Material identification methods, material sorting methods, and material sorting equipment

By acquiring material images using dual-energy X-ray technology, connected component detection and pixel-by-pixel category recognition are performed, solving the problem of low recognition accuracy for complex materials and improving the recognition accuracy and sorting performance of material sorting equipment.

CN123118629APending Publication Date: 2026-11-17BEIJING HONEST TECHNOLOGY CO LTD
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202611611072.9
Authority / Receiving Office
CN · China
Patent Type
Applications(China)
Current Assignee / Owner
Filing Date
2026-10-19
Publication Date
2026-11-17

AI Technical Summary

Technical Problem

Existing technologies struggle to accurately identify complex materials with irregular shapes, large particle size differences, and non-uniform internal structures, resulting in low classification accuracy and impacting the sorting performance of material sorting equipment.

Method used

High-energy and low-energy images are acquired using dual-energy X-ray technology. Connected component detection is performed using the low-energy images. Effective connected components are extracted by combining the minimum axis-aligned bounding rectangle. Pixel-by-pixel category detection is performed to determine the target material category and outer contour information. A pre-trained target classification model is used to improve recognition accuracy.

Benefits of technology

It improves the recognition accuracy under complex material conditions, rationally allocates recognition computing resources, achieves a balance between recognition accuracy and algorithm running efficiency, and ensures the sorting performance of material sorting equipment.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN123118629A_ABST
    Figure CN123118629A_ABST
Patent Text Reader

Abstract

This disclosure relates to the field of computer vision technology, specifically to a material identification method, a material sorting method, and a material sorting device. The material identification method includes: determining at least one effective connected component and its corresponding minimum axis-aligned bounding rectangle based on the results of connected component detection on a low-energy image; extracting a dual-energy target image sub-block corresponding to each effective connected component from a dual-energy image based on the position of the minimum axis-aligned bounding rectangle of each effective connected component; determining at least one predicted category and its corresponding pixel information for each dual-energy target image sub-block based on pixel-by-pixel category detection; and then, combining the effective number of predicted categories, determining at least one target material category corresponding to each effective connected component and the effective outer contour information corresponding to the target material category. This method can improve the identification accuracy under complex material conditions and ensure the sorting performance of subsequent material sorting equipment.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] This disclosure relates to the field of computer vision technology, specifically to a material identification method, a material sorting method, and a material sorting device. Background Technology

[0002] With the rapid development of the renewable resource recycling industry, the market demand for efficient and accurate automatic material sorting technologies and equipment is becoming increasingly urgent. Among related technologies, dual-energy X-ray technology obtains transmission data from both high-energy and low-energy bands to determine characteristic parameters related to the effective atomic number of materials, which can be used for the identification of materials of different materials.

[0003] However, when performing material identification, the relevant technologies mainly rely on manually designed features combined with traditional classifiers to determine the material category. The feature expression capability is limited. For materials with irregular shapes, large particle size differences, and non-uniform internal structures, the classification accuracy is difficult to meet the requirements of industrial sorting, which affects the sorting performance of material sorting equipment. Summary of the Invention

[0004] To overcome the problems existing in related technologies, an exemplary embodiment of this disclosure provides a material identification method applied to a material sorting device. The method includes: acquiring a dual-energy image of the material to be identified, wherein the dual-energy image includes a high-energy image and a low-energy image; determining at least one valid connected component and its corresponding minimum axis-aligned bounding rectangle based on the result of connected component detection of the low-energy image; extracting images from the high-energy image and the low-energy image based on the position of the minimum axis-aligned bounding rectangle of each valid connected component to obtain a dual-energy target image sub-block corresponding to each valid connected component; determining at least one predicted category and its corresponding pixel information included in the corresponding valid connected component based on the result of pixel-by-pixel category detection of each dual-energy target image sub-block; and determining at least one target material category and its corresponding valid outer contour information based on the at least one predicted category and its corresponding pixel information included in each valid connected component, as well as the effective number of predicted categories.

[0005] In some embodiments, based on the result of pixel-by-pixel category detection for each dual-energy target image sub-block, determining at least one predicted category and corresponding pixel information included in the corresponding effective connected component includes: performing category detection on the fusion features of the dual-energy target image sub-block pixel by pixel to determine the predicted category corresponding to each pixel; and dividing all pixels according to their corresponding predicted categories based on the position of each pixel in the dual-energy target image sub-block to obtain at least one predicted category and corresponding pixel information included in the effective connected component corresponding to the dual-energy target image sub-block.

[0006] In some embodiments, the dual-energy target image sub-block includes a high-energy target image sub-block and a low-energy target image sub-block. The fusion features of the dual-energy target image sub-block are subjected to class detection pixel-by-pixel to determine the predicted class corresponding to each pixel. This includes: stitching the high-energy target image sub-block and the low-energy target image sub-block to obtain a stitched tensor; extracting the multi-scale spatial features of the stitched tensor and reconstructing fusion features consistent with the original spatial resolution of the stitched tensor; and performing class detection pixel-by-pixel on the fusion features to determine the predicted class corresponding to each pixel.

[0007] In some embodiments, the predicted category corresponding to a pixel is obtained through inference by a pre-trained target classification model, wherein the input of the target classification model is a sub-block of a dual-energy target image, and the output is the predicted category corresponding to each pixel in the dual-energy target image sub-block.

[0008] In some embodiments, based on the position of each pixel in a dual-energy target image sub-block, all pixels are divided according to their corresponding predicted categories to obtain at least one predicted category included in the effective connected component of the dual-energy target image sub-block and the corresponding pixel information. This includes: determining whether the predicted categories corresponding to two adjacent pixels are the same based on the position of each pixel in the dual-energy target image sub-block; if the predicted categories corresponding to two adjacent pixels are the same, then the two adjacent pixels are grouped into the same group, and the pixel information of the corresponding predicted category is updated based on the position information of the two adjacent pixels; if the predicted categories corresponding to two adjacent pixels are different, then the two adjacent pixels are grouped into different groups, and the pixel information of the corresponding predicted category is updated based on the position information of each pixel; after all pixels are divided, at least one predicted category included in the effective connected component of the dual-energy target image sub-block and the corresponding pixel information are obtained, wherein the same predicted category includes at least one pixel group, and each pixel group includes at least one pixel.

[0009] In some embodiments, the predicted category includes background or valid category. Based on at least one predicted category included in each valid connected region and the corresponding pixel information, as well as the valid number of predicted categories, at least one target material category corresponding to each valid connected region and the valid outer contour information corresponding to the target material category are determined, including: counting the valid number of valid categories according to all predicted categories included in the valid connected region; if the valid number is greater than or equal to 1, then based on the pixel information corresponding to each valid category, at least one target material category corresponding to the valid connected region and the valid outer contour information corresponding to the target material category are determined.

[0010] In some embodiments, if the effective quantity is greater than or equal to 1, then based on the pixel information corresponding to each effective category, at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category are determined, including: if the effective quantity is equal to 1, then the corresponding effective category is directly used as the target material category corresponding to the effective connected region, and the effective outer contour information corresponding to the target material category is determined based on the pixel information corresponding to the effective category; if the effective quantity is greater than 1, then based on the pixel information corresponding to each effective category, the spatial positional relationship between the contour local regions corresponding to each effective category is determined, and based on the spatial positional relationship, at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category are determined.

[0011] In some embodiments, determining the spatial positional relationship between the contour local regions corresponding to each effective category based on the pixel information corresponding to each effective category includes: determining whether there are spatially adjacent contour local region pairs based on the pixel information corresponding to each effective category, wherein the contour local regions correspond one-to-one with the pixel information of the effective categories; if there are at least one spatially adjacent contour local region pair, then determining the spatial positional relationship between the corresponding contour local regions based on the overlap detection result of each contour local region pair; if there are no spatially adjacent contour local region pairs, then determining that the spatial positional relationship between the contour local regions corresponding to each effective category is mutually independent.

[0012] In some embodiments, based on the overlap detection results of each contour local region pair, the spatial positional relationship between corresponding contour local regions is determined, including: for each contour local region pair, extracting their respective corresponding pixel masks to obtain a first pixel mask and a second pixel mask; based on the first pixel mask, determining a first minimum axis-aligned bounding rectangle corresponding to the valid category, and based on the second pixel mask, determining a second minimum axis-aligned bounding rectangle corresponding to the valid category; and based on the comparison result between the intersection-union ratio (IU) of the first and second minimum axis-aligned bounding rectangles and a specified threshold, determining the spatial positional relationship between corresponding contour local region pairs.

[0013] In some embodiments, the spatial positional relationship between corresponding contour local regions is determined based on the comparison result between the intersection-union ratio (IU) between the first minimum axis-aligned bounding rectangle and the second minimum axis-aligned bounding rectangle and a specified threshold. This includes: if the comparison result indicates that the IU is greater than the specified threshold, then the spatial positional relationship between the corresponding contour local regions of the two valid categories is determined to be an overlapping relationship; if the comparison result indicates that the IU is less than or equal to the specified threshold, then the spatial positional relationship between the corresponding contour local regions of the two valid categories is determined to be an adhesive relationship.

[0014] In some embodiments, determining at least one target material category corresponding to an effective connected region and the effective outer contour information corresponding to the target material category based on spatial positional relationships includes: if the spatial positional relationship is overlapping, merging two effective categories into a whole and determining the overall target material category, and determining the effective outer contour information corresponding to the target material category based on the merging result of the corresponding contour local region pairs; if the spatial positional relationship is adhering, selectively performing image segmentation processing based on the adhesion detection result of the corresponding contour local region pairs, and determining the corresponding target material category and the effective outer contour information corresponding to the target material category respectively; if the spatial positional relationship is mutually independent, determining that the corresponding effective categories are all target material categories, and determining the effective outer contour information corresponding to the corresponding target material category based on the corresponding pixel information.

[0015] Secondly, this disclosure also provides a material sorting method applied to a material sorting device. The method includes: determining at least one target material category and the effective outer contour information corresponding to the target material category by means of the material identification method provided in the first aspect or any corresponding embodiment; generating a sorting control instruction based on each target material category and the corresponding effective outer contour information; and sorting materials according to the sorting control instruction.

[0016] Thirdly, this disclosure also provides a material sorting device, comprising: a transmission mechanism for conveying materials; a dual-energy X-ray recognition module for acquiring dual-energy images of the materials; a material recognition mechanism for identifying at least one target material category and the corresponding effective outer contour information in the dual-energy image using the material recognition method provided in the first aspect or any corresponding embodiment, and generating sorting control instructions based on each target material category and the corresponding effective outer contour information; and a sorting mechanism for receiving and sorting materials according to the sorting control instructions.

[0017] It should be understood that the above general description and the following detailed description are exemplary and explanatory only, and are not intended to limit this disclosure.

[0018] The technical solutions provided by the embodiments of this disclosure can include the following beneficial effects: According to the material identification method provided by this disclosure, by performing connected component detection on the low-energy image in the dual-energy image, it can be ensured that at least one effective connected component determined in the end is reliable data. Image extraction based on the minimum axis-aligned bounding rectangle corresponding to the effective connected component can ensure that the background pixel content in the extracted dual-energy target image sub-block is extremely low, reducing the interference of background information on material identification. By performing category detection pixel by pixel on the dual-energy target image sub-block, obtaining pixel-level predicted distribution with pixels as detection units, and then combining the number of predicted categories in the effective connected component with the pixel distribution to comprehensively determine at least one corresponding target material category and the effective outer contour information corresponding to the target material category, the identification accuracy under complex material conditions can be improved, identification computing resources can be reasonably allocated, a balance between identification accuracy and algorithm running efficiency can be achieved, and the sorting performance of subsequent material sorting equipment can be guaranteed. Attached Figure Description

[0019] This disclosure can be better understood by describing exemplary embodiments of the present disclosure in conjunction with the accompanying drawings, in which: Figure 1 This is a schematic diagram of the architecture of a material sorting device according to an exemplary embodiment of the present disclosure; Figure 2 This is a schematic flowchart illustrating a material identification method according to an exemplary embodiment of the present disclosure; Figure 3 This is a schematic flowchart illustrating a material sorting method according to an exemplary embodiment of the present disclosure; Figure 4 This is a schematic diagram of the structure of a material sorting device according to an exemplary embodiment of the present disclosure. Detailed Implementation

[0020] The following describes specific embodiments of this disclosure. It should be noted that, in order to provide a concise description, this specification cannot exhaustively describe all features of the actual embodiments. It should be understood that, in the actual implementation of any embodiment, just as in any engineering or design project, various specific decisions are often made to achieve the developer's specific goals and to meet system-related or business-related constraints, and this can change from one embodiment to another. Furthermore, it is understood that although the efforts made in this development process may be complex and lengthy, for those skilled in the art related to the content disclosed in this disclosure, changes in design, manufacturing, or production based on the technical content disclosed in this disclosure are merely conventional technical means and should not be construed as insufficient content of this disclosure.

[0021] Unless otherwise defined, the technical or scientific terms used in this disclosure shall have the ordinary meaning understood by one of ordinary skill in the art to which this disclosure pertains. The terms “first,” “second,” and similar terms used in this disclosure do not indicate any order, quantity, or importance, but are merely used to distinguish different components. The terms “a” or “one,” etc., do not indicate a quantity limitation, but rather indicate the presence of at least one. The terms “comprising,” “including,” etc., mean that the element or object preceding “comprising” or “including” encompasses the elements or objects listed following “comprising” or “including” and their equivalents, and do not exclude other elements or objects. The terms “connected,” “linked,” etc., are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

[0022] Material sorting equipment typically uses a conveyor belt to feed the materials to be identified into the equipment. A radiographic detection device emits X-rays to acquire X-ray images of the materials. These images are then analyzed and processed using a pre-defined material sorting algorithm to achieve material identification and sorting. For example... Figure 1 As shown, the material sorting equipment 100 may include a feeding mechanism 110, a conveying mechanism 120, a detection mechanism 130, and a sorting mechanism 140. The feeding mechanism 110 feeds the material to be sorted into the conveying mechanism 120. The conveying mechanism 120 may be a conveyor belt or a chute, etc., used to transport the material to be identified fed by the feeding mechanism 110. The detection mechanism 130 is used to identify the material and determine its category. The detection mechanism 130 includes a dual-energy X-ray identification module and a material identification module. The dual-energy X-ray identification module collects transmission data of the material in both high-energy and low-energy bands, generating corresponding dual-energy image data. The material identification module determines the material category based on the processing results of the dual-energy data. The sorting mechanism 140 sorts the material according to the identified material category.

[0023] In related technologies, dual-energy X-ray material identification schemes often rely on manual design of image features based on human experience, combined with traditional machine learning classifiers to determine the material's composition and category. These schemes have fixed feature extraction methods and weak model generalization capabilities, making them unable to adaptively fit the actual imaging distribution characteristics of complex recycled materials. For complex materials commonly found in recycled resource scenarios, characterized by irregular shapes, large particle size ranges, non-uniform internal components, and unstable imaging grayscale, traditional identification methods are prone to feature representation failure, material category confusion, and misidentification or omission of local areas. They struggle to accurately distinguish the boundaries and material distribution of adhered and stacked materials, failing to output precise material category information and effective contour information. Their identification accuracy and operational stability are limited, making it difficult to meet the requirements of high-precision industrial sorting and directly restricting the overall sorting performance and efficiency of material sorting equipment.

[0024] Therefore, this disclosure provides a material identification method applied to material sorting equipment. For example... Figure 2 As shown, the material identification method may include the following steps: Step S210: Obtain a dual-energy image of the material to be identified.

[0025] In this disclosure, the material sorting equipment is equipped with a dual-energy X-ray recognition module, which may include an X-ray source and a dual-energy X-ray linear array detector. The X-ray source emits high-energy X-rays and low-energy X-rays; the dual-energy X-ray linear array detector can achieve material imaging based on the difference in attenuation coefficients of different materials to high-energy and low-energy X-rays. The image acquisition area of ​​the dual-energy X-ray linear array detector corresponds to the position of the transmission mechanism of the material sorting equipment, enabling data acquisition during material transmission. The dual-energy X-ray linear array detector may employ a dual-layer scintillator detector architecture or a dual-layer sandwich detector architecture; the upper layer (or front layer) is sensitive to low-energy X-rays, and the lower layer (or rear layer) is sensitive to high-energy X-rays, allowing simultaneous acquisition of high- and low-energy transmission signals corresponding to the same spatial location.

[0026] During the material transport process, when the material to be identified enters the image acquisition area of ​​the dual-energy X-ray linear array detector, high-energy X-rays and low-energy X-rays emitted by the X-ray source penetrate the material and then enter the dual-energy X-ray linear array detector, forming dual-energy data corresponding to the material to be identified. This dual-energy data includes low-energy data and high-energy data.

[0027] Low-energy data and high-energy data are converted according to a preset grayscale mapping relationship to obtain a low-energy image corresponding to the low-energy data and a high-energy image corresponding to the high-energy data. The high-energy image and the low-energy image can be collectively referred to as a dual-energy image.

[0028] Step S220: Based on the results of connected component detection of the low-energy image, determine at least one valid connected component and its corresponding minimum axis-aligned bounding rectangle.

[0029] Compared to high-energy X-rays, low-energy X-rays are more likely to produce photoelectric effects, and the differences in X-ray attenuation between different materials are more significant, resulting in higher image grayscale contrast. Therefore, this disclosure chooses to perform material region detection based on low-energy images to improve the accuracy of connected component detection.

[0030] Specifically, the acquired low-energy images are binarized to generate corresponding binary mask images. The binarization process can employ an adaptive threshold binarization algorithm; for example, Otsu's maximum inter-class variance method can be used for image binarization segmentation.

[0031] After binarization, the low-energy image shows a significant difference in grayscale between the foreground and background material regions, with clear material edge contours, effectively distinguishing the material region from the background region. A connected component detection algorithm is used to segment the binary mask image, classifying adjacent foreground material pixels into the same connected component and spatially separated material regions into different candidate connected components. Candidate connected components are initially identified connected regions, which may include areas corresponding to the actual material, as well as false small regions formed by ray noise and dust imaging artifacts. For example, an 8-neighborhood connected component labeling algorithm can be applied to the binary mask image, traversing all candidate connected regions and eliminating noise regions with an area smaller than a preset threshold. The connected components retained after filtering are recorded as valid connected components, representing the actual material regions identified in the image, thus obtaining at least one valid connected component. The preset threshold can be determined in advance based on at least one factor such as the actual material sorting conditions and image resolution.

[0032] For each valid connected component, a corresponding minimum axis-aligned bounding rectangle is calculated and generated. This rectangle is then used to extract the corresponding dual-energy target image sub-blocks based on the region containing the minimum axis-aligned bounding rectangle. The minimum axis-aligned bounding rectangle is defined as a rectangle with sides parallel to the image coordinate axes, capable of completely enclosing all foreground pixels of the valid connected component, and having the smallest area. In some examples, this can be determined by obtaining the maximum and minimum x-coordinates and y-coordinates of all pixel coordinates within the valid connected component.

[0033] Step S230: Based on the position of the minimum axis-aligned bounding rectangle of each effective connected component, image extraction is performed on the high-energy image and the low-energy image respectively to obtain the dual-energy target image sub-block corresponding to each effective connected component.

[0034] For each valid connected component, obtain the coordinate region of its corresponding minimum axis-aligned bounding rectangle. Using this coordinate region as the cropping reference, locally crop the high-energy image to obtain a high-energy target image sub-block; similarly, locally crop the low-energy image to obtain a low-energy target image sub-block. Combine the high-energy target image sub-block and the low-energy target image sub-block to form the dual-energy target image sub-block corresponding to the valid connected component.

[0035] Step S240: Based on the results of pixel-by-pixel category detection for each dual-energy target image sub-block, determine at least one predicted category and corresponding pixel information included in the corresponding effective connected component.

[0036] In the context of recycled resources, the materials to be identified are complex in form, with highly random distribution of material components, and may contain a mixture of various materials. Furthermore, the orientation of the materials as they enter the conveying mechanism is random, and their landing point is not fixed, making them prone to adhesion and stacking. Directly classifying the entire effective connected region as a single material category can easily lead to misidentification.

[0037] Therefore, this disclosure uses pixels as the basic detection unit to perform class detection on each pixel of each dual-energy target image sub-block, and determines the predicted class corresponding to each pixel in the corresponding dual-energy target image sub-block. In this way, the distribution of predicted classes of different pixels in the corresponding effective connected domain can be determined, and the pixel information corresponding to each predicted class can be obtained.

[0038] Step S250: Based on at least one predicted category and corresponding pixel information included in each effective connected region, and the effective number of predicted categories, determine at least one target material category corresponding to each effective connected region and the effective outer contour information corresponding to the target material category.

[0039] After obtaining the predicted category distribution of each pixel within the effective connected region and the pixel information corresponding to each predicted category, the number of valid predicted categories within that effective connected region is counted. The predicted categories can include either background or valid categories. A valid category can be any one of the multiple candidate material categories required. In actual material identification, when the predicted category is background, it indicates that the corresponding material area is noise or other interference, and does not belong to the desired material. Therefore, to reduce unnecessary computational resource consumption, only the number of valid predicted categories is counted.

[0040] The larger the effective quantity, the more complex the material composition corresponding to the effective connected region. Therefore, this disclosure adopts a differentiated material identification strategy based on at least one predicted category and corresponding pixel information included in each effective connected region, as well as the effective quantity of predicted categories, to determine at least one target material category corresponding to each effective connected region and the effective outer contour information corresponding to the target material category. This not only enables reasonable allocation of computational resources for material identification and ensures material identification performance, but also effectively improves the accuracy and reliability of the identification results.

[0041] It is important to note that the predicted category is an intermediate result obtained based on pixel-level detection, while the target material category is the final recognition result obtained by integrating the predicted information of all pixels within the corresponding effective connected domain and employing a differentiated recognition strategy. The actual material categories corresponding to the two can be the same or different, but they deal with different object levels.

[0042] According to the material identification method provided in this disclosure, by performing connected component detection on the low-energy image in a dual-energy image, it is possible to ensure that at least one effective connected component is reliable data. Image extraction based on the minimum axis-aligned bounding rectangle corresponding to this effective connected component ensures that the background pixel content in the extracted dual-energy target image sub-blocks is extremely low, reducing the interference of background information on material identification. By performing category detection pixel-by-pixel on the dual-energy target image sub-blocks, obtaining pixel-level predicted distribution with pixels as detection units, and then combining the number of predicted categories within the effective connected component with the pixel distribution to comprehensively determine at least one corresponding target material category and the effective outer contour information corresponding to the target material category, the identification accuracy under complex material conditions can be improved, identification computing resources can be rationally allocated, a balance between identification accuracy and algorithm operating efficiency can be achieved, and the sorting performance of subsequent material sorting equipment can be guaranteed.

[0043] In some embodiments, step S240 may include: Step a1: Perform category detection on the fusion features of the dual-energy target image sub-blocks pixel by pixel to determine the predicted category corresponding to each pixel.

[0044] Each pixel within a dual-energy target image sub-block possesses corresponding high-energy and low-energy pixel information, and the imaging features corresponding to the dual-energy pixel information at different locations are distinct. Therefore, to improve the accuracy of subsequent material identification, class detection is performed pixel-by-pixel on the fusion features of the dual-energy target image sub-block. Material identification is then performed based on the local fusion features corresponding to each pixel, determining the predicted category from a set of multiple candidate material categories.

[0045] In some examples, the dual-energy target image sub-block includes a high-energy target image sub-block and a low-energy target image sub-block, and step a1 above may include: Step a11: stitch together the high-energy target image sub-blocks and the low-energy target image sub-blocks to obtain the stitched tensor; Step a12: Extract the multi-scale spatial features of the spliced ​​tensor and reconstruct the fused features that are consistent with the original spatial resolution of the spliced ​​tensor. Step a13: Perform class detection on the fused features pixel by pixel to determine the predicted class corresponding to each pixel.

[0046] Specifically, the high-energy and low-energy target image sub-blocks are image regions synchronously cropped from the same effective connected component. They have the same spatial size, pixel resolution, and corresponding pixel coordinates. By concatenating the size-matched high-energy and low-energy target image sub-blocks along the channel dimension, a concatenated tensor is obtained, enabling preliminary fusion of dual-energy pixel information. This concatenated tensor can simultaneously preserve the contour details of the low-energy image and the material penetration information of the high-energy image.

[0047] A multi-scale feature extraction method is employed to mine spatial feature information of different granularities within the splicing tensor, resulting in multi-scale spatial features of the splicing tensor. These multi-scale spatial features can include, but are not limited to, pixel-level fine textures, edge details, and local structural features, effectively adapting to the irregular shapes, large size variations, and mixed local features of recycled materials. The extracted multi-scale spatial features are upsampled and reconstructed to compensate for the spatial resolution information lost during feature extraction, ultimately restoring a fused feature that perfectly matches the original spatial resolution and pixel size of the splicing tensor. This fused feature integrates shallow imaging information and deep spatial features from dual-energy target image sub-blocks, effectively improving the distinguishability between different materials and reducing noise interference.

[0048] Using pixels as the smallest detection unit, class detection is performed pixel-by-pixel on the fused features. For the local fused features of each pixel, the candidate material category with the highest matching degree is selected from multiple candidate material categories and determined as the predicted category corresponding to that pixel.

[0049] Using this pixel-level detection method for material identification can identify the pixel distribution of different materials within the same effective connected region, avoiding misclassification caused by directly classifying the entire connected region as a single material category. It can provide reliable and accurate pixel information for subsequent determination of the target material category and effective outer contour information.

[0050] In other examples, the predicted category of each pixel can be inferred through a pre-trained target classification model, which can effectively improve the efficiency of pixel category determination, accelerate the overall material identification process, and reduce identification latency. The target classification model takes a sub-block of the dual-energy target image as input and outputs the predicted category of each pixel within that sub-block.

[0051] Specifically, when performing pixel-level detection using a target classification model, dual-energy target image sub-blocks can be directly input into the target classification model. This target classification model can receive high-energy target image sub-blocks through a high-energy channel and low-energy target image sub-blocks through a low-energy channel, and then perform channel concatenation between the two to obtain a concatenated tensor. Finally, based on the inference results of the concatenated tensor, the predicted category corresponding to each pixel in the dual-energy target image sub-block is determined.

[0052] The target classification model's framework can include, but is not limited to, deep learning models, self-attention mechanism (Transformer) models, lightweight networks, neural network models, or traditional machine learning models with inference capabilities. Traditional machine learning models can include, but are not limited to, random forests and support vector machines. In some optional application scenarios, the target classification model can be trained using a deep learning model as the framework. Compared to other model frameworks, deep learning models can autonomously learn the material feature differences between high-energy and low-energy pixels in dual-energy images, eliminating the need for manually designed complex feature operators. They possess stronger feature extraction capabilities and generalization performance for regenerated materials with varied shapes, sizes, and mixed adhesion, effectively improving the accuracy and stability of pixel-level classification results.

[0053] In other embodiments, the input to the target classification model includes not only dual-energy target image sub-blocks, but also combined data of the local original dual-energy data corresponding to the dual-energy target image sub-blocks, in order to ensure the integrity of the input data and thus help improve the output accuracy of the model.

[0054] Step a2: Based on the position of each pixel in the dual-energy target image sub-block, all pixels are divided according to their corresponding prediction categories to obtain at least one prediction category and corresponding pixel information of the effective connected components of the dual-energy target image sub-block.

[0055] After determining the predicted category of each pixel, we can combine the position of each pixel in the dual-energy target image sub-block with the predicted categories of adjacent pixels to determine whether the predicted category of each pixel is isolated or sporadically disordered, and thus determine whether the predicted category of the pixel is misidentified noise.

[0056] Since pixels that are adjacent in location and have the same predicted category are likely to belong to the same material body, while pixels with the same predicted category but not adjacent in spatial location may belong to the same material body, different material bodies that are geographically connected but do not belong to the same type of material, or isolated discrete noise interference. Therefore, to improve the accuracy of material identification and reduce the problems of false identification and missed identification, when clustering all pixels in a sub-block of the dual-energy target image according to the predicted category, pixels that belong to the same predicted category and are adjacent in location are prioritized for grouping and sorting. Finally, at least one predicted category is contained in the current effective connected region, and the pixel information corresponding to each predicted category is obtained. The pixel information includes, but is not limited to, key information such as pixel coordinates and the number of pixels.

[0057] In some examples, the specific process of pixel division can be as follows: Step a21: Based on the position of each pixel in the dual-energy target image sub-block, determine whether the predicted categories corresponding to two adjacent pixels are the same; Step a22: If two adjacent pixels correspond to the same predicted category, then the two adjacent pixels are grouped into the same group, and the pixel information corresponding to the predicted category is updated based on the position information of the two adjacent pixels. Step a23: If the predicted categories of two adjacent pixels are different, then the two adjacent pixels are divided into different groups, and the pixel information corresponding to the predicted category is updated based on the position information of each pixel. Step a24: After all pixels are divided, at least one predicted category and corresponding pixel information are obtained for the effective connected components corresponding to the dual-energy target image sub-blocks. The same predicted category includes at least one pixel group, and each pixel group includes at least one pixel.

[0058] Specifically, the process iterates through all pixels within a sub-block of the dual-energy target image, comparing the predicted category of adjacent pixels one by one based on their spatial location. During the traversal, pixels that are spatially adjacent and have the same predicted category are determined to belong to the same material, are grouped into the same pixel group, and their corresponding pixel coordinates, number of pixels, and other pixel information are simultaneously counted and updated. For pixels that are spatially adjacent but have different predicted categories, they are determined to belong to different materials, are assigned to different pixel groups, and their respective predicted category pixel information is maintained separately.

[0059] By using the above-mentioned pixel-by-pixel neighborhood comparison and grouping method, the pixel prediction results that are randomly distributed within the effective connected domain can be regularized and grouped according to the prediction category and spatial correlation, distinguishing between continuous effective pixels of the same category and scattered discrete pixels, filtering out data interference caused by isolated misidentification noise, and finally obtaining the pixel group and complete pixel information corresponding to each prediction category within the effective connected domain.

[0060] In some embodiments, the predicted category includes background or valid category, and step S250 above may include the following steps: Step b1: Count the number of valid categories based on all predicted categories included in the valid connected components.

[0061] Step b2: If the effective quantity is greater than or equal to 1, then based on the pixel information corresponding to each effective category, determine at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category.

[0062] If the effective quantity is greater than or equal to 1, it indicates that at least one required material exists within the effective connected domain. Then, based on the pixel information corresponding to each effective category, the position of each effective category in the effective connected domain can be determined, thereby determining the target material category corresponding to the effective connected domain. Based on the boundary pixel coordinates of each effective category pixel group, the edge contour of the material region can be extracted, and the effective outer contour information corresponding to the target material category can be obtained.

[0063] In some examples, if the effective quantity is equal to 1, the corresponding effective category is directly used as the target material category corresponding to the effective connected domain, and the effective outer contour information corresponding to the target material category is determined based on the pixel information corresponding to the effective category, thereby simplifying the recognition process and speeding up the recognition speed.

[0064] If the number of valid items is greater than 1, then based on the pixel information corresponding to each valid category, the spatial positional relationship between the local regions of the contour corresponding to each valid category is determined, and based on the spatial positional relationship, at least one target material category corresponding to the valid connected domain and the valid outer contour information corresponding to the target material category are determined.

[0065] That is, when the effective quantity is greater than 1, it indicates that there may be multiple required materials within the same effective connected region, and there are specific spatial relationships between different materials. Therefore, in order to improve the material recognition capability, the spatial position of the corresponding contour local region is determined based on the pixel information corresponding to each effective type. Then, based on the spatial relationship between the respective contour local regions, the corresponding clustering or segmentation method is used to process the data, thereby obtaining at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category.

[0066] In other examples, if the effective quantity is 0, it indicates that the material corresponding to the effective connected component is not the desired material. To save material identification resources, the effective connected component can be directly ignored, or its material category can be determined as interference.

[0067] The following will explain in detail the process of determining the spatial relationship between the local regions of the contours corresponding to each valid category: Step c1: Based on the pixel information corresponding to each valid category, determine whether there are pairs of spatially adjacent contour local regions.

[0068] Based on the pixel information corresponding to the valid category, the pixel distribution of that valid category within the dual-energy target image sub-block can be determined, thereby identifying its corresponding contour local region and spatial location within the valid connected domain. Each contour local region corresponds one-to-one with the pixel information of the valid category.

[0069] Based on the spatial location of the corresponding contour local region for each valid category, detect whether there are adjacent contour local region pairs around it.

[0070] Step c2: If there is at least one pair of spatially adjacent contour local regions, then based on the overlap detection results of each pair of contour local regions, determine the spatial positional relationship between the corresponding contour local regions.

[0071] For each pair of contour local regions, the coordinate information of the two contour local regions within the pair is extracted, and overlap detection is performed to determine whether there is regional overlap between the two contour local regions, obtaining the corresponding overlap detection result. Based on the overlap detection result, the spatial relationship between the pair of contour local regions is determined, thereby effectively identifying the distribution of adhered materials or different material regions on the same material, reducing the problem of missegmentation and misclassification of materials in multi-category mixed scenarios, and improving the reliability of material recognition and contour parsing. The spatial relationship may include, but is not limited to, overlapping, adhered, or independent relationships.

[0072] In some examples, step c2 above may include: Step c21: For each pair of contour local regions, extract their respective pixel masks to obtain the first pixel mask and the second pixel mask. Step c22: Based on the first pixel mask, determine the first minimum axis-aligned bounding rectangle corresponding to the valid category, and based on the second pixel mask, determine the second minimum axis-aligned bounding rectangle corresponding to the valid category; Step c23: Based on the comparison result between the intersection-union ratio of the first minimum axis-aligned bounding rectangle and the second minimum axis-aligned bounding rectangle and the specified threshold, determine the spatial positional relationship between the corresponding contour local region pairs.

[0073] Specifically, to facilitate the differentiation of local regions of each contour, the distribution range of corresponding effective category pixels is marked using pixel masks. Then, based on the corresponding pixel distribution range, the minimum axis-aligned bounding rectangle of the corresponding local contour region is extracted. The first and second pixel masks are generated in the same way, and the first and second minimum axis-aligned bounding rectangles are determined in the same way.

[0074] The intersection-union ratio (IoU) between the first minimum axis-aligned bounding rectangle and the second minimum axis-aligned bounding rectangle is determined to quantify the degree of overlap between them based on the magnitude of the IoU. In some application scenarios, the IoU between the first minimum axis-aligned bounding rectangle BB1 ​​and the second minimum axis-aligned bounding rectangle BB2 can be expressed by the following formula: IoU = Area(BB1∩BB2) / Area(BB1∪BB2), where Area(BB1∩BB2) represents the intersection area between the first minimum axis-aligned bounding rectangle BB1 ​​and the second minimum axis-aligned bounding rectangle BB2, and Area(BB1∪BB2) represents the union area between the first minimum axis-aligned bounding rectangle BB1 ​​and the second minimum axis-aligned bounding rectangle BB2.

[0075] A specified threshold is used to determine the degree of association between two minimum axis-aligned bounding rectangles. For example, if the comparison result indicates that the intersection-union ratio is greater than the specified threshold, it indicates that the overlapping area of ​​the two is large, which may be a wrapping adhesion or a stacked spatial position relationship. Therefore, the spatial position relationship between the local regions of the contours corresponding to the two valid categories can be determined to be an overlapping relationship. If the comparison result indicates that the intersection-union ratio is less than or equal to the specified threshold, it indicates that the overlapping area of ​​the two is small, which can be determined to be a adhesion relationship between the local regions of the contours corresponding to the two valid categories.

[0076] Therefore, by comparing the intersection-union ratio between the first minimum axis-aligned bounding rectangle and the second minimum axis-aligned bounding rectangle with a specified threshold, the spatial relationship between the corresponding local regions of the two can be quickly distinguished, thereby effectively improving the efficiency of spatial relationship determination in multi-category material scenarios.

[0077] Step c3: If there are no spatially adjacent contour local regions, then the spatial positional relationship between the contour local regions corresponding to each valid category is determined to be mutually independent.

[0078] When there are no adjacent contour local regions after detection, it indicates that the contour local regions corresponding to each effective category are not adjacent to each other in space, and there is no overlap or boundary contact between them. Therefore, the spatial positional relationship between each contour local region can be directly determined as an independent relationship.

[0079] In some embodiments, the process of determining at least one target material category corresponding to a valid connected region and the valid outer contour information corresponding to the target material category based on spatial location relationships can be as follows: Step d1: If the spatial relationship is overlapping, then merge the two valid categories into a whole and determine the overall target material category. Based on the merging result of the corresponding contour local region pairs, determine the valid outer contour information corresponding to the target material category.

[0080] When the contour local regions are overlapping, it indicates that the material regions corresponding to the two valid categories overlap significantly, belonging to the same whole material or multiple material regions of the same material. There is no need to physically separate them. Therefore, the two contour local regions within the contour local region pair are merged and combined to form a whole region.

[0081] The target material category corresponding to the whole is determined by combining the merged pixel information, and the closed boundary coordinates of the merged whole area are extracted as effective outer contour information.

[0082] By merging overlapping areas, the mis-splitting of multiple material regions of the same material into multiple independent materials can be avoided, thus ensuring the integrity of the material outline.

[0083] Step d2: If the spatial relationship is an adhesion relationship, then based on the adhesion detection results of the corresponding contour local region pairs, selective image segmentation processing is performed, and the corresponding target material category and the effective outer contour information corresponding to the target material category are determined respectively.

[0084] When the contour local regions are adhered to each other, it indicates that the materials corresponding to the two valid categories only have a small boundary contact and belong to multiple independent materials that are mutually adhered. Based on the boundary contact position obtained by adhesion detection, image segmentation processing is performed on the contour local region pair to separate the two mutually adhered contour local regions into independent regions.

[0085] For example, in image segmentation analysis, two locally connected contour regions within a contour pair are considered as a single overall contour. The discrete curvature of each contour point on this overall contour can be calculated using the k-curvature method, and local minima of curvature are selected as candidate concave points. False concave points are then eliminated based on the criterion that the concave point's direction points towards the region's centroid. Optimal pairing is then performed on the selected valid concave points, and the Euclidean distance between each pair of concave points is calculated.

[0086] If the Euclidean distance is less than a preset distance threshold, the location is determined to be a true physical adhesion neck formed by the compression or surface tension of multiple independent material particles. The range of the distance threshold can be determined based on the particle size distribution of the material. This adhesion neck is the weakest point where stress is concentrated and most easily broken during high-pressure gas injection, possessing the conditions for physical dissociation. Therefore, a dissociation operation is performed: a shortest path dissociation line segment is generated between the paired concave points, i.e., a dividing guide line; along this dissociation line, the merged connected region is divided into two independent sub-regions, and the contour point set of each sub-region is extracted, retaining the corresponding valid category label, and output independently. The boundary coordinates of each independent region are extracted to generate the corresponding outer contour point set, thus constructing the valid outer contour information for the corresponding target material category.

[0087] If the Euclidean distance is greater than or equal to the distance threshold, the indentation is determined to be an irregular geometric feature of a single material, such as bending, stamping dents, or tear edges of broken metal, and not formed by the adhesion of multiple particles. To avoid over-segmentation or misjudgment of the whole material, a separation operation is not performed. Instead, the regions of two or more materials are merged as a whole, the target material category corresponding to this whole is determined, and the merged outer contour point set is output to update the effective outer contour information of the corresponding target material category.

[0088] The target material category corresponding to each segmented region is determined, and the boundary coordinates of each independent segmented region are extracted to obtain their respective effective outer contour information. This segmentation method can distinguish between different materials that are stuck together, solving the problem of difficulty in separating and identifying materials in scenarios where multiple materials are mixed and stuck together.

[0089] Step d3: If the spatial positional relationship is independent, then the corresponding valid categories are all the target material categories, and the valid outer contour information corresponding to the target material category is determined based on the corresponding pixel information.

[0090] In some optional application scenarios, the process of material identification using an X-ray source and a dual-energy X-ray linear array detector may include: During the transmission of the material to be identified via the transmission mechanism, when the material enters the image acquisition area of ​​the dual-energy X-ray linear array detector, high-energy X-rays and low-energy X-rays emitted by the X-ray source penetrate the material and are incident on the dual-energy X-ray linear array detector, forming dual-energy data corresponding to the material. The low-energy data and high-energy data are then converted according to a preset grayscale mapping relationship to obtain a dual-energy image. The dual-energy image includes both a high-energy image and a low-energy image.

[0091] Connectivity detection is performed on low-energy images, and based on the detection results, at least one valid connected component and its corresponding minimum axis-aligned bounding rectangle are determined.

[0092] Based on the position of the minimum axis-aligned bounding rectangle of each effective connected component, image extraction is performed on the high-energy image and the low-energy image respectively to obtain the dual-energy target image sub-block corresponding to each effective connected component.

[0093] For each dual-energy target image sub-block, a pre-trained target classification model is used to perform pixel-by-pixel detection, obtaining at least one predicted category and corresponding pixel information for each effective connected component. Based on the at least one predicted category and corresponding pixel information for each effective connected component, as well as the effective number of predicted categories, at least one target material category and the effective outer contour information corresponding to each target material category are determined.

[0094] Based on the same inventive concept, this disclosure also provides a material sorting method applied to material sorting equipment. For example... Figure 3 As shown, this material sorting method may include the following steps: Step S310: Determine at least one target material category and the corresponding valid outer contour information of the target material category using a material identification method. The material identification method can be any one provided in this disclosure.

[0095] Step S320: Generate sorting control instructions based on each target material category and its corresponding effective outer contour information.

[0096] By combining the conveying speed of the transmission mechanism, the spatial distance between the spray valve and the detector, and the material throwing trajectory, the effective outer contour coordinate information can be converted into the actual physical position and trigger delay, and matched with the preset sorting execution type to generate sorting control instructions containing action time and action position parameters.

[0097] In some examples, the sorting execution type can be a jetting type, and the sorting control command is used to drive the spray valve to spray high-pressure gas at the corresponding time and position to blow the material into the corresponding collection tank; In other examples, the sorting execution type can be a pusher type, and the sorting control command is used to drive the pusher mechanism to perform a telescopic pushing action, extending the pusher at the corresponding time and position to push the target material into the corresponding collection tank.

[0098] Step S330: Sort materials according to sorting control instructions.

[0099] The material sorting method provided in this disclosure can improve the accuracy of material identification, rationally allocate identification computing resources, and achieve a balance between identification accuracy and algorithm running efficiency. Furthermore, it can calculate the sorting timing and spatial position parameters based on the identified target material category and effective outer contour information, accurately generate sorting control commands, accelerate the material identification process, improve the accuracy and efficiency of material sorting, and ensure the stable and reliable sorting performance of the material sorting equipment.

[0100] Based on the same inventive concept, this disclosure also provides a material sorting device. For example... Figure 4 As shown, the material sorting equipment 400 may include: Conveying mechanism 410 is used for conveying materials; Dual-energy X-ray recognition module 420 is used to acquire dual-energy images of materials; The material identification mechanism 430 is used to identify at least one target material category and the effective outer contour information corresponding to the target material category in a dual-energy image by any of the material identification methods provided in this disclosure, and to generate sorting control instructions based on each target material category and the corresponding effective outer contour information. The sorting mechanism 440 is used to receive and sort materials according to sorting control instructions.

[0101] The material sorting equipment provided in this disclosure can improve the material identification accuracy, rationally allocate identification computing resources, achieve a balance between identification accuracy and algorithm running efficiency, accelerate the material identification process, thereby improving the overall material sorting accuracy and sorting efficiency, and ensuring the stable and reliable sorting performance of the equipment.

[0102] This disclosure uses specific terms to describe embodiments of the present disclosure. Terms such as "an embodiment," "one embodiment," and / or "some embodiments" refer to a particular feature, structure, or characteristic associated with at least one embodiment of the present disclosure. Therefore, it should be emphasized and noted that references to "an embodiment," "one embodiment," or "an alternative embodiment" in different locations throughout this specification do not necessarily refer to the same embodiment. Furthermore, certain features, structures, or characteristics in one or more embodiments of the present disclosure can be appropriately combined.

[0103] In the context of this disclosure, unless the context clearly indicates otherwise, the words "a," "an," "an," and / or "the" do not specifically refer to the singular and may also include the plural. Generally speaking, the terms "comprising" and "including" only indicate the inclusion of expressly identified steps and elements, which do not constitute an exclusive list, and the method or apparatus may also include other steps or elements.

[0104] Similarly, it should be noted that, in order to simplify the description of this disclosure and thus aid in the understanding of one or more embodiments, the foregoing description of embodiments of this disclosure may sometimes combine multiple features into a single embodiment, drawing, or description thereof. However, this disclosure method does not imply that the subject matter of this disclosure requires more features than the features claimed. In fact, the embodiments contain fewer features than all the features of the single embodiments disclosed above.

[0105] The basic concepts have been described above. It is obvious that the above disclosure is merely illustrative and does not constitute a limitation of this disclosure. Although not explicitly stated herein, various modifications, improvements, and corrections may be made to this disclosure by those skilled in the art. Such modifications, improvements, and corrections are suggested in this disclosure and therefore remain within the spirit and scope of the embodiments of this disclosure.

Claims

1. A material identification method, characterized in that, The method, applied to material sorting equipment, includes: Acquire a dual-energy image of the material to be identified, wherein the dual-energy image includes a high-energy image and a low-energy image; Based on the results of connected component detection of the low-energy image, at least one valid connected component and its corresponding minimum axis-aligned bounding rectangle are determined. Based on the position of the minimum axis-aligned bounding rectangle of each effective connected component, image extraction is performed on the high-energy image and the low-energy image respectively to obtain the dual-energy target image sub-block corresponding to each effective connected component; Based on the results of pixel-by-pixel category detection for each of the dual-energy target image sub-blocks, at least one predicted category and the corresponding pixel information are determined for the corresponding effective connected component. Based on at least one predicted category and corresponding pixel information included in each effective connected region, and the effective number of predicted categories, at least one target material category corresponding to each effective connected region and the effective outer contour information corresponding to the target material category are determined respectively.

2. The material identification method according to claim 1, characterized in that, The step of determining at least one predicted category and corresponding pixel information for each effective connected component based on the result of pixel-by-pixel category detection for each of the dual-energy target image sub-blocks includes: The fusion features of the dual-energy target image sub-blocks are detected pixel by pixel to determine the predicted category corresponding to each pixel. Based on the position of each pixel in the dual-energy target image sub-block, all pixels are divided according to their corresponding prediction categories to obtain at least one prediction category and corresponding pixel information of the effective connected components corresponding to the dual-energy target image sub-block.

3. The material identification method according to claim 2, characterized in that, The dual-energy target image sub-block includes a high-energy target image sub-block and a low-energy target image sub-block. The pixel-by-pixel fusion feature of the dual-energy target image sub-block is subjected to category detection to determine the predicted category corresponding to each pixel, including: By stitching together the high-energy target image sub-block and the low-energy target image sub-block, a stitching tensor is obtained; Extract the multi-scale spatial features of the spliced ​​tensor and reconstruct the fused features that are consistent with the original spatial resolution of the spliced ​​tensor; The fused features are subjected to category detection pixel by pixel to determine the predicted category corresponding to each pixel.

4. The material identification method according to claim 2 or 3, characterized in that, The predicted category corresponding to the pixel is obtained by reasoning through a pre-trained target classification model, wherein the input of the target classification model is the dual-energy target image sub-block, and the output is the predicted category corresponding to each pixel in the dual-energy target image sub-block.

5. The material identification method according to claim 2, characterized in that, Based on the position of each pixel in the dual-energy target image sub-block, all pixels are divided according to their corresponding predicted categories to obtain at least one predicted category and corresponding pixel information of the effective connected components corresponding to the dual-energy target image sub-block, including: Based on the position of each pixel in the dual-energy target image sub-block, determine whether the predicted categories corresponding to two adjacent pixels are the same; If two adjacent pixels correspond to the same predicted category, then the two adjacent pixels are grouped into the same group, and the pixel information corresponding to the predicted category is updated based on the position information of the two adjacent pixels. If two adjacent pixels correspond to different predicted categories, the two adjacent pixels are divided into different groups, and the pixel information corresponding to the predicted category is updated based on the position information of each pixel. After all the pixels are divided, at least one predicted category and corresponding pixel information are obtained for the effective connected components corresponding to the dual-energy target image sub-block. The same predicted category includes at least one pixel group, and each pixel group includes at least one pixel.

6. The material identification method according to claim 1, characterized in that, The predicted category includes background or valid category. The determination of at least one target material category corresponding to each valid connected component and the corresponding pixel information, based on at least one predicted category included in each valid connected component and the valid number of predicted categories, and the determination of at least one target material category and the valid outer contour information corresponding to each target material category, includes: Based on all the predicted categories included in the effective connected components, count the number of effective categories; If the effective quantity is greater than or equal to 1, then based on the pixel information corresponding to each effective category, at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category are determined.

7. The material identification method according to claim 6, characterized in that, If the effective quantity is greater than or equal to 1, then based on the pixel information corresponding to each effective category, at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category are determined, including: If the effective quantity is equal to 1, then the corresponding effective category is directly used as the target material category corresponding to the effective connected domain, and the effective outer contour information corresponding to the target material category is determined according to the pixel information corresponding to the effective category. If the effective quantity is greater than 1, then based on the pixel information corresponding to each effective category, the spatial positional relationship between the contour local regions corresponding to each effective category is determined, and based on the spatial positional relationship, at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category are determined.

8. The material identification method according to claim 7, characterized in that, The step of determining the spatial positional relationship between the contour local regions corresponding to each of the effective categories based on the pixel information of each effective category includes: Based on the pixel information corresponding to each of the effective categories, it is determined whether there are pairs of spatially adjacent contour local regions, wherein the contour local regions correspond one-to-one with the pixel information of the effective categories. If there are at least one pair of contour local regions that are spatially adjacent, then based on the overlap detection result of each pair of contour local regions, the spatial positional relationship between the corresponding contour local regions is determined respectively. If there are no spatially adjacent pairs of contour local regions, then the spatial positional relationship between the contour local regions corresponding to each valid category is determined to be independent.

9. The material identification method according to claim 8, characterized in that, The step of determining the spatial relationship between corresponding contour local regions based on the overlap detection results of each pair of contour local regions includes: For each pair of contour local regions, the corresponding pixel mask is extracted to obtain the first pixel mask and the second pixel mask; Based on the first pixel mask, a first minimum axis-aligned bounding rectangle corresponding to the valid category is determined, and based on the second pixel mask, a second minimum axis-aligned bounding rectangle corresponding to the valid category is determined. Based on the comparison result between the intersection-union ratio of the first minimum axis-aligned bounding rectangle and the second minimum axis-aligned bounding rectangle and a specified threshold, the spatial positional relationship between the corresponding contour local region pairs is determined.

10. The material identification method according to claim 9, characterized in that, The step of determining the spatial relationship between corresponding contour local region pairs based on the comparison result between the intersection-union ratio (IU) of the first minimum axis-aligned bounding rectangle and the second minimum axis-aligned bounding rectangle and a specified threshold includes: If the comparison result indicates that the intersection-union ratio is greater than the specified threshold, then the spatial positional relationship between the local regions of the contours corresponding to the two valid categories is determined to be an overlapping relationship; If the comparison result indicates that the intersection-union ratio is less than or equal to the specified threshold, then the spatial positional relationship between the local regions of the contours corresponding to the two valid categories is determined to be an adhesion relationship.

11. The material identification method according to claim 10, characterized in that, The step of determining at least one target material category corresponding to the effective connected region and the effective outer contour information corresponding to the target material category based on the spatial position relationship includes: If the spatial relationship is overlapping, the two valid categories are merged into a whole and the overall target material category is determined. Based on the merging result of the corresponding contour local region pairs, the valid outer contour information corresponding to the target material category is determined. If the spatial relationship is an adhesion relationship, then image segmentation processing is selectively performed based on the adhesion detection results of the corresponding contour local region pairs, and the corresponding target material category and the effective outer contour information corresponding to the target material category are determined respectively. If the spatial positional relationships are independent of each other, then the corresponding valid categories are all determined to be the target material category, and the valid outer contour information corresponding to the target material category is determined based on the corresponding pixel information.

12. A material sorting method, characterized in that, The method, applied to material sorting equipment, includes: The material identification method according to any one of claims 1-11 determines at least one target material category and the effective outer contour information corresponding to the target material category; A sorting control instruction is generated based on each of the target material categories and the corresponding effective outer contour information; Materials are sorted according to the sorting control instructions.

13. A material sorting device, characterized in that, include: A conveying mechanism used to transport materials; A dual-energy X-ray recognition module is used to acquire dual-energy images of the material; A material identification mechanism is used to identify at least one target material category and the effective outer contour information corresponding to the target material category in the dual-energy image by the material identification method according to any one of claims 1-11, and to generate a sorting control instruction based on each target material category and the corresponding effective outer contour information; The sorting mechanism is used to receive and sort materials according to the sorting control instructions.