Storage safety management method and related device

Abstract

The invention belongs to the field of data processing, and particularly relates to a storage safety management method and related device.The method comprises the steps that in response to an access application instruction of a to-be-recognized object, real-time image information and a radio frequency identity label of the to-be-recognized object are obtained; performing identity recognition on the multi-angle figure image through a dynamic portrait recognition model to obtain first to-be-verified identity information of the to-be-recognized object; performing identification processing on the target equipment image through an equipment identification model to obtain target equipment information; on the basis of the first to-be-verified identity information, second to-be-verified identity information indicated by the radio frequency identity identifier and the target equipment information, verifying whether the to-be-identified object has the access authority of carrying the target equipment to go in and out of the target storage area; and if the to-be-identified object has the access permission, issuing access permission to the target storage area to the to-be-identified object. According to the method, the to-be-identified object can be automatically verified, the storage safety management efficiency is improved, and the storage safety is improved.

Description

Technical Field

[0001] This application belongs to the field of data processing, and in particular relates to a warehouse safety management method and related apparatus. Background Technology

[0002] Warehouse safety is one of the fundamental issues in warehouse management. Traditional warehouse management often relies on manual operation, which is easily affected by human factors and errors.

[0003] In related technologies, due to the large number of equipment in warehouses, manual entry and monitoring of equipment status is prone to errors, omissions, or oversights, leading to safety issues such as unclear equipment locations, lost equipment, or incorrect loading. Furthermore, traditional warehouse management often fails to monitor the location and status of equipment within the warehouse, making it difficult to promptly detect anomalies and take appropriate measures. It also lacks an effective tracking mechanism, making it difficult to immediately trace or recover lost or misplaced equipment. In summary, traditional management methods suffer from low efficiency and difficulty in monitoring.

[0004] Therefore, in order to improve warehouse security, there is an urgent need to propose a brand-new warehouse security management solution. Summary of the Invention

[0005] This application provides a warehouse security management method and related apparatus to achieve automated dual verification of the object to be identified in terms of both user identity and equipment possession status, thereby improving warehouse security management efficiency and enhancing warehouse security.

[0006] In a first aspect, this application provides a warehouse security management method, the method comprising:

[0007] In response to an access request instruction from an object to be identified to a target warehouse area, real-time image information and radio frequency identification (RFID) of the object to be identified are acquired; the real-time image information includes at least: multi-angle images of the object to be identified and images of the target equipment carried by the object to be identified; the real-time image information is acquired from multiple image acquisition devices with different perspectives pre-deployed within the target warehouse area; the RFID is acquired from a high-frequency radio frequency antenna pre-deployed within the target warehouse area;

[0008] The dynamic human image recognition model is used to identify the multi-angle human images to obtain the first identity information of the object to be identified.

[0009] By using an equipment recognition model, the target equipment image is processed to obtain the target equipment information carried by the object to be identified.

[0010] Based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information, verify whether the object to be identified has the access permission to carry the target equipment into and out of the target storage area.

[0011] If the object to be identified has the access permission, then an access permission for the target storage area is issued to the object to be identified.

[0012] Secondly, embodiments of this application provide a warehouse safety management device, which includes:

[0013] The acquisition unit is configured to acquire real-time image information and radio frequency identification (RFID) of the object to be identified in response to an access request instruction from the object to be identified to the target storage area. The real-time image information includes at least: multi-angle images of the object to be identified and images of the target equipment carried by the object to be identified. The real-time image information is acquired from multiple image acquisition devices with different perspectives pre-deployed within the target storage area. The RFID is acquired from a high-frequency radio frequency antenna pre-deployed within the target storage area.

[0014] The identity recognition unit is configured to perform identity recognition on the multi-angle human image through a dynamic human image recognition model to obtain the first identity information to be verified of the object to be identified.

[0015] The equipment identification unit is configured to perform identification processing on the target equipment image through an equipment identification model to obtain target equipment information carried by the object to be identified.

[0016] The verification unit is configured to verify whether the object to be identified has the access permission to carry the target equipment into and out of the target storage area based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information.

[0017] The authorization unit is configured to issue an access permission to the target storage area to the target object if the object to be identified has the access permission.

[0018] Thirdly, embodiments of this application provide a computing device, the computing device comprising:

[0019] At least one processor, memory, and input / output unit;

[0020] The memory is used to store computer programs, and the processor is used to call the computer programs stored in the memory to execute the warehouse security management method of the first aspect.

[0021] Fourthly, a computer-readable storage medium is provided, comprising instructions that, when executed on a computer, cause the computer to perform the warehouse security management method of the first aspect.

[0022] In the technical solution provided in this application embodiment, firstly, in response to the access request instruction from the object to be identified to the target storage area, real-time image information and radio frequency identification (RFID) of the object to be identified are acquired. The real-time image information includes at least: multi-angle images of the object to be identified and images of the target equipment carried by the object. The real-time image information is collected from multiple pre-deployed image acquisition devices with different perspectives within the target storage area, and the RFID is collected from a pre-deployed high-frequency radio frequency antenna within the target storage area. Then, through a dynamic facial recognition model, the multi-angle images of the object are used for identity recognition to obtain the first identity information to be verified for the object. Next, through an equipment recognition model, the image of the target equipment is processed to obtain the information of the target equipment carried by the object. Finally, based on the first identity information to be verified, the second identity information to be verified indicated by the RFID, and the target equipment information, it is verified whether the object to be identified has access permission to enter and exit the target storage area with the target equipment. If the object to be identified has access permission, an access permit for the target storage area is issued to the object.

[0023] This technical solution utilizes dynamic facial recognition and equipment recognition models to automatically identify the identity of the target object and the information of the target equipment it carries. This helps solve warehouse management problems such as cumbersome data entry processes, redundant security checks at entrances and exits, and the risk of missed detections in traditional control methods. It further enhances the supervision of entry and exit information, improves warehouse management efficiency, and ensures warehouse security. Simultaneously, by verifying the recognition information output by the two models, it also achieves automated dual verification of the target object's user identity and equipment possession status, further improving warehouse security management efficiency and ensuring warehouse security. Attached Figure Description

[0024] The accompanying drawings, which are included to provide a further understanding of this application and form part of this application, illustrate exemplary embodiments and are used to explain this application, but do not constitute an undue limitation of this application. In the drawings:

[0025] Figure 1 This is a flowchart illustrating a warehouse safety management method according to an embodiment of this application;

[0026] Figure 2 This is a schematic diagram illustrating the principle of a dynamic human face recognition model construction method according to an embodiment of this application;

[0027] Figure 3This is a schematic diagram illustrating the principle of an equipment identification model according to an embodiment of this application;

[0028] Figure 4 This is a schematic diagram illustrating the principle of a method for verifying equipment license permissions according to an embodiment of this application;

[0029] Figure 5 This is a schematic diagram illustrating the principle of a device state matching method according to an embodiment of this application;

[0030] Figure 6 This is a schematic diagram of the structure of a warehouse safety management device according to an embodiment of this application;

[0031] Figure 7 This is a schematic diagram of the structure of an electronic device according to an embodiment of this application. Detailed Implementation

[0032] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0033] Unless otherwise defined, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this application belongs. The terminology used herein is for the purpose of describing particular embodiments only and is not intended to be limiting of the application.

[0034] Warehouse safety is one of the fundamental issues in warehouse management. Traditional warehouse management often relies on manual operation, which is easily affected by human factors and errors.

[0035] In related technologies, due to the large number of equipment in warehouses, manual entry and monitoring of equipment status is prone to errors, omissions, or oversights, leading to safety issues such as unclear equipment locations, lost equipment, or incorrect loading. Furthermore, traditional warehouse management often fails to monitor the location and status of equipment within the warehouse, making it difficult to promptly detect anomalies and take appropriate measures. It also lacks an effective tracking mechanism, making it difficult to immediately trace or recover lost or misplaced equipment. In summary, traditional management methods suffer from low efficiency and difficulty in monitoring.

[0036] Therefore, in order to improve warehouse security, there is an urgent need to propose a brand-new warehouse security management solution.

[0037] To address at least one of the aforementioned technical problems, embodiments of this application provide a warehouse safety management method and related apparatus.

[0038] Specifically, in the warehouse security management solution, firstly, in response to the access request instruction from the object to be identified for the target warehouse area, real-time image information and RFID tags of the object to be identified are acquired. This real-time image information includes at least: multi-angle images of the object to be identified and images of the target equipment carried by the object. This real-time image information is collected from multiple pre-deployed image acquisition devices with different perspectives within the target warehouse area, and the RFID tags are collected from pre-deployed high-frequency radio frequency antennas within the target warehouse area. Then, using a dynamic facial recognition model, the multi-angle images of the object are used for identification to obtain the first identity information to be verified. Next, using an equipment recognition model, the image of the target equipment is processed to obtain the information of the target equipment carried by the object. Finally, based on the first identity information to be verified, the second identity information to be verified indicated by the RFID tags, and the target equipment information, it is verified whether the object to be identified has access permission to enter and exit the target warehouse area with the target equipment. If the object to be identified has access permission, an access permit for the target warehouse area is issued to the object.

[0039] This warehouse security management solution utilizes dynamic facial recognition and equipment recognition models to automatically identify the identity of the target and the equipment they are carrying. This helps solve warehouse management problems such as cumbersome data entry processes, redundant entrance and exit security checks, and the risk of missed detections, which are inherent in traditional control methods. It further enhances the supervision of entry and exit information, improves warehouse management efficiency, and ensures warehouse security. Simultaneously, by verifying the recognition information output by the two models, it also achieves automated dual verification of the target's identity and equipment possession status, further improving warehouse security management efficiency and ensuring warehouse security.

[0040] The warehouse security management solution provided in this application can be executed by an electronic device, which can be a server, server cluster, or cloud server. The electronic device can also be a terminal device such as a mobile phone, computer, tablet computer, wearable device, or dedicated device (such as a dedicated terminal device with a warehouse security management system). In an optional embodiment, the electronic device can be equipped with a service program for executing the warehouse security management solution.

[0041] Figure 1 This is a schematic diagram of a warehouse safety management method provided in an embodiment of this application, as shown below. Figure 1 As shown, the method includes the following steps:

[0042] 101, in response to the access request instruction from the object to be identified to the target storage area, acquires the real-time image information and radio frequency identification of the object to be identified.

[0043] In this embodiment, the object to be identified refers to an individual or group of people applying to enter the target storage area. Specifically, the object to be identified can be a person or a group of people. For example, the object to be identified can be personnel who need to enter the target storage area, such as warehouse staff, specific visitors, or other authorized personnel. The object to be identified can also be a vehicle or a convoy carrying target equipment. The target storage area can be a warehouse, such as an armory, ammunition depot, parts warehouse, or spare parts material warehouse.

[0044] In this embodiment, the object to be identified first applies to enter the target storage area and sends an entry / exit application instruction. Upon responding to this instruction, step 101 can acquire the real-time image information and radio frequency identification (RFID) of the object to be identified for identity verification and equipment identification. Therefore, by verifying the identity and equipment information of the object to be identified, it can be determined whether the object has the authority to carry the target equipment into or out of the target storage area.

[0045] The real-time image information of the object to be identified includes at least: multi-angle images of the object and images of the target equipment carried by the object. Multi-angle images refer to images of a person taken from different angles. By acquiring multi-angle images, a more comprehensive understanding of the person's facial features, posture, and movement trajectory can be obtained.

[0046] The real-time image information is acquired from multiple pre-deployed image acquisition devices with different perspectives within the target storage area. For example, cameras can be fixedly installed at key locations within the storage area, such as entrances, passageways, and storage areas. These cameras can provide fixed-perspective images of people and equipment. For instance, pan-tilt cameras can be installed in corners to allow remote control or automated programs to control the tilt angle and direction of the pan-tilt, capturing images of people and equipment from different angles and positions. Alternatively, intelligent surveillance cameras equipped with automatic identification of people and target equipment can be used to capture their images. Such cameras can automatically track and capture the target object to obtain images of people and equipment from multiple different angles. This multi-angle camera approach ensures comprehensive and accurate real-time image information for subsequent identification and equipment recognition processing. Further optionally, if the target object is a vehicle carrying the target equipment, the real-time image information of the target object can include multi-angle vehicle images. These multi-angle vehicle images should at least include identification information that can be used to identify the vehicle, such as vehicle number and license plate.

[0047] The radio frequency identification (RFID) data is collected from high-frequency radio frequency antennas pre-deployed in the target storage area. For example, the RFID data can be constructed using Radio Frequency Identification (RFID) technology. RFID technology is a technology that uses radio signals to automatically identify target objects. In warehouse management scenarios, RFID tags or RFID chips can be embedded in items carried by the object to be identified for subsequent identification. To enhance the monitoring of the storage status of target equipment, RFID tags or RFID chips can also be embedded in the target equipment to facilitate real-time acquisition of the target equipment's location and movement trajectory, overcoming the technical problem of difficulty in tracing the target equipment in related technologies.

[0048] In practical applications, high-frequency radio frequency (RF) antennas are pre-deployed in the target storage area, covering specific areas. When an object to be identified enters the coverage area of ​​the RF antenna, the RF signal triggers the RFID tag or chip, causing it to actively send an RF signal containing unique identification information. The advantage of RFID identification lies in its ability to achieve long-distance, contactless identification. When an object passes the RF antenna, the RF signal can instantly identify and acquire the object's identity information. This automatic identification method, requiring no manual intervention, improves the accuracy and efficiency of identification. By combining real-time image information and RFID identification, the warehouse management system can more accurately verify whether an object has the authority to enter or exit the target storage area. This integrated approach provides more reliable identity verification and access control, enhancing the security and management efficiency of the storage area.

[0049] In 102, a dynamic human image recognition model is used to identify the identity of a person from multiple angles in order to obtain the first identity information to be verified of the person to be identified.

[0050] In this embodiment, the dynamic facial recognition model refers to a model capable of identifying individuals from multi-angle images. The dynamic facial recognition model is a key technology; it effectively identifies human images from multiple angles and provides accurate identity information for subsequent identity verification and equipment verification. This model exhibits good adaptability and robustness in processing multi-angle human images of different individuals under various warehouse management scenarios.

[0051] As an optional embodiment, it is assumed that the dynamic facial recognition model includes at least the following structure: a first feature extraction layer, an angle-assisted classification layer, and an identity recognition layer.

[0052] Based on the above structure, in step 102, a dynamic facial recognition model is used to identify individuals from multi-angle images to obtain the first identity information to be verified for the object to be identified, such as... Figure 2 As shown, this can be achieved through the following steps:

[0053] 201. Through the first feature extraction layer, the feature vector of the person image is extracted from the multi-angle person image.

[0054] The image feature vector for a person includes at least three components: a face feature vector, a pose feature vector, and a behavioral inertia feature vector. For example, the face feature vector extracts facial features, which may include information such as the face's outline, eyes, and nose. The pose feature vector captures the body posture features of a person from different angles. The behavioral inertia feature vector reflects the person's movement patterns and dynamic characteristics.

[0055] For example, in step 201, a sequence of images of people from multiple angles is input into the first feature extraction layer, including images of the object to be identified from the front, side, and oblique overhead angles. For example, the input image sequence could be a video containing a person walking.

[0056] First, each frame of the image is preprocessed by scaling or cropping it to a fixed size. For example, each frame can be scaled to 224x224. Next, feature extraction is performed on the scaled images. Each frame is input into a convolutional layer to extract feature maps for multiple channels. These feature maps can represent different patterns and structural information in the image. For each frame, all feature map channels are concatenated to form a unified feature vector. For example, if a 224x224 image has 512 feature map channels, each 14x14, then all channels can be concatenated to form a 1x100352 feature vector. This process is repeated for each frame in the entire image sequence to extract the corresponding feature vector sequence. Finally, all feature vectors are used as the output of the first feature extraction layer for processing and analysis in subsequent steps, including the angle-assisted classification layer and the identity recognition layer.

[0057] 202. By using an angle-assisted classification layer, the feature vector of the person image is mapped to a spherical feature space to obtain a high-dimensional feature vector of the person image.

[0058] 203. By using the angle-assisted classification layer, the cosine distance metric similarity of the high-dimensional feature vector of the person image is obtained, and the cosine distance metric similarity is used to correct the high-dimensional feature vector of the person image, resulting in the optimized high-dimensional feature vector of the person image.

[0059] For example, suppose we use a dataset containing images of people for training, which includes images from different angles such as front, side, and back.

[0060] Based on this, step 202 first involves spherical mapping of the feature vectors of the person's image. Here, the spherical feature space is a special high-dimensional space with a spherical topological structure, which can better preserve the geometric relationships between features. By mapping the feature vectors of the person's image to the spherical feature space, the features of the person can be better expressed and distinguished from different angles, enhancing the differences between feature vectors from different angles and improving the accuracy of identity recognition.

[0061] Further suppose we have a feature vector for a person's image with the range [0.8, 0.3, 0.6], which contains the person's features at different angles. This feature vector can be mapped to a spherical feature space using spherical mapping.

[0062] In practical applications, as an optional embodiment, the cosine distance metric for similarity obtained in the angle-assisted classification layer is represented by the following expression:

[0063] s i =f(i)(ma·||ω i ||)cos(m·θ i +b)

[0064] Among them, s i Let f(i) represent the cosine distance metric for the similarity of the feature vectors of the i-th person's image, m represent the cosine value amplification factor, a represent the cosine value reduction factor, and b represent the cosine value balancing adjustment factor. m, a, and b are all adjustable hyperparameters. ω i Let θ represent the weight vector of the feature vector of the i-th person's image. i Let represent the angle between the feature vector and the weight vector of the i-th person's image.

[0065] In step 203, the high-dimensional feature vector of the person image will be corrected by measuring similarity using cosine distance. Cosine similarity is a similarity index that measures the angle between vectors and can convey the directional information of the feature vectors.

[0066] In another example, suppose we have two high-dimensional feature vectors of a person's image: A = [0.8, 0.3, 0.6] and B = [0.9, 0.2, 0.4]. We can calculate the cosine distance between these two feature vectors to obtain their cosine similarity. Let's assume the result is d = 0.1, where d is the cosine distance between the two feature vectors. By using cosine distance to measure similarity correction, feature vector A can be corrected. The corrected feature vector A' can be obtained by adjusting the value of A according to the cosine distance d, for example, A' = [0.8 - 0.1, 0.3 - 0.1, 0.6 - 0.1] = [0.7, 0.2, 0.5]. Finally, we obtain the optimized high-dimensional feature vector of the person's image: A' = [0.7, 0.2, 0.5]. This corrected feature vector can more accurately represent and distinguish the features of a person from different angles.

[0067] As demonstrated by the examples above, an angle-assisted classification layer can map the feature vectors of a person's image onto a spherical feature space. Furthermore, cosine distance is used to measure similarity and refine the feature vectors, resulting in an optimized high-dimensional feature vector for the person's image. This approach enhances the difference between feature vectors from different angles, improving the accuracy of identity recognition.

[0068] 204. Through the identity recognition layer, the optimized high-dimensional feature vector of the person image is input into the identity classifier to obtain the identity classification prediction distribution value of the object to be identified.

[0069] The identity recognition layer is one of the core layers in the model. It receives the output of the angle-assisted classification layer and inputs it into the identity classifier. The identity recognition layer can be a classification neural network model that matches optimized feature vectors with pre-trained identity labels to predict the identity of the object to be identified. The purpose of this layer is to determine the identity of a person through recognition and classification. For example, the identity classifier can be implemented as a classification neural network model that matches optimized feature vectors with pre-trained identity labels to predict the identity of the object to be identified.

[0070] Suppose there is an authorized user database containing multiple authorized users, each with a unique identity label. The image of the person to be identified is processed to obtain an optimized high-dimensional feature vector, which is then input into the identity classifier.

[0071] For example, in step 204, the optimized high-dimensional feature vector of the person image is input into the identity recognition layer to obtain an identity classification prediction distribution value. Assume that the predicted distribution value output by the model is [0.2, 0.7, 0.1], where the first component represents the probability that the object to be identified belongs to the first authorized user, the second component represents the probability that the object to be identified belongs to the second authorized user, and so on.

[0072] 205. Through the identity recognition layer, based on the identity classification prediction distribution value, authorized user information that meets the preset conditions is selected from the pre-entered authorized user database as the first identity information to be verified.

[0073] This step also involves setting thresholds and formulating decision-making strategies to ensure the accuracy and security of identity verification. For example, a threshold of 0.5 can be set, meaning that the identity information of the current target object can only be matched with the authorized user database when the highest probability of the identity classification prediction is greater than this threshold. In this case, the value of the second component in the identity classification prediction distribution, 0.7, is clearly higher than the threshold of 0.5, so the identity of the target object can be considered to match the second authorized user in the authorized user database. Finally, detailed identity information of the second authorized user can be obtained from the authorized user database, such as name, photo, affiliated unit, position, and type of equipment that can be held, and this can be regarded as the first identity information to be verified. With this identity information, subsequent identity verification and authorized access operations can be performed.

[0074] As can be seen from the above examples, the optimized high-dimensional feature vector of the person image can be input into the identity classifier through the identity recognition layer to obtain the identity classification prediction distribution value of the object to be identified. By selecting authorized user information that meets the preset conditions from the authorized user database as the first identity information to be verified, the identity recognition and verification of the object to be identified can be realized.

[0075] In section 103, the target equipment image is processed by the equipment recognition model to obtain the target equipment information carried by the object to be identified.

[0076] In this embodiment, the equipment identification model can automatically identify and classify target equipment without manual intervention, greatly improving equipment identification efficiency and further enhancing warehouse management efficiency. As an optional embodiment, it is assumed that the equipment identification model includes at least the following structure: a second feature extraction layer, a detection and localization layer, a classification and prediction layer, and an output layer.

[0077] Based on the above structure, in step 103, the target equipment image is processed using an equipment recognition model to obtain the target equipment information carried by the object to be identified, such as... Figure 3 As shown, this can be achieved through the following steps:

[0078] 301. Through the second feature extraction layer, the equipment feature vector is extracted from the target equipment image;

[0079] 302. Through the detection and localization layer, the equipment feature vector is input into multiple detection and localization units to obtain multiple localization boxes, and the positions of the multiple localization boxes are marked; each detection and localization unit consists of at least multiple convolutional layers and fully connected layers; the image size of different localization boxes is different;

[0080] 303. Through the classification prediction layer, bounding box coordinate regression is performed on multiple localization boxes to correct their positions.

[0081] 304. Through the classification prediction layer, multiple corrected bounding boxes are predicted to obtain the class probability values ​​of multiple bounding boxes.

[0082] 305. Through the output layer, non-maximum suppression is used to select the target bounding box with the highest category probability value from multiple bounding boxes.

[0083] 306. Through the output layer, obtain the target equipment information that matches the target positioning box from the equipment status management library.

[0084] For example, suppose we are performing target equipment identification on an image. Further suppose the image contains a handgun. Based on this assumption, in step 301, the target equipment image containing the handgun is transformed into a corresponding target equipment image through convolution using the second feature extraction layer, which is used for subsequent equipment localization and classification. In step 302, the equipment feature vector is input into multiple detection and localization units of the detection and localization layer, resulting in multiple initial localization boxes of different sizes and proportions. These localization boxes are used to locate and mark the position of the equipment in the image. Further, optionally, in practical applications, the detection and localization units can employ a multi-box detection algorithm.

[0085] In section 303, for each initial bounding box, a bounding box regression algorithm is used to correct its position. Specifically, the position of the initial bounding box is adjusted based on the correction vector (such as offset) predicted by the classification prediction layer. It can be understood that these correction vectors are learned during the training process of the classification prediction layer and can be adjusted according to the requirements of the specific task. After adjusting each initial bounding box using bounding box regression, these corrected bounding boxes will more accurately delineate the location of the target equipment. To further improve the accuracy of these bounding boxes, filtering strategies can be applied, such as deleting bounding boxes with low confidence or filtering bounding boxes according to certain criteria. The classification prediction layer can correct the position of the bounding boxes through bounding box coordinate regression to improve the quality of the equipment image contained within the bounding box. This provides more accurate images for subsequent equipment classification, improving the overall equipment recognition accuracy.

[0086] In step 304, the corrected bounding boxes are input into the classification network in the classification prediction layer. This classification network is a PyramidNet with Bottleneck Residual Units (SE), trained to classify different categories of equipment. The model employs a pyramid structure, including at least two SEs: Bottleneck Residual Units and Squeeze and Excite Units. This pyramid-shaped structure uses dense connections to distribute and converge information at different levels. The idea behind the pyramid structure is to converge information from deep feature maps across different branches, guiding information transmission end-to-end and improving feature reusability. The Bottleneck Residual Units consist of three convolutional layers: a 1x1 downsampled convolutional layer, a 3x3 convolutional layer, and a 1x1 pointwise convolutional layer. Using this unit reduces model parameters, increases model depth, and improves accuracy. Squeeze and Excite Units (SEs) are loaded within each PyramidNet model to enhance its discriminative power. In the compression phase, feature vectors are generated through 1x1 convolutional layers. In the activation phase, the feature vectors are sigmoid activated and element-wise multiplied to re-fuse them, strengthening the important feature layers of the model. Through forward propagation of the pyramid network model, probability values ​​for each bounding box belonging to different categories are obtained. These probability values ​​represent the likelihood of each bounding box belonging to a different equipment category. Simultaneously, the confidence score for each predicted category can be calculated, representing the model's level of confidence in the predicted probability value. In summary, the classification prediction layer can predict the category of multiple corrected bounding boxes, determine the equipment category of each bounding box, and provide specific classification information and confidence scores for each bounding box. This provides accurate target equipment classification data for subsequent applications and decision-making.

[0087] In step 305, non-maximum suppression (NMS) is employed to select the target bounding box with the highest class probability value from multiple bounding boxes via the output layer. Specifically, NMS selects the target bounding box with the highest class probability value or a class probability value within a preset position from multiple bounding boxes. Selecting a single target bounding box from multiple bounding boxes reduces false positives and improves accuracy. Furthermore, bounding boxes of different classes are processed separately, and the optimal bounding box is selected by calculating overlap. This process significantly reduces the number of bounding boxes and selects the bounding box with the class probability value that meets the criteria as the final target bounding box.

[0088] In section 306, target equipment information corresponding to a target location bounding box is extracted by matching it with the equipment status management database. First, the target area corresponding to the bounding box is cropped from the target equipment image and adjusted to an appropriate size and resolution. Then, the cropped bounding box is matched against information in the equipment status management database, and the relevant attribute information (i.e., target equipment information) is extracted from the matching query results, such as equipment name, model, color, size, etc. Through these steps, the target equipment information in section 306 can provide accurate target equipment information for subsequent applications and decision-making.

[0089] The steps described in sections 301 to 306 enable automated identification of target equipment, achieving automatic detection and classification of targets, avoiding manual intervention, and improving the efficiency of the equipment identification process. Specifically, feature extraction and classification of target equipment accurately identify different categories of targets and output target confidence information. The bounding box generation and non-maximum suppression methods in steps 303 to 305 accurately locate target positions, suppress false alarms and misjudgments, and improve identification accuracy. These steps can process a large number of target images in a short time, easily extracting detailed information about target equipment, especially useful for equipment identification scenarios that require rapid processing of large numbers of images. Finally, by matching and querying the equipment status management database in step 306, more detailed information about the specific equipment carried by the target object can be extracted, which is an important means of achieving refined target management and monitoring.

[0090] In summary, the target identification system consisting of steps 301 to 306 is characterized by high efficiency, accuracy, scalability, automation, and intelligence. It enables rapid, accurate, and precise identification and management of targets, thereby improving security in fields such as security and warehouse management.

[0091] In step 104, based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information, it is verified whether the object to be identified has the access permission to carry the target equipment into and out of the target storage area.

[0092] In step 105, if the object to be identified has access permissions, then an access permit for the target storage area is issued to the object to be identified.

[0093] In this embodiment of the application, access permissions include at least: user access permissions and equipment permission.

[0094] It is understood that, in this embodiment, user access permissions refer to granting access to specific users or a certain type of users, enabling them to enter and exit the target storage area and retrieve or manage the target equipment. Only personnel with the corresponding identity and permissions can enter the target storage area; otherwise, they will be denied entry.

[0095] Equipment licensing refers to the authorization granted to a specific user or type of user to carry or use a particular piece of equipment into or out of a specific area or location. In this embodiment, equipment licensing refers to the authorization granted to a specific user or type of user to carry this specific equipment into or out of the target storage area. Only personnel with equipment licensing authority can bring the target equipment into the target storage area; otherwise, entry will be denied. This ensures that only authorized personnel can bring the target equipment into the target storage area, thus ensuring the secure management of the target equipment.

[0096] Specifically, step 104 verifies whether the object to be identified has access permissions to enter and exit the target storage area with the target equipment, such as... Figure 4 As shown, this can be achieved through the following steps:

[0097] 401. Verify whether the first identity information to be verified is consistent with the second identity information to be verified indicated by the radio frequency identification.

[0098] 402. If the first identity information to be verified is consistent with the second identity information to be verified, then it is determined that the object to be identified is consistent with the user identity pre-bound to the radio frequency identification.

[0099] Specifically, prior to step 401, a binding relationship is pre-established between the RFID tag and the user's identity information. Therefore, step 401 can query the second identity information to be verified indicated by the RFID tag. Further assuming the second identity information to be verified indicated by the RFID tag is "Alice," step 401 checks whether the first identity information to be verified is "Alice." If so, it is determined that the first identity information to be verified matches the second identity information to be verified indicated by the RFID tag. In this case, the identity of the person to be identified can be confirmed as "Alice."

[0100] If the first identity information to be verified is not "Alice," it indicates that the first identity information to be verified is inconsistent with the second identity information to be verified indicated by the RFID tag. In this case, it can be confirmed that the identity of the object to be identified obtained through image recognition is inconsistent with the identity indicated by the RFID tag, posing a certain security risk. At this time, a security alarm can be remotely triggered, or a re-acquisition of multi-view images of the object to be identified can be triggered.

[0101] 403, Verify whether the object to be identified has user access permissions to the target warehouse area.

[0102] Based on the above example, and using the identity information query permission management system obtained from the above steps, confirm whether "Alice" has access permission to enter the target warehouse area.

[0103] 404. If the object to be identified has user access rights to the target storage area, then based on the target equipment information, verify whether the object to be identified has equipment permission to carry the target equipment into or out of the target storage area.

[0104] Assume the verification result from the access control system indicates that "Alice" has access to the target storage area. Then, query the target equipment management system to confirm whether "Alice" has the equipment permit permission to carry the target equipment into and out of the target storage area. The system verification result will then confirm that "Alice" has the equipment permit permission to carry the target equipment. Here, the equipment permit permission for the target equipment can be registered at the requisition site or applied for and authorized remotely via an app; this application is not limited to one method.

[0105] Optionally, in 404, based on the target equipment information, it is verified whether the object to be identified has the equipment permission to carry the target equipment into or out of the target storage area, such as... Figure 5 As shown, the implementation involves the following steps:

[0106] 501. Using the equipment search model, retrieve the target equipment object from the equipment status management library that matches the real-time image of the target equipment in the target equipment information.

[0107] As an alternative implementation, the equipment search model can be implemented using a region-based convolutional neural network (Faster R-CNN). Faster R-CNN is an object detection algorithm that can simultaneously locate and classify targets. By extracting candidate regions from real-time images of the target equipment and performing classification and bounding box regression on each region, the target equipment can be accurately located.

[0108] Alternatively, the equipment search model can be implemented using a computer vision algorithm (You Only Look Once, YOLO). YOLO is a real-time object detection algorithm that uses a single neural network to simultaneously predict the class and bounding boxes of multiple objects. YOLO is characterized by its speed and ability to accurately detect small objects.

[0109] 502. If the target equipment is in an authorized holding state, then obtain the equipment status information of the target equipment.

[0110] In this embodiment, the equipment status information includes at least: the user identification information of the authorized holder of the target equipment object, and pre-recorded historical images of the user. Equipment status information refers to relevant information about a specific piece of equipment, including but not limited to the equipment's location, status, attributes, affiliated unit / department, usage status, maintenance records, etc. This information helps users understand the equipment's status, such as whether it is usable, where it is used, and its maintenance status, enabling better management and utilization of the equipment. In the above embodiment, the equipment status information also includes the user identification information of the authorized holder of the target equipment object and pre-recorded historical images of the user. Further optionally, the equipment status information is updated in the equipment status management database when the user retrievals or borrows the target equipment object.

[0111] 503. Based on equipment status information, determine whether the object to be identified is authorized to hold the target equipment object.

[0112] Alternatively, prior to 503, when a user authorizes the receipt of a target equipment object, real-time images of the person from multiple perspectives can be acquired as the user's historical images by pre-deploying image acquisition devices at multiple different viewpoints.

[0113] Based on this, in section 503, a dynamic facial recognition model is used to assess the similarity between multi-angle images of a person and the user's historical images, resulting in a similarity score. Then, it is determined whether the radio frequency identification (RFID) tag matches the user's identification information. If the similarity score meets preset conditions and the RFID tag matches the user's identification information, it is determined that the person to be identified is authorized to possess the target equipment.

[0114] For example, suppose an employee uses their employee ID card (containing an RFID tag) to claim a piece of weaponry (the target equipment). When applying to claim it, multiple real-time images of the employee from different angles are captured by image acquisition devices and uploaded as historical images. Alternatively, the employee can take a selfie and upload multiple videos or images as one of their application credentials. When the employee approaches the weaponry again, a dynamic facial recognition model assesses the similarity between the current employee's multi-angle images and the historical images. If the similarity score meets preset conditions, and the RFID tag matches the user's identity information, the system determines that the employee is authorized to possess the weaponry and allows them to perform the corresponding operations. In this way, through dynamic facial recognition and identity verification, it ensures that only employees authorized to possess the weaponry can use it.

[0115] 504. If it is determined that the object to be identified is authorized to hold the target equipment object, then it is determined that the object to be identified has the equipment license authority to carry the target equipment into and out of the target storage area.

[0116] For example, suppose a military base has a supply storage area containing important military equipment. To ensure the security of the supplies, only authorized personnel can bring the equipment into or leave the area. When a soldier (the person to be identified) is determined to be authorized to possess a specific piece of equipment (the target equipment), it means he is authorized to bring that equipment into the target storage area. If this soldier needs to retrieve the equipment from the storage area, the system will verify his identity and confirm his authorization relationship with the target equipment through the appropriate identity verification and equipment management system. If the system confirms that the soldier is authorized to possess the equipment, it will grant him equipment permission to bring the equipment into and out of the storage area.

[0117] In this way, only personnel authorized to possess the target equipment can enter the storage area, and they only have the authority to carry the equipment in and out when necessary, thus ensuring the safety and management of the materials.

[0118] In related technologies, traditional warehouse management often cannot monitor the location and status of equipment in the warehouse in real time, making it difficult to detect abnormalities in a timely manner and take corresponding measures. It also lacks an effective tracking mechanism, so once equipment is lost or omitted, it is often impossible to trace or retrieve it immediately.

[0119] To address the aforementioned issues, in this embodiment of the application, optionally, if the object to be identified lacks access permissions and is carrying target equipment, multiple pre-deployed image acquisition devices within the target storage area can be used to perform real-time image positioning and tracking of the object based on the signal strength value of the radio frequency identification (RFID) identifier, thereby obtaining a real-time location image of the object. Subsequently, the real-time location image is uploaded to the storage management platform, and the platform dynamically displays the real-time location information and movement trajectory of the object to be identified, thus achieving the location and traceability of the object.

[0120] For example, suppose an equipment management platform is deployed within a military base. Assume all equipment has radio frequency identification (RFID), and the military base is equipped with multiple image acquisition devices to capture real-time images of the base. The equipment management platform also has the function of updating the real-time images captured by these acquisition devices in real time.

[0121] Based on the above assumptions, a soldier (the target of identification) enters a storage area carrying a piece of equipment, but he does not have permission to enter or leave the area. Since the equipment carried by the target has a radio frequency identification (RFID) tag, the approximate location and movement trajectory of the target can be determined by the signal strength value. Thus, multiple pre-deployed image acquisition devices perform real-time image positioning and tracking of the target. Once the system determines the soldier's real-time location, it sends the real-time location information to the equipment management platform, updating the target's location and movement trajectory information in real time. On the equipment management platform, managers can dynamically display the soldier's real-time location and track and trace his movement trajectory. Through this real-time positioning and tracking technology, equipment managers can immediately detect the soldier's movement and any anomalies related to the soldier, and take timely measures and adjustments. This helps to further improve the accuracy and real-time performance of equipment management.

[0122] Optionally, if the object to be identified does not have access permissions and is carrying the target equipment, a remote alarm can be sent to the warehouse management platform to notify the object of its current risky behavior. If the object is located passing through the entrance / exit of the target warehouse area, a preset alarm matching the object's current risky behavior is triggered.

[0123] For example, suppose a military base has an equipment storage area, and all equipment within this area is equipped with RFID tags and location tracking devices. When an individual to be identified enters this area carrying equipment but lacks access rights, the system automatically performs real-time image location tracking. When the identified individual passes through the area's entrance or exit, the equipment management platform triggers a pre-set alarm matching the potentially risky behavior of that individual. For instance, if the individual lacks authorization to retrieve the equipment, or if they deviate from the prescribed time and route, the equipment management platform can issue a corresponding alarm message, such as an audio alarm when passing through a doorway.

[0124] In this way, equipment managers can promptly identify risky actions by entities to be identified and take corresponding measures. Simultaneously, pre-set alarms matching the risky behavior can quickly attract the attention and response of relevant personnel, enhancing the security and reliability of equipment management.

[0125] It should be noted that the techniques and methods in the embodiments of this application are only one implementation method, and can be appropriately modified and adjusted according to the needs of different scenarios to improve the security and reliability of equipment management.

[0126] In this embodiment, a dynamic facial recognition model and an equipment recognition model are used to automatically identify the identity of the object to be identified and the information of the target equipment it carries. This helps to solve warehouse management problems such as cumbersome data entry processes, redundant security checks at entrances and exits, and the risk of missed detections in traditional control methods. It further enhances the supervision of entry and exit information, improves warehouse management efficiency, and ensures warehouse security. Simultaneously, by verifying the recognition information output by the two models, automated dual verification of the object to be identified in terms of both user identity and equipment possession status is achieved, further improving warehouse security management efficiency and ensuring warehouse security.

[0127] In another embodiment of this application, a warehouse security management device is also provided, see [link to relevant documentation]. Figure 6 As shown, the device includes the following units:

[0128] The acquisition unit is configured to acquire real-time image information and radio frequency identification (RFID) of the object to be identified in response to an access request instruction from the object to be identified to the target storage area. The real-time image information includes at least: multi-angle images of the object to be identified and images of the target equipment carried by the object to be identified. The real-time image information is acquired from multiple image acquisition devices with different perspectives pre-deployed within the target storage area. The RFID is acquired from a high-frequency radio frequency antenna pre-deployed within the target storage area.

[0129] The identity recognition unit is configured to perform identity recognition on the multi-angle human image through a dynamic human image recognition model to obtain the first identity information to be verified of the object to be identified.

[0130] The equipment identification unit is configured to perform identification processing on the target equipment image through an equipment identification model to obtain target equipment information carried by the object to be identified.

[0131] The verification unit is configured to verify whether the object to be identified has the access permission to carry the target equipment into and out of the target storage area based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information.

[0132] The authorization unit is configured to issue an access permission to the target storage area to the target object if the object to be identified has the access permission.

[0133] Further optionally, the access permissions include at least: user access permissions and equipment permission.

[0134] The verification unit is specifically configured to: verify whether the first identity information to be verified is consistent with the second identity information to be verified indicated by the radio frequency identification; if the first identity information to be verified is consistent with the second identity information to be verified, then determine that the object to be identified is consistent with the user identity pre-bound to the radio frequency identification; verify whether the object to be identified has user access rights to the target storage area; if the object to be identified has user access rights to the target storage area, then verify whether the object to be identified has equipment permission to carry the target equipment into and out of the target storage area based on the target equipment information.

[0135] Further optionally, the verification unit verifies, based on the target equipment information, whether the object to be identified has the equipment permission to carry the target equipment into or out of the target storage area, specifically configured as follows:

[0136] The target equipment object that matches the real-time image of the target equipment in the target equipment information is obtained from the equipment status management library through the equipment search model.

[0137] If the target equipment is in an authorized holding state, the equipment status information of the target equipment is obtained; the equipment status information includes at least: the user identity information of the user authorized to hold the target equipment, and the user's pre-recorded historical image; the equipment status information is updated to the equipment status management database when the user takes or borrows the target equipment.

[0138] Based on the equipment status information, determine whether the object to be identified is authorized to hold the target equipment object;

[0139] If it is determined that the object to be identified is authorized to hold the target equipment object, then it is determined that the object to be identified has the equipment license authority to carry the target equipment into and out of the target storage area.

[0140] Further optionally, before the verification unit determines whether the object to be identified is authorized to possess the target equipment object based on the equipment status information, it is further configured to:

[0141] When a user authorizes to receive the target equipment, real-time images of the person from multiple perspectives are acquired using image acquisition devices pre-deployed from multiple different viewpoints and used as the user's historical images.

[0142] The verification unit, based on the equipment status information, determines whether the object to be identified is authorized to possess the target equipment object, and is specifically configured as follows:

[0143] A dynamic portrait matching model is used to evaluate the similarity between the multi-angle portrait images and the user's historical images, and a similarity evaluation score is obtained.

[0144] Determine whether the radio frequency identification identifier matches the user identification identifier information;

[0145] If the similarity assessment score meets the preset conditions, and the radio frequency identification is consistent with the user identification information, then it is determined that the object to be identified is authorized to hold the target equipment object.

[0146] Further optionally, the dynamic facial recognition model includes at least the following structure: a first feature extraction layer, an angle-assisted classification layer, and an identity recognition layer. The identity recognition unit is specifically configured as follows:

[0147] The first feature extraction layer extracts human image feature vectors from the multi-angle human images; the human image feature vectors include at least: face feature vectors, posture feature vectors, and behavioral inertia feature vectors.

[0148] The angle-assisted classification layer maps the feature vector of the person image to a spherical feature space to obtain a high-dimensional feature vector of the person image; the cosine distance similarity of the high-dimensional feature vector of the person image is obtained, and the cosine distance similarity is used to correct the high-dimensional feature vector of the person image to obtain an optimized high-dimensional feature vector of the person image.

[0149] The optimized high-dimensional feature vector of the person image is input into the identity classifier through the identity recognition layer to obtain the identity classification prediction distribution value of the object to be identified; based on the identity classification prediction distribution value, authorized user information that meets the preset conditions is selected from the pre-entered authorized user database as the first identity information to be verified.

[0150] Further, optionally, the cosine distance metric for similarity obtained in the angle-assisted classification layer is represented by the following expression:

[0151] s i =f(i)(ma·||ω i ||)cos(m·θ i +b)

[0152] Among them, s i Let f(i) represent the cosine distance metric for the similarity of the feature vectors of the i-th person's image, m represent the cosine value amplification factor, a represent the cosine value reduction factor, and b represent the cosine value balancing adjustment factor. m, a, and b are all adjustable hyperparameters. ω i Let θ represent the weight vector of the feature vector of the i-th person's image. i Let represent the angle between the feature vector and the weight vector of the i-th person's image.

[0153] Further optionally, the equipment identification model includes at least the following structure: a second feature extraction layer, a detection and localization layer, a classification and prediction layer, and an output layer. The equipment identification unit is specifically configured as follows:

[0154] The second feature extraction layer extracts equipment feature vectors from the target equipment image;

[0155] The detection and localization layer inputs the equipment feature vector into multiple detection and localization units to obtain multiple localization boxes and marks the positions of the multiple localization boxes; each detection and localization unit consists of at least multiple convolutional layers and fully connected layers; the image size of different localization boxes is different;

[0156] The classification prediction layer performs bounding box coordinate regression on multiple localization boxes to correct their positions; the corrected localization boxes are then predicted to obtain their class probability values.

[0157] Through the output layer, non-maximum suppression processing is used to select the target location box with the highest category probability value from multiple location boxes; the target equipment information matching the target location box is obtained from the equipment status management library.

[0158] Optionally, the device further includes a real-time tracking unit configured to: if the object to be identified does not have the access permission and the object to be identified is carrying target equipment, then, based on the signal strength value of the radio frequency identification, perform real-time image positioning and tracking of the object to be identified using multiple image acquisition devices pre-deployed within the target storage area to obtain a real-time positioning image of the object to be identified; upload the real-time positioning image to the storage management platform, and dynamically display the real-time location information and movement trajectory of the object to be identified on the storage management platform to achieve the location and traceability of the object to be identified.

[0159] Further optionally, the real-time tracking unit is also configured to:

[0160] If the object to be identified does not have the access permission and is carrying the target equipment, a remote alarm will be sent to the warehouse management platform to notify that the object to be identified is currently engaging in risky behavior.

[0161] If the object to be identified is located passing through the entrance or exit of the target storage area, a preset alarm matching the current risk behavior of the object to be identified is triggered.

[0162] In this embodiment, a warehouse security management device utilizes a dynamic facial recognition model and an equipment recognition model to automatically identify the identity of the object to be identified and the information of the target equipment it carries. This helps solve warehouse management problems such as cumbersome data entry processes, redundant entrance and exit security checks, and the risk of missed detections in traditional control methods. It further enhances the supervision of entry and exit information, improves warehouse management efficiency, and ensures warehouse security. Simultaneously, the warehouse security management device verifies the identification information output by the two models, achieving automated dual verification of the object to be identified in terms of both user identity and equipment possession status, further improving warehouse security management efficiency and ensuring warehouse security.

[0163] In another embodiment of this application, an electronic device is also provided, including: a processor, a communication interface, a memory, and a communication bus, wherein the processor, the communication interface, and the memory communicate with each other through the communication bus;

[0164] Memory, used to store computer programs;

[0165] The processor, when executing a program stored in memory, implements the warehouse security management method described in the method embodiments.

[0166] The communication bus 1140 mentioned in the above electronic device can be a Peripheral Component Interconnect (PCI) bus or an Extended Industry Standard Architecture (EISA) bus, etc. This communication bus 1140 can be divided into an address bus, a data bus, a control bus, etc.

[0167] For ease of representation, Figure 7 The bus is represented by a single thick line, but this does not mean that there is only one bus or one type of bus.

[0168] The communication interface 1120 is used for communication between the above-mentioned electronic device and other devices.

[0169] The memory 1130 may include random access memory (RAM) or non-volatile memory, such as at least one disk storage device. Optionally, the memory may also be at least one storage device located remotely from the aforementioned processor.

[0170] The processor 1110 mentioned above can be a general-purpose processor, including a central processing unit (CPU), a network processor (NP), etc.; it can also be a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0171] Accordingly, embodiments of this application also provide a computer-readable storage medium storing a computer program, which, when executed, can implement the steps that can be performed by an electronic device in the above method embodiments.

Claims

1. A warehouse safety management method, characterized in that, include: In response to the access request instruction of the object to be identified to the target storage area, the real-time image information and radio frequency identification of the object to be identified are obtained; The real-time image information includes at least: multi-angle images of the person to be identified and images of the target equipment carried by the person to be identified; the real-time image information is collected from multiple image acquisition devices with different perspectives pre-deployed in the target storage area; the radio frequency identification is collected from high-frequency radio frequency antennas pre-deployed in the target storage area; The dynamic human image recognition model is used to identify the multi-angle human images to obtain the first identity information of the object to be identified. By using an equipment recognition model, the target equipment image is processed to obtain the target equipment information carried by the object to be identified. Based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information, verify whether the object to be identified has the access permission to carry the target equipment into and out of the target storage area. If the object to be identified has the access permission, then an access permission for the target storage area is issued to the object to be identified.

2. The warehouse safety management method according to claim 1, characterized in that, The access permissions include at least: user access permissions and equipment permission; The step of verifying whether the object to be identified has access permission to enter and exit the target storage area with the target equipment, based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information, includes: Verify whether the first identity information to be verified is consistent with the second identity information to be verified indicated by the radio frequency identification; If the first identity information to be verified is consistent with the second identity information to be verified, then it is determined that the object to be identified is consistent with the user identity pre-bound to the radio frequency identification. Verify whether the object to be identified has user access permissions to the target warehouse area; If the object to be identified has user access rights to the target storage area, then based on the target equipment information, it is verified whether the object to be identified has equipment permission to carry the target equipment into and out of the target storage area.

3. The warehouse safety management method according to claim 2, characterized in that, The step of verifying whether the object to be identified has the equipment permission to carry the target equipment into or out of the target storage area based on the target equipment information includes: The target equipment object that matches the real-time image of the target equipment in the target equipment information is obtained from the equipment status management library through the equipment search model. If the target equipment is in an authorized holding state, the equipment status information of the target equipment is obtained; the equipment status information includes at least: the user identity information of the user authorized to hold the target equipment, and the user's pre-recorded historical image; the equipment status information is updated to the equipment status management database when the user takes or borrows the target equipment. Based on the equipment status information, it is determined whether the object to be identified is authorized to hold the target equipment object; If it is determined that the object to be identified is authorized to hold the target equipment object, then it is determined that the object to be identified has the equipment license authority to carry the target equipment into and out of the target storage area.

4. The warehouse safety management method according to claim 3, characterized in that, Before determining whether the object to be identified is authorized to possess the target equipment object based on the equipment status information, the method further includes: When a user authorizes to receive the target equipment, real-time images of the person from multiple perspectives are acquired using image acquisition devices pre-deployed from multiple different viewpoints and used as the user's historical images. The step of determining whether the object to be identified is authorized to possess the target equipment object based on the equipment status information includes: A dynamic portrait matching model is used to evaluate the similarity between the multi-angle portrait images and the user's historical images, and a similarity evaluation score is obtained. Determine whether the radio frequency identification identifier matches the user identification identifier information; If the similarity assessment score meets the preset conditions, and the radio frequency identification is consistent with the user identification information, then it is determined that the object to be identified is authorized to hold the target equipment object.

5. The warehouse safety management method according to claim 1, characterized in that, The dynamic human face recognition model includes at least the following structure: a first feature extraction layer, an angle-assisted classification layer, and an identity recognition layer; The step of using a dynamic facial recognition model to perform identity recognition on the multi-angle human images to obtain the first identity information to be verified of the object to be identified includes: The first feature extraction layer extracts human image feature vectors from the multi-angle human images; the human image feature vectors include at least: face feature vectors, posture feature vectors, and behavioral inertia feature vectors. The angle-assisted classification layer maps the feature vector of the person image to a spherical feature space to obtain a high-dimensional feature vector of the person image; the cosine distance similarity of the high-dimensional feature vector of the person image is obtained, and the cosine distance similarity is used to correct the high-dimensional feature vector of the person image to obtain an optimized high-dimensional feature vector of the person image. The optimized high-dimensional feature vector of the person image is input into the identity classifier through the identity recognition layer to obtain the identity classification prediction distribution value of the object to be identified; based on the identity classification prediction distribution value, authorized user information that meets the preset conditions is selected from the pre-entered authorized user database as the first identity information to be verified.

6. The warehouse safety management method according to claim 5, characterized in that, The similarity metric obtained in the angle-assisted classification layer is represented by the following expression: s i =f(i)(ma·||ω i ||)cos(m·θ i +b) Among them, s i Let f(i) represent the cosine distance metric for the similarity of the feature vectors of the i-th person's image, m represent the cosine value amplification factor, a represent the cosine value reduction factor, and b represent the cosine value balancing adjustment factor. m, a, and b are all adjustable hyperparameters. ω i Let θ represent the weight vector of the feature vector of the i-th person's image. i Let represent the angle between the feature vector and the weight vector of the i-th person's image.

7. The warehouse safety management method according to claim 1, characterized in that, The equipment identification model includes at least the following structure: a second feature extraction layer, a detection and localization layer, a classification and prediction layer, and an output layer; The step of performing recognition processing on the target equipment image through an equipment recognition model to obtain target equipment information carried by the object to be identified includes: The second feature extraction layer extracts equipment feature vectors from the target equipment image; The detection and localization layer inputs the equipment feature vector into multiple detection and localization units to obtain multiple localization boxes and marks the positions of the multiple localization boxes; each detection and localization unit consists of at least multiple convolutional layers and fully connected layers; the image size of different localization boxes is different; The classification prediction layer performs bounding box coordinate regression on multiple localization boxes to correct their positions; the corrected localization boxes are then predicted to obtain their class probability values. Through the output layer, non-maximum suppression processing is used to select the target location box with the highest category probability value from multiple location boxes; the target equipment information matching the target location box is obtained from the equipment status management library.

8. The warehouse safety management method according to claim 1, characterized in that, If the object to be identified does not have the access permission, and the object to be identified carries a target equipment object, then the method further includes: By using multiple image acquisition devices pre-deployed within the target storage area, and based on the signal strength value of the radio frequency identification, the object to be identified is located and tracked in real time to obtain a real-time location image of the object to be identified. The real-time positioning image is uploaded to the warehouse management platform, and the real-time location information and movement trajectory of the object to be identified are dynamically displayed on the warehouse management platform to achieve the location and traceability of the object to be identified.

9. The warehouse safety management method according to claim 8, characterized in that, If the object to be identified does not have the access permission, and the object to be identified carries a target equipment object, then the method further includes: Remotely alert the warehouse management platform to notify the identified object of current risky behavior; If the object to be identified is located passing through the entrance or exit of the target storage area, a preset alarm matching the current risk behavior of the object to be identified is triggered.

10. A warehouse safety management device, characterized in that, The device includes: The acquisition unit is configured to acquire real-time image information and radio frequency identification (RFID) of the object to be identified in response to an access request instruction from the object to be identified to the target storage area. The real-time image information includes at least: multi-angle images of the object to be identified and images of the target equipment carried by the object to be identified. The real-time image information is acquired from multiple image acquisition devices with different perspectives pre-deployed within the target storage area. The RFID is acquired from a high-frequency radio frequency antenna pre-deployed within the target storage area. The identity recognition unit is configured to perform identity recognition on the multi-angle human image through a dynamic human image recognition model to obtain the first identity information to be verified of the object to be identified. The equipment identification unit is configured to perform identification processing on the target equipment image through an equipment identification model to obtain target equipment information carried by the object to be identified. The verification unit is configured to verify, based on the first identity information to be verified, the second identity information to be verified indicated by the radio frequency identification, and the target equipment information, whether the object to be identified has the access permission to carry the target equipment into and out of the target storage area. The authorization unit is configured to issue an access permission to the target storage area to the target object if the object to be identified has the access permission.

Images