Article identification method and device, storage medium and program product

Abstract

The invention provides an article identification method and device, a storage medium and a program product, and the method comprises the steps: obtaining a to-be-processed current article target, and enabling the current article target to be obtained through the target detection of a current frame image of a target region; target trajectory information of the current object target in continuous multi-frame images is determined, the continuous multi-frame images comprise the current frame image, and the target trajectory information comprises the corresponding interference suppression level of the current object target in the continuous multi-frame images; the interference suppression level is in positive correlation with the influence degree of an interferent on the current object target; calculating the proportion of image frames of which the interference suppression levels are greater than a preset level threshold value in the continuous multiple frames of images; and if the image frame proportion is smaller than a first preset proportion, carrying out loss prevention prompting on the current object target based on the interference suppression level. According to the invention, article false detection and false alarm caused by interferents are suppressed, and the article identification accuracy is improved.

Description

Item identification methods, devices, storage media and software products Technical Field

[0001] This application relates to the field of image processing technology, and in particular to an object recognition method, device, storage medium, and program product. Background Technology

[0002] With the development of smart retail technology, self-checkout loss prevention systems have been widely used in supermarkets, convenience stores, and other scenarios. Their core relies on computer vision algorithms to achieve automatic product identification and loss prevention monitoring. Currently, in self-checkout loss prevention systems, product detection tasks are typically implemented based on a general object detection framework.

[0003] However, in actual loss prevention operations, fixed and visually stable non-product objects (such as advertising stickers, cable storage boxes, disinfectant bottles, etc., hereinafter referred to as "interference objects") that are present on the countertop of self-checkout loss prevention equipment for a long time can affect the accuracy of product recognition when they enter the product detection range. For example, these interference objects may be misdetected as product targets, or the product recognition results may be incorrect because the interference objects obscure the product, thus reducing the accuracy of the self-checkout loss prevention system. Summary of the Invention

[0004] To overcome the problems existing in related technologies, this application provides an article identification method, device, storage medium, and program product, which suppresses false detections and false alarms of articles caused by interference and improves the accuracy of article identification.

[0005] In a first aspect, embodiments of this application provide an item recognition method, comprising: acquiring a current item target to be processed, wherein the current item target is obtained by target detection of a current frame image of a target region; determining target trajectory information of the current item target in a series of consecutive frames, wherein the target trajectory information includes an interference suppression level corresponding to the current item target in the series of consecutive frames, the interference suppression level being positively correlated with the degree to which the current item target is affected by interference; calculating the proportion of image frames in the series of consecutive frames where the interference suppression level is greater than a preset level threshold; and if the proportion of image frames is less than a first preset proportion, providing a damage prevention warning to the current item target based on the interference suppression level.

[0006] Secondly, embodiments of this application provide a product identification method, comprising: acquiring a product target to be processed, wherein the product target is obtained by detecting products in the current frame image of the checkout area during the checkout process; determining the target trajectory information of the product target in a series of consecutive frames, wherein the target trajectory information includes the interference suppression level corresponding to the product target in the series of consecutive frames, and the interference suppression level is positively correlated with the probability that the product target is an interfering object; calculating the proportion of image frames in the series of consecutive frames whose interference suppression level is greater than a preset level threshold; and if the proportion of image frames is less than a first preset proportion, providing a loss prevention prompt for the product target based on the interference suppression level.

[0007] Thirdly, embodiments of this application provide an item identification device, comprising:

[0008] The acquisition module is used to acquire the current item target to be processed, wherein the current item target is obtained by target detection of the current frame image of the target area;

[0009] The first determining module is used to determine the target trajectory information of the current object target in a series of consecutive frames of images. The target trajectory information includes the interference suppression level of the current object target in the series of consecutive frames of images. The interference suppression level is positively correlated with the degree to which the current object target is affected by interference.

[0010] The calculation module is used to calculate the proportion of image frames in the consecutive multi-frame images whose interference suppression level is greater than a preset level threshold;

[0011] The prompting module is used to provide a damage prevention prompt to the current object target based on the interference suppression level if the proportion of the image frame is less than a first preset proportion.

[0012] Fourthly, embodiments of this application provide an electronic device, including:

[0013] At least one processor; and

[0014] A memory that is communicatively connected to the at least one processor;

[0015] The memory stores instructions executable by the at least one processor, which is configured to execute the instructions to implement the method described in any of the above aspects.

[0016] Fifthly, embodiments of this application provide a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method described in any of the above aspects.

[0017] Sixthly, embodiments of this application provide a computer program product, including a computer program that, when executed by a processor, implements the method described in any of the above aspects.

[0018] Seventhly, embodiments of this application provide a cloud device, including:

[0019] At least one processor; and

[0020] A memory that is communicatively connected to the at least one processor;

[0021] The memory stores instructions that can be executed by the at least one processor, which, when executed by the at least one processor, cause the cloud device to perform the method described in any of the above aspects.

[0022] Eighthly, embodiments of this application provide an item identification system, including a server and a terminal, wherein the terminal and the server interact via data to implement the method described in any of the above aspects.

[0023] The item recognition method, device, storage medium, and program product provided in this application analyze the current item target detected in the current frame image within the spatiotemporal context of multiple consecutive frames. It not only tracks the target's trajectory but also assigns a quantified "interference suppression level" to each frame of the trajectory. This interference suppression level directly reflects the degree to which the current item target is affected by interfering objects in that frame image; that is, the greater the degree to which the current item target is affected by interfering objects in that frame image, the higher the corresponding interference suppression level. By calculating the proportion of image frames with high interference suppression levels and comparing it with a first preset proportion, it can automatically identify whether the target item is an interfering object. Only when the proportion of image frames with higher interference suppression levels is small is it indicated that the item is not an interfering object. At this point, a damage prevention warning for the item is triggered based on the interference suppression level, suppressing false detections and false alarms caused by interfering objects and improving the accuracy of target item recognition. Attached Figure Description

[0024] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application. It is obvious that the drawings described below are some embodiments of this application, and that those skilled in the art can obtain other drawings based on these drawings without any inventive effort.

[0025] Figure 1 is a schematic diagram of the structure of an electronic device provided in an embodiment of this application;

[0026] Figure 2 is a schematic diagram of an application scenario of an item recognition system provided in an embodiment of this application;

[0027] Figure 3 is a flowchart illustrating an article identification method provided in an embodiment of this application;

[0028] Figure 4 is a flowchart illustrating a process for determining the current state of a target area according to an embodiment of this application;

[0029] Figure 5 is a flowchart illustrating an article identification method provided in an embodiment of this application;

[0030] Figure 6 is a flowchart illustrating a product identification method provided in an embodiment of this application;

[0031] Figure 7 is a structural schematic diagram of an article recognition device provided in an embodiment of this application;

[0032] Figure 8 is a schematic diagram of the structure of a cloud device provided in an embodiment of this application.

[0033] The accompanying drawings illustrate specific embodiments of this application, which will be described in more detail below. These drawings and descriptions are not intended to limit the scope of the concept in any way, but rather to illustrate the concept of this application to those skilled in the art through reference to particular embodiments. Detailed Implementation

[0034] Exemplary embodiments will now be described in detail, examples of which are illustrated in the accompanying drawings. When the following description relates to the drawings, unless otherwise indicated, the same numbers in different drawings denote the same or similar elements. The embodiments described in the following exemplary embodiments do not represent all embodiments consistent with this application. Rather, they are merely examples of apparatuses and methods consistent with some aspects of this application as detailed in the appended claims.

[0035] In this article, the term "and / or" is used to describe the relationship between related objects. Specifically, it means that there can be three kinds of relationships. For example, A and / or B can mean: A exists alone, A and B exist at the same time, or B exists alone.

[0036] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data used for analysis, data stored, data displayed, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties. Furthermore, the collection, use and processing of the relevant data must comply with the relevant laws, regulations and standards of the relevant countries and regions, and corresponding operation entry points are provided for users to choose to authorize or refuse.

[0037] To clearly describe the technical solutions of the embodiments of this application, the terms involved in this application are first defined as follows:

[0038] ROI Align: Region of Interest Alignment.

[0039] IOU: Intersection over Union.

[0040] ID: Identifier, identification code, identifier.

[0041] SORT: Simple Online and Realtime Tracking, is a multi-target tracking algorithm.

[0042] SSIM: Structural Similarity.

[0043] ROI: Region of Interest.

[0044] The item recognition method of this application embodiment can be applied to any field that requires automatic target detection.

[0045] With the development of smart retail technology, self-checkout loss prevention systems have been widely used in supermarkets, convenience stores, and other scenarios. Their core relies on computer vision algorithms to achieve automatic product identification and loss prevention monitoring. Currently, in self-checkout loss prevention systems, product detection tasks are typically implemented based on a general object detection framework.

[0046] However, in actual loss prevention operations, fixed and visually stable non-product objects (such as advertising stickers, cable storage boxes, disinfectant bottles, etc., hereinafter referred to as "interference objects") that are present on the countertop of self-checkout loss prevention equipment for a long time can affect the accuracy of product recognition when they enter the product detection range. For example, these interference objects may be misdetected as product targets, or the product recognition results may be incorrect because the interference objects obscure the product, thus reducing the accuracy of the self-checkout loss prevention system.

[0047] In related technologies, static background modeling or manually labeled interference area masks are commonly used to filter structured interference. However, static background modeling is highly sensitive to changes in lighting and temporary occlusion, resulting in a high false suppression rate of interference. The interference area masking method is poorly adaptable to changes in the damage prevention equipment platform and has high labor costs. Furthermore, the filtering methods mentioned above lack modeling methods for the joint "spatial-appearance" features of structured interference, making it difficult to effectively distinguish between structured interference and temporarily placed goods, resulting in poor stability of the suppression methods.

[0048] To address at least one of the aforementioned problems, this application provides an item recognition scheme. The scheme analyzes a detected item target within the spatiotemporal context of multiple consecutive frames, tracking its trajectory and assigning a quantified "interference suppression level" to each frame. This level directly reflects the degree to which the item target is affected by interfering objects in that frame; the greater the interference, the higher the configured interference suppression level. By calculating the proportion of image frames with high interference suppression levels and comparing it to a first preset proportion, the scheme automatically identifies whether the target item is an interfering object. Only when the proportion of image frames with higher interference suppression levels is small is the item not considered an interfering object. At this point, a damage prevention alert is triggered based on the interference suppression level, suppressing false detections and false alarms caused by interfering objects and improving the accuracy of target item recognition.

[0049] The following detailed description of some embodiments of this application is provided in conjunction with the accompanying drawings. Where there is no conflict between the embodiments, the following embodiments and features can be combined with each other. Furthermore, the timing of the steps in the following method embodiments is merely an example and not a strict limitation.

[0050] As shown in Figure 1, this embodiment provides an electronic device 1, including at least one processor 11 and a memory 12. Figure 1 uses a single processor as an example. The processor 11 and the memory 12 are connected via a bus 10. The memory 12 stores instructions executable by the processor 11. These instructions are executed by the processor 11 to enable the electronic device 1 to perform all or part of the processes described in the following embodiments, thereby suppressing false detections and false alarms of items caused by interference and improving the accuracy of item identification.

[0051] In one embodiment, the electronic device 1 may be a self-service checkout device, or a mobile phone, tablet computer, laptop computer, desktop computer, or a large computing system composed of multiple computers.

[0052] Figure 2 is a schematic diagram of an application scenario 200 of an item recognition system provided in an embodiment of this application. As shown in Figure 2, the system includes: a server 210 and a terminal 220, wherein:

[0053] Server 210 can be a data center that provides item recognition services, such as a self-checkout service data center. In a real-world scenario, a self-checkout service data center may have multiple servers 210; Figure 2 uses one server 210 as an example.

[0054] Terminal 220 can be an electronic device that interacts with the self-service checkout data center, such as a computer, mobile phone, tablet or other device used when accessing the self-service checkout data center. There can be multiple terminals 220. Figure 2 illustrates one terminal 220 as an example.

[0055] The item recognition scheme of this application embodiment can be deployed on server 210, on terminal 220, or partially on server 210 and partially on terminal 220. The choice can be made based on actual needs in a real-world scenario, and this embodiment does not impose any limitations.

[0056] When the object recognition solution is deployed entirely or partially on server 210, an interface can be opened to terminal 220 to provide algorithm support to terminal 220.

[0057] The method provided in this application embodiment can be implemented by electronic device 1 executing corresponding software code, and is achieved through data interaction with a server. Electronic device 1 can be a local terminal device. When the method runs on a server, it can be implemented and executed based on a cloud interaction system, which includes a server and client devices.

[0058] In one possible implementation, the method provided in this application provides a graphical user interface through a terminal device, wherein the terminal device may be the aforementioned local terminal device or a client device in the aforementioned cloud interaction system.

[0059] Please refer to Figure 3, which illustrates an item recognition method according to an embodiment of this application. This method can be executed by the electronic device 1 shown in Figure 1 and can be applied to the item recognition application scenario shown in Figure 2 to suppress false detections and false alarms caused by interference, thereby improving the accuracy of item recognition. This embodiment takes a terminal as the execution end as an example, and the method includes the following steps:

[0060] Step 301: Obtain the current object target to be processed. The current object target is obtained by performing target detection on the current frame image of the target area.

[0061] In this step, the target area can refer to the self-checkout area, the current frame image can refer to the image of the area near the checkout counter taken during the self-checkout process, and the current item target can be the product target detected from the current frame image.

[0062] Optionally, a lightweight object detection model can be used as the basic architecture to perform object detection on the current frame image. The input of this object detection model is the current frame image of the cashier counter in the transaction state. The output is a set of product detection results. :

[0063]

[0064] in, The detection box of the current object target j after normalization in the current frame image (i.e., the t-th frame image) is represented by x, which is the horizontal coordinate of the detection box, y, w, and h. This represents the confidence level of the current object target box j. This represents the L2 normalized feature of the current item target j extracted by the ROI Align layer of the detection network. Indicates the current frame The number of goods to be inspected.

[0065] Step 302: Determine the target trajectory information of the current object in multiple consecutive frames of images. The target trajectory information includes the interference suppression level of the current object in multiple consecutive frames of images.

[0066] In this step, the interference suppression level is positively correlated with the likelihood that the current object target is an interfering element. That is, for current object targets that are more likely to be interfering elements, a higher interference suppression level is assigned, and this level is added to the target trajectory information, providing a data basis for distinguishing whether the current object target is an interfering element. Multiple consecutive frames can include the current frame image.

[0067] Taking a self-checkout scenario as an example, the state of a real item to be settled should be dynamic (picked up, scanned, put down), so the number of frames in the image that continuously assign a high interference suppression level to it should be limited. Conversely, a fixed interference object will continuously present a "high interference suppression level" in its trajectory. Therefore, by carrying the interference suppression level in the trajectory information, the current item target detected in the current frame image is placed in the spatiotemporal context of multiple consecutive frames for analysis. This not only tracks its trajectory but also assigns a quantified "interference suppression level" to each frame in the target trajectory information. This interference suppression level directly reflects the probability that the current item target is judged as an interference object in that frame image, thus providing a more accurate and richer information basis for accurately distinguishing whether the target item is an interference object.

[0068] In one embodiment, step 302 may specifically include: determining the current detection box and the current interference suppression level of the current object target in the current frame image; obtaining the trajectory set corresponding to each object target in the historical images of consecutive frames before the current frame image; selecting the matching historical trajectory information with the highest similarity to the current object target from the trajectory set based on the current detection box; adding the current detection box and the current interference suppression level to the matching historical trajectory information to obtain the target trajectory information of the current object target in consecutive frames of images.

[0069] In this embodiment, during continuous frame analysis, the system does not simply accumulate single-frame results. Instead, it dynamically matches the detected current item target (including its detection box position and initial interference suppression level) with the tracked item trajectories in historical images to find the most similar historical trajectory. This ensures that the state of the same item target in multiple consecutive frames is converged into a coherent trajectory. Adding information from the current frame image to this matched historical trajectory allows the generated target trajectory information to accumulate and update naturally over time. The evolution of the interference suppression level thus reflects the continuous behavioral characteristics of the item target. This enables the system to reliably distinguish between temporarily appearing goods and long-term lingering interference: for fixed interference, its trajectory continuously accumulates a high interference suppression level, forming a stable pattern; while for real goods, its trajectory dynamically changes with user operations. At the implementation level, this ensures the coherence and reliability of time-series-based interference assessment, thereby more accurately identifying and suppressing false alarms of fixed interference while ensuring that the normal notification process for real goods remains unaffected, enhancing the robustness of the entire loss prevention mechanism in real-world scenarios.

[0070] In one embodiment, determining the current interference suppression level of the current object target in the current frame image includes: calculating the first similarity between the current object target and each preset interference object in a preset interference feature library, wherein the first similarity includes the spatial location similarity and / or appearance feature similarity between the current object target and the preset interference object; determining the maximum value of the first similarity as the initial interference confidence of the current object target in the current frame image; performing temporal smoothing processing on the initial interference confidence based on a preset time sliding window to obtain the current interference confidence of the current object target in the current frame image; and determining the current interference suppression level of the current object target based on the current interference confidence, wherein the current interference suppression level is positively correlated with the current interference confidence.

[0071] In this embodiment, the interference feature library includes preset information on various preset interference objects within the target area, such as the spatial location information and appearance feature information of the preset interference objects, like the fixed position and visual features of an advertising sticker. The currently detected object target is compared with the preset interference feature library, and a first similarity score is calculated between the current object target and each preset interference object in terms of both spatial location and appearance features. The preset interference object with the highest first similarity score is most similar to the current object target; therefore, the highest first similarity score is used as the initial interference confidence score of the current object target in the current frame image. This allows the system to effectively utilize pre-recorded prior knowledge of interfering objects, thereby generating an evidence-based, quantified initial interference confidence level for the current object target at the single-frame level. This provides a direct and interpretable basis for distinguishing goods from known interfering objects. To avoid misjudgments caused by brief occlusions, lighting changes, or detection fluctuations in a single-frame image, a temporal smoothing process based on a time sliding window is further introduced. This process fuses the historical confidence information from the previous moment with the initial confidence level of the current frame, so that the output current interference confidence level no longer depends on instantaneous, sporadic similarity peaks, but reflects the interference characteristics that the current object target stably exhibits over a continuous period of time, enhancing the noise resistance and temporal consistency of the judgment results. The current interference suppression level is determined by mapping based on the current interference confidence level, ensuring the reliability of the interference level assigned to the current object target. Thus, this embodiment, from initial identification using prior knowledge to temporal fusion to smooth noise, and finally to generating a stable suppression level, works together to make the identification of known fixed interference objects not only more accurate, but also to ensure that its judgment output remains consistent and reliable in the time dimension, thereby providing a high-quality input signal for subsequent trajectory-based continuous analysis.

[0072] Taking a self-checkout scenario as an example, when the checkout counter is in a transaction state, for each product target detected in the current frame image of the checkout counter... The system performs a joint spatial-appearance matching with the interference feature library to determine whether it is a false positive of a structured interference object and outputs its interference confidence level. For commodity targets... Traverse all of the interference feature database One interference item The similarity score of their joint matching is calculated using the following formula (1):

[0073] (1)

[0074] in, Indicate product target The first similarity between the i-th preset interfering object and the interfering feature library. Indicate product target Spatial location characteristics, This represents the spatial location feature of the i-th preset interfering object in the interfering feature library, and the spatial similarity is represented by the overlap degree (IOU). The weighting coefficients representing spatial similarity can be set based on actual needs to adjust the contribution weight of spatial overlap in the first similarity score. It is the L2 normalized feature of the current item target j extracted by the ROI Align layer of the detection network, used to represent the item target. Appearance characteristics Let represent the appearance features of the i-th preset interference object, and use cosine similarity to characterize the similarity of appearance features. The weighting coefficients representing the similarity of appearance features can be set based on actual needs to adjust the contribution of appearance feature similarity in the first similarity score. .

[0075] Let the maximum first similarity between product target j and the preset interference be taken as the initial interference confidence level of product target j. This can be achieved using the following formula (2):

[0076] (2)

[0077] in, The number of interfering objects in the interference feature library is preset. It is a positive integer.

[0078] Then, based on a preset time sliding window, the initial interference confidence is processed by time smoothing to obtain the current interference confidence of the current object target in the current frame image; then, based on the current interference confidence, the current interference suppression level of the target object is determined.

[0079] In one embodiment, based on a preset time sliding window, the initial interference confidence is time-series smoothed to obtain the current interference confidence of the current object target in the current frame image. This includes: tracing back from the current frame image based on the time sliding window to determine the maximum number of consecutive frames of the target's historical images within the time sliding window, wherein the initial interference confidence of the current object target in the target's historical images is greater than a first confidence threshold; if the maximum number of consecutive frames is greater than or equal to a preset number of frames, a first average value of the initial interference confidence of the current object target in the target's historical images of the preset number of frames is calculated, and the first average value is used to determine the current interference confidence of the current object target in the current frame image; if the maximum number of consecutive frames is less than the preset number of frames but greater than 1, a second average value of the initial interference confidence of the current object target in the target's historical images of the maximum number of consecutive frames is calculated, and the second average value is used to determine the current interference confidence; if the maximum number of consecutive frames is zero, the initial interference confidence of the current object target in the current frame image is determined as the current interference confidence.

[0080] In this embodiment, a dynamic condition, "maximum number of consecutive high-confidence frames," is introduced as the basis for selecting the smoothing strategy, achieving adaptive and evidence-supported confidence fusion. Faced with single-frame interference confidence jumps caused by lighting changes, brief occlusions, or instantaneous detection fluctuations in real-world scenarios, simple moving averages may dilute crucial evidence or introduce hysteresis noise. In this embodiment, the process first backtracks within a time sliding window to identify the maximum number of consecutive frames in the target's historical image where the initial interference confidence level consistently exceeds the first confidence threshold. This captures the "consistently high suspicion" characteristic that a fixed interfering object should exhibit. Furthermore, differentiated smoothing strategies are implemented based on the deterministic differences reflected by different consecutive frame lengths.

[0081] When the maximum number of consecutive frames reaches or exceeds the preset number of frames, it indicates that the current object target has stably exhibited the characteristics of the interfering object. At this time, the first average value of the initial interference confidence in the target's historical images within the preset number of frames is taken as the current interference confidence.

[0082] When the maximum number of consecutive frames is less than the preset number of frames, but still exists, it indicates that the current object target may be in a state where interference characteristics are initially appearing or have not yet stabilized. In this case, a second average value is calculated based on the actual maximum number of consecutive frames, and this second average value determines the current confidence level of interference. This approach smooths out noise while preserving the sensitivity of its trends.

[0083] If the maximum number of consecutive frames is zero, it means that there are no historical images before the current frame where the initial interference confidence level is greater than the first confidence threshold. In this case, the current instantaneous value is directly adopted, that is, the initial interference confidence level of the current object target in the current frame image is determined as the current interference confidence level, avoiding delays in the judgment of sudden new targets. In this way, the conditional smoothing mechanism based on the continuity of evidence ensures that the final current interference confidence level fully absorbs the weight of historical valid evidence and can respond quickly to substantial changes in the state of the object target. This achieves a balance between suppressing instantaneous fluctuation interference and preventing judgment lag caused by erroneous smoothing, providing a key input that is temporally coherent and logically reliable for subsequent interference level determination and loss prevention decisions.

[0084] Taking the aforementioned self-checkout scenario as an example, based on a preset time-sliding window, the initial disturbed confidence level is smoothed over time. This allows for the use of target tracking algorithms to assign stable IDs to detected product targets. By tracing back from the current frame image using the time-sliding window, product targets are set. In the current frame The initial disturbed confidence sequence is As shown in formula (3):

[0085] (3)

[0086] in, Indicate product target In image frame The initial confidence level of the disturbance. It is an integer. The range of values ​​is [t-L+1, t]. L is the maximum number of historical image frames used during smoothing processing, and can be set based on the actual scene requirements.

[0087] The length of a valid continuous matching sequence is defined using the following formula (4). for:

[0088] (4)

[0089] in, Represent natural numbers; represents the first confidence threshold, k represents the total number of historical images traced back from the current frame, used to characterize the time sliding window. n is the frame order of the historical images traced back. Indicates the frame order of the current frame t. Indicates from the current frame Tracing back k frames, the following condition is continuously satisfied. The maximum number of consecutive frames in the target historical image.

[0090] Then, the time-smoothed commodity target is calculated using the following formula (5). Current Confidence of Disturbance in the Current Frame Image for:

[0091] (5)

[0092] in, Indicates the preset frame rate, which can be optionally... .

[0093] Furthermore, an integer attribute is defined attached to the product target j, namely the interference suppression level. This represents the degree to which the product target j is affected by structured interference in the current frame t. The current interference confidence level, after temporal smoothing, can be calculated using Equation 4. Based on this, the target of the product In the current frame The detection results are assigned an inhibition level attribute, which is expressed by the following formula (6):

[0094] (6)

[0095] in, The first suppression threshold is used to distinguish between detection targets that are less affected by interference and real goods; if This indicates that the detected product target j has a low similarity to preset interfering objects in the interference feature library, and can be determined to be a real product. Therefore, a suppression level is set. .

[0096] in This is the second suppression threshold, used to distinguish the detection target affected by interference from the preset interference; if If the similarity between the detected product target j and the preset interference is moderate, it can be determined that the detected product target is moderately affected by the structured interference, and a suppression level can be set. .

[0097] like If the similarity between the target product j and the preset interference is relatively high, then the target product can be determined as a false positive interference, and a suppression level should be set. .

[0098] Optionally, it can be assumed that the input to the object detection model in step 301 is the current frame image of the cashier counter in the transaction state. The output is a set of product detection results. :

[0099]

[0100] in, This represents the detection bounding box of the product target j after normalization of the current frame image (i.e., the t-th frame image). This represents the confidence level of the detection box for item j. This represents the L2 normalized feature of the product target j extracted through the ROI Align layer of the detection network. This indicates the number of product targets detected in the current frame.

[0101] Based on the product detection result set output in step 301, each product target in the current frame image is processed according to the above formulas (1) to (6) to obtain the extended suppression level information, which can be expressed by the following formula (7) to obtain the extended detection result set. :

[0102] (7)

[0103] in, Indicates the current frame The number of goods tested in the middle This represents the detection result information of item target j in the current frame t. This indicates the interference suppression level of commodity target j in the current frame t.

[0104] In one embodiment, obtaining the trajectory set corresponding to each object target in consecutive historical images before the current frame image includes: obtaining the historical detection box and corresponding historical interference suppression level of each object target in the consecutive historical images, generating the historical trajectory information of each object target in the consecutive historical images, and the trajectory set includes the historical trajectory information of each object target.

[0105] In this embodiment, continuous frame historical images refer to images of the target area that occurred before the current frame image. During the processing of preceding frames, not only is the location information (historical detection box) of each object target recorded, but its interference suppression state over time (historical interference suppression level) is also recorded simultaneously. These two types of information are fused to generate structured historical trajectory information, making the trajectory set constructed by the system no longer a simple sequence of motion path points, but a complete representation of each tracked object's "identity-location-state" in the spatiotemporal dimension. This provides continuous, structured, and information-rich input for the aforementioned temporal smoothing based on a time sliding window, conditional judgment based on continuous high-confidence frames, and the final determination of the persistence of interference, ensuring that the entire system's evaluation of the object's "behavioral pattern" is based on a complete and coherent chain of historical evidence.

[0106] Optionally, based on the embodiments shown in Formulas (1) to (7) above, the historical detection bounding boxes and corresponding historical interference suppression levels of each object target in the continuous frame historical image can be obtained respectively to form the historical trajectory information of each object target in the continuous frame historical image.

[0107] Optionally, in the aforementioned step of selecting the matching historical trajectory information with the highest similarity to the current item target from the trajectory set based on the current detection box, taking the aforementioned self-checkout scenario as an example, a multi-target tracking algorithm can be used to associate the product targets detected in multiple consecutive frames of images, assign a unique ID to the same product, and form a temporal trajectory. The specific process can be as follows:

[0108] First, assume the first p items were detected in the historical image of the frame, and the first p items were set as the target. The historical trajectory information of object target p in the frame historical image is as follows: Then the first The set of trajectories of p object targets in a frame of historical image is ={ From the aforementioned formula (7), the current frame can be obtained. The expanded set of target detection results is Inter-frame target association is achieved through the following steps:

[0109] (1) Calculate the association cost matrix:

[0110] For the The trajectory p of the historical image of the frame and the current product target in the current frame t (frame t). The IOU overlap is calculated as the basic matching cost, as shown in formula (8):

[0111] = (8)

[0112] in, For the first The latest detection bounding box of the trajectory p in the frame history image. For the first Frame of product target The detection frame;

[0113] Matching cost matrix Element is defined as:

[0114] (9)

[0115] (2) The optimal matching is solved using the Hungarian algorithm:

[0116] The cost matrix was analyzed using the Hungarian algorithm. Solving for the first... Trajectory p in the historical image of frames t and the target item in frame t Optimal matching pair In this context, the product targets that do not match the track trajectory in frame t are used as the initialization of the new product trajectory. Product trajectories in the historical image that do not match the product target in frame t are marked as "temporarily stored". If a trajectory in the historical image is continuous... If the detected target is not matched in the new image frame, the trajectory is determined to terminate. Optionally, This is to avoid storing excessive amounts of invalid data.

[0117] Optionally, taking the aforementioned self-checkout scenario as an example, the process of updating and generating product trajectory information can be as follows:

[0118] (1) Product trajectory initialization: For newly detected product targets (without matching trajectories), trajectory initialization is performed and a unique ID is assigned. Assuming that product target j in the current frame t does not match a historical trajectory, the new trajectory information is... It can be represented as follows:

[0119]

[0120] (2) Product trajectory update: update the historical trajectories that have been successfully matched. The detection information of the product target j in the t-th frame Add to historical trajectory In other words, assuming the historical matching trajectory information corresponding to product target j in the current frame t is... The trajectory is updated using the following formula. Obtain target trajectory information :

[0121]

[0122] (3) Representation of commodity trajectory information: the target trajectory information corresponding to the target j in multiple consecutive frames of images. It can be represented as:

[0123]

[0124] Where t is the frame order of multiple consecutive images. G is the total number of frames in a series of consecutive images.

[0125] Step 303: Calculate the proportion of image frames with an interference suppression level greater than a preset threshold in multiple consecutive frames;

[0126] In this step, the interference suppression level of the current object target is tracked across multiple consecutive frames. This interference suppression level quantifies the degree to which the current object target is affected by interfering objects (such as advertising stickers or disinfectant bottles). By calculating the proportion of image frames with interference suppression levels exceeding a preset threshold in the total consecutive frames, the persistence of the "interference characteristics" exhibited by the current object target is evaluated from a temporal perspective. The larger the proportion, the greater the likelihood that the current object target is an interfering object; the smaller the proportion, the less likely the current object target is an interfering object.

[0127] Optionally, taking the aforementioned self-service checkout scenario as an example, the product trajectory information based on the additional interference suppression level attribute implemented in the aforementioned embodiment can be calculated using the following formula (10) to determine the proportion of image frames in which the current item target has an interference suppression level greater than a preset level threshold in consecutive multi-frame images. :

[0128] (10)

[0129] Where m is the number of image frames in which the interference suppression level of the current object target is greater than the preset level threshold in multiple consecutive image frames.

[0130] Step 304: If the proportion of image frames is less than the first preset proportion, provide a damage prevention warning for the current object target based on the interference suppression level.

[0131] In this step, when the image proportion is less than the first preset proportion, it indicates that the current item target does not exhibit stable interference characteristics in most frames, but rather better matches the behavior pattern of dynamic goods. Therefore, corresponding loss prevention alerts can be triggered based on its configured interference suppression level. This allows the system to intelligently distinguish between long-term resident fixed interference and short-lived real goods: fixed interference will be continuously assigned a high interference suppression level in multiple frames, resulting in a high proportion, thus being suppressed by the system and not triggering false alarms; conversely, real goods, due to changes in state caused by manipulation, are often assigned a lower or unstable interference suppression level, resulting in a lower image proportion, thereby triggering necessary loss prevention alerts. In this way, the instantaneous judgment in single-frame detection, which is susceptible to noise, is transformed into a robust decision based on time consistency, reducing false alarms caused by false detection of fixed background objects, while ensuring timely attention to real traded goods, thus achieving more accurate and reliable loss prevention monitoring in complex retail environments.

[0132] In one embodiment, the target trajectory information includes the confidence level of the detection box of the current item target in multiple consecutive frames of images; if the proportion of image frames is less than a first preset proportion, a prompt message about the current item target is issued based on the interference suppression level, including: if the proportion of image frames is less than the first preset proportion, determining the comprehensive confidence level of the current item target in multiple consecutive frames of images based on the detection box confidence level and the interference suppression level, wherein the comprehensive confidence level is negatively correlated with the interference suppression level; if the comprehensive confidence level is greater than a second confidence threshold, and there is no transaction event information for the current item target, a prompt message about the current item target is issued.

[0133] In this embodiment, the detection box confidence reflects the reliability of the target detection algorithm in identifying the current item target. If the proportion of image frames is less than a first preset proportion, a comprehensive confidence score is generated for the current item target by fusing the detection box confidence score and the assigned interference suppression level. This ensures that the final comprehensive confidence score used for decision-making simultaneously considers both "what the current item target is" (detection box confidence) and "what the current item target is not" (interference suppression level). The higher the comprehensive confidence score, the greater the likelihood that the current item target is a product. Transaction event information includes, but is not limited to, product barcode scanning event information.

[0134] When the overall confidence level exceeds the second confidence threshold, it indicates that the current item target has a sufficiently high probability of being a commodity. Based on this, if the current item target lacks transaction event information, such as no valid matching scan event, it suggests that the current item target is likely a commodity missed during checkout. To avoid missing this commodity, a loss prevention alert is triggered. This ensures that the system only issues loss prevention alerts for items that have a high probability of being commodities but whose transaction process has not yet been completed. It achieves collaborative decision-making based on three dimensions: interference characteristics, detection reliability, and business status. This reduces false alerts caused by detection fluctuations, such as false alarms for settled goods or customer personal items, making loss prevention alerts more accurate, necessary, and consistent with the logic of actual checkout scenarios. Ultimately, while reducing the false alarm rate, it improves the detection rate of genuine missed scans and the overall credibility of the system.

[0135] Optionally, taking the aforementioned self-service checkout scenario as an example, assuming the first preset proportion is... If formula (11) is satisfied:

[0136] (11)

[0137] Then it is determined that the target trajectory information matched by the current item target may be the trajectory of the product. Selectable values .

[0138] Optionally, for target trajectory information that satisfies formula (11) Based on target trajectory information The confidence score of the detection bounding box of the target object j and the additional interference suppression level are used to calculate the overall confidence score of the current target object j in multiple consecutive frames of images. As shown in formula (12):

[0139] (12)

[0140] in, For target trajectory information The confidence weight coefficient can be configured according to the following formula (13):

[0141]

[0142] Optionally, the second confidence threshold can be multiple confidence thresholds with different levels, for example, it can be... The confidence threshold representing the high risk of missed product scanning is expressed as follows: This indicates the low-risk confidence threshold for missed product scanning. Tiered risk alerts can be generated based on the overall confidence level.

[0143] (1) If the target trajectory information No valid matching QR code scanning event, and If this indicates that the current target item j is highly likely to be a commodity, then a high-risk alarm message will be triggered.

[0144] (2) If the target trajectory information No valid matching QR code scanning event, and If the probability that the current item target j is a commodity is moderate, then a low-risk alarm message will be triggered.

[0145] (3) In other cases, it is considered normal and no alarm information needs to be triggered.

[0146] In this embodiment, the suppression decision-making mechanism is upgraded from a single-frame target to dynamic product trajectory analysis. By determining the proportion of strong suppression at the trajectory level and calculating the comprehensive confidence level, the system can effectively distinguish between temporarily placed products and structured interference, significantly reducing the false alarm rate. The trajectory-level analysis mechanism not only improves the system's robustness but also reduces computational power consumption, realizing a paradigm shift from single-frame detection to trajectory decision-making. Ultimately, risk assessment and alarms are completed based on the comprehensive confidence level of the trajectory and payment information, achieving a business closed loop.

[0147] Step 305: If the percentage of the image frame is greater than or equal to the first preset percentage, determine that the current object target is an interference object.

[0148] In this step, when the calculated image frame ratio is greater than or equal to the first preset ratio, it indicates that the current object target has consistently shown a high level of interference suppression in multiple consecutive image frames, that is, its state is highly stable in the time dimension, which is highly consistent with the behavior pattern of fixed interference objects (such as billboards or disinfectant bottles that are always placed on the table). Based on this, the current object target is directly determined to be an "interference object".

[0149] Optionally, items identified as interference targets will be actively ignored by the system or included in the background model, thus no longer being detected and tracked as suspicious targets in subsequent frames. This significantly reduces unnecessary computational resource consumption and improves the overall processing speed. In addition, the judgment result can be fed back to the interference feature library to update or enhance the feature records of the corresponding interference targets, enabling the system to have continuous self-learning and environmental adaptation capabilities. By accurately filtering out these fixed non-product targets, the system can concentrate its limited detection and prompting resources on real dynamic products, effectively reducing the false alarm rate and avoiding the interference to users caused by repeated warnings triggered by fixed objects. This improves the accuracy of loss prevention while also optimizing the human-computer interaction experience and the long-term operating performance of the system.

[0150] Taking the aforementioned self-checkout scenario as an example, assuming the first preset percentage is... The proportion of image frames in which the current target has an interference suppression level greater than a preset threshold across multiple consecutive frames is: If the following conditions are met:

[0151]

[0152] If the target trajectory information matched by the current item is determined to be the trajectory of an interfering object, then the current target item will not participate in the subsequent risk alarm resolution.

[0153] In one embodiment, before calculating the first similarity between the current object target and each preset interference object in the preset interference feature library, the method further includes: acquiring an interference object image of the target region; identifying the pixel category of each pixel in the interference object image, the pixel category including background category and interference object category; performing target detection on the interference object image to obtain candidate detection boxes of at least one candidate interference object in the interference object image; calculating the percentage of pixels in the candidate detection box that are identified as interference objects for each candidate interference object; determining the candidate interference objects whose pixel percentage is greater than a second preset percentage as preset interference objects; and constructing an interference feature library based on the preset interference objects, the interference feature library including the spatial location information and appearance feature information of at least one preset interference object.

[0154] In this embodiment, before comparing the current object target in the current frame image with the interference feature library, the interference feature library is first constructed. By acquiring actual interference images of the target area and identifying the pixel category of each pixel in the interference image (including background or interference categories), pixel-level semantic segmentation accurately distinguishes interference from the background, providing a reliable basis for subsequent screening. Then, by performing target detection on the same interference image, candidate interference objects and their detection boxes are obtained. The percentage of pixels belonging to the interference category in the candidate interference object detection box is calculated, effectively verifying the true attributes of each candidate interference object. Candidate interference objects whose interference pixels dominate (interference category pixel percentage greater than a second preset percentage) are automatically identified as preset interference objects, avoiding subjective errors and omissions from manual annotation, while ensuring that the interference objects included in the interference feature library have high authenticity and representativeness. Based on these screened preset interference objects, their spatial location and appearance feature information are extracted to construct the interference feature library, making the feature library not only closely match the distribution and visual characteristics of interference objects in the actual scene, but also easily updated by re-acquiring images as the environment changes. This reduces the manual costs of system initialization and maintenance, and provides a high-quality, high-reliability comparison benchmark for subsequent calculations of the similarity between the current target item and the interfering object. This fundamentally enhances the entire loss prevention system's ability to distinguish between fixed interfering objects and real goods, and improves the robustness of goods detection and the accuracy of loss prevention alerts.

[0155] Optionally, the step of identifying the pixel category of each pixel in the interference image may specifically include the following process:

[0156] (1) First, a semantic segmentation model is trained to extract and verify pixel-level interference features from interference images. This can be achieved using a semantic segmentation model. The training scheme for the semantic segmentation model can include: using idle-state platform videos as the training set for training the segmentation model, and using pre-labeled structured interference masks as the supervision signal. The segmentation model can use a weighted cross-entropy loss function. Specifically, as shown in formula (14):

[0157] (14)

[0158] in, This represents the total number of pixels in a sample image within the training set; This represents the total number of pixel categories in the segmentation. In this embodiment, the total number of pixel categories is... The value is 2, and the pixel categories are background and interference. Indicates the corresponding pixel category The weight coefficients of the loss function. Optionally, to enhance the learning effect of the interference class, the weight coefficients of the loss function for the interference class are set to be greater than the weight coefficients of the loss function for the background class; Indicates the first The true label of each pixel in pixel category c Indicates the first The predicted probability of a pixel in the true label of pixel class c; This is a numerical stability parameter that can be set according to actual needs.

[0159] (2) Input the interference image of the target region into the pre-trained semantic segmentation model in (1), perform inference on each frame of interference image, and generate a pixel-level interference probability map of the interference image. ,in Represents pixels The probability of being classified as a distractor in the distractor image of frame t;

[0160] For each pixel in the image of the interference Calculate the standard deviation of the interference probability in all interference images. Only when and average probability When this happens, the pixel is determined to be a stable interference pixel; This is the preset standard deviation threshold, which can be 0.1. This is a preset probability threshold, which can be 0.5.

[0161] (3) Based on the calculation in (2), the pixel-level interference mask is generated using the following formula (15). :

[0162] (15)

[0163] Optionally, during the semantic segmentation model training phase, an image synthesis enhancement strategy can be introduced. This involves pasting manually collected ROI images of typical products (such as bottled water and boxed milk) into non-interference areas such as the surface image of an idle checkout counter, generating synthetic training samples where interference and products coexist. Through this image enhancement strategy, the segmentation model can learn the semantic differences between interference and temporary products, improving its ability to distinguish between structured interference and temporary products.

[0164] Optionally, the step of performing target detection on the interference image to obtain candidate detection boxes for at least one candidate interference object present in the interference image may specifically include the following process:

[0165] (1) A lightweight target detection model is used to infer the interference image and the output is a set of candidate interference objects. :

[0166]

[0167] in, Let x represent the normalized detection box of the i-th candidate interference object, where x is the x-coordinate of the center point of the detection box, y is the y-coordinate of the center point of the detection box, w is the width of the center point of the detection box, and h is the height of the center point of the detection box. This represents the confidence level of the detection box for the i-th candidate interference object. Indicates passing through the detection network L2 normalized features of the i-th candidate interference object extracted from the layer;

[0168] Optionally, during the training phase of the object detection model, images of the cashier counter surface in an idle state can be introduced as negative samples. Through data augmentation, the model can learn the visual features of structured interference objects, thereby reducing the probability of false detection of structured interference objects on the counter surface and improving its robustness in real transaction scenarios.

[0169] (2) Smoothing the spatial features of candidate interference objects: In order to suppress the jitter of the detection box caused by illumination flicker, table reflection, etc., and improve the stability of spatial features, a smoothing strategy based on lightweight multi-target tracking and recursive Kalman filtering can be adopted, for example:

[0170] First, the SORT algorithm is used to establish trajectories for candidate interference objects with the same ID in consecutive frame interference images;

[0171] Then, for the trajectory of each candidate interference object, a Kalman filter is constructed, and its state vector is:

[0172]

[0173] The first four terms of X are the detection boxes for the i-th candidate interference object. , These are the components of the motion velocity of the i-th candidate interference object on the x-axis and the y-axis, respectively.

[0174] The state transition model adopts the assumption of uniform motion, and the state transition matrix is... The expression is as follows:

[0175]

[0176] in, This represents the time step for object detection, typically 1. , The frame rate of the video is processed for the self-checkout loss prevention system. 4 The identity matrix; 2 The identity matrix.

[0177] The expressions for the process noise covariance E1 and the observation noise covariance E2 of the Kalman filter are as follows:

[0178]

[0179]

[0180] in, Represents a diagonal matrix;

[0181] The Kalman filter output is the state estimate of the interference image in frame t. , The first four items are the original detection boxes for the i-th candidate interference object in the interference image of frame t. Result after smoothing , The stable spatial location features of the i-th candidate interference object can be used to construct an interference feature library, and the selected candidate interference objects are considered as preset interference objects. It can be used in the calculation of the first similarity in the aforementioned formula (1).

[0182] Alternatively, assume For each candidate interference object, the step of calculating the percentage of pixels identified as interference objects within the candidate detection box can include the following:

[0183] Pixel-level interference mask based on the aforementioned formula (15) The following formula (16) is used to calculate the percentage of pixels identified as interference objects in the candidate detection box:

[0184] (16)

[0185] in, Candidate detection boxes representing the i-th candidate interference object The pixel region within that area is identified as an interference object. Candidate detection boxes representing the i-th candidate interference object The total pixel area.

[0186] You can set semantic confidence. :when When the i-th candidate interference object is classified as a valid preset interference object, it is added to the interference feature; otherwise, the i-th candidate interference object is discarded.

[0187] Optionally, the step of constructing an interference feature library based on preset interferences may include performing multimodal feature fusion processing on the selected preset interferences to construct the final joint feature vector of the interferences. Specifically, this may include the following process:

[0188] 1) For the pre-selected interference, a mask-guided region attention mechanism is used for processing. As follows:

[0189] (a) Detection frame for the i-th preset interference object , This represents the original detection box for the i-th preset interference. The target detection box obtained after geometric normalization The cropping region on the feature map is defined; the detection box can be obtained by cropping using the ROI Align method. In the interference mask The area above Thus, the detection frame of the i-th preset interference object is obtained. Corresponding local interference mask features .in, This is the standard representation of the detection bounding box region after feature extraction using ROI Align. h represents the height of the ROI Align output region, and w represents the width of the ROI Align output region. The number of channels in the mask feature map.

[0190] (b) The spatial-semantic weight graph is calculated using the following formula (17):

[0191] (17)

[0192] in, It is the Sigmoid activation function. For convolution operations, for Global average pooling, Indicates the first Spatial confidence map guided by a mask corresponding to a preset interference object;

[0193] (c) Adaptive feature refinement of the spatial location features of preset interference objects based on the mask-guided spatial confidence map: In the reliable area of ​​the mask, the adopted features are... Sharpened detection features; enabled in areas where the mask is unreliable. The channel attention mechanism avoids erroneous suppression. Specifically, it can be implemented using the following formula (18):

[0194]

[0195] in, For the first Spatial location features after feature refinement of the detection bounding boxes of preset interference objects. For the process The locally normalized feature tensor extracted from the clipping region defined on the feature map is used to characterize the first... The original spatial features corresponding to the detection boxes of the preset interference objects For dot product operation, This is the attention mechanism module.

[0196] 2) Constructing an interference feature library through cross-modal interaction:

[0197] (a) will and The data is fed into the lightweight interactive attention module shown in formula (19) for fusion processing:

[0198] (19)

[0199] Where Q represents the query in the attention module, K represents the key in the attention module, and V represents the value in the attention module. and For a learnable linear projection function, for example, it can be implemented as Convolutional or fully connected layers.

[0200] (b) Combining formula (19), using the following formula (20), we obtain the fused first... Instance-level features of a preset interfering object :

[0201] (20)

[0202] Where D is the dimension of the attention latent space (typical value: or ); instance-level features It combines accurate boundary perception with strong discriminative ability.

[0203] (c) The interference feature library is modeled using the following formula (21):

[0204] (twenty one)

[0205] in, This represents the comprehensive features of the i-th preset interfering object in the feature library. This represents the spatial location characteristics of the i-th preset interference object. This indicates that the i-th preset interfering object in the interference feature library has passed through... Normalized mask-detection deep fusion features express Normalization This represents a multilayer perceptron. This indicates global average pooling. Let represent the local interference mask feature of the i-th preset interference. The above formula achieves a unified interference feature modeling that combines geometric accuracy and semantic discriminativeness.

[0206] This embodiment obtains more accurate pixel-level semantic features through a segmentation feature extraction process based on selecting stable frames. Combined with spatial features obtained from the object detection model, it introduces region attention guided by segmentation masks to construct an interference feature library containing a deep fusion feature vector of "mask-detection". Simultaneously, the feature library is dynamically updated through "idle state" initialization and "effective continuous frame verification," automatically adapting to changes in the tabletop layout to ensure the timeliness and accuracy of the feature library.

[0207] Optionally, a separate "structured interference object feature library" can be added to the overall process of the self-checkout loss prevention system. This feature library is manually uploaded and maintained by administrators through the management backend, and its content consists of pre-collected image samples of various structured interference objects (such as advertising stickers, shopping bags, etc.) that have been present on the counter for a long time. During system operation, when the product detection model outputs a target detection box, the features of the image region within the detection box are extracted and matched with the interference object sample features pre-stored in the feature library. Based on the similarity result, it is determined whether the detected target belongs to interference, and subsequent logic is executed accordingly.

[0208] In one embodiment, acquiring an interference image of a target area includes: determining the current state of the target area; if the current state is a non-transactional state, acquiring multiple frames of original image information of the target area; and selecting consecutive frame images from the multiple frames of original image information where the inter-frame pixel change is less than a preset change threshold as interference images.

[0209] In this embodiment, the non-transactional state can refer to an idle state where no one is performing a transaction. By determining whether the current state of the target area is a "non-transactional state," the timing of image acquisition is ensured to minimize interference from dynamic targets such as real goods and user hands. This allows the acquired raw images to purely reflect the inherent, static background and fixed interference objects in the scene, ensuring the relevance of the interference object data from the source. On the other hand, in the non-transactional state, the appearance and position of real fixed interference objects (such as stickers or fixed disinfectant bottles) should hardly change over time, resulting in extremely low pixel changes between consecutive frames. However, light flicker, sensor noise, or extremely slight disturbances can cause pixel changes. By selecting consecutive frames with "inter-frame pixel changes less than a preset change threshold" from multiple frames of raw image information as the final interference object images, the core physical characteristic of spatiotemporal stability that fixed interference objects should possess is fully utilized. Frames that are unstable or unrepresentative due to instantaneous interference or noise are automatically removed, retaining only those clear, stable, and consistent consecutive stable frame sequences that represent fixed interference objects. This ensures that the images of interfering objects used to build the feature library are not only clean in content but also stable in state, thus making the spatial location and appearance feature information extracted from them more consistent and reliable.

[0210] Taking the aforementioned self-checkout scenario as an example, when the checkout counter is idle, a multi-period idle state sampling strategy can be adopted to obtain images of interference objects in the checkout counter area. For example, the loss prevention system can automatically trigger the idle state sampling process by setting a time period before the self-checkout equipment opens each day (such as 30 minutes before the store opens).

[0211] (1) The sampling strategy continuously collects data at fixed time intervals (e.g., every 5 seconds). frame( The tabletop image is used to construct a set of multiple original images in the idle state.

[0212] (2) Calculate the pixel-level change measure between adjacent frames. For example, use SSIM to calculate the similarity index of two adjacent original images in terms of brightness, contrast and image structure information, and use it as a measure of the change between frames.

[0213] (3) Filtering stable frames: When consecutive Frame (e.g.) The pixel change measurement between () is lower than a preset threshold When, then determine this continuity If the countertop in the original frame is completely still, then this continuous... The original frame image is used as the interference image. Among them, the following can be selected: .

[0214] In one embodiment, determining the current state of the target area includes: acquiring the human body detection result of the target area, the transaction signal of the cash register, and the duration of no transaction; if the human body detection result is no human body, the transaction signal of the cash register is a transaction task completion signal, and the duration of no transaction is greater than a preset time threshold, determining the current state of the target area as a non-transaction state; and / or, in response to an idle state command input by the user, determining the current state of the target area as a non-transaction state.

[0215] In this embodiment, the human detection result can characterize the visual state of the target area, and the duration of no transaction can characterize the temporal continuity of the target area's state. The transaction signals of the POS device are used to characterize the state at the business logic level, including but not limited to user interaction signals with the POS device, payment success signals, or transaction cancellation signals. Interaction signals include, for example, the scanning device outputting a valid product code, or the user's touchscreen detecting a click operation.

[0216] By comprehensively considering three key dimensions—human detection results, POS transaction signals, and the duration of no transactions—a three-dimensional, mutually verifying state determination mechanism has been constructed. Specifically, a "non-transaction state" is only determined when visually confirming the absence of a human object, business logic confirming the completion or cancellation of the previous transaction, and the idle state has lasted for more than a preset time threshold. This multi-condition joint judgment method effectively avoids errors in data collection timing caused by single-dimensional misjudgments such as a person briefly leaving but items remaining on the counter, or transaction signal delays. It ensures that when image acquisition of interfering objects is initiated, the target area is indeed a clean, static scene with no human operation and no remaining goods.

[0217] Optionally, by providing a parallel or alternative path of idle state commands in response to user input, the system administrator is given the flexibility to actively intervene and control in specific situations (such as device debugging and scene reset), further enhancing the operability and adaptability of the system in actual deployment.

[0218] Figure 4 shows a flowchart illustrating the process of determining the current state of a target area according to an embodiment of this application. Taking the aforementioned self-service checkout scenario as an example, the process includes the following steps:

[0219] Step 401: The countertop status monitoring module of the cashier is activated to obtain real-time input information and build the state machine of the cashier equipment.

[0220] In this step, the core state vector of the POS device includes, but is not limited to: human target detection results, POS device transaction signals, and the duration of no transaction. The POS device state machine infers the current working state of the counter in real time through logical combinations and threshold judgments. This part can provide a decision-making basis for constructing an interference feature library and matching product target features.

[0221] Optionally, the human target detection result can be based on the human presence status. Characterization: Output from the visual sensor of the POS device. .in, This indicates that a human target has been detected in the area in front of the cashier counter; This indicates that no human target was detected in the area in front of the checkout counter.

[0222] Optionally, a transaction activity status can be constructed based on the transaction signals from the POS device. : ,in This indicates that a valid transaction has been detected at the POS terminal. This indicates that no valid transaction is currently being processed by the POS device. This status detection is achieved by continuously setting a signal from the POS device's interaction: for example, when the barcode scanner outputs a valid code, or when the touchscreen detects a click. When the POS device receives a payment success signal or a transaction cancellation signal, it sets... .

[0223] Optionally, the duration of no transactions It can be based on the current timestamp recorded by the POS system clock. The timestamp of the previous valid transaction. The calculation shows that:

[0224]

[0225] The previous valid transaction can be a transaction task with complete self-service checkout operations such as successful scanning and payment completion, and its completion timestamp is the payment completion timestamp.

[0226] Step 402: Check if an idle state command has been received from the user. If yes, proceed to step 405; otherwise, proceed to step 403.

[0227] Step 403: Determine if the current state of the checkout counter has changed. If yes, proceed to step 404; otherwise, return to step 401.

[0228] In this step, based on the three core state variables from step 401, the table state... The dynamic determination can be performed using the following formula (22):

[0229] (twenty two)

[0230] in, This represents the minimum continuous time interval without transactions required to enter an idle state; it also defines the state changes of the checkout counter as a mutual transition between the idle state and the transaction state.

[0231] Step 404: Output the tabletop status. Then proceed to step 406.

[0232] In this step, if the table surface state is determined dynamically by formula (22) If a change occurs, the process proceeds to the table state output stage. First, the currently determined table state is output. The output result;

[0233] Step 405: Manual trigger initialization. The administrator manually triggers the platform initialization through the management interface, forcibly setting the platform status to idle. Then proceed to step 406.

[0234] Step 406: Reset the core state variables of the above-mentioned POS device state machine, making... , , And maintain the current state of the tabletop. The output result. If the core state variables that satisfy the transaction state are subsequently monitored, the system will automatically exit the idle state and enter the transaction state.

[0235] In this embodiment, a dual mechanism combining "manual triggering" and "automatic monitoring" is constructed for countertop status detection. This enables effective monitoring of the countertop status of the POS equipment and updates the interference feature characteristics during the "idle state," ensuring the timeliness and accuracy of the interference feature library. By combining the segmentation feature extraction process of stable frames and the feature library construction process, the system can more accurately capture the multimodal features of structured interference, significantly overcoming the problems of poor adaptability and high labor costs of traditional manual annotation interference region masking schemes.

[0236] The aforementioned item identification method proactively models and constructs a feature library of interfering objects during idle periods at the checkout counter, capturing objects that remain stable over a long period. In transaction-state matching, a joint "space-appearance" criterion and dynamic trajectory analysis mechanism are introduced. A time-series smoothing mechanism is used to weighted average the matching confidence of consecutive frames, elevating the suppression decision from a single-frame target to the item trajectory level. This significantly suppresses false matches caused by transient interference, overcoming the sensitivity of traditional static background modeling methods to lighting / temporary occlusion. By integrating these solutions into the risk assessment module, a business closed loop is achieved. Through the trajectory-level suppression decision mechanism, the system can significantly reduce the false alarm rate caused by structured interfering objects while maintaining a high item detection rate. This improves the actual customer experience, brings economic benefits, and significantly reduces loss prevention labor costs.

[0237] Figure 5 shows a flowchart of an item recognition method provided in an embodiment of this application. Taking a self-service checkout scenario as an example, the method includes the following steps:

[0238] Step 501: Obtain the current state of the target area;

[0239] Step 502: Determine if the current state is idle. If yes, proceed to step 510; otherwise, proceed to step 503.

[0240] Step 503: Obtain the current object target to be processed. The current object target is obtained by performing target detection on the current frame image of the target area.

[0241] Step 504: Perform spatial-appearance feature matching between the current object target and each preset interference object in the preset interference feature library to determine the current interference suppression level of the current object target;

[0242] Step 505: Obtain the trajectory set corresponding to each item target in the consecutive historical images before the current frame image; select the matching historical trajectory information with the highest similarity to the current item target from the trajectory set;

[0243] Step 506: Add the current detection bounding box and current interference suppression level of the current object target in the current frame image to the matching historical trajectory information to obtain the target trajectory information of the current object target in multiple consecutive frames.

[0244] Step 507: Calculate the proportion of image frames with an interference suppression level greater than a preset threshold in multiple consecutive image frames.

[0245] Step 508: If the proportion of image frames is less than the first preset proportion, determine the comprehensive confidence of the current object target in multiple consecutive frames of images based on the detection box confidence and interference suppression level, and provide graded loss prevention alarm prompts based on the comprehensive confidence.

[0246] Step 509: If the percentage of the image frame is greater than or equal to the first preset percentage, determine that the current object target is an interference object.

[0247] Step 510: Collect multiple frames of original image information of the target area and construct an interference feature library. The interference feature library includes the spatial location information and appearance feature information of at least one preset interference object.

[0248] In this embodiment, the self-service checkout counter is first determined to be in an idle or transaction state by a counter monitoring module that combines core state vectors with manual triggering. If it is in an idle state, pixel-level semantic features of interfering objects on the counter are extracted by a target segmentation model. Stable spatial features of interfering objects are obtained by combining a lightweight target detection model and a Kalman filter smoothing strategy. Multimodal features are fused to construct a dynamically updated interfering object feature library. If it is in a transaction state, the "spatial-appearance" joint matching of the product targets detected by the target detection model with the interfering object feature library is performed. The confidence level of interference is calculated by a sliding window temporal smoothing mechanism, and a graded suppression level attribute is added to the product targets. The key innovation is to upgrade the suppression decision mechanism from single-frame target to dynamic product trajectory analysis. Based on a multi-target tracking algorithm, the single-frame detected targets are associated with product trajectories. The distribution of suppression levels in the trajectory is combined to calculate the proportion of strong suppression at the trajectory level and the overall confidence level, effectively filtering instantaneous matching noise. Finally, a loss prevention risk alarm is triggered based on the overall confidence level of the trajectory. By constructing a dynamic interference feature library and implementing a trajectory-level risk decision-making mechanism, a paradigm shift from single-frame detection to trajectory-based decision-making has been achieved. This significantly distinguishes between fixed interference and temporary goods, greatly reduces the false alarm rate of loss prevention alarms, improves the settlement accuracy and user experience of self-service checkout systems, and reduces the cost of manual review for loss prevention in stores.

[0249] For details of each step of the above method, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0250] Figure 6 shows a flowchart of a product identification method provided in an embodiment of this application. Taking a self-service checkout scenario as an example, the method includes the following steps:

[0251] Step 601: Obtain the target product to be processed. The target product is obtained by detecting products in the current frame image of the checkout area during the checkout process.

[0252] Step 602: Determine the target trajectory information of the product target in consecutive multi-frame images. The target trajectory information includes the interference suppression level of the product target in consecutive multi-frame images. The interference suppression level is positively correlated with the probability that the product target is an interference object.

[0253] Step 603: Calculate the proportion of image frames with an interference suppression level greater than a preset threshold in multiple consecutive image frames;

[0254] Step 604: If the proportion of image frames is less than the first preset proportion, provide a damage prevention warning for the target product based on the interference suppression level.

[0255] Step 605: If the percentage of the image frame is greater than or equal to the first preset percentage, the product target is determined to be an interference object.

[0256] For details of each step of the above method, please refer to the relevant descriptions in the above embodiments, which will not be repeated here.

[0257] Please refer to Figure 7, which shows an item recognition device 700 according to an embodiment of this application. This device can be applied to electronic device 1 and can be applied to the item recognition application scenario shown in Figure 2 to suppress false detections and false alarms caused by interference, thereby improving the accuracy of item recognition. The device includes: an acquisition module 701, a first determination module 702, a calculation module 703, and a prompting module 704. The functional principles of each module are as follows:

[0258] The acquisition module 701 is used to acquire the current object target to be processed. The current object target is obtained by target detection of the current frame image of the target area.

[0259] The first determining module 702 is used to determine the target trajectory information of the current object target in a series of consecutive frames of images. The target trajectory information includes the interference suppression level of the current object target in the series of consecutive frames of images. The interference suppression level is positively correlated with the degree to which the current object target is affected by interference.

[0260] The calculation module 703 is used to calculate the proportion of image frames with an interference suppression level greater than a preset level threshold in multiple consecutive image frames;

[0261] The prompt module 704 is used to provide a damage prevention prompt to the current object target based on the interference suppression level if the proportion of the image frame is less than the first preset proportion.

[0262] In one embodiment, the first determining module 702 is used to determine the current detection box and the current interference suppression level of the current object target in the current frame image; obtain the trajectory set corresponding to each object target in the historical images of consecutive frames before the current frame image; select the matching historical trajectory information with the highest similarity to the current object target from the trajectory set based on the current detection box; add the current detection box and the current interference suppression level to the matching historical trajectory information to obtain the target trajectory information of the current object target in consecutive frames of images.

[0263] In one embodiment, the first determining module 702 is specifically used to calculate the first similarity between the current object target and each preset interference object in the preset interference feature library, wherein the first similarity includes the spatial position similarity and / or appearance feature similarity between the current object target and the preset interference object; determine the maximum value of the first similarity as the initial interference confidence of the current object target in the current frame image; perform temporal smoothing processing on the initial interference confidence based on a preset time sliding window to obtain the current interference confidence of the current object target in the current frame image; and determine the current interference suppression level of the current object target based on the current interference confidence, wherein the current interference suppression level is positively correlated with the current interference confidence.

[0264] In one embodiment, the first determining module 702 is specifically configured to trace back from the current frame image based on a time sliding window to determine the maximum number of consecutive frames of the target historical image within the time sliding window, wherein the initial interference confidence of the current object target in the target historical image is greater than a first confidence threshold; if the maximum number of consecutive frames is greater than or equal to a preset number of frames, calculate the first average value of the initial interference confidence of the current object target in the target historical image of the preset number of frames, and determine the current interference confidence of the current object target in the current frame image using the first average value; if the maximum number of consecutive frames is less than the preset number of frames but greater than 1, calculate the second average value of the initial interference confidence of the current object target in the target historical image of the maximum number of consecutive frames, and determine the current interference confidence using the second average value; if the maximum number of consecutive frames is zero, determine the initial interference confidence of the current object target in the current frame image as the current interference confidence.

[0265] In one embodiment, the first determining module 702 is specifically used to obtain the historical detection box and the corresponding historical interference suppression level of each object target in the continuous frame historical image, and generate the historical trajectory information of each object target in the continuous frame historical image, the trajectory set including the historical trajectory information of each object target.

[0266] In one embodiment, the device further includes: a construction module, configured to: acquire an image of an interfering object in a target region before calculating a first similarity between the current object target and each preset interfering object in a preset interfering feature library; identify the pixel category of each pixel in the interfering object image, the pixel category including background and interfering object categories; perform target detection on the interfering object image to obtain a candidate detection box for at least one candidate interfering object present in the interfering object image; calculate the percentage of pixels identified as interfering objects in the candidate detection box for each candidate interfering object; determine the candidate interfering object whose pixel percentage is greater than a second preset percentage as a preset interfering object; and construct an interfering feature library based on the preset interfering objects, the interfering feature library including the spatial location information and appearance feature information of at least one preset interfering object.

[0267] In one embodiment, the construction module is specifically used to determine the current state of the target area; if the current state is a non-transactional state, it collects multiple frames of original image information of the target area; and selects continuous frame images from the multiple frames of original image information whose inter-frame pixel changes are less than a preset change threshold as interference images.

[0268] In one embodiment, the construction module is specifically used to obtain the human body detection result of the target area, the transaction signal of the cash register, and the duration of no transaction; if the human body detection result is no human body object, the transaction signal of the cash register is the transaction task end signal, and the duration of no transaction is greater than a preset time threshold, the current state of the target area is determined to be a non-transaction state.

[0269] In one embodiment, the device further includes a response module for determining, in response to a user-input idle state command, that the current state of the target area is a non-transactional state.

[0270] In one embodiment, the target trajectory information includes the detection box confidence of the current item target in multiple consecutive frames of images; the prompting module 704 is specifically used to determine the comprehensive confidence of the current item target in multiple consecutive frames of images based on the detection box confidence and the interference suppression level if the image frame ratio is less than a first preset ratio, the comprehensive confidence is negatively correlated with the interference suppression level; if the comprehensive confidence is greater than a second confidence threshold and there is no transaction event information for the current item target, a prompting message about the current item target is issued.

[0271] In one embodiment, the device further includes a second determining module, configured to determine that the current object target is an interfering object if the percentage of the image frame is greater than or equal to a first preset percentage.

[0272] For a detailed description of the above-mentioned item recognition device 700, please refer to the description of the relevant method steps in the above embodiments. The implementation principle and technical effect are similar, and will not be repeated here.

[0273] Figure 8 is a schematic diagram of the structure of a cloud device 80 provided in an exemplary embodiment of this application. The cloud device 80 can be used to run the methods provided in any of the above embodiments. As shown in Figure 8, the cloud device 80 may include a memory 804 and at least one processor 805, with one processor being used as an example in Figure 8.

[0274] The memory 804 is used to store computer programs and can be configured to store various other data to support operations on the cloud device 80. The memory 804 may be object storage (OSS).

[0275] The memory 804 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as static random access memory (SRAM), electrically erasable programmable read-only memory (EEPROM), erasable programmable read-only memory (EPROM), programmable read-only memory (PROM), read-only memory (ROM), magnetic storage, flash memory, magnetic disk or optical disk.

[0276] The processor 805, coupled to the memory 804, is used to execute the computer program in the memory 804 to implement the solution provided in any of the above method embodiments. The specific functions and technical effects that can be achieved will not be elaborated here.

[0277] Furthermore, as shown in Figure 8, the cloud device also includes other components such as a firewall 801, a load balancer 802, a communication component 806, and a power supply component 803. Figure 8 only schematically shows some of the components and does not imply that the cloud device includes only the components shown in Figure 8.

[0278] In one embodiment, the communication component 806 in FIG8 is configured to facilitate wired or wireless communication between the device containing the communication component 806 and other devices. The device containing the communication component 806 can access wireless networks based on communication standards, such as WiFi, 2G, 3G, 4G, LTE (Long Term Evolution), 5G, or combinations thereof. In an exemplary embodiment, the communication component 806 receives broadcast signals or broadcast-related information from an external broadcast management system via a broadcast channel. In an exemplary embodiment, the communication component 806 also includes a Near Field Communication (NFC) module to facilitate short-range communication. For example, the NFC module may be implemented based on Radio Frequency Identification (RFID), Infrared Data Association (IrDA), Ultra Wide Band (UWB), Bluetooth, and other technologies.

[0279] In one embodiment, the power supply component 803 of FIG8 provides power to various components of the device in which the power supply component 803 is located. The power supply component 803 may include a power management system, one or more power supplies, and other components associated with generating, managing, and distributing power to the device in which the power supply component is located.

[0280] This application also provides a computer-readable storage medium storing computer-executable instructions, which, when executed by a processor, implement the method of any of the foregoing embodiments.

[0281] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the method of any of the foregoing embodiments.

[0282] In the several embodiments provided in this application, it should be understood that the disclosed devices and methods can be implemented in other ways. For example, the device embodiments described above are merely illustrative. For instance, the division of modules is only a logical functional division, and there may be other division methods in actual implementation. For example, multiple modules may be combined or integrated into another system, or some features may be ignored or not executed.

[0283] The integrated modules described above, implemented as software functional modules, can be stored in a computer-readable storage medium. These software functional modules, stored in a storage medium, include several instructions to cause a computer device (which may be a personal computer, server, or network device, etc.) or processor to execute some steps of the methods of the various embodiments of this application.

[0284] It should be understood that the aforementioned processor can be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), etc. A general-purpose processor can be a microprocessor or any conventional processor. The steps of the method disclosed in the application can be directly manifested as being executed by a hardware processor, or executed by a combination of hardware and software modules within the processor. The memory may include high-speed RAM (Random Access Memory), and may also include non-volatile memory (NVM), such as at least one disk storage device, and may also be a USB flash drive, external hard drive, read-only memory, disk, or optical disc, etc.

[0285] The aforementioned storage media can be implemented from any type of volatile or non-volatile storage device or a combination thereof, such as Static Random-Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk. The storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0286] An exemplary storage medium is coupled to a processor, enabling the processor to read information from and write information to the storage medium. Alternatively, the storage medium can be an integral part of the processor. Both the processor and the storage medium can reside in an Application Specific Integrated Circuit (ASIC). Alternatively, the processor and storage medium can exist as discrete components in an electronic device or host device.

[0287] It should be noted that, in this document, the terms "comprising," "including," or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such a process, method, or apparatus. Unless otherwise specified, an element defined by the phrase "comprising one..." does not exclude the presence of other identical elements in the process, method, or apparatus that includes that element.

[0288] The sequence numbers of the embodiments in this application are for descriptive purposes only and do not represent the superiority or inferiority of the embodiments.

[0289] Through the above description of the embodiments, those skilled in the art can clearly understand that the methods of the above embodiments can be implemented by means of software plus necessary general-purpose hardware platforms. Of course, they can also be implemented by hardware, but in many cases the former is a better implementation method. Based on this understanding, the technical solution of this application, in essence, or the part that contributes to the prior art, can be embodied in the form of a software product. This computer software product is stored in a storage medium (such as ROM / RAM, magnetic disk, optical disk) and includes several instructions to cause a terminal device (which may be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods of the various embodiments of this application.

[0290] The collection, storage, use, processing, transmission, provision, and disclosure of user data and other information involved in the technical solution of this application all comply with the provisions of relevant laws and regulations and do not violate public order and good morals.

[0291] The above are merely preferred embodiments of this application and do not limit the patent scope of this application. Any equivalent structural or procedural transformations made using the content of this application's specification and drawings, or direct or indirect applications in other related technical fields, are similarly included within the patent protection scope of this application.

Claims

1. An item recognition method, characterized in that, include: The current object target to be processed is obtained by performing target detection on the current frame image of the target region; The target trajectory information of the current object target in a series of consecutive frames of images is determined. The target trajectory information includes the interference suppression level of the current object target in the series of consecutive frames of images. The interference suppression level is positively correlated with the degree of interference affecting the current object target. The proportion of image frames in the series of consecutive frames with the interference suppression level greater than a preset level threshold is calculated. If the proportion of image frames is less than a first preset proportion, a damage prevention warning is given to the current object target based on the interference suppression level.

2. The method according to claim 1, characterized in that, Determining the target trajectory information of the current object target in consecutive multi-frame images includes: determining the current detection box and the current interference suppression level of the current object target in the current frame image; obtaining the trajectory set corresponding to each object target in consecutive historical frames before the current frame image; selecting the matching historical trajectory information with the highest similarity to the current object target from the trajectory set based on the current detection box; adding the current detection box and the current interference suppression level to the matching historical trajectory information to obtain the target trajectory information of the current object target in consecutive multi-frame images.

3. The method according to claim 2, characterized in that, Determining the current interference suppression level of the current object target in the current frame image includes: calculating a first similarity between the current object target and each preset interference object in a preset interference feature library, wherein the first similarity includes spatial location similarity and / or appearance feature similarity between the current object target and the preset interference object; determining the maximum value of the first similarity as the initial interference confidence level of the current object target in the current frame image; performing temporal smoothing processing on the initial interference confidence level based on a preset time sliding window to obtain the current interference confidence level of the current object target in the current frame image; and determining the current interference suppression level of the current object target based on the current interference confidence level, wherein the current interference suppression level is positively correlated with the current interference confidence level.

4. The method according to claim 3, characterized in that, The step of performing time-series smoothing on the initial interference confidence based on a preset time sliding window to obtain the current interference confidence of the current object target in the current frame image includes: tracing back from the current frame image based on the time sliding window to determine the maximum number of consecutive frames of the target historical images within the time sliding window, wherein the initial interference confidence of the current object target in the target historical images is greater than a first confidence threshold; if the maximum number of consecutive frames is greater than or equal to a preset number of frames, calculating a first average value of the initial interference confidence corresponding to the current object target in the target historical images of the preset number of frames, and determining the current interference confidence of the current object target in the current frame image using the first average value; if the maximum number of consecutive frames is less than the preset number of frames but greater than 1, calculating a second average value of the initial interference confidence corresponding to the current object target in the target historical images of the maximum number of consecutive frames, and determining the current interference confidence using the second average value; if the maximum number of consecutive frames is zero, determining the initial interference confidence of the current object target in the current frame image as the current interference confidence.

5. The method according to claim 2, characterized in that, The step of obtaining the trajectory set corresponding to each object target in the consecutive historical images preceding the current frame image includes: obtaining the historical detection box and the corresponding historical interference suppression level of each object target in the consecutive historical images, generating historical trajectory information of each object target in the consecutive historical images, wherein the trajectory set includes the historical trajectory information of each object target.

6. The method according to claim 3, characterized in that, Before calculating the first similarity between the current target item and each preset interference object in the preset interference feature library, the method further includes: acquiring an interference object image of the target region; identifying the pixel category of each pixel in the interference object image, the pixel category including background category and interference object category; performing target detection on the interference object image to obtain candidate detection boxes of at least one candidate interference object in the interference object image; calculating the percentage of pixels in the candidate detection box identified as interference object for each candidate interference object; determining the candidate interference object whose pixel percentage is greater than a second preset percentage as a preset interference object; and constructing the interference feature library based on the preset interference objects, the interference feature library including the spatial location information and appearance feature information of at least one preset interference object.

7. The method according to claim 6, characterized in that, The step of acquiring the interference image of the target area includes: determining the current state of the target area; if the current state is a non-transactional state, acquiring multiple frames of original image information of the target area; and selecting consecutive frame images from the multiple frames of original image information whose inter-frame pixel change is less than a preset change threshold as the interference image.

8. The method according to claim 7, characterized in that, Determining the current state of the target area includes: acquiring the human detection result of the target area, the transaction signal of the cash register, and the duration of no transaction; if the human detection result indicates no human object, the transaction signal of the cash register is a transaction task completion signal, and the duration of no transaction is greater than a preset time threshold, determining the current state of the target area as a non-transactional state; and / or, in response to an idle state command input by the user, determining the current state of the target area as a non-transactional state.

9. The method according to claim 1, characterized in that, The target trajectory information includes the detection box confidence of the current item target in the consecutive multi-frame images; the step of providing a loss prevention warning to the current item target based on the interference suppression level if the image frame ratio is less than a first preset ratio includes: if the image frame ratio is less than the first preset ratio, determining the comprehensive confidence of the current item target in the consecutive multi-frame images based on the detection box confidence and the interference suppression level, wherein the comprehensive confidence is negatively correlated with the interference suppression level; if the comprehensive confidence is greater than a second confidence threshold and there is no transaction event information for the current item target, issuing a warning message about the current item target.

10. The method according to claim 1, characterized in that, Also includes: If the percentage of the image frame is greater than or equal to the first preset percentage, the current object target is determined to be an interference object.

11. A product identification method, characterized in that, include: Obtain the target product to be processed, which is obtained by detecting products in the current frame image of the checkout area during the checkout process; The target trajectory information of the product target in a series of consecutive frames is determined. The target trajectory information includes the interference suppression level of the product target in the series of consecutive frames. The interference suppression level is positively correlated with the probability that the product target is an interference object. The proportion of image frames in the series of consecutive frames with the interference suppression level greater than a preset level threshold is calculated. If the proportion of image frames is less than a first preset proportion, a damage prevention warning is given to the product target based on the interference suppression level.

12. An electronic device, characterized in that, include: At least one processor; And a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, the processor being configured to execute the instructions to implement the method of any one of claims 1-11.

13. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores computer-executable instructions, which, when executed by a processor, implement the method as described in any one of claims 1-11.

14. A computer program product, characterized in that, Includes a computer program that, when executed by a processor, implements the method as described in any one of claims 1-11.

Images