Waste fabric classification system based on machine vision and artificial intelligence

Abstract

The invention relates to the field of waste fabric recycling, and discloses a waste fabric classification system based on machine vision and artificial intelligence, which comprises a transmission subsystem, a hyperspectral image acquisition subsystem, a sorting execution subsystem and a control and processing subsystem which are sequentially arranged along a waste fabric conveying path. A hyperspectral image processing algorithm based on the combination of K-means clustering analysis and statistical filtering is adopted, a single-band image is segmented by setting a large number of clustering centers, spectral values of the clustering centers are sequenced, and two ends are cut off and removed. The system can effectively filter out abnormal spectral data generated by wrinkle shadow, highlight reflection and background noise on the surface of the fabric, and ensures that the finally extracted material characterization vector mainly originates from effective spectral response of a fabric body, so that high-purity feature data can be obtained under complex working conditions; and the matching result with the expert rule base is more accurate.

Description

Technical Field

[0001] This invention relates to the field of waste textile recycling technology, specifically a waste textile classification system based on machine vision and artificial intelligence. Background Technology

[0002] With the development of the textile industry and the improvement of residents' consumption levels, the amount of waste textiles generated has increased dramatically year by year. Effective material classification of waste textiles (such as distinguishing between pure cotton, polyester, and blends) is a prerequisite for achieving high-value recycling. Currently, waste textile sorting still mainly relies on manual labor. This method is not only labor-intensive and unhygienic, but its efficiency and accuracy are also limited by the experience and fatigue level of the operators, making it difficult to meet the needs of large-scale industrial processing. Although some companies have begun to try to introduce automated sorting equipment, many technical bottlenecks still exist in practical applications.

[0003] Existing machine vision-based material recognition technologies suffer from insufficient anti-interference capabilities when processing soft materials such as fabrics. Waste fabrics are prone to wrinkles, overlaps, or distortions during transport, resulting in shadowed or highly reflective areas in the image. Traditional image processing algorithms often employ simple spectral averaging or single-point sampling methods, which struggle to effectively separate the fabric's spectral information from shadows, reflections, and background noise. This signal mixing directly leads to inaccurate material feature extraction, causing a significant drop in recognition accuracy when dealing with old fabrics with complex surface conditions.

[0004] In terms of sorting execution and system control, existing equipment typically lacks high-precision timing coordination and automated linkage capabilities. Many devices can only perform simple, single actions and lack precise control logic for scenarios involving continuous high-speed transport of multiple types of materials. When conveyor belt speed fluctuates or material delivery intervals are uneven, the actuators often fail to act at the optimal time, leading to sorting failures or material jams, severely restricting the overall throughput and operational efficiency of the production line.

[0005] Furthermore, most existing sorting systems employ open-loop control, lacking real-time monitoring and feedback mechanisms for sorting results. The systems typically cannot detect missed sorting or mechanical malfunctions at the end of the process. When push rod failures or identification errors result in unsorted materials, these waste materials often slide directly to the end of the equipment, accumulating and even causing mechanical blockages. Due to the lack of automated anomaly detection and alarm mechanisms, operators struggle to identify equipment malfunctions promptly, leading to prolonged operation of the production line with inherent defects, thus reducing the overall stability and maintenance efficiency of the system. Summary of the Invention

[0006] To address the shortcomings of existing technologies, this invention provides a waste textile classification system based on machine vision and artificial intelligence, which solves the problems mentioned in the background section.

[0007] To achieve the above objectives, the present invention is implemented through the following technical solution: a waste textile classification system based on machine vision and artificial intelligence, comprising a transmission subsystem, a hyperspectral image acquisition subsystem, a sorting execution subsystem, and a control and processing subsystem arranged sequentially along the waste textile conveying path; The transmission subsystem is used to carry and transport waste textiles to be sorted; The hyperspectral image acquisition subsystem is located on the path of the transmission subsystem and is used to acquire hyperspectral image data of waste fabrics. The sorting execution subsystem is located downstream of the hyperspectral image acquisition subsystem and is used to perform physical sorting actions based on the material identification results. The control and processing subsystem includes a central control computer and a control unit that are interconnected. The central control computer is connected to the hyperspectral image acquisition subsystem and is used to receive image data and perform material recognition calculations. The control unit is electrically connected to the transmission subsystem, the hyperspectral image acquisition subsystem and the sorting execution subsystem respectively, and is used to perform hardware input and output control. The central control computer generates sorting control instructions based on the material identification results and sends them to the control unit, which then drives the sorting execution subsystem to perform its actions.

[0008] Preferably, the transmission subsystem includes a conveyor belt, a conveyor belt drive motor, and a frequency converter; The conveyor belt forms a material transport plane; the control unit is electrically connected to the conveyor belt drive motor and the frequency converter respectively; The central control computer sends speed commands to the frequency converter through the control unit to adjust the linear speed of the conveyor belt; The waste fabrics are placed in a single row on the conveyor belt, with a fixed physical distance between adjacent waste fabrics. The value of the distance depends on the acquisition time of a single frame image by the hyperspectral image acquisition subsystem and the computation time required by the central control computer to process a single material identification algorithm.

[0009] Preferably, the hyperspectral image acquisition subsystem includes a hyperspectral image acquisition box, and a hyperspectral camera, an illumination source, and a laser detection sensor installed inside the hyperspectral image acquisition box; The hyperspectral image acquisition box is positioned above the conveyor belt; the type of illumination source includes a strip light source, a ring light source, or a planar light source. The laser detection sensor is signal-connected to the control unit and is used to detect whether waste fabric has entered the collection area; The central control computer is equipped with camera triggering logic, which is used to control the hyperspectral camera to execute exposure acquisition commands after receiving the trigger signal from the laser detection sensor.

[0010] Preferably, the sorting execution subsystem includes several sorting and collection units arranged along the conveying direction; Each sorting and collection unit includes a collection bin, a push rod, a push rod assist mechanism, and a baffle; the push rod assist mechanism is electrically connected to the control unit. The push rod is mechanically connected to the output end of the push rod assist mechanism; The baffle is positioned between two adjacent sorting and collection units; The push rod assist mechanism is used to drive the push rod to move along a path perpendicular to the conveying direction, pushing the waste fabric into the corresponding collection bucket.

[0011] Preferably, the sorting execution subsystem further includes an end collection unit disposed in the end area of ​​the conveyor belt; The end collection unit includes an end laser detection sensor, an end collection bucket, an end push rod assist mechanism, and an end push rod. The end laser detection sensor is connected to the control unit and is used to monitor the waste fabrics that arrive at the end of the conveyor belt; When the end laser detection sensor detects the passage of material, the control unit controls the end push rod assist mechanism to start, driving the end push rod to push the waste fabric into the end collection bucket.

[0012] Preferably, the central control computer has an image processing module deployed therein. This image processing module is used to perform cluster analysis on the acquired hyperspectral image data, specifically including: The central control computer stores the acquired hyperspectral image data in a three-dimensional matrix format; The central control computer performs K-means clustering operations on each of all bands; Before performing clustering operations, a cluster center number parameter is set, which is greater than the preset number of material types to be classified. For the current band being processed, a set consisting of multiple cluster centers is calculated.

[0013] Preferably, the central control computer is also used to perform statistical filtering and noise reduction processing on the clustering output data, specifically including: The central control computer acquires multiple cluster center spectral values ​​for the current band and sorts them in ascending order according to their numerical values. The central control computer removes the cluster centers with the smallest values ​​and the cluster centers with the largest values ​​after sorting, based on a preset cutoff threshold ratio parameter. The central control computer calculates the arithmetic mean of the spectral values ​​of the remaining cluster centers, and determines the arithmetic mean as the effective spectral value of the waste fabric in a specific band. The central control computer sequentially executes the above steps for all bands and constructs the arranged spectral values ​​as a material characterization vector for waste fabrics.

[0014] Preferably, the central control computer has a pre-installed hyperspectral image expert rule library in its internal memory, and the hyperspectral image expert rule library contains standard material vectors for various standard materials; The central control computer is used to perform the following material classification and determination steps: Calculate the feature distance between the material representation vector of the waste fabric and each standard material vector in the expert rule base; The feature distance is calculated using either the Euclidean distance algorithm or the cosine distance algorithm; all calculated distance values ​​are compared, and the minimum value is determined. The material category represented by the standard material vector corresponding to the minimum value is determined as the final material classification result of the current waste fabric.

[0015] Preferably, the control unit is used to perform time-lag-based trigger control on the push rod assist mechanism: The control unit starts timing after receiving the detection signal from the laser detection sensor; When the timing duration reaches the preset trigger delay time, the control unit outputs an action command to the push rod assist mechanism at the target position; The trigger delay time depends on the physical distance between the laser detection sensor and the target push rod assist mechanism in the direction of the conveyor belt's operation and the current operating speed of the conveyor belt.

[0016] Preferably, the central control computer is used to perform system performance evaluation and alarms: The central control computer is configured with statistical time window parameters and fault alarm threshold parameters. The central control computer collects the trigger signal of the end laser detection sensor in real time within the statistical time window parameter and maintains the abnormal collection count value; The central control computer compares the abnormal collection count value with the fault alarm threshold parameter in real time. When the abnormal collection count value is greater than the fault alarm threshold parameter, it determines that the system is in an abnormal working state and triggers the alarm program.

[0017] This invention provides a waste textile sorting system based on machine vision and artificial intelligence. It has the following beneficial effects: 1. This invention employs a hyperspectral image processing algorithm based on a combination of K-means clustering analysis and statistical filtering. By setting a large number of cluster centers to over-segment single-band images, and sorting and truncating the spectral values ​​of the cluster centers, the system can effectively filter out abnormal spectral data caused by fabric surface wrinkles, shadows, high light reflection, and background noise. This ensures that the final extracted material characterization vector mainly originates from the effective spectral response of the fabric itself, thereby obtaining high-purity feature data even under complex working conditions, making the matching results with the expert rule base more accurate.

[0018] 2. This invention establishes a precise timing control mechanism through the coordinated operation of a central control computer, control unit, and multi-stage pusher assist mechanism. Based on the trigger signal from the laser detection sensor, the real-time speed of the conveyor belt, and the physical distance between each sorting unit, the system automatically calculates and executes millisecond-level action delays. This time-lag-based control logic ensures that the pushers can move accurately the instant the fabric passes through the sorting port at high speed. Combined with multi-stage serial sorting and collection units, it enables one-time streamlined sorting of fabrics of various materials, significantly reducing the labor intensity of manual sorting and increasing throughput.

[0019] 3. This invention features a remedial mechanism at the end of the conveyor belt, consisting of an end laser detection sensor and an end pusher. This mechanism automatically cleans up abnormal materials that were not successfully sorted by the preceding unit, effectively preventing the risk of blockage caused by material accumulation at the tail of the equipment. At the same time, the system uses the trigger data from the end sensor to establish a closed-loop performance evaluation logic. By statistically analyzing the number of abnormal material drops within a certain time window and comparing it with the alarm threshold, the system can automatically determine whether the front-end camera or pusher mechanism has malfunctioned and issue a maintenance alarm in a timely manner, ensuring the stable operation of the system over a long period of time. Attached Figure Description

[0020] Figure 1 This is a schematic diagram of the system layout of the present invention; Figure 2 This is a schematic diagram of the process of the present invention.

[0021] 1. Conveyor belt driven wheel; 2. Conveyor belt; 3. Waste fabric; 4. Hyperspectral image acquisition box; 5. Hyperspectral camera; 6. Light source; 7. Central control computer; 8. Control unit; 9. Laser detection sensor; 10. Conveyor belt drive motor; 11. Frequency converter; 12. Collection bin; 13. Push rod; 14. Push rod assist mechanism; 15. Baffle; 16. End laser detection sensor; 17. End collection bin; 18. End push rod assist mechanism; 19. End push rod; Detailed Implementation

[0022] 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.

[0023] Example: Please see the appendix Figure 1 This invention provides a waste textile sorting system based on machine vision and artificial intelligence, including a transmission subsystem, an image acquisition subsystem, a control and processing subsystem, and a sorting execution subsystem.

[0024] The conveying subsystem is used to carry and transport the waste textiles 3 to be sorted. The conveying subsystem mainly includes a conveyor belt driven wheel 1, a conveyor belt 2, a central control computer 7, a control unit 8, a conveyor belt drive motor 10, and a frequency converter 11. The conveyor belt 2 is tensioned and installed on the transmission roller shaft formed by the conveyor belt drive mechanism and the conveyor belt driven wheel 1, forming a material conveying plane.

[0025] The central control computer 7, acting as the system's host computer, establishes a communication connection with the control unit 8. The control unit 8 is electrically connected to the conveyor belt drive motor 10 and the frequency converter 11, respectively, and is used to execute the underlying hardware control logic. The central control computer 7 is equipped with a control interface for sending start commands and speed parameters.

[0026] During system operation, the central control computer 7 sends a start signal to the conveyor belt drive motor 10 through the control unit 8. The conveyor belt drive motor 10 operates and drives the driven wheel 1 of the conveyor belt to rotate, thereby driving the conveyor belt 2 to run smoothly in a preset direction. The waste fabric 3 is placed on the surface of the conveyor belt 2 and moves to the subsequent workstation by friction.

[0027] The frequency converter 11 is used to adjust the output speed of the conveyor belt drive motor 10. The central control computer 7 sends speed commands to the frequency converter 11 through the control unit 8 according to the set operating parameters, thereby controlling the linear speed of the conveyor belt 2.

[0028] Multiple pieces of waste fabric 3 are placed sequentially in a single row on conveyor belt 2. A fixed physical interval is set between adjacent pieces of waste fabric 3. The value of the physical interval depends on the acquisition time of a single frame image by the hyperspectral camera 5 and the computation time required for the central control computer 7 to process a single material identification algorithm. During the system initialization phase, the above time parameters are determined based on the hardware performance indicators of the central control computer 7, and the running speed of the conveyor belt 2 and the placement interval of the waste fabric 3 are set accordingly.

[0029] The hyperspectral image acquisition subsystem involved in this invention is located on the path of the transmission subsystem and is used to acquire spectral image information of waste fabric 3. The image acquisition subsystem mainly includes a hyperspectral image acquisition box 4, a hyperspectral camera 5, an illumination source 6, and a laser detection sensor 9.

[0030] The hyperspectral image acquisition box 4 is positioned above the conveyor belt 2, forming a darkroom or controlled light environment for image acquisition. The hyperspectral camera 5, illumination source 6, and laser detection sensor 9 are installed inside the hyperspectral image acquisition box 4. The illumination source 6 provides stable lighting conditions for image acquisition; the type of illumination source 6 includes a strip light source, a ring light source, or a planar light source.

[0031] Laser detection sensor 9 is used to detect whether waste fabric 3 has entered the acquisition area of ​​hyperspectral image acquisition box 4. Laser detection sensor 9 adopts a through-beam installation structure and is respectively set on both sides of conveyor belt 2. The vertical installation height of laser detection sensor 9 is adjustable, adapting to the height position of conveyor belt 2 to ensure accurate detection of passing waste fabric 3.

[0032] The laser detection sensor 9 establishes a signal connection with the control unit 8, and the control unit 8 establishes a data communication connection with the central control computer 7. When the waste fabric 3 moves into the detection range with the conveyor belt 2, the laser detection sensor 9 detects the presence of the material and generates a trigger signal. The trigger signal is transmitted to the control unit 8, and the control unit 8 forwards the signal to the central control computer 7.

[0033] The central control computer 7 is equipped with camera triggering logic. After receiving a signal from the laser detection sensor 9, the central control computer 7 sends an exposure acquisition command to the hyperspectral camera 5. The timing of sending the exposure acquisition command is set to be sent immediately after receiving the signal or after a preset time delay. The hyperspectral camera 5 responds to the command to complete hyperspectral image acquisition and transmits the acquired image data to the central control computer 7 for storage and algorithm processing.

[0034] The sorting execution subsystem provided by this invention is arranged along the conveying direction of the conveyor belt 2, located downstream of the hyperspectral image acquisition box 4, and is used to perform physical sorting actions based on material identification results. The sorting execution subsystem mainly consists of several classification collection units and an end collection unit located at the end of the conveyor belt 2.

[0035] Each sorting and collection unit includes a collection bin 12, a push rod 13, a push rod assist mechanism 14, and a baffle 15. The collection bin 12 is located on one side of the conveyor belt 2 and is used to hold waste textiles 3 of a specific material type. The push rod assist mechanism 14 is located on the other side of the conveyor belt 2 opposite to the collection bin 12, corresponding to the position of the collection bin 12.

[0036] The push rod 13 is mechanically connected to the output end of the push rod assist mechanism 14. The push rod assist mechanism 14 is electrically connected to the control unit 8 to receive control commands. The push rod assist mechanism 14 is driven by a pneumatic or electric drive. Upon receiving a start signal, the push rod assist mechanism 14 drives the push rod 13 to move along a path perpendicular to the conveyor belt 2, pushing the waste fabric 3 on the conveyor belt 2 into the corresponding collection bin 12.

[0037] The number of sorting and collection units is set according to the number of different types of fabrics to be sorted. A baffle 15 is installed between two adjacent sorting and collection units. The baffle 15 serves as a physical barrier to prevent waste fabrics 3 from deviating from the path and entering adjacent collection bins 12 during the pushing process.

[0038] The end collection unit includes an end laser detection sensor 16, an end collection bucket 17, an end push rod assist mechanism 18, and an end push rod 19. The end laser detection sensor 16 is located in the end area of ​​the conveyor belt 2 and is installed in a through-beam structure. Its installation height is adjusted according to the height of the conveyor belt 2.

[0039] The end-of-line laser detection sensor 16 is used to monitor the waste fabric 3 arriving at the end of the conveyor belt 2. The end-of-line laser detection sensor 16 is connected to the control unit 8, and the end-of-line push rod assist mechanism 18 is also connected to the control unit 8. When the end-of-line laser detection sensor 16 detects material passing through, the control unit 8 controls the end-of-line push rod assist mechanism 18 to start, driving the end-of-line push rod 19 to push the waste fabric 3 into the end collection bin 17, preventing materials that have not been sorted by the front-end sorting and collection unit from sliding directly from the end of the conveyor belt 2.

[0040] The control and processing subsystem provided by this invention serves as the core of the entire system's operation and instruction hub. It mainly consists of a central control computer 7 and a control unit 8, which establish a bidirectional data connection through an industrial communication interface.

[0041] The central control computer 7 serves as the system's host computer, responsible for receiving, storing, processing, and interacting with the system, as well as handling image data. The central control computer 7 is connected to the hyperspectral camera 5 to receive raw hyperspectral image data. The central control computer 7 contains an image processing and pattern recognition software module, capable of performing clustering analysis, feature extraction, and expert rule base matching operations as described in subsequent embodiments, thereby outputting the material classification results of the waste fabric 3.

[0042] Control unit 8 serves as the lower-level controller of the system, performing input / output control of the underlying hardware. Control unit 8 establishes electrical connections with conveyor belt drive motor 10, frequency converter 11, laser detection sensor 9, push rod assist mechanism 14, end laser detection sensor 16, and end push rod assist mechanism 18, forming a complete hardware control loop.

[0043] In terms of signal acquisition, the control unit 8 is responsible for acquiring the detection signals from the laser detection sensor 9 and the end laser detection sensor 16 in real time, and feeding back the status information to the central control computer 7. The central control computer 7 determines the current operating stage of the system based on the received sensor signals, in order to trigger the image acquisition process or perform abnormal status monitoring.

[0044] In terms of instruction execution, the central control computer 7 generates corresponding sorting control instructions based on the material classification results and sends these instructions to the control unit 8. The control unit 8 parses the instructions and outputs a drive signal to the push rod assist mechanism 14 at the designated location, controlling the push rod 13 to perform an extension / retraction action. Simultaneously, the central control computer 7 performs system performance evaluation based on data fed back by the end-effector laser detection sensor 16, triggering an alarm procedure when it determines that the system is malfunctioning.

[0045] Please see the appendix Figure 2 This invention provides a method for classifying waste textiles based on machine vision and artificial intelligence, comprising the following steps: S1. System Start-up and Transmission: The central control computer 7 sends a start command to the conveyor belt drive motor 10 through the control unit 8. The conveyor belt drive motor 10 drives the driven wheel 1 of the conveyor belt to rotate, causing the conveyor belt 2 to run in a preset direction. Waste fabrics 3 are placed sequentially on the conveyor belt 2, with a preset physical interval maintained between adjacent waste fabrics 3, and move with the conveyor belt 2 towards the hyperspectral image acquisition box 4.

[0046] S2. Trigger Detection and Signal Transmission: When the waste fabric 3 enters the internal area of ​​the hyperspectral image acquisition box 4 along with the conveyor belt 2, the laser detection sensor 9 located at that position detects the presence of the material. The laser detection sensor 9 generates a detection signal and transmits it to the control unit 8, which then forwards the signal to the central control computer 7 in real time.

[0047] S3. Image Acquisition: After receiving the detection signal, the central control computer 7 immediately or after a preset delay sends an acquisition command to the hyperspectral camera 5 according to the preset control logic. The hyperspectral camera 5 exposes and acquires images of the current waste fabric 3 under the illumination of the lighting source 6. The acquired hyperspectral image data is transmitted to the central control computer 7 for storage via the data interface.

[0048] S4. Material Identification Calculation: The central control computer 7 calls the internal preset image processing algorithm to perform calculations on the acquired hyperspectral image data. The calculation process specifically performs band-dimensional clustering analysis and statistical filtering to extract spectral feature vectors, and compares the spectral feature vectors with the pre-stored expert rule base to determine the specific material category of the waste fabric 3.

[0049] S5. Sorting Execution: The central control computer 7 generates corresponding hardware control commands based on the material identification results and sends the commands to the control unit 8. After parsing the commands, the control unit 8 controls the push rod assist mechanism 14 corresponding to the identified material to start immediately or after a delay. The push rod assist mechanism 14 drives the push rod 13 to move, pushing the waste fabric 3 from the conveyor belt 2 into the corresponding collection bin 12.

[0050] S6. End-of-line monitoring and remediation: If the waste fabric 3 reaches the end of the conveyor belt 2 without being sorted by the preceding pusher assist mechanism 14, the end laser detection sensor 16 detects the material signal. After receiving the material signal, the control unit 8 drives the end pusher assist mechanism 18 to operate, pushing the waste fabric 3 into the end collection bin 17 through the end pusher 19, thus completing the collection of abnormal materials.

[0051] The principle of the present invention will be specifically explained below in conjunction with the above steps, specifically including: When the waste fabric 3 moves with the conveyor belt 2 and enters the detection range of the laser detection sensor 9, the laser detection sensor 9 detects the material signal. The laser detection sensor 9 transmits the detection signal to the control unit 8, and the control unit 8 forwards the material signal to the central control computer 7.

[0052] After receiving the signal from the laser detection sensor 9, the central control computer 7 executes the camera trigger control logic. Based on the physical positional relationship between the laser detection sensor 9 and the hyperspectral camera 5, the central control computer 7 controls the hyperspectral camera 5 to either immediately perform data acquisition or perform acquisition after a preset delay time. The preset delay time is set according to the running speed of the conveyor belt 2 to ensure that exposure occurs when the waste fabric 3 moves into the acquisition field of view of the hyperspectral camera 5.

[0053] The hyperspectral camera 5 executes the acquisition command to obtain hyperspectral image data of the waste fabric 3, and transmits it to the central control computer 7 via the data interface. The central control computer 7 stores the acquired image data as... It is in the form of a three-dimensional matrix.

[0054] in, This indicates the number of bands in the hyperspectral image. In this example, It is set to an odd number, which can include values ​​such as 5, 7, or 9. The number of rows indicating image resolution. The number of columns indicates the image resolution. Each data point in the 3D matrix represents the spectral response value of the waste fabric 3 at a specific spatial coordinate and a specific wavelength band, serving as the basis for subsequent material identification algorithms.

[0055] The format obtained from the central control computer 7 is... After obtaining the hyperspectral image data, perform the cluster analysis described in step S12.

[0056] The central control computer 7 performs band-by-band processing on the stored image data. For any one of the C bands, it extracts all the bands within that band. The spectral response value of each pixel. The central control computer 7... Using the spectral response values ​​as the input data set, K-means clustering operation is performed.

[0057] Before performing clustering operations, a parameter S is set to indicate the number of cluster centers. The value of parameter S is set to be greater than the preset number of material types to be classified. In this embodiment, parameter S is set to 500. Using a larger S value for image data segmentation ensures that fabric texture areas, background noise areas, and unevenly lit areas in the image are grouped into different clusters, thus preserving detailed information.

[0058] The K-means algorithm iteratively calculates the bands within the band. Each pixel is divided into S categories, and the center value of each category is calculated. For the currently processed band, a set consisting of S cluster centers is calculated. This set is represented as... ,in This represents the spectral response value of the i-th cluster center.

[0059] The central control computer 7 traverses all C bands, performs the above K-means clustering steps for each band, and outputs the S cluster center values ​​for each band as the basis for subsequent statistical analysis and feature extraction.

[0060] The material identification method provided by this invention performs statistical filtering and noise reduction processing on the cluster output data after completing the aforementioned single-band clustering analysis.

[0061] The central control computer 7 acquires the spectral values ​​of the S cluster centers for the current processing band, calculated in the cluster analysis step. The central control computer 7 then performs statistical filtering on the S cluster center spectral values ​​based on the normal distribution assumption.

[0062] Specifically, the central control computer 7 first sorts the S cluster center spectral values ​​in ascending order of their numerical values. The system has a preset truncation threshold ratio parameter t%. In this embodiment, the truncation threshold ratio parameter t is set to 10, representing a rejection ratio of 10%.

[0063] The central control computer 7, based on the data sequence sorted in ascending order, removes the smallest value from the first... The cluster centers and the one with the largest value. Cluster centers are identified. By eliminating cluster centers, abnormal spectral response interference caused by background noise, shadow occlusion, or specular reflection in the image is removed, thereby preserving valid data in the middle of the distribution.

[0064] Subsequently, the central control computer 7 calculates the arithmetic mean of the remaining cluster center spectral values. The number of remaining data is... The central control computer 7 determines the arithmetic mean as the effective fabric spectral value of the waste fabric 3 in that specific wavelength band.

[0065] The central control computer 7 sequentially performs the aforementioned clustering analysis and statistical filtering steps on all C bands until the effective fabric spectral values ​​of the waste fabric 3 in each band are obtained, thereby completing the extraction of spectral features across the entire band.

[0066] The material identification method provided by this invention constructs a material characterization vector after obtaining the effective spectral feature values ​​of all preset bands.

[0067] The central control computer 7 summarizes the calculations obtained in the aforementioned steps, applicable to all... The effective fabric spectral values ​​for each band. The central control computer 7 processes these values ​​according to the wavelength numerical order or the preset channel index order. The individual spectral values ​​are arranged and combined. The central control computer 7 constructs the arranged spectral values ​​into a material characterization vector for waste fabrics. Material characterization vector of waste fabric Represented as:

[0068] In the formula, The total number of bands set for the system; Indicates the first The effective fabric spectral values ​​corresponding to the i-th band, i.e., the values ​​for the i-th band in the aforementioned statistical filtering step. The arithmetic mean of the cluster centers calculated for each band. (Subscript) This is the band index, and its value range is... .

[0069] The material characterization vector of the waste fabric The spectral response characteristics of waste fabric 3 in the selected spectral frequency band are characterized in the form of a multi-dimensional vector. The central control computer 7 will then construct the vector... It is stored in memory and used as input data for subsequent material classification decision algorithms.

[0070] The material identification method provided by this invention completes the construction of the material representation vector of waste fabric. Then, material classification is performed using a pre-set hyperspectral image expert rule library. The central control computer 7 has a pre-installed hyperspectral image expert rule library in its internal memory. This library contains pre-calibrated... The spectral feature vectors of a standard material are obtained by performing hyperspectral acquisition, cluster analysis, and statistical calculations on standard samples of known material types. The vectors in the expert rule base are respectively labeled as... ,in This represents the total number of material types the system can recognize. Each standard material vector... All have a material characterization vector related to waste fabrics. Same dimensions The central control computer 7 calculates the current material representation vector of the waste fabric. With each standard material vector in the expert rule base Feature distance between Subscript The range of values ​​is The feature distance can be calculated using either the Euclidean distance algorithm or the cosine distance algorithm. In this embodiment, the Euclidean distance algorithm is used, and its calculation formula is as follows:

[0071] In the formula, Material characterization vector for waste fabrics The Middle Spectral values ​​of each band, For the first Standard material vectors The Middle The standard spectral values ​​for each band were obtained through a 7-step calculation by the central control computer. Each distance value is compared with the stated distance values. Use distance values ​​to determine the minimum value among them.

[0072] The central control computer 7 identifies the index corresponding to the minimum distance. The index The following conditions must be met:

[0073] Central control computer 7 will index Corresponding standard material vector The material category represented is determined as the final material classification result for the current waste fabric 3. The central control computer 7 uses the final material classification result as the basis for generating subsequent control instructions for the sorting execution subsystem.

[0074] The automated sorting control strategy provided by this invention generates corresponding hardware execution signals based on the material classification results of the waste fabric 3 determined in the aforementioned steps.

[0075] The central control computer 7 has a pre-installed mapping table between material categories and control commands in its memory. This mapping table defines the corresponding logic between each material category that the system can recognize and a specific control command parameter. For example, 100% cotton material is associated with the first control command, and 100% polyester material is associated with the second control command. There is a one-to-one correspondence between the control commands and the physical location or hardware interface address of the sorting and collection unit, and this correspondence is set according to the physical arrangement order of the field equipment.

[0076] Based on the final material classification results output by the expert rule base, the central control computer 7 queries the mapping relationship table and extracts the target control command corresponding to the current waste fabric 3. The central control computer 7 sends the target control command to the control unit 8 through the data communication interface.

[0077] The control unit 8 receives control commands from the central control computer 7. The processor inside the control unit 8 parses the control commands and identifies the hardware address code of the target push rod assist mechanism 14 that responds to the command. Based on the hardware address code, the control unit 8 determines the classification and collection unit to be activated, preparing for subsequent output drive signals.

[0078] Based on the target hardware address determined by parsing, control unit 8 locks the corresponding push rod assist mechanism 14. Control unit 8 performs time-delay-based trigger control on push rod assist mechanism 14.

[0079] Specifically, the system pre-sets or calculates the trigger delay time in real time. This delay time depends on the physical distance between the laser detection sensor 9 and the target push rod assist mechanism 14 in the running direction of the conveyor belt 2, as well as the current running linear speed of the conveyor belt 2. After receiving the material detection signal, the control unit 8 starts timing, and when the timing duration reaches the delay time, it outputs an action command to the push rod assist mechanism 14. The control logic ensures that when the push rod 13 moves, the waste fabric 3 moves exactly within the effective stroke range of the push rod 13.

[0080] Upon receiving the instruction, the push rod assist mechanism 14 drives the push rod 13 to extend rapidly along a path perpendicular to the running direction of the conveyor belt 2. The actuating end of the push rod 13 contacts the waste fabric 3 and applies a lateral thrust, forcing the waste fabric 3 to change its direction of movement, detach from the conveyor belt 2, and fall into the designated collection bin 12. After completing the pushing action, the push rod assist mechanism 14 drives the push rod 13 to automatically retract to the initial waiting position, preparing for the next sorting.

[0081] The push rod assist mechanism 14 can be either a pneumatic drive assembly or an electric drive assembly. If pneumatic drive is used, the control unit 8 controls the solenoid valve to open, driving the cylinder piston rod to extend; if electric drive is used, the control unit 8 controls the motor to drive the push rod to move through a mechanical transmission mechanism.

[0082] During the sorting process, the baffles 15 placed between adjacent collection units serve as physical guides and isolate the materials. When the waste fabric 3 is moved by force, the baffles 15 restrict its movement trajectory, preventing the material from drifting into adjacent non-target collection areas due to inertia, thereby ensuring the accuracy of sorting and discharging.

[0083] The end-of-line remediation mechanism provided by the present invention is set at the end of the conveyor belt 2, downstream of all sorting and collection units, and is used to process waste fabrics 3 that have not been sorted by the preceding sorting and collection units. It mainly consists of an end laser detection sensor 16, an end collection bucket 17, an end push rod assist mechanism 18, and an end push rod 19. The end laser detection sensor 16 is installed in the end detection area of ​​the conveyor belt 2, and the end collection bucket 17 is located on one side of the conveyor belt 2, with its position corresponding to the end point of the stroke of the end push rod 19.

[0084] During system operation, if the waste fabric 3 moves to the end area with the conveyor belt 2, the end laser detection sensor 16 detects a material obstruction signal. The end laser detection sensor 16 transmits the material obstruction signal to the control unit 8.

[0085] After receiving the signal from the end laser detection sensor 16, the control unit 8 executes the remedial control logic. The control unit 8 controls the end push rod assist mechanism 18 to start immediately or after a preset delay time to ensure that the push head of the end push rod 19 accurately acts on the waste fabric 3.

[0086] The end pusher assist mechanism 18 drives the end pusher 19 to extend along the direction perpendicular to the running direction of the conveyor belt 2, pushing the waste fabric 3 from the conveyor belt 2 into the end collection bin 17. After the pushing is completed, the end pusher assist mechanism 18 drives the end pusher 19 to automatically retract to the initial waiting position, ensuring that the material at the end of the conveyor belt 2 is cleared and preventing unsorted material from slipping or causing equipment blockage.

[0087] See attached document Figure 1 , Figure 1 This is a schematic diagram of the overall layout of an automatic waste textile material sorting system according to an embodiment of the present invention. The system performance evaluation and alarm algorithm provided by the present invention runs in the central control computer 7, and performs closed-loop monitoring of the overall sorting efficiency of the system based on the feedback signal from the end laser detection sensor 16.

[0088] The central control computer 7 is configured with a statistical time window parameter T. In this embodiment, the value of the time window parameter T is set to 30 minutes. The central control computer 7 also has a preset fault alarm threshold parameter φ, which is set to 20 in this embodiment.

[0089] During system operation, the central control computer 7 collects the trigger signals of the end-effector laser detection sensor 16 in real time, with a time window parameter T as one statistical period. The central control computer 7 internally maintains an anomaly collection count value Q, which is used to record the total number of times the end-effector laser detection sensor 16 is triggered within the current statistical period.

[0090] The central control computer 7 compares the abnormal collection count value Q with the fault alarm threshold parameter φ in real time: when Q > φ, the central control computer 7 determines that the system is in an abnormal working state. The central control computer 7 then triggers the alarm program, drives the alarm device to issue an audible and visual warning, or displays a maintenance prompt on the operation interface, instructing the operator to check whether the hyperspectral camera 5, the lighting source 6, or the various levels of push rod assist mechanism 14 at the front end are faulty.

[0091] When a statistical period T ends and Q ≤ φ, the central control computer 7 determines that the system has been working normally during that period. At this time, the central control computer 7 automatically performs a counter reset operation, setting Q = 0, and begins the statistics for the next time window T.

[0092] In addition, when the system is powered on again, or after an alarm is triggered and the fault is manually resolved, the central control computer 7 will force the counter initialization operation to set Q=0, ensuring that the system restarts the statistics and evaluation of performance data.

[0093] 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 waste textile sorting system based on machine vision and artificial intelligence, characterized in that, It includes a transmission subsystem, a hyperspectral image acquisition subsystem, a sorting execution subsystem, and a control and processing subsystem, which are arranged sequentially along the waste fabric conveying path; The transmission subsystem is used to carry and transport waste textiles to be sorted; The hyperspectral image acquisition subsystem is located on the path of the transmission subsystem and is used to acquire hyperspectral image data of waste fabrics. The sorting execution subsystem is located downstream of the hyperspectral image acquisition subsystem and is used to perform physical sorting actions based on the material identification results. The control and processing subsystem includes a central control computer and a control unit that are interconnected. The central control computer is connected to the hyperspectral image acquisition subsystem and is used to receive image data and perform material recognition calculations. The control unit is electrically connected to the transmission subsystem, the hyperspectral image acquisition subsystem and the sorting execution subsystem respectively, and is used to perform hardware input and output control. The central control computer generates sorting control instructions based on the material identification results and sends them to the control unit, which then drives the sorting execution subsystem to perform its actions.

2. The waste textile sorting system based on machine vision and artificial intelligence according to claim 1, characterized in that, The transmission subsystem includes a conveyor belt, a conveyor belt drive motor, and a frequency converter; The conveyor belt forms a material transport plane; the control unit is electrically connected to the conveyor belt drive motor and the frequency converter respectively; The central control computer sends speed commands to the frequency converter through the control unit to adjust the linear speed of the conveyor belt; The waste fabrics are placed in a single row on the conveyor belt, with a fixed physical distance between adjacent waste fabrics. The value of the distance depends on the acquisition time of a single frame image by the hyperspectral image acquisition subsystem and the computation time required by the central control computer to process a single material identification algorithm.

3. The waste textile sorting system based on machine vision and artificial intelligence according to claim 2, characterized in that, The hyperspectral image acquisition subsystem includes a hyperspectral image acquisition box, and a hyperspectral camera, an illumination source, and a laser detection sensor installed inside the hyperspectral image acquisition box; The hyperspectral image acquisition box is positioned above the conveyor belt; the type of illumination source includes a strip light source, a ring light source, or a planar light source. The laser detection sensor is signal-connected to the control unit and is used to detect whether waste fabric has entered the collection area; The central control computer is equipped with camera triggering logic, which is used to control the hyperspectral camera to execute exposure acquisition commands after receiving the trigger signal from the laser detection sensor.

4. The waste textile sorting system based on machine vision and artificial intelligence according to claim 1, characterized in that, The sorting execution subsystem includes several sorting and collection units arranged along the conveying direction; Each sorting and collection unit includes a collection bin, a push rod, a push rod assist mechanism, and a baffle; the push rod assist mechanism is electrically connected to the control unit. The push rod is mechanically connected to the output end of the push rod assist mechanism; The baffle is positioned between two adjacent sorting and collection units; The push rod assist mechanism is used to drive the push rod to move along a path perpendicular to the conveying direction, pushing the waste fabric into the corresponding collection bucket.

5. A waste textile sorting system based on machine vision and artificial intelligence according to claim 2, characterized in that, The sorting execution subsystem also includes an end collection unit located in the end area of ​​the conveyor belt; The end collection unit includes an end laser detection sensor, an end collection bucket, an end push rod assist mechanism, and an end push rod. The end laser detection sensor is connected to the control unit and is used to monitor the waste fabrics that arrive at the end of the conveyor belt; When the end laser detection sensor detects the passage of material, the control unit controls the end push rod assist mechanism to start, driving the end push rod to push the waste fabric into the end collection bucket.

6. A waste textile sorting system based on machine vision and artificial intelligence according to claim 1, characterized in that, The central control computer is equipped with an image processing module, which is used to perform cluster analysis on the acquired hyperspectral image data, specifically including: The central control computer stores the acquired hyperspectral image data in a three-dimensional matrix format; The central control computer performs K-means clustering operations on each of all bands; Before performing clustering operations, a cluster center number parameter is set, which is greater than the preset number of material types to be classified. For the current band being processed, a set consisting of multiple cluster centers is calculated.

7. A waste textile sorting system based on machine vision and artificial intelligence according to claim 6, characterized in that, The central control computer is also used to perform statistical filtering and noise reduction processing on the clustering output data, specifically including: The central control computer acquires multiple cluster center spectral values ​​for the current band and sorts them in ascending order according to their numerical values. The central control computer removes the cluster centers with the smallest values ​​and the cluster centers with the largest values ​​after sorting, based on a preset cutoff threshold ratio parameter. The central control computer calculates the arithmetic mean of the spectral values ​​of the remaining cluster centers, and determines the arithmetic mean as the effective spectral value of the waste fabric in a specific band. The central control computer sequentially executes the above steps for all bands and constructs the arranged spectral values ​​as a material characterization vector for waste fabrics.

8. A waste textile sorting system based on machine vision and artificial intelligence according to claim 7, characterized in that, The central control computer has a pre-installed hyperspectral image expert rule library in its internal memory, which contains standard material vectors for various standard materials. The central control computer is used to perform the following material classification and determination steps: Calculate the feature distance between the material representation vector of the waste fabric and each standard material vector in the expert rule base; The feature distance is calculated using either the Euclidean distance algorithm or the cosine distance algorithm; Compare all the calculated distance values ​​and determine the minimum value among them; The material category represented by the standard material vector corresponding to the minimum value is determined as the final material classification result of the current waste fabric.

9. A waste textile sorting system based on machine vision and artificial intelligence according to claim 5, characterized in that, The control unit is used to perform time-lag-based trigger control on the push rod assist mechanism: The control unit starts timing after receiving the detection signal from the laser detection sensor; When the timing duration reaches the preset trigger delay time, the control unit outputs an action command to the push rod assist mechanism at the target position; The trigger delay time depends on the physical distance between the laser detection sensor and the target push rod assist mechanism in the direction of the conveyor belt's operation and the current operating speed of the conveyor belt.

10. A waste textile sorting system based on machine vision and artificial intelligence according to claim 5, characterized in that, The central control computer is used to perform system performance evaluation and alarms. The central control computer is configured with statistical time window parameters and fault alarm threshold parameters. The central control computer collects the trigger signal of the end laser detection sensor in real time within the statistical time window parameter and maintains the abnormal collection count value; The central control computer compares the abnormal collection count value with the fault alarm threshold parameter in real time. When the abnormal collection count value is greater than the fault alarm threshold parameter, it determines that the system is in an abnormal working state and triggers the alarm program.

Images