A method and system for short fiber anomaly detection based on machine learning

Abstract

The present application provides a short fiber anomaly detection method and system based on machine learning. By extracting feature objects from the spinning sliver image, a tracer fiber region image and a background fiber region image of the spinning sliver are obtained; based on the tracer fiber region image, connected region segmentation is performed to obtain multiple sub-images of the tracer fiber region, and the fiber radial break points corresponding to each sub-image of the tracer fiber region are extracted; furthermore, the hook shape of the tracer fiber region image is extracted to obtain the hook shape image score of the tracer fiber region image; according to the hook shape image score of the tracer fiber region image, the drafting quality of the spinning short fibers in the roving process is detected. The present application can perform a hook shape image score based on the hook shape of the tracer fiber image, thereby realizing the quality detection of the short fiber drafting process in the roving process and improving the accuracy of the anomaly detection of the spinning short fiber quality.

Description

Technical Field

[0001] The present application relates to the technical field of fiber image detection, and more specifically, to a short fiber anomaly detection method and system based on machine learning. Background Art

[0002] The yarn structure refers to the arrangement of fibers within the yarn, and the arrangement of fibers is highly related to the spinning process and fiber raw materials. In the traditional spinning process, randomly arranged raw cotton fibers are subjected to a series of processes such as opening, carding, drawing, roving, and spinning, and then presented as fiber arrangements. The straightness and parallelism of fibers in the spinning sliver are significantly correlated with the yarn quality. The drafting process will cause a small number of front hooks to form in the fibers, and at the same time, the back hooks will be straightened. Analyzing the hook shape and quantity of fibers in the sliver is important for evaluating the spinning drafting process.

[0003] In existing spinning fiber detection methods, the hook shape of fibers during the drafting process is mainly detected by manual visual inspection, that is, the image of the spinning sliver during the spinning process is collected and magnified, and the process evaluation and anomaly detection are carried out by manually observing the morphological characteristics of the fiber sliver. This method not only consumes a large amount of labor costs and takes a long time, but also relies on the experience of technicians, which greatly restricts the development of the spinning fiber drafting process. Summary of the Invention

[0004] The present application provides a short fiber anomaly detection method and system based on machine learning, which can perform hook shape image scoring based on the hook shape of the tracer fiber image, thereby realizing the quality detection of the short fiber drafting process in the spinning process and improving the accuracy of the anomaly detection of the spinning short fiber quality.

[0005] In the first aspect, the present application provides a short fiber anomaly detection method based on machine learning. The short fiber anomaly detection method includes the following steps:

[0006] After the drafting in the spinning process, image collection is performed on the spinning short fiber sliver output by the front roller to obtain a spinning sliver image;

[0007] Feature object extraction is performed on the spinning sliver image to obtain the tracer fiber region image and the background fiber region image of the spinning sliver;

[0008] Based on the tracer fiber region image, connected region segmentation is performed to obtain multiple tracer fiber region sub-images, and a machine learning algorithm is used to extract fracture features from each tracer fiber region sub-image to obtain the fiber radial fracture points corresponding to each tracer fiber region sub-image;

[0009] Obtain the background fiber features of the background fiber region image, and based on the background fiber features and the radial fracture positions corresponding to each fiber radial fracture point, extract the hook shape of the tracer fiber region image to obtain the hook shape image score of the tracer fiber region image;

[0010] According to the hook shape image score of the tracer fiber region image, detect the drafting quality of the spinning short fibers during the roving process.

[0011] In this embodiment, the specific steps of extracting the characteristic object of the roving image to obtain the tracer fiber region image and the background fiber region image of the roving are as follows:

[0012] Obtain the color level of the tracer fiber, and map to obtain the tracer threshold according to the threshold interval where the color level of the tracer fiber is located;

[0013] Perform image segmentation on the roving image according to the tracer threshold to obtain the tracer fiber region image and the background fiber region image of the roving.

[0014] In this embodiment, obtaining the color level of the tracer fiber specifically includes: using a standard sample of the tracer fiber blended with the background fiber for image acquisition, and after converting the sample image to a color space, extracting the hue angle from the tracer fiber region as the color level of the tracer fiber.

[0015] In this embodiment, the specific steps of performing connected region segmentation on the tracer fiber region image to obtain multiple tracer fiber region sub-images are as follows: Obtain the tracer fiber region image, define the image connectivity standard, and perform connected region segmentation on the tracer fiber region image based on the image connectivity standard to obtain multiple tracer fiber region sub-images.

[0016] In this embodiment, the specific steps of using a machine learning algorithm to extract the fracture features of each tracer fiber region sub-image to obtain the fiber radial fracture points corresponding to each tracer fiber region sub-image are as follows:

[0017] Input the training data set into a support vector machine for machine learning training, where the training data set includes standard tracer fiber region images and labeled fracture points;

[0018] For any tracer fiber region sub-image, use an edge detection algorithm to extract the edge gray mutation position corresponding to the tracer fiber region sub-image, extract the connected region shape attribute label of the tracer fiber region sub-image, and use a gray level co-occurrence matrix to extract the local texture features of the tracer fiber region sub-image;

[0019] Input the edge gray mutation positions, connected region shape attribute labels, and local texture features corresponding to the sub-graphs of the tracer fiber regions into the support vector machine, so as to extract the fiber radial break points of each sub-graph of the tracer fiber region.

[0020] In this embodiment, based on the background fiber features and the radial break positions corresponding to each fiber radial break point, extracting the hook shape of the tracer fiber region image, and obtaining the hook shape image score of the tracer fiber region image specifically includes:

[0021] Extract the hook shape of each sub-graph of the tracer fiber region in the tracer fiber region image according to the background fiber features, and obtain the initial hook shape image score corresponding to each sub-graph of the tracer fiber region;

[0022] Obtain the fiber radial break points corresponding to each sub-graph of the tracer fiber region; determine the radial break positions corresponding to each fiber radial break point based on the fiber radial break points corresponding to each sub-graph of the tracer fiber region;

[0023] Extract the radial interference degree according to the radial break positions and the initial hook shape image scores corresponding to each fiber radial break point, and obtain the radial interference degree between the radial break positions and the initial hook shape image scores corresponding to each fiber radial break point;

[0024] Fuse and correct the initial hook shape image scores corresponding to each sub-graph of the tracer fiber region according to the radial interference degree, and obtain the hook shape image score of the tracer fiber region image.

[0025] In this embodiment, according to the hook shape image score of the tracer fiber region image, detecting the drafting quality of the spinning short fibers in the roving process specifically includes: determining the corresponding drafting quality grade according to the score interval where the hook shape image score of the tracer fiber region image is located.

[0026] In this embodiment, an industrial camera is used to collect images of the spinning short fiber sliver output by the front roller to obtain a spinning sliver image.

[0027] In this embodiment, before extracting the feature objects from the spinning sliver image, it further includes: performing image preprocessing on the spinning sliver image.

[0028] In a second aspect, the present application provides a short fiber anomaly detection system based on machine learning for executing a short fiber anomaly detection method based on machine learning. The short fiber anomaly detection system includes:

[0029] An image acquisition module, configured to acquire an image of the spinning short fiber sliver output by the front roller after drafting in the roving process, so as to obtain a spinning sliver image;

[0030] An image processing module, configured to extract feature objects from the spinning sliver image, so as to obtain a tracer fiber region image and a background fiber region image of the spinning sliver;

[0031] The image processing module is further configured to perform connected region segmentation based on the tracer fiber region image to obtain a plurality of tracer fiber region sub-images, and use a machine learning algorithm to extract fracture features from each tracer fiber region sub-image, so as to obtain fiber radial fracture points respectively corresponding to each tracer fiber region sub-image;

[0032] An image scoring module, configured to obtain background fiber features of the background fiber region image, and based on the background fiber features and the radial fracture positions corresponding to each fiber radial fracture point, extract the hook shape of the tracer fiber region image, so as to obtain a hook shape image score of the tracer fiber region image;

[0033] A fiber drafting detection module, configured to detect the drafting quality of the spinning short fiber during the roving process according to the hook shape image score of the tracer fiber region image.

[0034] The technical solution provided by the embodiments disclosed in this application has the following beneficial effects:

[0035] It can be seen that this application utilizes the significant color feature difference between the tracer fiber and the background fiber, extracts the tracer fiber separately through region segmentation, thereby concentrating on analyzing the hook shape of the tracer fiber. By using a machine learning algorithm to extract the fracture points of the fiber, the specific positions where the fiber breaks during drafting can be accurately obtained, marking the key points of the change in drafting force. The positions of the radial fracture points refine the non-uniformity of the fiber force, providing basic data for subsequent hook shape scoring. And by scoring the hook shape image to quantify the straightening and parallelism degree of the fiber, and then correcting the initial hook shape image score by analyzing the position of the radial fracture point, reducing the error caused by uneven force, thereby improving the robustness of the scoring. Finally, according to the hook shape image score of the tracer fiber region image, the drafting quality of the spinning short fiber during the roving process is detected, avoiding the subjective error in manual detection and ensuring the consistency and reliability of the detection result.

[0036] In summary, this application performs hook shape image scoring based on the hook shape of the tracer fiber image, thereby realizing the quality detection of the short fiber drafting process in the roving process and improving the accuracy of the quality anomaly detection of the spinning short fiber. Description of the Drawings

[0037] To more clearly illustrate the technical solutions in the embodiments of the present application or the prior art, the following will briefly introduce the accompanying drawings required in the description of the embodiments or the prior art. Obviously, the accompanying drawings in the following description are only the embodiments of the present application. For those of ordinary skill in the art, without creative efforts, other drawings can also be obtained based on these drawings.

[0038] Figure 1 is a flowchart of a short fiber anomaly detection method based on machine learning provided by the present application;

[0039] Figure 2 is a module structure diagram of a short fiber anomaly detection system based on machine learning provided by the present application. Detailed implementation manners

[0040] The following will clearly and completely describe the technical solutions in the embodiments of the present application with reference to the accompanying drawings in the embodiments of the present application. Obviously, the described embodiments are only a part of the embodiments of the present application, rather than all of the embodiments. Based on the embodiments of the present application, all other embodiments obtained by those of ordinary skill in the art without creative efforts belong to the scope of protection of the present application.

[0041] The embodiments of the present application provide a short fiber anomaly detection method and system based on machine learning. The core is to collect images of the spinning short fiber sliver output by the front roller after drafting in the roving process to obtain the spinning sliver image; extract feature objects from the spinning sliver image to obtain the tracer fiber region image and the background fiber region image of the spinning sliver; perform connected region segmentation on the tracer fiber region image to obtain multiple sub-images of the tracer fiber region, and use machine learning algorithms to extract fracture features from each sub-image of the tracer fiber region to obtain the fiber radial fracture points corresponding to each sub-image of the tracer fiber region; obtain the background fiber features of the background fiber region image, and based on the background fiber features and the radial fracture positions corresponding to each fiber radial fracture point, extract the hook shape of the tracer fiber region image to obtain the hook shape image score of the tracer fiber region image; according to the hook shape image score of the tracer fiber region image, detect the drafting quality of the spinning short fibers in the roving process. The present application can perform hook shape image scoring based on the hook shape of the tracer fiber image, thereby realizing the quality detection of the short fiber drafting process in the roving process and improving the accuracy of the anomaly detection of the spinning short fiber quality.

[0042] Embodiment 1. To better understand the above technical solutions, the following will describe the above technical solutions in detail with reference to the accompanying drawings of the specification and specific implementation manners. Refer to Figure 1As shown, the figure is an exemplary flowchart of a short fiber anomaly detection method based on machine learning according to this embodiment of the present application. The short fiber anomaly detection method includes the following steps:

[0043] In step S1, after drafting in the roving frame process, image acquisition is performed on the spinning short fiber sliver output by the front roller to obtain a spinning sliver image.

[0044] It should be noted that drafting in the roving frame process is a key link in the spinning process, aiming to further refine and straighten the fiber sliver through a drafting device to make it reach the fineness and quality required for the final yarn. It is an important transition step from roving to yarn. In the present application, the roller drafting system consists of front, middle, and rear roller groups, and the fibers are stretched by different rotational speeds. The speed ratio between the rollers is set on the roving frame to determine the drafting multiple, that is, the ratio of the thickness of the input sliver to the output sliver. The rear roller outputs the roving at a slower speed, and the middle roller and the front roller are gradually accelerated, so that the fibers are elongated in the drafting zone. During the drafting process, the fibers are under the combined action of the control force and the traction force and gradually reach a straightened and parallel state.

[0045] In this embodiment, an industrial camera is used to perform image acquisition on the spinning short fiber sliver output by the front roller to obtain a spinning sliver image; in some other embodiments, other devices or equipment capable of realizing image acquisition can also be used, and the present application does not limit this.

[0046] In this embodiment, a conveyor belt device is used to convey the spinning short fiber sliver output by the front roller, and the spinning sliver image is acquired during the conveying process.

[0047] In step S2, feature object extraction is performed on the spinning sliver image to obtain a tracer fiber region image and a background fiber region image of the spinning sliver.

[0048] In this embodiment, before performing feature object extraction on the spinning sliver image, it further includes: performing image preprocessing on the spinning sliver image.

[0049] Preferably, in some embodiments, performing image preprocessing on the spinning sliver image specifically includes: after performing grayscale processing on the spinning sliver image, median filtering is used to filter the noise of the initial detection image.

[0050] In this embodiment, performing feature object extraction on the spinning sliver image to obtain a tracer fiber region image and a background fiber region image of the spinning sliver specifically includes:

[0051] Obtaining the color level of the tracer fiber, and mapping to obtain a tracer threshold according to the threshold interval where the color level of the tracer fiber is located;

[0052] Performing image segmentation on the spinning sliver image according to the tracer threshold to obtain the tracer fiber region image and the background fiber region image of the spinning sliver.

[0053] In this embodiment, obtaining the color level of the tracer fiber specifically includes: performing image acquisition on a standard sample of the tracer fiber blended with the background fiber, and after converting the sample image into the HSV color space, extracting the hue angle from the tracer fiber region as the color level of the tracer fiber.

[0054] It should be noted that the tracer fiber uses fluorescent fiber or dyed fiber with similar performance to the main fiber but obvious color contrast. The tracer fiber is uniformly mixed into the fiber raw material in a certain proportion to ensure that the distribution of the tracer fiber is random and can represent the overall fiber behavior. When specifically implemented, according to the threshold interval where the color level of the tracer fiber is located, gray mapping can be performed according to a preset gray mapping table to obtain the corresponding gray value as the tracer threshold. Then, taking advantage of the difference in optical characteristics between the tracer fiber and the background fiber, region segmentation is performed according to the tracer threshold, that is, the pixel points with gray values higher than the tracer threshold are classified into the tracer fiber region, and other pixels are classified into the background fiber region, thereby obtaining the tracer fiber region image and the background fiber region image of the spinning sliver.

[0055] In step S3, performing connected region segmentation on the tracer fiber region image to obtain multiple sub-images of the tracer fiber region, and using a machine learning algorithm to extract the fracture characteristics of each sub-image of the tracer fiber region to obtain the fiber radial fracture points corresponding to each sub-image of the tracer fiber region.

[0056] In this embodiment, performing connected region segmentation on the tracer fiber region image to obtain multiple sub-images of the tracer fiber region specifically includes: obtaining the tracer fiber region image, defining an image connectivity standard, and performing connected region segmentation on the tracer fiber region image based on the image connectivity standard to obtain multiple sub-images of the tracer fiber region.

[0057] When specifically implemented, the image connectivity standard can be defined as a 4-connectivity standard, that is, each pixel is associated with its adjacent pixels above, below, left, and right. Then, the Flood Fill algorithm is used to mark the connected regions in the connected regions of the tracer fiber region image, and each separate connected region is used as a sub-image of the tracer fiber region, thereby obtaining multiple sub-images of the tracer fiber region.

[0058] It should be noted that in this application, filaments with equally spaced weak rings in the length direction are fed from behind the front roller of the roving frame and spun into roving together with the sliver. The roving is drafted by the spinning draft device, and the weak-ring filaments break at the weak rings during the drafting process and become sequential tracer short fibers distributed in the spinning sliver. The tracer fibers are distributed in different cross-sections in the radial direction. The fiber radial fracture points described in this application are the fiber fracture points at different cross-section positions of the tracer fibers in the radial direction. In this embodiment, a machine learning algorithm is used to extract the fracture characteristics of each sub-graph of the tracer fiber region, and the specific steps for obtaining the fiber radial fracture points corresponding to each sub-graph of the tracer fiber region include:

[0059] Input the training data set into the support vector machine for machine learning training, where the training data set includes standard tracer fiber region images and labeled fracture points;

[0060] For any sub-graph of the tracer fiber region, use the edge detection algorithm to extract the position of the edge gray-scale mutation corresponding to the sub-graph of the tracer fiber region, extract the shape attribute label of the connected region of the sub-graph of the tracer fiber region, and use the gray-level co-occurrence matrix to extract the local texture features of the sub-graph of the tracer fiber region;

[0061] Input the position of the edge gray-scale mutation, the shape attribute label of the connected region, and the local texture features corresponding to the sub-graph of the tracer fiber region into the support vector machine, so as to extract the fiber radial fracture point of this sub-graph of the tracer fiber region.

[0062] When specifically implemented, the Canny edge detection algorithm can be used to extract the position or amplitude of the edge gray-scale value change corresponding to the sub-graph of the tracer fiber region to form a feature vector, extract the rectangularity of the sub-graph of the tracer fiber region as the shape attribute label of the connected region, and then use the gray-level co-occurrence matrix to extract the image contrast of the sub-graph of the tracer fiber region as the local texture feature. After splicing the above three features into a complete feature vector, input it into the support vector machine, and use support vector regression prediction to obtain the position of the fiber radial fracture point of the sub-graph of the tracer fiber region.

[0063] In some other embodiments, it is also possible to input the image features at the fiber fracture into the machine learning algorithm for training, so as to learn the image features at the fiber fracture and identify the fiber radial fracture points in the sub-graph of the tracer fiber region. This application does not limit this.

[0064] In step S4, obtain the background fiber features of the background fiber region image, and based on the background fiber features and the radial fracture positions corresponding to each fiber radial fracture point, extract the hook shape of the tracer fiber region image to obtain the hook shape image score of the tracer fiber region image.

[0065] In this embodiment, obtaining the background fiber features of the background fiber region image specifically includes: obtaining the fiber arrangement direction of the background region image as the background fiber features of the background fiber region image. When specifically implemented, the direction vector corresponding to the fiber arrangement direction of the background region image can be used as the background fiber features.

[0066] It should be noted that the tracer fibers are distributed in different cross-sections in the radial direction. When the forces in different radial positions are uneven, it is easy to interfere with the accuracy parameters of the hook shape image scoring. In this application, the radial fracture positions corresponding to the radial fracture points of each fiber are determined, and the radial interference degree is determined based on the degree of connection between the radial fracture position and the initial hook shape image scoring. The final hook shape image scoring is corrected based on the radial interference degree, thereby reducing the detection interference caused by uneven forces in different radial distribution positions and improving the accuracy of short fiber hook shape detection.

[0067] In this embodiment, based on the background fiber features and the radial fracture positions corresponding to the radial fracture points of each fiber, extracting the hook shape of the tracer fiber region image, and obtaining the hook shape image scoring of the tracer fiber region image specifically includes:

[0068] Extracting the hook shape of each tracer fiber region sub-image in the tracer fiber region image according to the background fiber features to obtain the initial hook shape image scoring corresponding to each tracer fiber region sub-image;

[0069] Obtaining the radial fracture points of each tracer fiber region sub-image; determining the radial fracture positions corresponding to the radial fracture points of each fiber based on the radial fracture points of each tracer fiber region sub-image;

[0070] Extracting the radial interference degree according to the radial fracture positions and the initial hook shape image scoring corresponding to each radial fracture point of each fiber to obtain the radial interference degree corresponding to each tracer fiber region sub-image;

[0071] Fusing and correcting the initial hook shape image scoring corresponding to each tracer fiber region sub-image according to the radial interference degree to obtain the hook shape image scoring of the tracer fiber region image.

[0072] Optionally, in some embodiments, extracting the hook morphologies of each sub - image of the tracer fiber region in the image of the tracer fiber region according to the background fiber characteristics, and obtaining the initial hook morphology image scores corresponding to each sub - image of the tracer fiber region specifically includes: for any sub - image of the tracer fiber region, extracting the fiber curvature and the number of hook points of this sub - image of the tracer fiber region according to the direction vector corresponding to the fiber arrangement direction in the background fiber characteristics, generating the initial hook morphology image score corresponding to this sub - image of the tracer fiber region based on the fiber curvature and the number of hook points, and then determining the initial hook morphology image scores corresponding to other sub - images of the tracer fiber region in the same way.

[0073] When specifically implemented, the pixel coordinates corresponding to the fiber radial fracture points of each sub - image of the tracer fiber region are used as the radial fracture positions.

[0074] Optionally, in some embodiments, extracting the radial interference degrees of each sub - image of the tracer fiber region according to the radial fracture positions and the initial hook morphology image scores corresponding to each fiber radial fracture point specifically includes:

[0075] Obtaining the radial fracture positions corresponding to each fiber radial fracture point and assigning corresponding sequence labels;

[0076] Determining the projection distance of each fiber radial fracture point on the direction vector in the background fiber characteristics as the radial distance, and obtaining a radial distance sequence composed of the radial distances corresponding to each fiber radial fracture point;

[0077] Obtaining the initial hook morphology image scores corresponding to each sub - image of the tracer fiber region, determining the interference degree detection windows corresponding to each sub - image of the tracer fiber region according to the sequence labels of the fiber radial fracture points corresponding to each sub - image of the tracer fiber region, and then within the interference degree detection windows corresponding to each sub - image of the tracer fiber region, determining the radial interference degrees corresponding to each sub - image of the tracer fiber region through the radial distances and the initial hook morphology image scores corresponding to each fiber radial fracture point.

[0078] When specifically implemented, the initial hook morphology image scores corresponding to each sub - image of the tracer fiber region can be sorted according to the sequence labels of the fiber radial fracture points corresponding to each sub - image of the tracer fiber region, and then according to the window size of the preset interference degree detection window, the initial hook morphology image scores and the radial distances of the tracer fiber region sub - images corresponding to the same number of adjacent sequence labels are used as window elements, and the Pearson correlation coefficient between the initial hook morphology image score and the radial distance is used as the radial interference degree corresponding to this sub - image of the tracer fiber region.

[0079] Optionally, in some embodiments, during the process of fusing and correcting the initial hook shape image scores corresponding to each sub - image of the tracer fiber region according to the radial interference degree to obtain the hook shape image score of the tracer fiber region image, the initial hook shape image scores corresponding to each sub - image of the tracer fiber region are weighted and fused according to the radial interference degree to obtain the final hook shape image score.

[0080] In step S5, according to the hook shape image score of the tracer fiber region image, the drafting quality of the spinning short fibers during the roving process is detected.

[0081] In this embodiment, detecting the drafting quality of the spinning short fibers during the roving process according to the hook shape image score of the tracer fiber region image specifically includes: determining the corresponding drafting quality level according to the score range where the hook shape image score of the tracer fiber region image is located, and performing quality assessment according to the drafting quality level.

[0082] Specifically, when implemented, according to a preset score mapping table, the hook shape image score can be mapped to the corresponding drafting quality level.

[0083] Embodiment 2. The present application provides a short - fiber anomaly detection system based on machine learning. Refer to Figure 2 As shown, this figure is a schematic diagram of the short - fiber anomaly detection system according to this embodiment of the present application. The short - fiber anomaly detection system includes:

[0084] An image acquisition module 100, configured to collect an image of the spinning short - fiber sliver output by the front roller after drafting in the roving process to obtain a spinning sliver image;

[0085] An image processing module 200, configured to extract feature objects from the spinning sliver image to obtain a tracer fiber region image and a background fiber region image of the spinning sliver;

[0086] The image processing module 200 is further configured to perform connected - region segmentation on the tracer fiber region image to obtain a plurality of sub - images of the tracer fiber region, and use a machine - learning algorithm to extract fracture features from each sub - image of the tracer fiber region to obtain fiber radial fracture points corresponding to each sub - image of the tracer fiber region;

[0087] An image scoring module 300, configured to obtain background fiber features of the background fiber region image, and based on the background fiber features and the radial fracture positions corresponding to each fiber radial fracture point, extract the hook shape of the tracer fiber region image to obtain the hook shape image score of the tracer fiber region image;

[0088] The fiber draft detection module 400 is used to detect the draft quality of the short spinning fibers in the roving process according to the hook shape image score of the tracer fiber area image.

[0089] This application is described with reference to the flowcharts and / or block diagrams of methods, apparatuses (systems) and computer program products according to embodiments of the present application. It should be understood that each flow and / or block in the flowchart and / or block diagram can be implemented by computer program instructions, and the combination of the flows and / or blocks in the flowchart and / or block diagram can also be implemented by computer program instructions. These computer program instructions can be provided to the processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing devices to generate a machine, so that the instructions executed by the processor of the computer or other programmable data processing devices generate means for implementing the functions specified in one Figure 1 flow or multiple flows and / or blocks Figure 1 or multiple blocks.

[0090] Those of ordinary skill in the art can understand that all or part of the steps in the various methods of the above embodiments can be completed by instructing relevant hardware through a program. This program can be stored in a computer-readable storage medium, and the storage medium includes read-only memory (ROM), random access memory (RAM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), one-time programmable read-only memory (OTPROM), electrically-erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc memories, magnetic disk memories, tape memories, or any other medium that can be used to carry or store data and is computer-readable.

[0091] It should also be noted that the term "comprising", "including" or any other variant thereof is intended to cover non-exclusive inclusion, so that a process, method, commodity or device comprising a series of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, commodity or device. Without further limitation, an element defined by the statement "comprising an..." does not exclude the presence of additional identical elements in the process, method, commodity or device comprising the element.

Claims

1. A method for detecting short fiber anomalies based on machine learning, characterized in that: The short fiber abnormality detection method comprises the following steps: After the spinning process is drafted, the spinning short fiber whiskers output by the front roller are imaged to obtain the spinning whisker images; Extracting characteristic objects from the spinning whisker image to obtain a tracer fiber region image and a background fiber region image of the spinning whisker; Based on the tracer fiber region image, connected region segmentation is performed to obtain multiple tracer fiber region sub-graphs, and fracture features of each tracer fiber region sub-graph are extracted using a machine learning algorithm to obtain fiber radial fracture points corresponding to each tracer fiber region sub-graph; Acquire background fiber features of the background fiber area image, perform hook morphology extraction on the tracer fiber area image based on the background fiber features and radial fracture positions corresponding to radial fracture points of each fiber, and obtain a hook morphology image score of the tracer fiber area image; According to the hook morphology image score of the tracer fiber region image, the drafting quality of the spinning staple fibers in the spinning process is detected; Among them, based on the background fiber characteristics and the radial fracture positions corresponding to the radial fracture points of each fiber, the hook morphology of the tracer fiber region image is extracted, and the hook morphology image score of the tracer fiber region image is obtained, which specifically includes: According to the background fiber characteristics, hook morphology is extracted from each tracer fiber region sub-image in the tracer fiber region image to obtain initial hook morphology image scores corresponding to each tracer fiber region sub-image; Obtaining fiber radial fracture points corresponding to each tracer fiber region sub-image; determining radial fracture positions corresponding to each fiber radial fracture point based on the fiber radial fracture points corresponding to each tracer fiber region sub-image; According to the radial fracture positions corresponding to the radial fracture points of each fiber and the initial hook morphology image score, the radial interference degree is extracted to obtain the radial interference degree between the radial fracture positions corresponding to the radial fracture points of each fiber and the initial hook morphology image score; The initial hook morphology image scores corresponding to each tracer fiber region sub-image are fused and corrected according to the radial interference degree to obtain the hook morphology image score of the tracer fiber region image.

2. The method for detecting short fiber abnormalities based on machine learning according to claim 1, characterized in that: Extracting feature objects from the spinning whisker image to obtain the tracer fiber region image and the background fiber region image of the spinning whisker specifically includes: Obtaining the color grade of the tracer fiber, and mapping to obtain the tracer threshold value according to the threshold interval where the color grade of the tracer fiber is located; The spinning whisker image is segmented according to the tracing threshold value to obtain a tracing fiber region image and a background fiber region image of the spinning whisker.

3. A method for detecting short fiber abnormalities based on machine learning as claimed in claim 2, characterized in that: Obtaining the color grade of the tracer fiber specifically includes: using a standard sample of the tracer fiber and background fiber blended to collect images, converting the sample image into a color space, and extracting a hue angle from the tracer fiber region as the color grade of the tracer fiber.

4. The method for detecting short fiber abnormalities based on machine learning according to claim 1, characterized in that: Performing connected region segmentation based on the tracer fiber region image to obtain multiple tracer fiber region subgraphs specifically includes: acquiring the tracer fiber region image, defining an image connectivity standard, performing connected region segmentation on the tracer fiber region image based on the image connectivity standard to obtain multiple tracer fiber region subgraphs.

5. The method for detecting short fiber abnormalities based on machine learning according to claim 4, characterized in that: The machine learning algorithm is used to extract the fracture features of each tracer fiber region sub-image, and the fiber radial fracture points corresponding to each tracer fiber region sub-image are obtained, including: Inputting a training data set into a support vector machine for machine learning training, wherein the training data set includes a standard tracer fiber region image and annotated breakpoints; For any tracer fiber region subgraph, an edge detection algorithm is used to extract the edge grayscale mutation position corresponding to the tracer fiber region subgraph, the shape attribute label of the connected region of the tracer fiber region subgraph is extracted, and a gray level co-occurrence matrix is ​​used to extract the local texture features of the tracer fiber region subgraph; The edge grayscale mutation position, the connected region shape attribute label and the local texture feature corresponding to the tracer fiber region sub-graph are input into the support vector machine, so as to extract the fiber radial break point of the tracer fiber region sub-graph.

6. The method for detecting short fiber abnormalities based on machine learning according to claim 1, characterized in that: According to the hook morphology image score of the tracer fiber area image, the drafting quality of the spinning staple fibers in the spinning process is detected, which specifically includes: determining the corresponding drafting quality grade according to the scoring interval of the hook morphology image score of the tracer fiber area image.

7. The method for detecting short fiber abnormalities based on machine learning according to claim 1, characterized in that: An industrial camera is used to collect images of the spinning staple fiber whiskers output by the front roller to obtain the spinning whisker images.

8. The method for detecting short fiber abnormalities based on machine learning according to claim 1, characterized in that: Before extracting the characteristic objects from the spinning yarn image, the method further includes: performing image preprocessing on the spinning yarn image.

9. A machine learning-based short fiber anomaly detection system, used to execute a machine learning-based short fiber anomaly detection method as claimed in any one of claims 1 to 8, characterized in that: The short fiber abnormality detection system comprises: An image acquisition module is used to acquire images of the spinning short fiber whiskers output by the front roller after drafting in the spinning process to obtain images of the spinning whiskers; An image processing module is used to extract feature objects from the spinning whisker image to obtain a tracer fiber region image and a background fiber region image of the spinning whisker; The image processing module is further used to perform connected area segmentation based on the tracer fiber area image to obtain multiple tracer fiber area sub-graphs, and use a machine learning algorithm to extract fracture features of each tracer fiber area sub-graph to obtain fiber radial fracture points corresponding to each tracer fiber area sub-graph; An image scoring module is used to obtain background fiber features of the background fiber area image, and based on the background fiber features and radial fracture positions corresponding to radial fracture points of each fiber, perform hook morphology extraction on the tracer fiber area image to obtain a hook morphology image score of the tracer fiber area image; The fiber draft detection module is used to detect the draft quality of the spinning short fibers during the spinning process according to the hook morphology image score of the tracer fiber area image.

Images