A method and system for sorting waste textiles based on spatial reliability fusion within an object

Abstract

The application discloses a kind of object space reliability fusion's waste textiles sorting method and system, belong to waste textiles resource sorting technical field, comprising: obtaining waste textiles two-dimensional real foreground area and the material prediction information of effective measurement pixel, according to the position of pixel relative object contour Determine the spatial reliability weight and form weighted class evidence, according to this Determine object winner class and class index;Further based on the local winner class of space unit, obtain object space consistency, and judge whether it meets sorting condition in combination with class index, improve the stability and accuracy of waste textiles object level material identification result, improve the reliability of sorting decision, reduce the probability of abnormal object into target material flow.

Description

Technical Field

[0001] This invention relates to the field of waste textile resource recycling and sorting technology, and in particular to a waste textile sorting method and system that integrates spatial reliability within an object. Background Technology

[0002] The fiber composition of waste textiles directly affects the applicability of subsequent mechanical regeneration, solvent recycling, or chemical recycling processes. Existing online sorting systems typically utilize near-infrared or hyperspectral detection technologies to obtain local or pixel-level material information of fabric pieces, and then further form a complete fabric object to provide object-level material results. For high-speed conveying scenarios, relying solely on a single pixel or a single scan line is easily affected by factors such as fabric wrinkles, curling edges, local occlusion, surface contamination, mixed spectra of the conveyor belt, and changes in lighting. Therefore, it is usually necessary to aggregate the prediction results of multiple pixels to the object scale.

[0003] Existing object-level decision-making methods typically employ majority voting, winner pixel percentage, or fixed-width edge culling. Majority voting assumes that pixels at different locations within an object have approximately the same reliability, while fixed-width edge culling discards edge pixels within a preset width. However, in real-world applications, for irregular textile fabrics, the distance to the true contour, the degree of local warping, and the degree of background blending vary at different locations. Existing methods like majority voting and fixed-width edge culling may not adequately suppress boundary blending and may also lose valuable evidence of transitional regions.

[0004] On the other hand, it is difficult to distinguish between two different situations based solely on the winner category or a single confidence ratio: first, although the winner category has a high proportion, the second candidate category is still highly competitive; second, the evidence for the winner is concentrated in a local area of ​​the object, while it is not consistently supported in other areas of the object.

[0005] Therefore, the existing technology still has the following shortcomings: object-level majority voting is difficult to reflect the reliability differences of pixels at different spatial locations, a single confidence index is difficult to simultaneously characterize the competitive relationship between candidate categories and their spatial distribution characteristics within the object, and there is a lack of distinction between material determination uncertainty and the physical unsortable state of the object. Summary of the Invention

[0006] To address the technical problems existing in the prior art, embodiments of the present invention provide a method and system for sorting waste textiles based on spatial reliability fusion within an object, which addresses the issues that object-level majority voting cannot express pixel spatial reliability differences and that a single confidence ratio cannot characterize category competition and space.

[0007] To achieve the above objectives, the technical solution provided in this paper is as follows: A method for sorting waste textiles based on spatial reliability fusion within an object, specifically including: S1. Obtain the two-dimensional real foreground region of the waste textiles for which material prediction has been completed, as well as the material prediction information of each effective measurement pixel.

[0008] S2. Based on the contour of the two-dimensional real foreground region and the inward distance of each effective measurement pixel, determine the spatial reliability weight of each effective measurement pixel, and summarize the material prediction information of each effective measurement pixel based on the spatial reliability weight to obtain the weighted category evidence of each material candidate category, where the spatial reliability weight of effective measurement pixels closer to the contour is low.

[0009] S3. Determine the object winner category and category index based on the weighted category evidence. The object winner category is the material candidate category with the largest weighted category evidence, and the category index is used to characterize the reliability of the object winner category.

[0010] S4. Divide the two-dimensional real foreground region into multiple spatial units. Determine the local winner category of each spatial unit based on weighted category evidence. The local winner category is the material candidate category with the largest weighted category evidence in the current spatial unit. Based on the consistency relationship between the local winner category and the object winner category, obtain the object spatial consistency.

[0011] S5. Determine whether the waste textile objects meet the sorting conditions based on the category indicators and object space consistency. If they do, sort them.

[0012] In addition, a waste textile sorting system based on spatial reliability integration within an object is also provided, specifically including: Complete the object data interface module, which is used to obtain the two-dimensional real foreground area of ​​waste textiles with completed material prediction and the material prediction information of each effective measurement pixel.

[0013] The reliability evidence fusion module is used to determine the spatial reliability weight of each effective measurement pixel based on the contour of the two-dimensional real foreground region and the inward distance of each effective measurement pixel, and to summarize the material prediction information of each effective measurement pixel based on the spatial reliability weight to obtain the weighted category evidence of each material candidate category, wherein the spatial reliability weight of effective measurement pixels closer to the contour is low.

[0014] The category reliability assessment module is used to determine the object winner category and category index based on weighted category evidence. The object winner category is the material candidate category with the largest weighted category evidence, and the category index is used to characterize the reliability of the object winner category.

[0015] The object space consistency determination module is used to divide the two-dimensional real foreground region into multiple spatial units, determine the local winner category of each spatial unit based on weighted category evidence, where the local winner category is the material candidate category with the largest weighted category evidence in the current spatial unit, and obtain the object space consistency based on the consistency relationship between the local winner category and the object winner category.

[0016] The sorting judgment module is used to determine whether waste textile objects meet the sorting conditions based on category indicators and object space consistency. If they meet the conditions, they are sorted.

[0017] The beneficial effects of the technical solutions provided by the embodiments of the present invention include: (1) This invention provides a waste textile sorting method based on the spatial reliability fusion of objects. By combining the reliability differences of different spatial locations inside the waste textile objects, the material prediction results are fused, which can reduce the interference of object edge, contour changes and local abnormal areas on the overall material judgment results, improve the stability and accuracy of the waste textile object-level material identification results, and enhance the adaptability to irregular, wrinkled or morphologically different objects.

[0018] (2) By comprehensively utilizing the category determination results and the spatial distribution characteristics within the object, the present invention further evaluates the material reliability of the object to be sorted, which can reduce the misjudgment caused by relying solely on a single classification result, and enable objects with insufficient material evidence, obvious category competition, or inconsistent internal performance to be more reasonably identified and processed, thereby improving the reliability of sorting decisions.

[0019] (3) By distinguishing the physical sortable state of the object and the material judgment result, the present invention can adopt corresponding sorting, return or rejection methods for waste textiles in different states, avoid using a unified processing strategy for uncertain objects caused by different reasons, improve the fault tolerance and operational stability of the sorting process, and reduce the probability of abnormal objects entering the target material flow.

[0020] (4) By differentiating the sorting conditions based on the identification difficulty and sorting risk between different material categories, the present invention enables objects with similar or easily confused material compositions to adopt a more robust processing method, thereby improving the purity of the target material stream while ensuring sorting efficiency, and enhancing the applicability of the present invention to different waste textile material systems and online sorting scenarios. Attached Figure Description

[0021] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings: Figure 1This is a schematic diagram of the method flow involved in Embodiment 1 of the present invention.

[0022] Figure 2 This is a schematic diagram of the system modules involved in Embodiment 1 of the present invention.

[0023] Figure 3 This is a schematic diagram of the weighted partitioning of the true contour depth and spatial reliability of the object involved in Embodiment 1 of the present invention.

[0024] Figure 4 This is a schematic diagram of object region meshing and local winner consistency calculation involved in Embodiment 1 of the present invention.

[0025] Figure 5 This is a schematic diagram of the three-state routing determination of physical validity and material evidence layering involved in Embodiment 2 of the present invention. Detailed Implementation

[0026] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numerals in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with one or more embodiments of this specification. Rather, they are merely examples of apparatuses and methods consistent with some aspects of one or more embodiments of this specification as detailed in the appended claims.

[0027] It should be noted that the steps of the corresponding methods are not necessarily performed in the order shown and described in this specification in other embodiments. In some other embodiments, the methods may include more or fewer steps than described in this specification. Furthermore, a single step described in this specification may be broken down into multiple steps in other embodiments; and multiple steps described in this specification may be combined into a single step in other embodiments.

[0028] Example 1: Please see Figure 1 As shown in the figure, this embodiment provides a waste textile sorting method based on the spatial reliability fusion within an object.

[0029] S1. Obtain the two-dimensional real foreground region of the waste textiles for which material prediction has been completed, as well as the material prediction information of each effective measurement pixel.

[0030] It should be noted that this invention is for object-level sorting of waste textiles for which material prediction has already been completed.

[0031] This step associates the actual foreground region of the object with the material prediction information through effective pixel measurement, so that each material prediction result has a corresponding spatial location within the object.

[0032] The material prediction information includes at least the material prediction category corresponding to each valid measured pixel. Candidate material categories are pre-determined based on the waste textile sorting task and may include cotton, polyester, cotton-based blends containing elastic fibers, polyester-based blends containing elastic fibers, and cotton / polyester blends categorized according to a preset composition range. The two-dimensional true foreground region represents the actual area occupied by the waste textile in the detection plane. Valid measured pixels are those pixels within the two-dimensional true foreground region that have obtained valid material prediction results and can be used for subsequent category evidence calculations. The material prediction information records the predicted material for each valid measured pixel.

[0033] For example, the actual foreground region of the object is denoted as Its boundary is denoted as ,Right now Let represent the set of contour pixels of the two-dimensional real foreground region, where the contour pixels in the set are denoted as . The set of effective measured pixels is denoted as P. o Each effective measurement pixel Each has corresponding spatial coordinates and material prediction information, effectively measuring pixels. The predicted material category is denoted as .

[0034] It should be noted that when upstream material detection outputs non-material categories such as conveyor belt and background, the background category will not be included as a material candidate category in the subsequent weighted category evidence summary.

[0035] S2. Based on the contour of the two-dimensional real foreground region and the inward distance of each effective measurement pixel, determine the spatial reliability weight of each effective measurement pixel, and summarize the material prediction information of each effective measurement pixel based on the spatial reliability weight to obtain the weighted category evidence of each material candidate category, where the spatial reliability weight of effective measurement pixels closer to the contour is low.

[0036] Unlike ordinary majority voting, this scheme ensures that predictions located near the true contour do not contribute as much to the decision as predictions inside the object. This scheme differentiates the spatial reliability weights of each valid measurement pixel by using inward distance, resulting in lower spatial reliability weights for valid measurement pixels closer to the contour. The spatial reliability of different valid measurement pixels can be adjusted by their location.

[0037] The two-dimensional real foreground region is represented as a binary region. Pixels that belong to the two-dimensional real foreground region and have at least one location in their neighborhood that does not belong to the two-dimensional real foreground region are taken as contour pixels. The contour pixels form the outline of the current waste textile.

[0038] In this embodiment, the Euclidean distance between the effectively measured pixel and the nearest position of the contour is used as the inward distance: ; in, Indicates the number of valid measured pixels The inward distance, Represents the outline of a two-dimensional real foreground region. Indicates the number of valid measured pixels Two-dimensional coordinates, Represents outline pixels Two-dimensional coordinates.

[0039] When the two-dimensional coordinates are pixel coordinates, the inward distance is in pixels; when the detection area has been spatially calibrated, the inward distance can also be converted into the actual length.

[0040] In some embodiments, as a preferred technical solution, the spatial reliability weights of effective measurement pixels monotonically increase sequentially from the contour to the center. Alternatively, the two-dimensional real foreground region is divided into different regions according to inward distance, with the spatial reliability weights of effective measurement pixels within the same region being the same, and the spatial reliability weights of regions closer to the contour being less than those closer to the center.

[0041] The spatial reliability weight is a dimensionless quantity used to represent the relative contribution of the material prediction information of effective measured pixels to subsequent object-level weighted category evidence, and does not represent the prediction probability. Therefore, effective measured pixels closer to the contour have lower spatial reliability weights, while the spatial reliability weights of effective measured pixels gradually increase or remain unchanged as they move into the two-dimensional true foreground region.

[0042] For irregular, wrinkled, or rolled-up waste textiles, the spatial reliability weight is determined based on the actual contour of the current two-dimensional real foreground region, rather than based on a fixed image edge width. Therefore, waste textiles of different shapes have spatial reliability distributions corresponding to their own contours. For waste textiles with irregular contours, wrinkles, or rolled-up edges, the spatial reliability evaluation can change with the actual contour of the current object.

[0043] The spatial reliability weights in this scheme can be implemented using either a continuous monotonic function or a piecewise constant function.

[0044] like Figure 3As shown, in this specific embodiment, a first contour depth threshold and a second contour depth threshold are preset, and the two-dimensional real foreground region is sequentially divided into an edge region, a transition region, and an interior region according to the inward distance of the effective measured pixels. In this embodiment, the edge region is a region that is easily affected by both the object material spectrum and the transport belt or background spectrum; the transition region is a region where the background mixing effect gradually decreases but has not yet reached the stability level of the interior region; and the interior region is a region with a large distance from the actual contour and a relatively weak background mixing effect.

[0045] Specifically, the first contour depth threshold and the second contour depth threshold satisfy 0 < the first contour depth threshold. Second contour depth threshold Regions with an inward distance less than the first contour depth threshold are considered edge regions; regions with an inward distance between the first and second contour depth thresholds are considered transition regions; and regions with an inward distance greater than the second contour depth threshold are considered interior regions. This is represented as follows: ; ; ; in, For spatial reliability weights, The first contour depth threshold, The second contour depth threshold, Indicates the number of valid measured pixels The inward distance, The weights corresponding to the edge regions, The weights corresponding to the transition zone, The weights are for the internal regions.

[0046] For example, in one online operating configuration, the spatial reliability weights corresponding to the edge zone, transition zone, and interior zone are set to 0, 0.35, and 1.00, respectively.

[0047] Furthermore, after obtaining the spatial reliability weights of each effective measured pixel, the material prediction information is summarized to obtain the weighted category evidence of the material candidate category c, which is represented as: ; in, This represents the weighted category evidence for material candidate category c. Represents the set of valid measurement pixels. Represents the set of valid measurement pixels The effective measurement pixels in Indicates the number of valid measured pixels The corresponding spatial reliability weight, Indicates the number of valid measured pixels Material prediction category Indicates an indicator function, when When the value is 1, The value is 0 at that time.

[0048] In some embodiments, if the material prediction information is displayed as an output conveyor belt or a background category, the background category is not included in the calculation of the weighted category evidence.

[0049] S3. Determine the object winner category and category index based on the weighted category evidence. The object winner category is the material candidate category with the largest weighted category evidence, and the category index is used to characterize the reliability of the object winner category.

[0050] In some embodiments, the category metrics include a weighted dominance ratio and a weighted first-two category interval, wherein the weighted dominance ratio is the ratio of the sum of weighted category evidence for the object winner category to the sum of weighted category evidence for all candidate material categories, and the weighted first-two category interval is the normalized result of the difference between the object winner category and the second largest weighted category evidence relative to the sum of weighted category evidence.

[0051] Specifically, the weighted dominance ratio characterizes the degree of dominance of the winning category among all material candidate categories. The weighted first-two-category interval characterizes the degree of separation between the winning category and the second candidate category. Therefore, even if two objects have the same winning percentage, their material uncertainty can still be distinguished by the weighted first-two-category interval if the evidence for the second candidate category differs.

[0052] The material candidate evidence is sorted from largest to smallest according to its weighted category. The largest weighted category evidence is designated as the object winner category, the second largest weighted category evidence is designated as the second weighted category evidence, and the material candidate category corresponding to the second weighted category evidence is determined as the second candidate category.

[0053] The second candidate category is not the result of the second independent classification, but rather the material candidate category that ranks second in the object-level weighted category evidence ranking, reflecting the material category that poses the strongest competition to the object winner category.

[0054] In some embodiments, the object winner category is represented as: ; in, For the winner category, This represents the weighted category evidence for material candidate category c.

[0055] The method for determining the weighted dominant ratio is as follows: ; in, Indicates the weighted dominant ratio. This indicates the first weighted category of evidence, that is, the weighted category of evidence corresponding to the winner category of the object. This represents the set of all candidate material categories. This represents the weighted category evidence for material candidate category c.

[0056] The method for determining the weighted interval between the first two categories is as follows: ; in, This indicates the interval between the first two weighted categories. This indicates the first weighted category of evidence, that is, the weighted category of evidence corresponding to the winner category of the object. This indicates the second weighted category evidence, that is, the weighted category evidence corresponding to the second candidate category.

[0057] Furthermore, when the weighted sum of evidence for all candidate material categories is zero, the above ratio calculation is not performed, and the current object is treated as an object with insufficient material evidence to avoid the denominator being zero.

[0058] Furthermore, when two waste textile products have similar evidence proportions for the winner category, but the weighted category evidence for the second candidate category is different, the degree of category competition experienced by the winner category can be further distinguished by weighting the interval between the first two categories.

[0059] Furthermore, when the weighted sum of evidence for all material candidate categories is zero, the weighted dominant ratio and the ratio of the weighted interval between the first two categories are not calculated, and the waste textile is determined to be a material evidence condition that does not meet the requirements for direct sorting.

[0060] S4. Divide the two-dimensional real foreground region into multiple spatial units. Determine the local winner category for each spatial unit based on weighted category evidence. The local winner category is the material candidate category with the highest weighted category evidence within the current spatial unit. Based on the consistency relationship between the local winner category and the object winner category, obtain the object spatial consistency. Object spatial consistency reflects whether the global object winner receives broad spatial support within the object.

[0061] As mentioned above, in order to avoid the object winner category being contributed by only local regions, the present invention further divides the two-dimensional real foreground region into regular grids or adaptive spatial partitions to obtain multiple spatial units, so as to calculate the local winner category within the spatial unit.

[0062] like Figure 4 As shown, in this embodiment, a regular grid is established within the outer boundary of the two-dimensional real foreground region, and each grid forms multiple spatial units. For each spatial unit, only valid measurement pixels located within the two-dimensional real foreground region are counted. The grid size can be preset or adaptively adjusted according to the object size, spatial resolution, and computing resources.

[0063] As a preferred technical solution, after dividing the two-dimensional real foreground region into multiple spatial units, the method further includes obtaining the weighted effective support of each spatial unit, determining the spatial unit whose weighted effective support reaches the preset minimum unit support threshold as an effective spatial unit, and determining the object spatial consistency only within the effective spatial units.

[0064] Specifically, the spatial reliability weights of each effective measured pixel within the current spatial cell are summarized, and the summarized result is used as the weighted effective support for that spatial cell.

[0065] The weighted effective support of a spatial unit is compared with a preset minimum unit support threshold. When the weighted effective support of a spatial unit reaches the preset minimum unit support threshold, the corresponding spatial unit is determined as a valid spatial unit. When the weighted effective support of a spatial unit does not reach the preset minimum unit support threshold, the spatial unit is not used to determine the spatial consistency of the object.

[0066] It should be noted that this scheme also assesses the overall effective material evidence of the current waste textile object to avoid allowing the current object to enter direct sorting judgment only because some material candidate categories have a relative advantage when the total amount of effective material evidence is insufficient. The object refers to the current waste textile object that has formed a two-dimensional real foreground region and completed effective pixel material prediction.

[0067] Let the set of all candidate material categories be denoted as . Since the aforementioned steps have already obtained weighted category evidence for each material candidate category, the weighted category evidence for each material candidate category is summed to obtain the total weighted valid support for the object: ; in, This represents the total weighted effective support of the current waste textile objects; This represents the set of all candidate material categories; 'c' represents the material candidate category, and 'c' represents the weighted category evidence.

[0068] It should be noted that when the upstream material detection results also include conveyor belts, backgrounds, or other non-material categories, the valid measurement pixels corresponding to the non-material categories are not included in the total weighted valid support of the object.

[0069] The total weighted support for objects effectively represents the total amount of weighted material evidence that is actually obtained for the current object and can be used for object-level material judgment.

[0070] Furthermore, to evaluate the sufficiency of the weighted material evidence actually obtained for the current object relative to its theoretically obtainable material evidence, the maximum theoretical weighted support is determined based on the two-dimensional true foreground region of the current waste textile object and the established spatial reliability weighting rules. The specific process is as follows: The theoretically measurable pixel set is composed of the pixel positions within the current two-dimensional real foreground region that can theoretically generate material prediction information according to the current sampling method of the detection system. The difference between the theoretically measurable pixel set and the actual effective measurable pixel set lies in the following: the theoretically measurable pixel set is determined by the current two-dimensional real foreground region and the sampling spatial position of the detection system, without using whether the corresponding pixel actually obtains an effective material prediction result during the current detection process as a selection criterion; the actual effective measurable pixel set only includes pixels that have obtained effective material prediction information in this detection. Following the same spatial reliability weight determination method as the actual effective measurable pixel set, the corresponding spatial reliability weight is determined for each pixel position in the theoretically measurable pixel set, and these weights are accumulated to obtain the theoretical maximum weighted support. , can be represented as: ; in, This provides the most weighted theoretical support for the current object of waste textiles. For the theoretically measurable set of pixels, To effectively measure pixels, This represents the spatial reliability weight.

[0071] The relative weighted effective support of objects is obtained based on the total weighted effective support of objects and the theoretical maximum weighted support. The specific method of obtaining it is as follows .

[0072] When all spatial reliability weights are non-negative and the location where effective material prediction results are obtained actually belongs to the theoretically measurable pixel set, the object's relative weighted effective support satisfies: It should be noted that when At that time, the ratio calculation of the relative weighted effective support of the object is not performed, and the current object is determined to not meet the object-level effective support condition.

[0073] Furthermore, a preset absolute support threshold is defined. and relative support threshold , This is used to limit the minimum absolute weighted evidence required for an object to make a material determination, that is, to limit the total amount of weighted material evidence that the current object should actually obtain. This is used to limit the minimum proportion of weighted material evidence actually obtained by the current object relative to its theoretical maximum weighted support. The absolute support threshold prevents an object from obtaining a high relative support proportion but with insufficient actual material evidence; the relative support threshold prevents objects that are large in size or theoretically have a large amount of material evidence from obtaining some absolute material evidence but actually obtaining only a small proportion of the theoretically obtainable evidence. It should be noted that in this embodiment, the absolute support threshold and relative support threshold can be preset using a calibration object with known material composition before the system is put into operation.

[0074] when and When the current object meets the valid support conditions at the object level, the judgment of local winner category, object space consistency and sorting conditions is continued; when any of the above conditions are not met, the current object is determined to be an object with insufficient material evidence and direct sorting according to the object winner category is not allowed.

[0075] In one operating configuration, the default minimum unit support threshold is set to 8.

[0076] For each valid spatial cell, the valid measurement pixels belonging to different material candidate categories within that valid spatial cell are summarized. Specifically, the weighted category evidence for each material candidate category within the current valid spatial cell is obtained according to the spatial reliability weight corresponding to each valid measurement pixel. The material candidate category with the largest weighted category evidence within the current valid spatial cell is determined as the local winner category for that valid spatial cell.

[0077] Therefore, both the object winner category and the local winner category are based on the weighted category evidence after spatial reliability weight correction. The statistical range of the object winner category is the entire two-dimensional true foreground region, and the statistical range of the local winner category is the corresponding effective spatial unit.

[0078] Furthermore, this scheme obtains object spatial consistency based on the consistency relationship between local winner categories and object winner categories. Object spatial consistency is used to characterize the degree of spatial distribution of object winner categories within the two-dimensional real foreground region. The higher the proportion of effective spatial units in the two-dimensional real foreground region where local winner categories and object winner categories are consistent, the higher the object spatial consistency.

[0079] Specifically, object space consistency S o It can be represented as: ; in For an effective set of spatial units, The larger the value, the wider and more stable the spatial distribution of the winner categories. A lower value indicates that the evidence for the winner may be concentrated only in a local area, where q is a spatial unit. For total weighted support, The largest category of locally weighted evidence. For the winner category of the object.

[0080] The higher the proportion of valid spatial units in the two-dimensional true foreground region that are consistent with the local winner category and the object winner category, the higher the object spatial consistency, indicating that the object winner category is more widely distributed in the two-dimensional true foreground region. When the object winner category is identified as a local winner category only in a few valid spatial units, the object spatial consistency decreases, indicating that the object-level weighted category evidence may mainly come from local regions.

[0081] When no valid spatial unit exists, the ratio calculation for object spatial consistency is not performed, and the current waste textiles are determined to not meet the spatial consistency conditions required for direct sorting.

[0082] S5. Determine whether the waste textile objects meet the sorting conditions based on the category indicators and object space consistency. If they do, sort them.

[0083] In some embodiments, sorting conditions include at least: a dominance ratio threshold corresponding to a weighted dominance ratio, a category interval threshold corresponding to a weighted first two category intervals, and a spatial consistency threshold corresponding to object spatial consistency.

[0084] Among these, the dominance ratio threshold controls the degree of dominance that the winning category should achieve in the overall material evidence. The category interval threshold controls the degree of category separation that the winning category should have relative to the second candidate category. The spatial consistency threshold controls the degree of spatial distribution that the winning category should have within the object.

[0085] In this embodiment, the process of setting sorting conditions includes: Before the system is put into operation, calibrated objects with known material composition are acquired. For each calibrated object, the dominant ratio threshold, category interval threshold, and spatial consistency threshold are calculated according to steps S1 to S4 of this embodiment, and it is recorded whether the object's winner category is consistent with the known material category. Multiple candidate threshold combinations are pre-set. For each candidate threshold combination, the number of calibrated objects that meet the direct sorting conditions and the number of calibrated objects whose winner category is consistent with the known material category are counted. This yields the direct sorting coverage and direct sorting correctness corresponding to the candidate threshold combination. Based on the target purity required by the target material flow, threshold combinations that can achieve the target purity requirements are selected.

[0086] Example 2: Please see Figure 5As shown, the difference between this embodiment and Embodiment 1 is that, before determining whether the waste textile objects meet the sorting conditions based on the category indicators and object space consistency, the physical validity status of the waste textile objects is also determined, and three processing results are distinguished based on the physical validity status of the objects: rejection, return for re-sorting, and direct sorting. The remaining steps are the same as in Embodiment 1.

[0087] That is, as a preferred technical solution, before determining whether the waste textile objects meet the sorting conditions, the physical validity state of the waste textile objects is also determined; the physical validity state of the objects is used to characterize whether the waste textile objects have the physical conditions to perform sorting. This avoids classifying objects with uncertain material properties and objects that cannot be reliably sorted into the same processing result.

[0088] In this second embodiment, determining whether waste textiles meet the sorting conditions specifically includes: When the physical validity of a waste textile object is invalid, the rejection result is determined.

[0089] When the physical validity status of the waste textile object is valid, and the consistency of the category index and the object space does not meet the sorting conditions, the result of re-sorting is determined.

[0090] When the physical validity status of an object is valid, and the category index and object space consistency meet the sorting conditions, the direct sorting result is determined.

[0091] The physical validity status of an object is used to characterize whether the current waste textile object possesses the physical conditions for sorting by a sorting execution mechanism. That is, the physical validity status of an object is used to evaluate whether the current waste textile object itself has a feasible physical basis for sorting. In this embodiment, the physical validity status of the object is determined based on the object's projected area and effective measurement coverage.

[0092] First, the object's projected area is determined based on the two-dimensional real foreground region. The detection system pre-establishes a spatial calibration relationship between image pixel size and actual size, and converts the pixel area corresponding to the two-dimensional real foreground region into the object's projected area based on this spatial calibration relationship.

[0093] The projected area of ​​the object is compared with the preset minimum physical projected area. When the projected area of ​​the object reaches the preset minimum physical projected area, the current waste textile object is determined to meet the area requirement; when the projected area of ​​the object is lower than the preset minimum physical projected area, the current waste textile object is determined not to meet the area requirement. In one operating configuration, the preset minimum physical projected area is set to 3000 mm².

[0094] The effective measurement coverage is further determined based on the number of foreground pixels in the two-dimensional real foreground region and the number of effective measurement pixels. Specifically, the number of effective measurement pixels is compared with the number of foreground pixels in the two-dimensional real foreground region to characterize the coverage of the area where effective material prediction information can be obtained in the current waste textile object.

[0095] The effective measurement coverage is compared with the preset minimum effective measurement coverage condition. If the effective measurement coverage reaches the preset minimum effective measurement coverage condition, the current waste textile object is determined to meet the measurement coverage condition; otherwise, the current waste textile object is determined not to meet the measurement coverage condition.

[0096] When both the area condition and the measurement coverage condition are met, the physical valid state of the object is determined to be valid; when at least one of the area condition and the measurement coverage condition is not met, the physical valid state of the object is determined to be invalid.

[0097] In another embodiment, the physical effective state of an object can be further restricted based on geometric integrity and / or observability.

[0098] When an object's physical validity status is invalid, a rejection result is determined. A rejection result indicates that the current object will not enter the normal material sorting channel determined by the object's winner category, but will instead enter the rejection processing channel. This is because the current object itself lacks the physical conditions to reliably perform sorting, rather than simply due to insufficient material evidence for the object's winner category.

[0099] When an object's physical validity is valid, but its category index and object spatial consistency do not meet the sorting conditions, a re-sorting result is determined. The re-sorting result indicates that the current object has executable physical sorting conditions, but the material evidence obtained in this detection is insufficient to support direct entry into the material flow corresponding to the object's winner category. Therefore, the current object is sent to the re-sorting channel and re-enters the subsequent material flow to be tested. When the re-sorted object re-enters the detection area, it re-obtains the material prediction information of the two-dimensional true foreground area and each valid measurement pixel, and re-determines the spatial reliability weight, weighted category evidence, category index, and object spatial consistency. It is not required to inherit the object identity corresponding to the previous detection.

[0100] When the physical validity status of an object is valid, and the category index and object spatial consistency both meet the sorting conditions, the direct sorting result is determined, and the current waste textile object is directed to the corresponding material flow according to the object winner category.

[0101] This means that during the routing decision phase, the physical validity status of an object is first determined based on its projected area, effective measurement coverage, geometric integrity, or observability. If the object's physical validity status does not meet the physical validity conditions, the object enters the rejection channel. If the object's physical validity status meets the physical validity conditions, it is then determined whether the weighted dominance ratio, the weighted first two-class interval, and the object spatial consistency have reached their respective weighted dominance ratio threshold, weighted first two-class interval threshold, and object spatial consistency threshold. When all three are satisfied, a direct sorting instruction is generated; otherwise, a return and re-sorting instruction is generated. This can be represented as: ; in, For the routing decision results, This indicates an instruction to enter the rejection channel. This indicates a direct sorting instruction. This indicates a re-sorting instruction. For the physical effective state of the object, For the weighted dominant ratio, m o S is the weighted interval between the first two categories. o For object space consistency, For the weighted dominant ratio threshold, The threshold for the weighted interval between the first two categories. The object space consistency threshold. It represents any other case besides the other two.

[0102] Example 3: The difference between this embodiment and Embodiment 1 is that, after obtaining the object winner category, the second candidate category, the category index, and the object space consistency, it is further determined whether the object winner category and the second candidate category constitute a preset composition boundary category pair; when they constitute a preset composition boundary category pair, a special sorting condition that is stricter than the general sorting condition is adopted, and the remaining steps are the same as in Embodiment 1.

[0103] That is, as a preferred technical solution, before determining whether the waste textile object meets the sorting conditions, it is also determined whether the winning category of the object and the second candidate category corresponding to the evidence of the second largest weighted category constitute a preset composition boundary category pair.

[0104] When the winner category of an object and the second candidate category form a preset composition boundary category pair, a special sorting condition that is stricter than the sorting condition is adopted.

[0105] As a preferred technical solution, determining whether the winner category of the object and the second candidate category corresponding to the evidence of the second largest weighted category constitute a preset composition boundary category pair includes the following process: Based on the material composition range corresponding to the winner category and the second candidate category, determine whether the two belong to adjacent composition categories under the same material system.

[0106] When both belong to adjacent constituent categories, and the arrangement relationship between the object winner category and the second candidate category conforms to a preset direction, the two are determined to constitute a preset constituent boundary category pair.

[0107] Before the system is put into operation, corresponding material composition intervals and their arrangement relationships are established based on the material composition range of the material candidate categories. For the winning object category and the second candidate category, their corresponding material composition intervals are obtained. It is then determined whether the winning object category and the second candidate category belong to adjacent composition categories within the same material system.

[0108] "Same material system" means that two candidate material categories are divided into intervals based on the same basic material composition; "adjacent composition categories" means that after sorting according to the corresponding material composition parameters, there are no other composition categories between the two categories. For example, cotton / polyester blended materials can be divided into multiple adjacent composition categories according to cotton content. In this case, each category has a corresponding cotton content interval, arranged in ascending order of cotton content.

[0109] After determining that two categories are adjacent constituent categories, the next step is to determine whether the arrangement relationship between the winning category and the second candidate category conforms to a preset direction. The preset direction represents a pre-determined direction of category competition that requires stricter direct sorting conditions. For example, for two adjacent cotton / polyester blend categories, if the lower cotton category is the winning category and the higher cotton category is the second candidate category, this arrangement relationship can be determined as the preset direction. In this case, the winning category and the second candidate category constitute a preset constituent boundary category pair.

[0110] When the winning category and the second candidate category form a preset boundary category pair, a special sorting condition, stricter than the sorting condition in Example 1, is applied. The special sorting condition may include at least one of a special dominance ratio condition, a special first-two category interval condition, and a special spatial consistency condition. The special dominance ratio condition is used to increase the minimum dominance requirement of the winning category in the overall material evidence. The special first-two category interval condition is used to increase the minimum evidence separation requirement between the winning category and the second candidate category. The special spatial consistency condition is used to increase the minimum spatial distribution requirement of the winning category in the two-dimensional real foreground region.

[0111] Preset component boundary category pairs and dedicated sorting conditions do not participate in recalculating the object winner category, but are only used to control whether the material evidence corresponding to the current object winner category is sufficient to allow direct entry into the target material flow.

[0112] In one specific operating configuration, the dedicated dominance ratio threshold is set to 0.750, the dedicated first two category interval threshold is set to 0.600, and the dedicated space consistency threshold is set to 0.850.

[0113] When the weighted dominant ratio, the weighted interval between the first two categories, and the object space consistency meet the corresponding dedicated sorting conditions, direct sorting is allowed according to the already determined object winner category.

[0114] When the specific sorting conditions are not met, the winner category of the object is not changed, and the current waste textiles are identified as the result of recycling and re-sorting.

[0115] Example 4: Verification of waste textile sorting under online conveying conditions The system obtains completed objects from the upstream object formation and material inference module. Each completed object contains at least a two-dimensional real foreground region M. o Effective measurement pixel set P o Pixel coordinates and material prediction category In this embodiment, the detection object comes from a line-scan near-infrared hyperspectral detection system.

[0116] The running categories in this experiment include cotton, polyester, cotton / elastic fiber blend, polyester / elastic fiber blend, and two cotton / polyester blend categories, resulting in a total of 90 completed material objects.

[0117] For the completed material object, the true outline of the object is calculated based on the two-dimensional real foreground region, and the inward Euclidean distance to the true outline is calculated for each valid measured pixel. . use Establish a spatial reliability field. For irregular, wrinkled, rolled, or locally bulging fabric pieces, the reliability field adapts to the actual object contour, instead of being uniformly deleted based on a fixed horizontal edge width.

[0118] For each completed material object, the corresponding two-dimensional real foreground region and the material prediction information of each effective measured pixel are obtained. The spatial reliability weight is determined based on the inward distance of each effective measured pixel relative to the contour of the two-dimensional real foreground region. In the running configuration used in this experimental example, the spatial reliability weights for the edge region, transition region, and interior region are set to 0, 0.35, and 1.00, respectively, and the weighted category evidence for each material candidate category is summarized based on the spatial reliability weight.

[0119] To compare the impact of spatial reliability weighting on the determination of the winner category of the object, 90 completed material objects were processed using both the object majority voting method without spatial reliability weighting and the weighted category evidence determination method based on spatial reliability weights.

[0120] When using the unweighted majority voting method, the winning category of 85 completed material objects was consistent with the corresponding independent component reference category, with a consistency rate of 94.4%. After determining the winning category using weighted category evidence based on spatial reliability weights, the winning category of 88 completed material objects was consistent with the corresponding independent component reference category, with a consistency rate of 97.8%. Compared to the unweighted majority voting method, the three completed material objects that were corrected were all cotton / polyester blended objects whose original statistics were affected by evidence from adjacent categories near the contour.

[0121] The results above show that by changing the contribution of effective measured pixels to object-level material evidence based on their position relative to the object contour, the abnormal influence of material prediction results near the contour on the object winner category can be reduced, while retaining the main evidence contribution of effective measured pixels inside the object to the object-level judgment.

[0122] After determining the object winner category, the weighted dominance ratio and weighted first two category interval are obtained based on the weighted category evidence of each material candidate category, and the two-dimensional true foreground region is divided into spatial units. In the operating configuration used in this experimental example, the minimum weighted effective support of the spatial unit is set to 8. Spatial units whose weighted effective support does not meet this condition do not participate in the determination of object spatial consistency. For spatial units that meet the condition, the corresponding local winner category is determined, and the object spatial consistency is obtained based on the consistency relationship between the local winner category and the object winner category.

[0123] Subsequently, based on the weighted dominance ratio, the weighted interval between the first two categories, and the consistency of the object space, it is determined whether each completed material object has sufficient material evidence to perform direct sorting. For objects that are physically valid but whose material evidence does not meet the conditions for direct sorting, a return and re-sorting result is generated; for objects that are physically valid and whose material evidence meets the conditions for direct sorting, a direct sorting result is generated. When the physical validity of an object does not meet the corresponding physical conditions, a rejection result is generated.

[0124] Of the 90 completed material objects, 78 met the conditions for direct sorting and were sorted directly. 12 objects were returned for re-sorting due to insufficient material evidence. No objects were rejected due to invalid physical states during this verification. The sorting categories of the 78 directly sorted objects were consistent with their corresponding independent component reference categories.

[0125] The above results indicate that, based on reducing the impact of low-reliability prediction information near the contour on the object winner category by using spatial reliability weights, further utilizing weighted dominance ratios to characterize the dominance of the object winner category in the overall material evidence, utilizing weighted first-two category intervals to characterize the degree of competition between the object winner category and the second candidate category, and utilizing object spatial consistency to characterize the spatial distribution of the object winner category within the two-dimensional real foreground region, it is possible to screen out objects with insufficient material evidence from the directly sorted objects and send them into the return and re-sorting process.

[0126] For two cotton / polyester blended fabric operation categories, when the winning category and the second candidate category form a preset composition boundary category pair, stricter special sorting conditions than general sorting conditions can be further applied. In one operation configuration, the special dominant ratio threshold, the special first two category interval threshold, and the special spatial consistency threshold are set to 0.750, 0.600, and 0.850, respectively; if the special sorting conditions are not met, the winning category remains unchanged and a re-sorting result is generated.

[0127] Example 5: Please see Figure 2 As shown, this solution also provides a waste textile sorting system that integrates spatial reliability within the object, specifically including: Complete the object data interface module, which is used to obtain the two-dimensional real foreground area of ​​waste textiles with completed material prediction and the material prediction information of each effective measurement pixel.

[0128] The reliability evidence fusion module is used to determine the spatial reliability weight of each effective measurement pixel based on the contour of the two-dimensional real foreground region and the inward distance of each effective measurement pixel, and to summarize the material prediction information of each effective measurement pixel based on the spatial reliability weight to obtain the weighted category evidence of each material candidate category, wherein the spatial reliability weight of effective measurement pixels closer to the contour is low.

[0129] The category reliability assessment module is used to determine the object winner category and category index based on weighted category evidence. The object winner category is the material candidate category with the largest weighted category evidence, and the category index is used to characterize the reliability of the object winner category.

[0130] The object space consistency determination module is used to divide the two-dimensional real foreground region into multiple spatial units, determine the local winner category of each spatial unit based on weighted category evidence, where the local winner category is the material candidate category with the largest weighted category evidence in the current spatial unit, and obtain the object space consistency based on the consistency relationship between the local winner category and the object winner category.

[0131] The sorting judgment module is used to determine whether waste textile objects meet the sorting conditions based on category indicators and object space consistency. If they meet the conditions, they are sorted.

[0132] It should be noted that the specific examples in this embodiment can refer to the examples described in the above embodiments and optional implementations, and will not be repeated here.

[0133] Generally, various embodiments can be implemented in hardware or dedicated circuitry, software, logic, or any combination thereof. Some aspects of the invention can be implemented in hardware, while others can be implemented by firmware or software executed by a controller, microprocessor, or other computing device, but the invention is not limited thereto. Although various aspects of the invention may be shown and described as block diagrams, flowcharts, or using some other graphical representation, it should be understood that, by way of non-limiting example, these blocks, apparatuses, systems, techniques, or methods described herein can be implemented in hardware, software, firmware, dedicated circuitry or logic, general-purpose hardware or controllers or other computing devices, or some combination thereof.

[0134] Embodiments of the present invention can be implemented by computer software, which may be executable by a data processor of a mobile device, such as a processor entity, or by hardware, or by a combination of software and hardware. Computer software or programs (also referred to as program products), including software routines, applets, and / or macros, can be stored in any device-readable data storage medium, and they include program instructions for performing specific tasks. A computer program product may include one or more computer-executable components configured to perform embodiments when the program is run. One or more computer-executable components may be at least one piece of software code or a portion thereof. Additionally, it should be noted that any block in the logical flow of the figures may represent a program step, or interconnected logical circuitry, blocks and functions, or a combination of program steps and logical circuitry, blocks and functions. The software may be stored on physical media such as memory chips or blocks of storage implemented within a processor, magnetic media such as hard disks or floppy disks, and optical media such as, for example, DVDs and their data variants, CDs, etc. The physical medium is a non-transient medium.

[0135] Those skilled in the art should understand that the technical features of the above embodiments can be combined in any way. For the sake of brevity, not all possible combinations of the technical features in the above embodiments have been described. However, as long as there is no contradiction in the combination of these technical features, they should be considered to be within the scope of this specification.

[0136] The above embodiments are merely illustrative of several implementation methods of this application, and their descriptions are relatively specific and detailed, but they should not be construed as limiting the scope of this application. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of this application, and these all fall within the protection scope of this application. Therefore, the protection scope of this application should be determined by the appended claims.

Claims

1. A method for sorting waste textiles based on spatial reliability fusion within an object, characterized in that, include: S1. Obtain the two-dimensional real foreground region of the waste textile with completed material prediction and the material prediction information of each effective measurement pixel. S2. Based on the contour of the two-dimensional real foreground region and the inward distance of each effective measurement pixel, determine the spatial reliability weight of each effective measurement pixel, and summarize the material prediction information of each effective measurement pixel based on the spatial reliability weight to obtain the weighted category evidence of each material candidate category, wherein the spatial reliability weight of effective measurement pixels closer to the contour is low. S3. Determine the object winner category and category index based on the weighted category evidence, wherein the object winner category is the material candidate category with the largest weighted category evidence, and the category index is used to characterize the reliability of the object winner category; S4. Divide the two-dimensional real foreground region into multiple spatial units, determine the local winner category of each spatial unit based on weighted category evidence, wherein the local winner category is the material candidate category with the largest weighted category evidence in the current spatial unit, and obtain the object spatial consistency based on the consistency relationship between the local winner category and the object winner category. S5. Determine whether the waste textile objects meet the sorting conditions based on the category indicators and the object space consistency. If they do, sort them.

2. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 1, characterized in that: The spatial reliability weights of effective measurement pixels monotonically increase sequentially from the contour to the center; or, the two-dimensional real foreground region is divided into different regions according to the inward distance, and the spatial reliability weights of effective measurement pixels in the same region are the same, while the spatial reliability weights of regions closer to the contour are less than those closer to the center.

3. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 1, characterized in that: The process of summarizing each effective measurement pixel based on the spatial reliability weight to obtain weighted category evidence for each material candidate category specifically includes: The category metrics include a weighted dominance ratio and a weighted first-two category interval. The weighted dominance ratio is the ratio of the sum of weighted category evidence for the winner category to the sum of weighted category evidence for all candidate material categories. The weighted first-two category interval is the normalized result of the difference between the winner category and the second largest weighted category evidence relative to the sum of weighted category evidence.

4. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 1, characterized in that: After dividing the two-dimensional real foreground region into multiple spatial units, the method further includes obtaining the weighted effective support of each spatial unit, determining the spatial unit whose weighted effective support reaches a preset minimum unit support threshold as a valid spatial unit, and determining the object spatial consistency only within the valid spatial unit.

5. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 1, characterized in that: The object spatial consistency is used to characterize the spatial distribution of the object winner category within the two-dimensional real foreground region; the higher the proportion of effective spatial units where the local winner category is consistent with the object winner category within the two-dimensional real foreground region, the higher the object spatial consistency.

6. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 1, characterized in that: Before determining whether the waste textile object meets the sorting conditions, the physical validity status of the waste textile object is also determined; the physical validity status is used to characterize whether the waste textile object has the physical conditions to perform sorting.

7. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 6, characterized in that: Determining whether the waste textile objects meet the sorting conditions specifically includes: When the physical validity state of the waste textile object is invalid, a rejection result is determined; When the physical validity status of the waste textile object is valid, and the consistency between the category index and the object space does not meet the sorting conditions, the return and re-sorting result is determined. When the physical validity status of the object is valid, and the consistency between the category index and the object space satisfies the sorting conditions, the direct sorting result is determined.

8. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 1, characterized in that: Before determining whether the waste textile object meets the sorting conditions, it is also determined whether the winner category of the object and the second candidate category corresponding to the evidence of the second largest weighted category constitute a preset composition boundary category pair; When the object winner category and the second candidate category constitute the preset composition boundary category pair, a special sorting condition that is stricter than the sorting condition is adopted.

9. The method for sorting waste textiles based on spatial reliability fusion within an object as described in claim 8, characterized in that: The specific process of determining whether the winner category of the object and the second candidate category corresponding to the evidence of the second largest weighted category constitute a preset composition boundary category pair includes: Based on the material composition range corresponding to the object winner category and the second candidate category, determine whether the two belong to adjacent composition categories under the same material system; When both belong to adjacent constituent categories, and the arrangement relationship between the object winner category and the second candidate category conforms to a preset direction, it is determined that the two constitute a preset constituent boundary category pair.

10. A waste textile sorting system based on spatial reliability fusion within an object, characterized in that, Specifically, it includes: Complete the object data interface module, which is used to obtain the two-dimensional real foreground area of ​​the waste textiles with completed material prediction and the material prediction information of each effective measurement pixel; The reliability evidence fusion module is used to determine the spatial reliability weight of each effective measurement pixel based on the contour of the two-dimensional real foreground region and the inward distance of each effective measurement pixel, and to summarize the material prediction information of each effective measurement pixel based on the spatial reliability weight to obtain the weighted category evidence of each material candidate category, wherein the spatial reliability weight of effective measurement pixels closer to the contour is low. The category reliability assessment module is used to determine the object winner category and category index based on the weighted category evidence, wherein the object winner category is the material candidate category with the largest weighted category evidence, and the category index is used to characterize the reliability of the object winner category; The object space consistency determination module is used to divide the two-dimensional real foreground region into multiple spatial units, determine the local winner category of each spatial unit based on weighted category evidence, wherein the local winner category is the material candidate category with the largest weighted category evidence in the current spatial unit, and obtain the object space consistency according to the consistency relationship between the local winner category and the object winner category; The sorting determination module is used to determine whether the waste textile objects meet the sorting conditions based on the category indicators and the object space consistency. If they meet the conditions, sorting is performed.

Images