A machine vision-based cone yarn color identification and sorting equipment

Abstract

The application relates to the technical field of sorting equipment, and discloses a bobbin color identification and sorting equipment based on machine vision, which comprises a machine arm, the movable end of the machine arm is provided with a gripper and a cylinder, the gripper is circumferentially distributed on the outer side of the cylinder, a connecting plate is connected with the output end of the cylinder in the middle of one side and connected with the gripper at the edge, the connecting plate is used for driving the gripper to move along with the operation of the cylinder, a camera is installed in the middle of the other side of the connecting plate and used for collecting the color of the bobbin, and a support is installed on the outer wall of the connecting plate, the inside of the support is provided with a light supplementing lamp used for providing light for the camera. The Mask-R-CNN model is adopted to separate the target, the channel attention mechanism is used to weaken the visual interference on the surface of the bobbin, the CIE-Lab color space which is not sensitive to light is combined, and therefore, various colors can be accurately and stably identified under complex light and surface conditions, and similar colors with extremely small color difference can be effectively distinguished.

Description

Technical Field

[0001] This invention relates to the field of sorting equipment technology, specifically to a machine vision-based yarn color recognition and sorting equipment. Background Technology

[0002] Yarn cones, in the textile production process, refer to a form of yarn packaging formed by winding yarn onto bobbins, facilitating storage, transportation, and subsequent processing (such as weaving and knitting). In modern textile industry, color sorting of yarn cones is a crucial quality control step to produce fabrics with uniform color that meet design requirements. Especially in the production of multi-colored fabrics or scenarios requiring the splicing of different colored yarns, it is essential to ensure that each batch of yarn cones has an accurate and consistent color code to avoid quality defects such as color differences and color variations in the final product.

[0003] Traditional manual sorting methods rely on visual judgment, which has problems such as low efficiency, high labor intensity, and strong subjectivity in color difference judgment. Especially in multi-color and small-batch customized production scenarios, it is easy to cause color mixing and missorting, resulting in poor dyeing consistency, reduced product quality and waste of raw materials. Furthermore, accurate identification of similar color numbers remains a challenge. Summary of the Invention

[0004] To address the shortcomings of existing technologies, this invention provides a machine vision-based yarn color recognition and sorting equipment, which solves the problems of low efficiency, high labor intensity, strong subjectivity in color difference judgment, and inability to accurately distinguish similar color codes in traditional manual yarn sorting methods.

[0005] To achieve the above objectives, the present invention provides the following technical solution: a machine vision-based yarn color recognition and sorting equipment, comprising:

[0006] The robotic arm integrates an end effector at its movable end for gripping and identifying yarn packages. This end effector includes a cylinder and multiple grippers arranged circumferentially around the cylinder.

[0007] The middle part of one side of the connecting plate is connected to the output end of the cylinder, while its edge is connected to each gripper.

[0008] The camera is mounted on the other side of the connecting plate in the middle, with its lens facing outward, and is used to capture images of the yarn package within the gripper's grasping range; multiple brackets are mounted on the outer wall of the connecting plate, distributed alternately with the gripper, and supplementary lights are installed inside the brackets to provide illumination for the camera's acquisition area;

[0009] As a preferred layout, the bracket is configured as multiple units and symmetrically distributed in a ring around the outside of the camera. The multiple built-in fill lights together form a ring light source, and the ring light source is coaxially arranged around the lens of the camera. This structure can provide uniform and shadowless lighting conditions for the cylindrical surface of the yarn package, minimizing the interference of uneven lighting and high light reflection on color recognition.

[0010] It also includes a controller, which establishes communication or electrical connections with the robotic arm, cylinder, camera, and supplementary lighting; the controller contains the computational logic to achieve the purpose of this invention, and can be specifically divided into the following functional modules:

[0011] The image data acquisition module is used to receive digital images containing one or more yarn packages captured in real time by the camera;

[0012] The signal processing and color recognition module is used to perform algorithmic processing on the received yarn package image to determine the final color number of each yarn package;

[0013] The sorting control module is used to generate and send motion control commands to the robotic arm and cylinder based on the determined final color number, so as to drive them to perform the corresponding sorting actions.

[0014] The signal processing and color recognition module is key to achieving high-precision color recognition in this invention, and its internal algorithm flow specifically includes the following units:

[0015] The instance segmentation unit uses the Mask-R-CNN (a convolutional neural network based on masked regions) model. After receiving the original image, it outputs the pixel-level mask region (mask) of each yarn bobbin target in the image, thereby effectively separating the yarn bobbins.

[0016] The feature enhancement unit, located within the Mask-R-CNN model, is integrated after the backbone network. This unit employs a spatial-channel attention mechanism. After the backbone network extracts the initial feature map of the image, this feature map is fed into this unit. Through the spatial attention mechanism, the model focuses its computational resources more on the spatial location features corresponding to the masked regions; simultaneously, through the channel attention mechanism, it enhances the weights of the feature channels most sensitive to color information and suppresses noise irrelevant to the color recognition task, such as feathers on the yarn surface or textured features.

[0017] The color feature extraction unit receives image data containing only the yarn mask area, processed by the feature enhancement unit. To eliminate the severe impact of changes in ambient lighting on color recognition, this unit first converts the image data from the light-sensitive RGB color space to the CIE-Lab color space, which is less sensitive to lighting changes and more in line with human visual perception.

[0018] The color matching unit performs the final identification and matching of the aforementioned color features. The specific steps are as follows:

[0019] First, to obtain a single quantized value representing the color of the entire yarn package, the average color feature of all pixels within the masked area in the CIE-Lab color space is calculated. If the color value of any pixel within the masked area is: Let N be the total number of pixels in the region, then the quantized color value of this yarn package... The calculation is as follows:

[0020]

[0021] In the formula, C i This represents the color value vector of the i-th pixel within the mask region; This represents the luminance component of the i-th pixel in the CIE-Lab color space; This represents the green-red color components of the i-th pixel in the CIE-Lab color space; The i-th pixel represents the blue-yellow color component in the CIE-Lab color space; i represents the index of the pixel, which is an integer from N; N represents the total number of pixels contained in the specific yarn mask area output by the instance segmentation unit. This represents the quantized color value vector calculated from the final values ​​of the entire yarn package. This represents the average brightness component of the entire yarn package; This represents the average green-red color component of the entire yarn package; This represents the average blue-yellow color component of the entire yarn package.

[0022] Then, the obtained quantized color values The color difference distance is calculated between the yarn and each standard color number in a preset color number library. Preferably, the CIEDE color difference formula is used for calculation, which can simulate the human eye's perception of color difference. The calculated color difference distance of 2000 is compared with a preset threshold, and the standard color number with the smallest color difference distance that is less than the threshold is selected as the final color number of the yarn package.

[0023] In particular, the preset color code library is a configurable database. Operators can easily add, delete, or modify the standard color codes and their corresponding CIE-Lab values ​​in the library through a human-machine interface according to different production batches or specific customer needs.

[0024] Subsequently, after determining the final color number of the yarn package, the sorting control module generates instructions to drive the equipment to perform physical sorting. The robotic arm first moves above the yarn package to be grasped, then the cylinder extends its output end, pushing the connecting plate downwards. The connecting plate, through leverage, squeezes the inner sides of multiple grippers, causing them to contract synchronously towards the center, thereby firmly clamping the yarn package. Afterwards, the robotic arm transports the yarn package to the target area determined by its color number. Upon reaching the target position, the cylinder extends its output end, the connecting plate retracts accordingly, and pulls the grippers through the connection relationship, causing them to open synchronously, completing the release of the yarn package.

[0025] This invention provides a machine vision-based equipment for yarn bobbin color recognition and sorting. It offers the following advantages:

[0026] 1. This invention uses the Mask-R-CNN model to segment the yarn package, which can separate the target from the complex background. Combined with the channel attention mechanism, it can effectively suppress visual noise interference caused by physical characteristics such as feathers and entanglement textures on the surface of the yarn package. Furthermore, by utilizing the characteristic that the CIE-Lab color space is insensitive to changes in illumination, it can ultimately achieve a highly accurate and stable color recognition capability under different illumination conditions and yarn surface conditions, with significant advantages, especially when dealing with similar color numbers with very small color differences.

[0027] 2. This invention uses a camera for visual recognition and works in conjunction with a robotic arm and grippers to automate the entire process from image acquisition of the sorting area, target recognition, color matching to final physical grasping and placement. It replaces traditional manual sorting operations, not only fundamentally eliminating errors and inconsistencies caused by subjective judgment, but also increasing sorting speed by orders of magnitude and significantly improving the overall efficiency of the textile production line.

[0028] 3. This invention integrates the gripper, camera, and supplementary light into the execution end of the robotic arm. The supplementary light provides imaging conditions for the camera, making the equipment easy to deploy and maintain in complex industrial environments. At the same time, by setting a preset color code library that can be dynamically configured by the user, the equipment can flexibly adapt to the color standards of different production batches and quickly switch sorting tasks without retraining the algorithm model, demonstrating strong industrial practicality and flexibility. Attached Figure Description

[0029] Figure 1 This is a three-dimensional structural diagram of the present invention;

[0030] Figure 2This is a schematic diagram of an embodiment of the present invention;

[0031] Figure 3 This is a schematic diagram of the gripper portion of the present invention;

[0032] Figure 4 This is a schematic diagram of the support structure distribution according to the present invention;

[0033] Figure 5 This is a schematic diagram of the support structure of the present invention;

[0034] Figure 6 This is a schematic diagram of the system architecture of the present invention.

[0035] Among them, 1. robotic arm; 101. cylinder; 2. gripper; 3. connecting plate; 4. camera; 5. bracket. Detailed Implementation

[0036] The technical solutions in the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0037] To better understand the present invention, the above content will be described in detail below with reference to specific embodiments.

[0038] Example 1, please refer to the appendix. Figure 1 - Appendix Figure 5 This invention provides a machine vision-based yarn color recognition and sorting equipment, comprising: a robotic arm 1, with grippers 2 and a cylinder 101 mounted on its movable end, the grippers 2 being circumferentially distributed outside the cylinder 101; a connecting plate 3, one side of which is connected to the output end of the cylinder 101, and the edge of which is connected to the grippers 2, the connecting plate 3 being used to drive the grippers 2 to move along with the cylinder 101; a camera 4, which is mounted on the other side of the connecting plate 3, for collecting yarn color; and the grippers 2 being located outside the camera 4; and a bracket 5, which is mounted on the outer wall of the connecting plate 3 and is staggered with the grippers 2, the bracket 5 having a supplementary light installed inside, for providing illumination to the collection area of ​​the camera 4.

[0039] In this embodiment, the robotic arm can be used to operate the gripper 2 and the cylinder 101. The cylinder 101 can drive the gripper 2 to operate, thereby causing the gripper 2 to open and close, thus achieving the clamping and release of the yarn package. Subsequently, in conjunction with the movement of the robotic arm, and with the camera 4 collecting and analyzing the color of the yarn package, the robotic arm can be controlled to move yarn packages of different colors to different positions, thereby achieving the sorting action of the yarn package.

[0040] When the robotic arm 1 and cylinder 101 are running, the robotic arm 1 is used to drive the cylinder 101 and gripper 2 to move. After the cylinder 101 retracts its output end and pulls the connecting plate 3 back, the connecting plate 3 pulls the gripper 2 to close the gripper 2 and clamp the yarn. After the cylinder 101 extends its output end and pushes the connecting plate 3 out, the connecting plate 3 squeezes the gripper 2 to open the gripper 2 and release the yarn.

[0041] In this embodiment, the connecting plate 3 can be moved by the cylinder 101, and the connecting plate 3 can squeeze or pull the gripper 2, thereby realizing the opening and closing of the gripper 2.

[0042] Multiple supports 5 are provided, arranged in a ring around the outside of the camera 4 to provide a ring light source, and the supports 5 are coaxially arranged around the lens of the camera 4.

[0043] In this embodiment, by setting up multiple brackets 5, multiple supplementary lights can be set coaxially along the lens of the camera 4, thereby providing a ring light source for the camera 4, thus providing good lighting conditions for the camera 4 and improving the accuracy of yarn package image acquisition.

[0044] Working principle: During use, the robotic arm can drive the cylinder 101 and gripper 2 to move. When it is necessary to sort the yarn package, firstly, the robotic arm moves the cylinder 101 and gripper 2 above the yarn package. At this time, the camera and supplementary light are running. The supplementary light can illuminate the yarn package, which is conducive to the camera 4 to collect images of the yarn package. After the collection is completed, the controller can analyze the collected images of the yarn package and drive the robotic arm and cylinder 101 to run based on the analysis results.

[0045] Specifically: First, the robotic arm moves the gripper 2 close to the yarn package and wraps it around the yarn. Then, the drive cylinder 101 retracts its output end, causing the connecting plate 3 to move closer to the robotic arm. At this time, the connecting plate 3 pulls the gripper 2, causing it to move. Subsequently, the gripper 2 closes, clamping the yarn package. Then, the robotic arm moves to pick up the yarn package and move it to the designated position. Finally, the drive cylinder 101 extends its output end, causing the connecting plate 3 to move away from the robotic arm and squeeze the gripper 2. At this time, the gripper 2 opens, releasing the yarn package. This completes one sorting operation. After placing the yarn package, the next sorting operation can begin.

[0046] Example 2, based on Example 1, please refer to the appendix. Figure 6 In this embodiment:

[0047] The robotic arm 1 also has a controller installed inside, which includes:

[0048] The image data acquisition module is used to receive images of yarn bobbins captured by camera 4;

[0049] In this embodiment, to achieve automated operation and intelligent recognition of the aforementioned equipment, specific information about the controller is also provided. This controller receives information from external sensors, executes algorithms, and issues control commands to the robotic arm. The controller mainly includes: an image data acquisition module, a signal processing and color recognition module, and a sorting control module.

[0050] Specifically, the image data acquisition module is connected to the camera 4 and receives the raw image data containing the yarn package acquired by the camera 4.

[0051] Specifically, this module is responsible for initializing the parameters of camera 4 and establishing a communication link with camera 4. In one possible implementation, the image acquisition process can be triggered by an event. For example, when robotic arm 1 moves above a preset shooting position in the sorting area, camera 4 acquires an image of the yarn package. Alternatively, the acquisition process can be continuous. Camera 4 continuously transmits a video stream to the image data acquisition module at a preset frame rate (e.g., 30 frames per second).

[0052] Typically, the images received by this module are digital images, such as an RGB image with a resolution of 1920x1080 pixels and 24-bit true color encoding. Upon receiving the raw image data, the module does not simply forward the data. It first performs an integrity check on the data frame to ensure that what is received is a complete image without any corrupted or missing data.

[0053] Furthermore, to ensure the accuracy and repeatability of the measurement results, this module invokes the camera 4 calibration program upon the system's first startup or after camera 4 is replaced. This program corrects the geometric distortions of camera 4 itself, such as barrel distortion or pincushion distortion, ensuring that straight lines in the final image are also straight lines in the real world, providing an accurate image basis for subsequent instance segmentation and dimensional measurements.

[0054] Finally, the image data acquisition module stores the verified, possibly pre-processed, timestamped, complete image data into a specific high-speed memory buffer or data queue for the signal processing and color recognition modules to access at any time. This modular design ensures a clear and efficient data stream, providing a stable and reliable data source for subsequent complex image processing and recognition algorithms.

[0055] The signal processing and color recognition module receives and processes the yarn package image to determine the final color code of the yarn package.

[0056] In this embodiment, after the image data acquisition module transmits stable and clear yarn package image data to the controller, the signal processing and color recognition module takes over the processing flow. The aim is to accurately and stably extract the final color code information of each yarn package from the complex original image.

[0057] Specifically, the signal processing and color recognition module is broken down into several sequentially executed units, including: an instance segmentation unit, a feature enhancement unit, a color feature extraction unit, and a color matching unit. These units work together to form a complete processing pipeline from target localization, feature purification, feature transformation to final decision-making.

[0058] Among them, instance segmentation unit

[0059] First, the image data is fed into the instance segmentation unit. This unit is used to accurately locate and separate each target object of interest in a complex scene.

[0060] Specifically, this unit employs the Mask-R-CNN (Mask Region Convolutional Neural Network) model. This model receives the entire yarn package image from the image data acquisition module and performs depth analysis on it. Its output is not a simple rectangular box, but rather outlines the contour of each yarn package in the image, thus completely separating it from the background, adjacent yarn packages, and any debris at the data level.

[0061] In one possible implementation, the Mask-R-CNN model specifically includes:

[0062] The backbone network employs mature convolutional neural network structures such as ResNet (Residual Network) or EfficientNet. It is responsible for performing preliminary feature extraction on the input yarn package image, generating an initial feature map containing information from low to high levels (such as edges, textures, and components) in the image.

[0063] The feature enhancement unit, located after the backbone network, receives the initial feature map output from the backbone network and processes it using a channel attention mechanism. This mechanism enables the model to adaptively learn which regions (spatial dimensions) and features (channel dimensions) in the image are most important for identifying the color of the yarn package. For example, it enhances the focus on the main area of ​​the yarn package while suppressing the response to noise features such as feathers caused by yarn fluffiness and fine textures caused by uneven winding on the yarn surface. In this way, the core features related to the target color of the yarn package are greatly enhanced, providing higher-quality data for subsequent processing.

[0064] The segmentation head receives the enhanced feature maps processed by the feature enhancement unit and, based on these feature maps, generates an independent and accurate binary mask region for each yarn bobbin instance in the image. The mask region data is then passed to the subsequent color feature extraction unit.

[0065] Color feature extraction unit

[0066] After obtaining the precise mask area for each yarn bobbin, the color feature extraction unit extracts stable and reliable color features that can resist interference from changes in ambient lighting in the workshop.

[0067] Generally, the image data directly output by an industrial camera is based on the RGB (red, green, blue) color space. However, RGB values ​​are extremely sensitive to changes in light intensity and color temperature. Even small fluctuations in light can cause significant changes in RGB values, leading to color misjudgment.

[0068] Therefore, in this embodiment, the color feature extraction unit first performs a crucial color space conversion on the image data within the mask area, that is, converting it from the RGB color space to the CIE-Lab color space. The CIE-Lab color space is a color model that is more in line with human visual perception. After separating brightness and color information, even if the ambient light changes (mainly affecting the color component), the brightness and color components can remain relatively stable, thereby enabling the extraction of pure color features that are not sensitive to changes in light.

[0069] Color matching unit

[0070] Once stable and reliable color features are extracted, the final decision-making stage begins—executed by the color matching unit. This unit uses a classifier to match the extracted color features with preset standard color codes to determine the final color code of the yarn package.

[0071] Specifically, the steps by which the color matching unit matches color features with preset color codes include:

[0072] First, in order to obtain a single quantized value that represents the overall color of the yarn from a mask region containing thousands of pixels, the unit calculates the arithmetic mean of the color features of all pixels in the mask region.

[0073] In the CIE-Lab color space, the color value of any pixel within the mask region can be represented as:

[0074]

[0075] If the total number of pixels in the area is N, then the quantized color value of the yarn package... The calculation is as follows:

[0076]

[0077] In the formula, C i This represents the color value vector of the i-th pixel within the mask region; This represents the luminance component of the i-th pixel in the CIE-Lab color space; This represents the green-red color components of the i-th pixel in the CIE-Lab color space; This represents the blue-yellow color component of the i-th pixel in the CIE-Lab color space; i represents the index of the pixel, which is an integer from N; N represents the total number of pixels contained in a specific yarn mask region output by the instance segmentation unit. This represents the quantized color value vector calculated from the final values ​​of the entire yarn package. This represents the average brightness component of the entire yarn package; This represents the average green-red color component of the entire yarn package; This represents the average blue-yellow color component of the entire yarn package.

[0078] In other embodiments, this embodiment may also set a preset color difference threshold. Only when the minimum color difference distance is less than the threshold is the corresponding standard color number confirmed as the final color number of the yarn package. If the minimum color difference is also greater than the threshold, the yarn package can be determined to be of unknown color or a defective product.

[0079] The preset color code library is a configurable database that allows users or field engineers to easily add new standard color codes, delete outdated color codes, or modify the CIE-Lab standard values ​​of existing color codes through a human-machine interface, based on the needs of the current production batch, without having to retrain the core algorithm model, which greatly improves the adaptability and practicality of the equipment.

[0080] The sorting control module is used to generate control commands based on the final color number of the yarn package to drive the robotic arm 1 and cylinder 101 to sort the yarn package to the corresponding target position.

[0081] In this embodiment, the sorting control module receives the yarn color number and its position information in the image coordinate system output by the signal processing and color recognition module. Based on these inputs, it generates control commands to drive the robotic arm 1 and cylinder 101 to complete the entire sorting operation from grasping to placement.

[0082] Specifically, this module maintains a color code-target location mapping table. This table can be a configurable data structure, such as a look-up table or an associative array, that associates each preset yarn color code with one or more specific three-dimensional spatial coordinate points. These coordinate points correspond to different collection bins, conveyor belts, or packaging areas in the production site.

[0083] In one possible implementation, after the sorting control module receives a specific yarn color number (e.g., "bright red-01") and its location information, its workflow is as follows:

[0084] First, the module queries the mapping table to obtain the three-dimensional coordinates of the target placement position corresponding to "positive red-01".

[0085] Subsequently, the module initiates a path planning algorithm. This algorithm calculates the trajectory of robotic arm 1 from its current position to directly above the yarn to be grasped. Generating this trajectory requires considering the kinematic model of robotic arm 1, and the required rotation angles of each joint are determined through inverse kinematics calculations.

[0086] In some embodiments, the path planning algorithm can also load static obstacle information within the workspace. When generating the trajectory, the algorithm actively avoids these obstacles to generate a collision-free, safe path.

[0087] Once the end effector of the robotic arm 1 reaches the predetermined gripping position, the sorting control module generates and sends a gripping action command. This command first drives the retracting end of the cylinder 101 to extend, which in turn pushes the connecting plate 3 to squeeze the gripper 2 to retract, thereby completing the clamping action on the yarn package.

[0088] Alternatively, the control signal sent to cylinder 101 can include parameterized control of the force or speed. This allows for flexible gripping of the yarn package, preventing damage to the yarn or deformation of the package due to excessive clamping force.

[0089] After confirming that the yarn package is stably clamped, the module calls the path planning algorithm again to calculate the motion trajectory of the robotic arm 1 from the current gripping position to the target placement position obtained by the aforementioned table lookup.

[0090] Normally, robotic arm 1 transports the yarn bobbin to the target position along this trajectory. At this point, the module generates and sends a release command. This command drives the output end of cylinder 101 to extend, causing connecting plate 3 to be pushed out and squeeze gripper 2 to open, thereby releasing the yarn bobbin into the designated collection bin.

[0091] Throughout the process, the sorting control module continuously receives and processes feedback signals from encoders at each joint of the robotic arm 1, position sensors at cylinder 101, and pressure sensors at gripper 2. This closed-loop control method ensures that every instruction is executed accurately. If an anomaly is detected at any stage (e.g., gripping) (e.g., insufficient pressure from the sensor indicating gripping failure), the module immediately suspends the current task flow and executes a preset anomaly handling procedure, such as retrying the gripping or sending an alarm to the central monitoring system.

[0092] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A machine vision-based yarn color recognition and sorting equipment, characterized in that, include: The robotic arm (1) has a gripper (2) and a cylinder (101) installed at its movable end. The gripper (2) is distributed in a circular pattern on the outside of the cylinder (101). The connecting plate (3) has its center connected to the output end of the cylinder (101) and its edge connected to the gripper (2). The connecting plate (3) is used to drive the gripper (2) to move along with the cylinder (101). A camera (4) is mounted on the other side of the connecting plate (3) and is used to collect the color of the yarn package; and the gripper (2) is located outside the camera (4); The bracket (5) is installed on the outer wall of the connecting plate (3) and is staggered with the gripper (2). The bracket (5) is equipped with a supplementary light to provide illumination to the acquisition area of ​​the camera (4). The robotic arm (1) is also equipped with a controller, which includes: The image data acquisition module is used to receive the image of the yarn package acquired by the camera (4); A signal processing and color recognition module is used to receive and process the yarn package image to determine the final color code of the yarn package. The sorting control module is used to generate control commands based on the final color number of the yarn package to drive the robotic arm (1) and cylinder (101) to sort the yarn package to the corresponding target position. The signal processing and color recognition module includes: An instance segmentation unit is used to process the image using a Mask-R-CNN model to generate a mask region for locating the yarn package and separating it from the background. The feature enhancement unit is used to enhance the feature attention of the mask region and suppress background noise through a channel attention mechanism; The color feature extraction unit is used to perform color space conversion on the image data within the mask area output by the feature enhancement unit in order to extract color features that are not sensitive to changes in illumination. The color matching unit is used to identify the extracted color features through a classifier and match them with a preset color number to determine the final color number of the yarn package. The bracket (5) is configured as a plurality of brackets, which are arranged in a ring around the outside of the camera (4) to provide a ring light source, and the brackets (5) are coaxially arranged around the lens of the camera (4); The color feature extraction unit is used to convert the image data within the mask area from the RGB color space to the CIE-Lab color space in order to extract color features that are not sensitive to changes in illumination. The step of the color matching unit matching the color feature with a preset color number includes: First, the average value of the color features of all pixels in the mask area output by the color feature extraction unit is calculated to obtain the quantized color value of the current yarn package. Subsequently, the color difference distance between the quantified color value and each standard color number in the preset color number library is calculated, and the standard color number with the smallest color difference distance and less than the preset threshold is selected as the final color number of the yarn package. The preset color code library is a configurable database that allows users to adjust the standard color codes according to the needs of production batches.

2. The machine vision-based yarn color recognition and sorting equipment according to claim 1, characterized in that, When the robotic arm (1) and cylinder (101) are running, the robotic arm (1) is used to drive the cylinder (101) and gripper (2) to move. After the cylinder (101) extends its output end to push out the connecting plate (3), the connecting plate (3) squeezes the gripper (2) to open the gripper (2) and release the yarn. After the cylinder (101) retracts its output end and retracts the connecting plate (3), the connecting plate (3) pulls the clamp (2) to close the clamp (2) and clamp the yarn.

3. The machine vision-based yarn color recognition and sorting equipment according to claim 1, characterized in that, The Mask-R-CNN model specifically includes: The backbone network is used to extract the initial feature map of the yarn package image; The feature enhancement unit, integrated after the backbone network, is used to receive the initial feature map and process the initial feature map through a spatial channel attention mechanism to enhance features related to the yarn target and suppress background noise. A segmentation head is used to generate the mask region for locating the yarn bobbin based on the enhanced feature map output by the feature enhancement unit.

Images