Article identification method and device, electronic equipment, medium and program product

Abstract

The invention provides an article identification method for warehouse-in and warehouse-out, and relates to the field of artificial intelligence. The method comprises the following steps: acquiring N radio frequency identification results obtained by performing radio frequency scanning on respective tags of N first articles; collecting at least one target image containing the N first articles; extracting M image features associated with each first article from the at least one target image; the radio frequency identification result of each first article and the M image features are input into a target detection model, a model identification result of each first article is obtained, and the target detection model is obtained based on machine learning; and carrying out warehouse-out operation or warehouse-in operation on the first article of which the radio frequency identification result is consistent with the model identification result. The invention further provides an article identification method for precious metal warehouse-in and warehouse-out, and a device, equipment, a storage medium and a program product corresponding to the method.

Description

Technical Field

[0001] The present disclosure relates to the field of artificial intelligence, and more specifically, to an object recognition method, device, equipment, medium and program product. Background Art

[0002] With the development of Internet of Things technology, object identification and management have become key requirements in many fields. In the field of warehouse management, traditional object identification methods mainly rely on RFID (Radio Frequency Identification) technology, which uses radio waves to identify and track items with electronic tags. However, RFID tags usually only provide limited object information and are easily interfered with.

[0003] Take the example of the in-and-out management of the vault of a commercial bank, which involves the flow and management of a large number of items. RFID technology, as a common tag recognition solution, is often used in bank box recognition and other scenarios. However, due to site factors and the presence of a large number of boxes, the possibility of RFID mis-scanning is relatively high, which brings certain troubles to the in-and-out management of the vault. Summary of the invention

[0004] In view of the above problems, the present disclosure provides an object identification method, apparatus, device, medium and program product that integrates radio frequency scanning technology and machine learning machine.

[0005] According to a first aspect of the present disclosure, there is provided a method for identifying objects for entering and exiting a warehouse, comprising: obtaining N radio frequency identification results obtained by performing radio frequency scanning on respective tags of N first objects; collecting at least one target image containing the N first objects; extracting M image features associated with each of the first objects from the at least one target image, where N and M are integers greater than or equal to 1; inputting the radio frequency identification result and the M image features of each of the first objects into a target detection model to obtain a model recognition result of each of the first objects, wherein the target detection model is obtained based on machine learning; and performing an out-of-warehouse operation or an in-warehouse operation on the first objects whose radio frequency identification results are consistent with the model recognition results.

[0006] According to an embodiment of the present disclosure, each of the first objects is used as a container to contain a second object; wherein, extracting M image features associated with each of the first objects from the at least one target image includes: performing the following operations on each of the first objects as a target container to extract at least one image feature of the target container and its label; and when the target container is made of a transparent material, extracting at least one image feature of the second object contained inside the target container through the target container.

[0007] According to an embodiment of the present disclosure, after extracting at least one image feature of the target container and its label, the method further includes: identifying a material category of the target container based on the at least one image feature extracted from the target container, the material category including a transparent material or a non-transparent material.

[0008] According to an embodiment of the present disclosure, the second object includes a precious metal, and extracting at least one image feature of the second object contained inside the target container through the target container includes: extracting at least one of a volume feature, a color feature, a texture feature, and a shape feature of the precious metal.

[0009] According to an embodiment of the present disclosure, the second object includes precious metals, the tag includes information of the precious metals, and the RFID result includes RFID information of the precious metals; wherein, inputting the RFID result and M image features of each of the first objects into the target detection model includes: inputting the RFID information of the precious metals, the material category of the target container, at least one image feature of the target container and its label, and at least one image feature of the precious metal into the target detection model.

[0010] According to an embodiment of the present disclosure, when N is greater than or equal to 2, at least two of the N first objects are stacked on each other.

[0011] According to an embodiment of the present disclosure, before the RFID result and M image features of each of the first items are input into the target detection model, the method also includes training the target detection model, specifically including: inputting the model recognition result of the training sample to be trained and the actual category of the training sample into a comprehensive loss function to obtain a loss value, wherein the training sample includes images and RFID results of historical items entering and leaving the warehouse; updating the parameters of the target detection model based on the loss value; wherein the comprehensive loss function includes a cross entropy loss function and its first weight, a bounding box regression loss function and its second weight, and a binary cross entropy loss function and its third weight.

[0012] Another aspect of an embodiment of the present disclosure provides an object identification method for the entry and exit of precious metals into and out of a warehouse, comprising: acquiring at least one target image containing a precious metal container; extracting at least one image feature of the precious metal container and its label from the at least one target image; when the precious metal container is made of a transparent material, extracting at least one image feature of the precious metal contained inside the precious metal container through the precious metal container; inputting at least one image feature of the precious metal container and its label, and at least one image feature of the precious metal, into a target detection model to obtain a model recognition result of the precious metal, wherein the target detection model is obtained based on machine learning; when the model recognition result meets a preset condition, performing an out-of-warehouse operation or an in-warehouse operation on the precious metal.

[0013] Another aspect of an embodiment of the present disclosure provides an object identification device for entering and exiting a warehouse, comprising: a radio frequency scanning module, used to obtain N radio frequency identification results obtained by performing radio frequency scanning on the labels of each of N first objects; a first acquisition module, used to acquire at least one target image containing the N first objects; a first extraction module, used to extract M image features associated with each of the first objects from the at least one target image, where N and M are integers greater than or equal to 1; a first prediction module, used to input the radio frequency identification result and the M image features of each of the first objects into a target detection model to obtain a model recognition result of each of the first objects, wherein the target detection model is obtained based on machine learning; a first entry and exit module, used to perform an exit operation or an entry operation on the first objects whose radio frequency identification results are consistent with the model recognition results.

[0014] Another aspect of an embodiment of the present disclosure provides an object identification device for precious metals entering and leaving a warehouse, comprising: a second acquisition module, used to acquire at least one target image containing a precious metal container; a second extraction module, used to extract at least one image feature of the precious metal container and its label from the at least one target image; a third extraction module, used to extract at least one image feature of the precious metal contained inside the precious metal container through the precious metal container when the precious metal container is made of a transparent material; a second prediction module, used to input at least one image feature of the precious metal container and its label, and at least one image feature of the precious metal, into a target detection model to obtain a model recognition result of the precious metal, wherein the target detection model is obtained based on machine learning; a second entry and exit module, used to perform an out-of-warehouse operation or an entry and exit operation on the precious metal when the model recognition result meets a preset condition.

[0015] Another aspect of an embodiment of the present disclosure provides an electronic device, including: one or more processors; and a memory for storing one or more programs, wherein when the one or more programs are executed by the one or more processors, the one or more processors execute the method described above.

[0016] Another aspect of an embodiment of the present disclosure provides a computer-readable storage medium having a computer program stored thereon. When the computer program is executed by a processor, the processor executes the method as described above.

[0017] Another aspect of an embodiment of the present disclosure provides a computer program product, including a computer program, which implements the above method when executed by a processor.

[0018] One or more of the above embodiments have the following beneficial effects:

[0019] 1) By combining radio frequency identification technology and machine learning technology, it can adapt to the appearance of different objects and changing environmental conditions. Compared with RFID tags, image features provide richer object information, effectively reducing the misidentification problem that may be caused by relying on a single RFID. By comparing the consistency of radio frequency identification results and model recognition results, the possibility of misoperation is further reduced, which not only improves the overall recognition rate, but also provides stronger fault tolerance.

[0020] 2) Regarding the identification of precious metals, considering their small size and the use of transparent containers, in addition to extracting the image features of precious metal containers and labels through image acquisition, the transparent characteristics of the containers can also be used to extract the image features of the precious metals through the precious metal containers, thereby improving the recognition efficiency and accuracy of precious metals by obtaining richer information. BRIEF DESCRIPTION OF THE DRAWINGS

[0021] The above contents and other objects, features and advantages of the present disclosure will become more apparent through the following description of the embodiments of the present disclosure with reference to the accompanying drawings, in which:

[0022] Figure 1 The application scenario diagram of the object identification method according to the embodiment of the present disclosure is schematically shown;

[0023] Figure 2 A flowchart of a method for identifying items entering and leaving a warehouse according to an embodiment of the present disclosure is schematically shown;

[0024] Figure 3 Schematically shows a flow chart of extracting image features of a second object according to an embodiment of the present disclosure;

[0025] Figure 4A flowchart of an object identification method for precious metal storage in and out according to an embodiment of the present disclosure is schematically shown;

[0026] Figure 5 Schematically shows a flow chart of training a target detection model according to an embodiment of the present disclosure;

[0027] Figure 6 A schematic diagram of a management flow chart for entering and exiting a vault according to an embodiment of the present disclosure is shown;

[0028] Figure 7 The structure block diagram of the device for identifying objects in and out of a warehouse according to an embodiment of the present disclosure is schematically shown;

[0029] Figure 8 A schematic diagram showing a structural block diagram of an object identification device for precious metal storage and retrieval according to an embodiment of the present disclosure; and

[0030] Fig. 9 A block diagram of an electronic device suitable for implementing an object identification method according to an embodiment of the present disclosure is schematically shown. DETAILED DESCRIPTION

[0031] Hereinafter, embodiments of the present disclosure will be described with reference to the accompanying drawings. However, it should be understood that these descriptions are exemplary only and are not intended to limit the scope of the present disclosure. In the following detailed description, for ease of explanation, many specific details are set forth to provide a comprehensive understanding of the embodiments of the present disclosure. However, it is apparent that one or more embodiments may also be implemented without these specific details. In addition, in the following description, descriptions of known structures and technologies are omitted to avoid unnecessary confusion of the concepts of the present disclosure.

[0032] Figure 1 The following schematically shows an application scenario of the object identification method according to an embodiment of the present disclosure. It should be noted that: Figure 1 What is shown are merely examples to which the embodiments of the present disclosure can be applied, so as to help those skilled in the art understand the technical content of the present disclosure, but it does not mean that the embodiments of the present disclosure cannot be used in other devices, systems, environments or scenarios.

[0033] Among them, in application scenario 100, the object identification method for entering and exiting a warehouse and the object identification method for entering and exiting a warehouse for precious metals involved in the present disclosure can be implemented.

[0034] like Figure 1As shown, the application scenario 100 according to this embodiment may include a first camera 111, a second camera 112, a server 120, a warehouse 131, a plurality of items 132 to be put in and out of the warehouse, a radio frequency scanning device 133, and a network 140. The network 140 is used to provide a medium for a communication link between the first camera 111, the second camera 112, and the server 120. The network 140 may include various connection types, such as wired, wireless communication links, or optical fiber cables, etc.

[0035] The terminal device (not shown) may be used to receive the video or image captured by the first camera 111 and the second camera 112, and may also be used to interact with the server 120, such as sending a storage entry and exit instruction or receiving an item recognition result, etc. Various communication client applications may be installed on the terminal device, such as shopping applications, web browser applications, search applications, instant messaging tools, email clients, social platform software, etc. (only as examples).

[0036] The terminal device may be any electronic device having a display screen and supporting web browsing, including but not limited to a smart phone, a tablet computer, a laptop computer, a desktop computer, etc. In some embodiments, the first camera 111 and the second camera 112 may be implemented by a terminal device having a camera function.

[0037] The server 120 may be a server that provides various services, such as a background management server that provides support for websites browsed by users using terminal devices (for example only). The background management server may analyze and process the received data such as user requests, and feed back the processing results (such as web pages, information, or data obtained or generated according to user requests) to the terminal device. For example, the server 120 may be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud computing, network services, and middleware services.

[0038] It should be noted that you can use Figure 1 The radio frequency scanning device 133 in the storage performs radio frequency scanning on the multiple items 132 to be entered and exited from the warehouse to obtain the radio frequency identification result of each item. The item identification method provided in the embodiment of the present disclosure can generally be executed by at least one of the terminal device or the server. Accordingly, the item identification device provided in the embodiment of the present disclosure can generally be set in at least one of the terminal device or the server. The terminal device can be a device for edge computing arranged on the side of the warehouse 131, which can complete the calculation of large amounts of data on the warehouse side, thereby improving the efficiency of entering and exiting the warehouse.

[0039] It should be understood that Figure 1The number of terminal devices, networks and servers in the embodiment is only for illustration. Any number of terminal devices, networks and servers may be provided according to implementation requirements.

[0040] The following will be based on Figure 1 The scene described by Figure 2~Figure 3 The object identification method for entering and exiting a warehouse according to an embodiment of the present disclosure is described in detail.

[0041] Figure 2 The flowchart of the method for identifying items entering and leaving a warehouse according to an embodiment of the present disclosure is schematically shown.

[0042] like Figure 2 As shown, this embodiment includes:

[0043] In operation S210, N radio frequency identification results are obtained by performing radio frequency scanning on respective tags of N first objects.

[0044] For example, in a bank vault scenario, gold bars, cash, vouchers, bills, etc. can all be equipped with RFID tags, and the tag information can be read by the radio frequency scanning device 133 to obtain the information of the item.

[0045] When N is greater than or equal to 2, the plurality of first items may include items of the same type or different types.

[0046] In operation S220 , at least one target image including N first objects is acquired.

[0047] For example, one image may include multiple first objects to be put in and taken out of the warehouse, and then a target image of each first object is cut out from the image. Alternatively, multiple target images may be taken of each first object from one or more angles.

[0048] In some embodiments, when N is greater than or equal to 2, at least two of the N first objects are stacked on top of each other.

[0049] For example, in a bank vault, multiple items may be put in and taken out of the vault at the same time. During transportation, some items are placed on top of other items, forming a stacked state, which increases the complexity of identification and operation. In order to obtain as much image information as possible, the camera angle can be adjusted or multiple cameras can be used to collect images.

[0050] In operation S230 , M image features associated with each first object are extracted from at least one target image, where N and M are integers greater than or equal to 1, respectively.

[0051] For example, the M image features may include the features of the first object itself, such as shape, texture, color, volume, etc.; they may also include the features of the label thereon, such as the position of the label on the first object, the information displayed in the label area, such as the cash box number, line number, branch number and branch name, transfer date, cash box type, handler, reviewer, and QR code information corresponding to the RFID, etc.

[0052] For example, in the case of stacking objects, the edge of each first object may be identified to distinguish and identify the image features of each first object.

[0053] In operation S240, the radio frequency identification result of each first object and the M image features are input into a target detection model to obtain a model recognition result of each first object, wherein the target detection model is obtained based on machine learning.

[0054] For example, the target detection model may include a deep learning network that has been pre-trained to recognize one or more image features and RFID results and is capable of outputting a prediction result. For example, an image of each first object may be cut out from at least one target image. For example, taking a cash box as an example, the image of the cash box is cut out and the image features are extracted. Then, the M image features of the cash box and all RFID results of all first objects (or the RFID results of the cash box) are input into the target detection model to obtain a model recognition result that predicts whether its category is a cash box.

[0055] In the bank vault scene, gold bars, cash boxes, vouchers, bills, etc. have different image features because they belong to different categories and uses. If the RFID results are accurate, their respective label information can be scanned. The RFID results can be used as prior knowledge. On the basis of the target detection model's prediction based on M image features, it can be considered whether the RFID scanning results contain the same prediction results, and then the model outputs the model recognition results. These can enable the target detection model to learn the ability to recognize objects based on image features and RFID results as input during the training phase.

[0056] In some embodiments, the extraction of M image features may also be performed by a target detection model, or the M image features may be extracted using other deep learning models independent of the target detection model.

[0057] In operation S250, a warehouse-out operation or a warehouse-in operation is performed on the first item whose radio frequency identification result is consistent with the model identification result.

[0058] For each first item, the RFID result and the model recognition result can be compared to determine if there is a false scan. If the RFID result and the model recognition result are consistent, the item can be put in and out of the warehouse normally.

[0059] According to the embodiments of the present disclosure, by combining radio frequency identification technology and machine learning technology, it is possible to adapt to the appearance of different objects and changing environmental conditions, and the image features provide richer object information compared to RFID tags, effectively reducing the misidentification problem that may be caused by relying on a single RFID. By comparing the consistency of radio frequency identification results and model recognition results, the possibility of misoperation is further reduced, which not only improves the overall recognition rate, but also provides more powerful fault tolerance.

[0060] It can be understood that the first item can be the item itself that is in and out of the warehouse. The first item can also be a container, and the items contained therein (such as gold bars, cash, vouchers, bills, etc.) are the actual items in and out of the warehouse, and the M image features associated with each first item include the image features of the items contained therein. The following further describes one of the latter situations.

[0061] Figure 3 The flowchart of extracting the image features of the second object according to an embodiment of the present disclosure is schematically shown.

[0062] Each first object is used as a container to hold a second object. Figure 3 As shown, this embodiment is one of the embodiments of operation S230, including taking each first object as a target container and performing the following operations:

[0063] In operation S310 , at least one image feature of a target container and its label is extracted.

[0064] In operation S320 , when the target container is made of a transparent material, at least one image feature of the second object contained in the target container is extracted through the target container.

[0065] For example, one or more of gold bars, cash, vouchers, bills, etc. may use a transparent container, which may be a box with one or more transparent sides, or a transparent plastic bag, etc. When the target container is made of a transparent material, it can not only serve as a container for attaching a label, but also display the second object contained inside.

[0066] The transparent material involved in the present disclosure refers to a substance that allows light to pass through, such as certain plastics or glass.

[0067] For example, first, the contrast between the transparent container and the second object in the image is increased to make the second object inside easier to identify. Then, based on the color and depth information, the transparent container can be segmented from the remaining containers, for example, considering the occlusion, part or all of the transparent container displayed in the target image can be segmented. Then, based on the color and depth information, the transparent container is distinguished from the second object. Then, an edge detection algorithm is used to identify the outline of the second object. And the texture, volume, shape and other features of the second object are identified.

[0068] According to the embodiments of the present disclosure, by distinguishing between containers made of transparent materials and the items contained therein and extracting image features thereof respectively, it is helpful to obtain richer information and improve recognition accuracy.

[0069] In some embodiments, during the training phase, training samples of historical items using transparent containers can be provided so that the target detection model can learn the representation of light passing through transparent containers for different types of second objects, thereby effectively extracting image features of transparent containers and image features of different types of second objects.

[0070] In some embodiments, the second object includes a precious metal (eg, a gold bar), and extracting at least one image feature of the second object contained inside the target container through the target container includes extracting at least one of a volume feature, a color feature, a texture feature, and a shape feature of the precious metal.

[0071] In some embodiments, after extracting at least one image feature of the target container and its label, the method further includes: identifying a material category of the target container based on the at least one image feature extracted from the target container, where the material category includes a transparent material or a non-transparent material.

[0072] For example, a category recognition model independent of the target detection model can also be provided, which can be obtained based on a neural network algorithm by collecting the depth and light intensity of transparent materials and scene information. Due to the characteristics of transparent materials, the collected depth will be distorted. By estimating the refractive index, reflectivity and transmittance, the material features that express the characteristics of highly transparent materials can be extracted and input into the category recognition model to realize the classification of transparent materials.

[0073] According to the embodiments of the present disclosure, transparent containers and non-transparent containers may be mixed and stacked together. Through a two-stage processing method, whether the material is transparent is first identified, and then the image features of the internal objects are extracted, thereby avoiding processing erroneous information as the image features of the second object, saving computing resources and improving accuracy.

[0074] In some embodiments, the second object includes precious metals, the tag includes information about the precious metals, and the RFID result includes RFID information about the precious metals.

[0075] Among them, inputting the radio frequency identification result and M image features of each first object into the target detection model includes: inputting the radio frequency identification information of the precious metal, the material category of the target container, at least one image feature of the target container and its label, and at least one image feature of the precious metal into the target detection model.

[0076] According to the embodiments of the present disclosure, by effectively identifying and classifying target containers made of transparent materials and non-transparent materials, effective prior knowledge can be used to identify specific categories of items. For example, if only precious metals use transparent containers, the accuracy of precious metal identification can be improved by determining the category of the container.

[0077] The items in and out of the vault include banknotes, precious metals, certificates, etc. Different items have different values ​​and characteristics, and may use different outer packaging. Precious metals are of higher value and smaller in size. If transparent plastic is used as the outer packaging, the shape is irregular and it is easy to be blocked, which may lead to misoperation.

[0078] In some embodiments, the target detection model can also be used alone to realize the identification of precious metals. Figure 4 Further explanation.

[0079] Figure 4 The flowchart of the method for identifying items for entering and exiting a warehouse of precious metals according to an embodiment of the present disclosure is schematically shown.

[0080] like Figure 4 As shown, this embodiment includes:

[0081] In operation S410, at least one target image including a precious metal container is acquired.

[0082] For example, the precious metal containers to be put in and out of the warehouse may be mixed and stacked with cash boxes, voucher boxes, ticket boxes, etc., and the volume is smaller. A target image containing precious metal containers, cash boxes, voucher boxes, ticket boxes, etc., or multiple target images at multiple angles can be collected. In the case where the precious metal container is covered by other containers, the radio frequency identification result can be obtained, and the precious metal container can be exposed to the camera by manual or mechanical means. Alternatively, in the same batch of items to be put in and out of the warehouse, all are precious metal containers, and the number of items to be put in and out of the warehouse in this batch can be one or more.

[0083] In operation S420, at least one image feature of the precious metal container and its label is extracted from at least one target image.

[0084] For example, features that help identify containers can be extracted, such as shape, size, color, texture, etc. Features of labels can also be extracted, such as the location on the container, information displayed in the label area, such as the number, line number, branch number and branch name, transfer date, type, handler, reviewer, and QR code information corresponding to the RFID.

[0085] In operation S430 , when the precious metal container is made of a transparent material, at least one image feature of the precious metal contained in the precious metal container is extracted through the precious metal container.

[0086] In operation S440, at least one image feature of the precious metal container and its label, and at least one image feature of the precious metal are input into a target detection model to obtain a model recognition result of the precious metal, wherein the target detection model is obtained based on machine learning.

[0087] In operation S450, when the model recognition result meets the preset conditions, a storage-out operation or storage-in operation is performed on the precious metal.

[0088] For example, the preset condition may be that the RFID result of precious metals is consistent with the model recognition result, such as both are gold bars or silver ingots. Alternatively, the preset condition may be that the result is consistent with the manually created entry and exit application record. Alternatively, the preset condition may be that the predicted probability value in the model recognition result is greater than 80%, which is only an example.

[0089] It can be understood that the means of the corresponding steps in the above method for identifying items entering and leaving the warehouse can be applied to this embodiment and will not be repeated here.

[0090] According to the embodiments of the present disclosure, the recognition of precious metals takes into account the characteristics of precious metals such as small size and the use of containers made of transparent materials. In addition to extracting image features of precious metal containers and labels through image acquisition, it is also possible to utilize the transparent characteristics of the containers to extract image features of the precious metals through the precious metal containers, thereby improving the recognition efficiency and accuracy of precious metals by obtaining richer information.

[0091] The process of pre-training the object detection model is further explained below.

[0092] Figure 5 A flowchart of training a target detection model according to an embodiment of the present disclosure is schematically shown.

[0093] like Figure 5 As shown, this embodiment includes:

[0094] In operation S510, a radio frequency identification result is obtained by performing radio frequency scanning on the tag of the training sample. The training sample includes images of historical in-and-out warehouse items and the radio frequency identification result in operation S510.

[0095] For historical in-and-out items, the radio frequency scanning module is used to capture the signal of the RFID tag, extract the object ID and related information in the tag as the radio frequency identification result, and collect its image through the camera.

[0096] Based on the historical data, an image training set containing labels of historical in-and-out items and radio frequency scanning results can be collected, thereby obtaining a training set including one or more training samples. Each training sample can also include the actual category of the historical in-and-out items.

[0097] For example, to further enrich the training set and help the model learn and generalize better, one or more of the following data enhancement operations can be performed on the basis of historical data:

[0098] Random cropping: simulates the situation where labels are in different positions. The labels on the containers entering and leaving the vault are randomly affixed, and the labels may appear in different positions on the containers.

[0099] Rotation: simulates the situation of labels at different angles. The containers entering and leaving the vault are placed manually, and it is impossible to place them neatly and fixedly, so the labels will appear at any angle.

[0100] Scaling: Make the model adaptable to labels of different sizes. The containers entering and leaving the warehouse are boxes or bags of various sizes, and the label sizes vary according to the different containers.

[0101] Blur and noise: simulate image quality problems that may occur in real environments. The vault site operation scene is complex, with many items and people being handed over, which may cause problems such as image quality degradation.

[0102] For example, the YOLOv8 model can be used as a target detection model for real-time detection. It can detect and classify objects by performing a single forward pass on the input image.

[0103] In the process of building the training set, the RFID tags of historical in-and-out items are converted into a format that can be recognized by the YOLOv8 model. YOLOv8 files are usually text files. Each file corresponds to an image. Each line in the file represents an item and contains information such as the category, the center coordinates of the bounding box, the width, and the height. The format is as follows:

[0104] <object-class> <x-center> <y-center> <width> <height>

[0105] in: <object-class>: Represents the category label of the container, numbered from 0, 0 represents the transfer payment box, 1 represents the ATM bag, and 2 represents the door-to-door collection bag.

[0106] <x-center> 、 <y-center>: Relative coordinates of the center of the bounding box (relative to the image width and height).

[0107] <width> 、 <height>: Relative width and height of the bounding box (relative to the image width and height).

[0108] The conversion steps are as follows:

[0109] a) Data labeling: Use the labeling tool to label the box labels in the image. Make sure the labeled bounding box tightly covers the label area and contains all key information (box number, etc.). When labeling, assign a unique category number to each label category.

[0110] b) Label file generation: Export the label files generated by the annotation tool (such as JSON format). Write a Python script to convert these label files into the format required by the YOLOv8 model.

[0111] In operation S520, at least one target image in the training sample is acquired; that is, images of historical in-and-out warehouse items included in the training sample.

[0112] In operation S530, M image features associated with the training sample are extracted from at least one target image.

[0113] In operation S540, the radio frequency recognition result of the training sample and the M image features are input into the object detection model to be trained to obtain the model recognition result of the training sample.

[0114] For example, the operation of extracting M image features may be performed at once using a YOLOv8 model.

[0115] In the process of model construction, the pre-trained YOLOv8 model is used as the basic network. According to the number of categories in the training set, the output layer of the model is adjusted, and the final classifier is modified to match the categories of the training set. The specific steps are as follows:

[0116] a) Load the pre-trained model. b) Modify the model configuration: Modify the model configuration according to the number of training set categories and specific needs. For example, set the number of categories to 3 and test the scenarios of branch transfer, ATM and door-to-door collection. c) Fine-tune the model: Use the training set to fine-tune the model so that the model can adapt to specific vault scenarios. For example, set the training parameters according to the requirements of vault in and out container recognition. Including image size, number of training rounds, batch size, etc. Set the resolution of the input image and use a high-resolution camera to capture the image to ensure that the label area is clearly visible.

[0117] In operation S550, the model recognition result of the object detection model to be trained on the training sample and the actual category of the training sample are input into the comprehensive loss function to obtain a loss value.

[0118] Among them, the comprehensive loss function includes the cross entropy loss function and its first weight, the bounding box regression loss function and its second weight, and the binary cross entropy loss function and its third weight.

[0119] In the vault entry and exit object recognition scenario, high-precision bounding box positioning and reliable classification are required. Choosing a suitable loss function is critical because it directly affects the training effect and final performance of the model.

[0120] Exemplarily, the cross entropy loss function is as follows:

[0121]

[0122] in: represents the number of samples; It is a sample Belongs to category the actual category of (0 or 1); is the weight of the category; is the probability that the model predicts that sample i belongs to category C.

[0123] For example, you can set weights for different categories. For example, this scene has 3 categories, and the weights are set to 2.0, 1.0, and 1.5 respectively. As follows:

[0124] class_weights = torch.tensor([2.0, 1.0, 1.5])

[0125] For example, in the vault entry and exit recognition scenario, the accuracy of the bounding box is more important, especially for the positioning of the label, so the weighted smooth L1 loss is chosen. The bounding box regression loss function is as follows:

[0126]

[0127] in: represents the jth parameter of the model-predicted and true bounding box for sample i; the bounding box is usually represented by the coordinates of the center point and the width and height, or the coordinates of the two diagonal points. j is the weight of the jth parameter, which is used to adjust the loss contribution of different parameters. It is the smooth L1 loss function (Smooth L1 Loss).

[0128] You can set weights for different parameters. Suppose the center coordinate weight is 1.0, and the width and height weights are 2.0, for example:

[0129] box_weights = torch.tensor([1.0, 1.0, 2.0, 2.0])

[0130] For example, in the vault entry and exit recognition scenario, the confidence of the object is also an important indicator, especially in high-risk areas, so the weighted binary cross-entropy loss (Weighted Binary Cross-EntropyLoss) is chosen. The binary cross-entropy loss function is as follows:

[0131]

[0132] in, It is a sample the true confidence level of (0 or 1); It is a sample Weight; is the probability that the model predicts that sample 𝑖 belongs to the positive class (label 1).

[0133] This scenario has 3 categories, and the weights are set to 2.0, 1.0, and 1.5, as follows:

[0134] confidence_weights = torch.tensor([2.0, 1.0, 1.5])

[0135] In the vault entry and exit recognition scenario, the above three loss functions are used in combination, and different weights are set for each loss function to ensure that the model can achieve good results in all aspects. The comprehensive loss function Total loss is as follows:

[0136]

[0137] in, is the first weight, and Classification loss is the classification loss, which corresponds to the value of the cross entropy loss function. is the second weight, and Bounding box regression loss corresponds to the value of the bounding box regression loss function. The third weight, Object confidence target confidence loss, corresponds to the value of the binary cross entropy loss function. The values ​​of the first weight, the second weight, and the third weight can be automatically optimized during the training process or can be debugged and determined based on expert experience.

[0138] In other embodiments, the YOLOv8 model can be trained using a conventional loss function (such as mean error).

[0139] In operation S560, parameters of the object detection model are updated based on the loss value.

[0140] For example, among all the samples, 80% of the data is used as the training set and 20% is used as the validation set. The model is trained using the training set and the model parameters are updated through the back propagation algorithm. The performance of the model on the validation set is monitored to avoid overfitting. A stricter non-maximum suppression (NMS) threshold can be used to filter out detection frames with more overlaps to further reduce false scans. In addition, the performance of the model is evaluated using the comprehensive accuracy, false alarm rate, and false negative rate. Among them, the comprehensive accuracy is the ratio of the number of objects that are correctly detected and correctly read the label to the total number of objects. The false alarm rate is the ratio of the number of objects that are incorrectly detected or incorrectly read the label to the total number of objects. The false alarm rate is the ratio of the number of objects that are not detected or read the label to the total number of objects. The false negative rate is the ratio of the number of objects that are not detected or read the label to the total number of objects.

[0141] The test set is 100 containers in and out of the vault, of which 20 object labels are partially blocked and 10 objects are tested under different lighting conditions. The results are shown in the following table:

[0142] Table 1

[0143]

[0144] In some embodiments, historical in-and-out warehouse items using transparent containers can also be provided as training samples, so that the target detection model can have the ability to accurately detect the type of material contained in the transparent container. Figure 4 and Figure 5 , which is different from Figure 5 The point is not to use the RFID results as model input, which will not be discussed here.

[0145] In other embodiments, historical in-and-out warehouse items using transparent containers can be provided as training samples, and the radio frequency identification results can be used as model inputs, that is, the present disclosure relates to an object identification method for in-and-out warehouse items and an object identification method for precious metal in-and-out warehouse items, which can use the same target detection model, that is, Figure 1 The method for identifying items in and out of the warehouse shown includes Figure 4 The technical solution shown.

[0146] Figure 6 The following schematically shows a management flow chart for vault entry and exit according to an embodiment of the present disclosure.

[0147] like Figure 6 As shown, the safe contains multiple lines (such as line 1 to line n, n is greater than 1), and each line is distributed with multiple in-and-out items, such as line 1 is distributed with box 1 to box n, and line 2 is distributed with package 1 to package n.

[0148] For each inbound and outbound item on line 1 to line n, in response to the inbound and outbound instruction of the inbound and outbound management system (used to record and manage inbound and outbound data), the radio frequency scanning device 133 configured at the entrance of the vault reads the RFID tag. At the same time as the RFID scanning, one or more cameras next to the radio frequency scanning device 133 are used to visually capture the inbound and outbound items, and the captured images are transmitted to the YOLOv8 model for target detection to achieve object recognition.

[0149] In the process of target detection, preprocessing is first performed, and preprocessing steps such as resizing and normalization can be performed to meet the input requirements of the model. The preprocessed image is then input into a deep convolutional neural network, which passes through multiple layers of convolutional layers, activation layers, pooling layers, and possible other layers (such as skip connections and upsampling layers) to learn and extract image features. Next, bounding box prediction and category recognition, for example, on each feature map, the YOLOv8 model uses a set of prediction layers to predict the bounding box and object category in the image. The YOLOv8 model then uses the non-maximum suppression (NMS) algorithm to reduce overlapping bounding boxes and only retain the bounding boxes with the highest confidence. Combined with the category confidence score and a preset threshold, the bounding boxes with lower confidence are filtered out. After the NMS and post-processing steps, the YOLOv8 model obtains the final tag recognition result. Each tag recognition result typically includes the location (coordinates) of the bounding box, the object category, and the confidence score. Optionally, the radio frequency identification result can be input into the YOLOv8 model.

[0150] In the data comparison stage, the RFID results of each item to be in and out of the warehouse are compared with the model recognition results to determine whether they are consistent, and to determine whether they are mis-scanned. If they are consistent, the scan is confirmed to be successful and the in and out operation is performed. If they are inconsistent, it is confirmed as a scan anomaly and feedback is given to the in and out warehouse management system.

[0151] As an additional logic layer, data comparison can improve the recognition accuracy of the entire system and reduce the problem of false scans. For example, a dynamically adjusted confidence threshold can be set. When the RFID signal strength is insufficient or interfered, the model recognition results provided by YOLOv8 will be relied on. It can intelligently optimize the recognition process and reduce false scans caused by changes in the signal environment. In addition, a feedback mechanism can be introduced to evaluate the scanning results in real time and perform correction processing when anomalies (such as false scans or inconsistent data) are detected. The self-correction capability is improved, making the management process more robust.

[0152] According to the embodiments of the present disclosure, multimodal data fusion is realized. When the RFID does not scan the tag correctly or other errors occur, YOLOv8 can verify and supplement the RFID data through visual information. This mode can not only reduce the risks brought by relying on a single technology (such as relying only on RFID), but also improve the overall recognition accuracy. If there is a problem with the RFID recognition system, YOLOv8 can re-identify the object and reconfirm the status of the item through visual information. Even when the RFID tag cannot be read or is damaged, the correct management of the item can be guaranteed to reduce the situation of mis-scanning and missed scanning. The computing resource requirements can be reduced and the efficiency can be improved through model pruning, quantization technology and edge computing.

[0153] Figure 7 The structural block diagram of the object identification device for entering and exiting a warehouse according to an embodiment of the present disclosure is schematically shown.

[0154] like Figure 7 As shown, the device 700 for identifying items entering and exiting a warehouse includes a radio frequency scanning module 710 , a first acquisition module 720 , a first extraction module 730 , a first prediction module 740 , and a first entry and exit module 750 .

[0155] The radio frequency scanning module 710 may perform operation S210 to obtain N radio frequency identification results obtained by performing radio frequency scanning on respective tags of the N first objects.

[0156] The first acquisition module 720 may perform operation S220 for acquiring at least one target image including N first objects.

[0157] The first extraction module 730 may perform operation S230 to extract M image features associated with each first object from at least one target image, where N and M are integers greater than or equal to 1, respectively.

[0158] The first prediction module 740 may perform operation S240 for inputting the radio frequency identification result and the M image features of each first object into a target detection model to obtain a model identification result of each first object, wherein the target detection model is obtained based on machine learning.

[0159] The first storage-in / out module 750 may perform operation S250 to perform a storage-out operation or a storage-in operation on the first item whose radio frequency identification result is consistent with the model identification result.

[0160] In some embodiments, the first extraction module 730 may also be used to perform operations S310 to S320. The first extraction module 730 may also be used to extract at least one of volume features, color features, texture features, and shape features of the precious metal.

[0161] In some embodiments, the object identification device 700 may further include a category prediction module for identifying a material category of the target container based on at least one image feature extracted from the target container, where the material category includes a transparent material or a non-transparent material.

[0162] In some embodiments, the first prediction module 740 may also be used to input the precious metal RFID information, the material category of the target container, at least one image feature of the target container and its label, and at least one image feature of the precious metal into the target detection model.

[0163] In some embodiments, the object identification device 700 may further include a training module, which may at least perform operations S550 to S560.

[0164] Figure 8 The structural block diagram of the object identification device for precious metals entering and exiting the warehouse according to an embodiment of the present disclosure is schematically shown.

[0165] like Figure 8 As shown, the article identification device 800 for precious metals entering and exiting the warehouse in this embodiment includes a second acquisition module 810 , a second extraction module 820 , a third extraction module 830 , a second prediction module 840 , and a second entry and exit module 850 .

[0166] The second acquisition module 810 may perform operation S410 to acquire at least one target image including a precious metal container.

[0167] The second extraction module 820 may perform operation S420 for extracting at least one image feature related to the precious metal container and its label from at least one target image.

[0168] The third extraction module 830 may perform operation S430 to extract at least one image feature of the precious metal contained in the precious metal container through the precious metal container when the precious metal container is made of a transparent material.

[0169] The second prediction module 840 can perform operation S440, which is used to input at least one image feature of the precious metal container and its label, and at least one image feature of the precious metal, into the target detection model to obtain a model recognition result of the precious metal, wherein the target detection model is obtained based on machine learning.

[0170] The second storage-in / out module 850 may execute operation S450 for executing a storage-out operation or a storage-in operation on the precious metals when the model recognition result meets the preset conditions.

[0171] For the parts not mentioned in the device part, they can be understood with reference to the various embodiments of the above method. That is, the device part includes modules for executing the various steps of any one of the method embodiments described above. In addition, the implementation methods, technical problems solved, functions implemented, and technical effects achieved of each module / unit / subunit, etc. in the device part embodiment are respectively the same or similar to the implementation methods, technical problems solved, functions implemented, and technical effects achieved of each corresponding step in the method part embodiment, and will not be repeated here.

[0172] According to the embodiments of the present disclosure, any multiple modules in the object identification device 700 or the object identification device 800 can be combined into one module for implementation, or any one of the modules can be split into multiple modules. Alternatively, at least part of the functions of one or more of the modules can be combined with at least part of the functions of other modules and implemented in one module.

[0173] According to an embodiment of the present disclosure, at least one module in the item identification device 700 or the item identification device 800 may be at least partially implemented as a hardware circuit, such as a field programmable gate array (FPGA), a programmable logic array (PLA), a system on a chip, a system on a substrate, a system on a package, an application specific integrated circuit (ASIC), or may be implemented by hardware or firmware in any other reasonable manner of integrating or packaging the circuit, or implemented in any one of the three implementation methods of software, hardware and firmware or in an appropriate combination of any of them. Alternatively, at least one of the item identification device 700 or the item identification device 800 may be at least partially implemented as a computer program module, and when the computer program module is executed, the corresponding function may be executed.

[0174] Fig. 9 A block diagram of an electronic device suitable for implementing an object identification method according to an embodiment of the present disclosure is schematically shown.

[0175] like Fig. 9 As shown, the electronic device 900 according to an embodiment of the present disclosure includes a processor 901, which can perform various appropriate actions and processes according to a program stored in a read-only memory (ROM) 902 or a program loaded from a storage part 908 to a random access memory (RAM) 903. The processor 901 may include, for example, a general-purpose microprocessor (e.g., a CPU), an instruction set processor and / or a related chipset and / or a dedicated microprocessor (e.g., an application-specific integrated circuit (ASIC)), etc. The processor 901 may also include an onboard memory for caching purposes. The processor 901 may include a single processing unit or multiple processing units for performing different actions of the method flow according to an embodiment of the present disclosure.

[0176] In RAM 903, various programs and data required for the operation of electronic device 900 are stored. Processor 901, ROM 902 and RAM 903 are connected to each other via bus 904. Processor 901 performs various operations of the method flow according to the embodiment of the present disclosure by executing the program in ROM 902 and / or RAM 903. It should be noted that the program can also be stored in one or more memories other than ROM 902 and RAM 903. Processor 901 can also perform various operations of the method flow according to the embodiment of the present disclosure by executing the program stored in one or more memories.

[0177] According to an embodiment of the present disclosure, the electronic device 900 may further include an input / output (I / O) interface 905, which is also connected to the bus 904. The electronic device 900 may further include one or more of the following components connected to the input / output (I / O) interface 905: an input portion 906 including a keyboard, a mouse, etc.; an output portion 907 including a cathode ray tube (CRT), a liquid crystal display (LCD), etc., and a speaker, etc.; a storage portion 908 including a hard disk, etc.; and a communication portion 909 including a network interface card such as a LAN card, a modem, etc. The communication portion 909 performs communication processing via a network such as the Internet. A drive 910 is also connected to the input / output (I / O) interface 905 as needed. A removable medium 911, such as a magnetic disk, an optical disk, a magneto-optical disk, a semiconductor memory, etc., is installed on the drive 910 as needed, so that a computer program read therefrom is installed into the storage portion 908 as needed.

[0178] The present disclosure also provides a computer-readable storage medium, which may be included in the device / apparatus / system described in the above embodiments; or may exist independently without being assembled into the device / apparatus / system. The above computer-readable storage medium carries one or more programs, and when the above one or more programs are executed, the method according to the embodiment of the present disclosure is implemented.

[0179] According to an embodiment of the present disclosure, a computer-readable storage medium may be a non-volatile computer-readable storage medium, for example, may include but is not limited to: a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof. In the present disclosure, a computer-readable storage medium may be any tangible medium containing or storing a program, which may be used by or in combination with an instruction execution system, an apparatus or a device. For example, according to an embodiment of the present disclosure, a computer-readable storage medium may include the ROM 902 and / or RAM 903 described above and / or one or more memories other than ROM 902 and RAM 903.

[0180] The embodiment of the present disclosure also includes a computer program product, which includes a computer program, and the computer program contains program code for executing the method shown in the flowchart. When the computer program product is run in a computer system, the program code is used to enable the computer system to implement the method provided by the embodiment of the present disclosure.

[0181] The above functions defined in the system / device of the embodiment of the present disclosure are performed when the computer program is executed by the processor 901. According to the embodiment of the present disclosure, the system, device, module, unit, etc. described above can be implemented by a computer program module.

[0182] In one embodiment, the computer program may be based on a tangible storage medium such as an optical storage device, a magnetic storage device, etc. In another embodiment, the computer program may also be transmitted and distributed in the form of a signal on a network medium, and downloaded and installed through the communication part 909, and / or installed from a removable medium 911. The program code contained in the computer program may be transmitted using any appropriate network medium, including but not limited to: wireless, wired, etc., or any suitable combination of the above.

[0183] In such an embodiment, the computer program can be downloaded and installed from the network through the communication part 909, and / or installed from the removable medium 911. When the computer program is executed by the processor 901, the above functions defined in the system of the embodiment of the present disclosure are performed. According to the embodiment of the present disclosure, the system, device, means, module, unit, etc. described above can be implemented by a computer program module.

[0184] According to an embodiment of the present disclosure, the program code for executing the computer program provided by the embodiment of the present disclosure can be written in any combination of one or more programming languages. Specifically, these computing programs can be implemented using high-level process and / or object-oriented programming languages, and / or assembly / machine languages. Programming languages ​​include, but are not limited to, Java, C++, python, "C" language or similar programming languages. The program code can be executed entirely on the user computing device, partially on the user device, partially on the remote computing device, or entirely on the remote computing device or server. In the case of a remote computing device, the remote computing device can be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or can be connected to an external computing device (for example, using an Internet service provider to connect through the Internet).

[0185] The flow charts and block diagrams in the accompanying drawings illustrate the possible architecture, functions and operations of the systems, methods and computer program products according to various embodiments of the present disclosure. In this regard, each box in the flow chart or block diagram can represent a module, a program segment, or a part of a code, and the above-mentioned module, program segment, or a part of a code contains one or more executable instructions for realizing the specified logical function. It should also be noted that in some alternative implementations, the functions marked in the box can also occur in a different order from the order marked in the accompanying drawings. For example, two boxes represented in succession can actually be executed substantially in parallel, and they can sometimes be executed in the opposite order, depending on the functions involved. It should also be noted that each box in the block diagram or flow chart, and the combination of the boxes in the block diagram or flow chart can be implemented with a dedicated hardware-based system that performs a specified function or operation, or can be implemented with a combination of dedicated hardware and computer instructions.

[0186] It will be appreciated by those skilled in the art that the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways, even if such combinations or combinations are not explicitly described in the present disclosure. In particular, without departing from the spirit and teachings of the present disclosure, the features described in the various embodiments of the present disclosure may be combined and / or combined in a variety of ways. All of these combinations and / or combinations fall within the scope of the present disclosure.

[0187] The embodiments of the present disclosure are described above. However, these embodiments are only for illustrative purposes and are not intended to limit the scope of the present disclosure. Although the embodiments are described above separately, this does not mean that the measures in the various embodiments cannot be used in combination to advantage. The scope of the present disclosure is defined by the attached claims and their equivalents. Without departing from the scope of the present disclosure, those skilled in the art may make a variety of substitutions and modifications, which should all fall within the scope of the present disclosure.< / height> < / width> < / x-center> < / height> < / width> < / y-center> < / x-center> < / object-class>

Claims

1. A method for identifying items entering and leaving a warehouse, comprising: Obtaining N radio frequency identification results obtained by performing radio frequency scanning on the tags of the N first objects; Acquire at least one target image including the N first objects; Extracting M image features associated with each of the first objects from the at least one target image, where N and M are integers greater than or equal to 1; Inputting the radio frequency identification result and the M image features of each of the first objects into a target detection model to obtain a model identification result of each of the first objects, wherein the target detection model is obtained based on machine learning; A warehouse-out operation or a warehouse-in operation is performed on the first item whose radio frequency identification result is consistent with the model identification result.

2. The method according to claim 1, characterized in that Each of the first objects is used as a container to contain a second object; The step of extracting M image features associated with each of the first objects from the at least one target image includes: taking each of the first objects as a target container and performing the following operations: extracting at least one image feature of the target container and its label; In the case that the target container is made of a transparent material, at least one image feature of the second object contained in the target container is extracted through the target container.

3. The method according to claim 2, characterized in that After extracting at least one image feature of the target container and its label, the method further includes: Based on at least one image feature extracted from the target container, a material category of the target container is identified, where the material category includes a transparent material or a non-transparent material.

4. The method according to claim 2, characterized in that: The second object includes a precious metal, and the step of extracting at least one image feature of the second object contained in the target container through the target container includes: At least one of the volume feature, color feature, texture feature and shape feature of the precious metal is extracted.

5. The method according to claim 3, characterized in that: The second object includes a precious metal, the tag includes information about the precious metal, and the RFID result includes RFID information about the precious metal; The step of inputting the radio frequency identification result and the M image features of each of the first objects into the target detection model comprises: The radio frequency identification information of the precious metal, the material type of the target container, at least one image feature of the target container and its label, and at least one image feature of the precious metal are input into the target detection model.

6. The method according to any one of claims 1 to 5, characterized in that When N is greater than or equal to 2, at least two of the N first objects are stacked on top of each other.

7. The method according to claim 1, characterized in that Before inputting the radio frequency identification result and the M image features of each of the first objects into the target detection model, the method further includes training the target detection model, specifically including: Inputting the model recognition results of the target detection model to be trained on the training samples and the actual categories of the training samples into the comprehensive loss function to obtain the loss value, wherein the training samples include images of historical in-and-out items and radio frequency recognition results; Updating parameters of the target detection model based on the loss value; Among them, the comprehensive loss function includes a cross entropy loss function and its first weight, a bounding box regression loss function and its second weight, and a binary cross entropy loss function and its third weight.

8. A method for identifying items for precious metals entering and leaving a warehouse, comprising: capturing at least one target image including a precious metal container; extracting at least one image feature of the precious metal container and its label from the at least one target image; In the case where the precious metal container is made of a transparent material, at least one image feature of the precious metal contained in the precious metal container is extracted through the precious metal container; Inputting at least one image feature of the precious metal container and its label, and at least one image feature of the precious metal, into a target detection model to obtain a model recognition result of the precious metal, wherein the target detection model is obtained based on machine learning; When the model recognition result meets the preset conditions, a warehousing-out operation or a warehousing-in operation is performed on the precious metal.

9. An object identification device for entering and exiting a warehouse, comprising: A radio frequency scanning module, used to obtain N radio frequency identification results obtained by performing radio frequency scanning on the respective tags of the N first objects; A first acquisition module, used to acquire at least one target image containing the N first objects; A first extraction module, configured to extract M image features associated with each of the first objects from the at least one target image, where N and M are integers greater than or equal to 1 respectively; a first prediction module, configured to input the radio frequency identification result and the M image features of each of the first objects into a target detection model to obtain a model identification result of each of the first objects, wherein the target detection model is obtained based on machine learning; The first storage-in / out module is used to perform a storage-out operation or a storage-in operation on the first item whose radio frequency identification result is consistent with the model identification result.

10. An object identification device for precious metals entering and leaving a warehouse, comprising: A second acquisition module, used to acquire at least one target image containing a precious metal container; a second extraction module, configured to extract at least one image feature related to the precious metal container and its label from the at least one target image; A third extraction module is used to extract at least one image feature of the precious metal contained in the precious metal container through the precious metal container when the precious metal container is made of a transparent material; a second prediction module, for inputting at least one image feature of the precious metal container and its label, and at least one image feature of the precious metal, into a target detection model to obtain a model recognition result of the precious metal, wherein the target detection model is obtained based on machine learning; The second storage-in / out module is used to perform a storage-out operation or a storage-in operation on the precious metal when the model recognition result meets the preset conditions.

11. An electronic device, comprising: one or more processors; a memory for storing one or more computer programs, It is characterized in that the one or more processors execute the one or more computer programs to implement the steps of the method according to any one of claims 1 to 8.

12. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

13. A computer program product, comprising a computer program, characterized in that When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 8 are implemented.

Images