area with periodic grid texture and the ABS main body area with isotropic random texture; and extract the material regions based on the original color reference benchmark. Net yellowing amount and zoning confidence; using the original color reference as the color starting point, calculate the offset direction angle of the current color of ABS and PP relative to their respective references in the chromaticity plane to obtain cross-material degradation path features; construct a two-dimensional feature vector based on the net yellowing amount difference signal and cross-material degradation path features, and map it to the pre-built age feature space to complete the manufacturing age classification; obtain the coverage index based on the number of effective overlap boundaries and zoning confidence; when the coverage index is insufficient, output the age range of the interval and generate supplementary shooting guidance; terminate the iteration when the deterministic classification conditions are met, and output the material classification results of the recycled materials.

[0020] Furthermore, the acquisition process of the net yellowing difference signal includes: extracting the cross-boundary pixel sequence along the direction perpendicular to the overlap boundary; using the stability of the brightness channel value in each side pixel region as a criterion, defining stable pixel bands on both the exposure side and the occlusion side of the boundary, and excluding mixed pixels with continuously changing brightness in the boundary transition region; after removing outliers from the yellow and blue channel values ​​of all pixels in each stable pixel band, taking the average value to obtain the average yellow and blue channel values ​​on the exposure side and the occlusion side, and using the average yellow and blue channel value on the occlusion side as the original color reference benchmark for the material; subtracting the average yellow and blue channel values ​​on the exposure side and the occlusion side to obtain the single net yellowing difference at the overlap boundary; taking the median of the single net yellowing differences of all effective overlap boundaries in the image, and outputting the net yellowing difference signal.

[0021] Specifically, after acquiring images of old home appliances, the first step is to detect panel overlap boundaries in the images. This involves first performing edge enhancement filtering and brightness normalization on the image, then calculating brightness gradient projection curves along multiple candidate directions. Lines or near-line positions where the gradient projection curve shows abrupt amplitude changes within a local window and where the texture statistical features on both sides are continuous are considered candidate overlap boundaries. In the image, overlap boundaries appear as straight-line brightness steps with a certain orientation; the surface texture on both sides of the step is continuous, but there are observable differences in brightness. For each detected overlap boundary, a cross-boundary pixel sequence is extracted along a direction perpendicular to the boundary's orientation.

[0022] In a uniform color space, the luminance channel values ​​of cross-boundary pixel sequences are analyzed pixel-by-pixel along the sequence direction. The luminance channel values ​​within the transition zone of the overlapping boundary continuously change due to geometric occlusion. After leaving the transition zone, the luminance channel values ​​of the exposed pixels tend to stabilize, and the luminance channel values ​​of the occluded pixels also tend to stabilize. Based on the condition that the variation in luminance channel values ​​within a consecutive number of pixels is lower than the stability threshold, stable pixel bands are defined on both the exposed and occluded sides, excluding mixed pixels in the boundary transition region where luminance is still changing.

[0023] The stability threshold is determined as follows: In the sample images of old home appliances of known age used to establish the age feature space, the sample images should cover the main brands, main shell color types and typical shooting lighting conditions. In the homogeneous surface area far from the overlapping boundary in the sample image, the natural fluctuation range of the pixel brightness channel value in space is statistically analyzed. The upper quartile of this fluctuation range is taken as the stability threshold to ensure that the delineation of the stable pixel band excludes the brightness fluctuation caused by normal surface texture and only retains the pixels in the real stable area.

[0024] For all pixels within each stable pixel band, abnormal pixels deviating from the average yellow-blue channel value of that pixel band are first removed. The boundary for removing abnormal values ​​is determined by statistically analyzing the reasonable upper limit of the normal fluctuation range of yellow-blue channel values ​​on a homogeneous surface of similar materials, to eliminate the influence of local smudged pixels on the mean calculation. To avoid color distortion caused by severe smudges or repainting in the masked area, before using the average yellow-blue channel value of the masked side as an approximate original color reference, it is further determined whether the color distribution of the stable pixel band on the masked side falls within the pre-statistically obtained reasonable color range. When the average yellow-blue channel value or its variance on the masked side significantly exceeds the upper limit of this reasonable range, the corresponding overlap boundary is marked as an abnormal boundary and is not included in the calculation of the net yellowing differential signal. The average value of the remaining pixels is taken to obtain the average yellow-blue channel values ​​on the exposed side and the masked side, respectively. The average yellow-blue channel value on the masked side is used as the approximate original color reference for the material for subsequent net yellowing calculation.

[0025] The mean of the yellow and blue channels on the exposed side is subtracted from the mean of the yellow and blue channels on the occluded side to obtain the single net yellowing difference at the overlapping boundary. The median of the single net yellowing differences for all valid overlapping boundaries in the image is then taken to obtain the net yellowing difference signal. Median processing is naturally robust to a few abnormal single difference values ​​caused by stain coverage, ensuring the stability of the estimated net yellowing difference signal.

[0026] Furthermore, the material segmentation process refers to... Figure 2Specifically, this includes: dividing the image into several local regions; statistically analyzing the gray-level difference intensity of adjacent pixels in multiple directions for each region; calculating the normalized difference ratio using the sum of the maximum and minimum gray-level difference intensities in each direction as the denominator and the difference between the maximum and minimum as the numerator, which serves as the texture anisotropy feature value of the region; and recording the direction corresponding to the maximum gray-level difference intensity as the main texture direction of the region; marking the set of regions whose anisotropy feature values ​​exceed the discrimination boundary and whose spatially adjacent regions have the same main texture direction as candidate ventilation structures; and marking the set of regions whose anisotropy feature values ​​are below the discrimination boundary and have no stable main texture direction as candidate body shell regions; performing connectivity analysis on each candidate region, merging spatially adjacent regions of the same type, and removing isolated fragment regions with too small an area to obtain the ventilation structure material segmentation region and the body shell material segmentation region; and using the consistency ratio of the texture classification judgment results of all regions within each material segmentation region as the partition confidence of the segmentation region.

[0027] Specifically, the main body shell typically employs a fine, random texture with isotropic properties, while the ventilation grille structure exhibits a periodic texture with a clear directional advantage. The normalized difference ratio is calculated for each region block as the texture anisotropic characteristic value. The minimum contrast threshold is set based on the noise characteristics of the image acquisition system, and is a multiple of the standard deviation of the luminance channels measured in a uniform region. When the luminance contrast within a region block is lower than this value, it is determined to be a low-contrast invalid region.

[0028] The image is divided into local regions. The side length of each region should be chosen to ensure that it contains at least one complete texture period, thus guaranteeing the effectiveness of texture feature extraction. For each region, the gray-level difference intensity of adjacent pixels is statistically analyzed in multiple uniformly distributed directions. A normalized difference ratio is calculated using the sum of the maximum and minimum gray-level difference intensities in each direction as the denominator and the difference between the maximum and minimum as the numerator. This normalized difference ratio serves as the texture anisotropy feature value for that region. The value of this normalized difference ratio is between zero and one. A value closer to one indicates stronger texture anisotropy, while a value closer to zero indicates a tendency towards isotropy. This avoids the risk of numerical instability when the denominator of a simple ratio approaches zero. When the denominator is lower than a preset minimum contrast threshold, the region is marked as a low-contrast invalid region and is not included in subsequent material segmentation. The minimum contrast threshold is determined based on the lowest measurable contrast of the effective texture region in the sample image. The direction corresponding to the maximum gray-level difference intensity is recorded as the main texture direction of that region.

[0029] The method for determining the discrimination boundary is as follows: On old home appliance sample images that have been labeled with material types and cover major home appliance categories and shell design types, the distribution of texture anisotropy feature values ​​of the main shell area block and the ventilation grille area block are statistically analyzed. Candidate discrimination boundary values ​​are traversed, the material classification error ratio corresponding to each candidate value is calculated, and the candidate value with the lowest error ratio is taken as the discrimination boundary to ensure the reliability of material segmentation under normal shooting conditions.

[0030] A set of regions whose anisotropic eigenvalues ​​exceed the discrimination boundary and whose spatially adjacent regions have consistent texture main direction heights are marked as candidate ventilation structure regions; a set of regions whose anisotropic eigenvalues ​​are below the discrimination boundary and have no stable texture main direction are marked as candidate main shell regions. The criterion for determining that adjacent regions have consistent texture main direction heights is that the angle difference between the texture main directions of adjacent regions is lower than the direction consistency threshold, which is determined by statistically analyzing the maximum angle difference between the texture main directions of adjacent regions within a known ventilation grille area.

[0031] When old home appliances have large areas of directional scratches or printed striped patterns on their surface, these areas may also exhibit high anisotropic eigenvalues ​​and a consistent main texture direction, posing a risk of confusion with the texture characteristics of ventilation grilles. In this case, the geometric regularity of the area can be used as an auxiliary factor in the judgment (ventilation grilles are usually arranged in regular rectangles or hexagons in a periodic pattern), thereby eliminating interference from non-grille textures such as scratches.

[0032] Connectivity analysis was performed on each candidate region, merging spatially adjacent candidate regions of the same type into connected regions. Isolated fragment regions with areas below the minimum area threshold were removed, resulting in ventilation structure material segmentation regions and main body shell material segmentation regions. The minimum area threshold was determined based on the minimum visible area of ​​the effective ventilation grille region in known old home appliance samples.

[0033] The proportion of the number of regions whose texture classification results match the final labeled category of the segmented region out of the total number of regions in the segmented region is used as the partition confidence output of the segmented region.

[0034] Furthermore, the process of obtaining the net yellowing amount and zoning confidence of each material region includes: for the segmented region of the main body shell material, extracting the yellow and blue channel values ​​of all pixels and taking the average value, using the original color reference of the main body shell material as the starting point, calculating the difference between the current average value of the yellow and blue channels and the original color reference to obtain the net yellowing amount of the main body shell material; performing the same process for the segmented region of the ventilation structure material: when there is a corresponding effective overlapping boundary in the segmented region, using the average value of the yellow and blue channels on the boundary occlusion side as the original color reference of the ventilation structure material, calculating the net yellowing amount of the ventilation structure material; when there is no corresponding effective overlapping boundary in the segmented region of the ventilation structure material, marking the net yellowing amount of the ventilation structure material as missing, and setting the zoning confidence of the region to the lowest value, triggering reshoot guidance by the coverage index mechanism.

[0035] Specifically, for all pixels within the segmented area of ​​the main body shell material, extract their yellow and blue channel values ​​and take the average value. Starting from the reference base of the approximate original color of the main body shell material, calculate the difference between the current average value of the yellow and blue channels and the approximate original color reference base to obtain the net yellowing amount of the main body shell material.

[0036] For the segmented area of ​​the ventilation structure material, determine whether there is a corresponding effective overlap boundary within the segmented area. When there is an effective overlap boundary within the segmented area of ​​the ventilation structure material, use the average value of the yellow-blue channel on the shaded side of the overlap boundary as the approximate original color reference benchmark for the ventilation structure material, and calculate the net yellowing amount of the ventilation structure material using the same calculation method as the main shell.

[0037] When there is no corresponding valid overlap boundary within the segmented area of ​​the ventilation structure material, the net yellowing amount of the ventilation structure material is marked as missing. At the same time, the zoning confidence of the segmented area is set to the lowest value. This missing state is passed to the downstream coverage index calculation step, and the coverage index mechanism determines whether to trigger reshoot guidance. The zoning confidence and net yellowing missing state of each material area are output downstream together.

[0038] Furthermore, the process of obtaining cross-material degradation path characteristics refers to... Figure 3This includes: using the coordinates of the original color reference of the main shell material in the chromaticity plane as the starting point of the color offset of the main shell material, and using the average of the current chromaticity plane coordinates of the segmented region of the main shell material as the offset endpoint, calculating the offset direction angle of the offset vector relative to the positive yellow-blue axis in the chromaticity plane to obtain the offset direction angle of the main shell material; using the coordinates of the original color reference of the ventilation structure material in the chromaticity plane as the starting point of the offset, and using the average of the current chromaticity plane coordinates of the segmented region of the ventilation structure material as the offset endpoint, calculating the offset direction angle of the ventilation structure material in the same way; calculating the difference between the offset direction angle of the main shell material and the offset direction angle of the ventilation structure material to obtain the cross-material offset direction angle difference value, which is used as the cross-material degradation path feature; when there are multiple sets of valid material pairs in the image, taking the average of the cross-material offset direction angle differences of each set, and using the dispersion of each set of differences as the reliability index of the feature.

[0039] Specifically, during the photo-oxidative degradation process, the main shell material and the ventilation structure material exhibit systematic differences in their color shift directions within the chromaticity plane due to variations in the chemical structure of their respective polymer chains and the formulations of the added stabilizers. The photo-oxidative degradation of the main shell material is primarily driven by the oxidation of the polybutadiene component, producing conjugated carbonyl chromophores, resulting in a significant shift in the yellow-blue channels and a relatively smaller shift in the red-green channels. In contrast, the photo-oxidative degradation of the ventilation structure material (polypropylene) is primarily driven by the decomposition pathway of methylene hydroperoxides, producing aldehyde and ketone chromophores, and the ratio of its red-green to yellow-blue channel shifts differs distinguishably from that of the main shell material. Furthermore, the stabilizer formulations corresponding to different manufacturing eras further influence the relative relationship between the shift directions of the two materials, making the cross-material shift angle difference an effective feature for encoding manufacturing era information.

[0040] The coordinates of the approximate original color reference of the main body shell material in the chromaticity plane are used as the starting point of the color offset of the main body shell material; the average coordinates of the current chromaticity values ​​of all pixels in the segmented area of ​​the main body shell material are used as the ending point of the color offset of the main body shell material; the direction angle of this offset vector relative to the positive yellow-blue axis in the chromaticity plane is calculated to obtain the offset direction angle of the main body shell material.

[0041] The coordinates of the approximate original color reference of the ventilation structure material in the chromaticity plane are used as the starting point of the color offset of the ventilation structure material; the average coordinates of the current chromaticity values ​​of all pixels in the segmented area of ​​the ventilation structure material are used as the ending point of the color offset of the ventilation structure material; the offset direction angle of the ventilation structure material is calculated in the same way.

[0042] The difference between the offset direction angle of the main shell material and the offset direction angle of the ventilation structure material is calculated to obtain the cross-material offset direction angle difference. This difference depends only on the direction of each offset vector and is independent of the vector amplitude, thereby eliminating the influence of cumulative ultraviolet irradiation on feature extraction. This allows the feature to encode only the relative relationship between the degradation paths of the two materials, i.e., information determined by the differences in the era of stabilizer formulation.

[0043] When there are multiple effective material pairs of the main body shell and ventilation structure in the image, the average value of the cross-material offset direction angle difference of each group is taken as the final cross-material degradation path feature; the statistical measure of the difference between each group's difference and the mean value is output as the reliability index of the feature.

[0044] Furthermore, the process of obtaining the coverage index includes: taking the ratio of the number of effective overlapping boundaries in the image from which the net yellowing differential signal was successfully extracted to the number of boundaries required for sufficient coverage as the overlapping coverage component; taking the smaller value of the confidence scores of the segmented areas of the main body shell material and the segmented areas of the ventilation structure material as the segmented area component; taking the smaller value of the overlapping coverage component and the segmented coverage component as the coverage index; when the coverage index reaches the sufficient coverage determination threshold, determining that the current image data meets the deterministic grading conditions; when the coverage index is lower than the sufficient coverage determination threshold, statistically analyzing the image orientation distribution of the missing effective overlapping boundaries in the image, and generating supplementary shooting guidance information output based on the image orientation description.

[0045] Specifically, the overlap coverage component is the ratio of the number of effective overlap boundaries successfully extracted from the net yellowing differential signal in the image to the preset number of boundaries required for sufficient coverage. The preset number of boundaries required for sufficient coverage is determined as follows: in a sample of old household appliances of known age, the minimum number of effective overlap boundaries required to make the fluctuation range of the net yellowing differential signal estimate reach an acceptable accuracy is statistically analyzed, and this number is taken as the preset value.

[0046] The smaller of the confidence scores for the areas segmented by the main shell material and the ventilation structure material is used as the zonal coverage component. The smaller of the overlap coverage component and the zonal coverage component is taken as the coverage index to ensure that the coverage index responds sensitively to any weak link.

[0047] The method for determining the sufficient coverage threshold is as follows: In a sample of old home appliances of known age, the correspondence between the coverage index and the age classification accuracy is statistically analyzed; the age classification accuracy requirement is determined based on the acceptable upper limit of the misjudgment rate of hazardous substances according to the age classification and disposal process; the lowest coverage index value corresponding to the age classification accuracy meeting this requirement is taken as the sufficient coverage threshold.

[0048] When the coverage index reaches the sufficient coverage determination threshold, the current image data is determined to meet the deterministic grading conditions and enters the age grading process.

[0049] When the coverage index is below the sufficient coverage threshold, the directional distribution of missing effective overlapping boundaries in the image is statistically analyzed. Based on the directional location of the missing effective overlapping boundaries in the image, corresponding supplementary shooting guidance information is generated to prompt the inclusion of the area in the shooting range. The supplementary shooting guidance information is output as an image directional description to ensure the generality of the method under the condition of no prior equipment model.

[0050] When the number of supplementary shooting attempts reaches the preset upper limit and the coverage index still has not reached the sufficient coverage threshold, the time range of the currently available features is output and the iteration terminates to avoid the process from falling into an endless loop. The upper limit of the number of supplementary shooting attempts is configured during system initialization based on the maximum number of supplementary shooting attempts that are acceptable in actual use.

[0051] Furthermore, the method for obtaining the material classification results of recycled materials includes: constructing a two-dimensional feature vector by combining the net yellowing differential signal and cross-material degradation path characteristics; mapping the two-dimensional feature vector to a pre-built age feature space, which is constructed based on the two-dimensional feature vectors corresponding to old home appliance samples with known manufacturing years, and dividing the space into several age interval clusters by the time nodes corresponding to major updates of material stabilizer formulas in different historical stages of the home appliance manufacturing industry; calculating the distance between the current two-dimensional feature vector and the cluster centers of each age interval, and outputting the age interval with the smallest distance as the manufacturing age classification result when the coverage index meets the deterministic classification conditions; outputting the age range result of the closest adjacent age intervals as the age range result when the coverage index still does not meet the deterministic classification conditions after the number of supplementary shooting guidance times reaches the upper limit; and performing a probabilistic mapping between the manufacturing age classification result and the corresponding hazardous substance content level of each age interval to output the material classification result of recycled materials and the corresponding confidence level.

[0052] Specifically, a two-dimensional feature vector is constructed based on the net yellowing difference signal and cross-material degradation path characteristics, serving as the input for manufacturing age classification. The two components of the two-dimensional feature vector encode manufacturing age information from the dimensions of yellowing rate and degradation path direction difference, respectively, providing complementary constraints from different physical dimensions.

[0053] The pre-constructed chronological feature space is built upon the two-dimensional feature vectors corresponding to old home appliance samples from known manufacturing years. The samples should cover different manufacturing years, major brands, and typical usage environments to ensure the representativeness of the chronological feature space. Using the time nodes corresponding to major updates in the material stabilizer formulations of the home appliance manufacturing industry at different historical stages as boundaries, the samples are divided into several chronological intervals. The cluster centers of the two-dimensional feature vectors of the samples within each chronological interval are calculated, forming the cluster distribution of each chronological interval in the chronological feature space.

[0054] Calculate the distance between the current two-dimensional feature vector and the cluster centers of each chronological interval in the chronological feature space. When the coverage index meets the deterministic grading condition, the chronological interval with the smallest distance to the current feature vector is taken as the manufacturing chronological grading result. When the coverage index still does not meet the deterministic grading condition after the number of reshoots reaches the upper limit, the two adjacent chronological intervals with the closest distance to the current feature vector are output as the chronological range result.

[0055] The manufacturing age classification results are probabilistically mapped to the corresponding hazardous substance content levels for each age range. This probabilistic mapping relationship can be directly determined based on publicly available industry statistics on hazardous substance content in used household appliances or the corresponding regulations for age boundaries and hazardous substance content in national standards. No independent chemical testing is required; the mapping relationship can be established directly by referring to publicly available data. The percentage of samples within each age range whose hazardous substance content exceeds the limit standard is used as the probability of exceeding the hazardous substance limit for that age range.

[0056] The final output is the material classification result of the recyclables and the corresponding confidence level. The confidence level is determined by the minimum value among the coverage index, the characteristic reliability index and the age classification distance margin, so as to ensure the sensitivity of the confidence level to the weakest link and to maintain consistency with the logic of taking the smaller value of the coverage index. The obtained classification result and confidence level are transmitted to the downstream classification and disposal process for use in classification and disposal decision-making.

[0057] Example 2: In images, the overlap boundaries of old home appliance panels often appear as discontinuous fragments rather than complete, continuous boundary lines due to localized dirt, surface scratches, or changes in shooting angle. Directly including these discontinuous fragments in the count of effective overlap boundaries will underestimate the actual coverage, leading to a lower coverage index and unnecessarily triggering reshoots. Global trend statistical analysis integrates discontinuous fragments belonging to the same physical overlap boundary into complete boundaries, improving the accuracy of the effective overlap boundary count.

[0058] The process of detecting panel overlap boundaries includes a global consistency integration step for overlap boundary orientation: Initial detection of overlap boundaries in the image yields several boundary segments, each carrying its position coordinates, orientation angle, and segment length in the image; statistical analysis of the orientation angles of all boundary segments is performed, extracting the dominant orientation angles that appear more frequently than a preset significance threshold. When all orientation angles are evenly distributed and there is no dominant orientation, the integration step is skipped, and the process proceeds directly to the next step; in the case of a dominant orientation, boundary segments whose orientation deviation from any dominant orientation angle is within the directional tolerance range are grouped into the same orientation group; within the same orientation group, discontinuous boundary segments with consistent orientations are integrated into complete overlap boundaries, provided that the distance between the endpoints of adjacent segments is less than a preset distance threshold. The preset distance threshold is determined based on the maximum discontinuity width of overlap boundaries truncated by local stains in known old appliance samples, and is strictly less than the minimum possible distance between two adjacent independent overlap boundaries; the integrated complete overlap boundaries replace the original discontinuous segments and participate in subsequent effective overlap boundary quantity statistics and net yellowing differential signal extraction, updating the coverage index calculation with the integrated effective overlap boundary quantity.

[0059] Specifically, initial boundary detection is performed on the image to obtain several boundary segments. Each boundary segment carries its position coordinates, orientation angle, and segment length in the image. The orientation angles of all boundary segments are statistically analyzed. The number of segments in each orientation angle interval is counted, and orientation angles with a frequency higher than a significance threshold are extracted as dominant orientation angles.

[0060] The significance threshold is determined as follows: In known old appliance sample images, the random distribution frequency of non-overlapping boundary noise segments within each directional angle interval is statistically analyzed. A reasonable upper limit of this random distribution frequency is taken as the significance threshold, ensuring that the determination of the dominant directional angle excludes the interference of random noise segments. When the frequency distribution of all directional angle intervals is uniform and no interval exceeds the significance threshold, it is determined that there is no dominant directional angle in the image, and the integration step is skipped, with the segments obtained from the initial detection directly participating in subsequent processes.

[0061] In cases where a dominant orientation exists, boundary segments whose angular deviation from any dominant orientation is within the directional tolerance range are grouped into the corresponding orientation group. The directional tolerance range is determined by statistically analyzing the maximum mutual deviation of the orientation angles of boundary segments on the same physical overlap boundary in known samples of old home appliances. The samples should cover major home appliance categories (including television, washing machine, refrigerator, and air conditioner panels) and typical shooting angle ranges to cover the maximum impact of perspective distortion on orientation angle deviations.

[0062] Within the same orientation group, boundary segments belonging to that orientation group are sorted according to their position coordinates perpendicular to the dominant orientation direction, and the endpoint spacing of adjacent segments is calculated. Discontinuous segments meeting the condition of having an endpoint spacing lower than a spacing threshold are integrated into a single complete overlapping boundary. The spacing threshold is determined based on the maximum discontinuity width caused by local stains truncating overlapping boundaries in known old appliance samples, and this threshold is strictly less than the minimum possible spacing between two adjacent independent overlapping boundaries in known samples to ensure that the integration operation does not incorrectly merge segments belonging to different physical boundaries. The minimum possible spacing is obtained by statistically measuring the spacing of independent overlapping boundaries in known sample images, and the discontinuity width is obtained by statistically measuring the discontinuous spacing of boundaries truncated by stains in known sample images.

[0063] The integrated complete overlapping boundary replaces the original discontinuous segment and participates in the subsequent statistics of the number of effective overlapping boundaries and the extraction of net yellowing differential signal. The number of effective overlapping boundaries after integration is updated in the coverage index calculation step, so that the coverage index accurately reflects the actual usable physical overlapping boundary coverage in the image.

[0064] The color temperature of the lighting environment in which users photograph old home appliances varies. This color temperature deviation introduces a systematic overall bias in the yellow and blue channels of the image, affecting the accuracy of the net yellowing difference signal and the calculation of the cross-material chromaticity offset direction angle. Therefore, by utilizing the inherent neutral color reference area in the image, the lighting color temperature bias is estimated and corrected, improving the consistency of feature extraction results under different lighting conditions.

[0065] The acquisition of the net yellowing differential signal also includes an image illumination color temperature normalization correction step: Locating a neutral color reference region within a non-overlapping area of ​​the image; the neutral color reference region is a material region whose original color in the chromaticity plane is close to neutral gray, including the areas where metal supports, aluminum nameplates, or gray rubber seals are located in the image; when a neutral color reference region meeting the conditions cannot be located in the image, the illumination color temperature normalization correction step is skipped, and the uncertainty caused by the uncorrected illumination color temperature is included in the weighting of the feature reliability index; when the neutral color reference region is successfully located, the current chromaticity plane coordinates of that region are extracted. The value is used as the yellow-blue channel component of the offset vector of the mean value relative to the origin of the chromaticity plane, as the yellow-blue bias of the current image's illumination color temperature. The illumination color temperature yellow-blue bias is subtracted from the mean values ​​of the yellow-blue channels on the exposed side and the mean values ​​of the yellow-blue channels on the occluded side to obtain the normalized mean values ​​of the yellow-blue channels on each side. The difference between the two normalized values ​​is used as the single net yellowing difference after illumination color temperature correction, which is used in subsequent median calculations. The correction steps are applicable to diffuse or nearly uniform illumination environments. Under strong local direct light conditions, the uncertainty caused by the decrease in correction accuracy is included in the weighting of the feature reliability index.

[0066] Specifically, a neutral color reference region is located in the image. The neutral color reference region is a material area whose original color is close to neutral gray in the chromaticity plane. The metal bracket or aluminum nameplate area visible in the image is preferred. When the metal area is not visible, the area where the gray rubber seal is located is used as an alternative. The additional uncertainty caused by the long-term aging color drift of the rubber area is included in the deweighting of the feature reliability index.

[0067] The factory chromaticity coordinates of various neutral color reference areas are pre-calibrated in known samples. This calibration value is recorded as the neutral reference chromaticity benchmark for the corresponding material to eliminate interference from the deviation between the material's own hue and strictly neutral gray. When the neutral color reference area is successfully located, the mean coordinates of the current chromaticity values ​​of all pixels within that area are extracted. The yellow-blue channel component of the offset vector of this mean coordinate relative to the neutral reference chromaticity benchmark coordinates of the corresponding material is used as the yellow-blue bias of the current image's illumination color temperature.

[0068] When a neutral color reference region that meets the conditions cannot be located in the image, the illumination color temperature normalization correction step is skipped, and the uncertainty caused by the uncorrected illumination color temperature is included in the weighting process of the feature reliability index. The weighting magnitude is determined as follows: in a sample of old household appliances of known age, images are acquired under standard illumination and color-biased illumination conditions, respectively. The deviation distribution of the net yellowing difference signal under the two conditions is statistically analyzed. The weighting coefficient is determined by the decrease in age classification accuracy corresponding to the upper quartile of the deviation distribution, so that the weighted feature reliability index accurately reflects the classification uncertainty under the uncorrected illumination color temperature condition.

[0069] Subtract the illumination color temperature yellow-blue bias from both the exposure-side and shading-side yellow-blue channel mean values ​​to obtain the illumination color temperature normalized mean values ​​for each side's yellow-blue channels. Subtract the normalized shading-side yellow-blue channel mean from the normalized exposure-side yellow-blue channel mean to obtain the illumination color temperature corrected single-shot net yellowing difference, which is then used in subsequent multi-boundary median calculations.

[0070] The physical premise of this correction step is that the illumination in the image acquisition environment is nearly diffuse and uniform, and the color temperature of the illumination received by the neutral color reference area and the overlapping boundary area is consistent. Under strong local direct light conditions, the color temperature of the illumination received by different areas of the image may have spatial non-uniformity. In this case, the correction accuracy decreases, and the corresponding uncertainty is incorporated into the weighting process of the feature reliability index. The weighting method is the same as in the case of uncorrected illumination color temperature.

[0071] When old home appliances are relatively recent in manufacturing or have a short service life, the color shift amplitude between the two materials may be extremely small, close to the magnitude of the measurement noise inherent in the image acquisition process. In this case, the estimated offset direction angle will be mainly dominated by measurement noise rather than the true degradation direction, and the difference in offset direction angle across materials will lose its effective age coding significance. By judging the validity of the offset vector amplitude, we can avoid including the noise-dominated direction angle difference in the age classification features and ensure the reliability of the cross-material degradation path features.

[0072] The acquisition of cross-material degradation path features also includes an offset vector validity screening step: calculating the amplitude of the color offset vector of the main shell material and the amplitude of the color offset vector of the ventilation structure material; the minimum reliable offset amplitude is determined based on the measurement noise standard deviation of the mean chromaticity coordinates obtained by repeated acquisition of a known color chart under typical shooting conditions, and the offset amplitude required to make the offset direction angle estimation error lower than the upper limit of acceptable error is taken as the minimum reliable offset amplitude; when the amplitude of any material color offset vector is lower than the minimum reliable offset amplitude, it is determined that the material offset direction angle is dominated by measurement noise, the offset direction angle of the corresponding material is marked as invalid, the cross-material degradation path feature is marked as low confidence, the degradation path coverage component is updated with the lowest value, and then the coverage index is updated, triggering the corresponding processing flow; when the offset direction angles of both materials are valid, the cross-material offset direction angle difference is calculated normally as the cross-material degradation path feature.

[0073] Specifically, the magnitude of the color offset vector of the main body shell material is calculated, which is the chromaticity plane distance between the current chromaticity mean coordinates of the main body shell material and its approximate original color reference coordinates; the magnitude of the color offset vector of the ventilation structure material is calculated in the same way.

[0074] The minimum reliable offset amplitude is determined as follows: Color charts of known colors are repeatedly sampled under typical shooting conditions. These typical shooting conditions should cover both indoor natural lighting and indoor artificial lighting environments, as well as the typical distance range for handheld shooting in the applicable scenario of this invention. The calibration result is taken as the maximum value of the standard deviation of the mean chromaticity coordinate measurement noise under all condition combinations, to ensure that the minimum reliable offset amplitude covers the most unfavorable shooting conditions. Based on this noise standard deviation, and based on the assumption that chromaticity measurement noise is approximately isotropic, and taking into account the monotonic relationship that the offset direction angle estimation error decreases as the offset amplitude increases, the offset amplitude required to make the offset direction angle estimation error lower than the upper limit of acceptable error is taken as the minimum reliable offset amplitude. The upper limit of acceptable error is determined by statistically analyzing the influence range of the offset direction angle error on the accuracy of age classification in known age samples, with the maximum acceptable direction angle error corresponding to the age classification accuracy not being lower than the preset requirement as the upper limit. In actual calibration, the applicability of the isotropic assumption of colorimetric measurement noise should be verified. If significant anisotropy exists, the minimum reliable offset amplitude should be determined by correcting the noise standard deviation calibrated for each channel.

[0075] When the amplitude of any color offset vector of the main shell material or ventilation structure material is lower than the minimum reliable offset amplitude, it is determined that the offset direction angle of the material is dominated by measurement noise. The offset direction angle of the corresponding material is marked as invalid, and the cross-material degradation path feature is marked as low reliability. The partition coverage component in the coverage index calculation is updated with the lowest value, thereby reducing the coverage index and triggering the reshoot guidance or interval age output process. The subsequent branches are uniformly processed by the coverage index mechanism.

[0076] When the amplitudes of the two color offset vectors of the main shell material and the ventilation structure material are not lower than the minimum reliable offset amplitude, it is determined that both offset direction angles are valid, and the cross-material offset direction angle difference is calculated normally and used as a cross-material degradation path feature to participate in the subsequent age classification process.

[0077] Those skilled in the art will readily understand that the above description is merely a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, and improvements made within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for automatic identification and classification of recyclable materials based on image processing, characterized in that, include: Acquire images of old home appliances, detect panel overlap boundaries, extract the mean difference in yellow and blue channels between the exposed and shaded areas of the same material at the overlap boundary, obtain the net yellowing difference signal, and use the mean color of the shaded side as the original color reference benchmark of the material. Texture anisotropic features are extracted from local regions of the image. Material segmentation is performed on the PP ventilation structure region with periodic grid texture and the ABS main body region with isotropic random texture. Combined with the original color reference benchmark, the net yellowing amount and partition confidence of each material region are extracted. Using the original color reference as the starting point, the offset direction angle of the current color of ABS and PP relative to their respective references in the chromaticity plane is calculated to obtain the cross-material degradation path characteristics. A two-dimensional feature vector is constructed based on the net yellowing differential signal and cross-material degradation path characteristics, and mapped to a pre-built age feature space to complete the manufacturing age classification; the coverage index is obtained based on the number of effective overlap boundaries and the confidence of the partition; when the coverage index is insufficient, the age range of the interval is output and supplementary shooting guidance is generated; the iteration terminates when the deterministic classification conditions are met, and the material classification results of the recycled materials are output.

2. The method for automatic identification and classification of recyclable materials based on image processing according to claim 1, characterized in that, The process of acquiring the net yellowing difference signal includes: extracting the cross-boundary pixel sequence along the direction perpendicular to the overlap boundary; using the stability of the brightness channel value in each side pixel region as a criterion, defining stable pixel bands on both the exposed and occluded sides of the boundary, and excluding mixed pixels with continuously changing brightness in the boundary transition region; after removing outliers from the yellow and blue channel values ​​of all pixels in each stable pixel band, taking the average value to obtain the average yellow and blue channel values ​​on the exposed and occluded sides, and using the average yellow and blue channel value on the occluded side as the material's original color reference for output; subtracting the average yellow and blue channel values ​​on the exposed and occluded sides to obtain the single net yellowing difference at the overlap boundary; taking the median of the single net yellowing differences of all effective overlap boundaries in the image, and outputting the net yellowing difference signal.

3. The method for automatic identification and classification of recyclable materials based on image processing according to claim 1, characterized in that, The material segmentation process includes: dividing the image into local regions; statistically analyzing the gray-level difference intensity of adjacent pixels in multiple directions for each region; calculating the normalized difference ratio using the sum of the maximum and minimum gray-level difference intensities in each direction as the denominator and the difference between the maximum and minimum as the numerator, which serves as the texture anisotropy feature value of the region; and recording the direction corresponding to the maximum gray-level difference intensity as the main texture direction of the region; marking the set of regions whose anisotropy feature values ​​exceed the discrimination boundary and whose spatially adjacent regions have the same main texture direction as candidate ventilation structures; and marking the set of regions whose anisotropy feature values ​​are below the discrimination boundary and have no stable main texture direction as candidate body shell regions; performing connectivity analysis on each candidate region; merging spatially adjacent regions of the same type and removing isolated fragment regions with too small an area to obtain the material segmentation regions for ventilation structures and the material segmentation regions for the body shell; and using the consistency ratio of the texture classification results of all regions within each material segmentation region as the partition confidence of the segmentation region.

4. The method for automatic identification and classification of recyclable materials based on image processing according to claim 1, characterized in that, The process of obtaining the net yellowing amount and zoning confidence of each material region includes: for the segmented region of the main body shell material, extracting the yellow and blue channel values ​​of all pixels and taking the average value, using the original color reference of the main body shell material as the starting point, calculating the difference between the current average value of the yellow and blue channels and the original color reference, to obtain the net yellowing amount of the main body shell material; the same process is performed for the segmented region of the ventilation structure material: when there is a corresponding effective overlapping boundary in the segmented region, the average value of the yellow and blue channels on the boundary occlusion side is used as the original color reference of the ventilation structure material to calculate the net yellowing amount of the ventilation structure material; when there is no corresponding effective overlapping boundary in the segmented region of the ventilation structure material, the net yellowing amount of the ventilation structure material is marked as missing, and the zoning confidence of the region is set to the lowest value, and the coverage index mechanism triggers the reshoot guidance.

5. The method for automatic identification and classification of recyclable materials based on image processing according to claim 1, characterized in that, The process of obtaining cross-material degradation path features includes: using the coordinates of the original color reference of the main shell material in the chromaticity plane as the starting point of the color offset of the main shell material, and using the average value of the current chromaticity plane coordinates of the segmented region of the main shell material as the offset endpoint, calculating the offset direction angle of the offset vector relative to the positive yellow-blue axis in the chromaticity plane to obtain the offset direction angle of the main shell material; using the coordinates of the original color reference of the ventilation structure material in the chromaticity plane as the starting point of the offset, and using the average value of the current chromaticity plane coordinates of the segmented region of the ventilation structure material as the offset endpoint, calculating the offset direction angle of the ventilation structure material in the same way; calculating the difference between the offset direction angle of the main shell material and the offset direction angle of the ventilation structure material to obtain the cross-material offset direction angle difference value, which is used as the cross-material degradation path feature; when there are multiple sets of valid material pairs in the image, the average value of the cross-material offset direction angle difference values ​​of each set is taken, and the dispersion of each set of differences is used as the reliability index of the feature.

6. The method for automatic identification and classification of recyclable materials based on image processing according to claim 1, characterized in that, The process of obtaining the coverage index includes: using the ratio of the number of effective overlapping boundaries successfully extracted from the net yellowing differential signal in the image to the number of boundaries required for sufficient coverage as the overlapping coverage component; using the smaller value of the confidence scores of the segmented areas of the main body shell material and the segmented areas of the ventilation structure material as the partition coverage component; taking the smaller value of the overlapping coverage component and the partition coverage component as the coverage index; when the coverage index reaches the sufficient coverage judgment threshold, determining that the current image data meets the deterministic grading conditions; when the coverage index is lower than the sufficient coverage judgment threshold, statistically analyzing the image orientation distribution of the missing effective overlapping boundaries in the image, and generating supplementary shooting guidance information output based on the image orientation description.

7. The method for automatic identification and classification of recyclable materials based on image processing according to claim 1, characterized in that, The method for obtaining the material classification results of recyclables includes: constructing a two-dimensional feature vector by combining the net yellowing difference signal and cross-material degradation path features; mapping the two-dimensional feature vector to a pre-built age feature space, which is constructed based on the two-dimensional feature vectors corresponding to old home appliance samples with known manufacturing years, and dividing the age intervals into clusters based on the time nodes corresponding to major updates of material stabilizer formulas in different historical stages of the home appliance manufacturing industry; calculating the distance between the current two-dimensional feature vector and the cluster centers of each age interval; when the coverage index meets the deterministic classification conditions, the age interval with the smallest distance is output as the manufacturing age classification result; when the coverage index still does not meet the deterministic classification conditions after the number of supplementary shooting guidance times reaches the upper limit, the two closest adjacent age intervals are output as the age range result of the interval; and performing a probabilistic mapping between the manufacturing age classification result and the corresponding hazardous substance content level of each age interval to output the material classification result of recyclables and the corresponding confidence level.

Images