Method and System for Generating Training Plan for Security Inspection Image Judgment Personnel

Abstract

A method for generating a training plan for security inspection image interpreters is provided, including: obtaining operation process data in a security inspection image interpretation system, where the operation process data includes image interpretation interface display data and image interpreter operation data; performing data processing on the obtained operation process data, and the data processing includes at least one of cleaning, screening, and feature extraction; for the operation process data after data processing, classifying and calibrating according to the characteristics of different operation process data to establish a data model; performing an initial image interpretation behavior identification on the trainee to generate the trainee's initial image interpretation process data; substituting the trainee's initial image interpretation process data into the data model, classifying and calibrating the trainee's image interpretation process to obtain the type of the trainee; and generating the trainee's training plan based on the type of the trainee.

Description

Technical Field

[0001] The present invention relates to the field of security inspection, and more particularly to a method for generating a training program for security inspection judges, a system thereof, an electronic device, and a computer-readable medium. Background Art

[0002] Currently, in the security inspection field, image interpreter training generally utilizes the same question bank, which is composed of a combination of collected historical image data or artificially generated image data for trainees to learn, practice, and take exams. Image data can be divided into purely on-site images, purely artificially generated image data (laboratory scans of luggage and parcels, simulations), and image data synthesized by adding artificially added suspects to the background of on-site images. When students complete the test, the system can randomly generate questions of varying difficulty levels, and the difficulty of the questions can be dynamically adjusted based on the accuracy rate. Training effectiveness is measured by having trainees take exams, and the system scores and determines whether the trainees have passed.

[0003] This existing training method for image judges has the following problems. The training content and approach are the same for all trainees, and no training program is tailored to the actual work of judges. Different people have different habits and approaches to image interpretation, so the training format and content should be tailored accordingly; one training program may not be suitable for everyone. The training content is primarily developed based on the training team's thinking, which has certain limitations. Training often relies on exercises. These rigid questions often fail to reflect the real-world context of image interpretation and the judge's response process, preventing judges from fully benefiting from the training and effectively improving their professional skills in the workplace. Some training programs also use pre-training grading exams to determine the level of trainees and assign them to training programs at different levels. Like training questions, grading exams cannot guarantee diversity and accuracy, making such tiered training programs inaccurate and unfair. Finally, judging trainees' qualifications based on these exams lacks persuasiveness. In most cases, training time is limited. To ensure effective training, it is necessary to quickly impart image interpretation experience to trainees so they can quickly master the skills, but current training programs are not yet able to do this. Training is not a one-dimensional task and should not be conducted based on only one standard, but should evolve into more diversified training. Summary of the Invention

[0004] Existing image judge training does not make good use of the large amount of operational process data accumulated in the centralized image judgment system, and does not provide training tailored to the specific circumstances of each trainee. To address the above issues, the present invention proposes a method and system for more effectively generating training programs for security image judges.

[0005] In the first aspect of the present invention, a method for generating a training plan for security inspection image interpreters is provided, including: obtaining operation process data in a security inspection image interpretation system, where the operation process data includes image interpretation interface display data and image interpreter operation data; performing data processing on the obtained operation process data, and the data processing includes at least one of cleaning, screening, and feature extraction; for the operation process data after data processing, classifying and calibrating according to the characteristics of different operation process data to establish a data model; performing an initial image interpretation behavior identification on a trainee to generate the trainee's initial image interpretation process data; substituting the trainee's initial image interpretation process data into the data model, classifying and calibrating the trainee's image interpretation process to obtain the type of the trainee; and generating a training plan for the trainee based on the type of the trainee.

[0006] In the second aspect of the present invention, a system for generating a training plan for security inspection image interpreters is provided. The system includes: an obtaining unit for obtaining operation process data in a security inspection image interpretation system, where the operation process data includes image interpretation interface display data and image interpreter operation data; a processing unit for performing data processing on the obtained operation process data, and the data processing includes at least one of cleaning, screening, and feature extraction; a modeling unit for classifying and calibrating according to the characteristics of different operation process data after data processing to establish a data model; an identification unit for performing an initial image interpretation behavior identification on a trainee to generate the trainee's initial image interpretation process data; a substitution unit for substituting the trainee's initial image interpretation process data into the data model, classifying and calibrating the trainee's image interpretation process to obtain the type of the trainee; and a generating unit for generating a training plan for the trainee based on the type of the trainee.

[0007] In the third aspect of the present invention, an electronic device is provided, including: one or more processors; a storage device for storing executable instructions, and when the executable instructions are executed by the processors, the method according to the first aspect of the present invention is implemented.

[0008] In the fourth aspect of the present invention, a computer-readable medium is provided, on which executable instructions are stored, and when the instructions are executed by a processor, the method according to the first aspect of the present invention is implemented.

[0009] Based on the various aspects provided by the present invention, a training plan for security inspection image interpreters can be generated more effectively. This training plan makes good use of a large amount of operation process data accumulated in the centralized image interpretation system and conducts training according to the specific situation of each trainee. BRIEF DESCRIPTION OF THE DRAWINGS

[0010] Figure 1 FIG. shows a schematic block diagram of a system for generating a training plan for security inspection image interpreters according to an embodiment of the present invention.

[0011] Figure 2 A conceptual diagram of a system for generating a security inspection image interpretation operator training plan according to an embodiment of the present invention is shown.

[0012] Figure 3 Another conceptual diagram of a system for generating a security inspection image interpretation operator training plan according to an embodiment of the present invention is shown.

[0013] Figure 4 A flowchart of a method for generating a security inspection image interpretation operator training plan according to an embodiment of the present invention is shown. Detailed implementation manners

[0014] Specific embodiments of the present invention will be described in detail below. It should be noted that the embodiments described here are only for illustrative purposes and do not limit the present invention. In the following description, in order to provide a thorough understanding of the present invention, a large number of specific details are set forth. However, it will be apparent to those of ordinary skill in the art that: the present invention does not have to employ these specific details. In other instances, well-known circuits, materials, or methods have not been described in detail in order to avoid obscuring the present invention.

[0015] Throughout the specification, the mention of "one embodiment", "an embodiment", "one example", or "an example" means that: the specific features, structures, or characteristics described in connection with that embodiment or example are included in at least one embodiment of the present invention. Thus, the phrases "in one embodiment", "in an embodiment", "one example", or "an example" appearing throughout the specification do not necessarily all refer to the same embodiment or example. In addition, the specific features, structures, or characteristics may be combined in any suitable combination and / or sub - combination in one or more embodiments or examples.

[0016] It should be understood that when an element is referred to as being "coupled to" or "connected to" another element, it can be directly coupled or connected to the other element or there can be intervening elements. In contrast, when an element is referred to as being "directly coupled to" or "directly connected to" another element, there are no intervening elements.

[0017] In addition, the term "and / or" used here includes any and all combinations of one or more of the associated listed items.

[0018] It will be understood that a noun in the singular form corresponding to a term may include one or more things, unless the relevant context clearly indicates otherwise. As used herein, each of the phrases such as "A or B", "at least one of A and B", "at least one of A or B", "A, B or C", "at least one of A, B and C", and "at least one of A, B or C" may include all possible combinations of the items listed together in the corresponding one of the plurality of phrases. As used herein, terms such as "first" and "second" or "1st" and "2nd" may be used to simply distinguish corresponding components from another component, and do not otherwise limit the components (e.g., importance or order).

[0019] As used herein, the term "module" may include a unit implemented in hardware, software, or firmware, and may be used interchangeably with other terms (e.g., "logic", "logic block", "part", or "circuit"). A module may be a single integrated component adapted to perform one or more functions or the smallest unit or part of the single integrated component. For example, according to an embodiment, a module may be implemented in the form of an application specific integrated circuit (ASIC).

[0020] It should be understood that the various embodiments of the present disclosure and the terms used therein are not intended to limit the technical features set forth herein to specific embodiments, but include various changes, equivalent forms, or alternative forms for the corresponding embodiments. Unless otherwise clearly defined herein, all terms will be given their broadest possible interpretation, including the meanings implied in the specification and understood by those skilled in the art and / or defined in dictionaries, treatises, etc.

[0021] In addition, those of ordinary skill in the art should understand that the drawings provided herein are for illustrative purposes only, and the drawings are not necessarily drawn to scale. For the description of the drawings, like reference numerals may be used to refer to like or related elements. The present disclosure will be described below with reference to the drawings by way of example.

[0022] The present invention makes full use of a large amount of operation process data accumulated in the centralized image judgment system, and formulates more suitable training and training plans for security inspection image judgment personnel under different business scenarios and different business capabilities through big data technologies such as subscription, feature extraction, cleaning, screening, and analysis, adapting measures to local conditions and teaching students in accordance with their aptitude, so as to achieve the goal of improving the training efficiency and quality.

[0023] Figure 1 A schematic block diagram of a system 100 for generating a training plan for security inspection image judgment personnel according to an embodiment of the present invention is shown. The following will refer to Figure 1 Describe in detail a system 100 for generating a training plan for security inspection image judgment personnel according to an embodiment of the present invention.

[0024] The system 100 for generating a training program for security inspection judges according to an embodiment of the present invention includes:

[0025] An acquisition unit 101 acquires operation process data in the security inspection image interpretation system, wherein the operation process data includes image interpretation interface display data and image interpreter operation data;

[0026] The processing unit 102 performs data processing on the acquired operation process data, wherein the data processing includes at least one of cleaning, screening, and feature extraction;

[0027] The modeling unit 103 classifies and calibrates the processed operation process data according to the characteristics of different operation process data to establish a data model;

[0028] The identification unit 104 performs an initial image recognition behavior identification on the trainee to generate initial image recognition process data of the trainee;

[0029] a substitution unit 105, substituting the student's initial image recognition process data into the data model, classifying and calibrating the student's image recognition process to obtain the student's type; and

[0030] The generating unit 106 generates a training plan for the trainee based on the type of the trainee.

[0031] First, the system for generating a training plan for security inspectors, according to an embodiment of the present invention, determines the scope of operational process data to be collected, supplemented by supporting information collection measures, records and stores relevant data within the system, and continuously updates it based on actual needs and analysis results. The following table lists some representative data or dimensions.

[0032]

[0033]

[0034]

[0035]

[0036]

[0037]

[0038] Determining the aforementioned data categories will help the system better establish data models. The system for generating security inspection judge training programs according to an embodiment of the present invention performs a series of processing on the historical and real-time data (i.e., operational process data) collected in the production environment, including cleaning, screening, and feature extraction, to facilitate data modeling.

[0039] The system classifies and calibrates the operation process data after data processing according to the characteristics of different operation process data, and completes the modeling process. The purpose of the modeling in this invention is to better discover and abstract the influence of the image judgment process data mainly including the data displayed on the image judgment interface and the operation data of the image judge on the image judgment work, so as to better conduct targeted training. The key element of the modeling process in this invention is to classify and calibrate according to the characteristics of different processes in the above table.

[0040] A specific example of the modeling in this invention is classification tree learning. In classification tree learning, the operation process data in the above-mentioned security inspection image judgment system collected is used as the training data set, and the characteristics of different operation process data in this training data set are classified and calibrated, so as to generalize a set of classification rules from this training data set and construct a classification tree model, enabling this classification tree model to correctly classify and calibrate the instances. In the embodiment of this invention, as a specific example of the modeling process, first, the correlation between the operation process data in the above-mentioned security inspection image judgment system collected is found through attribute screening, and the attributes with significant correlation are combined together for modeling. On the one hand, it is possible to discover whether there are certain regularities between these attribute features according to the results of machine learning; on the other hand, according to the professional security inspection business meaning, the regularities between these attribute features can be explained and discovered. Based on these two aspects, a classification hypothesis is established for these attribute features, that is, what type of image judge the image judge with these attribute features belongs to.

[0041] At least one classification prediction model can be selected to form a model combination, and the established classification hypothesis is used to train this model combination. Based on the classification hypothesis, the modeling will, through the way of ensemble learning, continuously try to combine different classification prediction models and model parameters to conduct learning and prediction, preventing a single model from overfitting. During the learning process, the parameters with unsatisfactory modification effects are modified, and the base models with unsatisfactory effects are replaced to find more optimized and more suitable base models and corresponding parameters to achieve the expected prediction results. The modeling process will be a constantly changing process, and the data collected and analyzed at each stage will have a more suitable model, parameters and combination corresponding to that stage.

[0042] In an embodiment of the present invention, at a certain stage of the modeling process, for example, two classification prediction models, namely decision tree and K-Nearest Neighbor (KNN), can be selected to form a set of model combinations for ensemble learning. The operation process data in the security inspection image judgment system collected above is divided into a training data set and a test data set, and a set of multiple attributes with significant correlation selected according to the above attribute screening method, such as "time on duty when receiving a task" (set as "time on duty") in the data displayed on the above image judgment interface, and "actual image judgment duration" (set as "image judgment duration"), "whether there is no operation timeout" (set as "timeout"), etc. in the operation data of the image judge, are used as the features of the data records in the data set and substituted into the model. On the premise that the model evaluation quality meets the standard, by comparing the prediction results of the two sets, and at the same time according to professional business understanding, records with the same data style and features are searched, such as records with "time on duty" above 40 minutes. For the records corresponding to the "timeout", there is a certain probability that "yes" indicates timeout, and if "timeout" is "no", it indicates no timeout, and its "image judgment duration" will probably exceed 4 seconds. For records with "time on duty" within 40 minutes, there are few timeouts and the image judgment duration rarely exceeds 4 seconds. In addition, there are also a small number of image judgment tasks corresponding to records with "time on duty" above 40 minutes that are completed within 4 seconds. Generally, the time on duty of an image judge once is 45 minutes, and the security inspection department sets the rotation time for each shift generally not less than 40 minutes and not more than 1.5 hours. Therefore, it can be judged that "time on duty" can be bounded by 40 minutes. 40 minutes belongs to the normal time on duty, and more than 40 minutes may lead to fatigue and slackness, resulting in a decline in image judgment quality. In the system for generating a training plan for security inspection image judges according to an embodiment of the present invention, records with "time on duty" above 40 minutes, and at the same time "image judgment duration" greater than 4 seconds or "timeout" being "yes", and another part of the records with "time on duty" above 40 minutes, but "timeout" being "no" and "image judgment duration" less than or equal to 4 seconds, are calibrated as two classifications, namely "image judgment tasks completed by image judges greatly affected by working hours" and "image judgment tasks completed by image judges less affected by working hours". Further, these two types of image judgment task records correspond to two types of image judges, namely "image judges greatly affected by working hours" and "image judges less affected by working hours". Finally, the data records with these two types of calibration labels are substituted into the model according to the data set for evaluation, and the learning process of the model is completed. When the prediction results of the records with these two data styles are relatively stable, the model quality is evaluated again. If the evaluation result fails to reach good, the model is adjusted again until the evaluation result is good, and thus the data modeling is completed, forming a classification prediction model of "affected by working hours".

[0043] According to the above classification prediction model of "affected by working hours", it can be predicted that when the attributes of the data records of the image judgment tasks completed by a certain image judge are such that the "starting work time" is more than 40 minutes, and the "image judgment duration" is greater than 4 seconds or the "overtime" is "yes", this image judge is very likely to be an "image judge significantly affected by working hours". Then, during the training process, this model can effectively predict whether a trainee is an "image judge significantly affected by working hours" or an "image judge less affected by working hours". The trainees labeled as "image judges significantly affected by working hours" can then receive training on how to stay focused during long-term image judgment work.

[0044] It should be noted that the quality of the above model evaluation is measured by the accuracy rate, recall rate, ROC (Receiver Operator Characteristic Curve) curve and the area under its AUC (Area Under Curve). The higher the accuracy rate, the lower the recall rate, and the larger the AUC, the better the model classification effect. Among them, the AUC and accuracy rate are the main criteria, and the recall rate is used as a reference to measure the model quality. The value range of AUC is between 0.5 and 1. When AUC is between 0.7 and 0.8, the model can be considered usable and there is still room for optimization; when AUC is between 0.8 and 0.9, the model is considered good and no excessive readjustment is required; when AUC is above 0.9, the model may be overfitted.

[0045] Of course, the above-mentioned attribute screening, ensemble learning and classification prediction are only an example of the modeling process of the present invention, and are not intended to impose any limitations on the modeling process of the present invention. Those skilled in the art can think of using other similar technologies to establish data models.

[0046] With the data model, it is possible to start from the actual situation of each image judge or each type of image judge, predict the training needs based on the data analysis results, and formulate a training plan for the image judges. Figure 2 A conceptual diagram of a system for generating a training plan for security inspection image judges according to an embodiment of the present invention is shown. As Figure 2 shown, the system performs modeling analysis based on the past and real-time data collected in the production environment, obtains classification and calibration information, and establishes a question bank based on the classification and calibration information. In addition, the system predicts the training needs according to the data analysis results, generates a training plan from the question bank to train the trainees, and the training data can be further fed back into the modeling analysis.

[0047] In addition, the learning and analysis of the data model is a cyclic process. Usually, based on the analysis results, some potential meanings and relationships of the data, as well as their impacts on image judgment, will be discovered. At the same time, some problems encountered in the modeling process will also be found. For example: First, the system fails to collect comprehensive enough data to demonstrate the results, and data needs to be supplemented or adjusted; Second, there are still some interfering or redundant data items affecting the final results; Third, there are unknown data associations between a certain target value and some independent variables. Therefore, the content and logic of each stage of data processing will be updated and improved in continuous learning, and the final data model will also be continuously updated, and sometimes new modeling will be carried out.

[0048] When a trainee first uses the system for generating the training plan for security inspection image judges according to the embodiments of the present invention, the system will conduct an image judgment behavior identification based on the trainee's input. The system will substitute the trainee's image judgment process data into the data model (for example, the classification tree described above), classify and calibrate the trainee's image judgment behavior, quickly analyze the trainee's type, ability, etc., and accordingly generate a training plan for the trainee. According to the trainee's personal situation, endow the trainee with the most needed new abilities, strengthen the abilities that need to be improved urgently, and more quickly and effectively improve the image judgment level. At the same time, point out the advantages and disadvantages and impacts of certain habits and operations, and try to correct the trainee's incorrect habits and bad and meaningless operations through training. And continuously obtain new analysis results according to the trainee's progress during the entire training period.

[0049] Specifically, based on the results of data modeling analysis, basic and special training subjects can be generated and organically combined to train image judges in all aspects. For example, basic training is mostly carried out by simulating the image judgment business in the production environment. The analysis shows that more urgent image judgment requirements will make image judges speed up actively to avoid task overtime and affect performance. Accordingly, some high-intensity training tasks can be generated to improve the working ability and efficiency of the overall trainees, enable trainees to adapt to the working intensity of the actual production environment, and be competent for the work of image judges.

[0050] Specifically, the classification and calibration of image judges will help the system plan specialized training for trainees and establish connections with each trainee. For example, a meticulous image judge has a certain degree of accuracy, but may be overly cautious and need to improve their speed. Therefore, training in speed will be strengthened. The system will plan to assign this trainee more traffic-intensive image judgment tasks. The judge may observe a large backlog of pending tasks and a short remaining time for incoming tasks, forcing them to complete judgments and reach conclusions more quickly. Furthermore, the items in the task images are primarily small packages, which are generally unsuspicious. These items may also be accompanied by more AI-powered prompts, allowing for quick release. Meanwhile, for fast and efficient image judges, their speed is guaranteed, but their accuracy needs improvement, and they may occasionally make misjudgments or miss detections. Training can help these judges slow down their work and improve their judgment quality. The training plan will include more slow-paced task combinations, resulting in a less busy system with fewer backlogs and longer remaining time. Task images may be complex and require careful judgment by the human interpreter. For packages with long lengths and numerous items, the AI may fail to identify suspicious items, resulting in no AI prompt. This allows the interpreter's existing habits to be corrected to a certain extent, balancing speed and quality. If no improvement is seen after training, the system will issue a warning to alert the interpreter.

[0051] Furthermore, tasks with high traffic density can be divided into two categories. The first category, during periods of low passenger traffic density, involves non-peak periods, with low density. This involves the image segmentation of long, connected packages, and individual packages. However, this requires fewer image analysis stations and limited resources. The second category, during peak periods, with high passenger density, involves the image segmentation of long, connected packages. This requires a larger number of image analysis stations and abundant resources. The image analysis behavior characteristics of these two categories are completely different, and the training content tailored to the trainees is therefore completely different. The tasks corresponding to the training program for meticulous image analyzers described above should be categorized as the first category.

[0052] Additionally, we list some types of image recognition behaviors. For the "pressure-gets-stronger" type, the more greedy the task allocation strategy, the stronger the image recognition ability. Training can focus on tapping their potential and further developing this ability. For the "lack of persistence" type, the longer they work, the slower their speed or the higher their error rate. Training is needed to strengthen their ability to consistently maintain high image recognition skills. For the "lack of attention" type, their eyes can't focus on the countdown and column expansion on the screen, affecting their image recognition efficiency. Therefore, specialized training on focusing should be implemented. For the "high-performance late-night but low-performance daytime" type, daytime training should be strengthened. For certain types of packages with high error rates, image recognition capabilities for these packages should be improved. Specific training subjects can be categorized accordingly.

[0053] In summary, an analysis report will be generated for each trainee's training session, listing the specific strengths and weaknesses of that trainee, analyzing the deficiencies, and providing improvement suggestions. All conclusions are supported by specific data. Finally, an overall training report will be generated based on the training effect of the image interpreter after each training session, giving suggestions according to the situation and characteristics of the image interpreter, enabling the image interpreter to better understand themselves and know where to pay attention to maintain and improve in actual work and future training processes. The report will also recommend the intensity of image interpretation work suitable for the image interpreter, the suitable industry fields, and the suitable time period for taking up the post, for the reference of the image interpreter himself and the owner. Thus, through continuous data accumulation, analysis, and practice, the system can establish a standardized question bank and continuously update and expand it. When a trainee retrieves training resources from the system, the system will load the corresponding customized training project according to the trainee information recorded in the system (if not recorded, registration entry, information completion, and initial assessment are required). This process is as Figure 3 shown Figure 3 Another conceptual diagram of the system for generating a security inspection image interpreter training plan according to an embodiment of the present invention is shown.

[0054] In the system for generating a security inspection image interpreter training plan according to an embodiment of the present invention, since the training content is more in line with the actual production environment and the degree of customization is high, the quality and efficiency of training will be significantly improved, and a certain amount of training time can be saved. Not only the professional ability of the image interpreter is improved, but the personalized training plan also upgrades the experience of the entire training process, thus cultivating senior professional image interpreters and improving the quality of the entire security inspection link.

[0055] Figure 4 A flowchart of the method for generating a security inspection image interpreter training plan according to an embodiment of the present invention is shown. The following will refer to Figure 4 and describe in detail the method for generating a security inspection image interpreter training plan according to an embodiment of the present invention.

[0056] In step S401, operation process data in the security inspection image interpretation system is obtained, and the operation process data includes image interpretation interface display data and image interpreter operation data;

[0057] In step S402, data processing is performed on the obtained operation process data, and the data processing includes at least one of cleaning, screening, and feature extraction;

[0058] In step S403, for the operation process data after data processing, classification and calibration are performed according to the characteristics of different operation process data to establish a data model;

[0059] In step S404, an initial image interpretation behavior identification is performed on the trainee to generate the initial image interpretation process data of the trainee;

[0060] In step S405, substitute the initial image judgment process data of the trainee into the data model to classify and calibrate the image judgment process of the trainee, so as to obtain the type of the trainee; and

[0061] In step S406, generate a training plan for the trainee based on the type of the trainee.

[0062] In the method for generating a security inspection image judgment operator training plan according to an embodiment of the present invention, a security inspection image judgment operator training plan can be generated more effectively. This training plan makes good use of a large amount of operation process data accumulated in the centralized image judgment system and conducts training according to the specific situation of each trainee.

[0063] Although multiple components are shown in each of the above block diagrams, those skilled in the art should understand that embodiments of the present invention can be implemented with one or more components missing or with some components combined.

[0064] Although the above steps are described in the order shown in the drawings, those skilled in the art should understand that the above steps can be executed in a different order, or embodiments of the present invention can be implemented without one or more of the above steps.

[0065] It can be understood from the foregoing that electronic components of one or more systems or devices may include, but are not limited to, at least one processing unit, a memory, and a communication bus or communication device that couples the various components including the memory to the processing unit. The system or device may include or may have access to various device-readable media. The system memory may include device-readable storage media in the form of volatile and / or non-volatile memory (such as, read-only memory (ROM) and / or random access memory (RAM)). By way of example and not limitation, the system memory may also include an operating system, application programs, other program modules, and program data.

[0066] Embodiments may be implemented as a system, a method, or a program product. Therefore, embodiments may take the form of an all-hardware embodiment or an embodiment including software (including firmware, resident software, microcode, etc.), which may be collectively referred to herein as "circuitry", "module", or "system". In addition, embodiments may take the form of a program product embodied in at least one device-readable medium having device-readable program code embodied thereon.

[0067] Combinations of device-readable storage media may be used. In the context of this document, a device-readable storage medium ("storage medium") may be any tangible non-signal medium that can contain or store a program consisting of program code configured to be used by or in conjunction with an instruction execution system, apparatus, or device. For the purposes of this disclosure, a storage medium or device should be construed as non-transitory, i.e., not including signals or propagating media.

[0068] This disclosure has been presented for purposes of illustration and description, but is not intended to be exhaustive or limiting. Many modifications and variations will be apparent to those of ordinary skill in the art. The embodiments were chosen and described in order to illustrate the principles and practical applications and to enable others of ordinary skill in the art to understand the disclosure with various modifications suited to the particular purposes contemplated.

Claims

1. A method for generating a training program for security inspection judges, comprising: Acquire operation process data in the security inspection image interpretation system, wherein the operation process data includes image interpretation interface display data and image interpreter operation data; performing data processing on the acquired operation process data, wherein the data processing includes at least one of cleaning, screening, and feature extraction; For the processed operation process data, classify and calibrate them according to the characteristics of different operation process data to establish a data model; Conducting initial image recognition behavior assessment on the trainee to generate initial image recognition process data of the trainee; Substituting the student's initial image recognition process data into the data model, classifying and calibrating the student's image recognition process to obtain the student's type; as well as Based on the type of the trainee, generate a training plan for the trainee. The classification and calibration according to the characteristics of different operation process data to establish a data model includes: Filter out the correlation between different operation process data, combine multiple operation process data with significant correlation to build a data model, The step of combining a plurality of operation process data with significant correlation to establish a data model includes: Establish classification hypotheses based on multiple operational process data with significant correlation; At least one classification prediction model is selected to form a model combination, the model combination is trained using the classification hypothesis, and during the learning process, parameters with unsatisfactory effects are modified and base models with unsatisfactory effects are replaced.

2. The method according to claim 1, wherein Acquiring the operation process data in the security inspection image judgment system includes: using a tracking and recording device to track and record the image judge's operation process on the image.

3. The method according to claim 1, wherein Acquiring the operation process data in the security inspection image judgment system includes: using an eye tracker to track and record the area and time where the judge's gaze stays in the security inspection image.

4. The method according to claim 1, wherein The data displayed on the image judgment interface includes at least one of the following: the real-time system time when the image judgment task is received; the time on the job when the task is received; the number of inspections / releases after taking up the job when the task is received; the real-time number of pending tasks queued when the task is received; the remaining image judgment time when the task is received; whether the task image is a long image segmentation image; the number of image columns displayed when the task is received; the number of image columns displayed when the conclusion is submitted; the AI image judgment conclusion; the remaining image judgment time when the AI conclusion is received; and the number of columns that have been removed when the AI conclusion is received.

5. The method according to claim 1, wherein The judge's operation data includes at least one of the following: actual judgment time; the final judgment conclusion given by the judge; the time difference between receiving the AI conclusion and submitting the judgment conclusion; whether there is no operation timeout; voice message; image processing effect transformation operation; image coordinate transformation operation; and visual attention.

6. The method according to claim 1, wherein Screening out the correlation between different operation process data includes at least one of the following: Discover whether there are patterns between data from different operation processes based on the results of machine learning; as well as Interpret and discover patterns between data from different operational processes based on the meaning of security inspection business.

7. The method according to claim 1, wherein The combining of a plurality of operation process data with significant correlation to establish a data model further comprises: Perform a quality assessment on the established data model. If the assessment result fails to reach a good level, adjust the model again until the assessment result reaches a good level.

8. A system for generating a training program for security inspection judges, the system comprising: An acquisition unit, which acquires operation process data in the security inspection image judgment system, wherein the operation process data includes image judgment interface display data and image judge operation data; a processing unit, performing data processing on the acquired operation process data, wherein the data processing includes at least one of cleaning, screening, and feature extraction; The modeling unit classifies and calibrates the processed operation process data according to the characteristics of different operation process data to establish a data model; an identification unit, performing an initial image identification behavior identification on the trainee to generate initial image identification process data of the trainee; a substitution unit, substituting the student's initial image recognition process data into the data model, classifying and calibrating the student's image recognition process to obtain the student's type; as well as A generating unit generates a training plan for the trainee based on the type of the trainee, The classification and calibration according to the characteristics of different operation process data to establish a data model includes: Filter out the correlation between different operation process data, combine multiple operation process data with significant correlation to build a data model, The step of combining a plurality of operation process data with significant correlation to establish a data model includes: Establish classification hypotheses based on multiple operational process data with significant correlation; At least one classification prediction model is selected to form a model combination, the model combination is trained using the classification hypothesis, and during the learning process, parameters with unsatisfactory effects are modified and base models with unsatisfactory effects are replaced.

9. An electronic device comprising: one or more processors; A storage device for storing executable instructions, wherein when the executable instructions are executed by the processor, the method according to any one of claims 1 to 7 is implemented.

10. A computer-readable medium having executable instructions stored thereon, wherein the instructions are executed by a processor to implement the method according to any one of claims 1 to 7.

Images