System and method for automated sorting of recycled garments using artificial intelligence

Abstract

An automated system and method for sorting recycled garments are disclosed. The system includes a conveyance mechanism for transporting suspended garments along a processing path, and at least one imaging station positioned along the path. The imaging station comprises a plurality of cameras and an illumination system for capturing image data of a passing garment. A computing system in communication with the cameras executes a trained artificial intelligence (AI) model to analyse the captured image data and identify at least one characteristic of the garment, such as defect, brand, material composition, or colour. Based on the identified characteristic, the computing system generates a sorting signal. A sorting mechanism coupled to the computing system receives the signal and routes the garment via the conveyance mechanism into a predetermined sorting area according to that signal.

Description

1 SYSTEM AND METHOD FOR AUTOMATED SORTING OF RECYCLED GARMENTS USING ARTIFICIAL INTELLIGENCE TECHNICAL FIELD The present disclosure relates generally to systems and methods for automated sorting of recycled garments, and more particularly to an imaging and artificial intelligence-based system for identifying garment characteristics for sorting. BACKGROUND As the global textile industry accelerates its transition toward a circular economy, the recycling and sorting of post-consumer garments have become critical links in the value chain. The ability to efficiently and accurately categorize used garments based on their material composition, physical condition, brand value, and other characteristics is essential for determining their optimal path for reuse, recycling, or other forms of recovery. Currently, the industry predominantly relies on manual sorting operations. This approach suffers from fundamental problems, including low efficiency, high labor costs, and inconsistent sorting standards. Manual sorting is inherently subjective, leading to variability in classification outcomes, and is unable to keep pace with the growing volume of textile waste generated globally. Existing automated technical solutions generally suffer from limited functionality and insufficient detection dimensions. Material sorting techniques in existing systems primarily rely on near-infrared spectroscopy. While near-infrared spectroscopy can identify material composition such as cotton, polyester, or wool, it is incapable of detecting other critical garment attributes including brand, defects, or surface condition. This single-dimensional analysis fails to capture the full value proposition of a garment and limits the potential for optimized sorting decisions. Conventional defect detection systems are largely based on visible light imaging. These systems struggle to reliably identify subtle imperfections such as pilling, which requires enhanced contrast to visualize effectively. Furthermore, they are unable to detect certain types of stains that are invisible to the naked eye under normal lighting conditions, such as oil stains and blood stains, which become visible only under ultraviolet illumination. Additionally, these systems exhibit insufficient robustness when detecting defects against complex textured backgrounds, leading to inconsistent detection accuracy. Existing brand recognition approaches are heavily dependent on Optical Character Recognition technology. Optical Character Recognition is limited to processing clear, text- based labels and is completely ineffective for graphical trademarks, stylized fonts, or garments without visible labels. This limitation severely restricts the coverage and accuracy HK 30137917 A 2 of brand identification, missing valuable information that could influence sorting decisions for resale markets and reducing the economic value that can be extracted from the sorting process. Current technologies lack the capability to perform multi-dimensional fusion analysis that integrates brand value, physical condition, material properties, and aesthetic characteristics of a garment. Consequently, there is no standardized, quantifiable assessment metric available to guide sorting decisions. This absence of a unified value indicator prevents sorting systems from making economically optimized decisions about whether a garment should be directed toward resale, material recycling, or other recovery pathways, thereby limiting the potential for high-value utilization of recovered materials. These technological fragmentation issues have created a significant barrier in the industry: existing automated sorting systems cannot meet the practical demand for rapid, comprehensive, and accurate value assessment of garments. This technological bottleneck constrains the efficiency of textile recycling operations and impedes the high-value utilization of recovered materials, ultimately hindering the transition toward a circular economy in the textile sector. In view of the foregoing technical deficiencies in the prior art, the present disclosure aims to address or ameliorate the foregoing; or at least provide the public with a further choice. SUMMARY OF THE INVENTION Features and advantages of the disclosure will be set forth in the description which follows, and in part will be obvious from the description, or can be learned by practice of the herein disclosed principles. The features and advantages of the disclosure can be realized and obtained by means of the instruments and combinations particularly pointed out in the appended claims. In accordance with a first aspect of the present disclosure, there is provided a system for automated sorting of recycled garments. The system comprises a conveyance mechanism configured to transport suspended a plurality of separate individual garments along a processing path. The system further comprises at least one imaging station positioned along the processing path. The at least one imaging station comprises a plurality of cameras configured to capture image data of a garment from the plurality of separate individual garments suspended on the HK 30137917 A 3 conveyance mechanism passing therethrough. The at least one imaging station further comprises an illumination system comprising one or more light sources. The system further comprises a computing system in communication with the plurality of cameras, the computing system comprising at least one processor and a memory storing instructions that, when executed, cause the computing system to: analyse captured image data of the garment by at least one trained artificial intelligence (AI) model executing on the processor, the AI model configured to identify at least one characteristic of the garment selected from the group comprising defect, brand, material composition, and colour; and generate a sorting signal based on the identified characteristic. The system further comprises a sorting mechanism coupled to the computing system configured to receive the sorting signal and route the garment via the conveyance mechanism into a predetermined sorting area according to the sorting signal. The at least one imaging station of the system may comprise a first inspection booth having at least one camera disposed therein, said at least one camera configured to capture image data of a front side of the garment, and a second inspection booth having at least one camera disposed therein, said at least one camera configured to capture image data of a rear side of the garment. At least some of the plurality of cameras of the system capture image data of the garment in a spectral band which is different from the spectral band of other cameras of the plurality of cameras. The system may further comprise at least one multi-spectral booth comprising at least one multi-spectral camera disposed therein configured to capture image data of the garment in a plurality of electromagnetic spectral bands. The plurality of cameras of the system may be selected from the group comprising a Red- Green-Blue (RGB) camera, an Ultraviolet (UV) camera, a Near-Infrared (NIR) camera, a Long-Wave Infrared (LIR) camera, a multispectral camera, and a Short-Wave Infrared (SWIR) camera. The material composition of the system may be identified using either the Short-Wave Infrared (SWIR) camera or the multispectral camera. The one or more light sources of the illumination system of the system may be selected from the group comprising Light Emitting Diode (LED) light source, strobe light source, visible monochromatic light source, backlight source, coaxial light source, diffuse light source, near- infrared (NIR) light source, short-wave infrared (SWIR) light source, ultraviolet (UV) light source, and multi-spectral light source. HK 30137917 A 4 The defect of the system may include at least one selected from the group comprising pilling, yellow stain, oil stain, blood stain, hole, colour fading, missing button, broken zipper and mildew. The colour of the system may be identified using dual RGB cameras, and the colour fading is identified by calculating a Delta E (ΔE) colour variation value. The at least one imaging station of the system may have at least a first Red-Green-Blue (RGB) camera and a second Red-Green-Blue (RGB) camera, wherein at least one of the first or second RGB cameras being movably mounted within the imaging station to scan the garment. The plurality of cameras and the light sources of the system may be fixed in position within the at least one imaging station. The system may further comprise one or more modules selected from: (a) a pilling detection module for detecting pilling using edge detection and contour extraction on image data comprising near-infrared (NIR) image data; (b) a yellow stain detection module configured for detecting the presence of yellow stain by performing colour segmentation in Hue-Saturation-Value (HSV) colour space on image data comprising Red-Green-Blue (RGB) image data, and analysis thereafter by a trained artificial Intelligence classification (AI) classification model; (c) a defect detection module for detecting general defects by analysing the image data using a defect detection Artificial Intelligence (AI) object detection model; and (d) a brand identification module for identifying a brand of the garment using Optical Caracter Recognition (OCR) to extract text from a label, and / or by using an object detection model to recognize a brand logo. The computing system of the system may be further configured to aggregate the identified characteristic(s) of the garment into a standardized dataset, and to calculate a Garment Circulability Index (GCI) score using a multi-feature regression model or a transformer-based regression model. The conveyance mechanism of the system may include a braking mechanism which automatically stops the conveyance mechanism when the garment enters the at least one imaging stations to enable image capture. HK 30137917 A 5 At least one camera of the plurality of cameras of the system may include an embedded processor for initial image processing. The system may further comprise: a data storage device for receiving data across a network from the computing system for storing thereon, an interface device communicatively coupled to the computing system for control thereof, and a remotely located training server for generating one or more artificial intelligence models using data received from the data storage device. The remotely located training server of the system may be configured for developing updated software for updating the computing system. The sorting mechanism of the system may comprise a diverter arm or gate configured to direct the garment into one of a plurality of sorting bins based on the sorting signal. In accordance with a second aspect of the present disclosure, there is provided a method for automated sorting of recycled garments. The method comprises moving by a conveyance mechanism a plurality of separate suspended individual garments along a processing path. The method further comprises capturing by a plurality of cameras image data of a garment of the plurality of separate suspended individual garments, said garment being illuminated by one or more light sources. The method further comprises analysing by a processor of a computing system the captured image data of the garment executing at least one trained artificial intelligence (AI) model thereon. The method further comprises identifying by the trained artificial intelligence model at least one characteristic of the garment selected from the group comprising defect, brand, material composition, and colour. The method further comprises generating by the processor a sorting signal based on the identified characteristic; and The method further comprises moving the garment into a predetermined sorting area according to the sorting signal. The method further comprises using a first inspection booth having at least one camera disposed therein, and a second inspection booth having at least one camera disposed therein. HK 30137917 A 6 The method further comprises capturing image data of the garment by at least some of the plurality of cameras in a spectral band, and capturing image data of the garment by other cameras of the plurality of cameras in a different spectral band. The method further comprises capturing image data of the garment in a plurality of electromagnetic spectral bands using at least one multi-spectral camera disposed in at least one multi-spectral booth. The plurality of cameras of the method may be selected from the group comprising a Red- Green-Blue (RGB) camera, an Ultraviolet (UV) camera, a Near-Infrared (NIR) camera, a Long-Wave Infrared (LIR) camera, a multispectral camera, and a Short-Wave Infrared (SWIR) camera. The identified characteristic of the method may be material composition; and said material composition identification may be performed by the trained AI model analysing image data captured by either the Short-Wave Infrared (SWIR) camera or the multispectral camera. The one or more light sources of the method may be selected from the group comprising a Light Emitting Diode (LED) light source, a strobe light source, a visible monochromatic light source, a backlight source, a coaxial light source, a diffuse light source, a near-infrared (NIR) light source, a short-wave infrared (SWIR) light source, an ultraviolet (UV) light source, and a multi-spectral light source. The identified characteristic of the method may be a defect in the garment; wherein the identified defect may be a defect selected from the group comprising pilling, yellow stain, oil stain, blood stain, hole, colour fading, missing button, broken zipper, and mildew. The identified characteristic of the method may be the colour of the garment; and wherein said colour may be identified using image data captured by dual RGB cameras, and wherein a defect of colour fading is identified by calculating a Delta E (ΔE) colour variation value. The method further comprises using at least one of a first or second RGB cameras to movably scan the garment. The method further comprises fixing the position of the plurality of cameras and the light sources. The method further comprises performing one or more of the following: (a) detecting by the trained artificial intelligence model executing of the processor the presence of pilling by using edge detection and contour extraction on image data comprising near-infrared (NIR) image data; HK 30137917 A 7 (b) detecting by the trained artificial intelligence model executing of the processor the presence of a yellow stain by performing colour segmentation in Hue-Saturation-Value (HSV) colour space on image data comprising RGB image data, followed by analysis using a trained artificial intelligence classification model; (c) detecting by the trained artificial intelligence model executing of the processor the presence of general defects by analysing the image data using a defect detection AI object detection model; and (d) identifying by the trained artificial intelligence model executing of the processor a brand of the garment by using Optical Character Recognition (OCR) to extract text from a label, and / or by using an object detection model to recognize a brand logo. The method further comprises aggregating the identified characteristic of the garment into a standardized dataset, and calculating a Garment Circulability Index (GCI) score using a multi-feature regression model or a transformer-based regression model. The method further comprises automatically stopping the conveyance mechanism to enable image capture. The method further comprises embedding a processor with at least one camera of the plurality of cameras for initial image processing. The method further comprises receiving data from a computing system and storing said data in a data storage device. The method further comprises controlling the computing system via an interface device communicatively coupled thereto. The method further comprises generating one or more artificial intelligence models using data received from the data storage device by a remotely located training server. The method further comprises developing updated software for updating the computing system by the remotely located training server. Moving the garment into a predetermined sorting area of the method may further comprise releasing the garment to one of a plurality of sorting bins based on the sorting signal from the conveyance mechanism. BRIEF DESCRIPTION OF THE DRAWINGS The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate several embodiments of the invention and together with the description, serve to explain the principles of the invention. HK 30137917 A 8 FIG. 1 depicts an exemplary system architecture of an automated sorting system according to an embodiment. FIG. 2A depicts an exemplary imaging station of an automated sorting system according to an embodiment. FIG. 2B-2C depict an exemplary imaging station of an automated sorting system according to another embodiment. FIG. 3A-3D depict exemplary camera configuration and light sources layout of an imaging station of the automated sorting system according to various embodiment. FIG. 4A depicts a schematic diagram of a computing system workflow of an automated sorting system according to an embodiment. FIG. 4B depicts a schematic diagram of a computing system workflow of an automated sorting system according to another embodiment. FIG. 5A depicts an exemplary user interface for user control and feedback of a computing system of an automated sorting system according to an embodiment. FIG. 5B depicts an exemplary user interface for user control and feedback of a computing system of an automated sorting system according to another embodiment. FIG. 6 depicts an exemplary dataset structure generated by the computing system for each garment to be sorted according to an embodiment. FIG. 7 depicts a schematic diagram of a sorting mechanism of an automated sorting system according to an embodiment. FIG. 8A depicts a workflow for pilling feature detection performed by an automated sorting system according to an embodiment. FIG. 8B depicts an exemplary image processing workflow for pilling feature detection performed by an automated sorting system according to an embodiment. FIG. 9A depicts an exemplary workflow for yellow stain detection performed by an automated sorting system according to an embodiment. FIG. 9B depicts an image processing workflow for yellow stain detection performed by an automated sorting system according to an embodiment. FIG. 9C depicts the use of an artificial intelligence classification model by an automated sorting system to distinguish yellow stains from other yellow objects according to an embodiment. HK 30137917 A 9 FIG. 10A depicts a workflow for general defect detection performed by an automated sorting system according to an embodiment. FIG. 10B depicts a visualized presentation of defect detection results performed by an automated sorting system according to an embodiment. FIG. 11 depicts a process flowchart according to the method of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS The present invention relates generally to systems and methods for automated sorting of recycled garments, and more particularly to a multi-spectral imaging and artificial intelligence-based system for identifying garment characteristics and generating sorting decisions. Preferred embodiments of the present invention will now be described in detail with reference to the accompanying drawings. The drawings form a part of this specification to help a fuller understanding of the invention. It is to be understood that the drawings are for illustrative purposes and are not intended to limit the scope of the invention. Figure 1 depicts an automated sorting system 100 for recycled garments, according to a preferred embodiment of the present invention. The system 100 is primarily used for the automated inspection and classification of post-consumer garments to facilitate subsequent reuse or recycling processes. As depicted in Figure 1, the system 100 generally comprises: a conveyance mechanism 110, at least one imaging station 120, an illumination system integrated within the imaging station 120, a computing system 130, and a sorting mechanism 140. The conveyance mechanism 110 is configured to hang individual garments 10 and transport them sequentially along a predetermined processing path P. An operator can simply hang the garments 10 to be sorted onto loading hooks or hangers 112 of the conveyance mechanism 110. In a specific embodiment, the number of loading hooks or hangers 112 can be flexible and is not limited to the number shown in the figures; it can be adjusted based on actual processing requirements or the system's throughput capacity, for example, to achieve optimal synchronization with upstream or downstream processes. The garments 10 to be sorted are suspended on the conveyance mechanism 110 via hangers or other suspension means. The conveyance mechanism 110 itself can be a motor- driven, continuously moving physical conveyor, such as a rotary or linear belt, which HK 30137917 A 10 transports the garments 10 along the processing path P into the imaging station 120 for inspection. In one specific application, a single automated sorting system apparatus can be equipped with three loading hooks or hangers 112, allowing three operators to load garments simultaneously in a collaborative manner, thereby improving efficiency. Furthermore, the rotation cycle time for each loading hook or hanger 112 can be adapted according to the overall processing capacity of the system, for instance, set to less than 15 seconds, ensuring continuous and stable system operation. To enable traceability and data association for each individual garment, each hook or hanger 112 may be equipped with a passive or active Radio‑Frequency Identification (RFID) tag. The tag stores a unique identifier (UID) that is permanently linked to the specific garment 10 suspended thereon. Before or at the loading stage, an operator associates the garment’s information (e.g., type, supplier, initial condition) with the UID of the hanger via an RFID reader / writer connected to the computing system 130. Once the association is established, the garment 10 carries its own digital identity throughout the entire sorting process. As the hook or hanger 112 travels along the conveyance mechanism 110, strategically placed RFID readers (e.g., at the entrance of the imaging station 120, at the sorting mechanism 140, and at unloading stations) automatically detect the UID without requiring line‑of‑sight. The computing system 130 uses this UID to log all data generated for that garment, including captured image data, AI‑based inspection results (defects, brand, material composition, color), sorting decisions, and timestamps. This RFID‑based approach provides a per‑garment “digital passport”, enabling real‑time monitoring, quality tracking, and downstream data analytics. It also facilitates error recovery, e.g., if a garment needs re‑sorting or manual verification, the system can retrieve its complete inspection history by reading the hanger’s UID. Compared to traditional loading methods that require manually placing garments flat, the present invention, through the conveyance mechanism 110, adopts an automated hanging loading method. This design ensures garments remain suspended throughout the process and are automatically fed into subsequent inspection stations, avoiding additional interference caused by folding or stacking and creating favourable conditions for subsequent multi-angle imaging. Figure 2A provides a detailed view of an exemplary embodiment of the imaging station 120 positioned on the conveyance mechanism 110 of the automated sorting system. In this embodiment, at least one imaging station 120 is provided and is used to systematically capture image data of an individual garment 10, enabling subsequent analysis to identify key HK 30137917 A 11 characteristics such as defects, brand, material composition, and colour. The defects may include a defect selected from the group consisting of pilling, yellow stain, oil stain, blood stain, hole, colour fading, missing button, broken zipper, and mildew. As shown in Figure 2A, the imaging station 120 is disposed along the processing path P on the conveyance mechanism 110 and preferably adopts a studio-like enclosure or booth structure 122. This structure features a rigid frame for precisely mounting and securing various cameras and lighting equipment, thereby enabling stable, repeatable, and systematic imaging of each passing garment. To acquire comprehensive and multi‑dimensional information, the imaging station 120 integrates a plurality of cameras of various types at different positions. The camera set may include any combination of the following: Red‑Green‑Blue (RGB) cameras, Ultraviolet (UV) cameras, Near‑Infrared (NIR) cameras, Long‑Wave Infrared (LIR) cameras, multispectral cameras, and Short‑Wave Infrared (SWIR) cameras. Depending on the specific application requirements (e.g., cost targets, detection accuracy, material complexity), one or more of these camera types can be selected or omitted. Each camera type is configured to capture image data in a distinct spectral band. For instance, RGB cameras cover the visible range, UV cameras operate in the ultraviolet region, NIR and SWIR cameras cover the near‑ and short‑wave infrared, LIR cameras detect thermal radiation, and multispectral cameras sample multiple narrow bands across the visible to SWIR range. By capturing images in different spectral bands, the system generates a rich, multi‑channel dataset that enables a much broader spectrum visualization of each garment than is possible with conventional visible‑light imaging. This approach reveals hidden features such as material composition, residual stains, structural defects, and even thermal anomalies, thereby significantly enhancing the robustness and reliability of subsequent AI‑based analysis. In one specific embodiment, the camera configuration within the imaging station 120 comprises: RGB Camera: Configured to capture high-resolution colour images, providing a baseline for identifying garment colour, texture, and visible brand logos or labels. Ultraviolet (UV) Camera: Configured to capture images of the garment under UV illumination. UV light is highly effective at revealing residues that are difficult to detect under normal lighting conditions, such as oil stains and blood stains. By minimizing background interference from visible light, UV imaging significantly enhances the accuracy and reliability of detecting specific types of soiling. HK 30137917 A 12 Near-Infrared (NIR) Camera: Configured to capture NIR images of the garment. Different materials (e.g., cotton, polyester, wool) reflect NIR light differently; therefore, NIR imaging can reveal material properties and surface characteristics that are difficult to observe with traditional lighting. For example, when NIR light is projected from a specific angle, it can enhance the contrast between the fabric surface and protruding fibers (such as pilling), thereby clearly highlighting common textile defects like pilling. Long-Wave Infrared (LIR) Camera: In some preferred embodiments, the imaging station 120 further comprises an LIR camera. LIR imaging, by detecting differences in thermal radiation or heat conductivity, can reveal the layered structure of a garment or hidden items. It can function effectively even in complete darkness, providing an additional dimension for comprehensive garment analysis. By integrating data from UV, visible, NIR, and LIR cameras, the system of the present invention achieves comprehensive and precise multi-spectral garment inspection. This multi- camera, multi-spectral fusion approach greatly enhances the robustness and reliability of identifying various garment characteristics and defects. It should be noted that the specific selection and number of cameras can vary across embodiments. Depending on the detection scenario and cost considerations, cameras that do not perform satisfactorily may be omitted. For instance, in a simplified yet effective embodiment, the imaging station 120 may include only an RGB camera, a UV camera, and an NIR camera, while still meeting basic defect and stain detection requirements. Conversely, for applications demanding rigorous material sorting, the SWIR and multispectral cameras are added to provide the necessary spectral resolution. In some embodiments, to enable reliable identification of material composition, the imaging station 120 further comprises a Short‑Wave Infrared (SWIR) camera and / or a multispectral camera. The SWIR camera: Different textile materials — such as cotton, polyester, wool, nylon, and elastane — exhibit distinct SWIR spectral signatures due to their unique molecular absorption characteristics. By analysing the SWIR reflectance spectrum of a garment, the computing system 130 can accurately determine its fiber composition without physical contact or destructive sampling. For example, cotton shows a characteristic absorption peak around 1490 nm, while polyester is readily distinguishable by its strong absorption at 1410 nm and 1670 nm. This capability is critical for recycling applications, where material purity directly affects the value of the recovered feedstock. The multispectral camera: This camera captures image data across a plurality of discrete, narrow spectral bands spanning the visible, near‑infrared, and short‑wave infrared regions. HK 30137917 A 13 Unlike a conventional RGB camera that records only three broad channels, the multispectral camera provides a high‑dimensional spectral profile for each pixel of the garment. The computing system 130 processes this profile using trained AI models to classify materials, detect contaminants (e.g., non‑textile components), and even identify blends with high precision. In a preferred embodiment, the multispectral camera is integrated within a dedicated multispectral booth equipped with appropriate broadband or switchable illumination sources. In one specific example, to balance cost and performance, the imaging station 120 may comprise four RGB cameras, two UV cameras, and one NIR camera, distributed at different angles and positions to ensure complete coverage of the garment's surface. To ensure comprehensive image capture of all sides of a garment, in some embodiments, as depicted in Figure 2A, the imaging station 120 may comprise two independent inspection stations: a first inspection booth 122A and a second inspection booth 122B. The first inspection booth 122A is configured to capture image data of the front side of a garment 10 when it enters. Subsequently, the conveyance mechanism 110 transports the garment 10 into the second inspection booth 122B, which is configured to capture image data of the rear side of the garment. Utilizing two independent booths effectively prevents light interference between the two capture events and simplifies the camera and lighting layout within each station, thereby ensuring high-quality image data is captured from both the front and back of the garment without blind spots. In other embodiments, as depicted in Figure 2B or Figure 2C, the imaging station 120 may be implemented as a single inspection booth 122 without requiring two separate booths. In this configuration, the single station 122 is equipped with a mixed arrangement of cameras and light sources strategically positioned to capture both the front and rear sides of the garment 10 simultaneously or sequentially within the same enclosure. For example, a first set of cameras (e.g., RGB, UV, NIR) and corresponding illuminators may be aimed at the front side of the passing garment, while a second set of cameras and illuminators is oriented toward the rear side. By carefully shielding or time‑synchronizing the lighting, the single‑station design avoids cross‑illumination and reduces hardware footprint and cost, while still achieving comprehensive inspection of the garment without blind spots. In some embodiments, to better capture material characteristics of the garment, a dedicated multispectral booth may be provided specifically for the multispectral camera and / or the SWIR camera to capture multispectral image date of the garment 10. This dedicated booth is equipped with appropriate broadband or switchable illumination sources tailored to the spectral requirements of these cameras. motorized filter wheels. By isolating the HK 30137917 A 14 multispectral and / or SWIR cameras in a separate booth, the system avoids optical interference from other lighting sources (e.g., RGB, UV, or NIR illuminators) used in the main imaging station, thereby improving signal‑to‑noise ratio and measurement repeatability. This configuration is particularly advantageous for high‑precision material composition analysis. Each inspection booth (122A, 122B) may contain a controlled background 124. The colour and material of the background 124 are selected to ensure uniformity and consistency across all captured images, facilitating subsequent image processing. The garment 10 being photographed is suspended in front of the background 124, positioned optimally for all cameras within the booth to achieve the best imaging results. In further embodiments, the number of inspection booths is not limited to one or two, but may be adapted according to actual detection requirements. For instance, the system may comprise three, four, or even more booths arranged along the processing path, each configured with different lighting conditions and corresponding camera setups tailored to specific inspection tasks. A first booth may be dedicated to RGB and UV imaging for stain detection under bright visible / ultraviolet illumination; a second booth may focus on NIR and SWIR imaging for material composition under near‑infrared lighting; a third booth may perform multispectral analysis with programmable narrow‑band illumination; and an optional fourth booth may include LIR cameras for thermal assessment. This modular, multi‑booth approach allows the system to be customized for complex sorting workflows, balancing throughput with detection precision, and enabling parallel or sequential specialized inspections without mutual interference. Figures 3A depicts an exemplary illumination system integrated within the imaging station 120. The illumination system comprises programmable lighting or non-programmable lighting. The illumination system is equipped with multiple types of light sources configured to provide controlled illumination across multiple wavelengths. Specifically, the lighting source includes, but is not limited to, Light Emitting Diode (LED) light source, strobe light source, visible monochromatic light source, backlight source, coaxial light source, grille fill light source, diffuse light source, near-infrared (NIR) light source, short-wave infrared (SWIR) light source, ultraviolet (UV) light source, and multi-spectral light source. These lights are carefully arranged within the imaging station 120 to minimize shadows and to reveal the characteristics of the garment being photographed under different wavelengths, thereby facilitating high-quality multi-spectral imaging. In an exemplary embodiment, as depicted in Figure 3A, the arrangement of the light sources within an inspection booth (e.g., 122A or 122B) is optimized as follows: HK 30137917 A 15 LED lights 210 are positioned at the top side of the booth, providing general-purpose, even illumination. Grille fill lights 220 are disposed on the two side walls of the inspection booth to provide fill light, further reducing shadows and ensuring uniform brightness across the garment. Ultraviolet (UV) lights 230 are positioned at the upper and lower portions of the side opposite the background 124, ensuring thorough UV exposure for detecting residues and stains. In further embodiments, additional Near-infrared (NIR) lights may be mounted on a bracket on one side of the inspection booth, preferably at an upper position relative to the side opposite the background 124. For example, as depicted in Figure 3B, four NIR lights 240 are positioned at the top corner locations opposite the background 124 (i.e., on the same side as the UV lights). These NIR lights 240 may be evenly spaced and oriented at an angle toward the garment 10. This angled arrangement from the upper corners provides directional near- infrared illumination, which can further enhance surface texture details and defect visibility from multiple directions. This specific arrangement for the NIR lights is particularly advantageous. By positioning the NIR lights to illuminate the garment from above, the contrast between the fabric surface and protruding fibers (such as those found in pilling) is significantly enhanced. Consequently, this configuration effectively highlights surface features and defects, such as pilling, that are difficult to observe under normal lighting conditions, as illustrated in the accompanying images. Figure 3C depicts a different arrangement of lighting sources within an inspection booth according to another embodiment. As shown, LED lights 210 are positioned at the upper and lower corner locations on the side opposite the background 124. UV lights 230 are also positioned at the upper and lower portions of the same side opposite the background 124. Additionally, grille fill lights 220 are disposed at the connecting corner regions between the side opposite the background 124 and the adjacent side walls. This configuration distributes illumination from multiple corner positions, which may help further reduce shadows and improve overall lighting uniformity across the garment. It should be noted that, in different embodiments, the arrangement of the light sources may be specifically adapted based on the actual imaging requirements (such as the characteristics of the garment 10 to be acquired) as well as the lighting conditions needed for camera operation. Therefore, a person skilled in the art would appreciate that the number and positions of the light sources are not limited to the above embodiments. Figure 3A also depicts an exemplary embodiment of the camera configuration within the inspection booth 122. As shown, three cameras may include RGB cameras 310 or UV camera 320 are positioned approximately level with the height of the garment 10 and directly HK 30137917 A 16 facing the garment 10. It would be appreciated that the position of these cameras and number of cameras is exemplary only. Additionally, as depicted an NIR camera 330 is disposed at an angle on the side opposite the background (124), toward a lower lateral position, so as to perform imaging tasks at an oblique angle. Figure 3B depicts a camera configuration within the inspection booth 122 according to another embodiment. In this embodiment, RGB camera 310 and NIR camera 330 are employed. Unlike the configuration shown in Figure 3A, a fifth RGB camera 310 is added, positioned at an angle from an upper location on the side opposite the background 124 to capture image data of the garment 10. Among the five NIR cameras 330, four cameras are arranged at positions substantially similar to those of the RGB camera 310, while the fifth NIR camera 330 is placed at a position approximately symmetrical to the fifth RGB camera 310. Figure 3C depicts another embodiment of an exemplary camera configuration within the inspection booth 122. In this embodiment, in contrast to the configuration of Figure 3A, four RGB cameras 310 are provided, each respectively disposed at the central positions of the upper, lower, left, and right regions on the side opposite the background 124, directly facing the garment 10. Two UV cameras 320 are placed at the middle portions of the upper and lower sections on the side opposite the background. In this embodiment, an NIR camera 330 may also be employed, positioned at an angle from the upper right side of the background 124 at a corner location to obliquely capture the image data of the garment 10 obliquely. In another embodiment, to achieve a balance between cost and performance, the imaging station 120 may be configured as an ‘economy’ version as shown in Figure 3D. Instead of employing multiple fixed cameras distributed at various angles, this implementation uses one or two movable RGB cameras 310. Specifically, the two RGB cameras 310 are mounted on a movable carriage or rail system within the imaging station 120, enabling the cameras to translate vertically (i.e., up and down) along the length of the suspended garment (not shown). As the garment enters the imaging station, the cameras scan the entire front and rear surfaces sequentially by moving from top to bottom (and optionally bottom to top), capturing overlapping image segments that are later stitched together by the computing system 130. This scanning mechanism eliminates the need for a large array of stationary cameras, thereby significantly reducing component costs while still providing full‑surface coverage. It should be noted that, in different embodiments, the arrangement of cameras may be specifically adapted according to actual imaging requirements (e.g., which features of the HK 30137917 A 17 garment 10 are to be captured). Therefore, the types, number, and positions of the cameras are not limited to those described in the foregoing embodiments. In some preferred embodiments, to obtain the sharpest possible images, the conveyance mechanism 110 is further configured to automatically stop movement when an individual garment 10 fully enters the imaging station 120 (e.g., into the first booth 122A or the second booth 122B). This pause allows the garment to be held stationary at a pre-determined optimal imaging position. After the imaging station 120 completes all image capture, the conveyance mechanism 110 restarts, transporting the garment to the next station. This "stop-shoot-go" operational mode prevents image blurring caused by garment swaying, to enhance quality. For example, the above‑described "stop‑shoot‑go" operational mode may be implemented using an infrared sensor. As shown in Figure 3B, an infrared sensor 410 is positioned within the booth directly facing the garment 10 to detect arrival of the garment at the imaging station. When the sensor detects that the garment has fully entered the predetermined optimal imaging position, it sends a signal to the conveyance mechanism 110 to pause movement. Upon completion of all image captures, the conveyance mechanism 110 is signalled to resume movement / transport. This sensor‑based triggering provides a reliable and cost‑effective solution for synchronizing garment conveyance with image acquisition. It should be noted that the background 124 within the booth 122 may be configured according to the imaging requirements of the cameras, and is not limited to being disposed on only one side. As shown in Figure 3B, a solid colour background 124 is provided on multiple sides of the booth to ensure accurate image capture. This multi-sided background arrangement helps reduce interference from external surroundings and provides a consistent imaging environment for cameras positioned at different angles, thereby improving the reliability of feature detection on the garment. The automated sorting system 100 further comprises the computing system 130. Figure 4A depicts an architecture diagram of the automated sorting system for implementing automated inspection and data management, according to one embodiment of the present invention. The computing system 130 includes at least one processor and a memory. The memory stores instructions that, when executed, configure the computing system 130 to perform the following operations: First, the computing system 130 communicates with one or more AI camera modules 132. Each AI camera module 132 comprises a camera unit (which may be an RGB, UV, NIR, or LIR camera) and an associated microcontroller or embedded processor. This AI camera HK 30137917 A 18 module 132 assembly is responsible for image acquisition and performing initial, on-device AI processing, such as preliminary defect detection and brand recognition. The results of this initial processing are then transmitted to a terminal computer 134 for further analysis. The terminal computer 134 acts as a central hub, connecting all system components and managing higher-level operations. It receives image data and preliminary results from the AI camera modules 132 and conducts more advanced AI analysis. This analysis may include, but is not limited to, detailed defect detection, brand name verification, pilling assessment, and yellow stain detection. The terminal computer 134 aggregates the results from these various detection modules to generate comprehensive output data. In some embodiments, the cameras may be not configured as AI camera modules as they do not contain an embedded AI chip or processor. Instead, each camera operates as a conventional image acquisition device. Upon receiving a capture command from the terminal computer 134, the camera performs the image capture task and transmits the raw image data back to the terminal computer 134 for all subsequent processing. In such a configuration, all image analysis — including defect detection, brand recognition, pilling assessment, stain identification, and material classification — is carried out entirely by the terminal computer 134 or by a remotely connected server. This approach simplifies the camera hardware and reduces per‑unit cost, but places higher computational demands on the central system. In other embodiments, the AI camera modules 132 may be equipped with more powerful embedded chips (e.g., a neural processing unit or a high‑performance microcontroller). These embedded processors can handle the vast majority of image analysis tasks, including not only preliminary detection but also advanced AI inferences such as detailed defect classification, material composition analysis via SWIR or multispectral data, and even calculation of the Garment Circulability Index (GCI) score. In such embodiment, the terminal computer 134 receives partially or fully processed results from each AI camera module and aggregates the data into a standardized dataset. The terminal computer may also perform final validation, store results in the data storage device, and manage system‑level coordination without undertaking heavy pixel‑level computations. Furthermore, the terminal computer 134 may be configured to perform Garment Circulability Index (GCI) score analysis and calculation, for example, by utilizing a self-learning neural network model. The terminal computer 134 also provides a user interface 135 for operator 137 control and feedback, and is responsible for distributing operational commands and tasks to the individual AI camera modules 132. HK 30137917 A 19 Finally, the computing system 130 is configured to send the data generated by the terminal computer 134 to a Network-Attached Storage (NAS) device 136 for persistent storage. The NAS device 136 receives data from the terminal computer 134 and provides high-capacity, reliable storage for future analysis, model training, and system upgrades. Figure 4B depicts an exemplary architecture diagram of the automated sorting system according to another embodiment of the present invention, which further enables device integration and automated firmware and model updates. In this embodiment, the computing system 130 may achieve enhanced integration and communication. Specifically, the terminal computer 134 is configured for bidirectional communication and collaboration with Internet of Things (IoT) devices 133 (such as sensors, cell phone and smart hardware) and operators 137. This allows the system to obtain real- time information, data, or instructions, enabling it to respond or adjust quickly based on the latest conditions or feedback. In the above embodiment, the operator 137 is also able to perform bidirectional communication and collaboration with the terminal computer 134 via the interaction interface 135. This interface 135 allows the system to obtain real‑time information, data, or operational instructions directly from the operator. Conversely, the terminal computer 134 can present detection results, analysis outputs, image data, and system status information to the operator in a clear and actionable manner. The operator can thus quickly respond or adjust system behaviour based on the latest conditions or feedback. For example, the operator may manually correct misclassified defects, add new labels to unmarked images, or trigger a capture command through the UI. This bidirectional human‑machine interaction not only provides a supervisory check on the automated processes but also enables dynamic corrections that can be fed back into the retraining loop, further enhancing the AI model’s accuracy and adaptability over time. Critically, the computing system 130 is configured for bidirectional communication with an external AI training server 138. The external AI training server 138 receives sample data requiring training from the NAS device 136. This sample data may include, for example, images of defects, brand labels, and other monitored data. The external AI training server 138 utilizes self-learning neural networks to train on this data, generating or updating AI algorithm models for tasks such as defect detection and brand recognition. Once a new or improved model is trained, the server 138 transmits the updated model or associated firmware back to the terminal computer 134. The terminal computer 134 then deploys this update, upgrading the AI algorithms used by the computing system in real- time. HK 30137917 A 20 This architecture creates a continuous optimization loop: data collected during operations is stored, then used by an external server for offline training, and the resulting improved models are deployed back to the on-site system. This allows the automated sorting system to continuously refine its algorithms from new data, progressively improving its detection accuracy over time. In the configuration shown in Figure 4B, while the NAS device 136 handles primary data storage, the external AI training server 138 acts as the "brain" responsible for algorithm iteration and distribution. It serves as the "training center" and "algorithm upgrade engine" for the entire system, not participating directly in real-time inspection but driving continuous performance improvement through accumulated data. Figure 5A depicts an exemplary user control and feedback interaction interface 135 of the automated sorting system according to one embodiment of the present invention. In this embodiment, the colour (RGB), ultraviolet (UV), and near‑infrared (NIR) camera modules are connected to the terminal PC (i.e., they are not equipped with onboard AI chips). The terminal computer’s user interface (UI) 135 provides real‑time status information of the cameras, displays live image feeds captured by the RGB, UV, and NIR camera modules, and shows adjacent windows that log camera device activity, confirm camera initialization status, and present other relevant system information. This arrangement gives the operator comprehensive situational awareness and control over the imaging and inspection process. Beyond monitoring, the UI 135 enables several direct operational commands. Specifically, the operator can manually label the dataset of a garment 10 or manually correct an existing dataset. For example, if the operator finds that the AI assigned an incorrect result, missed a stain or defect, or incorrectly marked a normal area, the operator can click a labeling tool on the UI 135 to manually correct the error. The corrected data can then be fed back into retraining to improve AI performance. Additionally, the operator can issue a capture command via the UI 135 to trigger the cameras to take images (e.g., when the operator manually clicks the capture command on the control panel, or automatically when the terminal PC 134 receives IoT information that the garment is in position). Regarding the tasks assigned to the AI camera, this part is mainly handled automatically by the software. The signals processed by the terminal PC 134 are expected to include only two main types: (1) a capture task signal — sent when the terminal PC 134 receives IoT information indicating that the garment is in position, or when the operator manually clicks the capture command on the control panel; and (2) an update task signal — sent when the terminal PC 134 receives notification that the object detection AI model is ready to be updated. Based on an AI model upgrade reminder displayed on the UI, the operator can HK 30137917 A 21 instruct the terminal PC to perform the algorithm upgrade (i.e., load the updated model for subsequent inference). Finally, the UI 135 also supports manual dataset operations: the operator can label images after data collection (e.g., marking defects with bounding boxes and linking relevant information such as brand, material, and quality), and the labelled data is then used for AI training. If any error is discovered during or after operation, the operator can immediately correct it through the UI 135, and the corrected data is fed back into retraining. This closed‑loop human‑in‑the‑loop mechanism continuously improves the AI’s detection accuracy over time. All image analysis — including defect detection, brand recognition, pilling assessment, stain identification, and material classification — is carried out entirely by the terminal PC 134, simplifying camera hardware and centralizing computational resources. In some other embodiments, the user interface (UI) 135 of the computing system 130 may be provided in a simplified version to enhance operational efficiency for routine sorting tasks. As depicted in Figure 5B, the exemplary UI is divided into left and right panels 510, 520. As depicted, the left panel 510 is configured to display image information, such as captured still images of the garment 10 from the RGB, UV, and NIR camera modules. As depicted, the right panel 520 presents the results of image analysis performed by the computing system 130, including detailed information and spatial locations of specific defects (e.g., stains, holes, pilling), as well as the detected colour, brand, and / or material composition of the garment. Additionally, the right panel 520 displays the corresponding GCI (Garment Condition Index) score, providing an overall quality assessment. For human‑machine interaction, the UI includes command windows that allow the operator to issue instructions, such as executing an analysis on the current image, switching to the next camera for further capture and analysis, or confirming a sorting decision. This simplified exemplary layout enables operators to quickly view both raw images and analytical results on a single screen, reducing cognitive load and improving throughput. The streamlined interface facilitates efficient manual review and intervention when necessary. As shown in Figure 6, the dataset 160 generated by the computing system 130, comprising entries for each identified garment based on the recognized characteristics, may contain various data fields. These fields can include, for example, a brand level classification 162, a pilling score 164, and defect characteristics 166. The brand level 162 can be categorized into groups such as high-grade, mid-grade, low-grade, and unknown. Defect characteristic 166 subclasses may include, but are not limited to, holes, pilling, and stains. In some embodiments, the data entry formed for each individual garment to be sorted is structured as a feature vector. In an exemplary implementation, this may be a 1×173 feature HK 30137917 A 22 vector. This vector comprises four features for brand level, one feature for the pilling score, and 168 features related to defects. These 168 defect features are generated by aggregating counts and average areas across several dimensions. Specifically, they are derived from a combination of: Three defect types: holes, pilling, and stains; Seven cameras: four RGB cameras, two UV cameras, and one NIR camera; Four regions of the garment (e.g., front-top, front-bottom, back-top, back-bottom); and Two numerical value types: count and average area. For each combination of camera, defect type, and region, both the count of detected instances and their average area are calculated, resulting in the 168-dimensional defect feature set (e.g., 3 defects × 7 cameras × 4 regions × 2 value types = 168 features). This comprehensive feature vector provides a rich, multi-dimensional representation of the garment's condition. The computing system 130 is further configured to calculate a Garment Circulability Index (GCI) score. This is achieved by aggregating outputs from multiple detection modules, which at least include brand recognition, pilling assessment, and defect detection. The system utilizes the data entry (e.g., the 1×173 feature vector described above) generated for each garment as input to a predictive model. This model may be, for example, a multi-feature regression model or a transformer-based regression model. The computing system 130 integrates the identified characteristics of the garment into a standardized dataset (the feature vector) and then uses this dataset to predict a single, unified quality score for each garment—the GCI score. In embodiments employing a multi-feature regression model, the model may incorporate an attention mechanism. This mechanism assigns weights to each input feature, enabling the system to focus on the most relevant garment attributes. The architecture is designed to ensure that features of higher importance, such as a large-area stain, contribute more significantly to the predicted quality score than less critical features. In embodiments employing a transformer-based model, the model utilizes multi-head attention layers. These layers are capable of capturing dependencies and relationships between all input features, even if those features are located far apart within the input feature sequence. This approach is particularly effective for handling complex, high- dimensional data, allowing the model to learn intricate interactions between, for example, a specific brand, a particular type of defect in a specific region, and the overall material composition, leading to a more nuanced and accurate GCI score. HK 30137917 A 23 As illustrated in Figure 7, the automated sorting system of the present invention further comprises a sorting mechanism 140. The sorting mechanism 140 may include, for example, a diverter arm or a sorting gate. The sorting mechanism 140 is operatively coupled to the computing system (not shown) and is configured to receive the sorting signal generated by the computing system based on the identified characteristics and the calculated GCI score. Upon receiving the signal, the sorting mechanism 140 physically diverts the garment 10 from the main processing path into a designated sorting bin 142. In some embodiments, the sorting mechanism 140 directs the sorted garments into at least two primary storage areas based on the sorting signal: a recyclable garment storage area and a reusable garment storage area. Garments for which the sorting signal indicates they are suitable for "reusable" status (e.g., based on a high GCI score, minimal defects, desirable brand) are diverted to the reusable garment storage area. Meanwhile, other garments deemed only suitable for material recycling are diverted to the recyclable garment storage area for further classification. This further classification within the recyclable stream can be based on various factors, such as garment type (e.g., t-shirts, trousers), primary material composition (e.g., cotton, polyester), colour, or other relevant parameters. To further enhance automation and efficiency, in some embodiments, the hangers used to suspend the garments on the conveyance mechanism 110 are configured as rotatable or actuatable hangers. Specifically, for garments diverted to the recyclable garment storage area for further classification, the hanger's support end can be actuated to automatically release the garment. This automatic release mechanism allows the garment to be dropped or transferred into the appropriate downstream classification bin or conveyor without requiring manual intervention, thereby increasing the overall automation level of the system, improving sorting efficiency, and reducing labour participation. The automated sorting system for recycled garments according to the present invention, as described above, achieves full-process automation of garment handling—from loading and movement to sorting—through the cooperative operation of an automated hanging loading system and a conveyor mechanism. This design substantially reduces manual intervention, significantly improves sorting speed, and enhances system throughput. By enabling precise sorting, the system improves the quality of recyclable materials, thereby supporting the development of a circular economy. Furthermore, the system adopts a modular design, making it easy to integrate into existing sorting lines and supporting future functional expansions, such as the addition of thermal imaging or RFID identification modules. Consequently, the system possesses excellent compatibility and scalability. HK 30137917 A 24 In some embodiments, a single automated sorting system apparatus according to the present invention can process up to 1.7 tons of garments per day. In contrast to traditional sorting methods that rely on manual labour or single-modal optical sorting—which suffer from low efficiency, poor consistency, and often feature closed designs that are difficult to upgrade or expand—the present system reduces processing time to the second level, making it highly suitable for large-scale recycling scenarios. The automated sorting system according to the present invention integrates multi-spectral cameras, including RGB, Near-Infrared (NIR), and Ultraviolet (UV) cameras, in combination with AI vision algorithms. This integration enables high-precision identification of garments across multiple dimensions, including material composition, color, defects, and brand. While traditional near-infrared spectroscopy can only identify material composition and cannot detect defects or brands, the present system achieves comprehensive judgment through multi-spectral fusion and AI analysis. This results in higher accuracy, particularly in recycling scenarios involving blended fabrics and garments with complex patterns. The automated sorting system according to the present invention generates a Garment Circulability Index (GCI) through its algorithms. Based on factors such as brand, quality, and material, the GCI provides a standardized assessment of a garment's value, offering data- driven support for decisions regarding resale, recycling, and other downstream processes. Traditional sorting methods typically classify garments based solely on material or appearance, lacking a dimension for value assessment. The GCI introduced by the present invention provides a quantifiable quality and value metric for the garment circulation market, enabling more informed and economically viable sorting decisions. In some embodiments, the automated sorting system utilizes computer vision techniques to process video streams and extract key images. This is necessary because, during the few seconds a garment passes through the imaging station 120, swinging motion may occur, potentially generating blurred frames. Therefore, the system is configured to select the frame where the garment appears flattest, most stationary, and sharpest for subsequent analysis of brand, defects, material, and other characteristics by the computing system 130. The frame selection process may proceed as follows: First, a background image of the empty imaging station is loaded and converted to a grayscale image. Subsequently, video frames are read in a loop. Each current colour image frame is converted to grayscale to simplify calculations, and background subtraction is performed against the loaded background image to isolate the garment. HK 30137917 A 25 Advantageously, the computing system maintains a fixed-length frame buffer, for example, storing the difference results (indicating motion) from the most recent 10 frames. It then analyses the data within this current sliding window buffer, such as calculating the proportion of frames within the buffer that are classified as containing "motion." The system applies a dynamic threshold to determine whether the video stream at a given moment is in a "motion state" or a "stationary state." Generally, frames with less motion are more stable and clearer. Finally, the computing system compares the cumulative motion metric for all frames within a detected stationary segment. The frame with the smallest cumulative motion value is selected as the moment within that segment with the least motion, the most stable image, and the highest likelihood of being sharp. This frame is then extracted as the key image from the video segment for further analysis. This workflow combines background subtraction, sliding window analysis, and dynamic thresholding to detect significant video segments. Motion tracking can be efficiently implemented using a double-ended queue (deque), while segment detection may utilize a state machine to ensure robustness. Finally, stability assessment techniques are applied to select the optimal frame for output. In some embodiments, when analysing spectral image data, the automated sorting system detects pilling on garments by combining near-infrared (NIR) imaging with standard image processing techniques. An exemplary workflow is illustrated in Figures 8A-B. First, an NIR image of the garment is captured by an NIR camera at step A1. The computing system 130 then processes the acquired image at step A2 by applying sharpening and Gaussian blur filters. Sharpening enhances relevant pilling features, while Gaussian blur reduces irrelevant noise. Subsequently, an edge detection algorithm, such as Canny, Laplacian, HED (Holistically- Nested Edge Detection), or RCF (Richer Convolutional Features), is applied at step A3 to highlight the subtle, localized texture variations characteristic of pilling. The detected edge details corresponding to pilling features are further processed at step A4 to extract contours that correspond to raised fibre areas (pills). These contours can then be overlaid onto the original image for visualization. Finally, the number and area of the pilling regions are quantified to derive an objective pilling score which is output at step A5. This quantitative assessment effectively distinguishes between normal garments and those affected by pilling. HK 30137917 A 26 According to another embodiment, the automated sorting system employs a multi-stage computer vision pipeline to accurately detect yellow stains on garments. An exemplary workflow is illustrated in Figures 9A-B. First, the computing system applying the steps depicted applies a Gaussian blur to an input RGB image to reduce feature noise and smooth the image at step B1. The resulting smoothed image is then converted to a colour model that aligns more closely with human visual perception, specifically the Hue, Saturation, Value (HSV) colour space. A predefined range of hue values corresponding to the colour yellow is used at step B2 to create a binary mask, thereby highlighting all yellow regions within the image. The step B2 may be followed by further morphological operations to clean the mask at step B3, removing isolated noise pixels and ensuring only significant yellow areas are retained. The yellow regions identified by the mask may include various yellow objects, such as stains, labels, buttons, or design elements. Therefore, the extracted yellow regions require further analysis at step B4. As illustrated in Figure 9C, an AI classification model, such as YOLOv11n-cls, a Vision Transformer (ViT), or a Convolutional Neural Network (CNN), can be employed to distinguish between actual yellow stains and other benign yellow objects at step B5. The model is trained on labelled datasets of stains and non-stain yellow items to perform this classification accurately. In some embodiments, the automated sorting system performs garment brand recognition using Optical Character Recognition (OCR). This process may include the following steps: First, a clear image of the garment's brand label is captured by a camera within the system. An OCR recognition engine, stored in the memory of the computing system 130, then processes the image to extract textual brand names. Finally, the extracted text is cross- referenced against a pre-stored internal brand database or list to match and classify the garment's brand. In other embodiments, particularly for recycled garments featuring logos or stylized brand labels that may be challenging for OCR, the automated sorting system employs object detection methods for brand recognition. The process for using object detection may include the following steps: Initially, a large and diverse dataset of images containing various garment brands is collected. This dataset should encompass different brand tiers (e.g., high- grade, mid-grade, low-grade), various brand identification locations (e.g., neck labels, care labels, chest prints), diverse shooting conditions (lighting, angle, sharpness), and different states of the brand identification (e.g., wrinkled, folded, partially occluded). HK 30137917 A 27 Subsequently, this collected dataset is used to train an object detection model, such as YOLOv11n, Faster R-CNN, or RetinaNet, to detect and classify brand logos or labels directly within images. In some embodiments, the defect detection process within the automated sorting system may include steps as illustrated in Figure 10A. First, the computing system loads video or image data acquired by the multiple cameras under multi-spectral illumination conditions (e.g., RGB, UV, NIR) at step C1. The computing system then processes the images using one or more trained object detection models to identify and localize target defects at step C2, such as holes, stains, or tears. Finally, as shown in Figure 10B, the computing system can present the output results visually on the system's user control and feedback UI (User Interface) at step C3. This visualized output may include bounding boxes or segmentation masks indicating the location and type of each detected defect, providing operators with clear, interpretable feedback on the system's analysis. According to another embodiment, the automated sorting system further comprises a Short- Wave Infrared (SWIR) camera configured to detect the material composition of a garment. The SWIR camera can capture SWIR images or, in more advanced implementations, hyperspectral data cubes containing spatial information and a complete SWIR spectrum for each pixel. The computing system is configured to extract key spectral features from this data, such as absorption depths at specific wavelengths and overall spectral shape. These extracted features are compared against a pre-established spectral database of textile materials (e.g., cotton, polyester, wool, nylon) to identify the primary material composition of the garment, which is particularly useful for analysing blended fabrics. In some embodiments, the automated sorting system employs dual RGB cameras for colour detection. When a garment is suspended, folds and shadows are inevitable. A single camera perspective might lead to uneven colour sampling. Using a dual-camera setup allows the system to acquire more surface information. By algorithmically fusing the data from both cameras, the system obtains a more comprehensive and accurate representation of the garment's true colour. Additionally, side-mounted cameras are advantageous for capturing the colour of labels inside collars or sleeves, or the colour of linings, which can be auxiliary information for brand and style recognition. HK 30137917 A 28 The computing system is configured to extract RGB values from all non-background pixels in the pre-processed images from both cameras. These RGB values are then converted to a perceptually uniform colour space, such as CIELAB (Lab). The system maintains a pre- stored palette database containing multiple (e.g., 6-12) standard colours, defined by their Lab values. This palette is preferably established based on industry standards, such as Pantone textile colour cards, or common garment colours. The computing system is configured to calculate the colour difference between the average Lab value of the garment's main region and the Lab value of each standard colour in the palette, typically using an advanced colour difference formula like CIEDE2000. The palette colour yielding the smallest colour difference is identified as the garment's primary colour. Furthermore, the automated sorting system can also detect the degree of fading on a garment. The specific steps may include: First, the system identifies the garment's brand and style using an object detection model. Based on this identification and the primary colour matching result (e.g., "dark denim blue"), the computing system retrieves a reference standard "new" Lab value for that specific brand, style, and colour from a database. This value serves as the baseline. Finally, using the representative Lab value of the current garment's main colour area obtained from the colour inspection module, the system calculates the colour difference (e.g., using CIEDE2000) between the current garment and the standard reference. The magnitude of this calculated colour difference quantifies the degree of fading. According to further embodiments, the automated sorting system may include additional sensors to enhance its analytical capabilities. For instance, a structured light camera or a high-resolution RGB camera coupled with specific image analysis algorithms may be employed to detect the fabric weave structure of the garment, distinguishing between types such as woven, knitted, or non-woven. Additionally, a depth-sensing camera (e.g., a time-of-flight or stereo vision camera) may be included to capture three-dimensional information about the garment's shape. This depth data can be analysed by the computing system 130 to assist in determining the garment type, such as distinguishing between trousers, shirts, T-shirts, jackets, dresses, or sweaters, based on its overall form and proportions. According to another aspect of the present invention, a computer-based method for automated sorting of recycled garments is provided. Figure 11 depicts a flow chart depicting an exemplary embodiment of such a method. The method begins at step S1 (Start). HK 30137917 A 29 Step S2: Transporting the Garment In step S2, individual garments are suspended on a conveyance mechanism and transported along a processing path. As described in previous embodiments, the conveyance mechanism 110 is configured to hang garments 10 using hangers or hooks and move them sequentially, for example, via a motor-driven conveyor, towards the imaging station 120. This automated hanging transport ensures that garments are presented consistently for subsequent inspection. Step S3: Capturing Image Data In step S3, multi-spectral image data of each garment is captured using a plurality of cameras. As detailed in Sections 2 and 3, the imaging station 120 comprises multiple cameras, including but not limited to RGB, UV, NIR, and optionally LIR cameras. The illumination system, with its programmable and multi-wavelength lighting (e.g., LED, UV, NIR lights), is activated to ensure optimal and consistent imaging conditions. The capture process may involve the "stop-shoot-go" mechanism described in Section 2.3, where the conveyance mechanism pauses to allow for stable, blur-free image acquisition. Furthermore, as described in Section 10.1, the system may employ intelligent frame selection techniques to extract the sharpest, most stable image from a short video sequence captured during the pause. Step S4: Analysing Captured Image Data with AI Models In step S4, the captured multi-spectral image data is analysed using at least one trained artificial intelligence (AI) model to identify at least one characteristic of the garment. The computing system 130, which may include on-device AI camera modules and a central terminal computer as described in Section 4, processes the image data. The identified characteristics are selected from the group consisting of a defect, a brand, a material composition, and a colour. This analysis step may encompass several specialized sub-processes, including but not limited to: Defect Detection: Utilizing object detection models (e.g., Faster R-CNN, YOLOv11n) as described in Section 10.5, or specialized algorithms for specific defects like pilling (Section 10.2) and yellow stains (Section 10.3). Brand Recognition: Employing OCR for text-based labels or object detection models for logos, as detailed in Section 10.4. Material Composition Analysis: Analysing SWIR or NIR spectral data against material databases, as described in Section 10.6. HK 30137917 A 30 Colour and Fading Analysis: Using dual RGB camera data and colour space conversion (e.g., Lab) with colour difference formulas (e.g., CIEDE2000) to determine primary colour and fading level, as described in Section 10.7. GCI Score Calculation: Aggregating the outputs from the various detection modules to calculate a Garment Circulability Index (GCI) score, for example, by using a multi-feature regression or transformer-based model on a generated feature vector (e.g., the 1×173 vector), as detailed in Section 5. Step S5: Generating a Sorting Signal In step S5, a sorting signal is generated based on the identified characteristic(s) and / or the calculated GCI score. The computing system 130, specifically the terminal computer in the described architectures, consolidates the analysis results and determines an appropriate destination for the garment (e.g., reusable vs. recyclable, or a specific bin for a particular material or brand tier). This determination is encoded into the sorting signal. Step S6: Automatically Sorting the Garment In step S6, the garment is automatically diverted into a designated sorting bin based on the sorting signal. As illustrated in Figure 7 and described in Section 6, the sorting mechanism 140, which may be a diverter arm or a sorting gate, receives the signal from the computing system 130 and physically directs the garment 10 from the main processing path P into the appropriate sorting bin 142. In some embodiments, this step may further involve the automatic release of the garment from specialized rotatable hangers, as described in Section 6, to facilitate further downstream processing. The method concludes at step S7 (End). This method, implemented by the system described in the preceding embodiments, enables a fully automated, high-throughput, and intelligent sorting process for recycled garments, achieving the advantages of efficiency, accuracy, and scalability previously discussed. The present disclosure solves several critical technical problems in the field of recycled garment sorting. The disclosure addresses the problem of labour-intensive, inefficient, and subjective traditional sorting processes. The system and method of the present disclosure achieve full‑process automation from garment loading through imaging, analysis, and physical sorting, thereby significantly improving sorting speed and consistency while reducing reliance on manual intervention. This provides an integrated, intelligent solution that addresses multiple technical problems simultaneously. HK 30137917 A 31 The disclosure teaches a solution which addresses functional limitations of existing systems that operate in isolation. Synchronized acquisition and comprehensive analysis is performed across multiple critical dimensions simultaneously, including material composition, defect detection, brand recognition, and colour assessment. The problem of the inability of conventional visible‑light systems to detect certain types of defects is also addressed. The present disclosure addresses this by introducing multi‑spectral imaging technology. Specifically, ultraviolet imaging is utilized to detect special stains such as oil and blood that become visible only under ultraviolet light, while near‑infrared imaging enhances the visibility of surface texture anomalies such as pilling that are difficult to perceive under standard illumination. The system and method of the disclosure addresses the problem of reliance of Optical Character Recognition (OCR) technology on pure text labels. The present disclosure overcomes this by combining object detection with OCR to enable broad brand recognition and classification for graphical logos, text‑based labels, and even garments without visible labels, thereby significantly expanding identification coverage. The absence of an objective, quantifiable evaluation system is also addressed by the present disclosure. The present disclosure addresses this by constructing a Garment Circulability Index (GCI) that fuses discrete multi‑source detection data including brand value, physical condition, and material properties into a unified value score, thereby providing a data‑driven basis for sorting decisions. The challenges of coordinating multi‑sensor systems, edge‑to‑cloud artificial intelligence collaboration, and integration with automated actuation mechanisms. The present disclosure ensures that the complete workflow from imaging and analysis to sorting decision and physical diversion can be completed within seconds to meet the throughput requirements of industrial‑scale recycling operations. By solving the foregoing and additional technical problems, the present disclosure constructs an efficient, accurate, and scalable intelligent sorting system that provides critical technical support for the circular economy transition within the textile and apparel industry. The present disclosure provides practical advantages for day-to-day recycling operations. By recording the images, detected features, sorting results, and optional RFID information for each garment, the system can build a useful digital history of the garments passing through the facility. This can help operators review sorting decisions, track garment quality, identify recurring issues, and better understand the types of clothing being received from different sources. HK 30137917 A 32 The system of the present disclosure is flexible. As new requirements arise, additional cameras, sensors, defect categories, brand databases, or AI models can be added without replacing the whole system. This allows the system to adapt over time as recycling standards, market demands, and garment materials continue to change. The system of the present disclosure improve itself with use. When an operator corrects an error or labels a new type of defect, brand, or material, that information can be used to retrain the AI models. As a result, the system may become more accurate and reliable over time, especially when it is used in different regions or facilities where garment types and conditions may vary. The present system helps increase the value recovered from used garments. Instead of sorting garments only by appearance or material, the system can consider several factors together, such as brand, condition, defects, colour, and material composition. This allows garments to be directed to a more suitable destination, such as resale, repair, reuse, or material recycling. In this way, the system can reduce unnecessary downgrading of garments and help recycling facilities make better commercial and environmental decisions. The present disclosure supports more connected and data-driven recycling operations. For example, information collected from multiple sorting systems could be used to improve AI models, compare sorting performance, forecast material supply, and support reporting requirements. This would allow the system to contribute not only to individual sorting facilities, but also to the wider development of a more efficient and transparent textile recycling industry. The above embodiments are described by way of example only. Many variations are possible without departing from the scope of the disclosure as defined in the appended claims. For clarity of explanation, in some instances the present technology may be presented as including individual functional blocks including functional blocks comprising devices, device components, steps or routines in a method embodied in software, or combinations of hardware and software. Methods according to the above-described examples can be implemented using computer- executable instructions that are stored or otherwise available from computer readable media. Such instructions can comprise, for example, instructions and data which cause or otherwise configure a general purpose computer, special purpose computer, or special purpose processing device to perform a certain function or group of functions. Portions of computer resources used can be accessible over a network. The computer executable instructions may be, for example, binaries, intermediate format instructions such as assembly language, HK 30137917 A 33 firmware, or source code. Examples of computer-readable media that may be used to store instructions, information used, and / or information created during methods according to described examples include magnetic or optical disks, flash memory, Universal Serial Bus (USB) devices provided with non-volatile memory, networked storage devices, and so on. Devices implementing methods according to these disclosures can comprise hardware, firmware and / or software, and can take any of a variety of form factors. Typical examples of such form factors include laptops, smart phones, small form factor personal computers, personal digital assistants, and so on. Functionality described herein also can be embodied in peripherals or add-in cards. Such functionality can also be implemented on a circuit board among different chips or different processes executing in a single device, by way of further example. The instructions, media for conveying such instructions, computing resources for executing them, and other structures for supporting such computing resources are means for providing the functions described in these disclosures. Although a variety of examples and other information was used to explain aspects within the scope of the appended claims, no limitation of the claims should be implied based on particular features or arrangements in such examples, as one of ordinary skill would be able to use these examples to derive a wide variety of implementations. Further and although some subject matter may have been described in language specific to examples of structural features and / or method steps, it is to be understood that the subject matter defined in the appended claims is not necessarily limited to these described features or acts. For example, such functionality can be distributed differently or performed in components other than those identified herein. Rather, the described features and steps are disclosed as examples of components of systems and methods within the scope of the appended claims. HK 30137917 A Claims 1. A system for automated sorting of recycled garments, comprising: a conveyance mechanism configured to transport suspended a plurality of separate individual garments along a processing path; at least one imaging station positioned along the processing path, the imaging station comprising: a plurality of cameras configured to capture image data of a garment from the plurality of separate individual garments suspended on the conveyance mechanism passing therethrough; an illumination system comprising one or more light sources; a computing system in communication with the plurality of cameras, the computing system comprising at least one processor and a memory storing instructions that, when executed, cause the computing system to: analyse captured image data of the garment by at least one trained artificial intelligence (AI) model executing on the processor, the AI model configured to identify at least one characteristic of the garment selected from the group comprising defect, brand, material composition, and colour; and generate a sorting signal based on the identified characteristic; and a sorting mechanism coupled to the computing system configured to receive the sorting signal and route the garment via the conveyance mechanism into a predetermined sorting area according to the sorting signal. 2. The system according to claim 1, wherein the at least one imaging station comprises a first inspection booth having at least one camera disposed therein, said at least one camera configured to capture image data of a front side of the garment, and a second inspection booth having at least one camera disposed therein, said at least one camera configured to capture image data of a rear side of the garment. 3. The system according to claim 1 or 2, wherein at least some of the plurality of cameras capture image data of the garment in a spectral band which is different from the spectral band of other cameras of the plurality of cameras. 4. The system according to any one of the preceding claims, wherein the system further comprises at least one multi-spectral booth comprising at least one multi-spectral camera disposed therein configured to capture image data of the garment in a plurality of electromagnetic spectral bands. 1 HK 30137917 A 5. The system according to any one of the preceding claims, wherein the plurality of cameras is selected from the group comprising a Red-Green-Blue (RGB) camera, an Ultraviolet (UV) camera, a Near-Infrared (NIR) camera, a Long-Wave Infrared (LIR) camera, a multispectral camera, and a Short-Wave Infrared (SWIR) camera. 6. The system according to claim 5, wherein the material composition is identified using either the Short-Wave Infrared (SWIR) camera or the multispectral camera. 7. The system according to any one of the preceding claims, wherein the one or more light sources of the illumination system are selected from the group comprising Light Emitting Diode (LED) light source, strobe light source, visible monochromatic light source, backlight source, coaxial light source, diffuse light source, near-infrared (NIR) light source, short-wave infrared (SWIR) light source, ultraviolet (UV) light source, and multi-spectral light source. 8. The system according to any one of the preceding claims, wherein the defect includes at least one selected from the group comprising pilling, yellow stain, oil stain, blood stain, hole, colour fading, missing button, broken zipper and mildew. 9. The system according to claim 8, wherein the colour is identified using dual RGB cameras, and the colour fading is identified by calculating a Delta E (ΔE) colour variation value. 10. The system according to claim 9, wherein the at least one imaging station has at least a first Red-Green-Blue (RGB) camera and a second Red-Green-Blue (RGB) camera, wherein at least one of the first or second RGB cameras being movably mounted within the imaging station to scan the garment. 11. The system according to claim 9, wherein the plurality of cameras and the light sources are fixed in position within the at least one imaging station. 12. The system according to any one of the preceding claims, wherein the system further comprises one or more modules selected from: (a) a pilling detection module for detecting pilling using edge detection and contour extraction on image data comprising near-infrared (NIR) image data; (b) a yellow stain detection module configured for detecting the presence of yellow stain by performing colour segmentation in Hue-Saturation-Value (HSV) colour space on image data comprising Red-Green-Blue (RGB) image data, and analysis thereafter by a trained artificial Intelligence classification (AI) classification model; 2 HK 30137917 A (c) a defect detection module for detecting general defects by analysing the image data using a defect detection Artificial Intelligence (AI) object detection model; and (d) a brand identification module for identifying a brand of the garment using Optical Caracter Recognition (OCR) to extract text from a label, and / or by using an object detection model to recognize a brand logo. 13. The system according to any one of the preceding claims, wherein the computing system is further configured to aggregate the identified characteristic(s) of the garment into a standardized dataset, and to calculate a Garment Circulability Index (GCI) score using a multi-feature regression model or a transformer-based regression model. 14. The system according to any one of the preceding claims, wherein the conveyance mechanism includes a braking mechanism which automatically stops the conveyance mechanism when the garment enters the at least one imaging stations to enable image capture. 15. The system according to any one of the preceding claims, wherein at least one camera of the plurality of cameras includes an embedded processor for initial image processing. 16. The system according to any one of the preceding claims, further comprises: a data storage device for receiving data across a network from the computing system for storing thereon; an interface device communicatively coupled to the computing system for control thereof; and a remotely located training server for generating one or more artificial intelligence models using data received from the data storage device. 17. The system according to claim 16, wherein the remotely located training server is configured for developing updated software for updating the computing system. 18. The system according to any one of the preceding claims, wherein the sorting mechanism comprises a diverter arm or gate configured to direct the garment into one of a plurality of sorting bins based on the sorting signal. 19. A method for automated sorting of recycled garments, comprising: moving by a conveyance mechanism a plurality of separate suspended individual garments along a processing path; 3 HK 30137917 A capturing by a plurality of cameras image data of a garment of the plurality of separate suspended individual garments, said garment being illuminated by one or more light sources; analysing by a processor of a computing system the captured image data of the garment executing at least one trained artificial intelligence (AI) model thereon; identifying by the trained artificial intelligence model at least one characteristic of the garment selected from the group comprising defect, brand, material composition, and colour; generating by the processor a sorting signal based on the identified characteristic; and moving the garment into a predetermined sorting area according to the sorting signal. 20. The method according to claim 19, further comprising using a first inspection booth having at least one camera disposed therein, and a second inspection booth having at least one camera disposed therein. 21. The method according to claim 19 or 20, further comprising capturing image data of the garment by at least some of the plurality of cameras in a spectral band, and capturing image data of the garment by other cameras of the plurality of cameras in a different spectral band. 22. The method according to any one of claims 19 to 21, further comprising capturing image data of the garment in a plurality of electromagnetic spectral bands using at least one multi-spectral camera disposed in at least one multi-spectral booth. 23. The method according to any one of claims 19 to 22, wherein the plurality of cameras is selected from the group comprising a Red-Green-Blue (RGB) camera, an Ultraviolet (UV) camera, a Near-Infrared (NIR) camera, a Long-Wave Infrared (LIR) camera, a multispectral camera, and a Short-Wave Infrared (SWIR) camera. 24. The method according to claim 23, wherein the identified characteristic is material composition; and said material composition identification is performed by the trained AI model analysing image data captured by either the Short-Wave Infrared (SWIR) camera or the multispectral camera. 25. The method according to any one of the preceding claims, wherein the one or more light sources are selected from the group comprising a Light Emitting Diode (LED) light source, a strobe light source, a visible monochromatic light source, a backlight source, a coaxial light source, a diffuse light source, a near-infrared (NIR) light source, a short-wave infrared (SWIR) light source, an ultraviolet (UV) light source, and a multi-spectral light source. 4 HK 30137917 A 26. The method according to any one of claims 19 to 25, wherein the identified characteristic is a defect in the garment; wherein the identified defect is a defect selected from the group comprising pilling, yellow stain, oil stain, blood stain, hole, colour fading, missing button, broken zipper, and mildew. 27. The method according to claim 26, wherein the identified characteristic is the colour of the garment; and wherein said colour is identified using image data captured by dual RGB cameras, and wherein a defect of colour fading is identified by calculating a Delta E (ΔE) colour variation value. 28. The method according to claim 27, further comprising using at least one of a first or second RGB cameras to movably scan the garment. 29. The method according to claim 27, further comprising fixing the position of the plurality of cameras and the light sources. 30. The method according to any one of claims 19 to 29, further comprising performing one or more of the following: (a) detecting by the trained artificial intelligence model executing of the processor the presence of pilling by using edge detection and contour extraction on image data comprising near-infrared (NIR) image data; (b) detecting by the trained artificial intelligence model executing of the processor the presence of a yellow stain by performing colour segmentation in Hue-Saturation-Value (HSV) colour space on image data comprising RGB image data, followed by analysis using a trained artificial intelligence classification model; (c) detecting by the trained artificial intelligence model executing of the processor the presence of general defects by analysing the image data using a defect detection AI object detection model; and (d) identifying by the trained artificial intelligence model executing of the processor a brand of the garment by using Optical Character Recognition (OCR) to extract text from a label, and / or by using an object detection model to recognize a brand logo. 31. The method according to any one of claims 19 to 30, further comprising aggregating the identified characteristic of the garment into a standardized dataset, and calculating a Garment Circulability Index (GCI) score using a multi-feature regression model or a transformer-based regression model. 32. The method according to any one of claims 19 to 31, further comprising automatically stopping the conveyance mechanism to enable image capture. 5 HK 30137917 A 33. The method according to any one of claims 19 to 32, further comprising embedding a processor with at least one camera of the plurality of cameras for initial image processing. 34. The method according to any one of claims 19 to 33, further comprising: receiving data from a computing system and storing said data in a data storage device; controlling the computing system via an interface device communicatively coupled thereto; and generating one or more artificial intelligence models using data received from the data storage device by a remotely located training server. 35. The method according to claim 34, further comprising developing updated software for updating the computing system by the remotely located training server. 36. The method according to any one of claims 19 to 35, wherein moving the garment into a predetermined sorting area further comprises releasing the garment to one of a plurality of sorting bins based on the sorting signal from the conveyance mechanism. 6 HK 30137917 A FIG. 1 FIG. 2A 1 HK 30137917 A FIG. 2B FIG. 2C 2 HK 30137917 A FIG. 3A 3 HK 30137917 A FIG. 3B 4 HK 30137917 A FIG. 3C 5 HK 30137917 A FIG. 3D 6 HK 30137917 A FIG. 4A FIG. 4B 7 HK 30137917 A FIG. 5A 8 HK 30137917 A FIG. 5B 9 HK 30137917 A FIG. 6 FIG. 7 10 HK 30137917 A FIG. 8A FIG. 8B 11 HK 30137917 A FIG. 9A FIG. 9B 12 HK 30137917 A FIG. 9C FIG. 10A 13 HK 30137917 A FIG. 10B 14 HK 30137917 A FIG. 11 15 HK 30137917 A