An artificial intelligence-based cross-border overseas warehouse cargo monitoring method

Abstract

This invention relates to an artificial intelligence-based method for monitoring goods in cross-border overseas warehouses. The method first collects IoT sensor data of the goods using IoT devices deployed at the overseas warehouse, and collects and identifies image data to obtain identification data. Based on the IoT sensor data and identification data, multi-dimensional cross-validation of goods information is performed on customs clearance data. A goods inventory monitoring model is established, extracting inventory data in real time from the distributed ledger. Based on the inventory data, a preset evaluation algorithm is used to calculate the stability index of the current inventory. When the stability index deviates from a preset safety range, an anomaly report is generated and output to the domestic terminal. Compared with existing technologies, this invention has advantages such as improved automation capabilities and timely risk warnings in cross-border logistics.

Description

Technical Field

[0001] This invention relates to the field of cargo monitoring technology, and in particular to an artificial intelligence-based method for monitoring cargo in cross-border overseas warehouses. Background Technology

[0002] With the rapid development of global cross-border e-commerce and supply chain finance, overseas warehouses, as crucial nodes in cross-border logistics, have seen a rapid expansion in scale and number. Warehouse receipt pledging refers to a method where companies use goods stored in their warehouses as collateral to apply for financing from financial institutions. Along with the booming development of cross-border e-commerce, warehouse receipt pledging, as an important financing tool, is gradually expanding its application from domestic warehousing to overseas warehousing, and is expected to become a key tool for solving cross-border trade financing challenges.

[0003] However, overseas warehouse receipt pledging presents several challenges: First, its cross-border nature. Overseas warehouse receipt pledging involves cross-border goods storage and financing. The goods are located in overseas warehouses, while the financing entity is typically a domestic cross-border e-commerce company or trading company. This model requires simultaneous compliance with the laws and regulations of both Chinese financial institutions and the country where the overseas warehouse is located. Second, multi-category mixed storage. Overseas warehouses store a wide variety of goods, including cross-border e-commerce retail items, bulk trade commodities, and customized production materials. Different categories of goods vary significantly in value, specifications, and storage conditions, increasing the complexity of managing the pledged goods. Third, the difficulty of tracing and confirming ownership. Tracing ownership of cross-border goods involves multiple legal systems, making the confirmation process complex. Financial institutions need to ensure clear ownership of the pledged goods to avoid legal disputes. Fourth, cross-border data transmission and compliance. Monitoring data for goods in overseas warehouses needs to be synchronized in real time between domestic financial institutions and overseas warehouses.

[0004] The existing field of cross-border overseas warehouse cargo identification and monitoring has the following shortcomings: 1. Insufficient accuracy in overseas warehouse cargo recognition, particularly when handling irregularly shaped goods and high-density small-package goods. In terms of cargo categories, cross-border overseas warehouses store a wide variety of goods, covering multiple fields such as electronics, apparel, food and beverages, and household goods. Different cargo categories vary significantly in shape, material, and packaging, placing extremely high demands on recognition algorithms. Taking video recognition as an example, regular goods, such as electronics, are relatively easy to identify due to their regular shape; however, irregular goods, such as apparel, with their varied shapes and materials, are much more difficult to identify. Especially for the recognition of small-package goods, cross-border trade often uses small packaging to meet personalized needs, leading to a surge in demand for cargo recognition. Existing recognition technologies are prone to errors or missed readings in high-density, small-particle packaging scenarios, making it difficult to meet the needs of rapid inbound and outbound operations.

[0005] 2. Barriers exist in the integration and interaction of cross-border cargo data, and an effective multi-source data cross-verification mechanism is lacking. Currently, the level of digitalization in overseas warehousing is generally high, with basic implementations of barcode scanning, RFID (Radio Frequency Identification) tag identification, AGV (Automated Guided Vehicle) automatic positioning, and automated storage and retrieval systems. However, the lack of effective collaboration between these functional modules leads to obstacles in data integration and interaction, creating data silos and making it difficult to comprehensively grasp information such as the flow and storage status of goods within the warehouse. Furthermore, there is poor coordination between customs declaration data (such as category and quantity) and data from the warehouse's internal management system (such as the time and quantity of goods entering the warehouse). Users find it difficult to verify the authenticity of goods and changes in their storage status in overseas warehouses through dynamic comparison of multi-source data. Due to data inconsistencies and the lack of cross-verification mechanisms, users struggle to track changes in cargo status in a timely manner, easily leading to problems such as inflated / depleted inventory, misdelivered or lost goods, significantly increasing management difficulty and operational risks.

[0006] 3. Shortcomings in data security and architecture design of overseas warehouse systems. Currently, some cross-border overseas warehouse systems have significant deficiencies in data security and architecture design: First, access control is lax, lacking fine-grained control, allowing internal personnel to freely access and modify core data, posing a risk of privilege abuse; second, disaster recovery capabilities are insufficient, lacking a robust data backup and recovery mechanism, making it difficult to quickly restore business operations in the event of hardware failure or cyberattacks, posing a single point of failure risk; finally, encryption and anti-tampering mechanisms are weak, with simple data encryption algorithms and a lack of advanced anti-tampering technologies such as blockchain, leaving data vulnerable to tampering during transmission and storage.

[0007] For example, the invention patent with publication number CN119831298A discloses a smart customs management method and system for digital ports based on artificial intelligence. This solution collects multi-source heterogeneous data (such as container positioning and customs clearance documents), uses federated learning to predict risks, and uses blockchain to verify the consistency of documents. However, this existing technology has significant drawbacks when applied to overseas warehouse receipt pledging scenarios: First, its monitoring granularity remains at the macro level, such as "containers" and "port customs clearance documents," without addressing the microscopic visual feature identification of irregular, high-density small-packaged goods within overseas warehouses, thus failing to solve the problems of missed inspections and misdeliveries at the physical warehouse level; second, its consistency verification is limited to a superficial comparison of documents and logistics documents, lacking multi-dimensional verification of customs clearance data and physical sensor data within the warehouse, resulting in the inability to confirm the absolute binding between physical goods and digital identities, posing a risk of malicious substitution of physical goods; finally, the solution does not construct a dynamic inventory health assessment system for the financial attributes of "warehouse receipt pledging," and does not provide a robust end-to-end security protection mechanism for cross-border data transmission, making it difficult to meet the needs of financial institutions for monitoring the stability of pledged value and ensuring compliant cross-border data transmission.

[0008] In summary, the current identification and monitoring of goods in cross-border overseas warehouses faces challenges such as the difficulty in accurately identifying complex and diverse goods and the lack of rigorous logical verification of data. Furthermore, the automated processing capabilities and timely risk warnings of cross-border logistics need to be improved. Summary of the Invention

[0009] The purpose of this invention is to overcome the shortcomings of the existing technology and provide an artificial intelligence-based method for monitoring goods in cross-border overseas warehouses.

[0010] The objective of this invention can be achieved through the following technical solutions: According to one aspect of the present invention, an artificial intelligence-based method for monitoring goods in cross-border overseas warehouses is provided, the method comprising the following steps: IoT sensor data of goods is collected by IoT devices deployed in overseas warehouses, and image data of goods is obtained through the collection devices; the image data is processed by a deep learning-based target detection algorithm to identify the outer packaging features and text labels of the goods, and thus obtain identification data. The system acquires customs clearance data for goods and performs multi-dimensional cross-validation of goods information based on IoT sensor data and identification data to determine whether the goods verification status is normal or abnormal. If the verification status is normal, the system updates the corresponding inventory data of the goods to the distributed ledger using a hierarchical collaborative distributed processing logic. If the verification status is abnormal, the system generates an interception command to block the goods warehousing process and outputs an alarm signal containing the abnormality type to the management terminal. Establish an inventory monitoring model to extract inventory data from the distributed ledger in real time. Based on the inventory data, use a preset evaluation algorithm to calculate the stability index of the current inventory. When the stability index deviates from the preset safe range, generate an anomaly report and output the report to the domestic terminal. Performing multi-dimensional cross-validation of cargo information on customs clearance data includes a spatiotemporal logic verification process and a triple verification process for warehousing. If both the spatiotemporal logic check and the triple check for warehousing pass, the goods verification status is determined to be normal; otherwise, the goods verification status is determined to be abnormal.

[0011] As a preferred technical solution, the IoT devices include radio frequency identification (RFID) readers and writers deployed at the entrances and exits of overseas warehouses, AGVs (Automated Guided Vehicles) inside overseas warehouses, and photoelectric sensors arranged within overseas warehouses; the IoT sensing data of the goods includes: RFID tag information of the goods, timestamps of goods entering and leaving the warehouse, goods handling information, and trigger information for the status of goods entering and leaving the warehouse. The specific process of collecting IoT sensor data of goods includes: Using an RFID reader, the electronic tags attached to the goods are read, and the RFID tag information and the timestamps of the goods entering and leaving the warehouse are obtained. Using AGVs, real-time operation data is collected during the handling of tasks to obtain cargo handling information, including the starting position, the target position, and the handling time. Using photoelectric sensors, the system detects obstruction or presence signals generated when goods pass through a preset position, converts the signals into electrical signals, and outputs them to obtain the trigger information for the goods entering or leaving the warehouse.

[0012] As a preferred technical solution, the acquisition device is a video acquisition device, and the image data of the goods is in video stream format; the specific steps for processing the image data include: The image data is preprocessed for denoising and image stabilization; and image frames are extracted according to a preset frame rate or time interval. The image frame is input into a pre-set YOLOv5 (You Only Look Once version 5, a single-stage object detection algorithm) object detection model to identify and parse the cargo area, the text on the outer packaging of the cargo, and the barcode information to obtain the parsing results; and based on the parsing results, the cargo location and category name are marked in the image using rectangular or polygonal boxes as recognition data. The identified data is manually corrected to obtain manually corrected data, which is then fed back into the target detection model for iterative training.

[0013] As a preferred technical solution, when identifying and parsing the cargo area, the text on the outer packaging of the cargo, and the barcode information: For goods that are not rigid or have irregular contours, a DCN (Deformable Convolutional Networks) layer is configured in the feature extraction layer of the YOLOv5 model. By learning the offset of the input feature map, the sampling points of the convolution kernel undergo adaptive deformation to match the geometric deformation features of the goods' edges. For multiple goods with overlapping packaging areas, an attention mechanism module is embedded at the P4 and P3 feature fusion nodes in the FPN (Feature Pyramid Network) path of the YOLOv5 object detection model. In the prediction output stage of the YOLOv5 object detection model, the non-maximum suppression (NMS) algorithm based on center distance is used to replace the conventional non-maximum suppression algorithm. The non-maximum suppression algorithm based on center distance introduces a center distance penalty term when calculating the overlap of candidate boxes. When the intersection-union ratio (IU) of two candidate boxes is higher than the preset IU threshold but the center distance is greater than the preset center distance judgment value, the two candidate boxes are retained as different independent goods targets to distinguish multiple independent units with overlapping spatial locations.

[0014] As a preferred technical solution, the pre-configured YOLOv5 object detection model is a pre-trained YOLOv5 object detection model, and its pre-training process includes: Construct an initial sample dataset including regular cargo, irregular soft-packaged cargo, and tightly stacked individually packaged cargo; The cargo targets in the initial sample dataset are labeled to generate a training set including cargo category labels and location coordinate boxes. The YOLOv5 neural network is iteratively trained using the training set to obtain the basic target detection model. Acquire false positive or false negative image frames generated by the basic target detection model in actual operation, and receive manually corrected labeling data for false positive or false negative image frames; Manually corrected labeled data is fed back as incremental samples to the basic object detection model for retraining, resulting in a pre-set YOLOv5 object detection model.

[0015] As a preferred technical solution, the spatiotemporal logic verification process specifically includes: Extract cargo declaration information and trade document information from customs clearance data, and extract cargo RFID tag information, cargo entry and exit timestamps, cargo handling information and cargo entry and exit status trigger information from IoT sensor data; The goods are associated with the declaration information and RFID tag information. The trade document information of each associated goods is aligned with the timestamps of goods entering and leaving the warehouse. The preset circulation path corresponding to the declaration information is compared with the goods handling information and the trigger information of goods entering and leaving the warehouse. If the deviation of the time axis and the deviation of the spatial trajectory are both within the preset threshold range, the spatiotemporal logic verification is deemed to have passed; otherwise, the spatiotemporal logic verification is deemed to have failed.

[0016] As a preferred technical solution, the triple verification process for warehousing specifically includes: First verification: The text and barcode information on the outer packaging of the goods in the identification data are compared with the goods declaration information in the customs clearance data. The first verification is deemed to have passed only if the comparison results show that the features are consistent. The barcode information includes barcode information and QR code information. The second verification is to compare the cargo location information in the identification data with the cargo entry and exit status trigger information in the IoT sensor data to verify the consistency of the cargo's physical location; the second verification is deemed to have passed only if the verification results show that the two are consistent in spatiotemporal logic. The third verification step involves obtaining the cargo serial number from the identification data and comparing it with the cargo RFID tag information to see if their hash values ​​match. The third verification step is considered successful if and only if the hash values ​​match. If the first to third checks all pass, the three-stage entry check is considered successful; otherwise, the three-stage entry check is considered unsuccessful.

[0017] As a preferred technical solution, the specific steps for updating the distributed ledger using a layered and collaborative distributed processing logic include: The verified inventory data is submitted as incremental data to the regional endorsement node of the blockchain network, and the regional endorsement node generates a temporary ledger fragment and returns it to the terminal for verification. Incremental data will only be submitted to the sorting service cluster once the terminal verification is successful. The sorting service cluster performs topological sorting on the incremental data according to the time sequence of the goods operations, generates a block containing multi-dimensional verification anchors, and synchronously writes the block to the accounting nodes of the distributed ledger.

[0018] As a preferred technical solution, the specific process of calculating the stability index of the current inventory using a preset evaluation algorithm includes: The system retrieves inventory data from the distributed ledger and analyzes the inventory data to obtain real-time physical quantity data for each item. Query the pre-set cargo attribute database to obtain the dynamic value weight coefficients corresponding to various types of cargo. Perform a weighted calculation on the real-time physical quantity data of cargo and the dynamic value weight coefficients to generate the current inventory value data. Read the system's preset pledge benchmark value, calculate the ratio between the inventory value data and the pledge benchmark value, and define this ratio as the inventory coverage factor; Acquire data on the frequency of goods entering and leaving the warehouse within a preset monitoring period, and generate inventory turnover coefficient through time series analysis; A multidimensional state vector is constructed using the inventory coverage coefficient and inventory turnover coefficient as a stability indicator representing the current inventory status of goods in overseas warehouses.

[0019] As a preferred technical solution, the status anomaly report includes inventory data and stability indicators; when outputting this report to a domestic endpoint, an encryption process is included, specifically: The AES (Advanced Encryption Standard) symmetric encryption algorithm is used to preprocess the status anomaly report, converting plaintext into ciphertext, and then transmitting the ciphertext to the domestic end through a secure channel; During transmission, the TLS (Transport Layer Security) protocol is introduced, and after key exchange is completed through an asymmetric encryption algorithm, a symmetric encryption algorithm is used to protect the data end-to-end, and the data integrity is verified by combining a message authentication code. The domestic end decrypts the ciphertext using the key and obtains a status anomaly report.

[0020] Compared with the prior art, the present invention has the following beneficial effects: 1. In this invention, by integrating IoT sensor data and image recognition technology, and combining customs clearance data for multi-dimensional cross-verification, intelligent management of the entire chain from goods warehousing to inventory monitoring is realized. This enables automatic identification and blocking of abnormal goods from entering the warehouse, and utilizes a hierarchical collaborative distributed ledger to ensure real-time synchronization and stability monitoring of inventory data, significantly improving the automated handling capability and risk warning timeliness of cross-border logistics.

[0021] 2. In this invention, by integrating RFID reading and writing, AGV operation data and photoelectric sensor signals, a multimodal Internet of Things sensing system covering identity, time, location and dynamic behavior is constructed, which enables the real-time capture of the dynamic handling trajectory of goods and the trigger status of entering and leaving the warehouse, providing detailed and multi-dimensional underlying physical data support for subsequent logical verification, and ensuring a high degree of synchronization between the physical world and digital information.

[0022] 3. In this invention, video stream preprocessing is combined with the YOLOv5 deep learning model, and an iterative training mechanism with manual correction feedback is introduced to achieve efficient recognition of cargo packaging and text labels. Through the closed loop of manual intervention and algorithm feedback, the system can continuously optimize the accuracy of target detection and maintain high robustness in cargo feature recognition even in complex overseas warehouse environments.

[0023] By introducing DCN adaptive convolutional layers and attention mechanisms into the model, and optimizing the non-maximum suppression algorithm, efficient recognition of irregularly shaped goods and overlapping small particle packaging is achieved. For goods with diverse shapes or tightly stacked items such as electronic products and clothing, it can accurately distinguish independent units, effectively solving the problems of false detection and missed detection in complex goods under overlapping and deformation conditions.

[0024] By constructing a full-sample dataset containing regular, irregular, and tightly stacked goods and implementing incremental learning, the model is ensured to be widely adaptable to various commodities in cross-border trade. This enables the method to accurately identify everything from large equipment to various small and fragmented goods, significantly improving the recognition accuracy in scenarios with irregular shapes and diverse materials.

[0025] 4. This invention utilizes the association between cargo declaration information and RFID information, and aligns and compares trade documents with measured timelines and spatial trajectories to achieve closed-loop verification of cargo flow logic. This effectively identifies false declarations or abnormal logistics paths, ensuring the authenticity and compliance of the cargo flow process through spatiotemporal matching. Furthermore, this invention employs a triple verification system—image and text feature comparison, physical location consistency verification, and serial number hash value verification—which, through multi-source data fusion and cross-validation mechanisms, greatly enhances the rigor of identity verification. This feature ensures that every piece of inbound cargo completely matches the customs declaration information in terms of content, location, and digital identity, eliminating the risk of misdelivery, omission, or cargo swapping from the source.

[0026] 5. In this invention, a layered collaborative distributed processing logic is adopted. The distributed ledger is updated through regional endorsement nodes, sorting services and consensus mechanisms. Blockchain technology is introduced to realize the decentralized storage and transparent flow of goods data, ensuring the temporal consistency of inventory information during the update process, and guaranteeing the immutability and traceability of data from a technical perspective.

[0027] 6. This invention utilizes dynamic value weights, inventory coverage coefficients, and turnover rate coefficients to construct stability indicators, achieving a shift from quantity monitoring to value and liquidity assessment. This algorithm can analyze the inventory stability of overseas warehouses in real time, automatically sending warnings to the domestic end when the indicators deviate from the safe range, providing scientific and intelligent data support for the supply chain decisions of cross-border enterprises.

[0028] 7. This invention constructs an end-to-end data protection barrier by combining AES symmetric encryption, TLS security protocol, and asymmetric key exchange, and implementing message authentication code verification at the decryption end. This encryption mechanism ensures the confidentiality, integrity, and non-repudiation of sensitive data such as status anomaly reports during cross-border transmission. Combined with strict access control, it fundamentally protects the security of corporate trade secrets and cross-border trade data. Attached Figure Description

[0029] Figure 1 This is a schematic diagram of the cross-border overseas warehouse cargo monitoring method based on artificial intelligence in this invention; Figure 2 This is a schematic diagram of the AI-based cross-border overseas warehouse cargo identification and monitoring process in the example. Figure 3 This is a schematic diagram illustrating the cross-validation and updating of cargo information in the example; Figure 4 This is a schematic diagram illustrating the establishment and monitoring process of the goods inventory monitoring model in this embodiment. Detailed Implementation

[0030] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort should fall within the scope of protection of the present invention.

[0031] Overseas warehouse receipt pledging, as a new cross-border trade financing model, provides crucial financial support for cross-border e-commerce enterprises. However, it also places higher demands on cargo supervision capabilities, involving aspects such as "cargo ownership confirmation and traceability," "cargo identification and monitoring," "data security and cross-border transmission," "risk control and cargo preservation," and "compliance with multiple regulations." Therefore, the successful operation of the overseas warehouse receipt pledging model relies on efficient and accurate supervision of the pledged goods. By introducing advanced technologies such as the Internet of Things, blockchain, and distributed storage, a cross-border cargo identification and monitoring system can be built, effectively solving the regulatory challenges in overseas warehouse receipt pledging and providing a reliable technological foundation for the financialization of cross-border trade.

[0032] This solution provides a comprehensive and innovative method for monitoring and building a network for cross-border overseas warehouse cargo information. Extensive data integration and collection are fundamental. Through overseas customs clearance data collection, real-time sensing by IoT devices, and video image recognition, it accurately captures basic cargo information and dynamic location. Cross-verification of cargo information is crucial to ensuring its authenticity. A three-dimensional evidence chain of customs clearance data, IoT data, and video recognition data, combined with distributed ledger technology, achieves spatiotemporal consistency verification of multi-source data. The establishment of a cargo inventory monitoring model and verification of collateral value are key to ensuring asset liquidity. Real-time monitoring of dynamic inventory changes ensures stable inventory value. Data encryption and cross-border transmission are critical aspects of data security. End-to-end encrypted transmission is achieved through AES and TLS protocols, combined with blockchain evidence storage and local backup strategies to ensure data security.

[0033] It should be stated that the customs clearance data involved in this plan is data to be processed obtained from legally public data interfaces or authorized data, and these data themselves have legal source attributes.

[0034] Furthermore, the cross-border data transfer method described in this application targets data packets that have met the data export security assessment conditions stipulated by relevant laws and regulations or have completed the statutory cross-border transfer compliance procedures. In other words, this transfer method is applied to legally cross-border transferable data, and this solution does not include any steps for circumventing regulatory oversight in the transfer of non-compliant data.

[0035] Example In this embodiment, an AI-based method for monitoring cross-border overseas warehouse goods is adopted. This method provides AI-based identification and monitoring of cross-border overseas warehouse goods. By utilizing AI, IoT, blockchain, and encryption technologies, and through data cross-verification, real-time monitoring, and blockchain evidence storage, the authenticity of cross-border goods, the security and immutability of data are ensured. This further proves that the goods are indeed in the overseas warehouse and that no loss or omission has occurred, thus providing strong support for overseas warehouse receipt pledging.

[0036] First, based on IoT devices and video imaging equipment, information on goods in cross-border overseas warehouses is collected from multiple dimensions, covering a wide range of products.

[0037] Secondly, relying on local customs clearance data, and integrating IoT sensor and identification data, a trusted chain of evidence is constructed across the entire supply chain through spatiotemporal logic verification and a triple verification process for warehousing, and this evidence is stored locally. Overseas warehouse data updates employ a layered, collaborative, distributed processing logic, and are synchronously written to the distributed ledger after being orchestrated by a sorting service cluster and verified by accounting nodes.

[0038] Then, an inventory monitoring model is established to extract inventory data from the distributed ledger in real time. Based on the inventory data, a preset evaluation algorithm is used to calculate the stability index of the current inventory. When the stability index deviates from the preset safe range, an abnormal status report is generated and the report is output to the domestic end to ensure the security of the pledge.

[0039] Finally, a security system for local data in overseas warehouses is constructed. The AES symmetric encryption algorithm is used to preprocess the aforementioned anomaly reports, and key exchange is completed through a combination of TLS protocol and asymmetric encryption to securely transmit data to the domestic terminal.

[0040] To achieve the above goals, an AI-based method for monitoring cross-border overseas warehouse goods is adopted. The method's process is as follows: Figure 1 As shown, it specifically includes: IoT sensor data of goods is collected by IoT devices deployed in overseas warehouses, and image data of goods is obtained through the collection devices; the image data is processed by a deep learning-based target detection algorithm to identify the outer packaging features and text labels of the goods, and thus obtain identification data. The system acquires customs clearance data for goods and performs multi-dimensional cross-validation of goods information based on IoT sensor data and identification data to determine whether the goods verification status is normal or abnormal. If the verification status is normal, the system updates the corresponding inventory data of the goods to the distributed ledger using a hierarchical collaborative distributed processing logic. If the verification status is abnormal, the system generates an interception command to block the goods warehousing process and outputs an alarm signal containing the abnormality type to the management terminal. Establish an inventory monitoring model to extract inventory data from the distributed ledger in real time. Based on the inventory data, use a preset evaluation algorithm to calculate the stability index of the current inventory. When the stability index deviates from the preset safe range, generate an anomaly report and output the report to the domestic terminal. Performing multi-dimensional cross-validation of cargo information on customs clearance data includes a spatiotemporal logic verification process and a triple verification process for warehousing. If both the spatiotemporal logic check and the triple check for warehousing pass, the goods verification status is determined to be normal; otherwise, the goods verification status is determined to be abnormal.

[0041] The specific implementation process of this method is as follows: Figure 2 As shown, it includes: Extensive data integration and collection. The system uses IoT devices, including RFID readers, AGVs, and photoelectric sensors, to collect multi-dimensional data on goods in cross-border overseas warehouses in real time; video imaging equipment uses image recognition algorithms to analyze barcodes, QR codes, and text labels on outer packaging, covering a wide range of products from electronics to apparel.

[0042] A multi-dimensional cross-verification mechanism for cargo information is implemented. First, customs clearance data and IoT sensor data undergo spatiotemporal logical verification to ensure consistency between the cargo movement trajectory and the declared information. Second, a video image system analyzes the outer packaging and customs clearance documents, compares the textual and image information, and combines this with triple verification using RFID and photoelectric sensors to ensure the accuracy of the information, location, and identity of the goods entering the warehouse. Simultaneously, the raw data is stored locally according to compliance requirements. Next, overseas warehouse data updates employ a layered collaborative mechanism. Terminal devices submit verified incremental data to regional endorsement nodes to generate a temporary ledger. After being sorted by the sorting service cluster according to the operation sequence, it is written into the distributed ledger.

[0043] The establishment of an inventory monitoring model and verification of collateral value are as follows: First, a data monitoring system is established to collect data on inventory categories, quantities, turnover, and market conditions, and to set core indicators such as the collateral ratio. Second, an intelligent model is built to track changes in inventory value in real time, and intelligent algorithms are used to calculate the ratio of inventory value to collateral value to maintain it within a safe range. An automatic warning is issued when the inventory value falls below the set collateral ratio.

[0044] Data Encryption and Cross-Border Transmission. To ensure the security of local data in overseas warehouses, a user authorization mechanism and robust data backup and recovery mechanisms are required. This solution employs the AES symmetric encryption algorithm to preprocess information such as whether goods are stored in overseas warehouses and whether the value of inventory meets the pledge value, converting plaintext into ciphertext and transmitting it to the domestic headquarters through a secure channel. Simultaneously, the TLS protocol is introduced in conjunction with asymmetric encryption, such as RSA (Rivest-Shamir-Adleman, an asymmetric encryption algorithm), to complete key exchange.

[0045] Step 1: Broadly connect with and collect data to ensure data integrity; 11) Collect external data for overseas warehouses, such as overseas customs clearance data, trade compliance documents, and cargo transportation data. First, connect with customs systems, electronic port platforms, and logistics service provider systems of various countries through APIs (Application Programming Interfaces) to obtain basic information about goods, including commodity name, specifications, quantity, and value. Second, obtain compliance document data, such as import and export licenses and inspection and quarantine certificates. In addition, dynamic logistics data is also crucial, including transport documents (such as bills of lading and waybill numbers), carrier information, and cargo transportation routes.

[0046] 12) Real-time collection of cargo flow information within overseas warehouses using IoT devices. Fixed RFID readers are installed at the warehouse entrances and exits. When goods enter or leave the warehouse, the readers automatically read the RFID tag information on the goods, enabling rapid identification and data collection. AGVs (Automated Guided Vehicles) upload data such as the starting position, destination position, and handling time of the goods to the warehouse management system, enabling real-time monitoring and data recording of the cargo flow process. Furthermore, photoelectric sensors are installed at key locations such as shelves and conveyor belts in the overseas warehouse. When goods pass through these locations, the sensors detect their presence and convert the detection signal into an electrical signal, which is then sent to the control system. The control system determines the cargo entry and exit actions based on the sensor feedback.

[0047] 13) Acquire image data (e.g., video streams) of goods using acquisition devices. Deploy video acquisition devices at key nodes in the warehouse. First, preprocess the acquired image data by denoising and stabilizing, and extract a series of image frames according to a preset frame rate or time interval. Second, use the YOLOv5 deep learning-based object detection algorithm to detect objects in the extracted image frames, identifying the goods regions in the image. For irregular goods, the model needs to be able to accurately detect their irregular shapes and boundaries; for high-density small-package goods, it needs to be able to distinguish the closely packed small packages. During the detection process, an appropriate confidence threshold can be set according to the type of goods to filter out low-confidence detection results.

[0048] The image frame is input into the preset YOLOv5 object detection model to identify and parse the cargo area, the text on the outer packaging of the cargo, and the barcode information to obtain the identification data.

[0049] In terms of model processing details: For non-rigid or irregularly shaped goods with irregular contours, a DCN layer is configured in the feature extraction layer of the YOLOv5 model to adaptively deform the sampling points of the convolution kernel by learning the offset of the input feature map; for multiple goods with overlapping packaging areas, an attention mechanism module is embedded at the P4 and P3 feature fusion nodes in the FPN path of the model's feature fusion network; and in the prediction output stage, a non-maximum suppression algorithm based on center point distance is adopted, introducing a center point distance penalty term to distinguish multiple independent units with overlapping spatial positions.

[0050] Subsequently, the detected cargo areas are marked as identification data. Professional personnel review the identification data to obtain manually corrected data, which is then used as incremental samples to feed into the base object detection model for iterative training, updating the pre-built YOLOv5 object detection model. This enables accurate identification of barcodes, QR codes, and text labels on cargo packaging from video images, covering various commodities from electronic products to clothing.

[0051] Step 2: Cross-verify cargo information to ensure data authenticity; 21) Perform a spatiotemporal logic verification process. Extract cargo declaration information and trade document information from customs clearance data, and extract cargo RFID tag information, cargo entry / exit timestamps, cargo handling information, and cargo entry / exit status trigger information from IoT sensor data. Associate cargo based on the cargo declaration information and cargo RFID tag information, and align the trade document information of each associated cargo with the cargo entry / exit timestamps on the timeline. Compare the spatial trajectory of the preset flow path corresponding to the cargo declaration information with the cargo handling information and cargo entry / exit status trigger information. If both the timeline deviation and the spatial trajectory deviation are within the preset threshold range, the spatiotemporal logic verification is considered successful, verifying the logical consistency between the cargo's physical movement trajectory and the declaration information. Through this spatiotemporal logic verification process, ensure that the location and time information of the cargo during transportation are highly consistent with the declaration content, thereby proving that the storage status and logistics trajectory of the cargo in the overseas warehouse meet expectations.

[0052] 22) A three-tiered verification process is executed upon goods entering the warehouse. The first verification involves comparing the text and barcode information on the outer packaging of the goods in the identification data with the goods declaration information in the customs clearance data. If the text and image information identified by the video recognition matches the customs clearance document, the goods' entry information is consistent with the customs declaration, and the verification passes. The second verification compares the goods' location information in the identification data with the goods' entry / exit status trigger information in the IoT sensor data to verify the consistency of the goods' physical location. If the spatiotemporal logic matches, the verification passes. The third verification obtains the goods' serial number from the identification data and compares it with the goods' RFID tag information, checking if their hash values ​​match. If they match, the verification passes. If all the above verifications pass, the goods' verification status is determined to be normal. Thus, the multi-dimensional cross-verification mechanism of customs clearance data, IoT sensor data, and video recognition data constructs a complete and credible chain of evidence. In the overseas warehousing stage, based on data export compliance requirements, the monitoring video stream data and original sensor data will be fully retained on overseas local servers.

[0053] 23) The overseas warehouse data update process follows a layered, collaborative, distributed processing logic. First, each data source terminal (customs declaration system client, RFID reader / writer, AGV terminal, video node) submits the verified incremental data to the regional endorsement node. The regional endorsement node simulates the update to generate a temporary ledger fragment, encrypts it, and returns it to the terminal for secondary verification. After successful verification, the terminal submits it to the sorting service cluster, which performs topological sorting according to the cargo operation sequence (arrival time → warehousing time → shelving), forming blocks containing multi-dimensional verification anchors. Verified blocks are written to the distributed ledger and updated synchronously, ensuring data traceability throughout its entire lifecycle. Cross-validation and updating of cargo information are as follows: Figure 3 As shown.

[0054] Step 3: Establishing an inventory monitoring model and verifying collateral value to ensure asset liquidity; 31) Data Collection and Indicator Setting. First, connect to the overseas warehouse management system to collect inventory data from overseas warehouses, including information such as product types, inventory quantities, and inbound / outbound frequencies. Simultaneously, based on the pledged amount, set key indicators such as the ratio of inventory value to pledged value (target ratio), safety stock level, and inventory turnover rate.

[0055] The system retrieves inventory data from the distributed ledger, analyzes the inventory data to obtain real-time physical quantity data for each type of goods, and queries a pre-set goods attribute database to obtain dynamic value weight coefficients corresponding to each type of goods.

[0056] 32) Model Construction and Dynamic Monitoring. Real-time physical quantity data of goods is weighted and calculated with dynamic value weighting coefficients to generate the current inventory value data. The system's preset collateral benchmark value is read, and the ratio between the inventory value data and the collateral benchmark value is calculated, defined as the inventory coverage coefficient. Simultaneously, the frequency data of goods entering and leaving the warehouse within a preset monitoring period is acquired to generate the inventory turnover rate coefficient. Finally, a multi-dimensional state vector is constructed using the inventory coverage coefficient and the inventory turnover rate coefficient, serving as a stability indicator representing the current inventory status of goods in the overseas warehouse. When the stability indicator deviates from the preset safe range, the model automatically generates an anomaly report.

[0057] By collecting and analyzing inventory turnover data in real time, and combining this with intelligent algorithms to dynamically calculate the ratio of inventory value to pledged value, an early warning mechanism is constructed to ensure that the ratio remains within a preset safe fluctuation range. When the inventory value falls below a set proportion of the pledged value due to factors such as sales or market fluctuations, the model automatically triggers an early warning, thereby ensuring that the inventory value meets the pledged value. The inventory monitoring model is as follows: Figure 4 As shown.

[0058] Step 4: Data encryption and cross-border transmission to ensure data security; A secure system for local data in overseas warehouses is established. To ensure the security and controllability of external data, internal cargo flow information, and product image data in overseas warehouses, the system includes an embedded user authorization mechanism, allowing users to choose whether to grant data access permissions to regulatory authorities in cross-border business scenarios. When a user authorizes access, regulatory authorities can access cloud data by decrypting the encrypted metadata (such as storage address and symmetric key) stored in the blockchain, achieving transparent supervision. Simultaneously, the system has a robust data backup and recovery mechanism. Data backups are performed regularly, employing a multi-copy storage strategy to back up data to servers or cloud storage in different geographical locations. Regarding data recovery, the system conducts regular recovery drills according to a recovery plan to ensure rapid and complete data recovery, effectively enhancing data resilience and forming a security protection system covering the entire data lifecycle.

[0059] 42) The technical solution for cross-border data compliance comprehensively utilizes encryption technology and secure transmission protocols. First, the generated status anomaly reports (including inventory data and stability indicators) are preprocessed using the AES symmetric encryption algorithm, converting plaintext into ciphertext, and then transmitted to the domestic end via a secure channel. During transmission, the TLS protocol is introduced, and after key exchange via an asymmetric encryption algorithm, a symmetric encryption algorithm is used for end-to-end data protection. Combined with a Message Authentication Code (MAC) to verify data integrity, this prevents theft or tampering during transmission. Upon receiving the data, the domestic end decrypts the ciphertext using the key to obtain the final status anomaly report.

[0060] In summary, this method utilizes deep learning and artificial intelligence technologies to achieve efficient identification and monitoring of complex goods and small-particle packaging. Employing advanced image recognition algorithms and multimodal data fusion technology, it can accurately identify various commodities, from electronic products to clothing, and even distinguish different styles and sizes when the shapes are irregular and the materials are diverse.

[0061] Secondly, advanced data verification technology is employed, utilizing multi-source data fusion and cross-validation mechanisms to ensure the authenticity and accuracy of cargo information. Customs clearance data and warehouse management system data are integrated, achieving seamless connection and effective comparison between different data sources through a unified data format and standardized processes.

[0062] Finally, advanced encryption algorithms and strict access control mechanisms are employed to ensure the integrity and confidentiality of data during storage and transmission. Access and modification operations by internal personnel are strictly restricted and recorded. The system incorporates blockchain technology to achieve data immutability and traceability, fundamentally guaranteeing long-term data integrity. Simultaneously, a robust data backup and recovery mechanism further enhances system reliability, enabling rapid data recovery to ensure business continuity even in the event of hardware failure or data loss.

[0063] The above description is merely a specific embodiment of the present invention, but the scope of protection of the present invention is not limited thereto. Any person skilled in the art can easily conceive of various equivalent modifications or substitutions within the technical scope disclosed in the present invention, and these modifications or substitutions should all be covered within the scope of protection of the present invention. Therefore, the scope of protection of the present invention should be determined by the scope of the claims.

Claims

1. An artificial intelligence-based cross-border overseas warehouse cargo monitoring method, characterized in that, The method steps include: IoT sensor data of goods is collected by IoT devices deployed in overseas warehouses, and image data of goods is obtained through the collection devices; the image data is processed by a deep learning-based target detection algorithm to identify the outer packaging features and text labels of the goods, and thus obtain identification data. The system acquires customs clearance data for goods and performs multi-dimensional cross-validation of goods information based on IoT sensor data and identification data to determine whether the goods verification status is normal or abnormal. If the verification status is normal, the system updates the corresponding inventory data of the goods to the distributed ledger using a hierarchical collaborative distributed processing logic. If the verification status is abnormal, the system generates an interception command to block the goods warehousing process and outputs an alarm signal containing the abnormality type to the management terminal. An inventory monitoring model is established to extract inventory data from the distributed ledger in real time. Based on the inventory data, a preset evaluation algorithm is used to calculate the stability index of the current inventory. When the stability index deviates from the preset safe range, an abnormal status report is generated and the report is output to the domestic terminal. The aforementioned multi-dimensional cross-validation of cargo information for customs clearance data includes a spatiotemporal logic verification process and a triple verification process for warehousing. If both the spatiotemporal logic check and the triple check for warehousing pass, the goods verification status is determined to be normal; otherwise, the goods verification status is determined to be abnormal.

2. The method of claim 1, wherein the method is based on artificial intelligence. The IoT devices include RFID readers deployed at the entrances and exits of overseas warehouses, AGVs inside overseas warehouses, and photoelectric sensors arranged inside overseas warehouses; the IoT sensing data of the goods includes: RFID tag information of goods, timestamps of goods entering and leaving the warehouse, goods handling information, and trigger information for goods entering and leaving the warehouse. The specific process of collecting IoT sensor data of goods includes: Using the aforementioned radio frequency identification (RFID) reader / writer, the electronic tags attached to the goods are read, and the RFID tag information and the timestamps of the goods entering and leaving the warehouse are obtained. Using the AGV, real-time operation data is collected during the execution of the handling task to obtain cargo handling information, which includes the starting position of handling, the target position of handling, and the handling operation time. The photoelectric sensor is used to detect the obstruction signal or presence signal generated when the goods pass through the preset position, and the signal is converted into an electrical signal for output to obtain the goods entry and exit status trigger information.

3. The method of claim 1, wherein the method further comprises: The acquisition device is a video acquisition device, and the image data of the goods is in video stream format; the specific steps for processing the image data include: The image data is preprocessed for denoising and image stabilization; and image frames are extracted according to a preset frame rate or time interval. The image frame is input into a preset YOLOv5 object detection model to identify and parse the cargo area, the cargo packaging text, and the barcode information to obtain the parsing results. Based on the parsing results, the cargo location and category name are marked in the image using rectangular or polygonal boxes as recognition data. The identified data is manually corrected to obtain manually corrected data, and this manually corrected data is fed back into the target detection model for iterative training.

4. The method of claim 3, wherein the method further comprises: When identifying and parsing the cargo area, the text on the outer packaging of the cargo, and the barcode information: For goods that are not rigid or have irregular contours, a DCN layer is configured in the feature extraction layer of the YOLOv5 model. By learning the offset of the input feature map, the sampling points of the convolution kernel undergo adaptive deformation to match the geometric deformation features of the goods' edges. For multiple goods with overlapping packaging areas, an attention mechanism module is embedded at the P4 and P3 feature fusion nodes in the FPN path of the YOLOv5 object detection model. In the prediction output stage of the YOLOv5 object detection model, a non-maximum suppression algorithm based on center point distance is used to replace the conventional non-maximum suppression algorithm. The non-maximum suppression algorithm based on center point distance introduces a center point distance penalty term when calculating the overlap of candidate boxes. When the intersection-union ratio (IU) of two candidate boxes is higher than a preset IU threshold but the center point distance is greater than a preset center point distance judgment value, the two candidate boxes are retained as different independent goods targets to distinguish multiple independent units with overlapping spatial locations.

5. The method of claim 3, wherein the method further comprises: The pre-set YOLOv5 object detection model is a pre-trained YOLOv5 object detection model, and its pre-training process includes: Construct an initial sample dataset including regular cargo, irregular soft-packaged cargo, and tightly stacked individually packaged cargo; The cargo targets in the initial sample dataset are labeled to generate a training set including cargo category labels and location coordinate boxes. The YOLOv5 neural network is iteratively trained using the training set to obtain a basic target detection model. Acquire false-detection or false-detection image frames generated by the basic target detection model in actual operation, and receive manually corrected labeling data for the false-detection or false-detection image frames; The manually corrected labeled data is fed back as incremental samples to the basic target detection model for retraining, thus obtaining the preset YOLOv5 target detection model.

6. The method for monitoring cross-border overseas warehouse goods based on artificial intelligence according to claim 1, characterized in that, The spatiotemporal logic verification process specifically includes: Extract cargo declaration information and trade document information from the customs clearance data, and extract cargo RFID tag information, cargo entry and exit timestamps, cargo handling information and cargo entry and exit status trigger information from the IoT sensor data; The goods are associated with the declaration information and RFID tag information. The trade document information of each associated goods is aligned with the timestamps of goods entering and leaving the warehouse. The preset circulation path corresponding to the declaration information is compared with the goods handling information and the trigger information of goods entering and leaving the warehouse. If the deviation of the time axis and the deviation of the spatial trajectory are both within the preset threshold range, the spatiotemporal logic verification is deemed to have passed; otherwise, the spatiotemporal logic verification is deemed to have failed.

7. The method for monitoring cross-border overseas warehouse goods based on artificial intelligence according to claim 1, characterized in that, The aforementioned triple verification process for warehousing specifically includes: First verification: The text and barcode information on the outer packaging of the goods in the identification data are compared with the goods declaration information in the customs clearance data; the first verification is deemed to have passed only if the comparison results show that the features are consistent; the barcode information includes barcode information and QR code information; The second verification is to compare the cargo location information in the identification data with the cargo entry and exit status trigger information in the IoT sensor data to verify the consistency of the cargo's physical location; the second verification is deemed to have passed only if the verification results show that the two are consistent in spatiotemporal logic. The third verification step involves obtaining the cargo serial number from the identification data and comparing it with the cargo RFID tag information to see if their hash values ​​match. The third verification step is considered successful if and only if the hash values ​​match. If the first to third checks all pass, the three-stage entry check is considered successful; otherwise, the three-stage entry check is considered unsuccessful.

8. The method for monitoring cross-border overseas warehouse goods based on artificial intelligence according to claim 1, characterized in that, The specific steps for updating the distributed ledger using the layered collaborative distributed processing logic include: The verified inventory data is submitted as incremental data to the regional endorsement node of the blockchain network, and the regional endorsement node generates a temporary ledger fragment and returns it to the terminal for verification. The incremental data is submitted to the sorting service cluster only when the terminal verification is successful; The sorting service cluster performs topological sorting on the incremental data according to the time sequence of the goods operations, generates a block containing multi-dimensional verification anchor points, and synchronously writes the block to the accounting node of the distributed ledger.

9. The method for monitoring cross-border overseas warehouse goods based on artificial intelligence according to claim 1, characterized in that, The specific process of calculating the stability index of the current inventory using a preset evaluation algorithm includes: The inventory data in the distributed ledger is called up, and the inventory data is analyzed to obtain the real-time physical quantity data of each item; Query the pre-set cargo attribute database to obtain the dynamic value weight coefficients corresponding to various types of cargo. Perform a weighted calculation on the real-time physical quantity data of cargo and the dynamic value weight coefficients to generate the current inventory value data. Read the system's preset pledge benchmark value, calculate the ratio between the inventory value data and the pledge benchmark value, and define this ratio as the inventory coverage factor; Acquire data on the frequency of goods entering and leaving the warehouse within a preset monitoring period, and generate inventory turnover coefficient through time series analysis; A multidimensional state vector is constructed using the inventory coverage coefficient and the inventory turnover rate coefficient, serving as a stability indicator characterizing the current inventory status of goods in overseas warehouses.

10. The method for monitoring cross-border overseas warehouse goods based on artificial intelligence according to claim 1, characterized in that, The aforementioned status anomaly report includes inventory data and stability indicators; when this report is exported to a domestic terminal, it includes an encryption process, specifically: The AES symmetric encryption algorithm is used to preprocess the status anomaly report, converting plaintext into ciphertext, and then transmitting the ciphertext to the domestic end through a secure channel; During transmission, the TLS protocol is introduced, and after key exchange is completed through an asymmetric encryption algorithm, a symmetric encryption algorithm is used to protect the data end-to-end, and the data integrity is verified by combining a message authentication code. The domestic end decrypts the ciphertext using the key and obtains a status anomaly report.

Images