Data monitoring method and system based on image recognition

Abstract

The invention is suitable for the technical field of image recognition and translation task scheduling, and provides a data monitoring method and system based on image recognition, and the method comprises the steps: obtaining an original scheduling priority value, a historical translation record and a current image segment of a target translator when determining that the target translator has a specific equipment switching behavior, and determining the type of the current translation task. According to the method, two behavior adjustment factors, namely the equipment adaptation index and the translation input index, are introduced, so that an equipment switching scene-oriented translator scheduling priority dynamic correction mechanism is constructed. Compared with a traditional mode which only depends on static scoring or single state evaluation, the method has the advantages that the long-term adaptation trend of the translator in the cross-device task and the instant cut-in efficiency of the current task can be sensed at the same time, and the accuracy and robustness of scheduling judgment are improved.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition and translation task scheduling, and particularly relates to a data monitoring method and system based on image recognition. Background Art

[0002] In existing translation task assignment systems, fixed scheduling strategies or static scoring mechanisms are usually adopted to assign tasks to translators. Such systems often perform priority sorting based on dimensions such as the translator's language direction, historical evaluation scores, task completion rate, etc., and then achieve automatic task matching. However, existing methods generally lack the ability to perceive the current actual state of translators and cannot determine whether they are in a working state suitable for immediately accessing tasks. Especially in a multi-terminal hybrid environment, the system cannot effectively identify whether a translator has completed the adaptation to the current device, which may lead to a state deviation between scheduling and execution.

[0003] At the same time, although some technologies attempt to introduce real-time video monitoring or behavior recognition modules to obtain the online status or simple behavior changes of translators, such methods often remain at a rough recognition of "whether on duty" or "whether active", and do not establish a dynamic evaluation model from deep-level features such as behavior evolution trends and task access efficiency, nor do they establish a systematic behavior analysis mechanism for the key interference factor of "device switching" commonly found in remote or multi-terminal collaborative translation tasks. Therefore, existing technologies have significant perception blind spots and scheduling delays when dealing with complex work scenarios such as frequent cross-platform work of translators and uncertain access status. Summary of the Invention

[0004] The purpose of the present invention is to provide a data monitoring method and system based on image recognition, aiming to solve the problems raised in the background art.

[0005] The present invention is implemented as follows. A data monitoring method based on image recognition, the method includes: When it is determined that a target translator has a specific device switching behavior, obtain the original scheduling priority value, historical translation records, and current image segment of the target translator, and determine the type of the current translation task; Based on the current image segment, perform image recognition processing on the target translator to generate its current behavior label; According to the type of the current translation task, the current behavior label, and the specific device switching behavior, screen out several associated translation segments from the historical translation records; Based on the historical translation records, extract the device adaptation index of the target translator in each associated translation segment, and calculate and generate a first adjustment factor according to the time series change trend of the device adaptation index; Intelligently evaluate each associated translation segment, generate a corresponding translation input index, obtain a preset benchmark index, and calculate a second adjustment factor based on the difference between each translation input index and the preset benchmark index; Based on the first adjustment factor and the second adjustment factor, comprehensively correct the original scheduling priority value.

[0006] As a further limitation of the technical solution of the embodiment of the present invention, based on the historical translation records, extract the device adaptation index of the target translator in each associated translation segment, and the steps of calculating and generating the first adjustment factor based on the time series change trend of the device adaptation index include: Parse the historical translation records, and extract the image behavior segments of the target translator within a preset time period after completing a specific device switching behavior in each associated translation segment; Based on image recognition technology, analyze the image behavior segments, extract the behavior parameters of face orientation, lip movement rhythm, posture stability, and eye movement trajectory, calculate the convergence degree of each behavior parameter within a preset time period, generate corresponding behavior stability scores, and weight and fuse the various stability scores to generate a device adaptation index; Construct a time series based on the device adaptation indices corresponding to several associated translation segments, and use the average slope of the time series as the first adjustment factor.

[0007] As a further limitation of the technical solution of the embodiment of the present invention, the steps of intelligently evaluating each associated translation segment, generating a corresponding translation input index, obtaining a preset benchmark index, and calculating a second adjustment factor based on the difference between each translation input index and the preset benchmark index include: Parse each associated translation segment, monitor the behavior of the target translator at the start stage of the translation task, identify the behavioral change process experienced by the target translator from task startup to entering a stable translation state, and measure the duration required for the target translator to reach the translation focus state, and generate a corresponding translation input index based on this duration; Obtain a preset benchmark index, where the preset benchmark index refers to the average input duration statistically obtained from several high-quality translation behavior segments; Calculate the deviation amplitude of each translation input index compared to the preset benchmark index, and use the average value of all deviation amplitudes as the second adjustment factor.

[0008] As a further limitation of the technical solution of the embodiment of the present invention, the steps of comprehensively correcting the original scheduling priority value based on the first adjustment factor and the second adjustment factor include: Retrieve the preset priority value correction formula, and in combination with the first adjustment factor and the second adjustment factor, perform weighted correction on the original scheduling priority value to obtain the corrected scheduling priority value; Apply the corrected scheduling priority value to the task assignment process of the target translator to optimize the scheduling order or task matching decision of the target translator.

[0009] As a further limitation of the technical solution of the embodiment of the present invention, the priority value correction formula is: , where refers to the corrected scheduling priority value, refers to the original scheduling priority value, refers to the first adjustment factor, that is, the average slope of the time series, refers to the adjustment weight of the first adjustment factor, refers to the total number of associated translation segments, refers to the translation input index corresponding to the th associated translation segment, refers to the preset reference index, refers to the second adjustment factor, that is, the average value of all deviation amplitudes, and are both greater than 0.

[0010] As a further limitation of the technical solution of the embodiment of the present invention, a data monitoring system based on image recognition, the system includes: a data acquisition module, a label setting module, a data screening module, a first adjustment factor determination module, a second adjustment factor determination module, and a priority value correction module, where: The data acquisition module is used to obtain the original scheduling priority value, historical translation records, and current image segments of the target translator when it is determined that the target translator has a specific device switching behavior, and determine the type of the current translation task; The label setting module is used to perform image recognition processing on the target translator based on the current image segment to generate its current behavior label; The data screening module is used to screen out several associated translation segments from the historical translation records according to the type of the current translation task, the current behavior label, and the specific device switching behavior; The first adjustment factor determination module is used to extract the device adaptation index of the target translator in each associated translation segment based on the historical translation records, and calculate and generate the first adjustment factor according to the time series change trend of the device adaptation index; The second adjustment factor determination module is used to perform intelligent evaluation on each associated translation segment, generate the corresponding translation input index, obtain the preset reference index, and calculate the second adjustment factor based on the difference between each translation input index and the preset reference index; A priority value correction module, configured to comprehensively correct the original scheduling priority value based on a first adjustment factor and a second adjustment factor.

[0011] As a further limitation of the technical solution of the embodiment of the present invention, the first adjustment factor determination module specifically includes: An image segment extraction unit, configured to parse historical translation records, and extract image behavior segments of the target translator within a preset time period after completing a specific device switching behavior in each associated translation segment; A device adaptation index generation unit, configured to parse the image behavior segments based on image recognition technology, extract behavior parameters such as facial orientation, lip movement rhythm, posture stability, and eye movement trajectory, calculate the convergence degree of each behavior parameter within a preset time period, generate corresponding behavior stability scores, and weight and fuse the stability scores of each item to generate a device adaptation index; An average slope calculation unit, configured to construct a time series based on the device adaptation indexes corresponding to a plurality of associated translation segments, and use the average slope of the time series as the first adjustment factor.

[0012] As a further limitation of the technical solution of the embodiment of the present invention, the second adjustment factor determination module specifically includes: A translation investment index generation unit, configured to parse each associated translation segment, monitor the behavior of the target translator at the start stage of the translation task, identify the behavior change process experienced by the target translator from task startup to entering a stable translation state, measure the duration required for the target translator to reach the translation focus state, and generate a corresponding translation investment index based on the duration; A preset reference index acquisition unit, configured to acquire a preset reference index, where the preset reference index refers to the average investment duration statistically obtained from a plurality of high-quality translation behavior segments; A deviation amplitude calculation unit, configured to calculate the deviation amplitude between each translation investment index and the preset reference index, and use the average value of all deviation amplitudes as the second adjustment factor.

[0013] As a further limitation of the technical solution of the embodiment of the present invention, the priority value correction module specifically includes: A priority value correction unit, configured to retrieve a preset priority value correction formula, and combine the first adjustment factor and the second adjustment factor to perform weighted correction on the original scheduling priority value to obtain a corrected scheduling priority value; A corrected priority value application unit, configured to apply the corrected scheduling priority value to the task assignment process of the target translator for optimizing the scheduling order or task matching decision of the target translator.

[0014] As a further limitation of the technical solution of the embodiment of the present invention, the priority value correction formula is: , where refers to the corrected scheduling priority value, refers to the original scheduling priority value, refers to the first adjustment factor, i.e., the average slope of the time series, refers to the adjustment weight of the first adjustment factor, refers to the total number of associated translation segments, refers to the translation input index corresponding to the th associated translation segment, refers to the preset reference index, refers to the second adjustment factor, i.e., the average value of all deviation amplitudes, and both are greater than 0.

[0015] Compared with the prior art, the present invention has the following beneficial effects: By introducing two behavioral adjustment factors, namely the device adaptation index and the translation input index, the present invention constructs a dynamic correction mechanism for the interpreter scheduling priority facing the device switching scenario. Compared with the traditional method that only relies on static scoring or single-state evaluation, the present invention can simultaneously perceive the long-term adaptation trend of the interpreter in the cross-device task and the immediate cut-in efficiency of the current task, improving the accuracy and robustness of the scheduling judgment. This mechanism is applicable to the translation task allocation in the multi-terminal collaborative environment, can effectively avoid the behavioral instability or response delay caused by device switching, improve the intelligent level of system scheduling and the rationality of interpreter matching, and has obvious practical value and promotion prospects. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] Figure 1 is a flowchart of the method provided by the embodiment of the present invention; Figure 2 is a flowchart of generating the first adjustment factor in the method provided by the embodiment of the present invention; Figure 3 is a flowchart of generating the second adjustment factor in the method provided by the embodiment of the present invention; Figure 4 is a flowchart of correcting the original scheduling priority value in the method provided by the embodiment of the present invention; Figure 5 is an application architecture diagram of the system provided by the embodiment of the present invention; Figure 6 is a structural block diagram of the first adjustment factor determination module in the system provided by the embodiment of the present invention; Figure 7 is a structural block diagram of the second adjustment factor determination module in the system provided by the embodiment of the present invention; Figure 8It is a structural block diagram of the priority value correction module in the system provided by the embodiments of the present invention. Detailed implementation manners

[0017] In order to make the objectives, technical solutions and advantages of the present invention more clear and understandable, the present invention will be further described in detail below with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are only used to explain the present invention and are not used to limit the present invention.

[0018] Figure 1 It shows a flowchart of the method provided by the embodiments of the present invention.

[0019] Specifically, a data monitoring method based on image recognition, the method specifically includes the following steps: Step S100, when it is determined that the target interpreter has a specific device switching behavior, obtain the original scheduling priority value, historical translation records and current image segment of the target interpreter, and determine the type of the current translation task.

[0020] In the embodiments of the present invention, this case is applicable to a scheduling scenario that includes multiple interpreters and is oriented to dynamic task allocation. Usually, this system is applied to a remote or multi-platform translation task environment, where there are multiple interpreters in a standby state, waiting for the system to allocate new translation tasks. Each interpreter may be in a different terminal device (such as a desktop computer, a notebook, a tablet, a mobile terminal, etc.). The system needs to dynamically evaluate the adaptability of each interpreter according to their actual status and historical performance, and accordingly optimize the interpreter scheduling priority to achieve the optimization of the person-task matching for translation tasks.

[0021] The specific device switching behavior refers to the behavior that the target interpreter switches from one translation terminal device to another within a short period of time. For example, switching from a mobile device to a desktop terminal, or switching from an offline simultaneous interpretation system to a remote video platform, etc. Since the change of the device environment usually affects the operation fluency, behavior rhythm and task adaptability of the interpreter, introducing such a behavior recognition mechanism in the scheduling decision can effectively avoid assigning tasks to interpreters who have not yet completed adaptation or have unstable operations, thereby improving the translation quality and the reliability of the system scheduling.

[0022] The original scheduling priority value is the reference sorting basis preset by the system for the interpreter in the current state of unassigned tasks. Usually, it can be obtained by the existing translation platform management system, intelligent scheduling system or task matching module through statistical scoring methods. For example, it can be initially evaluated based on factors such as the language matching degree, historical completion rate, average response time, feedback quality score, etc. of the interpreter, and an initial priority score is generated. Such indicators have been applied in existing remote translation service platforms and have a good practical availability and engineering implementation basis.

[0023] The historical translation record is a data set formed by the target translator during the execution of past tasks, usually including information such as the historical task numbers participated by the translator, the corresponding task types, participation time, device environment, behavior trajectories (such as image frame sequences), input performance data (such as the time from the start of the task to the focused state), and device adaptation-related performance (such as operation stability after switching devices). These basic data serve as the basis for behavior modeling and trend extraction, and are the original sources for subsequent extraction of device adaptation indices and translation input indices.

[0024] The current image segment is sourced from the image acquisition device of the target translator's current terminal, usually including the video frame sequence obtained by the camera module. The system periodically acquires the translator's image data during the waiting period for task allocation, and after detecting a device switching behavior, intercepts a continuous image frame sequence for a specific time period (e.g., 30 seconds to 60 seconds after switching) as the input segment for subsequent image recognition processing and behavior feature extraction.

[0025] The type of the current translation task can be determined by the scheduling platform based on task source information, usually including classification information such as task theme, language direction, business type (such as legal, medical, technical, conference, etc.), task duration or urgency. This type of metadata is generally provided by the task initiator or user interface when the task is generated or accessed into the system, and belongs to the category of task description information already available in the existing platform, which can be used as one of the context conditions for scheduling decisions.

[0026] Furthermore, the data monitoring method based on image recognition further includes the following steps: Step S200, performing image recognition processing on the target translator based on the current image segment to generate his current behavior label.

[0027] In the embodiment of the present invention, the core objective of step S200 is: performing image recognition analysis on the target translator based on the current image segment, and accordingly judging his current working state, so as to match a behavior label with semantic description for subsequent comparison with the labels of historical translation segments and behavior trend evaluation.

[0028] This step mainly relies on existing image recognition and pose recognition technologies. The system obtains the current image segment of the target translator through the camera, that is, the video frame data within a time window after a specific device switching behavior, and uses pose estimation algorithms (such as OpenPose, MediaPipe, etc.) to extract key point parameters, including facial orientation, eye movement direction, head rotation angle, mouth opening and closing conditions, and changes in shoulder and hand positions. Further extract feature indicators such as action persistence, frequency, and amplitude in the image frame sequence.

[0029] Combining these features, the system can determine whether the target interpreter is currently in a state of stable stillness, frequent movement, turning to talk, eating, long-term head-down, etc., and match the corresponding behavior labels accordingly. The behavior labels may include, but are not limited to, the following categories: "Concentrated sitting", "Non-interactive state", "Eating", "Moving", "Inattentive", "Communicating with others", "Device adjustment in progress", "Frequent hand movements", "Away from the device", etc.

[0030] The generation of these labels depends on the classification rules or lightweight behavior classification models obtained through training, which can be constructed based on existing image samples through supervised learning. The labels themselves are not used to infer the translation quality of the interpreter, but only as a semantic expression of behavioral characteristics, serving as matching conditions for subsequent screening of historical behaviors, so as to ensure that the historical segments are highly similar to the current state of the interpreter, which helps to generate more valuable device adaptation indices and translation input indices.

[0031] Furthermore, the data monitoring method based on image recognition further includes the following steps: Step S300, according to the type of the current translation task, the current behavior label, and specific device switching behaviors, screen out several associated translation segments from the historical translation records.

[0032] In the embodiments of the present invention, the purpose of step S300 is to screen out several associated translation segments that match the current situation from the historical translation records of the target interpreter, as the data basis for subsequent calculation of the device adaptation index and the translation input index, so as to improve the pertinence and effectiveness of the behavior trend evaluation.

[0033] The specific screening process includes the following aspects: First, the system determines the semantic category to which the task belongs according to the type of the current translation task, such as legal, medical, conference, technical, etc. This task type is provided by the task metadata and is used as one of the preliminary screening conditions to screen out historical segments that are significantly inconsistent with the current task type from the historical translation records.

[0034] Secondly, the system calls the current behavior label identified in step S200 as the second layer of screening condition. The system extracts all translation segments with the same behavior label in the initial stage of task access from the historical translation records. For example, if the current label is "Concentrated sitting", only the segments with the same "Concentrated sitting" label at the beginning of the task in the history are retained to ensure the comparability of the behavior states.

[0035] Again, the system combines the currently recognized specific device switching behaviors, such as "switching from a mobile device to a desktop terminal" or "switching from an offline device to a remote platform", and further limits the translation segments in the historical translation records to only those with the same device switching type as the current one. The device switching behaviors can be matched through the task access logs or the device change records in the historical behavior records to ensure that the screening results have structural consistency in terms of device adaptability.

[0036] Finally, based on meeting the three conditions of task type matching, behavior label consistency, and device switching type compliance, the system selects multiple translation segments that meet the conditions from the historical translation records as associated translation segments, and sorts and numbers them according to the time sequence or the task quality level, providing input data for the calculation of the device adaptation index and the translation input index in the subsequent steps.

[0037] Through the above screening mechanism, the system can ensure that the extracted historical behavior samples are highly similar to the real state of the current translator, thereby improving the accuracy of behavior trend modeling and the reference value of priority correction.

[0038] Furthermore, the data monitoring method based on image recognition further includes the following steps: Step S400, based on the historical translation records, extract the device adaptation index of the target translator in each associated translation segment, and calculate and generate the first adjustment factor according to the time series change trend of the device adaptation index.

[0039] Specifically, Figure 2 shows the flowchart for generating the first adjustment factor.

[0040] Among them, based on the historical translation records, extracting the device adaptation index of the target translator in each associated translation segment and calculating and generating the first adjustment factor according to the time series change trend specifically includes the following steps: Step S401, parse the historical translation records, and extract the image behavior segments of the target translator within a preset time period after completing the specific device switching behavior in each associated translation segment; Step S402, based on image recognition technology, parse the image behavior segments, extract the behavior parameters of facial orientation, lip movement rhythm, posture stability, and eye movement trajectory, calculate the convergence degree of each behavior parameter within the preset time period, generate the corresponding behavior stability scores, and weight and fuse the stability scores of each item to generate the device adaptation index; Step S403, construct a time series based on the device adaptation indexes corresponding to several associated translation segments, and use the average slope of the time series as the first adjustment factor.

[0041] In an embodiment of the present invention, step S402 analyzes the image behavior fragments through image recognition technology to extract multiple behavior parameters of the target interpreter within a preset period of time after the specific device is switched. These behavior parameters include facial orientation, lip movement rhythm, posture stability and eye movement trajectory, which can be extracted by existing image posture estimation and feature tracking algorithms, such as recognition based on OpenPose, MediaPipe or other deep learning-based skeleton key point detection models. The system identifies the facial deflection angle, mouth opening and closing frequency, upper body posture amplitude fluctuation and eye sight changes by analyzing continuous images of key areas such as the head, eyes, mouth, shoulders, etc. in the video frame.

[0042] Within the set time period, the system constructs a parameter sequence for each behavior parameter and calculates the stability indicators of the sequence, such as the amplitude of change, fluctuation frequency, and the difference between the maximum and minimum values, to determine whether it shows a trend of change from fluctuation to stability. The system uses this to determine the degree of convergence of each type of parameter, that is, whether the behavior after the device switch tends to be stable in a short period of time. After assigning a stability score to each parameter, the multiple scores are weighted and fused according to the preset weights to generate a device adaptation index that comprehensively represents the translator's adaptability in the current device switching situation.

[0043] In step S403, the system arranges the device adaptation indexes generated in the aforementioned multiple related translation segments in chronological order to construct a time series of device adaptation indexes. The time series can reflect whether there is an obvious trend in the adaptation performance of the translator after completing the device switch in different historical tasks. The system performs linear fitting on the time series and obtains the average slope of the fitting line as the first adjustment factor.

[0044] The core reason for using the average slope as the first adjustment factor is that the slope can quantify the changing trend of the interpreter's device adaptability. If the slope is positive, it means that the interpreter's performance in the device switching situation is gradually improving, and there is a trend of improving adaptability; if the slope is negative, it indicates that the adaptability is decreasing; if the slope is close to zero, it can be regarded as stable device adaptability. In this way, the system can dynamically perceive the target interpreter's adaptation evolution trend in multiple device switching tasks, thereby providing a quantifiable and trend-oriented adjustment basis for the correction of the priority value.

[0045] Furthermore, the data monitoring method based on image recognition also includes the following steps: Step S500 , intelligently evaluate each associated translation segment, generate a corresponding translation input index, obtain a preset benchmark index, and calculate a second adjustment factor based on the difference between each translation input index and the preset benchmark index.

[0046] Specifically, Figure 3A flowchart for generating a second adjustment factor is shown.

[0047] Among them, for each associated translation segment, an intelligent evaluation is performed to generate a corresponding translation input index, a preset reference index is obtained, and based on the difference between each translation input index and the preset reference index, calculating the second adjustment factor specifically includes the following steps: Step S501: Analyze each associated translation segment, monitor the behavior of the target translator at the starting stage of the translation task, identify the behavioral change process experienced by the translator from task startup to entering a stable translation state, and measure the duration required for the translator to reach the translation focus state. Based on this duration, generate a corresponding translation input index; Step S502: Obtain a preset reference index, where the preset reference index refers to the average input duration statistically obtained from several high-quality translation behavior segments; Step S503: Calculate the deviation amplitude of each translation input index compared to the preset reference index, and take the average value of all deviation amplitudes as the second adjustment factor.

[0048] In the embodiment of the present invention, the system continuously monitors the behavior of the image behavior sequence of the associated translation segment in step S501 to identify the behavioral change process experienced by the target translator from task startup to entering a stable translation state. Specifically, the system analyzes the consecutive image frames after the start of the translation task based on image recognition technology, and extracts key action features in areas such as the head, face, eyes, mouth, shoulders, etc., including whether the face orientation is stable, whether the eyes are focused on the screen, whether the mouth opens and closes regularly, whether the upper body is stationary, etc.

[0049] During the behavior monitoring process, the system identifies the starting point where the translator's behavior gradually changes from unstable to stable by setting a set of behavior stability thresholds, such as the face offset angle being continuously lower than the set range, the mouth rhythm maintaining a fixed frequency, the line of sight continuously concentrating on the screen area, etc. When these behavior parameters first simultaneously meet the stability requirements in the image sequence, the system defines this time point as the decision node for entering the stable translation state, and calculates the time length from the start of the task to this node, that is, the duration required for the translator to complete the state transition.

[0050] This duration can be directly used as the translation input index after normalization processing. If it is necessary to improve the sensitivity of the index, the system can also introduce a weighting mechanism to perform weighted correction on factors such as the state fluctuation amplitude, fluctuation frequency, or smoothness of the transition path at different stages, so as to generate a more distinguishable input performance value.

[0051] In step S502, the system obtains a preset reference index, which is statistically obtained based on a set of historical high-quality translation behavior segments. The high-quality segments are derived from the behavioral process data of other interpreters with high scoring performance when translating the same type of tasks (i.e., the task types are the same). These interpreters usually have the characteristics of fast response speed, stable behavior entry, and good task completion evaluation, so they can be used as a reference template for evaluating the current interpreter's input state. By statistically calculating the average duration from task start to the focused state in this set of high-quality segments, the system obtains a preset reference index for comparison.

[0052] The second adjustment factor reflects the overall deviation degree between the target interpreter's current state and the efficient behavior reference. If this value is positive, it indicates that the interpreter usually takes more time than the high-quality reference interpreters to enter the translation state, and the task response speed is slow; if this value is negative, it indicates that the interpreter can enter the focused state faster and has strong input efficiency. By quantifying and averaging these deviation degrees, it can effectively measure the general input performance of the target interpreter in different tasks recently, provide an objective basis for lowering or raising the scheduling priority, and make the task allocation strategy more robust and data-driven.

[0053] Furthermore, the data monitoring method based on image recognition further includes the following steps: Step S600, comprehensively correct the original scheduling priority value based on the first adjustment factor and the second adjustment factor.

[0054] Specifically, Figure 4 shows a flowchart for correcting the original scheduling priority value.

[0055] Among them, comprehensively correcting the original scheduling priority value based on the first adjustment factor and the second adjustment factor specifically includes the following steps: Step S601, retrieve the preset priority value correction formula, and combine the first adjustment factor and the second adjustment factor to perform weighted correction on the original scheduling priority value to obtain the corrected scheduling priority value; Step S602, apply the corrected scheduling priority value to the task allocation process of the target interpreter to optimize the scheduling order or task matching decision of the target interpreter.

[0056] The priority value correction formula is: , where refers to the corrected scheduling priority value, refers to the original scheduling priority value, refers to the first adjustment factor, that is, the average slope of the time series, refers to the adjustment weight of the first adjustment factor, refers to the total number of associated translation segments, refers to the translation input index corresponding to the th associated translation segment, refers to the preset benchmark index, refers to the second adjustment factor, that is, the average value of all deviation amplitudes, refers to the adjustment weight of the second adjustment factor, and and both are greater than 0.

[0057] In the embodiments of the present invention, the first adjustment factor and the second adjustment factor are designed to jointly participate in the correction of the scheduling priority value, having obvious technical advantages and application values. The behavior dimensions measured by the two do not overlap with each other, but are highly correlated in the specific context of "device switching", and can jointly reflect the real adaptation ability of the translator from two levels of behavior trend and instant state, so as to construct a more complete, stable and dynamically responsive scheduling optimization mechanism.

[0058] First of all, the first adjustment factor reflects the change trend of the translator's adaptation ability in multiple historical device switching scenarios. Its essence is to construct a time series through the device adaptation index and calculate its average slope, so as to judge whether the behavior performance of the target translator is gradually stabilizing, improving or degrading in similar device migration tasks. This trend factor based on the historical sequence can reveal the impact of device switching on the translator's long-term behavior state. For example, for some translators, frequently switching from a mobile terminal to a desktop terminal may form a stable adaptation path, manifested as a successive increase in the adaptation index; while for other translators, there may be a repeated adjustment period after each switch, manifested as large fluctuations or even a decline in the adaptation index. The introduction of this factor enables the system to not only focus on the current performance during scheduling, but also identify in advance whether the translator has the ability to work stably across devices, thereby reducing task interruption or quality fluctuations caused by scheduling mistakes.

[0059] Secondly, the second adjustment factor evaluates the "cut-in speed" required for the translator to enter a fully stable translation state from the start of the current translation task. This factor calculates the input deviation amplitude of the translator in different tasks by comparing with the average input duration extracted from a group of high-quality translation samples. It does not rely on long-term historical trends, but focuses more on the state response ability in the "early stage of the current task", and is an important indicator for measuring the translator's instant input efficiency. After device switching, the translator often needs a certain amount of time to re-adapt to the technical environment such as the terminal interface, voice input, and auditory feedback. Therefore, this factor is extremely crucial for judging whether the translator has completed the state transition.

[0060] The reason for selecting these two factors as the basis for scheduling correction is that "device switching" itself has the following two typical characteristics: First, there is a certain degree of reconstruction of the operating environment, which disturbs the translator's behavior in the short term; Second, this kind of disturbance will not only be reflected as local response delay in each task, but may also accumulate to form a habitual impact in long-term tasks. Therefore, relying on only one perspective is not enough to comprehensively reflect the translator's scheduling adaptation ability. If only relying on the historical trend (the first adjustment factor), the immediate risk that the translator has just completed a high-intensity task and the state has declined before a certain task may be ignored; if only relying on the current input performance (the second adjustment factor), the overall device adaptation stability of the translator may be ignored, resulting in frequent improper scheduling. Therefore, the two factors are combined and used to form a dual-verification logic of "long-term behavior trend + current input efficiency", which can achieve a balance between trend prediction and real-time state response in the scheduling process, and effectively improve the reliability and context adaptability of scheduling behavior.

[0061] Furthermore, this dual-factor mechanism is not only applicable to the translator task scheduling scenario, but also has high generality in other intelligent allocation systems with multi-terminal and multi-task switching, and can be extended and applied to multiple fields such as remote collaboration, virtual customer service, and cross-device human-computer interaction. Thus, it can be seen that the dual-factor joint scheduling correction mechanism proposed by the present invention is particularly adapted to the unique working background of "device switching", and has significant advantages in terms of the depth of behavior understanding, the efficiency of system response, and the stability of algorithms.

[0062] The correction formula proposed by the present invention adopts a linear weighted exponential correction model, which has the advantages of intuitiveness, adjustability, and small computational overhead, and is particularly suitable for embedded scheduling scenarios and real-time system applications. However, the present invention does not limit the adoption of this linear structure. According to actual application requirements, the following several types of alternative or extended forms can also be designed: For example, based on proportional normalization processing, an exponential smoothing function or a piecewise non-linear mapping method can be adopted to improve the robustness to extreme values; in the context of preferring "trend-driven" or "efficiency-first", a weight adaptive adjustment mechanism can also be designed, that is and can be dynamically updated according to historical performance; a threshold control mechanism can also be introduced to make the correction behavior only trigger when the adjustment factor exceeds the set critical value, so as to avoid frequent adjustment in the fluctuation range. In addition, if the system supports the embedding of deep models, a priority correction function can also be learned through a neural network or a regression model based on multi-dimensional behavior data, so as to realize a more complex non-linear dynamic allocation logic.

[0063] Further, Figure 5 shows the application architecture diagram of the system provided by the embodiment of the present invention.

[0064] Among them, in another preferred embodiment provided by the present invention, a data monitoring system based on image recognition includes: A data acquisition module 100, configured to obtain the original scheduling priority value, historical translation records, and current image segment of the target interpreter when it is determined that the target interpreter has a specific device switching behavior, and determine the type of the current translation task.

[0065] Furthermore, the data monitoring system based on image recognition further includes: A label setting module 200, configured to perform image recognition processing on the target interpreter based on the current image segment to generate its current behavior label.

[0066] Furthermore, the data monitoring system based on image recognition further includes: A data screening module 300, configured to screen out a number of associated translation segments from the historical translation records according to the type of the current translation task, the current behavior label, and the specific device switching behavior.

[0067] Furthermore, the data monitoring system based on image recognition further includes: A first adjustment factor determination module 400, configured to extract the device adaptation index of the target interpreter in each associated translation segment based on the historical translation records, and calculate and generate a first adjustment factor according to the time series change trend of the device adaptation index.

[0068] Specifically, Figure 6 shows the structural block diagram of the first adjustment factor determination module 400 in the system provided by the embodiment of the present invention.

[0069] Among them, in the preferred embodiment provided by the present invention, the first adjustment factor determination module 400 specifically includes: An image segment extraction unit 401, configured to parse the historical translation records and extract the image behavior segments of the target interpreter within a preset time period after completing the specific device switching behavior in each associated translation segment; A device adaptation index generation unit 402, configured to parse the image behavior segments based on image recognition technology, extract the behavior parameters of facial orientation, lip movement rhythm, posture stability, and eye movement trajectory, calculate the convergence degree of each behavior parameter within a preset time period, generate the corresponding behavior stability scores, and weight and fuse the behavior stability scores to generate a device adaptation index; An average slope calculation unit 403, configured to construct a time series based on the device adaptation indices corresponding to a number of associated translation segments, and use the average slope of the time series as the first adjustment factor.

[0070] Furthermore, the data monitoring system based on image recognition further includes: The second adjustment factor determination module 500 is configured to intelligently evaluate each associated translation segment, generate a corresponding translation input index, obtain a preset benchmark index, and calculate a second adjustment factor based on the difference between each translation input index and the preset benchmark index.

[0071] Specifically, Figure 7 FIG. shows a structural block diagram of the second adjustment factor determination module 500 in the system provided by the embodiment of the present invention.

[0072] Among them, in the preferred embodiment provided by the present invention, the second adjustment factor determination module 500 specifically includes: The translation input index generation unit 501 is configured to parse each associated translation segment, monitor the behavior of the target translator at the starting stage of the translation task, identify the process of behavioral changes experienced by the translator from the start of the task to entering the stable translation state, measure the duration required for the translator to reach the translation focus state, and generate a corresponding translation input index based on the duration; The preset benchmark index acquisition unit 502 is configured to obtain a preset benchmark index, where the preset benchmark index refers to the average input duration statistically obtained from a number of high-quality translation behavior segments; The deviation amplitude calculation unit 503 is configured to calculate the deviation amplitude between each translation input index and the preset benchmark index, and use the average value of all deviation amplitudes as the second adjustment factor.

[0073] Furthermore, the data monitoring system based on image recognition further includes: The priority value correction module 600 is configured to comprehensively correct the original scheduling priority value based on the first adjustment factor and the second adjustment factor.

[0074] Specifically, Figure 8 FIG. shows a structural block diagram of the priority value correction module 600 in the system provided by the embodiment of the present invention.

[0075] Among them, in the preferred embodiment provided by the present invention, the priority value correction module 600 specifically includes: The priority value correction unit 601 is configured to retrieve a preset priority value correction formula, and combine the first adjustment factor and the second adjustment factor to perform weighted correction on the original scheduling priority value to obtain a corrected scheduling priority value; The corrected priority value application unit 602 is configured to apply the corrected scheduling priority value to the task assignment process of the target translator to optimize the scheduling order or task matching decision of the target translator.

[0076] The priority value correction formula is: , where refers to the corrected scheduling priority value, refers to the original scheduling priority value, refers to the first adjustment factor, i.e., the average slope of the time series, refers to the adjustment weight of the first adjustment factor, refers to the total number of associated translation segments, refers to the translation input index corresponding to the nth associated translation segment, refers to the preset reference index, refers to the adjustment weight of the second adjustment factor, and and both are greater than 0.

[0077] It should be understood that although the steps in the flowcharts of the embodiments of the present invention are shown in sequence according to the indication of the arrows, these steps do not necessarily execute in the order indicated by the arrows. Unless there is a clear indication in this article, the execution of these steps has no strict order limitation, and these steps can be executed in other orders. Moreover, at least a part of the steps in each embodiment may include multiple sub-steps or multiple stages. These sub-steps or stages do not necessarily execute at the same moment, but can execute at different moments. The execution order of these sub-steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or sub-steps or stages of other steps.

[0078] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The program can be stored in a non-volatile computer-readable storage medium. When the program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, storage, database, or other medium used in the various embodiments provided in the present application can include non-volatile and / or volatile memories. Non-volatile memory can include read-only memory (ROM), programmable ROM (PROM), electrically programmable ROM (EPROM), electrically erasable programmable ROM (EEPROM), or flash memory. Volatile memory can include random access memory (RAM) or external cache memory. By way of illustration and not limitation, RAM is available in many forms, such as static RAM (SRAM), dynamic RAM (DRAM), synchronous DRAM (SDRAM), double data rate SDRAM (DDR SDRAM), enhanced SDRAM (ESDRAM), synchronous link DRAM (SLDRAM), Rambus direct RAM (RDRAM), direct memory bus dynamic RAM (DRDRAM), and Rambus dynamic RAM (RDRAM), etc.

[0079] The technical features of the above embodiments can be combined arbitrarily. For the sake of concise description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope described in this specification.

[0080] The above embodiments merely represent several implementation manners of the present invention. The description is relatively specific and detailed, but it should not be construed as a limitation on the scope of the patent of the present invention. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present invention, several modifications and improvements can still be made, and these all belong to the protection scope of the present invention. Therefore, the protection scope of the patent of the present invention should be subject to the appended claims.

[0081] The above is only the preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent replacements, and improvements made within the spirit and principle of the present invention should be included in the protection scope of the present invention.

Claims

1. A data monitoring method based on image recognition, characterized in that: The method comprises: When it is determined that the target translator has a specific device switching behavior, the original scheduling priority value, historical translation records, and current image fragment of the target translator are obtained, and the type of the current translation task is determined; Based on the current image segment, perform image recognition processing on the target translator to generate its current behavior label; According to the type of the current translation task, the current behavior label, and the specific device switching behavior, several related translation segments are selected from the historical translation records; Based on the historical translation records, the device adaptation index of the target translator in each associated translation segment is extracted, and the first adjustment factor is calculated and generated according to the time series change trend of the device adaptation index; Intelligently evaluate each associated translation segment, generate a corresponding translation input index, obtain a preset benchmark index, and calculate a second adjustment factor based on the difference between each translation input index and the preset benchmark index; Based on the first adjustment factor and the second adjustment factor, the original scheduling priority value is comprehensively modified.

2. The data monitoring method based on image recognition according to claim 1, characterized in that: Based on the historical translation records, the step of extracting the device adaptation index of the target translator in each associated translation segment, and calculating and generating the first adjustment factor according to the time series change trend of the device adaptation index includes: Parse historical translation records and extract image behavior segments within a preset period of time after the target translator completes a specific device switching behavior in each associated translation segment; Based on image recognition technology, image behavior fragments are analyzed to extract behavior parameters such as facial orientation, lip movement rhythm, posture stability, and eye movement trajectory. The convergence degree of each behavior parameter within a preset time period is calculated to generate a corresponding behavior stability score, and each stability score is weighted and fused to generate a device adaptation index. A time series is constructed based on the device adaptation indexes corresponding to a number of associated translation segments, and the average slope of the time series is used as the first adjustment factor.

3. The data monitoring method based on image recognition according to claim 2 is characterized in that: The steps of intelligently evaluating each associated translation segment, generating a corresponding translation input index, obtaining a preset benchmark index, and calculating a second adjustment factor based on a difference between each translation input index and the preset benchmark index include: Analyze each relevant translation segment, monitor the behavior of the target translator at the beginning of the translation task, identify the behavioral changes from the start of the task to the stable translation state, and measure the duration required for the translator to reach the translation focus state, and generate the corresponding translation engagement index based on the duration; Obtaining a preset benchmark index, wherein the preset benchmark index refers to an average investment time calculated based on a number of high-quality translation behavior segments; The deviation amplitude of each translation input index compared with the preset benchmark index is calculated, and the average value of all the deviation amplitudes is used as the second adjustment factor.

4. The data monitoring method based on image recognition according to claim 3 is characterized in that: The step of comprehensively correcting the original scheduling priority value based on the first adjustment factor and the second adjustment factor includes: Retrieving a preset priority value correction formula, and combining the first adjustment factor and the second adjustment factor to perform weighted correction on the original scheduling priority value to obtain a corrected scheduling priority value; The corrected scheduling priority value is applied to the task allocation process of the target translator to optimize the scheduling order or task matching decision of the target translator.

5. The data monitoring method based on image recognition according to claim 4 is characterized in that: The priority value correction formula is: ,in Refers to the revised scheduling priority value, refers to the original scheduling priority value, refers to the first adjustment factor, i.e. the average slope of the time series, Refers to the adjustment weight of the first adjustment factor, Refers to the total number of associated translation segments, Refers to the The translation investment index corresponding to the associated translation segments, Refers to the preset benchmark index, refers to the second adjustment factor, which is the average of all deviation amplitudes, refers to the adjustment weight of the second adjustment factor, and and Both are greater than 0.

6. A data monitoring system based on image recognition, characterized in that: The system includes: a data acquisition module, a label setting module, a data screening module, a first adjustment factor determination module, a second adjustment factor determination module and a priority value correction module, wherein: A data acquisition module is used to obtain the original scheduling priority value, historical translation records and current image fragment of the target translator when it is determined that the target translator has a specific device switching behavior, and determine the type of the current translation task; A label setting module is used to perform image recognition processing on the target translator based on the current image segment and generate its current behavior label; A data filtering module is used to filter out several related translation segments from historical translation records according to the type of the current translation task, the current behavior label, and the specific device switching behavior; A first adjustment factor determination module is used to extract the device adaptation index of the target translator in each associated translation segment based on the historical translation records, and calculate and generate the first adjustment factor according to the time series change trend of the device adaptation index; A second adjustment factor determination module is used to perform intelligent evaluation on each associated translation segment, generate a corresponding translation input index, obtain a preset benchmark index, and calculate a second adjustment factor based on the difference between each translation input index and the preset benchmark index; The priority value correction module is used to comprehensively correct the original scheduling priority value based on the first adjustment factor and the second adjustment factor.

7. The data monitoring system based on image recognition according to claim 6 is characterized in that: The first adjustment factor determination module specifically includes: An image segment extraction unit is used to parse the historical translation records and extract image behavior segments within a preset period of time after the target translator completes a specific device switching behavior in each associated translation segment; The device adaptation index generation unit is used to analyze the image behavior fragments based on the image recognition technology, extract the behavior parameters of facial orientation, lip movement rhythm, posture stability, and eye movement trajectory, calculate the convergence degree of each behavior parameter within a preset time period, generate the corresponding behavior stability score, and weightedly fuse each stability score to generate the device adaptation index; The average slope calculation unit is used to construct a time series based on the device adaptation indexes corresponding to the plurality of associated translation segments, and use the average slope of the time series as a first adjustment factor.

8. The data monitoring system based on image recognition according to claim 7, characterized in that: The second adjustment factor determination module specifically includes: A translation engagement index generation unit is used to analyze each associated translation segment, monitor the behavior of the target translator at the beginning of the translation task, identify the behavior change process from the start of the task to the stable translation state, and measure the duration required for the translator to reach the translation focus state, and generate a corresponding translation engagement index based on the duration; A preset benchmark index obtaining unit, used to obtain a preset benchmark index, wherein the preset benchmark index refers to an average investment time calculated based on a number of high-quality translation behavior segments; The deviation amplitude calculation unit is used to calculate the deviation amplitude of each translation input index compared with a preset reference index, and use the average value of all deviation amplitudes as the second adjustment factor.

9. The data monitoring system based on image recognition according to claim 8, characterized in that: The priority value correction module specifically includes: A priority value correction unit, used to retrieve a preset priority value correction formula, and perform weighted correction on the original scheduling priority value in combination with the first adjustment factor and the second adjustment factor to obtain a corrected scheduling priority value; The modified priority value application unit is used to apply the modified scheduling priority value to the task allocation process of the target translator, so as to optimize the scheduling sequence or task matching decision of the target translator.

10. The data monitoring system based on image recognition according to claim 9, characterized in that: The priority value correction formula is: ,in Refers to the revised scheduling priority value, refers to the original scheduling priority value, refers to the first adjustment factor, i.e. the average slope of the time series, Refers to the adjustment weight of the first adjustment factor, Refers to the total number of associated translation segments, Refers to the The translation investment index corresponding to the associated translation segments, Refers to the preset benchmark index, refers to the second adjustment factor, which is the average of all deviation amplitudes, refers to the adjustment weight of the second adjustment factor, and and Both are greater than 0.

Images