Fiber raw material tracing method and system based on big data

Abstract

The invention discloses a fiber raw material traceability method and monitoring system based on big data, and the method comprises the steps: collecting the multi-source data of a raw material through an Internet of Things device, including image data, sensor data and a unique identity; performing standardization processing on the collected data, and combining to generate a comprehensive feature vector; inputting the comprehensive feature vector into an improved convolutional neural network, and outputting parameters of the fiber raw material, including crimpness, length, color feature, region and generation time; the traceability information is encrypted and stored through a block chain technology; a user retrieves and displays fiber raw material parameters in real time through the query module. By introducing the big data analysis and block chain technology, the problems of data splitting, easy tampering and insufficient intelligence in the existing fiber raw material traceability and quality monitoring method are solved, the traceability transparency, data security and monitoring efficiency are significantly improved, and the method is suitable for supply chain management and quality control in the textile industry.

Description

Technical Field

[0001] The present invention relates to the field of big data technology, and in particular to a fiber raw material tracing method and system based on big data. Background Art

[0002] Fiber raw materials (such as wool, linen, silk, etc.) are important basic materials for the textile industry, and their quality and source have an important impact on the performance and market competitiveness of the final product. However, the traceability and quality monitoring of fiber raw materials in the current textile industry face many challenges, which are mainly reflected in the following aspects: Supply chain complexity. The production and circulation of fiber raw materials involve multiple links, including raw material collection, transportation, processing and storage. Due to the long supply chain chain, the fragmentation of information between links often makes it difficult to trace the source and quality of products. For example, a batch of wool may go through multiple middlemen, and the path from the place of origin to the final textile factory is complicated. There is a lack of a unified recording mechanism, which can easily cause information loss or tampering. In addition, the quality recording methods of different links are different, making it difficult to form a unified traceability system.

[0003] At present, the traceability of fiber raw materials mostly adopts traditional methods such as barcodes, labels or manual records. These methods have the following problems: Easy to tamper: barcode labels are easy to be forged or tampered with, and it is difficult to ensure the authenticity of the data. Lack of real-time: manual records rely on human operations, data upload is delayed and accuracy cannot be guaranteed. Single data source: traditional traceability methods often only record batch numbers or simple origin information, ignoring the monitoring of multi-dimensional data such as environmental conditions and transportation status, resulting in one-sided traceability information.

[0004] The quality assessment of fiber raw material parameters usually relies on manual inspection or laboratory testing, which is not only costly but also inefficient. In addition, manual inspection is susceptible to subjective factors, and different inspectors may have different assessments of the same batch of raw materials. Existing quality assessment methods lack automated and intelligent analysis methods for fiber morphology (such as curl, length, color, etc.). The potential of blockchain technology in traceability has not been fully utilized. Blockchain technology, with its advantages such as decentralization, data immutability and traceability, provides new ideas for supply chain traceability. However, in the field of fiber raw materials, the application of blockchain technology is still in its early stages, and the main problems include: Insufficient integration of data collection and blockchain: Existing blockchain traceability solutions mostly focus on data storage, but lack effective integration of data collection and processing. Performance bottleneck: The processing speed and storage capacity of blockchain technology limit its application in large-scale supply chains.

[0005] Artificial intelligence technology (such as convolutional neural networks) has broad application prospects in the fields of image recognition and data analysis, but its potential has not been fully explored in the quality monitoring of fiber raw materials. For example, morphological parameters such as fiber curl, length and color characteristics can be automatically extracted through image data, but the existing industry lacks intelligent analysis methods for these parameters. In addition, the fusion and collaborative processing capabilities between different data sources (such as image data, sensor data and unique identification) are insufficient, resulting in a low level of intelligence in traceability and quality assessment. In addition, in the existing technology, the geographical factors of fiber raw materials are not fully considered when training convolutional neural networks, resulting in inaccurate identification.

[0006] As consumers' requirements for product quality and sustainability increase, the market demand for transparency in the traceability of fiber raw materials is increasing. For example, in the field of high-end textiles, consumers are more concerned about the origin and production process of products, and want to know whether the raw materials come from high-quality sources and meet environmental and ethical standards. However, the existing traceability system is difficult to meet these needs, especially when it involves multi-national supply chains, where information acquisition and verification are particularly difficult.

[0007] In summary, the existing technology cannot effectively solve the problems in fiber raw material traceability and quality monitoring. Therefore, there is an urgent need for a new method that can comprehensively utilize the Internet of Things, big data, artificial intelligence and blockchain technology to build a full-process traceability system from raw material collection to terminal query, and improve the quality monitoring ability of fiber raw materials through multi-source data fusion and intelligent analysis. The present invention is proposed to solve the above problems, aiming to realize the intelligent, real-time and efficient traceability of fiber raw materials, and provide an innovative solution for the textile industry.

[0008] Therefore, there is an urgent need for a fiber raw material traceability method based on big data to achieve large-scale and accurate fiber material traceability identification. The present invention solves the problems of incomplete data collection, insufficient feature extraction and low model accuracy in the prior art by introducing an improved convolutional neural network, providing a scientific basis and technical support for fiber raw material traceability, which has important practical application value and promotion prospects. Summary of the invention

[0009] In view of the above problems mentioned in the prior art, in order to solve the above technical problems, the present invention provides a fiber raw material traceability method and system based on big data, including: using Internet of Things devices to collect multi-source data of raw materials, including image data, sensor data and unique identity identification; standardizing the collected data and combining to generate a comprehensive feature vector; inputting the comprehensive feature vector into an improved convolutional neural network, and outputting the parameters of the fiber raw material, including curl, length, color characteristics, region and generation time; encrypting and storing the traceability information through blockchain technology; users retrieve and display the fiber raw material parameters in real time through a query module. The present invention solves the problems of data fragmentation, easy tampering and lack of intelligence in the existing fiber raw material traceability and quality monitoring methods by introducing big data analysis and blockchain technology, significantly improves the transparency of traceability, data security and monitoring efficiency, and is suitable for supply chain management and quality control in the textile industry.

[0010] A fiber raw material traceability method based on big data, comprising the steps of:

[0011] S1: Collect multi-source data of fiber raw materials, including image data, sensor data, and unique identity identification;

[0012] S2: compress the image data into a standardized JPEG format, encode the sensor data into JSON format, normalize the unique identifier to generate a standardized string, and combine the processed data to generate a comprehensive feature vector;

[0013] S3: Input the comprehensive feature vector into the trained improved convolutional neural network, and the convolutional neural network outputs fiber raw material parameters, which include the curliness of the fiber raw material, the length of the fiber, the color characteristics of the fiber raw material, the region, and the generation time; the loss function L used in the training of the improved convolutional neural network is:

[0014]

[0015] Where L is the total loss value, N is the total number of fiber raw material samples; α k is the weight of category k, y i,k is the true label of sample i belonging to category k, which takes the value of 1 or 0; y' i,k is the predicted probability that sample i belongs to category k, and its value range is [0,1]; K is the total number of categories; category k is the province where each fiber raw material sample is located determined by its GPS coordinates, and the fiber raw material samples are divided by province, and each province corresponds to a category k;

[0016] S4: Use blockchain technology to encrypt and store traceability information to ensure that the data cannot be tampered with; establish a distributed database in the cloud to store all traceability information for quick access and analysis: traceability information includes collected multi-source data and fiber raw material parameters;

[0017] S5: When the user queries, the fiber raw material parameters in the blockchain are retrieved and displayed through the terminal, and the information is sent to the customer terminal.

[0018] Preferably, the sensor data is obtained by installing sensors at the raw material collection site using IoT devices to record the ambient temperature, humidity, and light intensity data in real time; the unique identity is assigned to each batch of raw materials through radio frequency identification RFID or QR code, including the GPS coordinates of the source, timestamp and batch number; the location information, timestamp, ambient temperature and humidity during the transportation of raw materials are collected; the image data is obtained by recording the appearance of the raw materials through a high-definition camera to generate image data corresponding to the raw material batch.

[0019] Preferably, the method of using blockchain technology to encrypt and store the collected multi-source data includes: formatting the collected multi-source data, that is, storing the image data in a standardized JPEG format, storing the sensor data in a JSON format, encoding the unique identity identifier, and forming a standardized string; encrypting the formatted data by using the SHA-256 hash algorithm to generate a hash value of a fixed length; storing the encrypted data records in the blockchain, and verifying the legitimacy of the new block through a proof-of-work consensus mechanism; and updating the distributed ledgers of all nodes to ensure that the data cannot be tampered with.

[0020] Preferably, the processed data is combined to generate a comprehensive feature vector, including: using a color histogram to extract color distribution features of the image data, a grayscale co-occurrence matrix to extract texture features, an edge detection algorithm to extract edge features, and merging the extracted feature vectors into an image feature vector; normalizing and standardizing the sensor data to form a sensor feature vector of a fixed length; hash encoding the unique identity identifier and converting the encoding result into a numerical feature vector of a fixed length; and combining the image feature vector, the sensor feature vector and the unique identity identifier feature vector into a comprehensive feature vector through a splicing operation.

[0021] Preferably, the y' i,k The predicted probability that sample i belongs to category k includes: inputting the comprehensive feature vector into the trained convolutional neural network; extracting features from the comprehensive feature vector through the convolution layer and the pooling layer in turn, and flattening it to form a one-dimensional feature vector; inputting the flattened one-dimensional feature vector into the fully connected layer, and calculating the unnormalized score z for each category k i,k; Use the Softmax function to convert the unnormalized score into a predicted probability. The calculation formula is as follows:

[0022]

[0023] Among them, j is a variable with a value range of [1, K]; z i,j is the unnormalized score of sample i belonging to category k.

[0024] The present application also provides a fiber raw material traceability system based on big data, including:

[0025] The data acquisition module collects multi-source data of fiber raw materials, including image data, sensor data, and unique identification;

[0026] A comprehensive vector generation module compresses image data into a standardized JPEG format, encodes sensor data into JSON format, normalizes unique identifiers to generate a standardized string, and combines the processed data to generate a comprehensive feature vector;

[0027] The fiber raw material parameter recognition module inputs the comprehensive feature vector into the trained improved convolutional neural network, and the convolutional neural network outputs the fiber raw material parameters, which include the curliness of the fiber raw material, the length of the fiber, the color characteristics of the fiber raw material, the region, and the generation time. The loss function L used in the training of the improved convolutional neural network is:

[0028]

[0029] Where L is the total loss value, N is the total number of fiber raw material samples; α k is the weight of category k, y i,k is the true label of sample i belonging to category k, which takes the value of 1 or 0; y' i,k is the predicted probability that sample i belongs to category k, and its value range is [0,1]; K is the total number of categories; category k is the province where each fiber raw material sample is located determined by its GPS coordinates, and the fiber raw material samples are divided by province, and each province corresponds to a category k;

[0030] The blockchain data storage module uses blockchain technology to encrypt and store traceability information to ensure that the data cannot be tampered with; a distributed database is established in the cloud to store all traceability information for quick access and analysis: traceability information includes collected multi-source data and fiber raw material parameters;

[0031] In the query module, when the user queries, the fiber raw material parameters in the blockchain are retrieved and displayed through the terminal, and the information is sent to the customer terminal.

[0032] Preferably, the sensor data is obtained by installing sensors at the raw material collection site using IoT devices to record the ambient temperature, humidity, and light intensity data in real time; the unique identity is assigned to each batch of raw materials through radio frequency identification RFID or QR code, including the GPS coordinates of the source, timestamp and batch number; the location information, timestamp, ambient temperature and humidity during the transportation of raw materials are collected; the image data is obtained by recording the appearance of the raw materials through a high-definition camera to generate image data corresponding to the raw material batch.

[0033] Preferably, the method of using blockchain technology to encrypt and store the collected multi-source data includes: formatting the collected multi-source data, that is, storing the image data in a standardized JPEG format, storing the sensor data in a JSON format, encoding the unique identity identifier, and forming a standardized string; encrypting the formatted data by using the SHA-256 hash algorithm to generate a hash value of a fixed length; storing the encrypted data records in the blockchain, and verifying the legitimacy of the new block through a proof-of-work consensus mechanism; and updating the distributed ledgers of all nodes to ensure that the data cannot be tampered with.

[0034] Preferably, the processed data is combined to generate a comprehensive feature vector, including: using a color histogram to extract color distribution features of the image data, a grayscale co-occurrence matrix to extract texture features, an edge detection algorithm to extract edge features, and merging the extracted feature vectors into an image feature vector; normalizing and standardizing the sensor data to form a sensor feature vector of a fixed length; hash encoding the unique identity identifier and converting the encoding result into a numerical feature vector of a fixed length; and combining the image feature vector, the sensor feature vector and the unique identity identifier feature vector into a comprehensive feature vector through a splicing operation.

[0035] Preferably, the y' i,k The predicted probability that sample i belongs to category k includes: inputting the comprehensive feature vector into the trained convolutional neural network; extracting features from the comprehensive feature vector through the convolution layer and the pooling layer in turn, and flattening it to form a one-dimensional feature vector; inputting the flattened one-dimensional feature vector into the fully connected layer, and calculating the unnormalized score z for each category k i,k ; Use the Softmax function to convert the unnormalized score into a predicted probability. The calculation formula is as follows:

[0036]

[0037] Among them, j is a variable with a value range of [1, K]; z i,j is the unnormalized score of sample i belonging to category k.

[0038] The present invention provides a fiber raw material traceability method and system based on big data, which can achieve the following beneficial technical effects:

[0039] 1. The present invention uses IoT devices to collect multi-source data of raw materials, including image data, sensor data and unique identity identification; standardizes the collected data and combines them to generate comprehensive feature vectors; inputs the comprehensive feature vectors into the improved convolutional neural network, and outputs the parameters of fiber raw materials, including curl, length, color characteristics, region and generation time; encrypts and stores the traceability information through blockchain technology; and users retrieve and display fiber raw material parameters in real time through the query module. The present invention solves the problems of data fragmentation, easy tampering and lack of intelligence in existing fiber raw material traceability and quality monitoring methods by introducing big data analysis and blockchain technology, significantly improves the transparency of traceability, data security and monitoring efficiency, and is suitable for supply chain management and quality control in the textile industry.

[0040] 2. The present invention collects multi-source data of fiber raw materials during production, transportation and storage through Internet of Things devices, including image data, sensor data and unique identity information, and builds a full-process traceability system covering the raw material collection site to the end user. Users can query detailed information such as the source of raw materials, production environment, transportation route, etc. in real time to meet the needs of consumers and enterprises for supply chain transparency. It supports multi-source data fusion, improves the comprehensiveness of traceability and quality monitoring, generates comprehensive feature vectors through the fusion processing of image data, sensor data and unique identity information, comprehensively analyzes the multi-dimensional information that affects the quality of fiber raw materials, and builds a multi-dimensional quality traceability model, which solves the limitations of traditional single data source monitoring methods.

[0041] 3. The present invention uses blockchain technology to ensure the authenticity and non-tamperability of data, stores the collected multi-source data in the blockchain after formatting and encryption, and uses the decentralized and distributed ledger characteristics of the blockchain to ensure the authenticity and non-tamperability of the traceability data, effectively prevent the falsification of information in the supply chain, and improve the credibility of the data. 6. Through the distributed storage and automated intelligent analysis of the blockchain, the present invention reduces the workload of manual recording and repeated inspections, greatly improves the efficiency of supply chain management, and reduces the management costs of enterprises.

[0042] 4. The present invention introduces artificial intelligence algorithms to achieve intelligent analysis of quality parameters, processes comprehensive feature vectors through improved convolutional neural networks, and adds the weight of the category distribution of the fiber raw material's origin to the training considerations, effectively solving the problem of unbalanced data distribution and significantly improving the classification model's ability to identify small sample categories. The present invention can intelligently extract quality parameters such as curl, length, and color characteristics of fiber raw materials, replacing traditional manual inspection methods, greatly improving the efficiency and accuracy of quality assessment, and reducing subjective errors caused by human factors. BRIEF DESCRIPTION OF THE DRAWINGS

[0043] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0044] Figure 1 It is a schematic diagram of the steps of a fiber raw material tracing method based on big data of the present invention;

[0045] Figure 2 This is a fiber raw material traceability system diagram based on big data of the present invention. DETAILED DESCRIPTION

[0046] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0047] Embodiment 1:

[0048] In view of the above problems mentioned in the prior art, in order to solve the above technical problems, as shown in the attached Figure 1 As shown: A fiber raw material traceability method based on big data, comprising the steps of:

[0049] S1: Collect multi-source data of fiber raw materials, the multi-source data includes: image data, sensor data, unique identity; in one embodiment, image data acquisition equipment: install a high-definition camera (resolution not less than 1920x1080) at the wool shearing station of the ranch. Collection method During the wool shearing process, real-time image acquisition of the appearance of the wool of each sheep is performed. The color, curl and surface condition (such as gloss, stains, etc.) of the wool are photographed. Storage Compress the collected raw image data into a standardized JPEG format, and each image is bound to the unique identity of the batch of wool.

[0050] Sensor data collection equipment: The following sensors are deployed on the ranch and transport vehicles: Ambient temperature and humidity sensor: Real-time monitoring of ambient temperature and humidity during shearing (e.g., temperature is 22°C and humidity is 65%). Light intensity sensor: Records light intensity during shearing (e.g., light intensity is 1000 lux). GPS positioning device: Records the geographic coordinates of the location of the wool in the ranch and during transportation (e.g., latitude 34.0522, longitude 108.9370). Collection method: At the shearing station, the sensor records the shearing environment data in real time. During transportation, the sensors on the vehicle automatically collect location and environment data every 10 minutes. Storage The sensor data is stored in JSON format, containing the following information: ambient temperature (unit: °C): 22; ambient humidity (unit: %): 65; light intensity (unit: lux): 1000; GPS coordinates: latitude 34.0522, longitude 108.9370; timestamp: 2023-12-26T10:30:00Z. Unique identification, generation method: Each sheep's wool generates a unique RFID tag (RFID number is RF123456). The tag contains the following information: Origin: A ranch in Shaanxi Province, Cutting time: December 26, 2024 10:30 Batch number: Batch001; Binding method: The RFID tag corresponds to the image data and sensor data of the wool one by one. RFID number: RF123456; Image file name: RF123456.jpg; JSON data file name: RF123456.json

[0051] Data storage and transmission, all collected data is transmitted to the cloud server through the IoT gateway. Data organization: Image data is stored in a specified image database. Sensor data and unique identifiers are stored in a JSON database and associated with the file name of the image data.

[0052] In one embodiment, multi-source data collection of hemp fiber raw materials, image data collection equipment: install high-definition industrial cameras on the production line of the hemp fiber processing plant. Collection method During the hemp fiber separation process, the length and uniformity of the fibers are recorded by image. 10 random sample images are collected for each batch of hemp fibers. Storage: compressed into a standardized JPEG format and bound to the unique identity of the batch of hemp fibers. Sensor data collection equipment: the following sensors are deployed in the hemp fiber separation workshop: Temperature and humidity sensors: monitor the processing environment. Vibration sensors: record the vibration state of the processing equipment. GPS positioning device: record the geographical location of the processing workshop. Collection method The environmental sensors in the processing workshop record temperature and humidity data in real time. The vibration sensor collects equipment operation data once a minute to determine whether the processing is abnormal. Storage: Store data in JSON format, for example: ambient temperature: 25°C; ambient humidity: 55%; equipment vibration value (unit: g): 0.5; GPS coordinates: latitude 31.2304, longitude 121.4737; timestamp: 2024-12-26T14:15:00Z. Unique ID generation method: A QR code label is generated for each batch of hemp fiber. The label content includes: Origin: a processing plant; Processing time: 14:15, December 26, 2024; Batch number: Batch002; Data storage and transmission; Upload image data and sensor data to the data management system of the processing plant through the industrial Internet of Things gateway. The QR code label is associated with all the data of the batch of hemp fiber.

[0053] S2: compress the image data into a standardized JPEG format, encode the sensor data in JSON format, normalize the unique identity to generate a standardized string, and combine the processed data to generate a comprehensive feature vector; in one embodiment, the image data processes the raw data acquisition, and uses a high-definition camera to collect images of the appearance characteristics of the wool, including the color, surface gloss and curl of the wool. Each image is in RAW format with a resolution of 4000x3000 pixels. Data processing: The image data is large, and direct storage and analysis are not conducive to subsequent processing, so compression is required. The original image file is compressed into JPEG format and the resolution is adjusted to 1920x1080 pixels to ensure a small file size and appropriate image quality. Each image file is assigned a unique file name corresponding to the unique identity. The target result image file is unified and standardized to facilitate subsequent data analysis and storage. The image file of each batch of wool is bound to its RFID tag one by one, the sensor data processes the raw data acquisition, and the environmental sensor is used to collect the temperature, humidity and light intensity data of the wool when it is sheared in real time. The longitude and latitude information of the shearing location is collected using a GPS device. Data examples include: temperature: 22.4℃, humidity: 65%, light intensity: 800lux, GPS latitude and longitude: 34.0522 (latitude), 108.9370 (longitude), data encoding All sensor data are encoded in standard JSON format to facilitate association with image data and unique identification information. Unique identification processing Raw data collection, each batch of wool is assigned a unique identification, and the information is stored through the RFID tag. The unique identification includes the following: RFID number: RF123456, origin: a ranch in Shaanxi Province, clipping time: 10:30 on December 26, 2024, batch number: Batch001, data processing: This information is spliced ​​into a standardized string. To ensure security and uniformity, a standard encoding method (such as the SHA algorithm) is used to convert the spliced ​​string into a fixed format hash value. Target result: Generate a unique identification string or hash value that corresponds to each batch of wool data. Comprehensive feature vector generation, data combines image data (such as image file name or extracted features), sensor data (JSON format encoding) and unique identifier (encoded string) in a fixed order. The target form combines the above data into a comprehensive feature vector, for example: Image feature vector: color distribution, texture features, etc. (key information extracted from the image). Sensor data feature vector: ambient temperature, humidity, light intensity and GPS coordinates. Unique identifier feature vector: encoded unique identifier. The comprehensive feature vector is a unified data input, formatted as a fixed structure data object, which is convenient for subsequent convolutional neural network or other analysis modules to process.

[0054] S3: Input the comprehensive feature vector into the trained improved convolutional neural network, and the convolutional neural network outputs fiber raw material parameters, which include the curliness of the fiber raw material, the length of the fiber, the color characteristics of the fiber raw material, the region, and the generation time; the loss function L used in the training of the improved convolutional neural network is:

[0055]

[0056] Where L is the total loss value, N is the total number of fiber raw material samples; α k is the weight of category k, y i,k is the true label of sample i belonging to category k, which takes the value of 1 or 0; y' i,k is the predicted probability that sample i belongs to category k, and its value range is [0,1]; K is the total number of categories; category k is the province where each fiber raw material sample is located determined by its GPS coordinates, and the fiber raw material samples are divided by province, and each province corresponds to a category k; in one embodiment, the structure of the convolutional neural network includes an input layer: the input is a comprehensive feature vector, including image features, sensor features, and unique identity features. For example: Image features: 271 dimensions (color, texture, edge information). Sensor features: 5 dimensions (temperature and humidity, light intensity, GPS information). Unique identity features: 32 dimensions (encoded hash value). The total dimension of the comprehensive feature vector is 308 dimensions. Convolution layer: Use multiple groups of convolution kernels (such as 32 3x1 kernels) to process the comprehensive feature vector and extract local patterns of high-dimensional features. The convolution operation can identify specific patterns or correlations, such as the potential relationship between image features and sensor features. Pooling layer: Apply pooling operations (such as maximum pooling) on ​​the convolved feature map to further reduce dimensionality and reduce redundant information. The main function of the pooling layer is to retain important features while reducing the data dimension and improving computational efficiency. Fully connected layer: The feature map after convolution and pooling is flattened into a one-dimensional feature vector and passed to the fully connected layer. The fully connected layer comprehensively analyzes the features through the weight matrix to generate predicted values ​​for each output parameter. Output layer: The output layer consists of multiple neurons, each corresponding to a target parameter: Curl prediction: continuous value output (e.g. 0.1 to 1.0). Fiber length prediction: continuous value output (e.g. 20 to 100 mm). Color feature: multi-category classification output (e.g. white, gray, brown). Region prediction: multi-category classification output (e.g. Shaanxi, Inner Mongolia, Xinjiang, etc.). Generation time: time range classification output (e.g. morning, afternoon, evening). Loss function: The category-weighted cross entropy loss function is used to optimize classification tasks (e.g. color, region, generation time). The mean square error loss function is used for continuous value prediction tasks (e.g. curl, length).

[0057] Working principle of convolutional neural network, data input: the comprehensive feature vector is used as the network input and enters the convolution layer for processing. Image features provide visual features of the fiber, sensor features reflect environmental conditions, and unique identifiers ensure that each piece of data is unique. Feature extraction: the convolution layer identifies local patterns in the comprehensive feature vector, such as the correlation between image features and sensor features. The stacking of multiple layers of convolution can gradually extract high-level semantic features, for example: Convolution 1: Identify color and texture patterns. Convolution 2: Analyze the potential impact of environmental conditions on the appearance of the fiber. Feature dimensionality reduction: the pooling layer reduces the dimensionality of the convolutional features, retaining important information while reducing the amount of calculation. Feature fusion: the fully connected layer flattens the reduced feature map into a one-dimensional vector. Features from different sources are comprehensively analyzed in the fully connected layer to generate a high-dimensional feature representation. Output prediction: the output layer predicts the various parameters of the fiber based on the analysis results of the fully connected layer. The classification task generates the probability distribution of each category through the Softmax function, and the numerical prediction task outputs a continuous value. Model optimization: the model is optimized using the training data, and the network parameters are updated through the back propagation algorithm. During the training process, the category weights in the loss function are dynamically adjusted to solve the problem of uneven sample distribution.

[0058] Input data image features: The color distribution of a batch of wool is light white, with high curl and uniform edge texture. Sensor features: The ambient temperature during shearing is 22°C, the humidity is 65%, the light intensity is 800lux, and the GPS is located in Shaanxi. Unique identity: The corresponding RFID number is RF123456. Prediction output: Curl: The predicted value is 0.85 (high curl). Length: The predicted value is 55 mm (medium length). Color features: Predicted to be "white" category with a probability of 98%. Region: Predicted to be "Shaanxi" category with a probability of 95%. Generation time: Predicted to be "afternoon" category with a probability of 89%.

[0059] S4: Use blockchain technology to encrypt and store traceability information to ensure that the data cannot be tampered with; establish a distributed database in the cloud to store all traceability information for quick access and analysis: traceability information includes collected multi-source data and fiber raw material parameters; in one embodiment, the structural block composition of the blockchain: each block stores traceability information of a batch of fiber raw materials, including the following: hash value of multi-source data (such as image data, sensor data, unique identity). Hash value of the previous block (used to link blocks). Timestamp (record the generation time of the block). Block number and traceability data summary (brief summary of data content). Chain structure: Blocks form a chain structure through their hash value and the hash value of the previous block. Each block is closely associated with the previous block to form an unalterable blockchain. For example: Block 1: records the traceability information of wool batch 001, including its hash value. Block 2: records the information of wool batch 002, including its own hash value and the hash value of block 1. Distributed storage: The blockchain is stored on multiple nodes, and all nodes have a complete copy of the blockchain. Nodes include corporate headquarters, processing plants and supply chain partners. The working principle of blockchain, data encryption: After collecting multi-source data of wool, the system formats the data (such as JPEG, JSON encoding). The formatted data is generated into a hash value of fixed length through an encryption algorithm (such as SHA-256). The hash value is the unique identifier of the data and is used to verify the integrity and authenticity of the data. Block generation: The system generates a new block for each batch of fiber raw materials. The block content includes: Hash value: records the summary of the batch data. The hash value of the previous block: ensures the continuity of the blockchain. Timestamp: records the specific time when the block is generated. Block number: identifies the order of the block in the chain. For example: Block number: Block002; Timestamp: 10:30, December 26, 2024; Current block hash value: abc123…; Previous block hash value: xyz789…; Data summary: Multi-source data hash of wool batch 002; Consensus mechanism: The system uses the Proof of Work (PoW) consensus mechanism to verify the legitimacy of the new block. Each node needs to verify whether the hash value of the new block meets specific rules through calculation. After verification, all nodes synchronously update the blockchain. Data storage: The newly generated block is stored in the blockchain copy of each node to ensure the distributed storage of data. When querying data, you only need to retrieve the blockchain from any node. Data query: The user enters a unique identity (such as an RFID number) through the query interface, and the system locates the corresponding block. After the system verifies the integrity of the data, it extracts multi-source data and fiber parameters from the block and returns them to the user. In the storage process, the traceability information of a batch of wool (batch number: Batch002) is as follows: Image data: hash value of the JPEG file. Sensor data: Hash value of the data in JSON format. Unique ID: Hash value of the RFID number.The block content generated by the system: Block number: Block002; Current block hash value: e1a3d…; Previous block hash value: d9c5b…; Data summary: {"image hash":"abc123","sensor hash":"xyz456","RFID hash":"pqr789"}; Timestamp: 2024-12-26T10:30:00Z; In the query process, the user queries the traceability information of the batch number Batch002. The system searches the blockchain and finds the block related to Batch002. The multi-source data hash value in the block is extracted and decoded into user-readable information: Image: The wool color is "light white". Sensor: Cutting temperature is 22.4℃ and humidity is 65%. Unique identity: The source is "a ranch in Shaanxi Province".

[0060] S5: When the user queries, the fiber raw material parameters in the blockchain are retrieved and displayed through the terminal, and the information is sent to the client terminal. In one embodiment, the query process includes query initiation: the user accesses the fiber raw material traceability system through a smart terminal (such as a mobile phone, computer or tablet) and enters the specified query conditions. For example: Condition 1: unique identity (such as RFID number RF123456). Condition 2: batch number (such as Batch002). After the query is initiated, the system sends the request to the blockchain node. Blockchain retrieval: The system scans blocks one by one in the blockchain according to the input query conditions to find matching traceability information. Retrieval basis: If the RFID number is entered, the system compares the data summary field in each block to find the block corresponding to RF123456. If the batch number is entered, the system locates the block through the association between the unique identity and the batch number. Data verification: After retrieving the target block, the system verifies the data hash value in the block to ensure that the data has not been tampered with. Verification method: By decoding the hash value in the block, it is compared with the hash value recalculated from the original collected data. If the verification is successful, it proves that the data is authentic; if the verification fails, the system issues a warning. Parameter extraction: Extract key information of fiber raw materials in the target block, including: Image features: color and curl of wool appearance image. Sensor data: temperature (22.4℃), humidity (65%), light intensity (800lux), GPS location (a ranch in Shaanxi Province) at the time of cutting. Unique identity: RFID number RF123456, generated at 10:30 on December 26, 2024. Data display: The system displays the extracted fiber parameters and traceability information in a structured manner on the user terminal. Display content: Batch number: Batch002RFID; Number: RF123456; Color: Light white; Curl: High; Length: 55 mm; Temperature: 22.4℃; Humidity: 65%; Light intensity: 800lux; Origin: A ranch in Shaanxi Province; Cutting time: 10:30 on December 26, 2024; Users can view specific parameter values ​​and verify the origin and quality information. Data transmission: The system sends the query results to the customer terminal (such as the management system of the corporate customer) through an encrypted network channel. The data is formatted as JSON or XML for easy reception and further processing by the customer system. Corporate customers can directly use the data for quality traceability, supply chain management or customer service.

[0061] In some embodiments, the sensor data is obtained by installing sensors at the raw material collection site using Internet of Things devices to record environmental temperature, humidity, and light intensity data in real time; the unique identity is assigned to each batch of raw materials through radio frequency identification RFID or QR code, including the GPS coordinates of the source, timestamp and batch number; the location information, timestamp, ambient temperature and humidity during the transportation of raw materials are collected; the image data is obtained by recording the appearance of the raw materials through a high-definition camera to generate image data corresponding to the raw material batch.

[0062] In some embodiments, the method of using blockchain technology to encrypt and store the collected multi-source data includes: formatting the collected multi-source data, that is, storing the image data in a standardized JPEG format, storing the sensor data in a JSON format, encoding the unique identity identifier, and forming a standardized string; encrypting the formatted data using the SHA-256 hash algorithm to generate a fixed-length hash value; storing the encrypted data records in the blockchain, and verifying the legitimacy of the new block through a proof-of-work consensus mechanism; updating the distributed ledgers of all nodes to ensure that the data cannot be tampered with. In some embodiments, the SHA-256 hash algorithm is defined as follows: SHA-256 is a cryptographically secure hash function that converts data input of any length into a fixed-length 256-bit (32-byte) hash value. Features: Irreversibility: The original data cannot be derived from the hash value. Uniqueness: Different inputs almost always generate different hash values. Fixed length: Regardless of the input length, the output is always 256 bits. Anti-collision: It is difficult to find two input data that generate the same hash value. Data encryption process: Data preparation, the collected multi-source data includes image files (JPEG format, file name: Batch002_image.jpg). Sensor data JSON format, unique identity string form, SHA-256 hash calculation, calculate the hash value for each piece of data separately, and combine the above generated hash value, timestamp and hash value of the previous block to form a new block. Proof of Work (PoW) consensus mechanism, Proof of Work definition: PoW is a blockchain consensus mechanism that requires participating nodes to solve a complex mathematical problem to verify the legitimacy of the new block. Solving the problem requires certain computing resources and time. Core idea: Nodes need to find a hash value that meets specific rules. The difficulty is determined by the specific format of the target hash value (such as the number of leading zeros). Proof of Work process, target setting, the blockchain system sets a difficulty target, for example: the generated block hash value must start with 4 zeros (for example: 0000xxxxxx...). Random number attempt: Each node adds a random number field to the new block, repeatedly adjusts the random number and recalculates the block hash value until the target condition is met. When the calculated hash value does not meet the condition, the random number is increased and retried, and finally a random number field is found so that the block hash value meets the target. Verification process: After receiving the new block, other nodes verify the following: Whether the hash value in the block meets the target condition. Whether the hash value of the data is consistent with the original content. Whether the blockchain maintains continuity (check the hash value of the previous block). Consensus reached: When the majority of nodes pass the verification, the new block is officially added to the blockchain, and all nodes synchronously update the distributed ledger.

[0063] In some embodiments, the processed data is combined to generate a comprehensive feature vector, including: using a color histogram to extract color distribution features from image data, a grayscale co-occurrence matrix to extract texture features, an edge detection algorithm to extract edge features, and merging the extracted feature vectors into an image feature vector; normalizing and standardizing the sensor data to form a sensor feature vector of a fixed length; hash encoding the unique identity identifier and converting the encoding result into a numerical feature vector of a fixed length; and combining the image feature vector, the sensor feature vector, and the unique identity identifier feature vector into a comprehensive feature vector through a splicing operation.

[0064] In some embodiments, the y' i,k The predicted probability that sample i belongs to category k includes: inputting the comprehensive feature vector into the trained convolutional neural network; extracting features from the comprehensive feature vector through the convolution layer and the pooling layer in turn, and flattening it to form a one-dimensional feature vector; inputting the flattened one-dimensional feature vector into the fully connected layer, and calculating the unnormalized score z for each category k i,k ; Use the Softmax function to convert the unnormalized score into a predicted probability. The calculation formula is as follows:

[0065]

[0066] Among them, j is a variable with a value range of [1, K]; z i,j is the unnormalized score of sample i belonging to category k.

[0067] Embodiment 2:

[0068] This application also provides a fiber raw material traceability system based on big data, such as Figure 2As shown, the hardware components of the system include: data acquisition module, high-definition camera function: collect appearance images of fiber raw materials, such as color, curl, and gloss. The resolution is not less than 1920x1080 and is installed at the fiber raw material collection station or production line. The sensor collects environmental parameters, including temperature and humidity, light intensity, GPS location, etc. Temperature and humidity sensor (such as DHT22): records the temperature and humidity of the collection environment. Light sensor (such as BH1750): monitors light intensity. GPS module (such as NEO-6M): obtains geographic location coordinates. RFID reader reads the unique identification (RFID tag) of each batch of fiber raw materials. The frequency range is 13.56MHz, which is used for tag reading. Data processing module: The edge computing device performs preliminary processing on the collected image data, sensor data, and unique identification. Embedded processor (such as NVIDIA JetsonNano) or micro industrial computer. The data formatting module compresses the image data into a standardized JPEG format, encodes the sensor data into a JSON format, and hashes the unique identification. Data storage and blockchain module, blockchain node server stores blockchain copies of traceability information, verifies and synchronizes blockchain data. Servers running Hyperledger Fabric or Ethereum blockchain platforms. Distributed databases provide the ability to quickly access and analyze traceability information. Cloud databases (such as MongoDB Atlas) are used to store summaries and query results of blockchain records. User query module: terminal devices provide query portals for users. Consumers use smartphones to scan QR codes. Enterprise users access the Web interface of the traceability system through computers. Display devices display traceability information in a visual way. LCD screens, mobile phone screens, industrial terminal screens. Network communication module IoT gateway connects the data acquisition module with the blockchain node server to upload and synchronize data. IoT communication gateways that support 4G / 5G / Wi-Fi. Cloud service interface realizes data interaction between blockchain and distributed database. Use HTTPAPI or WebSocket protocol.

[0069] System connection mode, data acquisition module connection, high-definition cameras, sensors and RFID readers are connected to edge computing devices through industrial buses (such as RS485 or CAN bus). The edge computing device is connected to the IoT gateway via Ethernet or Wi-Fi to transmit processed data in real time. Data processing module connection: The edge computing device compresses, encodes and normalizes the collected data, and then uploads it to the blockchain node server through the IoT gateway. The formatted data includes: image file: JPEG format. Sensor data: JSON encoding. Unique identity: hash value. Blockchain and database connection: The blockchain node server is connected to the distributed database through the cloud service interface: The blockchain stores complete traceability data records. The database stores simplified query indexes and provides quick access functions. User query module connection: Users access the query interface through smart terminals, interact with the distributed database, and obtain traceability information. The query conditions (such as RFID number or batch number) are sent to the block data acquisition module through the API to collect multi-source data of fiber raw materials. The multi-source data includes: image data, sensor data, and unique identity;

[0070] A comprehensive vector generation module compresses image data into a standardized JPEG format, encodes sensor data into JSON format, normalizes unique identifiers to generate a standardized string, and combines the processed data to generate a comprehensive feature vector;

[0071] The fiber raw material parameter recognition module inputs the comprehensive feature vector into the trained improved convolutional neural network, and the convolutional neural network outputs the fiber raw material parameters, which include the curliness of the fiber raw material, the length of the fiber, the color characteristics of the fiber raw material, the region, and the generation time. The loss function L used in the training of the improved convolutional neural network is:

[0072]

[0073] Where L is the total loss value, N is the total number of fiber raw material samples; α k is the weight of category k, y i,k is the true label of sample i belonging to category k, which takes the value of 1 or 0; y' i,k is the predicted probability that sample i belongs to category k, and its value range is [0,1]; K is the total number of categories; category k is the province where each fiber raw material sample is located determined by its GPS coordinates. The fiber raw material samples are divided by province, and each province corresponds to a category k.

[0074] The blockchain data storage module uses blockchain technology to encrypt and store traceability information to ensure that the data cannot be tampered with; a distributed database is established in the cloud to store all traceability information for quick access and analysis: traceability information includes collected multi-source data and fiber raw material parameters;

[0075] In the query module, when the user queries, the fiber raw material parameters in the blockchain are retrieved and displayed through the terminal, and the information is sent to the customer terminal.

[0076] In some embodiments, the sensor data is obtained by installing sensors at the raw material collection site using Internet of Things devices to record environmental temperature, humidity, and light intensity data in real time; the unique identity is assigned to each batch of raw materials through radio frequency identification RFID or QR code, including the GPS coordinates of the source, timestamp and batch number; the location information, timestamp, ambient temperature and humidity during the transportation of raw materials are collected; the image data is obtained by recording the appearance of the raw materials through a high-definition camera to generate image data corresponding to the raw material batch.

[0077] In some embodiments, the use of blockchain technology to encrypt and store the collected multi-source data includes: formatting the collected multi-source data, that is, storing the image data in a standardized JPEG format, storing the sensor data in a JSON format, encoding the unique identity identifier, and forming a standardized string; encrypting the formatted data by using the SHA-256 hash algorithm to generate a hash value of a fixed length; storing the encrypted data records in the blockchain, and verifying the legitimacy of the new block through a proof-of-work consensus mechanism; and updating the distributed ledgers of all nodes to ensure that the data cannot be tampered with.

[0078] In some embodiments, the processed data is combined to generate a comprehensive feature vector, including: using a color histogram to extract color distribution features from image data, a grayscale co-occurrence matrix to extract texture features, an edge detection algorithm to extract edge features, and merging the extracted feature vectors into an image feature vector; normalizing and standardizing the sensor data to form a sensor feature vector of a fixed length; hash encoding the unique identity identifier and converting the encoding result into a numerical feature vector of a fixed length; and combining the image feature vector, the sensor feature vector, and the unique identity identifier feature vector into a comprehensive feature vector through a splicing operation.

[0079] In some embodiments, the y' i,k The predicted probability that sample i belongs to category k includes: inputting the comprehensive feature vector into the trained convolutional neural network; extracting features from the comprehensive feature vector through the convolution layer and the pooling layer in turn, and flattening it to form a one-dimensional feature vector; inputting the flattened one-dimensional feature vector into the fully connected layer, and calculating the unnormalized score z for each category k i,k; Use the Softmax function to convert the unnormalized score into a predicted probability. The calculation formula is as follows:

[0080]

[0081] Among them, j is a variable with a value range of [1, K]; z i,j is the unnormalized score of sample i belonging to category k.

[0082] The present invention provides a fiber raw material traceability method and system based on big data, which can achieve the following beneficial technical effects:

[0083] 1. The present invention uses IoT devices to collect multi-source data of raw materials, including image data, sensor data and unique identity identification; standardizes the collected data and combines them to generate comprehensive feature vectors; inputs the comprehensive feature vectors into the improved convolutional neural network, and outputs the parameters of fiber raw materials, including curl, length, color characteristics, region and generation time; encrypts and stores the traceability information through blockchain technology; and users retrieve and display fiber raw material parameters in real time through the query module. The present invention solves the problems of data fragmentation, easy tampering and lack of intelligence in existing fiber raw material traceability and quality monitoring methods by introducing big data analysis and blockchain technology, significantly improves the transparency of traceability, data security and monitoring efficiency, and is suitable for supply chain management and quality control in the textile industry.

[0084] 2. The present invention collects multi-source data of fiber raw materials during production, transportation and storage through Internet of Things devices, including image data, sensor data and unique identity information, and builds a full-process traceability system covering the raw material collection site to the end user. Users can query detailed information such as the source of raw materials, production environment, transportation route, etc. in real time to meet the needs of consumers and enterprises for supply chain transparency. It supports multi-source data fusion, improves the comprehensiveness of traceability and quality monitoring, generates comprehensive feature vectors through the fusion processing of image data, sensor data and unique identity information, comprehensively analyzes the multi-dimensional information that affects the quality of fiber raw materials, and builds a multi-dimensional quality traceability model, which solves the limitations of traditional single data source monitoring methods.

[0085] 3. The present invention uses blockchain technology to ensure the authenticity and non-tamperability of data, stores the collected multi-source data in the blockchain after formatting and encryption, and uses the decentralized and distributed ledger characteristics of the blockchain to ensure the authenticity and non-tamperability of the traceability data, effectively prevent the falsification of information in the supply chain, and improve the credibility of the data. 6. Through the distributed storage and automated intelligent analysis of the blockchain, the present invention reduces the workload of manual recording and repeated inspections, greatly improves the efficiency of supply chain management, and reduces the management costs of enterprises.

[0086] 4. The present invention introduces artificial intelligence algorithms to achieve intelligent analysis of quality parameters, processes comprehensive feature vectors through improved convolutional neural networks, and adds the weight of the category distribution of the fiber raw material's origin to the training considerations, effectively solving the problem of unbalanced data distribution and significantly improving the classification model's ability to identify small sample categories. The present invention can intelligently extract quality parameters such as curl, length, and color characteristics of fiber raw materials, replacing traditional manual inspection methods, greatly improving the efficiency and accuracy of quality assessment, and reducing subjective errors caused by human factors.

[0087] The above is a detailed introduction to a fiber raw material traceability method and system based on big data. Specific examples are used in this article to illustrate the principles and implementation methods of the present invention. The description of the above embodiments is only used to help understand the core idea of ​​the present invention; at the same time, for general technical personnel in this field, according to the ideas and methods of the present invention, there will be changes in the specific implementation methods and application scopes. In summary, the content of this specification should not be understood as a limitation on the present invention.

Claims

1. A fiber raw material traceability method based on big data, characterized in that: Includes steps: S1: Collect multi-source data of fiber raw materials, including image data, sensor data, and unique identity identification; S2: compress the image data into a standardized JPEG format, encode the sensor data into JSON format, normalize the unique identifier to generate a standardized string, and combine the processed data to generate a comprehensive feature vector; S3: Input the comprehensive feature vector into the trained improved convolutional neural network, and the convolutional neural network outputs fiber raw material parameters, which include the curliness of the fiber raw material, the length of the fiber, the color characteristics of the fiber raw material, the region, and the generation time; the loss function L used in the training of the improved convolutional neural network is: Where L is the total loss value, N is the total number of fiber raw material samples; α k is the weight of category k, y i,k is the true label of sample i belonging to category k, which takes the value of 1 or 0; y' i,k is the predicted probability that sample i belongs to category k, and its value range is [0,1]; K is the total number of categories; category k is the province where each fiber raw material sample is located determined by its GPS coordinates, and the fiber raw material samples are divided by province, and each province corresponds to a category k; S4: Use blockchain technology to encrypt and store traceability information to ensure that the data cannot be tampered with; establish a distributed database in the cloud to store all traceability information for quick access and analysis: traceability information includes collected multi-source data and fiber raw material parameters; S5: When the user queries, the fiber raw material parameters in the blockchain are retrieved and displayed through the terminal, and the information is sent to the customer terminal.

2. A fiber raw material traceability method based on big data as claimed in claim 1, characterized in that: The sensor data is obtained by installing sensors at the raw material collection site using IoT devices to record the ambient temperature, humidity, and light intensity data in real time; the unique identity is assigned to each batch of raw materials through radio frequency identification RFID or QR code, including the GPS coordinates of the source, timestamp and batch number; the location information, timestamp, ambient temperature and humidity during the transportation of raw materials are collected; the image data is obtained by recording the appearance of the raw materials through a high-definition camera to generate image data corresponding to the raw material batch.

3. A fiber raw material traceability method based on big data as claimed in claim 1, characterized in that: The method uses blockchain technology to encrypt and store the collected multi-source data, including: formatting the collected multi-source data, that is, storing the image data in a standardized JPEG format, storing the sensor data in a JSON format, encoding the unique identity identifier, and forming a standardized string; encrypting the formatted data by using the SHA-256 hash algorithm to generate a hash value of a fixed length; storing the encrypted data records in the blockchain, and verifying the legitimacy of the new block through a proof-of-work consensus mechanism; and updating the distributed ledgers of all nodes to ensure that the data cannot be tampered with.

4. The fiber raw material tracing method based on big data according to claim 1, characterized in that: The method combines the processed data to generate a comprehensive feature vector, including: extracting color distribution features from image data using a color histogram, extracting texture features from a grayscale co-occurrence matrix, extracting edge features from an edge detection algorithm, and merging the extracted feature vectors into an image feature vector; normalizing and standardizing the sensor data to form a sensor feature vector of a fixed length; performing hash coding on a unique identity identifier and converting the coding result into a numerical feature vector of a fixed length; and combining the image feature vector, the sensor feature vector, and the unique identity identifier feature vector into a comprehensive feature vector through a splicing operation.

5. The fiber raw material tracing method based on big data according to claim 1, characterized in that: The y' i,k The predicted probability that sample i belongs to category k includes: inputting the comprehensive feature vector into the trained convolutional neural network; extracting features from the comprehensive feature vector through the convolution layer and the pooling layer in turn, and flattening it to form a one-dimensional feature vector; inputting the flattened one-dimensional feature vector into the fully connected layer, and calculating the unnormalized score z for each category k i,k ; Use the Softmax function to convert the unnormalized score into a predicted probability. The calculation formula is as follows: Among them, j is a variable with a value range of [1, K]; z i,j is the unnormalized score of sample i belonging to category k.

6. A fiber raw material traceability system based on big data, characterized in that: include: The data acquisition module collects multi-source data of fiber raw materials, including image data, sensor data, and unique identification; A comprehensive vector generation module compresses image data into a standardized JPEG format, encodes sensor data into JSON format, normalizes unique identifiers to generate a standardized string, and combines the processed data to generate a comprehensive feature vector; The fiber raw material parameter recognition module inputs the comprehensive feature vector into the trained improved convolutional neural network, and the convolutional neural network outputs the fiber raw material parameters, which include the curliness of the fiber raw material, the length of the fiber, the color characteristics of the fiber raw material, the region, and the generation time. The loss function L used in the training of the improved convolutional neural network is: Where L is the total loss value, N is the total number of fiber raw material samples; α k is the weight of category k, y i,k is the true label of sample i belonging to category k, which takes the value of 1 or 0; y' i,k is the predicted probability that sample i belongs to category k, and its value range is [0,1]; K is the total number of categories; category k is the province where each fiber raw material sample is located determined by its GPS coordinates, and the fiber raw material samples are divided by province, and each province corresponds to a category k; The blockchain data storage module uses blockchain technology to encrypt and store traceability information to ensure that the data cannot be tampered with; a distributed database is established in the cloud to store all traceability information for quick access and analysis: traceability information includes collected multi-source data and fiber raw material parameters; In the query module, when the user queries, the fiber raw material parameters in the blockchain are retrieved and displayed through the terminal, and the information is sent to the customer terminal.

7. The fiber raw material traceability system based on big data according to claim 1, characterized in that: The sensor data is obtained by installing sensors at the raw material collection site using IoT devices to record the ambient temperature, humidity, and light intensity data in real time; the unique identity is assigned to each batch of raw materials through radio frequency identification RFID or QR code, including the GPS coordinates of the source, timestamp and batch number; the location information, timestamp, ambient temperature and humidity during the transportation of raw materials are collected; the image data is obtained by recording the appearance of the raw materials through a high-definition camera to generate image data corresponding to the raw material batch.

8. The fiber raw material traceability system based on big data according to claim 1, characterized in that: The method uses blockchain technology to encrypt and store the collected multi-source data, including: formatting the collected multi-source data, that is, storing the image data in a standardized JPEG format, storing the sensor data in a JSON format, encoding the unique identity identifier, and forming a standardized string; encrypting the formatted data by using the SHA-256 hash algorithm to generate a hash value of a fixed length; storing the encrypted data records in the blockchain, and verifying the legitimacy of the new block through a proof-of-work consensus mechanism; and updating the distributed ledgers of all nodes to ensure that the data cannot be tampered with.

9. The fiber raw material traceability system based on big data according to claim 1, characterized in that: The method combines the processed data to generate a comprehensive feature vector, including: extracting color distribution features from image data using a color histogram, extracting texture features from a grayscale co-occurrence matrix, extracting edge features from an edge detection algorithm, and merging the extracted feature vectors into an image feature vector; normalizing and standardizing the sensor data to form a sensor feature vector of a fixed length; performing hash coding on a unique identity identifier and converting the coding result into a numerical feature vector of a fixed length; and combining the image feature vector, the sensor feature vector, and the unique identity identifier feature vector into a comprehensive feature vector through a splicing operation.

10. The fiber raw material traceability system based on big data according to claim 1, characterized in that: The y' i,k The predicted probability that sample i belongs to category k includes: inputting the comprehensive feature vector into the trained convolutional neural network; extracting features from the comprehensive feature vector through the convolution layer and the pooling layer in turn, and flattening it to form a one-dimensional feature vector; inputting the flattened one-dimensional feature vector into the fully connected layer, and calculating the unnormalized score z for each category k i,k ; Use the Softmax function to convert the unnormalized score into a predicted probability. The calculation formula is as follows: Among them, j is a variable with a value range of [1, K]; z i,j is the unnormalized score of sample i belonging to category k.

Images