5G technology-based stereoscopic warehouse in-out secondary checking method and system

Abstract

The invention provides a 5G technology-based stereo warehouse in-out warehouse secondary checking method and system. The method comprises the following steps of: acquiring warehouse location information of a stacker for performing in-out warehouse operation in real time; when the storage location information shows that the stacking machine completes the storage location operation, an image acquisition device is triggered to acquire a cargo image of the storage location; extracting a material identifier and a tray identifier based on the cargo image; storing the material identifier, the tray identifier and the corresponding cargo image in a local database, and updating an inventory record; comparing the updated inventory record in the local database with the corresponding inventory record in the warehouse management module in real time to obtain a comparison result; and when the comparison result shows inconsistency, the alarm module is triggered to give an alarm, and inconsistency of accounts and reality caused by data transmission delay or operation errors can be effectively identified. And process monitoring and instant verification of warehouse-in and warehouse-out operation are realized.

Description

Technical Field

[0001] This invention relates to the field of intelligent warehousing and logistics technology, and in particular to a method and system for secondary verification of inbound and outbound operations in an automated warehouse based on 5G technology. Background Technology

[0002] In the current field of intelligent warehousing and logistics, automated storage and retrieval systems (AS / RS) have become core equipment for improving warehousing efficiency and space utilization. Their core equipment, the stacker crane, performs inbound and outbound operations under the instructions of a warehouse management system (WMS). However, in actual operation, due to various reasons such as delays in WMS system data entry, minor deviations in stacker crane handling, and wear or misidentification of cargo information labels, discrepancies may arise between the inventory records in the WMS system and the actual inventory information in the automated storage and retrieval system locations. This discrepancy directly affects inventory accuracy, leading to subsequent problems such as incorrect shipments and production interruptions, severely restricting the operational reliability and overall efficiency of the automated storage and retrieval system.

[0003] To address these issues, traditional solutions primarily rely on periodic or ad-hoc manual inventory checks. This method requires operators to use stacker cranes to reach each storage location and closely verify the physical inventory against system records. This process is not only extremely labor-intensive and time-consuming, but also necessitates the interruption of normal inbound and outbound operations, severely impacting the efficient and continuous operation of automated warehouses. Summary of the Invention

[0004] This invention provides a method and system for secondary verification of inbound and outbound operations in automated warehouses based on 5G technology, in order to solve the technical problems of low efficiency caused by relying on traditional manual inventory methods and inconsistencies between inventory records and physical information in automated warehouses.

[0005] On one hand, this invention provides a secondary verification method for automated warehouse entry and exit based on 5G technology. The method is used in an automated warehouse secondary verification device, which includes an image acquisition device installed on a stacker crane, a 5G communication module, a warehouse management module, and an alarm module. The method includes: Real-time acquisition of storage location information for stacker crane inbound and outbound operations; When the storage location information indicates that the stacker crane has completed the storage location operation, the image acquisition device is triggered to acquire an image of the goods in that storage location; Extract material markings and pallet markings based on cargo images; Store material labels, pallet labels, and corresponding cargo images in a local database and update inventory records. The updated inventory records in the local database are compared in real time with the corresponding inventory records in the warehouse management module to obtain the comparison results. When the comparison results are inconsistent, the alarm module is triggered.

[0006] According to the present invention, a secondary verification method for inbound and outbound operations of an automated warehouse based on 5G technology is provided, which extracts material identification and pallet identification based on cargo images, including: Multi-scale feature fusion processing is performed on cargo images to construct a fused feature representation that includes local texture features and global semantic features; Parallel decoding of the fused feature representations is performed to generate spatial attention maps for the material identification region and the pallet identification region, respectively. Based on the spatial attention map, adaptive region cropping is performed on the original cargo image to obtain material identification sub-images and pallet identification sub-images; A joint optimization recognition strategy is adopted, in which the material identification sub-image and the pallet identification sub-image are input into the shared feature extraction backbone network, and the material identification and pallet identification are simultaneously recognized through the material identification branch and the pallet identification branch, respectively, to obtain the material identification and pallet identification.

[0007] According to the present invention, a secondary verification method for inbound and outbound operations of an automated warehouse based on 5G technology is provided. Based on a spatial attention map, adaptive region cropping is performed on the original cargo image to obtain material identification sub-images and pallet identification sub-images, including: Based on the spatial attention map of the material identification area, the minimum bounding rectangle of the material identification area is calculated, and the area of ​​interest of the material identification area is generated by expanding it outward according to the first preset expansion ratio. Based on the spatial attention map of the pallet label area, the minimum bounding rectangle of the pallet label area is calculated, and the pallet label area of ​​interest is generated by expanding it outward according to the second preset expansion ratio. Based on the coordinate information of the region of interest of the material identifier, the first image region is cropped from the original cargo image, and the resolution of the first image region is enhanced to obtain the material identifier sub-image; Based on the coordinate information of the region of interest of the pallet label, the second image region is cropped from the original cargo image, and the resolution of the second image region is enhanced to obtain the pallet label sub-image; The first preset expansion ratio is smaller than the second preset expansion ratio.

[0008] According to the present invention, a secondary verification method for inbound and outbound operations of an automated warehouse based on 5G technology is provided. The stacker crane is also equipped with an RFID reading device, which extracts material identification and pallet identification based on cargo images. The method further includes: The electronic tag information of the goods is obtained through an RFID reader; the electronic tag information includes at least the goods' identification data. Visual recognition results are obtained by identifying cargo images. The electronic tag information is compared with the visual recognition results to obtain the comparison results; When the comparison results are consistent, the visual recognition result is used as the final recognition result; When the comparison results are inconsistent, the electronic tag information is used as the final identification result, and a manual review instruction is generated.

[0009] According to the present invention, a secondary verification method for entry and exit of an automated warehouse based on 5G technology compares electronic tag information with visual recognition results to obtain comparison results, including: Extract the first cargo identification feature from the electronic tag information; Extract the second cargo identification features from the visual recognition results; Determine the similarity score between the first cargo identity feature and the second cargo identity feature; The similarity score is compared with a preset score threshold to obtain the comparison result.

[0010] According to the present invention, a secondary verification method for entry and exit of an automated warehouse based on 5G technology, after generating a manual verification instruction, further includes: The electronic tag information, visual recognition results, corresponding cargo images, and storage location information are linked and bound to generate a data package; Data packets are uploaded to the cloud management platform in real time via the 5G communication module; Data packets are displayed in the task dashboard of the cloud management platform and assigned to designated terminal devices; Receive the manual verification confirmation result returned by the designated terminal device, and update the inventory records of the local database and warehouse management module based on the manual verification confirmation result.

[0011] According to the present invention, a secondary verification method for entry and exit of an automated warehouse based on 5G technology is provided, which performs multi-scale feature fusion processing on cargo images to construct a fused feature representation including local texture features and global semantic features, including: A semantic distribution map is generated based on cargo images. This semantic distribution map is used to identify the location of material labels and pallet labels in the image and their corresponding confidence levels. Based on the region coordinate information of the semantic distribution map, the corresponding local image region is cropped from the original cargo image; The cropped local image region is processed to extract texture features containing identifying text, barcodes, or patterns; Based on the region confidence in the semantic distribution map, the extracted texture features are weighted, and the region with higher confidence corresponds to a larger feature weight. The weighted texture features are fused with the semantic features represented by the semantic distribution map to generate a fused feature representation that includes local details and overall semantic information.

[0012] According to the present invention, a secondary verification method for inbound and outbound operations in an automated warehouse based on 5G technology is provided, which performs parallel decoding on the fused feature representation to generate spatial attention maps of the material identification area and the pallet identification area, respectively, including: The fused feature representations are simultaneously input into the material identification decoding channel and the pallet identification decoding channel for parallel processing. A cross-attention mechanism is established between the two decoding channels, so that the material identification decoding process can refer to the feature information of the pallet area, and the pallet identification decoding process can refer to the feature information of the material area. Based on the output of the cross-attention mechanism, spatial attention maps of the material identification area and the pallet identification area are generated respectively. The two spatial attention maps are superimposed and fused, and the conflict in the overlapping area is resolved to generate the final dual-target spatial attention map.

[0013] According to the present invention, a secondary verification method for inbound and outbound operations in an automated warehouse based on 5G technology employs a joint optimization identification strategy. Material identification sub-images and pallet identification sub-images are input into a shared feature extraction backbone network, and synchronous identification is performed through the material identification branch and pallet identification branch respectively to obtain material identification and pallet identification, including: The material identification sub-image and the pallet identification image are input into the shared feature extraction backbone network to extract the shared feature map; The shared feature map is input into the material identification branch and the pallet identification branch respectively for synchronous identification, and a two-way information interaction mechanism is established between the two branches during the identification process to obtain their respective output results; Based on the output results of the two branches, the final recognition results of the material identifier and pallet identifier are obtained by decoding.

[0014] On the other hand, the present invention also provides a secondary verification system for automated warehouse entry and exit based on 5G technology. The system is used for secondary verification devices in automated warehouses. The devices include image acquisition equipment and a 5G communication module installed on a stacker crane, a warehouse management module, and an alarm module. The system includes: The acquisition module is used to acquire the storage location information for stacker crane inbound and outbound operations in real time. The image module is used to trigger the image acquisition device to capture images of the goods in the storage location when the stacker crane completes the storage location operation; The extraction module is used to extract material identification and pallet identification based on cargo images; The storage module is used to store material labels, pallet labels, and corresponding cargo images to the local database and update inventory records. The comparison module is used to compare the updated inventory records in the local database with the corresponding inventory records in the warehouse management module in real time to obtain the comparison results. The alarm module is used to trigger an alarm when the comparison results are inconsistent.

[0015] This invention provides a 5G-based automated warehouse inbound / outbound secondary verification method and system. Automatic verification is triggered by stacker crane operation nodes, utilizing 5G communication to achieve millisecond-level data synchronization, enabling real-time verification while maintaining continuous operation. Compared to single-system data verification, the dual comparison mechanism between the local database and the remote system effectively identifies discrepancies between records and actual inventory caused by data transmission delays or operational errors. It achieves process monitoring and real-time verification of inbound / outbound operations, reducing inventory discrepancy identification time from several hours for manual inventory checks to seconds after operation completion. The dual data verification mechanism reduces the probability of misjudgments due to single-point data errors, and 5G communication ensures the real-time transmission of massive image data and inventory records, effectively improving the accuracy of automated warehouse inventory data and operational reliability. Attached Figure Description

[0016] To more clearly illustrate the technical solutions in this invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are some embodiments of this invention. For those skilled in the art, other drawings can be obtained from these drawings without creative effort.

[0017] Figure 1 This is a flowchart illustrating the secondary verification method for inbound and outbound operations of a three-dimensional warehouse based on 5G technology provided in this embodiment of the invention. Figure 2 This is a schematic diagram of the structure of the 5G-based automated warehouse entry and exit secondary verification system provided in this embodiment of the invention; Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention. Detailed Implementation

[0018] To make the objectives, technical solutions, and advantages of this invention clearer, the technical solutions of this 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 this invention. All other embodiments obtained by those skilled in the art based on the embodiments of this invention without creative effort are within the scope of protection of this invention.

[0019] Figure 1 This is a flowchart illustrating the secondary verification method for automated warehouse entry and exit based on 5G technology provided in this embodiment of the invention. The secondary verification method for automated warehouse entry and exit is used in an automated warehouse secondary verification device, which includes an image acquisition device installed on a stacker crane, a 5G communication module, a warehouse management module, and an alarm module.

[0020] A stacker crane is a specialized crane that uses forks or booms to pick up, move, and stack goods. It is mainly divided into two types: bridge stacker cranes and aisle stacker cranes (also known as lane cranes). They are primarily used in warehouses, workshops, and other locations to pick up, move, and stack unitized goods or retrieve them from high-rise racks. An automated warehouse (AS / RS) refers to a warehouse that uses several, dozens, or even hundreds of layers of racks to store unitized goods and uses corresponding material handling equipment for inbound and outbound operations. It can effectively utilize limited space to store goods, hence the name "automated warehouse."

[0021] See Figure 1 The secondary verification method for entry and exit of automated warehouses based on 5G technology may include the following steps.

[0022] Step 101: Obtain the storage location information for the stacker crane's inbound and outbound operations in real time.

[0023] Step 102: When the storage location information indicates that the stacker crane has completed the storage location operation, the image acquisition device is triggered to acquire the image of the goods in that storage location.

[0024] Step 103: Extract material identification and pallet identification based on the cargo image.

[0025] Step 104: Store the material identification, pallet identification, and corresponding cargo image to the local database, and update the inventory records.

[0026] Step 105: Compare the updated inventory records in the local database with the corresponding inventory records in the warehouse management module in real time to obtain the comparison results.

[0027] Step 106: When the comparison results are inconsistent, trigger the alarm module to sound an alarm.

[0028] The image acquisition equipment refers to the optical imaging device installed on the stacker crane, which can be an industrial-grade CCD camera or an infrared imager. It captures surface feature information of the goods, providing immediate data on the goods' status after the operation and supporting subsequent identification and extraction. The 5G communication module is a transmission device supporting fifth-generation mobile communication technology, using an integrated 5G module to achieve high-speed data interaction between the local database and the remote warehouse management module. It eliminates the wiring limitations of traditional wired transmission, ensuring real-time synchronization of inventory data. The local database is a storage system deployed in the stacker crane control unit, which can be an embedded database. It temporarily stores the goods information collected on-site, establishing an independent data copy of the operation site and providing a comparison benchmark for dual verification.

[0029] Specifically, when the stacker crane performs storage or retrieval operations, the control system records the target storage location coordinates and operation type in real time. After the robotic arm completes the grabbing or placing of goods, it triggers the image acquisition device to take multi-angle pictures of the current storage location. The image processing unit extracts the material number barcode and pallet positioning mark through feature recognition algorithms, and stores the structured data and the original image together in a local embedded database. The 5G module simultaneously uploads the updated local inventory records to the warehouse management module, triggering the system-level data comparison engine. When a discrepancy is detected in the material code, storage location, or quantity information, the audible and visual alarm module is immediately activated and the current operation process is frozen.

[0030] In this embodiment, automatic verification is triggered by the stacker crane operation node, and millisecond-level data synchronization is achieved using 5G communication, enabling real-time verification while maintaining continuous operation. Compared to single-system data verification, the dual comparison mechanism of the local database and the remote system can effectively identify discrepancies between records and actual inventory caused by data transmission delays or operational errors. This embodiment realizes process monitoring and real-time verification of inbound and outbound operations, reducing the inventory discrepancy identification time from several hours in traditional manual inventory checks to a second-level response after the operation is completed. The dual data verification mechanism reduces the probability of misjudgment due to single-point data errors, and 5G communication ensures the real-time transmission of massive image data and inventory records, effectively improving the accuracy of automated warehouse inventory data and operational reliability.

[0031] In one embodiment of this specification, based on an image of the goods, material identification and pallet identification are extracted, including: Step 1: Perform multi-scale feature fusion processing on the cargo image to construct a fused feature representation that includes local texture features and global semantic features; Step 2: Perform parallel decoding on the fused feature representation to generate spatial attention maps for the material identification area and the pallet identification area, respectively; Step 3: Based on the spatial attention map, perform adaptive region cropping on the original cargo image to obtain the material identification sub-image and the pallet identification sub-image; Step 4: Using a joint optimization recognition strategy, the material identification sub-image and the pallet identification sub-image are input into the shared feature extraction backbone network, and synchronous recognition is performed through the material identification branch and the pallet identification branch respectively to obtain the material identification and pallet identification.

[0032] Multi-scale feature fusion processing refers to extracting and combining local details and global semantic information from feature maps of different scales. This can be implemented using a pyramid network structure, capturing visual features of different granularities through multi-level convolutional operations. Parallel decoding involves processing two target regions simultaneously: material identification and pallet identification. This can be implemented using a dual-branch neural network architecture, with each branch independently generating a spatial attention map for its corresponding region. The spatial attention map is a heatmap representing the importance of target regions in an image. It can be generated using an attention mechanism, locating the target region by calculating the weight distribution at each position in the feature map. Adaptive region cropping dynamically adjusts the cropping range based on the attention map. This can be achieved by calculating the minimum bounding rectangle of the target region and expanding it to generate a region of interest, ensuring complete coverage of the identification area. Joint optimization recognition strategy involves sharing backbone network parameters while performing dual-task collaborative training. This can be implemented using a multi-task learning framework, improving feature reuse efficiency through parameter sharing.

[0033] Specifically, the cargo image first passes through a multi-scale feature extraction network to generate feature maps containing semantic information at different levels. A feature pyramid structure is then used to fuse low-level texture features with high-level semantic features across scales. The fused feature representation is simultaneously input into both the material identification decoder and the pallet identification decoder. The two decoders exchange regional feature information through a cross-attention mechanism, dynamically adjusting their respective focus areas during the decoding process. The spatial attention maps output from the decoder are then superimposed and fused to generate a dual-target attention heatmap. The minimum bounding rectangle is calculated based on the coordinates of high-response regions in the heatmap, and then expanded outward according to a preset expansion ratio to generate the coordinates of the cropped region. The corresponding cropped region in the original image is then enhanced through super-resolution reconstruction and input into a shared backbone network. Character recognition and barcode parsing are performed in the material identification branch and the pallet identification branch, respectively, ultimately outputting the dual-identification recognition result.

[0034] In this embodiment, the problem of accurate positioning and collaborative recognition of dual-target labels in complex warehousing environments is effectively solved. The accuracy of regional positioning is enhanced by parallel decoding and cross-attention mechanism. Multi-scale feature fusion is used to improve the recognition robustness of image quality degradation scenarios. A joint optimization strategy is adopted to reduce the computational resource consumption of the dual-task system, and efficient synchronous verification of materials and pallet labels is achieved.

[0035] In one embodiment of this specification, adaptive region cropping is performed on the original cargo image based on a spatial attention map to obtain a material identification sub-image and a pallet identification sub-image, including: Step 1: Based on the spatial attention map of the material identification area, calculate the minimum bounding rectangle of the material identification area, and expand it outward according to the first preset expansion ratio to generate the material identification region of interest; Step 2: Based on the spatial attention map of the pallet label area, calculate the minimum bounding rectangle of the pallet label area, and expand it outward according to the second preset expansion ratio to generate the pallet label region of interest; Step 3: Based on the coordinate information of the region of interest of the material identifier, crop out the first image region from the original cargo image, and perform resolution enhancement processing on the first image region to obtain the material identifier sub-image; Step 4: Based on the coordinate information of the region of interest of the pallet label, crop out the second image region from the original cargo image, and perform resolution enhancement processing on the second image region to obtain the pallet label sub-image; The first preset expansion ratio is smaller than the second preset expansion ratio.

[0036] In this embodiment, the minimum bounding rectangle refers to the smallest rectangle that can completely enclose the target area. This can be achieved using a rotating caliper algorithm or an edge detection combined with geometric fitting algorithm to accurately locate the spatial range of the marked area. The preset expansion ratio refers to the magnitude of outward expansion relative to the original area boundary. This can be implemented using a fixed value or a dynamic calculation method. For example, the first expansion ratio can be 5%-10%, and the second expansion ratio can be 15%-25%. This expansion operation avoids the truncation of the marked information due to image acquisition angle deviation. Resolution enhancement processing refers to the operation of improving the clarity of image details. This can be achieved using super-resolution reconstruction algorithms or deep learning-based image inpainting models to solve the image blurring problem caused by long-distance shooting.

[0037] Specifically, once the spatial attention map locates the potential areas for material and pallet labels, the minimum bounding rectangles of each are first calculated as the basic bounding boxes. Since pallet labels typically have larger physical dimensions and relatively fixed positions, a larger expansion ratio is used to ensure complete coverage of the pallet code. For material labels, which may be tilted or partially occluded, a smaller expansion ratio is used to avoid introducing excessive background interference. The cropped image area undergoes resolution enhancement processing to effectively recover details lost due to stacker crane vibrations or long-distance shooting, such as barcode edge sharpening and text stroke completion, thus providing high-quality input for subsequent recognition.

[0038] In this embodiment, by adapting the different expansion ratios to the physical characteristics of various signs, and combining resolution enhancement processing to specifically optimize image quality, stable recognition performance can be maintained even under complex working conditions. Through the above technical solution, this application effectively solves the problem of sign information truncation caused by positioning deviation in the image acquisition area, avoids the risk of misjudgment due to image blurring during manual review, and significantly improves the processing efficiency and accuracy of automated warehouses in the secondary verification of goods information.

[0039] In one embodiment of this specification, the stacker crane is further equipped with an RFID reading device, which extracts material identification and pallet identification based on cargo images, and also includes: Step 1: Obtain the electronic tag information of the goods through an RFID reader; the electronic tag information shall contain at least the goods' identification data; Step 2: Perform recognition based on the cargo image to obtain visual recognition results; Step 3: Compare the electronic tag information with the visual recognition results to obtain the comparison results; Step 4: When the comparison results are consistent, the visual recognition result is used as the final recognition result; Step 5: When the comparison results are inconsistent, the electronic tag information is used as the final identification result, and a manual review instruction is generated.

[0040] The RFID reading device refers to a device that automatically identifies target objects and acquires relevant data through radio frequency signals. Specifically, it can be implemented using an UHF reader in conjunction with passive electronic tags, enabling rapid reading of cargo identification data under non-contact conditions. Electronic tag information refers to the digital identification information stored in the tag attached to the cargo, specifically implemented using EPC encoding format, including the cargo's unique identifier, batch number, and specifications. Visual recognition results refer to the identification information parsed from cargo images using image analysis algorithms. Specifically, it can be implemented using optical character recognition combined with barcode decoding algorithms, used to extract material numbers and pallet codes. Comparison results refer to the consistency judgment of the output information from two recognition methods, specifically implemented using feature vector similarity calculation, with a threshold set to determine whether a match exists. Manual review instructions refer to the operation commands that trigger the anomaly handling process, specifically implemented using a work order generation interface, pushing abnormal data to the manual review queue.

[0041] Specifically, after the stacker crane completes its storage operation, the RFID reader automatically scans the electronic tags on the goods to obtain coded data, while the image acquisition device captures images of the goods for visual recognition. After extracting the goods' identity features using both identification methods, the system calculates the feature similarity between the RFID data and the visual recognition data. If the similarity reaches a preset threshold, it is considered a match, and the inventory record is updated using the visual recognition result. If the similarity does not reach the threshold, it is considered an anomaly, and the more interference-resistant RFID data is prioritized as the baseline value. Simultaneously, a manual review task containing anomaly information is generated. This process effectively reduces the risk of misjudgment using a single identification method through a dual verification mechanism, automatically triggering the review process in case of data conflicts to ensure the accuracy of inventory records.

[0042] In this embodiment, a redundant verification mechanism is constructed through data fusion verification of RFID and visual recognition. By automatically selecting a reliable data source and generating verification instructions through preset logic, the response time for anomaly handling is significantly shortened. By associating and storing abnormal data packets, complete information support is provided for remote verification. Through the above technical solution, this application effectively solves the problem of discrepancies between accounts and actual inventory caused by a single identification error, reducing the probability of misidentification through cross-verification of dual data sources. In the event of data conflict, RFID data with stronger anti-interference capabilities is automatically selected as a temporary benchmark, preventing erroneous data from being directly entered into the database. Abnormal situations automatically trigger the verification process and associate and store a complete chain of evidence, significantly improving the standardization and traceability of anomaly handling and reducing inventory record deviations caused by identification errors.

[0043] In one embodiment of this specification, the electronic tag information is compared with the visual recognition result to obtain the comparison result, including: Extract the first cargo identification feature from the electronic tag information; Extract the second cargo identification features from the visual recognition results; Determine the similarity score between the first cargo identity feature and the second cargo identity feature; The similarity score is compared with a preset score threshold to obtain the comparison result.

[0044] The first cargo identification feature refers to the unique identifier data of the cargo parsed from the electronic tag information. Specifically, this can be achieved by using a hash algorithm to extract feature vectors of key information such as the cargo number and batch number, used to characterize the cargo's identity attributes. The second cargo identification feature refers to the visual features extracted from the cargo image through image recognition technology. Specifically, this can be achieved by using a convolutional neural network to extract the depth feature vectors of text, barcodes, or patterns on the cargo surface, used to characterize the cargo's visual attributes. The similarity score is a quantified value of the matching degree between two types of feature vectors, specifically calculated using cosine similarity or Euclidean distance algorithms, used to determine the consistency between the electronic tag and the visual recognition result. The preset score threshold is a pre-set judgment threshold, specifically determined using a dynamic threshold adjustment algorithm combined with historical recognition data, used to distinguish whether the comparison results are consistent.

[0045] Specifically, after the stacker crane completes the storage location operation, the RFID reader acquires the electronic tag information of the goods, while the image acquisition device captures images of the goods and performs visual recognition. The goods identification data in the electronic tag information is converted into a first feature vector using a hash algorithm, and the visual recognition result is extracted into a second feature vector using a neural network. Both feature vectors are input into a similarity calculation module, which uses a cosine similarity algorithm to generate a similarity score between 0 and 1. This score is compared with a preset threshold; if it is higher than the threshold, it is considered a match; otherwise, it is considered an inconsistency. For example, when the preset threshold is set to 0.85, if the similarity score is 0.92, the comparison result is considered a match, and the system uses the visual recognition result to update the inventory record; if the similarity score is 0.78, the comparison result is considered an inconsistency, and the system prioritizes the electronic tag information and triggers a manual review process.

[0046] In this embodiment, by integrating the dual feature comparison of electronic tags and visual recognition, conflicts between the two data sources can be effectively identified, avoiding discrepancies between records and actual inventory caused by a single identification error. Automated feature comparison and threshold determination significantly reduce the frequency of manual intervention. Through the above technical solution, this application can solve the problem of inventory record deviations caused by single identification errors in traditional methods. By using dual feature comparison and dynamic threshold determination, the accuracy and reliability of goods identification verification are improved, ensuring consistency between electronic tag information and visual recognition results, thereby reducing the risk of shipping errors or production interruptions caused by discrepancies between records and actual inventory.

[0047] In one embodiment of this specification, after generating the manual review instruction, the method further includes: The electronic tag information, visual recognition results, corresponding cargo images, and storage location information are linked and bound to generate a data package; Data packets are uploaded to the cloud management platform in real time via the 5G communication module; Data packets are displayed in the task dashboard of the cloud management platform and assigned to designated terminal devices; Receive the manual verification confirmation result returned by the designated terminal device, and update the inventory records of the local database and warehouse management module based on the manual verification confirmation result.

[0048] In this context, a data packet refers to a transmission unit that structurally integrates information from different sources. It can be implemented using JSON or XML formats to ensure the integrity and relevance of information during transmission. A cloud management platform refers to a data processing center deployed on a cloud server. It can be implemented using a distributed microservice architecture to support high-concurrency task processing and real-time data synchronization. A task dashboard is a visual task scheduling interface. It can be dynamically updated using the WebSocket protocol to intuitively display the status and allocation of tasks awaiting review. A designated terminal device refers to a mobile device with manual review capabilities. It can be implemented using an industrial tablet or smartphone to receive tasks and provide review results.

[0049] Specifically, when the electronic tag information is inconsistent with the visual recognition result, the system automatically triggers a data association operation, encapsulating the cargo identification data, image evidence, and operation location information into a structured data packet. Leveraging the high bandwidth of the 5G network, the data packet is transmitted to the cloud in real time, generating a work order in the task dashboard. The cloud platform automatically assigns the work order to idle terminal devices based on preset rules. Operators view the cargo image and identify discrepancies through the device interface, complete manual verification, and submit confirmation. The system synchronously updates the local database and warehouse management module based on the verification results, ensuring that inventory records match the actual cargo status.

[0050] In this embodiment, 5G communication enables real-time uploading and remote verification of abnormal data. Operators can complete the verification without entering the automated warehouse, and the cloud-based task allocation mechanism avoids the efficiency bottleneck of manual order dispatch. Through the above technical solution, this application realizes contactless verification operations in abnormal situations, effectively reducing the operation interruption time caused by manual intervention, ensuring the real-time and accurate updating of inventory data, and improving the response efficiency of abnormal events through cloud-based collaborative processing.

[0051] In one embodiment of this specification, a multi-scale feature fusion process is performed on a cargo image to construct a fused feature representation that includes local texture features and global semantic features, including: A semantic distribution map is generated based on cargo images. This semantic distribution map is used to identify the location of material labels and pallet labels in the image and their corresponding confidence levels. Based on the region coordinate information of the semantic distribution map, the corresponding local image region is cropped from the original cargo image; The cropped local image region is processed to extract texture features containing identifying text, barcodes, or patterns; Based on the region confidence in the semantic distribution map, the extracted texture features are weighted, and the region with higher confidence corresponds to a larger feature weight. The weighted texture features are fused with the semantic features represented by the semantic distribution map to generate a fused feature representation that includes local details and overall semantic information.

[0052] The semantic distribution map refers to a two-dimensional probability map generated by a deep learning model. Specifically, it can be implemented using a convolutional neural network to perform pixel-level classification of cargo images, predicting the spatial distribution probability of material and pallet labels within the image. Local image regions are image segments cropped based on the coordinate information of the semantic distribution map. This can be achieved using bilinear interpolation algorithms for region cropping, focusing on image portions that may contain labeling information. Texture features represent the visual features of label text, barcodes, or patterns. This can be implemented using histogram of oriented gradients or convolutional feature extractors, capturing the microscopic details of the labeling region. Weighting adjusts the feature contribution based on region confidence, using learnable attention coefficients or preset linear weights to suppress noise interference in low-quality regions. Fusion feature representation is a composite data structure combining texture and semantic features, implemented using channel concatenation or feature map overlay, simultaneously preserving local details and global contextual information.

[0053] Specifically, after the stacker crane completes the storage operation, the cargo image is input into a feature extraction network to generate a semantic distribution map. This map, in the form of a heatmap, marks the possible locations of material and pallet labels and their detection confidence levels. Based on the coordinate boundaries of high-confidence regions in the distribution map, local regions containing the labels are precisely cropped from the original image. For each cropped region, an image enhancement algorithm is used to improve the resolution, and then a feature extractor captures micro-texture features such as text and barcodes. Based on the region confidence parameters provided by the semantic distribution map, features in low-confidence regions are given smaller weights to reduce their impact on the final recognition. Finally, the weighted texture features and semantic features representing the overall layout are fused to form a composite feature vector that simultaneously contains detail resolution and semantic understanding capabilities, which serves as the input for subsequent label recognition.

[0054] In this embodiment, the semantic distribution map-guided local feature extraction and confidence-weighted mechanism effectively focuses on key details in high-confidence areas. Simultaneously, feature fusion preserves global semantic relationships, solving the problems of inaccurate identification region positioning and insufficient feature representation capabilities in complex scenarios. Through this technical solution, this application can accurately extract and fuse visual features at different scales even when cargo images are partially occluded or unevenly lit, significantly improving the robustness of material and pallet identification. This multi-scale feature collaboration mechanism effectively reduces the identification error rate caused by local image quality defects, providing a more reliable data foundation for subsequent inventory record comparisons, thereby reducing false alarms.

[0055] In one embodiment of this specification, parallel decoding is performed on the fused feature representation to generate spatial attention maps of the material identification region and the pallet identification region, respectively, including: The fused feature representations are simultaneously input into the material identification decoding channel and the pallet identification decoding channel for parallel processing. A cross-attention mechanism is established between the two decoding channels, so that the material identification decoding process can refer to the feature information of the pallet area, and the pallet identification decoding process can refer to the feature information of the material area. Based on the output of the cross-attention mechanism, spatial attention maps of the material identification area and the pallet identification area are generated respectively. The two spatial attention maps are superimposed and fused, and the conflict in the overlapping area is resolved to generate the final dual-target spatial attention map.

[0056] Parallel decoding refers to using two independent branch networks to handle the localization tasks of material tags and pallet tags respectively. Specifically, this can be implemented using a dual-branch convolutional neural network architecture, improving the localization accuracy of different tags through independent parameter learning. The cross-attention mechanism establishes an information exchange channel between the two branches during the decoding process. This can be implemented using a multi-head attention module, enabling material tag decoding to utilize the spatial context information of the pallet area, and pallet tag decoding to reference the semantic features of the material area. The spatial attention map is a two-dimensional heatmap representing the probability distribution of target regions in an image. It can be generated using convolutional layers and activation functions to guide subsequent image cropping operations. Conflict resolution involves performing logical operations on the superimposed attention maps to eliminate decision-making contradictions in overlapping areas. This can be implemented using non-maximum suppression algorithms or region priority rules.

[0057] Specifically, a dual-channel parallel processing structure is adopted in the feature decoding stage. The material identification decoding channel and the pallet identification decoding channel share the initial feature input but have independent parameter update capabilities. Cross-channel information interaction is established through a cross-attention mechanism. For example, during the material identification decoding process, key-value pairs of the pallet area feature map are introduced to calculate their correlation with the current decoding state, thereby dynamically adjusting the attention distribution. After decoding, the two spatial attention maps generated are superimposed at the pixel level to form a composite heatmap. For overlapping areas, a conflict resolution strategy based on confidence ranking is adopted to retain the attention weights of high-confidence areas.

[0058] In this embodiment, dual-target collaborative localization is achieved through parallel decoding and cross-attention mechanisms. This enhances the recognition of regional correlations while maintaining processing efficiency, and avoids misjudgment of overlapping areas through conflict resolution strategies. Through the above technical solutions, this application can effectively solve the problem of mutual interference between material tags and pallet tags in complex warehousing environments, improve the accuracy and robustness of dual-target localization, provide accurate regions of interest for subsequent tag recognition, and thus reduce the risk of inconsistencies between inventory records and physical information.

[0059] In one embodiment of this specification, a joint optimization recognition strategy is adopted. The material identification sub-image and the pallet identification sub-image are input into a shared feature extraction backbone network, and synchronous recognition is performed through the material identification branch and the pallet identification branch, respectively, to obtain the material identification and pallet identification, including: The material identification sub-image and the pallet identification image are input into the shared feature extraction backbone network to extract the shared feature map; The shared feature map is input into the material identification branch and the pallet identification branch respectively for synchronous identification, and a two-way information interaction mechanism is established between the two branches during the identification process to obtain their respective output results; Based on the output results of the two branches, the final recognition results of the material identifier and pallet identifier are obtained by decoding.

[0060] The shared feature extraction backbone network refers to a general feature extraction structure used to simultaneously process material label sub-images and pallet label sub-images. Specifically, it can be implemented using a lightweight network architecture based on depthwise separable convolutions, reducing computational resource consumption through parameter sharing. The bidirectional information interaction mechanism is a communication module that allows feature information exchange between the material label branch and the pallet label branch. This can be implemented using a cross-branch attention mechanism, enhancing the synergy of the recognition process by establishing correlations between feature channels. The adaptive weight adjustment strategy is a control method that dynamically adjusts the intensity of information interaction based on real-time recognition confidence. Specifically, it can be implemented using a dynamic weight allocation algorithm based on gated recurrent units, automatically optimizing the information fusion ratio by evaluating the recognition reliability of each branch.

[0061] Specifically, the material label sub-image and the pallet label sub-image are input into the same feature extraction network for parallel processing. Sharing network parameters reduces model complexity. During feature extraction, the material branch and the pallet branch exchange intermediate feature data through a bidirectional interaction channel. For example, the pallet branch passes contextual features containing location information to the material branch to help it accurately locate the label area; the material branch, in turn, feeds back feature maps containing semantic information to the pallet branch to help it parse the label content. During the interaction, the system monitors the recognition confidence index of the two branches in real time. When the confidence of one branch falls below a preset threshold, it automatically increases its information dependency weight on the other branch, achieving error suppression and result optimization through dynamic adjustment.

[0062] In some specific implementations, the shared feature extraction backbone network can be trained end-to-end using a multi-task learning framework, and the recognition accuracy of material labels and pallet labels can be optimized synchronously through a joint loss function. The bidirectional information interaction mechanism can be designed as a cross-branch feature fusion module, using a channel attention mechanism to filter effective information. The adaptive weight adjustment strategy can be implemented by combining the probability distribution of branch outputs and using a soft threshold mechanism to control the intensity of information interaction.

[0063] In this embodiment, redundant computation is reduced by sharing the backbone network, spatial and semantic relationships between identifiers are mined by a two-way interaction mechanism, and recognition robustness in complex scenarios is improved by dynamic weight adjustment, effectively solving the problem of overall verification failure caused by unilateral recognition errors.

[0064] Based on the same general inventive concept, this invention also protects a 5G-based automated warehouse entry and exit secondary verification system, such as... Figure 2 As shown, Figure 2 This is a schematic diagram of the structure of the automated warehouse entry and exit secondary verification system based on 5G technology provided in an embodiment of the present invention. The automated warehouse entry and exit secondary verification system based on 5G technology provided by the present invention is described below. The automated warehouse entry and exit secondary verification system based on 5G technology described below can be referred to in correspondence with the automated warehouse entry and exit secondary verification method based on 5G technology described above.

[0065] The 5G-based automated warehouse entry and exit secondary verification system is used for automated warehouse secondary verification devices. The device includes image acquisition equipment and a 5G communication module, a warehouse management module, and an alarm module installed on the stacker crane.

[0066] The automated warehouse entry and exit secondary verification system based on 5G technology includes: The acquisition module 201 is used to acquire the storage location information of the stacker crane for inbound and outbound operations in real time; Image module 202 is used to trigger image acquisition device to acquire images of goods in the storage location when the stacker crane completes the storage location operation; Extraction module 203 is used to extract material identification and pallet identification based on cargo images; The storage module 204 is used to store material identification, pallet identification and corresponding goods images to the local database and update inventory records; The comparison module 205 is used to compare the updated inventory records in the local database with the corresponding inventory records in the warehouse management module in real time to obtain the comparison results; The alarm module 206 is used to trigger an alarm when the comparison results are inconsistent.

[0067] Figure 3 This is a schematic diagram of the structure of the electronic device provided in an embodiment of the present invention.

[0068] like Figure 3 As shown, the electronic device may include a processor 310, a communication interface 320, a memory 330, and a communication bus 340. The processor 310, communication interface 320, and memory 330 communicate with each other via the communication bus 340. The processor 310 can call logical instructions from the memory 330 to execute a 5G-based automated warehouse entry and exit secondary verification method.

[0069] Furthermore, the logical instructions in the aforementioned memory 330 can be implemented as software functional units and, when sold or used as independent products, can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, essentially, or the part that contributes to the prior art, or a part of the technical solution, can be embodied in the form of a software product. This computer software product is stored in a storage medium and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute all or part of the steps of the methods described in the various embodiments of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0070] On the other hand, the present invention also provides a computer program product, which includes a computer program that can be stored on a non-transitory computer-readable storage medium. When the computer program is executed by a processor, the computer is able to execute the 5G-based three-dimensional warehouse entry and exit secondary verification method provided by the above methods.

[0071] In another aspect, the present invention also provides a non-transitory computer-readable storage medium storing a computer program thereon, which, when executed by a processor, implements the 5G-based automated warehouse entry and exit secondary verification method provided by the above methods.

[0072] The device embodiments described above are merely illustrative. The units described as separate components may or may not be physically separate. The components shown as units may or may not be physical units; that is, they may be located in one place or distributed across multiple network units. Some or all of the modules can be selected to achieve the purpose of this embodiment according to actual needs. Those skilled in the art can understand and implement this without any creative effort.

[0073] Through the above description of the embodiments, those skilled in the art can clearly understand that each embodiment can be implemented by means of software plus necessary general-purpose hardware platforms, and of course, it can also be implemented by hardware. Based on this understanding, the above technical solutions, in essence or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product can be stored in a computer-readable storage medium, such as ROM / RAM, magnetic disk, optical disk, etc., and includes several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) to execute the methods described in the various embodiments or some parts of the embodiments.

[0074] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for secondary verification of inbound and outbound operations in an automated warehouse based on 5G technology, characterized in that, The method is used in a secondary verification device for automated warehouses. The device includes an image acquisition device and a 5G communication module installed on a stacker crane, a warehouse management module, and an alarm module. The method includes: Real-time acquisition of storage location information for stacker crane inbound and outbound operations; When the storage location information indicates that the stacker crane has completed the storage location operation, the image acquisition device is triggered to acquire an image of the goods in that storage location; Based on the cargo image, extract material identification and pallet identification; The material identifier, the pallet identifier, and the corresponding cargo image are stored in the local database, and the inventory records are updated. The updated inventory records in the local database are compared with the corresponding inventory records in the warehouse management module in real time to obtain the comparison results. When the comparison results are inconsistent, the alarm module is triggered to sound an alarm.

2. The method for secondary verification of entry and exit of automated warehouses based on 5G technology according to claim 1, characterized in that, Based on the cargo image, extract cargo identification and pallet identification, including: The cargo image is subjected to multi-scale feature fusion processing to construct a fused feature representation that includes local texture features and global semantic features; The fused feature representation is decoded in parallel to generate spatial attention maps of the material identification area and the pallet identification area, respectively. Based on the spatial attention map, adaptive region cropping is performed on the original cargo image to obtain material identification sub-images and pallet identification sub-images; A joint optimization recognition strategy is adopted, in which the material identification sub-image and the pallet identification sub-image are input into the shared feature extraction backbone network, and the material identification and pallet identification are simultaneously recognized through the material identification branch and the pallet identification branch, respectively, to obtain the material identification and pallet identification.

3. The method for secondary verification of entry and exit of an automated warehouse based on 5G technology according to claim 2, characterized in that, Based on the spatial attention map, adaptive region cropping is performed on the original cargo image to obtain a material identification sub-image and a pallet identification sub-image, including: Based on the spatial attention map of the material identification area, the minimum bounding rectangle of the material identification area is calculated, and the area of ​​interest of the material identification area is generated by expanding it outward according to the first preset expansion ratio. Based on the spatial attention map of the pallet label area, the minimum bounding rectangle of the pallet label area is calculated, and the pallet label area of ​​interest is generated by expanding it outward according to the second preset expansion ratio. Based on the coordinate information of the region of interest of the material identifier, a first image region is cropped from the original cargo image, and the resolution of the first image region is enhanced to obtain a material identifier sub-image. Based on the coordinate information of the region of interest of the pallet label, a second image region is cropped from the original cargo image, and the resolution of the second image region is enhanced to obtain a pallet label sub-image; The first preset expansion ratio is smaller than the second preset expansion ratio.

4. The method for secondary verification of entry and exit of automated warehouses based on 5G technology according to claim 1, characterized in that, The stacker crane is also equipped with an RFID reader, which extracts material identification and pallet identification based on the cargo image, and also includes: The electronic tag information of the goods is obtained through an RFID reader; the electronic tag information includes at least the goods' identification data. Based on the cargo image, visual recognition results are obtained; The electronic tag information is compared with the visual recognition results to obtain the comparison results; When the comparison results are consistent, the visual recognition result is used as the final recognition result; When the comparison results are inconsistent, the electronic tag information is used as the final identification result, and a manual review instruction is generated.

5. The method for secondary verification of entry and exit of an automated warehouse based on 5G technology according to claim 4, characterized in that, The electronic tag information is compared with the visual recognition results to obtain the comparison results, including: Extract the first cargo identification feature from the electronic tag information; Extract the second cargo identification features from the visual recognition results; Determine the similarity score between the first cargo identity feature and the second cargo identity feature; The similarity score is compared with a preset score threshold to obtain the comparison result.

6. The method for secondary verification of entry and exit of an automated warehouse based on 5G technology according to claim 4, characterized in that, After generating the manual review instruction, the process also includes: The electronic tag information, the visual recognition result, the corresponding cargo image, and the warehouse location information are associated and bound to generate a data package; The data packets are uploaded to the cloud management platform in real time via the 5G communication module; The data packet is displayed in the task dashboard of the cloud management platform and assigned to the designated terminal device; Receive the manual verification confirmation result returned by the designated terminal device, and update the inventory records of the local database and warehouse management module based on the manual verification confirmation result.

7. The method for secondary verification of entry and exit of an automated warehouse based on 5G technology according to claim 4, characterized in that, The cargo image undergoes multi-scale feature fusion processing to construct a fused feature representation that includes local texture features and global semantic features, including: A semantic distribution map is generated based on cargo images. This semantic distribution map is used to identify the location of material labels and pallet labels in the image and their corresponding confidence levels. Based on the region coordinate information of the semantic distribution map, the corresponding local image region is cropped from the original cargo image; The cropped local image region is processed to extract texture features containing identifying text, barcodes, or patterns; Based on the region confidence in the semantic distribution map, the extracted texture features are weighted, and the region with higher confidence corresponds to a larger feature weight. The weighted texture features are fused with the semantic features represented by the semantic distribution map to generate a fused feature representation that includes local details and overall semantic information.

8. The method for secondary verification of entry and exit of an automated warehouse based on 5G technology according to claim 2, characterized in that, Parallel decoding of the fused feature representation is performed to generate spatial attention maps of the material identification region and the pallet identification region, respectively, including: The fused feature representations are simultaneously input into the material identification decoding channel and the pallet identification decoding channel for parallel processing. A cross-attention mechanism is established between the two decoding channels, so that the material identification decoding process can refer to the feature information of the pallet area, and the pallet identification decoding process can refer to the feature information of the material area. Based on the output of the cross-attention mechanism, spatial attention maps of the material identification area and the pallet identification area are generated respectively. The two spatial attention maps are superimposed and fused, and the conflict in the overlapping area is resolved to generate the final dual-target spatial attention map.

9. The method for secondary verification of entry and exit of an automated warehouse based on 5G technology according to claim 2, characterized in that, The joint optimization recognition strategy involves inputting the material identification sub-image and the pallet identification sub-image into a shared feature extraction backbone network, and simultaneously recognizing them through the material identification branch and the pallet identification branch, respectively, to obtain the material identification and pallet identification, including: The material identification sub-image and the pallet identification image are input into the shared feature extraction backbone network to extract the shared feature map; The shared feature map is input into the material identification branch and the pallet identification branch for synchronous identification, and a two-way information interaction mechanism is established between the two branches during the identification process to obtain their respective output results; Based on the output results of the two branches, the final recognition results of the material identifier and pallet identifier are obtained by decoding.

10. A secondary verification system for inbound and outbound operations of an automated warehouse based on 5G technology, characterized in that, The system is used for secondary verification of automated warehouses. The device includes an image acquisition device and a 5G communication module installed on a stacker crane, a warehouse management module, and an alarm module. The system includes: The acquisition module is used to acquire the storage location information for stacker crane inbound and outbound operations in real time. The image module is used to trigger the image acquisition device to acquire an image of the goods in the storage location when the stacker crane completes the storage location operation; The extraction module is used to extract material identification and pallet identification based on the cargo image; The storage module is used to store the material identifier, the pallet identifier, and the corresponding cargo image to a local database, and to update the inventory records; The comparison module is used to compare the updated inventory records in the local database with the corresponding inventory records in the warehouse management module in real time to obtain the comparison results. An alarm module is used to trigger an alarm when the comparison results are inconsistent.

Images