Intelligent management wardrobe system based on AIAgent

Abstract

The application discloses an AI Agent-based intelligent management wardrobe system and relates to the technical field of intelligent wardrobes. The application utilizes a cloud multi-modal large model to perform deep semantic analysis on clothes, automatically outputs multi-dimensional semantic tags such as categories, colors, materials, patterns, styles, suitable seasons, use scenarios and care needs, and generates a five-dimensional structured digital asset card, realizing complete mapping of clothes from a physical entity to a digital asset. Through the fusion of natural language requirements, real-time weather, schedule scenarios and multi-source fusion query vectors of personal preferences, retrieval is performed, and based on the three target weighted comprehensive scores of style uniformity, scene adaptability and season adaptability, candidate solutions are optimized and sorted, significantly improving the personalization of dressing recommendations. Furthermore, the AI Agent plans a collision-free optimal path based on the A* algorithm, combines material perception-based adaptive care procedures, automatically generates a standardized hardware execution task sequence, and realizes full-process automation from recommendation to delivery.

Description

Technical Field

[0001] This invention relates to the field of smart wardrobe technology, specifically to a smart wardrobe management system based on AI Agent. Background Technology

[0002] With the continuous improvement of people's living standards and the upgrading of consumption concepts, the number of personal clothes continues to grow. Daily management, matching, and care of clothing have become frequent pain points in modern family life. Traditional clothing management methods mainly rely on manual memory or simple paper / electronic lists, which are not only inefficient but also unable to provide intelligent outfit recommendations based on multi-dimensional contextual information such as weather changes, schedules, and personal style preferences. In recent years, with the rapid development of artificial intelligence, the Internet of Things, and robotics, smart wardrobes, as an important sub-sector of smart homes, have begun to receive widespread attention. However, existing technologies still have many shortcomings in areas such as digital clothing management, intelligent outfit recommendations, and the collaborative execution of software and hardware.

[0003] In terms of clothing recommendations, existing intelligent clothing recommendation systems mainly fall into two categories: collaborative filtering-based methods and rule engine-based methods. Collaborative filtering methods recommend clothing by analyzing users' historical clothing data and the clothing preferences of similar users. However, this type of method suffers from a cold start problem and struggles to integrate real-time contextual information such as weather and daily schedules. Rule engine methods, on the other hand, match and recommend clothing based on pre-set clothing rule templates. However, the coverage of these templates is limited, their flexibility is poor, and they cannot adapt to users' personalized and dynamically changing clothing needs. Furthermore, most existing clothing recommendation systems only output a flat list of recommendation results, lacking the ability to comprehensively evaluate and optimize the ranking of multiple items in the recommendation scheme based on multi-objective factors such as style consistency, scene adaptability, and seasonal suitability. The coordination and practicality of the recommendation results need to be improved. More importantly, there is a significant disconnect between existing clothing recommendation systems and the hardware execution layer of physical wardrobes. Recommendation results cannot be automatically converted into hardware-executable grabbing paths, care procedures, and outbound action sequences. Users still need to manually complete the entire process from recommendation to clothing retrieval, resulting in a severe lack of intelligence.

[0004] In terms of garment care and hardware execution, while existing smart wardrobe products have integrated some care functions, most use fixed care parameters. They cannot automatically match differentiated care process combinations based on the material type and care needs of different garments. For example, silk requires gentle wrinkle removal rather than high-temperature drying, and wool requires low-temperature sterilization rather than strong ultraviolet irradiation. Existing solutions lack adaptive care decision-making capabilities based on material perception. Furthermore, in scenarios involving the collaborative retrieval of multiple garments, existing solutions lack path planning capabilities based on the wardrobe's internal spatial topology. They cannot plan collision-free optimal movement paths for the robotic arm from multiple storage locations to the retrieval buffer zone, nor can they optimally sequence the grasping, care, and retrieval transfer actions, resulting in low execution efficiency and a high risk of mechanical collisions. In addition, existing solutions are weak in their post-action status feedback mechanisms. They lack a closed-loop management mechanism for structured data encapsulation, integrity verification, and multi-terminal synchronous updates of the execution results of each action node, making it difficult to guarantee the consistency and traceability of the system status. Summary of the Invention

[0005] To address the aforementioned technical issues, an AI Agent-based intelligent wardrobe management system is provided. This technical solution resolves the problems mentioned in the background section.

[0006] To achieve the above objectives, the technical solution adopted by the present invention is as follows:

[0007] In a first aspect of the present invention, an AI Agent-based intelligent wardrobe management system is provided, comprising:

[0008] The acquisition module uses a visual acquisition device to acquire multi-angle images of the clothing to be put into storage. The acquired multi-angle image data is preprocessed by sequentially performing illumination correction, size normalization and background removal. The main body area of ​​the clothing is extracted by semantic segmentation algorithm. The image of the main body area of ​​the clothing is input into a cloud multimodal big model for deep semantic analysis. The output includes multi-dimensional semantic tags containing category, color, material, pattern, style, suitable season, usage scenario and care requirements. The multi-dimensional semantic tags are associated and bound with the physical storage location code to generate structured digital asset cards and write them into the clothing digital asset library.

[0009] Outfit suggestion candidate set output module: Receives the user's outfit needs input in natural language, encodes the outfit needs into semantic vectors, integrates real-time weather data, user's daily scene tags, and personal outfit preference vectors to generate a multi-source fusion query vector, uses the query vector to retrieve a set of candidate clothes in the clothing digital asset library through vector similarity retrieval and inputs it into the large model, the large model optimizes and sorts the candidate clothes set based on multi-objective constraints of style consistency, scene adaptability, and seasonal adaptability, and outputs an outfit suggestion candidate set containing recommendation reasons;

[0010] Task sequence delivery module: In response to the user's confirmation of the target outfit plan, the AI ​​Agent parses the digital asset card of each garment in the target outfit plan, extracts the physical storage location code, plans the movement path from each storage location to the outbound buffer based on the internal spatial topology of the wardrobe, matches the combination of care procedures according to the material label and care requirement label of each garment, and arranges the storage location code, movement path and care procedure into a standardized hardware execution task sequence in chronological order and delivers it to the execution end;

[0011] Storage and update module: The execution end parses the action type, target storage location and timing relationship of each instruction in the task sequence, drives the hardware mechanism to complete the clothing grabbing, care and outbound transfer actions in sequence according to the arranged path. After each action node is completed, a status data packet containing action type, execution time and execution result is generated. After verification, it is synchronously written to the cloud database, local control terminal storage and mobile terminal cache, and the usage status field in the corresponding clothing digital asset card is synchronously updated.

[0012] Preferably, the preprocessing of the acquired multi-angle image data, including illumination correction, size normalization, and background removal, and the extraction of the main clothing region using a semantic segmentation algorithm, specifically includes:

[0013] The mechanical servo motor is controlled to rotate the clothing at no less than three preset angles. At each rotation angle, a high-definition camera captures images of the clothing. Histogram equalization is performed on the original images at each angle to eliminate uneven lighting. All angle images are uniformly mapped to a preset standard size through affine transformation.

[0014] A deep learning-based semantic segmentation network is used to perform pixel-level classification of images from each angle. Pixels that are not clothing in the classification results are set to zero. The zeroed images from multiple angles are then fused using a weighted average based on the spatial overlap of the clothing subject. The fusion weights are dynamically allocated according to the sharpness scores of each image angle. The output is standardized image data with zero background pixel values ​​and complete, unobstructed clothing subjects. The fusion weights are proportionally allocated according to the sharpness scores of each image angle.

[0015] ;

[0016] in, Indicates the first Fusion weights for images from different angles, This represents the total number of angles in the acquired images. Indicates the first The sharpness score of the images at each angle is calculated using Laplace variance:

[0017] ;

[0018] in, Indicates the first Sharpness rating of images from each angle. Represents the Laplace operator. Indicates the first An image at an angle in pixel coordinates The grayscale value at that location.

[0019] Preferably, the step of inputting the image of the main clothing area into a cloud-based multimodal large model for deep semantic analysis, outputting multidimensional semantic labels, and generating structured digital asset cards specifically includes:

[0020] Standardized image data and prior knowledge prompts for clothing categories are input into a cloud-based multimodal model. The model then outputs clothing category labels, color attribute values, material type identifiers, fit classification labels, style vectors, seasonal labels, usage scenario labels, and care requirement labels in parallel.

[0021] The aforementioned multidimensional semantic tags are associated and bound with the physical storage location code and the storage entry timestamp, respectively, to generate a five-dimensional structured digital asset card containing image identity field, semantic identity field, spatial storage location identity field, usage status identity field and care attribute identity field. The digital asset card is then written into the clothing digital asset library with a unique identifier.

[0022] Preferably, the large model optimizes and sorts the candidate clothing set based on multi-objective constraints of style consistency, scene adaptability, and seasonal adaptability, and outputs a candidate set of outfit schemes containing recommendation reasons, specifically including:

[0023] The user's natural language dressing needs are transformed into high-dimensional semantic vectors through a text encoding model. Temperature ranges and weather types from real-time weather data are obtained and encoded into weather feature vectors. Scene tags from the user's daily schedule are obtained and encoded into scene feature vectors. Style preference distribution from the user's historical dressing records is obtained and encoded into preference feature vectors.

[0024] Semantic vectors, weather feature vectors, scene feature vectors, and preference feature vectors are concatenated and fused to generate a multi-source fusion query vector. The query vector is then used to calculate cosine similarity in the style vector space of each garment in the clothing digital asset database.

[0025] ;

[0026] in, This represents a multi-source fusion query vector. Indicates the first Style vector of the garment and Represent the two vectors at the th... Dimensional components, Indicates the vector dimension;

[0027] Clothing items with similarity higher than a preset threshold are selected as a candidate clothing set. The semantic tags of each item in the candidate clothing set and the query vector are input into the large model, which calculates the style consistency score.

[0028] The scene adaptability score is calculated by combining the weather feature vector and the scene feature vector, and the season adaptability score is calculated by combining the season interval of the current date. The overall adaptability score is obtained by weighting the scene adaptability score, the season adaptability score and the style consistency score.

[0029] The candidate outfit ideas are output in descending order of scores, and the large model generates natural language recommendation reasons for each candidate idea, including explanations of style coordination, scene matching, and seasonal suitability.

[0030] Preferably, the formula for calculating the style consistency score is:

[0031] ;

[0032] in, The score indicates the consistency of style. This represents the total number of clothing items in the candidate scheme. and They represent the first Item and the first Style vector of the garment;

[0033] The formula for calculating scene adaptability score is:

[0034] ;

[0035] in, This represents the scenario adaptability score; a higher score indicates a higher degree of matching between the solution and the target scenario. This represents the total number of clothing items in the candidate scheme. Indicates the first The scene feature vector of a piece of clothing, This represents the target scenario feature vector of the user's daily schedule;

[0036] The formula for calculating the seasonal suitability score is as follows:

[0037] ;

[0038] in, This indicates the seasonality score. This indicates the number of clothing items in the candidate solution that successfully match the current season range with the appropriate season tag. This indicates the total number of clothing items in the candidate scheme;

[0039] The formula for calculating the overall fit score is as follows:

[0040] ;

[0041] in, This indicates the overall fit score. , , These are the preset weighting coefficients for style consistency, scene adaptability, and seasonal adaptability, respectively. .

[0042] Preferably, the step of arranging the storage location code, movement path, and nursing procedures into a standardized hardware execution task sequence according to time sequence and distributing it to the execution terminal specifically includes:

[0043] The AI ​​Agent reads the digital asset card of each garment in the target outfit plan, extracts the physical storage location code bound to each garment, divides the wardrobe into several spatial grids, and establishes a grid connectivity topology map.

[0044] Starting with the code of each storage location and ending with the code of the outbound buffer zone, the following method is adopted: The path planning algorithm searches for the collision-free optimal path on the topology graph, while reading the material type label and care requirement label of each garment, and searching for the set of basic care procedures that match the material type label in the preset care procedure library.

[0045] Based on the nursing needs tags, select the required procedures from the basic set, determine whether to add in-depth nursing procedures based on the storage time data of clothing, and form a nursing procedure chain by topologically sorting all the required and optional procedures according to the nursing dependencies.

[0046] Each storage location code, its corresponding optimal path, and the nursing procedure chain are concatenated and encoded according to the execution sequence to generate a standardized hardware execution task sequence containing action instruction codes, path codes, nursing instruction codes, and sequence codes. This sequence is then encapsulated using a unified instruction protocol and sent to the execution end.

[0047] Preferably, the step of synchronously writing the verified data to the cloud database, local control terminal storage, and mobile terminal cache, and synchronously updating the usage status field in the corresponding clothing digital asset card, specifically includes:

[0048] After receiving the standardized hardware execution task sequence, the execution end parses the action type code, target location code, and timing code in each instruction;

[0049] Motion control commands are sent sequentially to the stepper motor and servo motor according to the timing encoding order, driving the robotic arm to complete the clothing grabbing action. After the grabbing is completed, the digital asset card of the clothing is read to confirm its identity.

[0050] The garments are transferred to the care area, and the corresponding care peripherals are triggered sequentially according to the care instruction codes in the instructions to perform one or more care actions such as lint removal, hot air wrinkle removal, negative ion deodorization, and ultraviolet sterilization.

[0051] After the nursing action is completed, the robotic arm continues to transfer the clothes to the outbound buffer. After each action node is completed, the position sensor and status sensor collect the action completion confirmation signal.

[0052] The action type code, execution timestamp, and execution result status code are encapsulated into a status data packet, and the integrity of the data packet is verified using a CRC check algorithm.

[0053] ;

[0054] in, Indicates status data packet Cyclic Redundancy Check (CRC) code, Indicates the data packet number Bits of binary data, Indicates the total number of bits in the data packet. Represents the predefined generator polynomial;

[0055] After successful verification, the status data packet is published to the cloud message queue using the MQTT protocol. The cloud message queue then subscribes to and writes the data to the local control database and the mobile local cache, respectively. At the same time, the usage status identity field of the corresponding clothing in the clothing digital asset library is updated in an atomic manner.

[0056] Preferably, after the verification is successful, the status data packet is published to the cloud message queue using the MQTT protocol, specifically including:

[0057] After the action node is executed, the position sensor detects whether the robotic arm has reached the target storage location coordinates, and the status sensor detects whether the clothing gripping mechanism has completed the release or clamping action. If both detections pass, an action completion confirmation signal is generated. The action type code, the Unix timestamp of the action completion time, and the execution result status code are encapsulated into a status data packet according to a preset JSON field format. CRC-16 checksum calculation is performed on all bytes of the status data packet.

[0058] ;

[0059] in, Indicates status data packet CRC-16 checksum, Indicates the data packet number Bits of binary data, Indicates the total number of bits in the data packet. A polynomial is generated for the CRC-16 standard, and a checksum is appended to the end of the data packet. The execution end publishes the status data packet with the appended checksum to the cloud message queue of the specified topic via the MQTT protocol. After receiving the data packet, the cloud message queue first verifies the CRC-16 checksum. After the verification is successful, it is pushed to the local control terminal database and the mobile terminal local cache for data storage.

[0060] Preferably, updating the usage status identity field of the corresponding garment in the garment digital asset database using atomic operations specifically includes:

[0061] After the cloud message queue completes the data storage, it sends a status update request to the clothing digital asset database. The request carries the clothing's unique identifier and the target usage status code.

[0062] Upon receiving a request, the clothing digital asset library uses the unique identifier of the clothing as an index to lock the write permission of the corresponding digital asset card;

[0063] Replace the value of the status identity field with the target status code, and append a log record containing the operation type, operation time and status values ​​before and after the operation to the operation log field of the digital asset card. After unlocking the write permission, return an update confirmation response to the cloud message queue.

[0064] In a second aspect of the invention, an electronic device is also provided. The electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor to enable the at least one processor to perform the system of the first aspect of the invention.

[0065] Compared with existing technologies, this invention provides an AI Agent-based intelligent wardrobe management system, which has the following beneficial effects:

[0066] This invention effectively eliminates uneven lighting and background interference by combining multi-angle image acquisition with Laplacian sharpness scoring weighted fusion and semantic segmentation preprocessing, outputting high-quality standardized image data and laying a reliable foundation for subsequent semantic analysis. It utilizes a cloud-based multimodal large model to perform deep semantic analysis on clothing, automatically outputting multi-dimensional semantic tags such as category, color, material, fit, style, suitable season, usage scenario, and care requirements, and generating five-dimensional structured digital asset cards. This achieves a complete mapping of clothing from physical entities to digital assets, significantly improving the dimensionality and accuracy of digital clothing management. It performs retrieval through multi-source fusion query vectors that integrate natural language requirements, real-time weather, daily scenarios, and personal preferences, and optimizes and ranks candidate solutions based on a weighted comprehensive score of style consistency, scenario adaptability, and seasonal adaptability, significantly improving the personalization, coordination, and practicality of outfit recommendations. Furthermore, it leverages AI... The Agent uses the A* algorithm to plan collision-free optimal paths and combines material-aware adaptive nursing procedure matching to automatically generate standardized hardware execution task sequences, achieving full-process automation from recommendation to delivery and completely bridging the gap between software decision-making and hardware execution. Through CRC checksum and MQTT protocol, it achieves multi-terminal synchronous storage and atomic updates of action status data, ensuring the consistency, integrity and traceability of system status and building a complete closed-loop management system. Attached Figure Description

[0067] Figure 1 This is a schematic diagram of the system framework of the present invention;

[0068] Figure 2 This is a schematic diagram of the method flow for S201-S204 in this invention;

[0069] Figure 3 This is a schematic diagram of the method flow of S301-S303 in this invention;

[0070] Figure 4 This is a schematic diagram of the method flow for S401-S405 in this invention;

[0071] Figure 5 This is a schematic diagram of the method flow for S501-S504 in this invention. Detailed Implementation

[0072] The following description is intended to disclose the invention and enable those skilled in the art to implement it. The preferred embodiments described below are merely examples, and other obvious variations will occur to those skilled in the art.

[0073] Example 1

[0074] Please refer to Figure 1 As shown, in a first aspect of the present invention, an AI Agent-based intelligent wardrobe management system is provided, comprising:

[0075] Acquisition Module 101: The visual acquisition device acquires multi-angle images of the clothing to be put into storage. The acquired multi-angle image data is preprocessed by sequentially performing illumination correction, size normalization and background removal. The main body area of ​​the clothing is extracted by semantic segmentation algorithm. The image of the main body area of ​​the clothing is input into the cloud multimodal big model for deep semantic analysis. The output is a multi-dimensional semantic label containing category, color, material, pattern, style, suitable season, usage scenario and care requirements. The multi-dimensional semantic label is associated and bound with the physical storage location code to generate a structured digital asset card and written into the clothing digital asset library.

[0076] Outfit suggestion candidate set output module 102: Receives the user's outfit needs input in natural language, encodes the outfit needs into semantic vectors, integrates real-time weather data, user's daily scene tags and personal outfit preference vectors to generate a multi-source fusion query vector, uses the query vector to retrieve a set of candidate clothes in the clothing digital asset library through vector similarity retrieval and inputs it into the large model, the large model optimizes and sorts the candidate clothes set based on multi-objective constraints of style uniformity, scene adaptability and seasonal adaptability, and outputs an outfit suggestion candidate set containing recommendation reasons;

[0077] Task sequence distribution module 103: In response to the user's confirmation of the target outfit plan, the AI ​​Agent parses the digital asset card of each garment in the target outfit plan, extracts the physical storage location code, plans the movement path from each storage location to the outbound buffer based on the internal spatial topology of the wardrobe, and matches the combination of nursing procedures according to the material label and nursing requirement label of each garment. The storage location code, movement path and nursing procedure are arranged into a standardized hardware execution task sequence in chronological order and distributed to the execution end.

[0078] Storage and update module 104: The execution end parses the action type, target storage location and timing relationship of each instruction in the task sequence, drives the hardware mechanism to complete the clothing grabbing, care and outbound transfer actions in sequence according to the arranged path. After each action node is completed, a status data packet containing action type, execution time and execution result is generated. After verification, it is synchronously written to the cloud database, local control terminal storage and mobile terminal cache, and the usage status field in the corresponding clothing digital asset card is synchronously updated.

[0079] As will be understood by those skilled in the art, firstly, by combining multi-angle image acquisition with Laplacian sharpness score weighted fusion and semantic segmentation preprocessing, uneven lighting and background interference can be effectively eliminated, outputting high-quality standardized image data, laying a reliable foundation for subsequent semantic parsing.

[0080] Second, by utilizing a cloud-based multimodal big data model to perform deep semantic analysis on clothing, the system automatically outputs multi-dimensional semantic tags such as category, color, material, fit, style, suitable season, usage scenario, and care requirements, and generates five-dimensional structured digital asset cards. This enables a complete mapping of clothing from physical entities to digital assets, significantly improving the dimensionality and accuracy of digital clothing management.

[0081] Third, by integrating multi-source query vectors that combine natural language requirements, real-time weather, daily scenarios, and personal preferences, the system optimizes and ranks candidate solutions based on a weighted comprehensive score that considers three objectives: style consistency, scenario adaptability, and seasonal adaptability. This significantly improves the personalization, coordination, and practicality of outfit recommendations.

[0082] Fourth, the AI ​​Agent plans the collision-free optimal path based on the A* algorithm, and combines it with material-aware adaptive nursing procedure matching to automatically generate a standardized hardware execution task sequence, realizing full-process automation from recommendation to delivery, and completely breaking down the gap between software decision-making and hardware execution.

[0083] Fifth, by using CRC checksum and MQTT protocol to achieve multi-terminal synchronous storage and atomic updates of action status data, the consistency, integrity and traceability of system status are ensured, and a complete closed-loop management system is constructed.

[0084] Please refer to Figure 2 As shown, the acquired multi-angle image data undergoes preprocessing including illumination correction, size normalization, and background removal. The main clothing region is then extracted using a semantic segmentation algorithm, specifically including:

[0085] S201. Control the mechanical servo motor to drive the clothing to rotate at least three preset angles, and capture images of the clothing through a high-definition camera at each rotation angle;

[0086] S202. Perform histogram equalization on the original image of each angle to eliminate uneven lighting, and map all angle images to a preset standard size through affine transformation.

[0087] S203. A semantic segmentation network based on deep learning is used to perform pixel-level classification of each angle image, and the pixels of non-clothing categories in the classification results are set to zero.

[0088] S204. After the images from multiple angles have been zeroed out, a weighted average fusion is performed on the spatial overlapping areas of the main body of the clothing. The fusion weights are dynamically allocated according to the sharpness scores of the images from each angle, and standardized image data with zero background pixel values ​​and the main body of the clothing intact and unobstructed is output.

[0089] The fusion weights are proportionally allocated based on the sharpness scores of the images from each angle:

[0090] ;

[0091] in, Indicates the first Fusion weights for images from different angles, This represents the total number of angles in the acquired images. Indicates the first The sharpness score of the images at each angle is calculated using Laplace variance:

[0092] ;

[0093] in, Indicates the first Sharpness rating of images from each angle. Represents the Laplace operator. Indicates the first An image at an angle in pixel coordinates The grayscale value at that location.

[0094] Please refer to Figure 3 As shown, the image of the main clothing area is input into a cloud-based multimodal large model for deep semantic analysis, outputting multidimensional semantic labels and generating structured digital asset cards, specifically including:

[0095] S301. Input the standardized image data and the prior knowledge prompts for clothing categories into the cloud multimodal big model. The big model outputs clothing category labels, color attribute values, material type identifiers, pattern classification labels, dressing style vectors, suitable season labels, usage scenario labels and care requirement labels in parallel.

[0096] S302. Associate and bind the above multidimensional semantic tags with the physical storage location code and the storage entry timestamp respectively to generate a five-dimensional structured digital asset card containing image identity field, semantic identity field, spatial storage location identity field, usage status identity field and nursing attribute identity field.

[0097] S303. Write the digital asset card with a unique identifier into the clothing digital asset library.

[0098] Please refer to Figure 4 As shown, the large model optimizes and ranks the candidate clothing set based on multi-objective constraints of style consistency, scene adaptability, and seasonal adaptability, and outputs a candidate set of outfit schemes containing recommendation reasons, specifically including:

[0099] S401. Transform the user's natural language dressing needs into a high-dimensional semantic vector through a text encoding model, obtain the temperature range and weather type from the real-time weather data and encode them into a weather feature vector, obtain the scene tags from the user's daily schedule and encode them into a scene feature vector, and obtain the style preference distribution from the user's historical dressing records and encode it into a preference feature vector.

[0100] S402. Concatenate and fuse the semantic vector, weather feature vector, scene feature vector, and preference feature vector to generate a multi-source fusion query vector. Calculate the cosine similarity of the query vector in the style vector space of each garment in the clothing digital asset database.

[0101] ;

[0102] in, This represents a multi-source fusion query vector. Indicates the first Style vector of the garment and Represent the two vectors at the th... Dimensional components, Indicates the vector dimension;

[0103] S403. Select clothing items with similarity higher than a preset threshold as a candidate clothing set. Input the semantic label of each item in the candidate clothing set along with the query vector into the large model, and the large model calculates the style consistency score.

[0104] S404. Combine the weather feature vector and the scene feature vector to calculate the scene adaptability score, combine the seasonal interval of the current date to calculate the seasonal adaptability score, and use the weighted sum of the scene adaptability score, seasonal adaptability score and style consistency score as the comprehensive adaptability score.

[0105] S405. Output the candidate outfit schemes in descending order of scores, and the large model generates natural language recommendation reasons for each candidate scheme, including style coordination, scene matching, and seasonal suitability.

[0106] The formula for calculating the style consistency score is:

[0107] ;

[0108] in, The score indicates the consistency of style. This represents the total number of clothing items in the candidate scheme. and They represent the first Item and the first Style vector of the garment;

[0109] The formula for calculating scene adaptability score is:

[0110] ;

[0111] in, This represents the scenario adaptability score; a higher score indicates a higher degree of matching between the solution and the target scenario. This represents the total number of clothing items in the candidate scheme. Indicates the first The scene feature vector of a piece of clothing, This represents the target scenario feature vector of the user's daily schedule;

[0112] The formula for calculating the seasonality score is:

[0113] ;

[0114] in, This indicates the seasonality score. This indicates the number of clothing items in the candidate solution that successfully match the current season range with the appropriate season tag. This indicates the total number of clothing items in the candidate scheme;

[0115] The formula for calculating the overall fit score is as follows:

[0116] ;

[0117] in, This indicates the overall fit score. , , These are the preset weighting coefficients for style consistency, scene adaptability, and seasonal adaptability, respectively. .

[0118] Please refer to Figure 5 As shown, the warehouse location code, movement path, and nursing procedures are arranged in chronological order into a standardized hardware execution task sequence and distributed to the execution terminal. Specifically, this includes:

[0119] S501: The AI ​​Agent reads the digital asset card of each garment in the target outfit plan, extracts the physical storage location code bound to each garment, divides the wardrobe into several spatial grids, and establishes a grid connectivity topology map.

[0120] S502. Starting with the code of each storage location and ending with the code of the outbound buffer, adopt the following... The path planning algorithm searches for the collision-free optimal path on the topology graph, while reading the material type label and care requirement label of each garment, and searching for the set of basic care procedures that match the material type label in the preset care procedure library.

[0121] S503. Select mandatory procedures from the basic set based on nursing needs tags, determine whether to add in-depth nursing procedures based on clothing storage duration data, and form a nursing procedure chain by topologically sorting all mandatory and optional procedures according to nursing dependencies.

[0122] S504. The codes of each storage location, the corresponding optimal path and the nursing procedure chain are concatenated and encoded according to the execution time sequence to generate a standardized hardware execution task sequence containing action instruction codes, path codes, nursing instruction codes and time sequence codes. After being encapsulated by a unified instruction protocol, it is sent to the execution end.

[0123] After verification, the data is synchronously written to the cloud database, local control terminal storage, and mobile terminal cache, and the usage status field in the corresponding clothing digital asset card is synchronously updated, specifically including:

[0124] After receiving the standardized hardware execution task sequence, the execution end parses the action type code, target location code, and timing code in each instruction;

[0125] Motion control commands are sent sequentially to the stepper motor and servo motor according to the timing encoding order, driving the robotic arm to complete the clothing grabbing action. After the grabbing is completed, the digital asset card of the clothing is read to confirm its identity.

[0126] The garments are transferred to the care area, and the corresponding care peripherals are triggered sequentially according to the care instruction codes in the instructions to perform one or more care actions such as lint removal, hot air wrinkle removal, negative ion deodorization, and ultraviolet sterilization.

[0127] After the nursing action is completed, the robotic arm continues to transfer the clothes to the outbound buffer. After each action node is completed, the position sensor and status sensor collect the action completion confirmation signal.

[0128] The action type code, execution timestamp, and execution result status code are encapsulated into a status data packet, and the integrity of the data packet is verified using a CRC check algorithm.

[0129] ;

[0130] in, Indicates status data packet Cyclic Redundancy Check (CRC) code, Indicates the data packet number Bits of binary data, Indicates the total number of bits in the data packet. Represents the predefined generator polynomial;

[0131] After successful verification, the status data packet is published to the cloud message queue using the MQTT protocol. The cloud message queue then subscribes to and writes the data to the local control database and the mobile local cache, respectively. At the same time, the usage status identity field of the corresponding clothing in the clothing digital asset library is updated in an atomic manner.

[0132] After successful verification, the status data packet is published to the cloud message queue using the MQTT protocol, specifically including:

[0133] After the action node is executed, the position sensor detects whether the robotic arm has reached the target storage location coordinates, and the status sensor detects whether the clothing gripping mechanism has completed the release or clamping action. If both detections pass, an action completion confirmation signal is generated. The action type code, the Unix timestamp of the action completion time, and the execution result status code are encapsulated into a status data packet according to a preset JSON field format. CRC-16 checksum calculation is performed on all bytes of the status data packet.

[0134] ;

[0135] in, Indicates status data packet CRC-16 checksum, Indicates the data packet number Bits of binary data, Indicates the total number of bits in the data packet. A polynomial is generated for the CRC-16 standard, and a checksum is appended to the end of the data packet. The execution end publishes the status data packet with the appended checksum to the cloud message queue of the specified topic via the MQTT protocol. After receiving the data packet, the cloud message queue first verifies the CRC-16 checksum. After the verification is successful, it is pushed to the local control terminal database and the mobile terminal local cache for data storage.

[0136] Update the usage status identity field of the corresponding garment in the clothing digital asset repository using atomic operations, specifically including:

[0137] After the cloud message queue completes the data storage, it sends a status update request to the clothing digital asset database. The request carries the clothing's unique identifier and the target usage status code.

[0138] Upon receiving a request, the clothing digital asset library uses the unique identifier of the clothing as an index to lock the write permission of the corresponding digital asset card;

[0139] Replace the value of the status identity field with the target status code, and append a log record containing the operation type, operation time and status values ​​before and after the operation to the operation log field of the digital asset card. After unlocking the write permission, return an update confirmation response to the cloud message queue.

[0140] In a second aspect of the invention, an electronic device is also provided. The electronic device includes at least one processor; and a memory communicatively connected to the at least one processor; the memory stores instructions executable by the at least one processor, which, when executed by the at least one processor, enable the at least one processor to perform the system of the first aspect of the invention.

[0141] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the claimed invention. The scope of protection claimed by the appended claims and their equivalents is defined.

Claims

1. A smart wardrobe management method based on AI Agent, characterized in that, include: The acquisition module uses a visual acquisition device to acquire multi-angle images of the clothing to be put into storage. The acquired multi-angle image data is preprocessed by sequentially performing illumination correction, size normalization and background removal. The main body area of ​​the clothing is extracted by semantic segmentation algorithm. The image of the main body area of ​​the clothing is input into a cloud multimodal big model for deep semantic analysis. The output includes multi-dimensional semantic tags containing category, color, material, pattern, style, suitable season, usage scenario and care requirements. The multi-dimensional semantic tags are associated and bound with the physical storage location code to generate structured digital asset cards and write them into the clothing digital asset library. Outfit suggestion candidate set output module: Receives the user's outfit needs input in natural language, encodes the outfit needs into semantic vectors, integrates real-time weather data, user's daily scene tags, and personal outfit preference vectors to generate a multi-source fusion query vector, uses the query vector to retrieve a set of candidate clothes in the clothing digital asset library through vector similarity retrieval and inputs it into the large model, the large model optimizes and sorts the candidate clothes set based on multi-objective constraints of style consistency, scene adaptability, and seasonal adaptability, and outputs an outfit suggestion candidate set containing recommendation reasons; Task sequence delivery module: In response to the user's confirmation of the target outfit plan, the AI ​​Agent parses the digital asset card of each garment in the target outfit plan, extracts the physical storage location code, plans the movement path from each storage location to the outbound buffer based on the internal spatial topology of the wardrobe, matches the combination of care procedures according to the material label and care requirement label of each garment, and arranges the storage location code, movement path and care procedure into a standardized hardware execution task sequence in chronological order and delivers it to the execution end; Storage and update module: The execution end parses the action type, target storage location and timing relationship of each instruction in the task sequence, drives the hardware mechanism to complete the clothing grabbing, care and outbound transfer actions in sequence according to the arranged path. After each action node is completed, a status data packet containing action type, execution time and execution result is generated. After verification, it is synchronously written to the cloud database, local control terminal storage and mobile terminal cache, and the usage status field in the corresponding clothing digital asset card is synchronously updated.

2. The intelligent wardrobe management system based on AI Agent according to claim 1, characterized in that, The process involves sequentially performing illumination correction, size normalization, and background removal preprocessing on the acquired multi-angle image data, and then extracting the main clothing region using a semantic segmentation algorithm. Specifically, this includes: The mechanical servo motor is controlled to rotate the clothing at no less than three preset angles. At each rotation angle, a high-definition camera captures images of the clothing. Histogram equalization is performed on the original images at each angle to eliminate uneven lighting. All angle images are uniformly mapped to a preset standard size through affine transformation. A deep learning-based semantic segmentation network is used to perform pixel-level classification of images from each angle. Pixels that are not clothing in the classification results are set to zero. The zeroed images from multiple angles are then fused using a weighted average based on the spatial overlap of the clothing subject. The fusion weights are dynamically allocated according to the sharpness scores of each image angle. The output is standardized image data with zero background pixel values ​​and complete, unobstructed clothing subjects. The fusion weights are proportionally allocated according to the sharpness scores of each image angle. ； in, Indicates the first Fusion weights for images from different angles, This represents the total number of angles in the acquired images. Indicates the first The sharpness score of the images at each angle is calculated using Laplace variance: ； in, Indicates the first Sharpness rating of images from each angle. Represents the Laplace operator. Indicates the first An image at various angles in pixel coordinates The grayscale value at that location.

3. The intelligent wardrobe management system based on AI Agent according to claim 2, characterized in that, The process of inputting the image of the main clothing area into a cloud-based multimodal large model for deep semantic analysis, outputting multidimensional semantic labels, and generating structured digital asset cards specifically includes: Standardized image data and prior knowledge prompts for clothing categories are input into a cloud-based multimodal model. The model then outputs clothing category labels, color attribute values, material type identifiers, fit classification labels, style vectors, seasonal labels, usage scenario labels, and care requirement labels in parallel. The aforementioned multidimensional semantic tags are associated and bound with the physical storage location code and the storage entry timestamp, respectively, to generate a five-dimensional structured digital asset card containing image identity field, semantic identity field, spatial storage location identity field, usage status identity field and care attribute identity field. The digital asset card is then written into the clothing digital asset library with a unique identifier.

4. The intelligent wardrobe management system based on AI Agent according to claim 3, characterized in that, The large model optimizes and ranks the candidate clothing set based on multi-objective constraints of style consistency, scene adaptability, and seasonal adaptability, and outputs a candidate set of outfit schemes containing recommendation reasons, specifically including: The user's natural language dressing needs are transformed into high-dimensional semantic vectors through a text encoding model. Temperature ranges and weather types from real-time weather data are obtained and encoded into weather feature vectors. Scene tags from the user's daily schedule are obtained and encoded into scene feature vectors. Style preference distribution from the user's historical dressing records is obtained and encoded into preference feature vectors. Semantic vectors, weather feature vectors, scene feature vectors, and preference feature vectors are concatenated and fused to generate a multi-source fusion query vector. The query vector is then used to calculate cosine similarity in the style vector space of each garment in the clothing digital asset database. ； in, This represents a multi-source fusion query vector. Indicates the first Style vector of the garment and Represent the two vectors at the th... Dimensional components, Indicates the dimension of a vector; Clothing items with similarity higher than a preset threshold are selected as a candidate clothing set. The semantic tags of each item in the candidate clothing set and the query vector are input into the large model, which calculates the style consistency score. The scene adaptability score is calculated by combining the weather feature vector and the scene feature vector, and the season adaptability score is calculated by combining the season interval of the current date. The overall adaptability score is obtained by weighting the scene adaptability score, the season adaptability score and the style consistency score. The candidate outfit ideas are output in descending order of scores, and the large model generates natural language recommendation reasons for each candidate idea, including explanations of style coordination, scene matching, and seasonal suitability.

5. The intelligent wardrobe management system based on AI Agent according to claim 4, characterized in that, The formula for calculating the style consistency score is as follows: ； in, The score indicates the consistency of style. This represents the total number of clothing items in the candidate scheme. and They represent the first Item and the first Style vector of the garment; The formula for calculating scene adaptability score is: ； in, This represents the scenario adaptability score; a higher score indicates a higher degree of matching between the solution and the target scenario. This represents the total number of clothing items in the candidate scheme. Indicates the first The scene feature vector of a piece of clothing, This represents the target scenario feature vector of the user's daily schedule; The formula for calculating the seasonality score is as follows: ； in, This indicates the seasonality score. This indicates the number of clothing items in the candidate solution that successfully match the current season range with the appropriate season tag. This indicates the total number of clothing items in the candidate scheme; The formula for calculating the overall fit score is as follows: ； in, This indicates the overall fit score. , , These are the preset weighting coefficients for style consistency, scene adaptability, and seasonal adaptability, respectively. .

6. The intelligent wardrobe management system based on AI Agent according to claim 5, characterized in that, The process of arranging the storage location code, movement path, and nursing procedures into a standardized hardware execution task sequence according to time sequence and distributing it to the execution terminal specifically includes: The AI ​​Agent reads the digital asset card of each garment in the target outfit plan, extracts the physical storage location code bound to each garment, divides the wardrobe into several spatial grids, and establishes a grid connectivity topology map. Starting with the code of each storage location and ending with the code of the outbound buffer zone, the following method is adopted: The path planning algorithm searches for the collision-free optimal path on the topology graph, while reading the material type label and care requirement label of each garment, and searching for the set of basic care procedures that match the material type label in the preset care procedure library. Based on the nursing needs tags, select the required procedures from the basic set, determine whether to add in-depth nursing procedures based on the storage time data of clothing, and form a nursing procedure chain by topologically sorting all the required and optional procedures according to the nursing dependencies. Each storage location code, its corresponding optimal path, and the nursing procedure chain are concatenated and encoded according to the execution sequence to generate a standardized hardware execution task sequence containing action instruction codes, path codes, nursing instruction codes, and sequence codes. This sequence is then encapsulated using a unified instruction protocol and sent to the execution end.

7. The intelligent wardrobe management system based on AI Agent according to claim 6, characterized in that, The verified data is synchronously written to the cloud database, local control terminal storage, and mobile terminal cache, and the usage status field in the corresponding clothing digital asset card is synchronously updated, specifically including: After receiving the standardized hardware execution task sequence, the execution end parses the action type code, target location code, and timing code in each instruction; Motion control commands are sent sequentially to the stepper motor and servo motor according to the timing encoding order, driving the robotic arm to complete the clothing grabbing action. After the grabbing is completed, the digital asset card of the clothing is read to confirm its identity. The garments are transferred to the care area, and the corresponding care peripherals are triggered sequentially according to the care instruction codes in the instructions to perform one or more care actions such as lint removal, hot air wrinkle removal, negative ion deodorization, and ultraviolet sterilization. After the nursing action is completed, the robotic arm continues to transfer the clothes to the outbound buffer. After each action node is completed, the position sensor and status sensor collect the action completion confirmation signal. The action type code, execution timestamp, and execution result status code are encapsulated into a status data packet, and the integrity of the data packet is verified using a CRC check algorithm. ； in, Indicates status data packet Cyclic Redundancy Check (CRC) code, Indicates the data packet number Bits of binary data, Indicates the total number of bits in the data packet. Represents the predefined generator polynomial; After successful verification, the status data packet is published to the cloud message queue using the MQTT protocol. The cloud message queue then subscribes to and writes the data to the local control database and the mobile local cache, respectively. At the same time, the usage status identity field of the corresponding clothing in the clothing digital asset library is updated in an atomic manner.

8. The intelligent wardrobe management system based on AI Agent according to claim 7, characterized in that, After the verification is successful, the status data packet is published to the cloud message queue using the MQTT protocol, specifically including: After the action node is executed, the position sensor detects whether the robotic arm has reached the target storage location coordinates, and the status sensor detects whether the clothing gripping mechanism has completed the release or clamping action. If both detections pass, an action completion confirmation signal is generated. The action type code, the Unix timestamp of the action completion time, and the execution result status code are encapsulated into a status data packet according to a preset JSON field format. CRC-16 checksum calculation is performed on all bytes of the status data packet. ； in, Indicates status data packet CRC-16 checksum, Indicates the data packet number Bits of binary data, Indicates the total number of bits in the data packet. A polynomial is generated for the CRC-16 standard, and a checksum is appended to the end of the data packet. The execution end publishes the status data packet with the appended checksum to the cloud message queue of the specified topic via the MQTT protocol. After receiving the data packet, the cloud message queue first verifies the CRC-16 checksum. After the verification is successful, it is pushed to the local control terminal database and the mobile terminal local cache for data storage.

9. The intelligent wardrobe management system based on AI Agent according to claim 7, characterized in that, The process of updating the usage status identity field of the corresponding garment in the clothing digital asset database using atomic operations specifically includes: After the cloud message queue completes the data storage, it sends a status update request to the clothing digital asset database. The request carries the clothing's unique identifier and the target usage status code. Upon receiving a request, the clothing digital asset library uses the unique identifier of the clothing as an index to lock the write permission of the corresponding digital asset card; Replace the value of the status identity field with the target status code, and append a log record containing the operation type, operation time and status values ​​before and after the operation to the operation log field of the digital asset card. After unlocking the write permission, return an update confirmation response to the cloud message queue.

10. An electronic device, comprising at least one processor; and a memory communicatively connected to said at least one processor; characterized in that, The memory stores instructions that can be executed by the at least one processor to enable the at least one processor to perform the system according to any one of claims 1-9.

Images