Carding machine image online monitoring system and monitoring method

Abstract

The application provides a carding machine image online monitoring system and a monitoring method, comprising a carding machine, an image online monitoring device and a control system, and has the following characteristics: the image online monitoring device comprises a plurality of image acquisition modules, the control system comprises a man-machine interface, an image preprocessing module, an image feature extraction module, a classification and training module, a neps size calculation module and a neps statistics module, and the plurality of image acquisition modules arranged close to the front of the cotton web form an array combination. The cotton web output by the carding machine is synchronously detected by each image acquisition module to realize full-coverage and blind-area-free cotton web image data acquisition. The control system performs image preprocessing, image feature extraction, classification and training, neps size calculation and neps statistics to realize rapid detection and evaluation of the neps in the cotton web. The structure design is reasonable and compact, the failure rate is low, seamless full-coverage detection of the cotton web can be realized, and the reliability, stability, detection accuracy and detection efficiency can be improved.

Description

Technical Field

[0001] This invention belongs to the field of textile machinery manufacturing technology, and relates to online monitoring technology for carding machines, specifically an online image monitoring system and method for carding machines. Background Technology

[0002] In the textile machinery field, the carding machine, as the core equipment for fiber pretreatment, directly affects the quality of subsequent textile products due to the quality of its output fiber web. If defects such as neps and impurities in the fiber web are not detected in time, they will be difficult to eliminate in subsequent processes, leading to fabric appearance defects. Therefore, online nep detection technology is a key link in improving textile production efficiency and product quality. Fiber web quality mainly involves detecting indicators such as neps, impurities, broken webs, and clarity. Fiber web clarity is a comprehensive indicator reflecting the structural state of the fiber web. Through the clarity of the cotton web, the mechanical status of the carding machine and the rationality of its process parameters can be understood in real time. By detecting the fiber web quality, the work quality of the preceding processes and the operating status of the carding machine can also be determined. Based on the detection results, the carding machine's quantity, speed, spacing, and process parameters can be optimized, and problems existing in the carding machine can be resolved in a timely manner, thereby improving product quality.

[0003] Currently, most mainstream cotton nip detection solutions are based on image sensing technology, using cameras to dynamically acquire images of the fiber web and combining this with image processing algorithms to identify defects. A representative technology is the "Online Monitoring System for Fiber Defects on a Carding Machine" disclosed in Chinese Patent (Authorization Announcement No.: CN209323052U), which includes a doffer, stripping roller, upper roller, lower roller, and a camera system and reciprocating movement mechanism composed of a camera, lighting device, and controller. Its key features are: the camera system and reciprocating movement mechanism are installed in a sealed enclosure; a transparent window is installed on the sealed enclosure directly opposite the camera lens; the sealed enclosure is fixed to a bracket. The travel distance of the reciprocating movement mechanism is slightly less than the width of the doffer; the monitoring position of the online monitoring system is on the side of the lower roller and in front of the doffer; the camera is horizontally positioned, with the camera lens and transparent window perpendicular to and directly facing the doffer, allowing simultaneous monitoring of the doffer fiber web and doffer card cloth. This reduces obstruction from falling debris, ensures accurate monitoring data, improves monitoring quality, and reduces cleaning workload. Its core structure includes: 1. Single camera and cable chain moving mechanism: The camera is installed in a closed box and moves back and forth along the width of the doffer via cable chain drive, covering the transverse area of ​​the fiber mesh; 2. Vertical viewing window design: The lens faces the doffer, and the transparent viewing window is installed vertically to reduce fly debris coverage; 3. Cable cable chain protection: The highly flexible cable of the moving mechanism is managed through a closed cable chain to reduce the risk of bending.

[0004] Similar technologies include Chinese patents for "fiber defect monitoring device for carding machine" with application number CN96102829.7 and "online image detection system" with application number CN200720005144.1, both of which use a single sensor in conjunction with a mechanical moving mechanism to achieve detection.

[0005] Despite the progress made by the above-mentioned solutions in automated detection, the following significant problems still exist:

[0006] 1. Low mechanical reliability: The cable chain drive mechanism relies on moving parts such as motors and belts. During high-speed reciprocating movement, it is prone to jamming and wear (such as loosening of the drive belt and deformation of the guide rail), which leads to detection interruption and a high failure rate.

[0007] 2. Cable failure risk: Cables inside the drag chain are prone to breakage or poor contact due to repeated bending, especially under conditions with small turning radii, electrical faults occur frequently, requiring frequent cable replacement and increasing maintenance costs;

[0008] 3. Insufficient detection coverage: Horizontal missed detection: A single camera needs to stitch images together by moving it. Due to mechanical precision limitations, blind spots are prone to appear at the stitching points; Vertical missed detection: When the camera's exposure frequency does not match the speed of the fiber mesh movement, there are gaps in the vertical detection, and the missed detection rate can reach 5%-10%.

[0009] 4. Data errors and randomness:

[0010] Dynamically stitched images suffer from data misalignment due to time asynchrony, resulting in highly random statistical results that are difficult to accurately reflect the distribution of cotton knots.

[0011] 5. High maintenance complexity: The coating on its transparent window can reduce fly debris adhesion, but it still requires periodic cleaning. In addition, the cable chain mechanism is complicated to disassemble and assemble, and the downtime for maintenance is long.

[0012] The "fiber defect monitoring device for carding machine" with application number CN96102829.7 uses a mobile photoelectric sensor. Although it simplifies the mechanical structure, it has low detection resolution (it can only identify large-sized defects) and cannot meet the requirements for high precision.

[0013] The "Online Image Detection System" with application number CN200720005144.1 introduces multi-source illumination to improve image quality, but still relies on single-sensor motion detection and does not solve the coverage problem.

[0014] In summary, the key contradiction in existing technologies lies in the conflict between "mobile detection mechanisms" and "the need for full coverage and high reliability".

[0015] Therefore, there is an urgent need to design an online image monitoring system and method for carding machines to eliminate faults caused by mechanical moving parts, improve system reliability and stability, achieve seamless full coverage detection of cotton web in both the horizontal and vertical directions, optimize cable management to reduce the risk of electrical faults, improve data synchronization and detection accuracy, and increase detection efficiency so that the equipment can complete the detection in a shorter time. Summary of the Invention

[0016] To address the aforementioned problems in the prior art, this invention provides an online image monitoring system and method for carding machines. The system has a reasonable and compact structure, a low failure rate, and can achieve seamless full-coverage detection of the cotton web in both the transverse and longitudinal directions. It also improves reliability, stability, detection accuracy, and detection efficiency.

[0017] The objective of this invention is achieved through the following technical solution:

[0018] An online image monitoring system for a carding machine includes a carding machine, an online image monitoring device, and a control system. The online image monitoring device comprises several image acquisition modules, and the control system includes a human-machine interface, an image preprocessing module, an image feature extraction module, a classification and training module, a nip size calculation module, and a nip statistics module. Several image acquisition modules are fixedly arranged in an array near the front of the cotton web. Each image acquisition module synchronously detects the dynamically output cotton web from the carding machine, achieving full-coverage, blind-spot-free cotton web image data acquisition. The control system performs image preprocessing, image feature extraction, classification and training, nip size calculation, and nip statistics based on the acquired cotton web image data through its functional modules, enabling rapid detection and evaluation of nipples in the cotton web.

[0019] Improvement to the above technical solution: Several image acquisition modules are arranged horizontally and fixed inside a U-shaped long strip aluminum shell to form an aluminum shell module. A transparent window is provided in front of the aluminum shell module. The aluminum shell module is horizontally positioned in front of the cotton web to be tested. The length of the aluminum shell module is greater than the horizontal width of the cotton web to be tested. An LED lighting module is integrated inside the aluminum shell module and connected by a flexible circuit board and modular cables. The transparent window and the LED lighting module face the cotton web to be tested.

[0020] Further improvements to the above technical solution: The carding machine is a carding machine, which includes a doffer and a stripping roller. A frame beam is set below the stripping roller and directly in front of the doffer. A guide rail is installed on the frame beam. A slider is set below the aluminum shell module. The slider slides horizontally on the guide rail. During installation, the aluminum shell module is inserted into the gap and fixed by sliding along the guide rail with the slider.

[0021] Further improvements to the above technical solution: The image acquisition module includes a mounting base and a camera. The camera includes a back-illuminated CMOS image sensor and a wide-angle short-range lens. Each image acquisition module occupies a three-dimensional space dimension of no more than 30×30×20mm. The distance between the transparent window in front of the aluminum shell module and the cotton net to be tested is no more than 3mm.

[0022] Further improvement to the above technical solution: The formula for calculating the minimum number of image acquisition modules forming an array combination is:

[0023]

[0024] in:

[0025] L—Wide width of the cotton web in mm;

[0026] S1—Maximum horizontal viewing area of ​​a single image acquisition module (mm);

[0027] By optimizing the field of view and exposure frequency of the image acquisition module, continuous and full-coverage detection in the longitudinal direction is achieved. The formula for calculating the minimum exposure frequency of the image acquisition module is as follows:

[0028]

[0029] in:

[0030] V – Minimum speed of the cotton web (mm / s);

[0031] S2 – Maximum vertical field of view of a single area array image sensor (mm).

[0032] A further improvement to the above technical solution: the human-computer interaction interface adopts a touch screen.

[0033] The present invention provides a monitoring method for the above-mentioned online image monitoring system for a carding machine, characterized in that the monitoring method includes a cotton nip detection method, which includes the following steps:

[0034] Step 1: Acquire image data of the cotton web from the carding machine using several image acquisition modules;

[0035] Step 2: The image preprocessing module uses image enhancement technology to denoise, filter, and sharpen the acquired cotton web image data to improve image quality;

[0036] Step 3: Using image enhancement technology through the image feature extraction module, histogram equalization and edge detection algorithms are applied to the denoised and sharpened cotton web image data;

[0037] Step 4: Image recognition and classification are performed using segmentation technology through the classification and training module; Step 5: Image compression technology and image processing algorithms are used through the cotton knot size calculation module.

[0038] Calculation of cotton knot dimensions;

[0039] Step 6: Finally, calculate the number of cotton knots using the cotton knot statistics module.

[0040] Improvements to the above technical solution: In step 1, an on-chip switchable 12 / 10-bit digital-to-analog converter, an image signal processor, and a MIPI CSI-2 four-channel high-speed interface are integrated to achieve real-time image acquisition, processing, and output at full resolution of at least 60 fps; by enhancing the light flux collection capability under low illumination, high-precision identification and dynamic detection of tiny cotton knots in a wide-area cotton web are achieved; an array of area-array image sensors is arranged to synchronously acquire images of the entire cotton web;

[0041] In step 2, the weighted grayscale strategy is optimized to enhance edge texture details and compress the image data by at least 2 / 3, thereby improving storage and transmission efficiency, in response to the weak contrast characteristics of cotton knots. Gaussian filtering is optimized in combination with the texture characteristics of cotton web to achieve a balance between noise suppression and structure preservation, thereby improving the edge clarity and detection stability of cotton knots under high-speed imaging. The image pixel height is reduced and the influence of image noise is removed.

[0042] In step 3, an RFE structure is introduced to replace the original model's Bottleneck, enhancing the model's multi-scale context modeling capability and strengthening the expression of texture and edge features of tiny cotton knots, thereby improving the recognition accuracy of cotton knot detection; a channel and spatial dual attention mechanism is constructed to adaptively emphasize key feature regions, thereby enhancing the target features of cotton knots and improving detection accuracy.

[0043] In step 4, a high-speed and high-precision detection network for tiny cotton knots based on an attention mechanism is used; a cotton knot classification system is established by combining cotton knot size, morphological characteristics and generation mechanism; a large-scale cotton knot sample library including interferences such as short fibers and false cotton knots is constructed; and cotton knot samples are trained based on the database.

[0044] In step 5, a feature pyramid attention mechanism is introduced to optimize the upsampling network module; a deep neural network is trained using cotton knot image data to construct a cotton knot size evaluation algorithm; and the cotton knot size is calculated by combining the pixel ratio of the detected cotton knot sample to the image sample.

[0045] In step 6, the camera is centrally managed and accessed through a human-computer interaction interface, and key yarn data is displayed synchronously in real time. A mechanism for on-demand retrieval of camera images is constructed, and the real-time images and detection results of any camera are quickly switched through the display / hide control and RTSP list selection function, thereby enhancing the operability and debugging flexibility of multi-channel online detection. The size and number of cotton knots are recorded through the home page of the human-computer interaction interface.

[0046] Further improvement to the above technical solution: The carding machine is a carding machine, and the monitoring method includes a carding machine control method for the cotton nip detection process, the carding machine control method including the following steps:

[0047] Step S1: Power on the carding machine and the online image monitoring device;

[0048] Step S2: Set the cotton knot threshold in the human-computer interaction interface;

[0049] Step S3: The control system starts the carding machine;

[0050] Step S4: Dolph begins running;

[0051] Step S5: When the french reaches high speed, the control system starts the online image monitoring device;

[0052] Step S6: The online image monitoring device begins to acquire and process images of the cotton web output from the carding machine;

[0053] Step S7: If the number of detected cotton knots is greater than or equal to the set cotton knot threshold, the control system will control the doffer to stop running.

[0054] Step S8: If the number of detected cotton knots is less than the set cotton knot threshold, the control system controls the online image monitoring device to display the number of cotton knots through the human-machine interface.

[0055] Compared with the prior art, the present invention has the following advantages and positive effects:

[0056] 1. Advantageous installation location and improved system stability: The image sensor array is fixedly installed, and the selected location is free from interference from other factors, which fundamentally improves the signal-to-noise ratio and detection accuracy of the image, and effectively enhances the mechanical performance and electrical stability of the system.

[0057] 2. Achieve full coverage detection: Through the innovative hardware array design proposed in this invention, it is possible to achieve all-round online monitoring of cotton web without blind spots, which can greatly reduce the missed detection rate;

[0058] 3. Improved data accuracy: The system performs full-coverage detection and maintains synchronous data acquisition. Based on image optimization and evaluation algorithm research, it achieves high-precision data detection.

[0059] 4. Enhanced detection efficiency: The use of a dense array of image sensors and convolutional neural network algorithms enables the system to quickly and automatically detect cotton knots and accurately identify their characteristics;

[0060] 5. Enhanced compatibility: It can be adapted to different models of carding machines, improving the equipment's versatility and practicality.

[0061] In summary, the present invention has a reasonable and compact structural design, a low failure rate, and can achieve seamless full-coverage detection of cotton web in both the transverse and longitudinal directions, while improving reliability, stability, detection accuracy, and detection efficiency. Attached Figure Description

[0062] Figure 1 This is a schematic diagram of the installation structure of an aluminum shell module in an online image monitoring system for a combing machine according to the present invention;

[0063] Figure 2 This is a three-dimensional view of the U-shaped elongated aluminum shell of the aluminum shell module in the online image monitoring system for a combing machine according to the present invention;

[0064] Figure 3 This is a schematic diagram of an array of image acquisition modules within an aluminum shell module in an online image monitoring system for a combing machine according to the present invention.

[0065] Figure 4 This is a schematic diagram of the intelligent cotton nip detection system based on the attention mechanism in the online image monitoring system of a carding machine according to the present invention;

[0066] Figure 5 This is a block diagram of the online image monitoring system for a carding machine, which is used to start and participate in the control logic of the carding machine system according to the present invention.

[0067] Figure 6 This is a flowchart of an algorithm for acquiring, lightweighting, detecting, and evaluating cotton web images in an online image monitoring system for a carding machine according to the present invention.

[0068] Figure 7 This is a block diagram showing the relationship between various functional modules in an online image monitoring system for a combing machine according to the present invention.

[0069] In the picture, 1. U-shaped elongated aluminum shell; 2. Camera; 3. LED lighting module; 4. Transparent window; 5. Cotton net. Detailed Implementation

[0070] The present invention will now be described in further detail with reference to the accompanying drawings.

[0071] See Figures 1-7An embodiment of the online image monitoring system for a carding machine according to the present invention includes a carding machine, an online image monitoring device, and a control system. The online image monitoring device includes several image acquisition modules. The control system includes a human-computer interaction interface, an image preprocessing module, an image feature extraction module, a classification and training module, a nip size calculation module, and a nip statistics module. Several image acquisition modules are fixedly arranged in an array near the front of the cotton web 5. The cotton web dynamically output by the carding machine is detected synchronously by each image acquisition module, realizing full-coverage cotton web image data acquisition without blind spots. The control system performs image preprocessing, image feature extraction, classification and training, nip size calculation, and nip statistics through its functional modules based on the acquired cotton web image data, realizing rapid detection and evaluation of the nipples in the cotton web 5.

[0072] Furthermore, several image acquisition modules are arranged horizontally and fixed inside a U-shaped elongated aluminum shell 1 to form an aluminum shell module. A transparent window is provided in front of the aluminum shell module. The aluminum shell module is horizontally positioned in front of the cotton web 5 to be tested. The length of the aluminum shell module is greater than the horizontal width of the cotton web 5 to be tested. An LED lighting module is integrated inside the aluminum shell module and connected by a flexible circuit board and modular cables. The transparent window 4 and the LED lighting module 3 face the cotton web 5 to be tested.

[0073] Specifically: The aforementioned carding machine is a carding machine, which includes a doffer and a stripping roller. A frame beam is installed below the stripping roller and directly in front of the doffer. Guide rails are mounted on the frame beam. A slider is located below the aforementioned aluminum shell module. The slider slides horizontally along the guide rails. During installation, the aluminum shell module is inserted into the gap and fixed by the slider sliding along the guide rails. After the aluminum shell module is installed, it can be used for real-time monitoring of the cotton web 5 and the doffer carding cloth.

[0074] Furthermore, the aforementioned image acquisition module includes a mounting base and camera 2. Camera 2 includes a back-illuminated CMOS image sensor and a wide-angle short-range lens. Each image acquisition module occupies a three-dimensional space dimension of no more than 30×30×20mm. The distance between the transparent window 4 in front of the aluminum shell module and the cotton net 5 to be measured is no more than 3mm. The number of cameras 2 is usually around 20.

[0075] Furthermore, the formula for calculating the minimum number of image acquisition modules to form an array combination is:

[0076]

[0077] in:

[0078] L—Wide width of the cotton web in mm;

[0079] S1—Maximum horizontal viewing area of ​​a single image acquisition module (mm);

[0080] By optimizing the field of view and exposure frequency of the image acquisition module, continuous and full-coverage detection in the longitudinal direction is achieved. The formula for calculating the minimum exposure frequency of the image acquisition module is as follows:

[0081]

[0082] in:

[0083] V – Minimum speed of the cotton web (mm / s);

[0084] S2 – Maximum vertical field of view of a single area array image sensor (mm).

[0085] The imaging principle of the above image acquisition module:

[0086] 1) Each image acquisition module is responsible for capturing images of the corresponding field of view area;

[0087] 2) Through a single synchronous trigger, all image acquisition modules are exposed simultaneously, capturing the cotton web image from their own perspective;

[0088] 3) Maintain synchronization and calibration between multiple image acquisition modules.

[0089] After calculating the number of image acquisition modules based on the width of the combing machine, the image acquisition modules can be arranged within a closed module as follows: Figure 3 The array arrangement shown allows the image acquisition module to achieve full coverage detection of cotton knots.

[0090] Furthermore, the aforementioned human-computer interaction interface uses a touch screen.

[0091] During the specific installation and fabrication, the image acquisition modules are arranged in an array and fixed inside a U-shaped long aluminum shell 1 to form an aluminum shell module. The aluminum shell material can be aluminum alloy to meet the requirements of strength and heat dissipation. It is processed by bending and cutting, which is simple and quick and avoids the need for mold making. The aluminum shell module integrates an LED lighting module 3 to ensure a stable detection environment. At the same time, modular cable connection is adopted, and flexible circuit boards (FPC) are used to reduce cable bending. The key signal lines adopt a redundant design to improve system reliability.

[0092] See Figures 1-7 An embodiment of the monitoring method of the above-mentioned online image monitoring system for carding machines according to the present invention includes a cotton nip detection method, which includes the following steps:

[0093] Step 1: Acquire image data of the cotton web 5 of the carding machine through several image acquisition modules;

[0094] Step 2: Using image enhancement technology through the image preprocessing module, the acquired cotton web 5 image data is denoised, filtered, and sharpened to improve image quality;

[0095] Step 3: Using image enhancement technology through the image feature extraction module, histogram equalization and edge detection algorithms are applied to the denoised and sharpened cotton web image data;

[0096] Step 4: Image recognition and classification are performed using segmentation technology through the classification and training module; Step 5: Image compression technology and image processing algorithms are used through the cotton knot size calculation module.

[0097] Calculation of cotton knot dimensions;

[0098] Step 6: Finally, calculate the number of cotton knots using the cotton knot statistics module.

[0099] Furthermore, in step 1 above, an on-chip switchable 12 / 10-bit digital-to-analog converter, an image signal processor, and a MIPI CSI-2 four-channel high-speed interface are integrated to achieve real-time image acquisition, processing, and output at full resolution of 60 fps; by enhancing the light flux collection capability under low illumination, high-precision identification and dynamic detection of tiny cotton knots in the wide-format cotton web 5 are achieved; and an array of area array image sensors are arranged to synchronously acquire images of the entire cotton web 5.

[0100] In step 2 above, the weighted grayscale strategy is optimized to enhance edge texture details and compress the image data by at least 2 / 3, thereby improving storage and transmission efficiency, in response to the weak contrast characteristics of cotton knots. Gaussian filtering is optimized in combination with the texture characteristics of cotton web 5 to achieve a balance between noise suppression and structure preservation, thereby improving the edge clarity and detection stability of cotton knots under high-speed imaging. The image pixel height is reduced and the influence of image noise is removed.

[0101] In step 3 above, an RFE structure is introduced to replace the original model's Bottleneck structure, enhancing the model's multi-scale context modeling capability and strengthening the expression of texture and edge features of tiny cotton knots, thereby improving the recognition accuracy of cotton knot detection; a channel and spatial dual attention mechanism is constructed to adaptively emphasize key feature regions, thereby enhancing the target features of cotton knots and improving detection accuracy.

[0102] In step 4 above, a high-speed and high-precision detection network for tiny cotton knots based on an attention mechanism is used; a cotton knot classification system is established by combining cotton knot size, morphological characteristics and generation mechanism; a large-scale cotton knot sample library including interferences such as short fibers and false cotton knots is constructed; and cotton knot samples are trained based on the database.

[0103] In step 5 above, a feature pyramid attention mechanism is introduced to optimize the upsampling network module; a deep neural network is trained using cotton knot image data to construct a cotton knot size evaluation algorithm; and the cotton knot size is calculated by combining the pixel ratio of the detected cotton knot sample to the image sample.

[0104] In step 6 above, the camera 2 is centrally managed and accessed through a human-computer interaction interface, and key yarn data is displayed synchronously in real time. A mechanism for on-demand access to the camera 2 image is constructed, and the real-time image and detection results of any camera 2 can be quickly switched through the display / hide control and RTSP (Real-Time Streaming Protocol) list selection function, thereby enhancing the operability and debugging flexibility of multi-channel online detection. The size and number of cotton knots are recorded through the home page of the human-computer interaction interface.

[0105] Furthermore, the aforementioned carding machine is a carding machine, and the monitoring method also includes a carding machine control method for the cotton nip detection process, the carding machine control method including the following steps:

[0106] Step S1: Power on the carding machine and the online image monitoring device;

[0107] Step S2: Set the cotton knot threshold in the human-computer interaction interface;

[0108] Step S3: The control system starts the carding machine;

[0109] Step S4: Dolph begins running;

[0110] Step S5: When the french reaches high speed, the control system starts the online image monitoring device;

[0111] Step S6: The online image monitoring device begins to acquire and process images of the cotton web 5 output from the carding machine;

[0112] Step S7: If the number of detected cotton knots is greater than or equal to the set cotton knot threshold, the control system will control the doffer to stop running.

[0113] Step S8: If the number of detected cotton knots is less than the set cotton knot threshold, the control system controls the online image monitoring device to display the number of cotton knots through the human-machine interface.

[0114] The main innovation of this invention lies in the design of the online image monitoring device. The carding machine body in the online image monitoring system of the carding machine production process of this invention can be based on existing technology, so it will not be described in detail here.

[0115] For specific implementation, please refer to Figure 4 The following methods are used for rapid image processing and optimization:

[0116] To address the computational demands of image optimization, lightweight optimization algorithms were researched. Rapid image optimization was achieved through model parameter tuning and pruning. Lightweight cotton web images were implemented through grayscale and filtering processes, reducing storage requirements while smoothing the image, reducing noise, and outputting high-quality cotton web images. Parameters were adjusted to fully consider the characteristics of different hardware platforms, ensuring stable operation and maintaining good performance. An IPU / IVE parallel processing architecture and pipeline design were implemented, as shown in the figure below. Multiple frames were merged into a single batch, reducing model call overhead and improving GPU throughput. CUDA streaming was used to overlap with data transmission, reducing idle waiting time. Through parallel and pipelined design, parallel processing and transmission of real-time images from multiple sensor arrays can be achieved.

[0117] The following high-precision detection and evaluation algorithm is implemented:

[0118] To address the challenges of complex and varied characteristics of cotton knots, a high-speed, high-precision detection network for tiny cotton knots based on an attention mechanism is proposed, enabling high-precision measurement of cotton knots and improving the accuracy of cotton knot identification.

[0119] By constructing a convolutional neural network algorithm for cotton nip detection based on YOLOv5s, leveraging its excellent feature extraction capabilities for small targets, a solid foundation is provided for cotton nip height detection. Multi-dimensional optimizations enhance the accuracy of small target detection. Specifically, the CBAM (Convolutional Attention Module) attention mechanism is introduced into the Neck structure of YOLOv5s, which suppresses interference from irrelevant information, retains more key features of the target to be detected, and allows the network to focus on more small targets, thus improving detection accuracy. To further improve the model's detection accuracy while maintaining detection speed, the RFE (Feature Enhancement Module) is introduced into the C3 module. Through specific structural design, the RFE expands the receptive field of the convolutional kernel, forming a new C3RFE module, which replaces the last C3 module in the Backbone section. This enhances the feature representation of small targets such as cotton nipples, helping the model capture more contextual information from the input image and further enhancing features, thereby improving the model's accuracy in detecting and recognizing cotton nipples in the cotton web of the carding machine. Finally, by introducing channel attention and spatial attention mechanisms, the detection features are further enhanced. The channel attention mechanism allows the model to quickly focus on the knot information in different channels. The spatial attention mechanism can quickly focus on the specific spatial location of the knot in the image. The combination of the two can significantly improve the detection accuracy, achieving high-precision detection and improved accuracy of knot recognition.

[0120] Of course, the above description is not intended to limit the present invention, and the present invention is not limited to the examples given above. Any changes, modifications, additions or substitutions made by those skilled in the art within the scope of the present invention should also fall within the protection scope of the present invention.

Claims

1. A carding machine image online monitoring system, comprising a carding machine, an image online monitoring device, and a control system, characterized in that, The online image monitoring device includes several image acquisition modules, and the control system includes a human-computer interaction interface, an image preprocessing module, an image feature extraction module, a classification and training module, a cotton nip size calculation module, and a cotton nip statistics module. Several image acquisition modules are fixedly arranged in an array near the front of the cotton web. The cotton web dynamically output by the carding machine is detected synchronously by each image acquisition module, realizing full-coverage cotton web image data acquisition without blind spots. The control system performs image preprocessing, image feature extraction, classification and training, cotton nip size calculation, and cotton nip statistics through its functional modules based on the acquired cotton web image data, realizing rapid detection and evaluation of cotton nipples in the cotton web.

2. The online image monitoring system for a combing machine according to claim 1, characterized in that, Several image acquisition modules are arranged horizontally and fixed inside a U-shaped elongated aluminum shell to form an aluminum shell module. A transparent window is provided in front of the aluminum shell module. The aluminum shell module is horizontally positioned in front of the cotton web to be tested. The length of the aluminum shell module is greater than the horizontal width of the cotton web to be tested. An LED lighting module is integrated inside the aluminum shell module and connected by a flexible circuit board and modular cables. The transparent window and the LED lighting module face the cotton web to be tested.

3. The online image monitoring system for a combing machine according to claim 2, characterized in that, The carding machine is a carding machine, which includes a doffer and a stripping roller. A frame beam is set below the stripping roller and directly in front of the doffer. A guide rail is installed on the frame beam. A slider is set below the aluminum shell module. The slider slides horizontally on the guide rail. During installation, the aluminum shell module is inserted into the gap and fixed by sliding along the guide rail with the slider.

4. The online image monitoring system for a combing machine according to claim 2 or 3, characterized in that, The image acquisition module includes a mounting base and a camera. The camera includes a back-illuminated CMOS image sensor and a wide-angle short-range lens. Each image acquisition module occupies a three-dimensional space dimension of no more than 30×30×20mm. The distance between the transparent window in front of the aluminum shell module and the cotton net to be tested is no more than 3mm.

5. The online image monitoring system for a combing machine according to any one of claims 1-3, characterized in that, The minimum number of image acquisition modules required to form an array combination is calculated using the following formula: ； in: L—Wide width of the cotton web in mm; S1—Maximum horizontal viewing area of ​​a single image acquisition module (mm); By optimizing the field of view and exposure frequency of the image acquisition module, continuous and full-coverage detection in the longitudinal direction is achieved. The formula for calculating the minimum exposure frequency of the image acquisition module is as follows: ； in: V – Minimum speed of the cotton web (mm / s); S2 – Maximum vertical viewing area of ​​a single area array image sensor (mm).

6. The online image monitoring system for a combing machine according to any one of claims 1-3, characterized in that, The human-computer interaction interface uses a touch screen.

7. The online image monitoring system for a combing machine according to claim 5, characterized in that, The human-computer interaction interface uses a touch screen.

8. A monitoring method for an online image monitoring system for a combing machine as described in any one of claims 1-7, characterized in that, The monitoring method includes a cotton nip detection method, which includes the following steps: Step 1: Acquire image data of the cotton web from the carding machine using several image acquisition modules; Step 2: The image preprocessing module uses image enhancement technology to denoise, filter, and sharpen the acquired cotton web image data to improve image quality; Step 3: Using image enhancement technology through the image feature extraction module, histogram equalization and edge detection algorithms are applied to the denoised and sharpened cotton web image data; Step 4: Image recognition and classification are performed using segmentation technology through the classification and training module; Step 5: Image compression technology and image processing algorithms are used through the cotton knot size calculation module. Calculation of cotton knot dimensions; Step 6: Finally, calculate the number of cotton knots using the cotton knot statistics module.

9. The monitoring method of the online image monitoring system for a combing machine according to claim 8, characterized in that, In step 1, an on-chip switchable 12 / 10-bit digital-to-analog converter, an image signal processor, and a MIPI CSI-2 four-channel high-speed interface are integrated to achieve real-time image acquisition, processing, and output at full resolution of at least 60 fps; by enhancing the light flux collection capability under low illumination, high-precision identification and dynamic detection of tiny cotton knots in a wide-area cotton web are achieved; an array of area array image sensors are arranged to synchronously acquire images of the entire cotton web; In step 2, the weighted grayscale strategy is optimized to enhance edge texture details and compress the image data by at least 2 / 3, thereby improving storage and transmission efficiency, in response to the weak contrast characteristics of cotton knots. Gaussian filtering is optimized in combination with the texture characteristics of cotton web to achieve a balance between noise suppression and structure preservation, thereby improving the edge clarity and detection stability of cotton knots under high-speed imaging. The image pixel height is reduced and the influence of image noise is removed. In step 3, an RFE structure is introduced to replace the original model's Bottleneck, enhancing the model's multi-scale context modeling capability and strengthening the expression of texture and edge features of tiny cotton knots, thereby improving the recognition accuracy of cotton knot detection; a channel and spatial dual attention mechanism is constructed to adaptively emphasize key feature regions, thereby enhancing the target features of cotton knots and improving detection accuracy. In step 4, a high-speed and high-precision detection network for tiny cotton knots based on an attention mechanism is used; a cotton knot classification system is established by combining cotton knot size, morphological characteristics and generation mechanism; a large-scale cotton knot sample library including interferences such as short fibers and false cotton knots is constructed; and cotton knot samples are trained based on the database. In step 5, a feature pyramid attention mechanism is introduced to optimize the upsampling network module; a deep neural network is trained using cotton knot image data to construct a cotton knot size evaluation algorithm; and the cotton knot size is calculated by combining the pixel ratio of the detected cotton knot sample to the image sample. In step 6, the camera is centrally managed and accessed through a human-computer interaction interface, and key yarn data is displayed synchronously in real time. A mechanism for on-demand retrieval of camera images is constructed, and the real-time images and detection results of any camera are quickly switched through the display / hide control and RTSP list selection function, thereby enhancing the operability and debugging flexibility of multi-channel online detection. The size and number of cotton knots are recorded through the home page of the human-computer interaction interface.

10. The monitoring method of the online image monitoring system for a combing machine according to claim 8 or 9, characterized in that, The carding machine is a carding machine, and the monitoring method includes a carding machine control method for the cotton nip detection process. The carding machine control method includes the following steps: Step S1: Power on the carding machine and the online image monitoring device; Step S2: Set the cotton knot threshold in the human-computer interaction interface; Step S3: The control system starts the carding machine; Step S4: Dolph begins running; Step S5: When the french reaches high speed, the control system starts the online image monitoring device; Step S6: The online image monitoring device begins to acquire and process images of the cotton web output from the carding machine; Step S7: If the number of detected cotton knots is greater than or equal to the set cotton knot threshold, the control system will control the doffer to stop running. Step S8: If the number of detected cotton knots is less than the set cotton knot threshold, the control system controls the online image monitoring device to display the number of cotton knots through the human-machine interface.

Images