A textile shipment inspection control system

Abstract

The present invention relates to the field of operational control technology, and more specifically to a textile shipment inspection control system, comprising a data acquisition module, an inspection task generation module, and an inspection assignment module, wherein: the data acquisition module is used to automatically collect basic information and shipment requirements of the textiles to be inspected; the inspection task generation module dynamically generates inspection tasks based on the basic information and shipment requirements of the textiles, including inspection standards and inspection procedures; and the inspection assignment module assigns corresponding inspectors according to the inspection standards and inspection procedures, providing the inspectors with a workstation containing all relevant information, inspection standards, and inspection procedure guidance for the textiles to be inspected, supporting the inspectors in performing inspection tasks according to established standards. This invention not only significantly improves the accuracy of inspection task generation, but also ensures that each inspection task is highly consistent with the actual situation and market demand of the textiles.

Description

[0001] This application is a divisional application of the application filed on September 2, 2024, with application number 202411217590.3 and invention name “A Textile Shipment Inspection and Operation Control System”. Technical Field

[0002] The present invention relates to the technical field of operation control, and in particular to a textile shipment inspection control system. Background Art

[0003] The existing textile shipment inspection process often faces a series of challenges and limitations. First, the current inspection process often adopts a "one-size-fits-all" approach, applying the same inspection standards and procedures to every batch of textiles. This ignores the differences between textiles in terms of materials, applications, and customer requirements. This approach may not fully guarantee that the quality requirements of specific textiles are met, and it may also lead to waste of resources, such as excessive inspection of products that do not require strict inspection.

[0004] Secondly, traditional methods of assigning inspectors lack flexibility and personalization. Typically, inspector assignments don't consider individual expertise, experience, or current workloads, which can impact the efficiency and quality of inspections. For example, complex inspection tasks may require inspectors with specific skills, but traditional methods can assign tasks to individuals who lack the relevant skills, impacting the accuracy of inspection results.

[0005] Furthermore, with the continuous evolution of market demands and the application of new technologies, textile inspection standards and processes need to be continuously updated and optimized to adapt to these changes. However, existing inspection processes often lack sufficient flexibility and adaptability, making it difficult to promptly reflect changes in industry standards, special customer requirements, or the adoption of new inspection technologies.

[0006] Therefore, the existing textile shipment inspection process urgently needs a new method that can flexibly adapt to the characteristics of different textiles, personalize inspection needs, and optimize the allocation of inspectors to improve inspection efficiency, ensure product quality, and at the same time improve resource utilization and market response speed. Summary of the Invention

[0007] Based on the above objectives, the present invention provides a textile shipment inspection and control system.

[0008] A textile shipment inspection control system includes a data acquisition module, an inspection task generation module, and an inspection allocation module, wherein;

[0009] The data acquisition module is used to automatically collect basic information and shipping requirements of the textiles to be inspected;

[0010] The inspection task generation module dynamically generates inspection tasks based on basic textile information and shipping requirements, including inspection standards and inspection processes, specifically including:

[0011] Machine learning submodule: Based on a neural network autoencoder, it identifies and classifies basic information of textiles and the inspection requirements corresponding to shipping requirements;

[0012] Rule Engine Submodule: This module builds a rule base based on domain expert knowledge, including standard rules for textile inspection, industry best practices, and customer requirements. It uses the rule engine to analyze the output of the machine learning submodule to generate inspection tasks and processes that meet actual operational needs.

[0013] The inspection assignment module assigns corresponding inspectors according to the inspection standards and inspection procedures, and provides inspectors with a workstation that includes all relevant information, inspection standards and inspection procedure guidance for the textiles to be inspected, supporting inspectors to perform inspection tasks according to established standards.

[0014] Furthermore, the acquisition method of the data acquisition module includes:

[0015] By scanning the identification code on the textile to be inspected, the basic information of the textile, including material, size and color, can be automatically extracted;

[0016] Use OCR to read text information of shipping requirements from textile packaging or accompanying documents. Shipping requirements include durability, color fastness, and texture clarity standards. The scanned and recognized information is digitized and transmitted to the inspection task allocation module.

[0017] Furthermore, the neural network autoencoder specifically includes:

[0018] Collect basic information of textiles and data related to shipping requirements;

[0019] Encode the basic information and shipping requirements of textiles and convert them into a format suitable for neural network processing;

[0020] Autoencoder network architecture design:

[0021] Encoder: Uses convolutional neural network layers to process image inputs and extract texture, pattern, and color features from textile images. Non-image data (material type, size) is processed through fully connected layers and the output of the non-image data is merged with the image features to form a comprehensive feature representation.

[0022] Decoder: Design the decoder part to reconstruct the input data, including the reconstruction of the image as well as the basic information and shipping requirements of the textile to ensure that the encoding process captures all information;

[0023] Use the textile dataset to train the autoencoder, focusing on whether the encoder can extract features. The training goal is to minimize the reconstruction error, that is, the difference between the input data and the decoder output;

[0024] Feature extraction and classification: After training, the encoder is used as a feature extractor to extract features from new textile data. The extracted features reflect the basic properties and shipping requirements of the textiles.

[0025] Based on the extracted features, a classifier is trained to identify different inspection requirement categories;

[0026] According to the classification results, the corresponding inspection tasks are dynamically generated.

[0027] Furthermore, the autoencoder network architecture includes an encoder part consisting of two convolutional layers for extracting features from textile images, while processing basic information of textiles through a fully connected layer;

[0028] Image feature extraction: For the input textile image I, the calculation of the first convolution layer is expressed as: F1 = ReLU(W1*I+b1), where F1 is the output feature map of the first convolution layer, W1 is the weight of the convolution kernel, b1 is the bias term, ReLU is the activation function used to increase nonlinearity, and * represents the convolution operation;

[0029] The calculation of the second convolutional layer is expressed as: F2 = ReLU(W2*F1+b2), where F2 is the output feature map of the second convolutional layer, W2 and b2 are the weight and bias terms of the second convolutional layer respectively;

[0030] Basic information processing: The basic information B of the textile (material and size information processed by one-hot encoding) is processed through a fully connected layer and calculated as follows: F B =ReLU(W B B+b B ), where F B is the output feature of the fully connected layer, W B is the weight of the fully connected layer, b B is the bias term, · represents matrix multiplication;

[0031] After extracting image features and processing basic information, the two parts of features are fused together to test the required classification. F2 is converted into a fixed-size feature vector V2 through average pooling or global maximum pooling, and then V2 and F B Connect together to form a comprehensive feature vector F combined ;

[0032] The fully connected layer is used as a classifier to predict the inspection requirement category of textiles based on the comprehensive feature vector.

[0033] Furthermore, the inspection requirement category of textiles predicted based on the comprehensive feature vector is expressed as:

[0034] Y=Softmax(W C ·F combined +b C ), where Y is the probability distribution of the predicted inspection demand category, W C is the weight of the fully connected layer of the classifier, b C is the bias term, and the Softmax function is used to convert the output into a probability distribution.

[0035] Furthermore, the average pooling operation includes: assuming that the dimension of F2 is H×W×D, where H and W are the height and width of the feature map, respectively, and D is the depth (i.e., the number of channels) of the feature map. The goal of average pooling is to reduce the size of H and W while keeping the D dimension unchanged. The specific operation is as follows:

[0036] Select the pooling window size: First, determine the size of the pooling window p×p, where p is the side length of the pooling window. The window size affects the coarseness of the pooling.

[0037] Sliding window and calculating the average value: Slide the pooling window on F2 and calculate the average value of all elements in each window. The average value generated by each window corresponds to an element in the output feature vector V2.

[0038] Output feature vector: Repeat sliding the window and calculating the average value until the entire feature map F2 is covered. The final average value constitutes a fixed-size feature vector V2, which is expressed as follows:

[0039] Where V2(d) is the element of vector V2 at depth d, and F2(h, w, d) is the element of F2 at position (h, w) and depth d.

[0040] Furthermore, the rule engine submodule specifically includes:

[0041] Build a rule library: Collect and organize international and national standards related to textile inspection, as well as industry-recognized quality control standards. Rules include fiber composition, dimensional stability, color fastness, and abrasion resistance. It also brings together industry-recognized best practices and rules of thumb, including material processing methods, inspection focus for product types, and environmental standards and safety regulations for different requirements of different customers or markets.

[0042] Rule matching and application: The rule engine takes the output of the machine learning submodule as input and matches it against the rules in the rule base;

[0043] Inspection task and process generation: Based on the matching results, the rule engine generates specific inspection tasks. Each task includes inspection standards, required instruments, and expected results.

[0044] Furthermore, the inspection allocation module specifically includes:

[0045] Maintain an inspector database that records each inspector's skills, qualifications, experience, and past inspection performance;

[0046] Inspection task demand analysis: For each generated inspection task, analyze the required skills and qualifications, including inspection experience, proficiency in specific inspection equipment, and understanding of relevant standards;

[0047] Based on the personnel allocation optimization algorithm, the optimal inspection personnel are matched for each inspection task, taking into account the task urgency, personnel skill matching, and workload balance factors;

[0048] Determine the optimal match between inspection tasks and personnel, automatically schedule tasks, and notify the corresponding inspection personnel through the workstation.

[0049] Furthermore, the personnel allocation optimization algorithm includes the definition of constraints:

[0050] Personnel skill matching S: Each inspector has a skill set. Different inspection tasks require different skills. Use the binary relationship S ij Indicates that, where i represents the i-th inspector and j represents the j-th inspection task, if S ij =1, it means that the i-th inspector has the skills required to complete the j-th task; if S ij =0, it means not available;

[0051] Workload balance W: To ensure workload balance, the amount of tasks assigned to each inspector should not exceed the upper limit of their work capacity. i is the workload of the i-th inspector, L i For its maximum workload, it should meet W i ≤L i ;

[0052] Inspection time limit T: Each inspection task must be completed within the scheduled time. and are the start time and end time of the jth inspection task, respectively, and the inspector assigned to the task is available within this time range;

[0053] Based on the constraints, the objective function and constraints of the personnel allocation optimization algorithm are constructed. The objective function is to maximize the efficiency of task allocation or minimize the idle time of detection personnel. The maximized allocation efficiency is expressed as: Where n is the total number of inspectors, m is the total number of inspection tasks, and X ij is a decision variable. If the i-th inspector is assigned to the j-th task, then X ij =1, otherwise X ij =0; minimizing the idle time of all inspectors is to minimize the sum of the differences between the available working time and the working time of the assigned tasks: Where n is the total number of inspectors, m is the total number of inspection tasks, T i is the available working time of the i-th inspector, D j is the time required to complete the jth inspection task, X ij is a decision variable. If the i-th inspector is assigned to the j-th task, then X ij =1, otherwise X ij =0.

[0054] Furthermore, the constraint condition is expressed as:

[0055] Skill matching constraint: For each task j, there is at least one inspector assigned with matching skills:

[0056] Workload constraint: The total workload of each inspector does not exceed their maximum workload: Among them, W ij represents the workload of the i-th inspector performing the j-th task;

[0057] Time limit constraints: Ensure that the inspectors assigned to each task are available within the time frame of the task.

[0058] Beneficial effects of the present invention:

[0059] The present invention, by integrating the deep learning capabilities of a neural network autoencoder with a rule base built based on domain expert knowledge, conducts an in-depth analysis and understanding of the basic information of textiles and their shipping requirements. The application of the autoencoder enables the extraction of key features from complex textile images and attributes, while the rule base ensures that these features can be accurately connected with inspection standards and processes. This combination not only significantly improves the accuracy of inspection task generation, but also ensures that each inspection work is highly consistent with the actual situation of textiles and market demand. The result is a more efficient inspection process, reduces errors and omissions, improves customer satisfaction, and reduces the risk of unqualified products entering the market.

[0060] The present invention realizes the intelligent allocation of inspection tasks by applying a personnel allocation optimization algorithm combined with the optimization goal of minimizing the idle time of inspectors. It takes into account the skill set, experience level and current workload of inspectors, ensuring that each task can be performed by the most suitable person, rather than simply distributing the task load evenly. It effectively balances the workload and reduces the inefficiency caused by skill mismatch or workload overload. At the same time, by reducing idle time, it improves the utilization efficiency of human resources, which not only increases the speed and quality of inspection, but also improves the job satisfaction of employees and the overall morale of the team.

[0061] The present invention is designed to allow the rule base to be updated and expanded in a flexible and dynamic manner to adapt to the rapidly changing industry standards, evolving customer needs and the application of new technologies in the field of textile inspection. This adaptability enables the system to not only handle current inspection tasks, but also quickly adapt to future changes, whether it is the emergence of new materials, the development of new inspection technologies, or shifts in market demand. BRIEF DESCRIPTION OF THE DRAWINGS

[0062] In order to more clearly illustrate the technical solutions in the present invention or the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. Obviously, the drawings described below are only for the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying any creative work.

[0063] Figure 1 This is a schematic diagram of the functional modules of a control system according to an embodiment of the present invention;

[0064] Figure 2 Schematic diagram of a neural network autoencoder according to an embodiment of the present invention. DETAILED DESCRIPTION

[0065] In order to make the objectives, technical solutions and advantages of the present invention more clearly understood, the present invention is further described in detail below with reference to specific embodiments.

[0066] It should be noted that, unless otherwise defined, the technical or scientific terms used in the present invention should have the usual meanings understood by people with ordinary skills in the field to which the present invention belongs. The "first", "second" and similar words used in the present invention do not indicate any order, quantity or importance, but are only used to distinguish different components. "Include" or "comprise" and similar words mean that the elements or objects appearing before the word include the elements or objects listed after the word and their equivalents, without excluding other elements or objects. "Connect" or "connected" and similar words are not limited to physical or mechanical connections, but may include electrical connections, whether direct or indirect. "Up", "down", "left", "right" and the like are only used to indicate relative positional relationships. When the absolute position of the object being described changes, the relative positional relationship may also change accordingly.

[0067] like Figure 1-2 As shown, a textile shipment inspection control system includes a data acquisition module, an inspection task generation module, and an inspection allocation module, wherein;

[0068] The data acquisition module is used to automatically collect basic information and shipping requirements of textiles to be inspected;

[0069] The inspection task generation module dynamically generates inspection tasks based on basic textile information and shipping requirements, including inspection standards and inspection processes, specifically:

[0070] Machine learning submodule: Based on a neural network autoencoder, it identifies and classifies basic information of textiles and the inspection requirements corresponding to shipping requirements;

[0071] Rule Engine Submodule: This module builds a rule base based on domain expert knowledge, including standard rules for textile inspection, industry best practices, and customer requirements. It uses the rule engine to analyze the output of the machine learning submodule to generate inspection tasks and processes that meet actual operational needs.

[0072] The inspection allocation module assigns corresponding inspectors according to the inspection standards and inspection procedures, and provides inspectors with a workstation that includes all relevant information, inspection standards and inspection process guidance for the textiles to be inspected, supporting inspectors to perform inspection tasks in accordance with established standards.

[0073] The acquisition method of the data acquisition module includes:

[0074] By scanning the identification code on the textile to be inspected, the basic information of the textile, including material, size and color, can be automatically extracted;

[0075] Use OCR to read text information of shipping requirements from textile packaging or accompanying documents. Shipping requirements include durability, color fastness, and texture clarity standards. The scanned and recognized information is digitized and transmitted to the inspection task allocation module.

[0076] The neural network autoencoder specifically includes:

[0077] Collect basic information of textiles and data related to shipping requirements;

[0078] Encode the basic information and shipping requirements of textiles and convert them into a format suitable for neural network processing. One-hot encoding can be used to represent the type of textile materials and the various standards in the shipping requirements.

[0079] Autoencoder network architecture design:

[0080] Encoder: Uses convolutional neural network layers to process image inputs and extract texture, pattern, and color features from textile images. Non-image data (material type, size) is processed through fully connected layers and the output of the non-image data is merged with the image features to form a comprehensive feature representation.

[0081] Decoder: Design the decoder part to reconstruct the input data, including the reconstruction of the image as well as the basic information and shipping requirements of the textile to ensure that the encoding process captures all information;

[0082] Use the textile dataset to train the autoencoder, focusing on whether the encoder can extract features. The training goal is to minimize the reconstruction error, that is, the difference between the input data and the decoder output;

[0083] Feature extraction and classification: After training, the encoder is used as a feature extractor to extract features from new textile data. The extracted features reflect the basic properties and shipping requirements of the textiles.

[0084] Based on the extracted features, a classifier is trained to identify different inspection requirement categories. For example, based on the characteristics of the textile and shipping requirements, the classifier can determine whether a textile requires specific color fastness testing or abrasion resistance testing.

[0085] According to the classification results, the corresponding inspection tasks are dynamically generated.

[0086] The autoencoder network architecture includes an encoder part consisting of two convolutional layers to extract features from textile images, and a fully connected layer to process the basic information of the textile;

[0087] Image feature extraction: For the input textile image I, the calculation of the first convolution layer is expressed as: F1 = ReLU(W1*I+b1), where F1 is the output feature map of the first convolution layer, W1 is the weight of the convolution kernel, b1 is the bias term, ReLU is the activation function used to increase nonlinearity, and * represents the convolution operation;

[0088] The calculation of the second convolutional layer is expressed as: F2 = ReLU(W2*F1+b2), where F2 is the output feature map of the second convolutional layer, W2 and b2 are the weight and bias terms of the second convolutional layer respectively;

[0089] Basic information processing: The basic information B of the textile (material and size information processed by one-hot encoding) is processed through a fully connected layer and calculated as follows: F B =ReLU(W B B+b B ), where F B is the output feature of the fully connected layer, W B is the weight of the fully connected layer, b B is the bias term, · represents matrix multiplication;

[0090] After extracting image features and processing basic information, the two parts of features are fused together to test the required classification. F2 is converted into a fixed-size feature vector V2 through average pooling or global maximum pooling, and then V2 and F B Connect together to form a comprehensive feature vector F combined ;

[0091] The fully connected layer is used as a classifier to predict the inspection requirement category of textiles based on the comprehensive feature vector.

[0092] The inspection demand category of textiles predicted by the comprehensive feature vector is expressed as:

[0093] Y=Softmax(W C ·F combined +b C ), where Y is the probability distribution of the predicted inspection demand category, W C is the weight of the fully connected layer of the classifier, b C is the bias term, and the Softmax function is used to convert the output into a probability distribution.

[0094] The average pooling operation includes: assuming that the dimensions of F2 are H×W×D, where H and W are the height and width of the feature map respectively, and D is the depth of the feature map (i.e., the number of channels). The goal of average pooling is to reduce the dimensions of H and W while keeping the dimension D unchanged. The specific operation is as follows:

[0095] Select the pooling window size: First, determine the size of the pooling window p×p, where p is the side length of the pooling window. The window size affects the coarseness of the pooling.

[0096] Sliding window and calculating the average value: Slide the pooling window on F2 and calculate the average value of all elements in each window. The average value generated by each window corresponds to an element in the output feature vector V2.

[0097] Output feature vector: Repeat sliding the window and calculating the average value until the entire feature map F2 is covered. The final average value constitutes a fixed-size feature vector V2, which is expressed as follows:

[0098] Where V2(d) is the element of vector V2 at depth d, and F2(h, w, d) is the element of F2 at position (h, w) and depth d;

[0099] In this way, the average pooling operation can extract a global feature vector V2 representing the entire feature map from F2. This vector B2 captures the key visual features of the textile image and is suitable for subsequent inspection requirement classification tasks.

[0100] The rule engine submodule specifically includes:

[0101] Build a rule library: Collect and organize international and national standards related to textile inspection, such as ISO and ASTM specifications, as well as industry-recognized quality control standards. Rules include fiber composition, dimensional stability, color fastness, and abrasion resistance. It also compiles industry-recognized best practices and rules of thumb, including material processing methods, inspection focus for product types, and environmental standards and safety regulations for different requirements of different customers or markets.

[0102] Rule matching and application: The rule engine uses the output of the machine learning submodule as input and matches it against the rules in the rule library. For example, if the machine learning submodule indicates that a textile is high-density cotton fabric, the rule engine will match the dimensional stability and abrasion resistance test rules corresponding to that material.

[0103] Inspection task and process generation: Based on the matching results, the rule engine generates specific inspection tasks. Each task includes the inspection standard, required instruments, and expected results. At the same time, based on industry best practices and customer requirements, the rule engine will also guide the specific steps of the inspection process to ensure that the inspection work complies with standards and meets specific needs.

[0104] The rule engine can be implemented based on "if-then" rules (for example, if the material is "cotton", then the "shrinkage test" should be performed). To improve flexibility and scalability, the rule engine can adopt a visual rule editing tool and a dynamic rule loading mechanism so that domain experts can directly contribute and update the rule base.

[0105] The inspection allocation module specifically includes:

[0106] Maintain an inspector database that records each inspector's skills, qualifications, experience, and past inspection performance. For example, some inspectors may be particularly skilled in color fastness testing, while others may have deeper expertise in dimensional stability testing.

[0107] Inspection task demand analysis: For each generated inspection task, analyze the required skills and qualifications, including inspection experience, proficiency in specific inspection equipment, and understanding of relevant standards;

[0108] Based on the personnel allocation optimization algorithm, the optimal inspection personnel are matched for each inspection task, taking into account the task urgency, personnel skill matching, and workload balance factors;

[0109] Determine the optimal match between inspection tasks and personnel, automatically schedule tasks, and notify the corresponding inspection personnel through the workstation to ensure that they understand the specific requirements and time schedule of the upcoming inspection tasks.

[0110] The staff allocation optimization algorithm includes the definition of constraints:

[0111] Personnel skill matching S: Each inspector has a skill set. Different inspection tasks require different skills. Use the binary relationship S ij Indicates that, where i represents the i-th inspector and j represents the j-th inspection task, if S ij =1, it means that the i-th inspector has the skills required to complete the j-th task; if S ij =0, it means not available;

[0112] Workload balance W: To ensure workload balance, the amount of tasks assigned to each inspector should not exceed the upper limit of their work capacity. i is the workload of the i-th inspector, L i For its maximum workload, it should meet W i ≤L i ;

[0113] Inspection time limit T: Each inspection task must be completed within the scheduled time. and are the start time and end time of the jth inspection task, respectively, and the inspector assigned to the task is available within this time range;

[0114] Based on the constraints, the objective function and constraints of the personnel allocation optimization algorithm are constructed. The objective function is to maximize the efficiency of task allocation or minimize the idle time of detection personnel. The maximum allocation efficiency is expressed as: Where n is the total number of inspectors, m is the total number of inspection tasks, and X ij is a decision variable. If the i-th inspector is assigned to the j-th task, then X ij =1, otherwise X ij =0; Minimize the idle time of all inspectors, that is, minimize the sum of the differences between the available working time and the working time of the assigned tasks: Where n is the total number of inspectors, m is the total number of inspection tasks, T i is the available working time of the i-th inspector, D j is the time required to complete the jth inspection task, X ij is a decision variable. If the i-th inspector is assigned to the j-th task, then X ij =1, otherwise X ij =0; This objective function means that for each inspector i, their total available time T is calculated i Subtract the sum of the time worked for each assigned task j By summing up these differences, we get the total idle time of all inspectors, and our goal is to minimize this sum.

[0115] The constraints are expressed as:

[0116] Skill matching constraint: For each task j, there is at least one inspector assigned with matching skills:

[0117] Workload constraint: The total workload of each inspector does not exceed their maximum workload: Among them, W ij represents the workload of the i-th inspector performing the j-th task;

[0118] Time limit constraints: Ensure that the inspectors assigned to each task are available within the time frame of the task.

[0119] Those skilled in the art should understand that the discussion of any of the above embodiments is merely illustrative and is not intended to imply that the scope of the present invention is limited to these examples. Within the scope of the present invention, the technical features in the above embodiments or different embodiments may be combined, the steps may be implemented in any order, and there are many other variations of the different aspects of the present invention as described above, which are not provided in detail for the sake of simplicity.

[0120] The present invention is intended to cover all such substitutions, modifications and variations that fall within the broad scope of the claims. Therefore, any omissions, modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A textile shipment inspection and control system, characterized in that: It includes data acquisition module, inspection task generation module and inspection allocation module, among which; The data acquisition module is used to automatically collect basic information and shipping requirements of the textiles to be inspected; The inspection task generation module dynamically generates inspection tasks based on basic textile information and shipping requirements, including inspection standards and inspection processes, specifically including: Machine Learning Submodule: This module uses a neural network autoencoder to identify and classify basic information about textiles and corresponding inspection requirements for shipment. The neural network autoencoder specifically includes: Collect basic information of textiles and data related to shipping requirements; Encode the basic information and shipping requirements of textiles and convert them into a format suitable for neural network processing; Autoencoder network architecture design: Encoder: Uses convolutional neural network layers to process image inputs and extract texture, pattern, and color features from textile images. Non-image data is processed through fully connected layers and the output of the non-image data is merged with the image features to form a comprehensive feature representation. Decoder: Design the decoder part to reconstruct the input data, including the reconstruction of the image as well as the basic information and shipping requirements of the textile to ensure that the encoding process captures all information; Use the textile dataset to train the autoencoder, focusing on whether the encoder can extract features. The training goal is to minimize the reconstruction error, that is, the difference between the input data and the decoder output; Feature extraction and classification: After training, the encoder is used as a feature extractor to extract features from new textile data. The extracted features reflect the basic properties and shipping requirements of the textiles. Based on the extracted features, a classifier is trained to identify different inspection requirement categories; Dynamically generate corresponding inspection tasks based on the classification results; The autoencoder network architecture includes an encoder part consisting of two convolutional layers for extracting features from textile images, and a fully connected layer for processing basic information of textiles; Image feature extraction: For the input textile image I, the calculation of the first convolution layer is expressed as: F1 = ReLU(W1*I+b1), where F1 is the output feature map of the first convolution layer, W1 is the weight of the convolution kernel, b1 is the bias term, ReLU is the activation function used to increase nonlinearity, and * represents the convolution operation; The calculation of the second convolutional layer is expressed as: F2 = ReLU(W2*F1+b2), where F2 is the output feature map of the second convolutional layer, W2 and b2 are the weight and bias terms of the second convolutional layer respectively; Basic information processing: The basic information B of the textile is processed through a fully connected layer and calculated as follows: F B =ReLU(W B B+b B ), where F B is the output feature of the fully connected layer, W B is the weight of the fully connected layer, b B is the bias term, · represents matrix multiplication; After extracting image features and processing basic information, the two parts of features are fused together to test the required classification. F2 is converted into a fixed-size feature vector V2 through average pooling or global maximum pooling, and then V2 and F B Connect together to form a comprehensive feature vector F combined ; Use the fully connected layer as a classifier to predict the inspection requirement category of textiles based on the comprehensive feature vector; The inspection demand category of textiles predicted based on the comprehensive feature vector is expressed as: Y=Softmax(W C ·F combined +b C ), where Y is the probability distribution of the predicted inspection demand category, W C is the weight of the fully connected layer of the classifier, b C Is the bias term, the Softmax function is used to convert the output into a probability distribution; Rule Engine Submodule: This module builds a rule base based on domain expert knowledge, including standard rules for textile inspection, industry best practices, and customer requirements. It uses the rule engine to analyze the output of the machine learning submodule to generate inspection tasks and processes that meet actual operational needs. The rule engine submodule specifically includes: Build a rule library: Collect and organize international and national standards related to textile inspection, as well as industry-recognized quality control standards. Rules include fiber composition, dimensional stability, color fastness, and abrasion resistance. It also brings together industry-recognized best practices and rules of thumb, including material processing methods, inspection focus for product types, and environmental standards and safety regulations for different requirements of different customers or markets. Rule matching and application: The rule engine takes the output of the machine learning submodule as input and matches it against the rules in the rule base; Inspection task and process generation: Based on the matching results, the rule engine generates specific inspection tasks. Each task includes inspection standards, required instruments, and expected results. The inspection assignment module assigns corresponding inspectors according to the inspection standards and inspection procedures, and provides inspectors with a workstation that includes all relevant information, inspection standards, and inspection procedure guidance for the textiles to be inspected, supporting inspectors to perform inspection tasks according to established standards, including: Maintain an inspector database that records each inspector's skills, qualifications, experience, and past inspection performance; Inspection task demand analysis: For each generated inspection task, analyze the required skills and qualifications, including inspection experience, proficiency in specific inspection equipment, and understanding of relevant standards; Based on the personnel allocation optimization algorithm, the optimal inspection personnel are matched for each inspection task, taking into account the task urgency, personnel skill matching, and workload balance factors; Determine the optimal match between inspection tasks and personnel, automatically schedule tasks, and notify the corresponding inspection personnel through the workstation; The staff allocation optimization algorithm includes the definition of constraints: Personnel skill matching S: Each inspector has a skill set. Different inspection tasks require different skills. Use the binary relationship S ij Indicates that, where i represents the i-th inspector and j represents the j-th inspection task, if S ij =1, it means that the i-th inspector has the skills required to complete the j-th task; if S ij =0, it means not available; Workload balance W: To ensure workload balance, the amount of tasks assigned to each inspector should not exceed the upper limit of their work capacity. i is the workload of the i-th inspector, L i For its maximum workload, it should meet W i ≤L i ; Inspection time limit T: Each inspection task must be completed within the scheduled time. and are the start time and end time of the jth inspection task, respectively, and the inspector assigned to the task is available within this time range; The constraint condition is expressed as: Skill matching constraint: For each task j, there is at least one inspector assigned with matching skills: Workload constraint: The total workload of each inspector does not exceed their maximum workload: Among them, W ij represents the workload of the i-th inspector performing the j-th task; Time limit constraints: ensure that the inspectors assigned to each task are available within the time frame of the task; Based on the constraints, the objective function and constraints of the personnel allocation optimization algorithm are constructed. The objective function is to maximize the efficiency of task allocation or minimize the idle time of detection personnel. The maximized allocation efficiency is expressed as: Where n is the total number of inspectors, m is the total number of inspection tasks, and X ij is a decision variable. If the i-th inspector is assigned to the j-th task, then X ij =1, otherwise X ij =0; minimizing the idle time of all inspectors is to minimize the sum of the differences between the available working time and the working time of the assigned tasks: Where n is the total number of inspectors, m is the total number of inspection tasks, T i is the available working time of the i-th inspector, D j is the time required to complete the jth inspection task, X ij is a decision variable. If the i-th inspector is assigned to the j-th task, then X ij =1, otherwise X ij =0.

2. A textile shipment inspection and control system according to claim 1, characterized in that: The acquisition method of the data acquisition module includes: By scanning the identification code on the textile to be inspected, the basic information of the textile, including material, size and color, can be automatically extracted; Use OCR to read text information of shipping requirements from textile packaging or accompanying documents. Shipping requirements include durability, color fastness, and texture clarity standards. The scanned and recognized information is digitized and transmitted to the inspection task allocation module.

3. A textile shipment inspection and control system according to claim 1, characterized in that: The average pooling operation includes: assuming that the dimensions of F2 are H×W×D, where H and W are the height and width of the feature map respectively, and D is the depth of the feature map. The goal of average pooling is to reduce the dimensions of H and W while keeping the dimension of D unchanged. The specific operation is as follows: Select the pooling window size: First, determine the size of the pooling window p×p, where p is the side length of the pooling window. The window size affects the coarseness of the pooling. Sliding window and calculating the average value: Slide the pooling window on F2 and calculate the average value of all elements in each window. The average value generated by each window corresponds to an element in the output feature vector V2. Output feature vector: Repeat sliding the window and calculating the average value until the entire feature map F2 is covered. The final average value constitutes a fixed-size feature vector V2, which is expressed as follows: Where V2(d) is the element of vector V2 at depth d, and F2(h, w, d) is the element of F2 at position (h, w) and depth d.

Images