Image Recognition Method, Image Recognition Device, Material Sorting Equipment and Storage Medium

Abstract

The present disclosure relates to the field of image recognition technology, and particularly relates to an image recognition method, an image recognition device, a material sorting device, and a storage medium. The present disclosure provides an image recognition method, including: obtaining a to-be-processed image when materials are transported by a conveyor belt; determining a first bump image in the to-be-processed image and determining first contour information of the materials; determining target bump position information corresponding to the to-be-processed image based on a matching result between the first bump image and a reference bump image, where the reference bump image is a bump mosaic image of the conveyor belt including multiple rows of bumps; updating the first contour information based on a matching result between the first contour information and the target bump position information to obtain target contour information of the materials, so as to recognize the materials through the target contour information. This helps to promote the material recognition process, improve the material sorting efficiency, and ensure the material recognition performance of the material sorting device.

Description

Technical Field

[0001] The present disclosure relates to the field of image recognition technology, and particularly relates to an image recognition method, an image recognition device, a material sorting device, and a computer-readable storage medium. Background Art

[0002] When a material sorting device transports materials, in order to ensure the stability of material transportation and prevent the materials from rolling during transportation, some evenly distributed bumps are arranged on the conveyor belt to control the relative movement state of the materials on the conveyor belt through multiple bumps.

[0003] During the process of material transportation, generally, the currently transported materials can be determined by performing image recognition processing on the collected images. However, the bumps on the conveyor belt may affect the accuracy of material recognition and the recognition result of the materials. For example, during the transportation of plastics, if the size of the plastics is small or the thickness is too thin, and they adhere to the bumps on the conveyor belt during transportation, then during the image recognition processing, the bumps are likely to be misrecognized as part of the plastics, thereby affecting the recognition result of the plastics.

[0004] In the related art, a real-time matching method is adopted to determine the position of the bumps in the image, so as to achieve the purpose of improving the accuracy of material recognition. However, using this method to determine the position of the bumps will affect the efficiency of material recognition. Summary of the Invention

[0005] To overcome the problems existing in the related art, an exemplary embodiment of the present disclosure provides an image recognition method, which is applied to a material sorting device. The material sorting device includes a conveyor belt with multiple rows of bumps, and the conveyor belt is used to transport materials. The method includes: obtaining a to-be-processed image when transporting materials through the conveyor belt; determining a first bump image in the to-be-processed image and determining first contour information of the materials; determining target bump position information corresponding to the to-be-processed image based on the matching result between the first bump image and a reference bump image, where the reference bump image is a bump mosaic image of the conveyor belt including multiple rows of bumps; updating the first contour information based on the matching result between the first contour information and the target bump position information to obtain target contour information of the materials, so as to recognize the materials through the target contour information.

[0006] In some embodiments, if the image to be processed is the first-frame image, the target bump position information corresponding to the image to be processed is determined based on the matching result between the first bump image and the reference bump image, including: determining the first row interval bump image corresponding to the image to be processed in the reference bump image based on the matching result between the first bump image and the reference bump image; determining the first reference position information of each bump image in the first row interval bump image according to the reference position information of each bump image in the reference bump image; and determining the target bump position information corresponding to the image to be processed based on the first reference position information of each bump image in the first row interval bump image.

[0007] In some embodiments, if the image to be processed is a non-first-frame image, the target bump position information corresponding to the image to be processed is determined based on the matching result between the first bump image and the reference bump image, including: obtaining the position of the second row interval bump image corresponding to the previous frame image in the reference bump image; determining the third row interval bump image corresponding to the image to be processed in the reference bump image based on the position of the second row interval bump image in the reference bump image and the image acquisition order of the image to be processed; determining the second reference position information of each bump image in the third row interval bump image according to the reference position information of each bump image in the reference bump image; and determining the target bump position information corresponding to the image to be processed based on the second reference position information of each bump image in the third row interval bump image.

[0008] In some embodiments, the determination process of the reference bump image includes: obtaining a plurality of consecutive first images when the conveyor belt runs in a no-load state; splicing the plurality of first images to obtain an intermediate image, and the number of rows of bump images in the intermediate image is greater than the number of rows of bumps on the conveyor belt; and extracting the reference bump image corresponding to multiple rows of bumps from the intermediate image.

[0009] In some embodiments, extracting the reference bump image corresponding to multiple rows of bumps from the intermediate image includes: respectively matching each row of bump images in the intermediate image with a plurality of bump detection templates to obtain a similarity vector of each row of bump images, where the plurality of bump detection templates are determined based on the background images corresponding to multiple rows of bumps, and each bump detection template corresponds to each column of bump images; determining the reference bump image corresponding to multiple rows of bumps in the intermediate image based on the similarity vectors of each row of bump images in the intermediate image; and extracting the reference bump image.

[0010] In some embodiments, determining a reference bump image corresponding to multiple rows of bumps in the intermediate image based on the similarity vectors of the bump images in each row of the intermediate image includes: determining a target row bump image in the bump images of other rows in the intermediate image that has the greatest similarity to the bump image of the first row; and determining, according to the position of the target row bump image in the intermediate image, a reference bump image corresponding to multiple rows of bumps in the intermediate image, where the reference bump image includes the bump images of the rows from the first row bump image to the row above the target row bump image.

[0011] In some embodiments, extracting a reference bump image corresponding to multiple rows of bumps from the intermediate image includes: obtaining multiple consecutive second images when the conveyor belt runs in a no-load state; sequentially matching each frame of the second images with the intermediate image to determine an intermediate local image in the intermediate image corresponding to the second image, and determining a reference bump image corresponding to multiple rows of bumps based on the height relationship between two adjacent intermediate local images; and extracting the reference bump image from the intermediate image.

[0012] In some embodiments, the heights of each frame of the second images are the same; sequentially matching each frame of the second images with the intermediate image to determine an intermediate local image in the intermediate image corresponding to the second image, and determining a reference bump image corresponding to multiple rows of bumps based on the height relationship between two adjacent intermediate local images includes: matching the current second image with the intermediate image to determine a first intermediate local image in the intermediate image corresponding to the current second image and first position information of the first intermediate local image, where the first position information includes the first height of the first intermediate local image; determining second position information of a second intermediate local image, where the second intermediate local image is the local image in the intermediate image corresponding to the previous second image, and the second position information includes the second height of the second intermediate local image; if the difference between the second height and the first height is equal to the height of a frame image, then performing the step of matching the next second image with the intermediate image; if the difference between the second height and the first height is less than the height of a frame image, then determining a reference bump image corresponding to multiple rows of bumps in the intermediate image based on the acquisition order of the current second image in the multiple frames of second images, the first height, and the second height.

[0013] In some embodiments, the process of determining the reference position information of each bump image in the reference bump image includes: if the current bump image is not a bump image of the first row, then determining the reference position information of the current bump image according to the first row spacing between the row where the current bump image is located and the previous row; if the current bump image is a bump image of the first row, then determining, through the intermediate image, the second row spacing between the bump image of the last row of the reference bump image and the bump image of the next first row, and determining the reference position information of the current bump image according to the second row spacing.

[0014] In some embodiments, based on the matching result between the first contour information and the target bump position information, the first contour information is updated to obtain the target contour information of the material, including: if the target bump position overlaps with the first contour information locally, the contour information corresponding to the overlapping part of the target bump position in the first contour information is removed to obtain the target contour information of the material.

[0015] In a second aspect, the present disclosure also provides an image recognition device applied to a material sorting device. The material sorting device includes a conveyor belt with multiple rows of bumps for conveying materials. The device includes: a first acquisition module for acquiring a to-be-processed image when the materials are conveyed through the conveyor belt; a first determination module for determining the first bump image in the to-be-processed image and determining the first contour information of the material; a second determination module for determining the target bump position information corresponding to the to-be-processed image based on the matching result between the first bump image and a reference bump image, where the reference bump image is a bump mosaic image of the conveyor belt including multiple rows of bumps; and an update module for updating the first contour information based on the matching result between the first contour information and the target bump position information to obtain the target contour information of the material, so as to identify the material through the target contour information.

[0016] In a third aspect, the present disclosure also provides a material sorting device, including: a memory and a processor, which are communicatively connected to each other. The memory stores computer instructions, and the processor executes the computer instructions to perform the image recognition method provided in any of the above aspects.

[0017] In a fourth aspect, the present disclosure also provides a computer-readable storage medium storing the following program, and the program is used to execute the image recognition method provided in any of the above aspects.

[0018] It should be understood that the above general description and the following detailed description are only exemplary and explanatory, and cannot limit the present disclosure.

[0019] The technical solutions provided in the embodiments of the present disclosure may include the following beneficial effects: According to the material recognition method provided in the present disclosure, a reference bump image corresponding to multiple rows of bumps included in the conveyor belt is acquired in advance, so that the distribution of each bump image can be clarified. Then, after obtaining the first bump image of the to-be-processed image, the target bump position information corresponding to the to-be-processed image can be quickly determined through the matching result between the first bump image and the reference bump image, improving the information determination efficiency. Furthermore, combined with the matching result with the first contour information of the material, the interference of the bump image on the first contour information can be quickly eliminated or reduced, improving the recognition quality of the target contour information, thereby contributing to promoting the material recognition process, improving the material sorting efficiency, and ensuring the material recognition performance of the material sorting device. Description of the Drawings

[0020] The present disclosure can be better understood by describing exemplary embodiments thereof in conjunction with the accompanying drawings, in which:

[0021] Figure 1 is a schematic structural diagram of a material sorting device shown in an exemplary embodiment of the present disclosure;

[0022] Figure 2 is a schematic flow diagram of an image recognition method shown in an exemplary embodiment of the present disclosure;

[0023] Figure 3 is a schematic flow diagram of another image recognition method shown in an exemplary embodiment of the present disclosure;

[0024] Figure 4 is a schematic flow diagram of yet another image recognition method shown in an exemplary embodiment of the present disclosure;

[0025] Figure 5 is a schematic flow diagram of a method for determining a reference bump image shown in an exemplary embodiment of the present disclosure;

[0026] Figure 6 is a schematic flow diagram of another method for determining a reference bump image shown in an exemplary embodiment of the present disclosure;

[0027] Figure 7 is a schematic flow diagram of yet another method for determining a reference bump image shown in an exemplary embodiment of the present disclosure;

[0028] Figure 8 is a schematic contour diagram shown in an exemplary embodiment of the present disclosure;

[0029] Figure 9 is a schematic contour diagram of another kind shown in an exemplary embodiment of the present disclosure;

[0030] Figure 10 is a schematic structural diagram of an image recognition device shown in an exemplary embodiment of the present disclosure;

[0031] Figure 11 is a schematic structural diagram of another material sorting device shown in an exemplary embodiment of the present disclosure. Detailed Description

[0032] Specific embodiments of the present disclosure will be described below. It should be noted that in the specific description of these embodiments, for the sake of concise description, the present specification may not describe all features of the actual embodiments in detail. It should be understood that in the actual implementation of any embodiment, just as in the process of any engineering project or design project, in order to achieve the specific goals of the developer and to meet system-related or business-related restrictions, various specific decisions are often made, and these may vary from one embodiment to another. In addition, it should also be understood that although the efforts made in such a development process may be complex and time-consuming, for those of ordinary skill in the art related to the content disclosed in the present disclosure, some design, manufacturing, or production changes based on the technical content disclosed in the present disclosure are only conventional technical means and should not be understood as insufficient content of the present disclosure.

[0033] Unless otherwise defined, technical terms or scientific terms used in the present disclosure should have the ordinary meanings understood by those of ordinary skill in the art within the technical field to which the present disclosure belongs. The "first", "second", and similar terms used in the present disclosure do not denote any order, quantity, or importance, but are only used to distinguish different components. The terms "a" or "an", etc. do not denote a quantity limitation, but mean that there is at least one. The terms "comprising" or "including", etc. are intended to mean that the elements or items appearing before "comprising" or "including" cover the elements or items listed after "comprising" or "including" and their equivalent elements, and do not exclude other elements or items. The terms "connected" or "coupled", etc. are not limited to physical or mechanical connections, nor are they limited to direct or indirect connections.

[0034] When the material sorting device transports materials, in order to ensure the stability of material transportation and prevent the materials from rolling during transportation, some evenly distributed bumps are arranged on the conveyor belt to control the relative movement state of the materials on the conveyor belt through multiple bumps.

[0035] During the process of material transportation, generally, the currently transported materials can be determined by performing image recognition processing on the collected images. However, in the actual transportation process, part of the materials may adhere to the bumps, and then during the image recognition process, the bumps on the conveyor belt may be recognized as part of the materials, causing interference, and thus affecting the recognition accuracy of the materials and the recognition result.

[0036] For example, taking the material sorting device as a plastic sorting device as an example. The structure of the plastic sorting device 100 can be as Figure 1As shown in the figure, it includes a feeding mechanism 110, a conveying mechanism 120, a detection mechanism 130, a sorting device 140, and a collection device 150. The feeding mechanism 110 is used to feed the plastics to be sorted into the conveying mechanism 120. The conveying mechanism 120 at least includes a conveyor belt, which is used to convey the plastics fed by the feeding mechanism 110. The detection mechanism 130 is used to detect the plastics conveyed on the conveying mechanism 120 to identify the currently conveyed plastics. Among them, the detection mechanism 130 can detect the plastics conveyed on the conveying mechanism 120 by using near-infrared plastic identification technology. For example, an image to be processed of the plastics currently conveyed by the conveyor belt is collected through an infrared camera, and then through image recognition processing of the image to be processed, the contour of the plastics currently conveyed by the conveyor belt is determined, and then according to the obtained contour result, the currently conveyed plastics are identified. The sorting device 140 is used to remove impurities in the plastics. The collection device 150 is used to collect the plastics that need to be retained and the impurities in the plastics.

[0037] During the plastic conveying process, the plastics can be conveyed in the form of particles or sheets. If the size of the plastics is small, or the thickness is too thin, and they adhere to the bumps on the conveyor belt during the conveying process, then during the image recognition process, it is easy to misidentify the bumps as part of the plastics, thereby affecting the recognition accuracy of the plastics and the recognition result. The situations of adhesion can include, but are not limited to: plastic particles are vertically arranged above the bumps, or a plastic sheet partially covers the bumps on the conveyor belt, etc.

[0038] In the related art, in order to improve the recognition accuracy of the bumps, a real-time matching method can be adopted. The position information of the bump image corresponding to the image to be processed is determined by template matching, and then through information matching, the interference of the bump image position on the contour information of the material is eliminated or reduced, so as to achieve the purpose of improving the recognition accuracy of the material. However, using this method to determine the bump position will affect the material recognition efficiency, and thus will affect the material sorting performance of the material sorting equipment.

[0039] To solve the above problems, an exemplary embodiment of the present disclosure provides an image recognition method applied to a material sorting device. Among them, the material sorting device includes a conveyor belt with multiple rows of bumps, and the conveyor belt is used to convey materials. As Figure 2 shown, the image recognition method may include the following steps S210 to step S240:

[0040] Step S210, obtain an image to be processed when the material is conveyed by the conveyor belt.

[0041] The image to be processed refers to the image collected during the process of the material sorting device conveying the material through the conveyor belt. By obtaining the image to be processed, the current conveying situation of the material can be clarified, and then the currently conveyed material can be sorted specifically later.

[0042] Among them, the conveyor belt includes multiple rows of bumps to ensure that the relative movement state of the material on the conveyor belt can be controlled by the bumps on the conveyor mechanism during the transportation process. In some examples, the number of columns of each row of bumps is the same and corresponds one by one.

[0043] Step S220: Determine the first bump image in the image to be processed, and determine the first contour information of the material.

[0044] To determine the image content of the image to be processed, image recognition processing is performed on the image to be processed, and then the first bump image and the first contour information of the material are obtained. Among them, the first bump image can be understood as the image corresponding to the bumps on the conveyor belt that are not covered by the material in the acquisition area when the image to be processed is acquired. The first contour information can be understood as the information obtained by initially recognizing the location of the material. The first contour information can be obtained by performing a contour extraction algorithm on the image to be processed. For example, semantic segmentation algorithm, edge detection algorithm, connected region labeling algorithm, etc. The specific contour extraction algorithm used can be determined according to actual needs. In some examples, the algorithm used to determine the first bump image may be the same as or different from the algorithm used to determine the first contour information of the material, which can be specifically determined according to actual needs, and the order of determining the first bump image and the first contour information is not limited in this disclosure.

[0045] Step S230: Based on the matching result between the first bump image and the reference bump image, determine the target bump position information corresponding to the image to be processed.

[0046] Among them, the reference bump image is a bump mosaic image of the conveyor belt including multiple rows of bumps. That is, the reference bump image is a pre-created mosaic image. Through the reference bump image, the position information of the bump images of multiple rows of bumps relative to each other within one operating cycle of the conveyor belt can be clarified.

[0047] The first bump image can be understood as the correctly recognized bump image. However, during the actual image acquisition process, there may be bumps partially covered by the material in the acquisition area, and the corresponding bump images of such bumps may affect the material recognition result. Since the position information of each bump image in the reference bump image is known, in order to determine the position information of the bump images corresponding to each bump in the acquisition area in the image to be processed, the first bump image is matched with the reference bump image, and the distribution of the bump images corresponding to multiple bumps in the acquisition area in the image to be processed is quickly determined according to the obtained matching result, so as to obtain the target bump position information corresponding to the image to be processed.

[0048] Step S240: Based on the matching result between the first contour information and the target bump position information, update the first contour information to obtain the target contour information of the material, so as to identify the material through the target contour information.

[0049] Since the bumps are fixed on the conveyor belt, and during the process of transporting the material, there may be a situation where the material adheres to the bumps. As a result, when extracting the contour of the image to be processed, the bump image may be regarded as part of the material, which may lead to a deviation between the first contour information and the true contour information of the material.

[0050] Therefore, to improve the accuracy of material identification, the first contour information is matched with the target bump position information to determine whether there is a situation where the contour is misidentified due to the presence of the bump image according to the obtained matching result. After obtaining the matching result, updating the first contour information according to the matching result can make the updating process more targeted, effectively eliminate the influence of the bump image on material identification, and then improve the accuracy and reliability of contour extraction, so that the obtained target contour information can better conform to the true contour of the material. Thus, when using the target contour information to identify the material subsequently, the identification efficiency can be improved, and the material identification performance of the material sorting equipment can be guaranteed.

[0051] According to the material identification method provided by the present disclosure, by pre-acquiring the reference bump images corresponding to multiple rows of bumps included in the conveyor belt, the distribution of each bump image can be clarified. Then, after obtaining the first bump image of the image to be processed, through the matching result between the first bump image and the reference bump image, the target bump position information corresponding to the image to be processed can be quickly determined, improving the information determination efficiency. Furthermore, combined with the matching result with the first contour information of the material, the interference of the bump image on the first contour information can be quickly eliminated or reduced, improving the recognition quality of the target contour information, which helps to promote the material identification process, improve the material sorting efficiency, and guarantee the material identification performance of the material sorting equipment.

[0052] In some embodiments, if the image to be processed is the first frame image, then as Figure 3 shown, the above step S230 may include the following steps:

[0053] Step S231: Based on the matching result between the first bump image and the reference bump image, determine the first row interval bump image corresponding to the image to be processed in the reference bump image.

[0054] To improve the determination efficiency of the position information and narrow the screening range, the first bump image is matched with the reference bump image to determine the region image in the reference bump image that is closest to the first bump image, and the matching result is obtained.

[0055] Since the bumps on the conveyor belt are arranged in rows and the arrangement has a certain pattern, by matching the first bump image with the reference bump image, it is possible to determine which row bump images in the reference bump image are closest to the first bump image, and thus use the row bump image closest to the first bump image as the first row interval bump image, so that the distribution of multiple bump images in the image to be processed can be clarified through the first row interval bump image.

[0056] Step S232: Determine the first reference position information of each bump image in the first row interval bump image according to the reference position information of each bump image in the reference bump image.

[0057] The position information between the bump images in the reference bump image is relatively fixed. Therefore, through the reference bump image, the position distribution of the bump images of multiple rows of bumps in the reference bump image can be determined, and thus the reference position information of each bump image in the reference bump image can be determined.

[0058] When the reference position information of each bump image in the reference bump image is predetermined, according to the already determined first row interval bump image, the first reference position information of each bump image in the first row interval bump image can be quickly located and determined.

[0059] Step S233: Determine the target bump position information corresponding to the image to be processed based on the first reference position information of each bump image in the first row interval bump image.

[0060] The distribution of the bump images in the target row bump image can represent the distribution of the bumps on the conveyor belt during image acquisition of the image to be processed. Furthermore, when the first reference position information of each bump image in the first row interval bump image is clarified, the target bump position information corresponding to the image to be processed can be obtained. Among them, the target bump position information includes the position information of the first bump image already recognized in the image to be processed, and the position information of the bump images not recognized in the image to be processed. The position information of the bump images not recognized in the image to be processed can be understood as the image position information corresponding to the bumps that are completely or partially covered by the material in the acquisition area during image acquisition.

[0061] In some other embodiments, if the image to be processed is not the first frame image, then as Figure 4 shown, the above step S230 may include the following steps:

[0062] Step S234: Obtain the position of the second row interval bump image corresponding to the reference bump image in the previous frame image in the reference bump image.

[0063] Since the image to be processed is not the first frame image, and within one operating cycle of the conveyor belt, the distribution positions of the convex point images in each row are relatively fixed. Therefore, to improve the determination efficiency of the reference position information, the positions of the convex point images in the second row interval corresponding to the previous frame image and the image to be processed in the reference convex point image are obtained.

[0064] Step S235: Based on the positions of the convex point images in the second row interval in the reference convex point image and the image acquisition sequence of the image to be processed, determine the convex point images in the third row interval corresponding to the image to be processed in the reference convex point image.

[0065] When the image acquisition sizes of the frame images are the same, within one operating cycle of the conveyor belt, the number of rows of convex point images included in each frame image may vary. Since the previous frame image is adjacent to the image to be processed and the acquisition time is earlier than that of the image to be processed. Therefore, through the convex point images in the second row interval, the number of rows of convex point images corresponding to the image to be processed can be determined. Furthermore, based on the positions of the convex point images in the second row interval in the reference convex point image and the image acquisition sequence of the image to be processed, the row interval convex point images corresponding to the image to be processed are obtained by moving downward along the positions of the convex point images in the second row interval in the reference convex point image, so as to obtain the convex point images in the third row interval corresponding to the image to be processed in the reference convex point image. For example, assuming that the reference convex point image includes 128 rows of pixels, the position of the second row convex point image in the reference convex point image is the 50th row of pixels, and the height of the convex point images in the second row interval is 50 rows of pixels. Then, according to the acquisition sequence of the image to be processed, the position of the image to be processed in the reference convex point image can be determined as the 100th row. Furthermore, based on the position of the image to be processed in the reference convex point image, it can be determined that the 101 - 128th rows of pixels in the reference convex point image are the convex point images in the third row interval corresponding to the image to be processed. And, the 1 - 22nd rows in the next frame of the reference convex point image also belong to the row interval convex point images corresponding to the image to be processed.

[0066] Determining the convex point images in the third row interval in this way can simplify the difficulty of image positioning, improve the determination efficiency of the position information, and facilitate accelerating the processing efficiency of material identification.

[0067] Step S236: According to the reference position information of each convex point image in the reference convex point image, determine the second reference position information of each convex point image in the convex point images in the third row interval.

[0068] When the reference position information of each convex point image in the reference convex point image has been determined in advance, based on the already determined convex point images in the third row interval, the second reference position information of each convex point image in the convex point images in the third row interval can be quickly located and determined.

[0069] Step S237: Based on the second reference position information of each convex point image in the third-row interval convex point image, determine the target convex point position information corresponding to the image to be processed.

[0070] The distribution of each convex point image in the target row convex point image can represent the distribution of the convex points on the conveyor belt during image acquisition of the image to be processed. Therefore, when the second reference position information of each convex point image in the third-row interval convex point image is clear, the target convex point position information corresponding to the image to be processed can be obtained. Among them, the target convex point position information includes the position information of the first convex point image already recognized in the image to be processed, and the position information of the convex point image not recognized in the image to be processed. The position information of the convex point image not recognized in the image to be processed can be understood as the image position information corresponding to the convex points that are completely or partially covered by the material in the acquisition area during image acquisition.

[0071] The following embodiments will illustrate the specific determination process of the reference convex point image. As Figure 5 shown, the determination method of the reference convex point image includes:

[0072] Step S310: Obtain a continuous number of first images when the conveyor belt runs in an unloaded state.

[0073] Since the size of each image frame is limited, each image frame can only capture partial row convex point images when the conveyor belt runs in an unloaded state, and the number of convex point rows during one running cycle of the conveyor belt cannot be determined. Therefore, to ensure the integrity and accuracy of data acquisition, a continuous number of first images when the conveyor belt runs in an unloaded state are obtained to determine the number of convex point rows transmitted by the conveyor belt within one running cycle and the distribution of each convex point.

[0074] Step S320: Stitch multiple first images to obtain an intermediate image.

[0075] Since the conveyor belt rotates in a circle in the material sorter, to determine the image height corresponding to one rotation of the belt, the obtained multiple images are stitched to obtain an intermediate image for a more comprehensive view of the conveyor belt. Among them, the number of convex point images in the intermediate image is greater than the number of convex point rows of the conveyor belt to ensure the integrity and accuracy of the reference convex point image. One rotation of the belt is equivalent to one running cycle.

[0076] Step S330: Extract reference convex point images corresponding to multiple rows of convex points from the intermediate image.

[0077] Since the number of rows of the bump images in the middle image is greater than the number of rows of the bumps during one operating cycle of the conveyor belt, in order to facilitate improving the determination efficiency of the bump positions in the subsequent process, a reference bump image corresponding to multiple rows of bumps is extracted from the middle image, so that the distribution of the bump images in each image to be processed can be quickly determined through the reference bump image in the subsequent process, thereby improving the efficiency of material identification.

[0078] According to the method for determining the reference bump image provided by the present disclosure, the integrity and reliability of obtaining the reference bump image can be guaranteed.

[0079] In some embodiments, as Figure 6 shown, the above step S330 may include the following steps:

[0080] Step S331, respectively match each row of bump images in the middle image with a plurality of bump detection templates to obtain a similarity vector of each row of bump images.

[0081] Among them, the plurality of bump detection templates are determined based on the background images corresponding to multiple rows of bumps, and the bump detection templates correspond to each column of bump images. That is, since in the actual image acquisition process, the image acquisition device may be affected by factors such as the environment, light source, or its own hardware, resulting in uneven pixel brightness distribution in the image to be processed. Therefore, during the operation of the conveyor belt without placing any objects, the background image can be acquired through the image acquisition device for creating the background image. Since the bumps are evenly distributed on the conveyor belt, it can be clearly observed that the bump images are evenly distributed in each row and each column on the background image. To improve the detection accuracy and reduce the influence of factors such as the environment, light source, or its own hardware, a corresponding bump detection template is created for each image column corresponding to the bump images. Then, during the subsequent detection, the detection accuracy of the bump images can be improved. An image column may include at least one pixel column. Each image column corresponds to a bump detection template, and then all the bump images in its corresponding image column can be specifically detected through the bump detection template. In one example, the bump images corresponding to the plurality of bump detection templates may be multiple bump images in the same row in the background image, which can simplify the creation process of the bump detection templates and improve the creation efficiency. In another example, the bump images corresponding to the plurality of bump detection templates may not be multiple bump images in the same row in the background image, but may be the bump images that can best represent the image characteristics of the corresponding image column. Then, during the subsequent template matching, the bump images in the corresponding image column can be quickly determined.

[0082] To determine the distribution of the convex point images in the intermediate image, each row of convex point images in the intermediate image is respectively matched with multiple convex point detection templates to obtain the similarity between each row of convex points and their corresponding convex point detection templates, thereby obtaining the similarity vector of each row of convex point images. That is, for the convex point images in the same row, the similarity matching is respectively performed between the convex point images in each column and their corresponding convex point detection templates, and the similarity values between the convex point images in each column and the corresponding convex point detection templates can be obtained. Furthermore, according to the arrangement order of the convex point images in the convex point images of the same row, the similarity vector that can represent the similarity between the convex point images of this row and multiple convex point detection templates can be obtained.

[0083] Step S332: Based on the similarity vectors of each row of convex point images in the intermediate image, determine the reference convex point image corresponding to multiple rows of convex points in the intermediate image.

[0084] Since within the same operation cycle, the similarity vectors of the row convex point images collected for different rows of convex points are different, and the similarity vectors of the row convex point images collected for the same row of convex points are approximately the same. Therefore, based on the similarity vectors of each row of convex point images in the intermediate image, it can be identified whether there are row convex point images obtained by collecting the same row of images in the intermediate image. If there are different row convex point images with the same similarity vector in the intermediate image, it can be considered that the conveyor belt has completed an operation cycle. Furthermore, multiple rows of convex points corresponding to one rotation of the conveyor belt can be determined therefrom, so as to obtain the reference convex point image that can represent the distribution of multiple rows of convex points when the conveyor belt rotates one circle.

[0085] In some examples, the above step S332 may include the following steps:

[0086] Step a1: Determine the target row convex point image in the other row convex point images in the intermediate image that has the greatest similarity to the first row convex point image.

[0087] Step a2: According to the position of the target row convex point image in the intermediate image, determine the reference convex point image corresponding to multiple rows of convex points in the intermediate image.

[0088] Specifically, to ensure the integrity of the reference image, the similarity vectors of the first row convex point image in the intermediate image are respectively matched with the similarity vectors of the other row convex point images in the intermediate image, the similarities between the first row convex point image and the other row convex point images are respectively determined, and the row convex point image in the other row convex point images that has the greatest similarity to the first row convex point image is used as the target row convex point image. For example, the Euclidean distance between the similarity vector of the first row convex point image and the similarity vectors of the other row convex point images in the intermediate image can be compared to identify the row convex point image closest to the first row convex point image, and this row convex point image is used as the target row convex point image that best matches the first row convex point image.

[0089] Since the intermediate image contains the number of bump rows exceeding one operating cycle, it can be considered that when the similarity between the target row bump image and the first row bump image is determined to be the greatest, the bump row corresponding to the target row bump image and the bump row of the first row bump image are considered to be the same row. Therefore, according to the position of the target row bump image in the intermediate image, the reference bump images corresponding to multiple rows of bumps in the intermediate image are determined. The reference bump images include the bump images from the first row bump image to the row above the target row bump image. For example, if the first row bump image is the bump image in the 0th row of the intermediate image and the target row bump image is the bump image in the 100th row of the intermediate image, the reference bump images include the bump images from the 0th row to the 99th row.

[0090] Step S333, extract the reference bump images.

[0091] In some embodiments, as Figure 7 shown, the above step S330 may further include the following steps:

[0092] Step S334, obtain multiple consecutive second images when the conveyor belt runs in the no-load state. Among them, the multiple second images can be obtained in the same way as the multiple first images, and will not be elaborated here.

[0093] Step S335, sequentially match each frame of the second image with the intermediate image, determine the intermediate local image corresponding to the second image in the intermediate image, and determine the reference bump images corresponding to multiple rows of bumps based on the height relationship between two adjacent intermediate local images.

[0094] Since the second image and the first image are obtained in the same way, by sequentially matching each frame of the second image with the intermediate image stitched before, the intermediate local image that can match the second image can be determined from the intermediate image. During the same operating cycle, the height relationship between two adjacent intermediate local images can be the same. Therefore, the reference bump images corresponding to multiple rows of bumps during the same operating cycle can be determined based on the height relationship between two adjacent intermediate local images.

[0095] In some examples, the height of each frame of the second image is the same, and the above step S335 may include the following steps:

[0096] Step b1, match the current second image with the intermediate image, determine the first intermediate local image corresponding to the current second image in the intermediate image, and the first position information of the first intermediate local image, where the first position information includes the first height of the first intermediate local image;

[0097] Step b2, determine the second position information of the second intermediate local image.

[0098] Step b3, if the difference between the second height and the first height is equal to the height of the frame image, then execute the step of matching the next second image with the intermediate image;

[0099] Step b4, if the difference between the second height and the first height is less than the height of the frame image, then determine the reference bump image corresponding to multiple rows of bumps in the intermediate image based on the acquisition order of the current second image in multiple second images, the first height, and the second height.

[0100] Specifically, perform feature matching between the current second image and the intermediate image to find the corresponding part in the image. After successful matching, the corresponding area of the current second image in the intermediate image can be determined, and this area is called the first intermediate local image. Record the position information of the first intermediate local image, including its first height, that is, the distance from the bottom of the intermediate image to the bottom of the first intermediate local image.

[0101] To determine the height relationship between two adjacent intermediate local images, the second position information of the second intermediate local image is determined. Among them, the second intermediate local image is the local image in the intermediate image corresponding to the previous second image, and the second position information includes the second height of the second intermediate local image.

[0102] If the difference between the second height and the first height is equal to the height of the frame image, it indicates that the current second image and the previous second image belong to the images within the same operation cycle. Therefore, the second image of the next frame can be processed continuously.

[0103] If the difference between the second height and the first height is less than the height of the frame image, it indicates that the current second image includes two operation cycles. Therefore, determine the reference bump image corresponding to multiple rows of bumps in the intermediate image based on the acquisition order of the current second image in multiple second images, the first height, and the second height. Specifically, according to the first height and the second height, the target image height within the same operation cycle as the previous second image in the intermediate image can be determined. Then, based on the acquisition order of the current second image in multiple second images, the image height of the second image, and the target image height, start from the top of the intermediate image and obtain downward to get the reference bump image. For example, target image height = first height + image height of the second image - second height.

[0104] Step S336, extract the reference bump image from the intermediate image.

[0105] The following embodiments will illustrate the specific determination process of the reference position information of each bump image in the reference bump image:

[0106] Step c1, determine the reference position information of the current bump image in the reference bump image;

[0107] If the current convex dot image is not the convex dot image of the first row, determine the reference position information of the current convex dot image according to the first row spacing between the row where the current convex dot image is located and the previous row;

[0108] Step c2, if the current convex dot image is the convex dot image of the first row, determine the second row spacing between the last row convex dot image of the reference convex dot image and the next first row convex dot image through the intermediate image, and determine the reference position information of the current convex dot image according to the second row spacing.

[0109] Specifically, in the case of clarifying the reference position information, in order to quickly locate the position information of each convex dot image in the image to be processed during the subsequent material identification process, determine the position information of each convex dot image in the reference image respectively.

[0110] For example, you can start from the second row convex dot image, determine the first row spacing between the row where the current convex dot image is located and the previous row respectively, and then determine the reference position information of the current convex dot image according to the first row spacing.

[0111] Since the conveyor belt is always rotating, the last row convex dot image of the previous operation cycle is the row above the first row convex dot image of the current operation cycle. Therefore, in order to accurately calculate the subsequent convex dot positions, determine the second row spacing between the last row convex dot image of the reference convex dot image and the next first row convex dot image through the intermediate image, and determine the reference position information of the current convex dot image according to the second row spacing to ensure the coherence and stability of the recognition process.

[0112] In some examples, if the reference convex dot image is determined based on the similarity matching method of convex dot images, the storage expression of the position information of each convex dot image in the reference convex dot image can be: Start = 0; End = min(blocks_loc[p][j]) + min(d[1][j]) - min(blocks_loc[1][j]); where d[1][j] represents the distance stored in the jth column of the first row; blocks_loc[p][j] represents the upper left corner coordinates of the convex dot stored in the jth column of the pth row.

[0113] In other examples, if the reference convex dot image is determined based on the image feature matching method, the storage expression of the reference convex dot image can be: d[p][j] = image height - blocks_loc[p][j] + blocks_loc[1][j].

[0114] In some embodiments, the above step S240 may include: if the target bump position overlaps with the first contour information locally, then the contour information in the first contour information that overlaps with the target bump position correspondingly is removed to obtain the target contour information of the material. Through the matching results between the first contour information and multiple target bump positions, it can be determined whether there is a situation where the bump image is misidentified as part of the material during the process of identifying the material contour. If it is determined according to the matching results that there is a target bump position that overlaps with the first contour information locally, then the contour information in the first contour information that overlaps with the target bump position correspondingly is removed to reduce image noise, so as to eliminate the influence of the bump image on material identification, and further obtain the target contour information that can better fit the actual contour of the material, improving the accuracy of the target contour information. For example, the first contour information of the obtained material is as Figure 8 shown. When it is determined that there is a target bump position that overlaps with the first contour information locally, then the contour information in the first contour information that overlaps with the target bump position correspondingly is removed, and the obtained target contour information can be as Figure 9 shown.

[0115] In some other embodiments, if the number of the first contour information is multiple and there is a first contour information that overlaps with the target bump position locally, then the first contour information is removed to avoid the influence of the existence of the first contour information on the material identification by other first contour information, simplifying the difficulty of material identification, and further effectively improving the accuracy of material identification and ensuring the quality of material identification.

[0116] In some optional implementation scenarios, the process of identifying the material (plastic) currently transported by the material sorting device through the collected image to be processed may be as follows:

[0117] A reference bump image is created in advance, and the distribution information of each row of bump images is determined respectively according to the reference bump image.

[0118] During the actual image recognition process, first let the belt idle for a short period of time, and the idle time only needs to be greater than the acquisition time of one frame of image, and then place the plastic.

[0119] The first bump image in the first frame of image to be processed is determined, and the first contour information of the material is determined. Among them, the first contour information of the plastic in the image to be processed can be obtained through infrared image detection.

[0120] The first bump image is matched with the reference bump image to determine the first row interval bump image in the reference bump image corresponding to the first frame of image to be processed. Furthermore, according to the reference position information of each bump image in the reference bump image, the first reference position information of each bump image in the first row interval bump image is determined, so that the target bump position information corresponding to the first frame of image to be processed can be determined.

[0121] After obtaining the target convex point position information, the first contour information is matched with the target convex point position information. If there is a target position information that partially overlaps with the first contour information as shown in Figure 8 , the contour information corresponding to the target position information in the first contour information is removed, and the target contour information of the plastic as shown in Figure 9 is obtained.

[0122] According to the first row interval convex point image, the second row interval convex point image corresponding to the next image to be processed in the reference convex point image can be determined. Furthermore, the target convex point position information corresponding to the next image to be processed can be quickly determined without re-matching for the second time, thereby effectively improving the positioning efficiency of the convex point image, facilitating the acceleration of the subsequent process of identifying the plastic, improving the material identification efficiency, and ensuring the performance of the material sorting equipment.

[0123] Based on the same inventive concept, the present disclosure also provides an image recognition device applied to a material sorting device. The material sorting device includes a conveyor belt with multiple rows of convex points. As shown in Figure 10 , the image recognition device 400 may include: a first acquisition module 410, a first determination module 420, a second determination module 430, and an update module 440.

[0124] The first acquisition module 410 is configured to acquire an image to be processed when transporting materials through the conveyor belt;

[0125] The first determination module 420 is configured to determine the first convex point image in the image to be processed and determine the first contour information of the material;

[0126] The second determination module 430 is configured to determine the target convex point position information corresponding to the image to be processed based on the matching result between the first convex point image and the reference convex point image, where the reference convex point image is a convex point splicing image of the conveyor belt including multiple rows of convex points;

[0127] The update module 440 is configured to update the first contour information based on the matching result between the first contour information and the target convex point position information to obtain the target contour information of the material for identifying the material through the target contour information.

[0128] In some embodiments, if the image to be processed is the first-frame image, the second determination module 430 may include: a first determination unit configured to determine a first row interval bump image corresponding to the image to be processed in the reference bump image based on a matching result between the first bump image and the reference bump image; a second determination unit configured to determine first reference position information of each bump image in the first row interval bump image according to reference position information of each bump image in the reference bump image; and a third determination unit configured to determine target bump position information corresponding to the image to be processed based on the first reference position information of each bump image in the first row interval bump image.

[0129] In some embodiments, if the image to be processed is a non-first-frame image, the second determination module 430 may include: a position acquisition unit configured to acquire a position of a second row interval bump image corresponding to the reference bump image in the reference bump image for the previous frame image; a fourth determination unit configured to determine a third row interval bump image corresponding to the image to be processed in the reference bump image based on the position of the second row interval bump image in the reference bump image and an image acquisition sequence of the image to be processed; a fifth determination unit configured to determine second reference position information of each bump image in the third row interval bump image according to reference position information of each bump image in the reference bump image; and a sixth determination unit configured to determine target bump position information corresponding to the image to be processed based on the second reference position information of each bump image in the third row interval bump image.

[0130] In some embodiments, the apparatus for determining a reference bump image includes: a second acquisition module configured to acquire a plurality of consecutive first images when the conveyor belt runs in a no-load state; a splicing module configured to splice the plurality of first images to obtain an intermediate image, where the number of rows of bump images in the intermediate image is greater than the number of rows of bumps on the conveyor belt; and an extraction module configured to extract a reference bump image corresponding to multiple rows of bumps from the intermediate image.

[0131] In some embodiments, the extraction module includes: a first matching unit configured to respectively match each row of bump images in the intermediate image with a plurality of bump detection templates to obtain a similarity vector of each row of bump images, where the plurality of bump detection templates are determined based on a background image corresponding to multiple rows of bumps, and each bump detection template corresponds to each column of bump images; a first screening unit configured to determine a reference bump image corresponding to multiple rows of bumps in the intermediate image based on the similarity vectors of each row of bump images in the intermediate image; and a first extraction unit configured to extract the reference bump image.

[0132] In some embodiments, the first screening unit includes: a first execution unit configured to determine a target row bump image in the other row bump images of the intermediate image that has the highest similarity to the first row bump image; a second execution unit configured to determine a reference bump image corresponding to multiple rows of bumps in the intermediate image according to the position of the target row bump image in the intermediate image, where the reference bump image includes the bump images of the rows from the first row bump image to the row above the target row bump image.

[0133] In some embodiments, the extraction module includes: an image acquisition unit configured to acquire multiple consecutive second images when the conveyor belt is running in a no-load state; a second screening unit configured to sequentially match each second image with the intermediate image to determine an intermediate local image corresponding to the second image in the intermediate image, and determine a reference bump image corresponding to multiple rows of bumps based on the height relationship between two adjacent intermediate local images; a second extraction unit configured to extract the reference bump image from the intermediate image.

[0134] In some embodiments, the heights of each second image are the same; the second screening unit includes: a third execution unit configured to match the current second image with the intermediate image to determine a first intermediate local image corresponding to the current second image in the intermediate image and first position information of the first intermediate local image, where the first position information includes the first height of the first intermediate local image; a fourth execution unit configured to determine second position information of a second intermediate local image, where the second intermediate local image is a local image corresponding to the previous second image in the intermediate image, and the second position information includes the second height of the second intermediate local image; a fifth execution unit configured to, if the difference between the second height and the first height is equal to the height of a frame image, execute the step of matching the next second image with the intermediate image; a sixth execution unit configured to, if the difference between the second height and the first height is less than the height of a frame image, determine a reference bump image corresponding to multiple rows of bumps in the intermediate image based on the acquisition order of the current second image in the multiple second images, the first height, and the second height.

[0135] In some embodiments, the apparatus for determining the reference position information of each bump image in the reference bump image includes: a first information determination module configured to, if the current bump image is not the first row bump image, determine the reference position information of the current bump image according to the first row spacing between the row where the current bump image is located and the previous row; a second information determination module configured to, if the current bump image is the first row bump image, determine the second row spacing between the last row bump image of the reference bump image and the next first row bump image through the intermediate image, and determine the reference position information of the current bump image according to the second row spacing.

[0136] In some embodiments, the update module 440 includes: an information update unit configured to, if the target bump position overlaps with the first contour information locally, remove the contour information corresponding to and overlapping with the target bump position from the first contour information to obtain the target contour information of the material.

[0137] Regarding the image recognition device in the above embodiments, the specific manners in which each module performs operations have been described in detail in the embodiments related to the method, and will not be elaborated herein.

[0138] Based on the same inventive concept, as Figure 11 shown, an embodiment of the present disclosure provides a material sorting device 500. Among them, the material sorting device 500 includes a memory 510, a processor 520, and an input / output (I / O) interface 530. Among them, the memory 510 is configured to store instructions. The processor 520 is configured to call the instructions stored in the memory 510 to execute the image recognition method of the embodiments of the present disclosure. Among them, the processor 520 is respectively connected to the memory 510 and the I / O interface 530, and can be connected, for example, through a bus system and / or other forms of connection mechanisms (not shown). The memory 510 can be used to store programs and data, including the program of the image segmentation method involved in the embodiments of the present disclosure. The processor 520 executes various functional applications and data processing of the material sorting device 500 by running the program stored in the memory 510.

[0139] In the embodiments of the present disclosure, the processor 520 can be implemented in at least one hardware form of a digital signal processor (DSP), a field-programmable gate array (FPGA), and a programmable logic array (PLA). The processor 520 can be a central processing unit (CPU) or a combination of one or more of other forms of processing units having data processing capabilities and / or instruction execution capabilities.

[0140] In the embodiments of the present disclosure, the memory 510 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. Volatile memory may include, for example, random access memory (RAM) and / or cache memory. Non-volatile memory may include, for example, read-only memory (ROM), flash memory, hard disk drive (HDD), or solid-state drive (SSD).

[0141] In the embodiments of the present disclosure, the I / O interface 530 can be used to receive input instructions (such as numerical or character information, and key signal inputs related to user settings and function controls of the material sorting device 500), and can also output various information to the outside (such as images or sounds). In the embodiments of the present disclosure, the I / O interface 530 may include one or more of a physical keyboard, function keys (such as volume control keys, power on / off keys, etc.), a mouse, a joystick, a trackball, a microphone, a speaker, and a touch panel.

[0142] Based on the same inventive concept, the present disclosure also provides a computer-readable storage medium storing the following program, and the program is used to execute the image recognition method in any of the foregoing embodiments.

[0143] The present disclosure uses specific terms to describe the embodiments of the present disclosure. For example, "one embodiment", "an embodiment", and / or "some embodiments" mean a certain feature, structure, or characteristic related to at least one embodiment of the present disclosure. Therefore, it should be emphasized and noted that the "one embodiment" or "an embodiment" or "an alternative embodiment" mentioned twice or more at different positions in this specification does not necessarily refer to the same embodiment. In addition, certain features, structures, or characteristics in one or more embodiments of the present disclosure can be combined appropriately.

[0144] In the context of the present disclosure, unless the context clearly indicates an exception, the words "a", "an", "one", and / or "the" do not specifically refer to the singular, but may also include the plural. Generally speaking, the terms "include" and "comprise" only indicate the inclusion of the steps and elements that have been clearly identified, and these steps and elements do not constitute an exclusive list, and the method or device may also include other steps or elements.

[0145] Similarly, it should be noted that, in order to simplify the description of the present disclosure and thus assist in the understanding of one or more embodiments of the application, in the foregoing description of the embodiments of the present disclosure, multiple features are sometimes grouped into one embodiment, drawing, or description thereof. However, this disclosure method does not mean that the features required by the subject matter of the present disclosure are more than the features required to be protected. In fact, the features of the embodiment are fewer than all the features of the single embodiment disclosed above.

[0146] The basic concepts have been described above. Obviously, for those skilled in the art, the above disclosure is only an example and does not constitute a limitation to the present disclosure. Although not explicitly stated herein, those skilled in the art may make various modifications, improvements, and corrections to the present disclosure. Such modifications, improvements, and corrections are proposed in the present disclosure, so such modifications, improvements, and corrections still fall within the spirit and scope of the embodiments of the present disclosure.

Claims

1. An image recognition method, characterized in that: Applied to a material sorting device, the material sorting device comprises a conveyor belt with multiple rows of convex points, the conveyor belt is used to convey materials, and the method comprises: Acquire an image to be processed when the material is transmitted through the conveyor belt; Determine a first salient point image in the image to be processed, and determine first contour information of the material; Determine target salient point position information corresponding to the image to be processed based on a matching result between the first salient point image and a reference salient point image, wherein the reference salient point image is a salient point splicing image of the conveyor belt containing the multiple rows of salient points; Based on the matching result between the first contour information and the target salient point position information, the first contour information is updated to obtain the target contour information of the material, so as to identify the material through the target contour information.

2. The image recognition method according to claim 1, characterized in that: If the image to be processed is a first frame image, determining target salient point position information corresponding to the image to be processed based on a matching result between the first salient point image and a reference salient point image includes: Based on the matching result between the first salient point image and the reference salient point image, determining a first row interval salient point image in the reference salient point image corresponding to the image to be processed; Determining first reference position information of each salient point image in the first row interval salient point image according to reference position information of each salient point image in the reference salient point image; Based on the first reference position information of each salient point image in the first row interval salient point images, target salient point position information corresponding to the image to be processed is determined.

3. The image recognition method according to claim 1, characterized in that: If the image to be processed is not the first frame image, then determining the target salient point position information corresponding to the image to be processed based on the matching result between the first salient point image and the reference salient point image includes: Obtaining the position of the second row interval salient point image corresponding to the previous frame image and the reference salient point image in the reference salient point image; Determine, based on the position of the second row interval salient point image in the reference salient point image and the image acquisition order of the image to be processed, a third row interval salient point image in the reference salient point image corresponding to the image to be processed; Determining second reference position information of each salient point image in the third row interval salient point image according to the reference position information of each salient point image in the reference salient point image; Based on the second reference position information of each salient point image in the third row interval salient point images, target salient point position information corresponding to the image to be processed is determined.

4. The image recognition method according to claim 2 or 3, characterized in that: The process of determining the reference salient point image includes: Acquire a plurality of continuous first images when the conveyor belt is running in a no-load state; Splicing the multiple frames of first images to obtain an intermediate image, wherein the number of convex point image rows in the intermediate image is greater than the number of convex point rows of the transmission belt; A reference salient point image corresponding to the plurality of rows of salient points is extracted from the intermediate image.

5. The image recognition method according to claim 4, characterized in that: The step of extracting a reference salient point image corresponding to the plurality of rows of salient points from the intermediate image comprises: Matching each row of salient point images in the intermediate image with a plurality of salient point detection templates respectively to obtain a similarity vector of each row of salient point images, wherein the plurality of salient point detection templates are determined based on background images corresponding to the plurality of rows of salient points, and the salient point detection templates correspond to each column of salient point images; Determining reference salient point images corresponding to the multiple rows of salient points in the intermediate image based on the similarity vector of each row of salient point images in the intermediate image; The reference salient image is extracted.

6. The image recognition method according to claim 5, characterized in that: The determining, based on the similarity vector of each row of salient point images in the intermediate image, reference salient point images corresponding to the multiple rows of salient points in the intermediate image comprises: Determine a target row of salient point images having the greatest similarity to the first row of salient point images among other rows of salient point images in the intermediate image; According to the position of the target row of salient point images in the intermediate image, a reference salient point image corresponding to the multiple rows of salient points in the intermediate image is determined, and the reference salient point image includes a row of salient point images from the first row of salient point images to the previous row of salient point images of the target row of salient point images.

7. The image recognition method according to claim 4, characterized in that: The step of extracting a reference salient point image corresponding to the plurality of rows of salient points from the intermediate image comprises: Acquire a plurality of continuous second images when the conveyor belt is running in a no-load state; Matching each frame of the second image with the intermediate image in sequence, determining an intermediate partial image in the intermediate image corresponding to the second image, and determining the reference salient point image corresponding to the multiple rows of salient points based on a height relationship between two adjacent intermediate partial images; The reference salient image is extracted from the intermediate image.

8. The image recognition method according to claim 7, characterized in that: The height of the second image in each frame is the same; The step of sequentially matching each frame of the second image with the intermediate image, determining an intermediate partial image in the intermediate image corresponding to the second image, and determining the reference salient point image corresponding to the multiple rows of salient points based on a height relationship between two adjacent intermediate partial images, comprises: Matching the current second image with the intermediate image, determining a first intermediate partial image in the intermediate image corresponding to the current second image, and first position information of the first intermediate partial image, wherein the first position information includes a first height of the first intermediate partial image; Determine second position information of a second intermediate partial image, where the second intermediate partial image is a partial image in the intermediate image corresponding to a previous second image, and the second position information includes a second height of the second intermediate partial image; If the difference between the second height and the first height is equal to the frame image height, performing a step of matching the next second image with the intermediate image; If the difference between the second height and the first height is less than the frame image height, the reference salient point image corresponding to the multiple rows of salient points in the intermediate image is determined based on the acquisition order of the current second image in multiple frames of the second images, the first height and the second height.

9. The image recognition method according to claim 4, characterized in that: The process of determining the reference position information of each salient point image in the reference salient point image includes: If the current salient point image is not a first row salient point image, determining reference position information of the current salient point image according to a first row spacing between the row where the current salient point image is located and a previous row; If the current salient point image is a first row of salient point images, a second row spacing between a last row of salient point images of the reference salient point image and a next first row of salient point images is determined through the intermediate image, and reference position information of the current salient point image is determined according to the second row spacing.

10. The image recognition method according to claim 1, characterized in that: The updating of the first contour information based on the matching result between the first contour information and the target convex point position information to obtain the target contour information of the material includes: If the target convex point position partially overlaps with the first contour information, the contour information corresponding to and overlapping with the target convex point position in the first contour information is removed to obtain the target contour information of the material.

11. An image recognition device, characterized in that: Applied to material sorting equipment, the material sorting equipment includes a conveyor belt with multiple rows of convex points, the conveyor belt is used to transport materials, and the device includes: A first acquisition module, used for acquiring an image to be processed when the material is transmitted through the conveyor belt; A first determination module, used to determine a first salient point image in the image to be processed, and to determine first contour information of the material; A second determination module is used to determine the target salient point position information corresponding to the image to be processed based on the matching result between the first salient point image and a reference salient point image, wherein the reference salient point image is a salient point splicing image of the conveyor belt containing the multiple rows of salient points; An updating module is used to update the first contour information based on the matching result between the first contour information and the target salient point position information, so as to obtain the target contour information of the material and identify the material through the target contour information.

12. A material sorting device, characterized in that: include: A memory and a processor, wherein the memory and the processor are communicatively connected to each other, the memory stores computer instructions, and the processor executes the image recognition method according to any one of claims 1 to 10 by executing the computer instructions.

13. A computer-readable storage medium storing the following program, wherein the program is used to execute the image recognition method according to any one of claims 1 to 10.

Images