Method and system for quickly labeling and training APP recognition model

Abstract

The invention discloses a method and system for rapidly marking and training an APP recognition model, and the method comprises the steps: obtaining a to-be-processed target APP image data set, recognizing the type of a target APP, and calling a reference recognition model corresponding to the type from a preset model library; processing the target APP image data set by using the reference recognition model, and executing scene classification and element-level pre-labeling; displaying the pre-labeling result on the visual interaction interface, and receiving a correction instruction of a user for the pre-labeling result; generating an optimization training set by using the image data processed by the correction instruction and the annotation information of the image data, performing incremental training on the reference recognition model based on the optimization training set, and generating a special recognition model for the target APP; and storing the special identification model as an independent model version in a model library for subsequent automatic identification of newly added data of the same type of target APP.

Description

A method and system for rapid annotation and training of APP recognition models Technical Field

[0001] This invention relates to the field of artificial intelligence technology, and in particular to a method and system for rapid annotation and training of APP recognition models. Background Technology

[0002] With the rapid development of mobile internet technology, the number of mobile applications (APPs) has exploded, and the version iteration speed is extremely fast. In fields such as electronic data forensics, application feature analysis, interface compliance testing, and structured data acquisition, there is an increasing reliance on computer vision technology to automatically identify and extract various elements (such as buttons, icons, text boxes, avatars, speech bubbles, etc.) in APP interfaces.

[0003] However, the training and application of APP interface recognition models currently face the following two main technical challenges:

[0004] First, data annotation is costly and inefficient. Training high-precision AI recognition models typically requires massive amounts of high-quality labeled data. App interfaces usually contain numerous UI elements with complex and varied layouts. In current technologies, data annotation mainly relies on manual operation, using general annotation tools to draw boxes and label each element one by one. Although some OCR (Optical Character Recognition)-based auxiliary methods exist, such as recognizing text keywords or regular expressions to assist in sample construction, these methods are mainly limited to extracting text content and are difficult to effectively handle the structured recognition of non-text UI controls (such as avatar blocks, function icons, and interactive components). Furthermore, although there are methods that combine crowdsourced annotation with AI verification to control quality, this still essentially requires a large amount of human effort and lacks automated pre-annotation capabilities for specific app interface patterns, making annotation work time-consuming and labor-intensive, becoming a major bottleneck in model building.

[0005] Second, the model iteration cycle is long, making it unable to quickly respond to app updates. Apps are updated extremely frequently, and whenever a new app of the same type is launched or an existing app undergoes a redesign, the original recognition model often becomes ineffective. Under the current technological system, to adapt to new interface features, it is usually necessary to collect samples again, perform manual annotation from scratch, and train the model from the beginning. The current market lacks a collaborative mechanism that tightly integrates data annotation, manual verification, and model training. Existing simple template matching methods have poor generalization ability, while general-purpose deep learning models struggle to quickly adapt to specific new app interfaces with limited incremental data.

[0006] In summary, existing technologies lack a system and method that can significantly reduce the cost of manual annotation, create a closed loop between the annotation process and the model training process, and thus enable rapid model iteration for specific applications. Therefore, a new technical solution is urgently needed to address these issues. Summary of the Invention

[0007] To address the aforementioned technical problems in the existing technology, this invention proposes a method and system for rapid annotation and training of APP recognition models, thereby solving the above-mentioned technical problems.

[0008] According to a first aspect of the present invention, a method for rapid annotation and training of an APP recognition model is proposed, comprising:

[0009] S1: Obtain the target APP image dataset to be processed, identify the category of the target APP, and call the benchmark recognition model corresponding to the category from the pre-set model library;

[0010] S2: Process the target APP image dataset using the benchmark recognition model, and perform scene classification and element-level pre-annotation;

[0011] S3: Display the pre-annotation results in a visual interactive interface and receive correction instructions from the user for the pre-annotation results;

[0012] S4: Generate an optimized training set from the image data and its annotation information after the correction instructions, and incrementally train the benchmark recognition model based on the optimized training set to generate a dedicated recognition model for the target APP.

[0013] S5: Store the dedicated recognition model as an independent model version in the model library for subsequent automated recognition of new data from the same target APP.

[0014] In some specific embodiments, the benchmark identification model in S1 is pre-trained and generated based on the native data pattern features of similar APPs. Specifically, it includes: collecting historical electronic data forensic cases of different types of APPs and analyzing the native data patterns; extracting structured features of data structure, key field names, numerical formats and timestamp patterns in the native data patterns; defining an initial label system based on the extracted features; and using sample data with initial labels to train and generate a benchmark identification model.

[0015] In some specific embodiments, the scene classification in S2 is as follows: images are automatically classified into predefined interface scene categories based on image features. The predefined interface scene categories include: personal homepage, friend address book, chat history, group chat homepage, media files, and personal bills. The element-level pre-labeling is as follows: in the classified images, text boxes, image blocks, avatar blocks, and date blocks are automatically identified and selected; temporary labels are assigned to each selected area, and the confidence of the pre-labeling results is calculated based on the pre-trained weights.

[0016] In some specific embodiments, the correction instructions in S3 include adjustments to the scene classification, fine-tuning of element position coordinates, and addition, deletion, or modification of attribute tags.

[0017] In some specific embodiments, S3 also includes: displaying label configuration options during the pre-labeling process. The label configuration options include mandatory labels and optional labels. Mandatory labels are core feature elements that this type of APP must possess, determined based on the benchmark recognition model. Optional labels are feature elements that can be selectively added according to the APP variant.

[0018] In some specific embodiments, S4 specifically includes: automatically collecting high-quality labeled data that has been manually verified to form an optimized training set; using the optimized training set to adjust the internal parameters of the benchmark recognition model so that the model learns the manually corrected features; and saving the adjusted model as a new version model for a specific APP, with the new version model running independently of the benchmark recognition model.

[0019] In some specific embodiments, S5 further includes: when saving the dedicated recognition model, recording the model's metadata information, including: model type, required label set, optional label set, number of uses, and accuracy index.

[0020] According to a second aspect of the invention, a computer-readable storage medium is provided on which one or more computer programs are stored, which, when executed by a computer processor, implement the method described above.

[0021] A system for rapid annotation and training of APP recognition models includes:

[0022] The data access and model matching unit is configured to acquire the target APP image dataset to be processed, identify the category of the target APP, and call the benchmark recognition model corresponding to the category from the pre-set model library.

[0023] The pre-annotation unit is configured to process the target APP image dataset using a benchmark recognition model, performing scene classification and element-level pre-annotation;

[0024] The interactive proofreading unit is configured to display the pre-annotation results on a visual interactive interface and receive correction instructions from the user for the pre-annotation results.

[0025] The model incremental training unit is configured to generate an optimized training set from the image data processed by the correction instructions and its annotation information, and to incrementally train the benchmark recognition model based on the optimized training set to generate a dedicated recognition model for the target APP.

[0026] The model library management unit is configured to store dedicated recognition models as independent model versions in the model library for subsequent automated recognition of newly added data from the same target APP.

[0027] In some specific embodiments, the benchmark identification model is pre-trained and generated based on the native data pattern features of similar APPs. Specifically, it includes: collecting historical electronic data forensic cases of different types of APPs and analyzing the native data patterns; extracting structured features from the data structure, key field names, numerical formats and timestamp patterns in the native data patterns; defining an initial label system based on the extracted features; and training and generating the benchmark identification model using sample data with initial labels.

[0028] In some specific embodiments, scene classification specifically involves: automatically classifying images into predefined interface scene categories based on image features. The predefined interface scene categories include: personal homepage, friend address book, chat history, group chat homepage, media files, and personal bills. Element-level pre-labeling specifically involves: automatically identifying and selecting text boxes, image blocks, avatar blocks, and date blocks in the classified images; assigning temporary labels to each selected area; and calculating the confidence level of the pre-labeling results based on the pre-trained weights.

[0029] In some specific embodiments, the correction instructions include adjusting the scene classification, fine-tuning the element position coordinates, and adding, deleting, or modifying attribute labels; the interactive proofreading unit is also configured to display label configuration options during the pre-annotation process, the label configuration options include mandatory labels and optional labels; the mandatory labels are the core feature elements that this type of APP must have based on the benchmark recognition model, and the optional labels are feature elements that can be selectively added according to the APP variants.

[0030] In some specific embodiments, the incremental training unit is specifically configured to automatically collect high-quality labeled data that has been manually verified to form an optimized training set; use the optimized training set to adjust the internal parameters of the benchmark recognition model so that the model learns the manually corrected features; and save the adjusted model as a new version model for a specific APP, which runs independently of the benchmark recognition model.

[0031] In some specific embodiments, the model library management unit is also configured to record the metadata information of the model when saving a dedicated recognition model. The metadata information includes: model type, required tag set, optional tag set, number of times used, and accuracy index.

[0032] This invention has at least the following technical effects:

[0033] Significantly reducing annotation costs and greatly improving operational efficiency, this invention changes the traditional "starting from scratch" manual annotation model. By introducing a benchmark model for intelligent pre-annotation, it transforms manual operations into a more efficient "proofreading and correction" mode. Combined with a differential sample screening mechanism, users only need to focus on low-confidence or error-prone samples, thereby reducing the workload of manual annotation by more than 70%. When processing massive amounts of electronic data, this efficient collaborative mechanism can shorten the cycle of forensic analysis and model building from several weeks to several days, achieving significant "cost reduction and efficiency improvement" in electronic data forensic analysis.

[0034] This invention establishes a closed-loop learning mechanism to enable rapid adaptive iteration of the model. Addressing the pain points of rapid mobile app version updates and numerous interface variations, this invention constructs a closed-loop learning system of "labeling-training-optimization." Through an incremental training mechanism, when faced with new types of apps or new versions of data, the model does not need to be trained from scratch. It only needs to be fine-tuned based on the existing baseline model using manually calibrated data to quickly generate a dedicated model adapted to the specific app. This adaptive and self-learning mechanism gives the tool the vitality of continuous evolution, effectively solving the technical challenge that traditional fixed rules or feature libraries cannot cope with rapid app iteration and novel data encryption methods.

[0035] High accuracy and reduced technical barriers. This invention employs a core process of "machine pre-annotation + manual review," leveraging the efficiency of AI while retaining human oversight of key evidence. This ensures the accuracy and reliability of the final annotation results and model recognition, meeting the stringent quality requirements of judicial evidence collection. Furthermore, the system productizes the complex machine learning process, guiding operation through a visual interactive interface. This allows non-AI professionals to drive model optimization simply by participating in a simple verification process, reducing reliance on professional data annotators and further enhancing the system's usability and accessibility. Attached Figure Description

[0036] The accompanying drawings are included to provide a further understanding of the embodiments and are incorporated in and constitute a part of this specification. The drawings illustrate embodiments and, together with the description, serve to explain the principles of the invention. Other embodiments and many anticipated advantages of the embodiments will be readily recognized as they become better understood through reference to the following detailed description. Other features, objects, and advantages of this application will become more apparent from reading the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0037] Figure 1 is a flowchart of a method for fast annotation and training of an APP recognition model according to an embodiment of this application;

[0038] Figure 2 is an overall flowchart of a method for fast annotation and training of an APP recognition model according to a specific embodiment of this application;

[0039] Figure 3 is a schematic diagram of the data annotation page of a specific embodiment of this application;

[0040] Figure 4 is a schematic diagram of the workflow of the recognition model of a specific embodiment of this application;

[0041] Figures 5a-5c are interface diagrams of an experimental example of a specific embodiment of this application;

[0042] Figure 6 is a framework diagram of a system for rapid annotation and training of APP recognition models according to an embodiment of this application;

[0043] Figure 7 is a schematic diagram of the structure of a computer system used to implement the electronic device of this application. Detailed Implementation

[0044] The present application will now be described in further detail with reference to the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely illustrative of the invention and not intended to limit it. Furthermore, it should be noted that, for ease of description, only the parts relevant to the invention are shown in the accompanying drawings.

[0045] It should be noted that, unless otherwise specified, the embodiments and features described in this application can be combined with each other. This application will now be described in detail with reference to the accompanying drawings and embodiments.

[0046] Figure 1 shows a flowchart of a method for rapid annotation and training of an APP recognition model according to an embodiment of this application. As shown in Figure 1, the method includes the following steps:

[0047] S1: Obtain the target APP image dataset to be processed, identify the category of the target APP, and call the benchmark recognition model corresponding to the category from the pre-set model library.

[0048] In a specific embodiment, the benchmark recognition model is pre-trained based on the native data pattern features of similar apps. This includes: collecting historical electronic data forensics cases from different types of apps and analyzing native data patterns; extracting structured features from the data structure, key field names, numerical formats, and timestamp patterns in the native data patterns; defining an initial labeling system based on the extracted features; and training the benchmark recognition model using sample data with the initial labels. By combining historical case experience from the electronic data forensics industry, "native data patterns" (such as specific timestamp patterns and data structures) are extracted, giving the benchmark model industry know-how (domain knowledge). Compared to general OCR or general object detection models, this results in more accurate recognition of app interface elements. By defining the initial labeling system, complex app interface images are transformed into structured features that computers can understand, laying the foundation for subsequent automated processing and data analysis.

[0049] S2: Use the benchmark recognition model to process the target APP image dataset, and perform scene classification and element-level pre-annotation.

[0050] In a specific embodiment, scene classification involves automatically categorizing images into predefined interface scene categories based on image features. These categories include: personal homepage, friend address book, chat history, group chat homepage, media files, and personal bills. Element-level pre-annotation involves automatically identifying and selecting text boxes, image blocks, avatar blocks, and date blocks within the categorized images. A temporary label is assigned to each selected area, and the confidence level of the pre-annotation results is calculated based on pre-trained weights. This ability to distinguish specific semantic elements such as "text boxes," "avatar blocks," and "date blocks," rather than simply detecting objects, is crucial for extracting key relationships (avatar + nickname) and timelines (date blocks) in electronic forensics. By calculating the confidence level, the system can subsequently filter out low-confidence samples for focused manual review, further optimizing the allocation of human effort.

[0051] S3: Displays the pre-annotation results in a visual interactive interface and receives correction instructions from the user regarding the pre-annotation results.

[0052] In specific embodiments, the correction instructions include adjustments to scene classification, fine-tuning of element position coordinates, and addition, deletion, and modification of attribute labels, providing comprehensive correction dimensions (scene, coordinates, attributes), ensuring the accuracy of the data (Ground Truth) input to the model for training, solving the potential bias problem in AI pre-labeling, allowing users to supplement new elements that AI cannot recognize, or adjust misclassifications, enabling the system to adapt to various non-standard APP interface variations.

[0053] In specific embodiments, the method further includes: displaying label configuration options during the pre-annotation process. These options include mandatory labels and optional labels. Mandatory labels are core feature elements that an app of this type must possess, determined based on a baseline recognition model. Optional labels are feature elements that can be selectively added depending on app variations. Mandatory labels ensure consistency of core data across different app models, facilitating cross-app data analysis. Optional labels provide the model with the flexibility to handle unique functions (variants) of specific apps. By pre-configuring the labels, users are guided to annotate, avoiding confusion when faced with new apps and standardizing the annotation process.

[0054] S4: Generate an optimized training set from the image data and its annotation information after the correction instructions, and incrementally train the benchmark recognition model based on the optimized training set to generate a dedicated recognition model for the target APP.

[0055] In a specific implementation, high-quality labeled data, verified by manual review, is automatically collected to form an optimized training set. The internal parameters of the baseline recognition model are adjusted using this optimized training set, enabling the model to learn the manually corrected features. The adjusted model is then saved as a new version for a specific app, running independently of the baseline recognition model. Saving the trained model as a new version for a specific app and running it independently retains the general capabilities of the baseline model while generating a high-precision, dedicated model (such as a model specifically for Telegram recognition). This avoids interference from the general model's ability to recognize other apps when adapting to a specific app. Adjusting parameters using only the optimized training set significantly reduces computational costs compared to training with the full dataset, enabling rapid online evolution of the model.

[0056] S5: Store the dedicated recognition model as an independent model version in the model library for subsequent automated recognition of new data from the same target APP.

[0057] In a specific embodiment, when saving a dedicated recognition model, the model's metadata information is recorded. This metadata information includes: model type, required label set, optional label set, usage frequency, and accuracy metrics. By recording metrics such as accuracy and usage frequency, users can intuitively evaluate model quality and prioritize the best-performing model version in subsequent tasks. The metadata information provides a basis for maintaining the model library, facilitating the identification of high-value models and the iteration or elimination of low-quality or outdated models.

[0058] Figure 2 shows an overall flowchart of a method for fast annotation and training of an APP recognition model according to a specific embodiment of this application. As shown in Figure 2, the method specifically includes the following steps:

[0059] 210: Import App Asset Images. The process begins with the user importing app asset images to be processed into the system. At this point, the system will perform necessary preprocessing on the imported data, such as format standardization and data cleaning, in preparation for subsequent recognition.

[0060] 220: Enter the annotation training page. After importing, the system enters the annotation training main interface. This interface is mainly divided into four functional areas, corresponding to the branch processes 231 to 234 in the figure, realizing collaborative annotation with "human-computer interaction".

[0061] 231: The top area displays the data and model types.

[0062] 2311: Users can switch the current recognition model here. After switching, the system will call the baseline model to pre-annotate all imported images.

[0063] 2312: The system displays current data statistics in real time, including the ratio of labeled images to the total number of images, the number of labels, and the theoretical accuracy of the current model.

[0064] 232: Left-hand image list area.

[0065] 2321: The system automatically categorizes imported images into different folders based on the type of the app (e.g., "personal profile" or "chat history" in social apps).

[0066] 2322: Users can switch between different images in this list, which will visually show whether each image has been labeled.

[0067] 233: The annotated area in the middle of the image. This is the core area for users to manually proofread.

[0068] 2331: This area offers several auxiliary functions: Cursor-assisted annotation: Supports clicking on the pre-annotation box for adjustment. Example comparison: Displays standard example images for different types of apps for users to compare and annotate. Real-time interaction: Clicking on the annotation area will display the corresponding labels and information in real time. Effect preview: Supports previewing the effect after model recognition.

[0069] 234: Right-side labeled information area.

[0070] 2341: This section displays the labeling system of the current recognition model. Label information can be added and modified. Clicking a label establishes a one-to-one correspondence with the image annotation box in the central area (233). Labels are divided into mandatory labels and optional labels. The more complete the configured labels, the higher the accuracy of the recognition model. In a specific embodiment, Figure 3 shows a schematic diagram of the data annotation page of a specific embodiment of this application, illustrating the interface of the above annotation process.

[0071] 240: Click "Start Training". After the user completes the manual review and fine-tuning of the pre-labeled results on the above interface (i.e., correcting misclassifications, adjusting box positions, modifying labels, etc.), click "Start Training". At this time, the system will generate an optimized training set from the manually corrected high-quality data, and the training progress will be displayed at the top. The model will begin to learn the manually corrected features.

[0072] 250: Save the recognition model upon completion. After training is complete, the user saves the optimized model. This model can overwrite the original model or be saved as a new version, thereby continuously enriching the model library.

[0073] 260: Recognition Model List and Management. Saved models will be added to the recognition model list for management.

[0074] 271: Users can view detailed model information, including model type, required / optional tag set, usage frequency, and accuracy metrics.

[0075] 272: Users can edit, delete, or add models in the model library. For example, they can import new images based on existing models, make minor adjustments, and save them as new models to achieve continuous iteration and reuse of models.

[0076] Figure 4 illustrates a flowchart of the recognition model of a specific embodiment of this application. As shown in Figure 4, the process specifically includes:

[0077] Data preparation phase. The process begins with initial sample data (401), which typically comes from the collection and analysis of historical electronic data forensics cases. This is followed by the data preprocessing module (402), where the data undergoes format standardization, cleaning, and structuring based on experience and knowledge specific to different types of APP native data patterns (such as data structure, key field names, and numerical formats).

[0078] Model selection and architecture construction phase. The system performs model selection (403), supporting multiple algorithm paths to adapt to different scenarios:

[0079] Deep learning models (404): including CNN convolutional neural networks (406) suitable for image feature extraction and RNN recurrent neural networks (407) suitable for sequence information processing.

[0080] Traditional machine learning models (405): such as SVM / XGBoost (408), can be used to handle specific classification or regression tasks.

[0081] Initial model training phase. Before training, the system performs model configuration (409), including hyperparameter configuration, feature engineering settings, and training strategy configuration. Combined with sample data with initial labels, the system performs initial model training (410), thereby generating the initial recognition model V1.0 (411). This model has basic classification and recognition capabilities, serving as the starting point for subsequent optimization.

[0082] Closed-loop optimization and performance evaluation stage. This is the core step in model evolution, and specifically includes:

[0083] Manual data verification (412): The system uses the initial model to pre-label the new data and receives manual verification and correction.

[0084] Model optimization and update (413): The model continuously learns and accumulates data after manual correction, and adjusts internal parameters to perform autonomous optimization.

[0085] Performance evaluation (414): The system evaluates the updated model. If the indicators do not meet the standards, it returns to S413 to continue optimization and updating; if the indicators meet the standards, it proceeds to the next step to save the optimized model (415).

[0086] Save the optimized model (415): Persistently save the model that meets the performance requirements.

[0087] Model library management and application phase. Optimized models are stored in the model library (416), which provides multi-dimensional management functions including:

[0088] Version Management (417): Supports saving the updated model as a new version, enabling the accumulation of model knowledge and version backtracking.

[0089] Tag system management (418): Manages the tag names, attributes and category levels corresponding to the management model, and supports the addition and modification of tags.

[0090] Model call interface (419): Provides a standard interface for external use by the data annotation module (420), thereby enabling more accurate intelligent pre-annotation of new data using the optimized model, forming a positive cycle of "annotation-training-re-annotation".

[0091] Figures 5a-5c show the interface diagrams of an experimental example of a specific embodiment of this application. As shown in Figure 5a, the first step is to create a model, fill in relevant information, and import images; as shown in Figure 5b, the second step is the main program page for automatic image annotation. Specific functional modules include the following:

[0092] The system automatically identifies the type of app the image belongs to and calls the most suitable built-in recognition model. Users can manually modify the recognition model and automatically label the number of images, tags, and theoretical accuracy of the data after selecting the model.

[0093] After selecting a model, the left side will automatically categorize the main feature images of that type of APP, such as personal homepage information, address book, chat history, etc. for social chat APPs.

[0094] The system automatically pre-labels blocks and provides auxiliary functions to help users complete the block information labeling. Users can modify the labeling information and switch to view the labeling status of the next icon.

[0095] The right side displays the annotation information of the blocks. Clicking on a block will map it to the location of the annotated block in the image in real time. You can then make minor adjustments and click "Save As" to create a new model.

[0096] After the third step of annotation is completed, the management of the recognition model and the viewing of the model's training progress are supported. Multiple recognition models can be trained together, as shown in Figure 5c.

[0097] The method for rapid annotation and training of APP recognition models presented in this application possesses scalability and reusability across multiple scenarios. For different types of APP applications, it summarizes their core data features and label information, automatically classifying and pre-labeling the recognition model, enabling rapid adaptation to the APP recognition model of that type. In scenarios involving different APPs of the same type (such as variant APPs), the system's built-in mandatory and optional labels have high coverage. Furthermore, specific APP recognition models can be generated through manual fine-tuning based on existing models, improving overall work efficiency. The core mechanism of this application, the "closed-loop learning system with human-computer interaction," successfully combines human judgment with machine computing power, enabling the system to proactively discover new knowledge and adaptively evolve, "feeding back" this knowledge into the model. This addresses the pain point of traditional models' insufficient adaptability to the rapidly changing APP ecosystem.

[0098] Figure 6 shows a framework diagram of a system for rapid annotation and training of an APP recognition model according to an embodiment of this application. As shown in Figure 6, the system includes a data access and model matching unit 601, a pre-annotation unit 602, an interactive proofreading unit 603, a model incremental training unit 604, and a model library management unit 605. The data access and model matching unit 601 is configured to acquire the target APP image dataset to be processed, identify the category of the target APP, and call the benchmark recognition model corresponding to the category from a pre-set model library. The pre-annotation unit 602 is configured to use the benchmark recognition model to train the target APP. The image dataset is processed to perform scene classification and element-level pre-annotation; the interactive proofreading unit 603 is configured to display the pre-annotation results on a visual interactive interface and receive correction instructions from the user for the pre-annotation results; the incremental model training unit 604 is configured to generate an optimized training set from the image data and its annotation information after the correction instructions, and incrementally train the benchmark recognition model based on the optimized training set to generate a dedicated recognition model for the target APP; the model library management unit 605 is configured to store the dedicated recognition model as an independent model version in the model library for subsequent automated recognition of newly added data of the same target APP.

[0099] In terms of efficiency and cost, this application achieves "cost reduction and efficiency improvement" in evidence collection and analysis, greatly enhancing annotation efficiency. The system significantly reduces the manual time spent on data annotation, meaning that when processing massive amounts of electronic data, the investigation cycle can be shortened from weeks to days, which is crucial for case solving. In terms of quality and accuracy, it ensures the "precise and reliable" identification results. Starting from the initial model, with "machine pre-annotation + manual review" as the core process, it ensures the accuracy of the final annotation results and the continuous self-evolution of the model, meeting the stringent reliability requirements of judicial evidence. In terms of capability and adaptability, it endows the evidence collection tool with the vitality of "learning and evolution." Traditional evidence collection tools rely on fixed rules and feature libraries, which cannot effectively cope with rapidly updated apps and novel data encryption methods. This invention introduces a closed-loop learning mechanism, enabling the tool to have adaptive and self-learning capabilities, fundamentally solving this industry pain point.

[0100] Referring now to FIG7, a schematic diagram of a computer system suitable for implementing an electronic device according to embodiments of the present application is shown. The electronic device shown in FIG7 is merely an example and should not impose any limitation on the functionality and scope of use of the embodiments of the present application.

[0101] As shown in Figure 7, the computer system includes a central processing unit (CPU) 701, which can perform various appropriate actions and processes based on programs stored in read-only memory (ROM) 702 or programs loaded from storage section 708 into random access memory (RAM) 703. RAM 703 also stores various programs and data required for the operation of system 700. CPU 701, ROM 702, and RAM 703 are interconnected via bus 704. Input / output (I / O) interface 705 is also connected to bus 704.

[0102] The following components are connected to I / O interface 705: an input section 706 including a keyboard, mouse, etc.; an output section 707 including a liquid crystal display (LCD) and speakers, etc.; a storage section 708 including a hard disk, etc.; and a communication section 709 including a network interface card such as a LAN card and a modem, etc. The communication section 709 performs communication processing via a network such as the Internet. A drive 710 is also connected to I / O interface 705 as needed. A removable medium 711, such as a disk, optical disk, magneto-optical disk, semiconductor memory, etc., is installed on drive 710 as needed so that computer programs read from it can be installed into storage section 708 as needed.

[0103] Specifically, according to embodiments of this disclosure, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments of this disclosure include a computer program product comprising a computer program carried on a computer-readable storage medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via communication section 709, and / or installed from removable medium 711. When the computer program is executed by central processing unit (CPU) 701, it performs the functions defined in the methods of this application. It should be noted that the computer-readable storage medium of this application can be a computer-readable signal medium or a computer-readable storage medium or any combination thereof. The computer-readable storage medium can be, for example,—but not limited to—an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, apparatus, or device, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, random access memory (RAM), read-only memory (ROM), erasable programmable read-only memory (EPROM or flash memory), optical fiber, portable compact disk read-only memory (CD-ROM), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this application, a computer-readable storage medium can be any tangible medium containing or storing a program that can be used by or in connection with an instruction execution system, apparatus, or device. In this application, a computer-readable signal medium may include a data signal propagated in baseband or as part of a carrier wave, carrying computer-readable program code. Such propagated data signals can take various forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination thereof. A computer-readable signal medium can also be any computer-readable storage medium other than a computer-readable storage medium that can send, propagate, or transmit a program for use by or in connection with an instruction execution system, apparatus, or device. Program code contained on a computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wireless, wire, optical fiber, RF, etc., or any suitable combination thereof.

[0104] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof. Programming languages ​​include object-oriented programming languages—such as Java, Smalltalk, and C++—as well as conventional procedural programming languages—such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including a local area network (LAN) or a wide area network (WAN)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0105] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0106] The modules described in the embodiments of this application can be implemented in software or in hardware.

[0107] In another aspect, this application also provides a computer-readable storage medium, which may be included in the electronic device described in the above embodiments; or it may exist independently and not assembled into the electronic device. The aforementioned computer-readable storage medium carries one or more programs, which, when executed by the electronic device, cause the electronic device to: acquire a target APP image dataset to be processed, identify the category of the target APP, and call a benchmark recognition model corresponding to the category from a pre-set model library; process the target APP image dataset using the benchmark recognition model, performing scene classification and element-level pre-annotation; display the pre-annotation results on a visual interactive interface, and receive correction instructions from the user regarding the pre-annotation results; generate an optimized training set from the image data and its annotation information processed by the correction instructions, incrementally train the benchmark recognition model based on the optimized training set, and generate a dedicated recognition model for the target APP; store the dedicated recognition model as an independent model version in the model library for subsequent automated recognition of newly added data for the same target APP.

[0108] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of the invention involved in this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the above-described inventive concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions disclosed in this application.

Claims

1. A method for rapid annotation and training of APP recognition models, characterized in that, include: S1: Obtain the target APP image dataset to be processed, identify the category to which the target APP belongs, and call the benchmark recognition model corresponding to the category from the preset model library; S2: Use the benchmark recognition model to process the target APP image dataset, and perform scene classification and element-level pre-annotation; S3: Display the pre-annotation results in a visual interactive interface and receive correction instructions from the user regarding the pre-annotation results; S4: Generate an optimized training set from the image data and its annotation information after processing by the correction instruction, and incrementally train the benchmark recognition model based on the optimized training set to generate a dedicated recognition model for the target APP; S5: Store the dedicated recognition model as an independent model version in the model library for subsequent automated recognition of new data from the same target APP.

2. The method for rapid annotation and training of APP recognition models according to claim 1, characterized in that, The benchmark identification model in S1 is pre-trained and generated based on the native data pattern features of similar APPs. Specifically, it includes: collecting historical electronic data forensic cases of different types of APPs and analyzing the native data patterns; extracting structured features from the data structure, key field names, numerical formats and timestamp patterns in the native data patterns; defining an initial label system based on the extracted features; and training the benchmark identification model using sample data with initial labels.

3. The method for rapid annotation and training of APP recognition models according to claim 2, characterized in that, The scene classification in S2 specifically involves: automatically classifying images into predefined interface scene categories based on image features. These predefined interface scene categories include: personal homepage, friend address book, chat history, group chat homepage, media files, and personal bills. The element-level pre-labeling specifically involves: automatically identifying and selecting text boxes, image blocks, avatar blocks, and date blocks in the classified images; assigning temporary labels to each selected area; and calculating the confidence level of the pre-labeling results based on pre-trained weights.

4. The method for rapid annotation and training of APP recognition models according to claim 1, characterized in that, The correction instructions in S3 include adjustments to scene classification, fine-tuning of element position coordinates, and addition, deletion, and modification of attribute tags.

5. The method for rapid annotation and training of APP recognition models according to claim 2, characterized in that, S3 further includes: displaying label configuration options during the pre-labeling process, the label configuration options including mandatory labels and optional labels; the mandatory labels are core feature elements that this type of APP must have based on the benchmark recognition model, and the optional labels are feature elements that can be selectively added according to the APP variants.

6. The method for rapid annotation and training of APP recognition models according to claim 1, characterized in that, S4 specifically includes: automatically collecting high-quality labeled data that has been manually verified to form an optimized training set; using the optimized training set to adjust the internal parameters of the benchmark recognition model so that the model learns the manually corrected features; and saving the adjusted model as a new version model for a specific APP, wherein the new version model runs independently of the benchmark recognition model.

7. The method for rapid annotation and training of APP recognition models according to claim 1, characterized in that, S5 further includes: when saving the dedicated recognition model, recording the model's metadata information, which includes: model type, required tag set, optional tag set, usage frequency, and accuracy index.

8. A computer-readable storage medium having one or more computer programs stored thereon, characterized in that, When the one or more computer programs are executed by a computer processor, they perform the method according to any one of claims 1-7.

9. A system for rapid annotation and training of APP recognition models, characterized in that, include: The data access and model matching unit is configured to acquire the target APP image dataset to be processed, identify the category to which the target APP belongs, and call the benchmark recognition model corresponding to the category from the preset model library; The pre-annotation unit is configured to process the target APP image dataset using the benchmark recognition model, and perform scene classification and element-level pre-annotation. An interactive proofreading unit is configured to display the pre-annotation results on a visual interactive interface and receive correction instructions from the user regarding the pre-annotation results. A model incremental training unit is configured to generate an optimized training set from the image data processed by the correction instructions and its annotation information, and to perform incremental training on the benchmark recognition model based on the optimized training set to generate a dedicated recognition model for the target APP. A model library management unit is configured to store the dedicated recognition model as an independent model version in the model library for subsequent automated recognition of newly added data from the same target APP.

10. A system for rapid annotation and training of APP recognition models according to claim 9, characterized in that, The benchmark identification model is pre-trained based on the native data pattern features of similar apps. Specifically, it includes: collecting historical electronic data forensic cases of different types of apps and analyzing the native data patterns; extracting structured features from the data structure, key field names, numerical formats, and timestamp patterns in the native data patterns; defining an initial label system based on the extracted features; and training the benchmark identification model using sample data with the initial labels.

11. The system for rapid annotation and training of APP recognition models according to claim 9, characterized in that, The scene classification specifically involves: automatically classifying images into predefined interface scene categories based on image features. These predefined interface scene categories include: personal homepage, friend address book, chat history, group chat homepage, media files, and personal bills. The element-level pre-labeling specifically involves: automatically identifying and selecting text boxes, image blocks, avatar blocks, and date blocks in the classified images; assigning temporary labels to each selected area; and calculating the confidence level of the pre-labeling results based on pre-trained weights.

12. The system for rapid annotation and training of APP recognition models according to claim 9, characterized in that, The correction instructions include adjustments to scene classification, fine-tuning of element position coordinates, and addition, deletion, and modification of attribute labels. The interactive proofreading unit is also configured to display label configuration options during the pre-annotation process. The label configuration options include mandatory labels and optional labels. The mandatory labels are core feature elements that this type of APP must possess, as determined by the benchmark recognition model. The optional labels are feature elements that can be selectively added according to APP variants.

13. The system for rapid annotation and training of APP recognition models according to claim 9, characterized in that, The incremental training unit is specifically configured to automatically collect high-quality labeled data that has been manually verified and confirmed to form an optimized training set; the optimized training set is used to adjust the internal parameters of the benchmark recognition model so that the model learns the manually corrected features; The adjusted model is saved as a new version model for a specific app, and the new version model runs independently of the baseline recognition model.

14. The system for rapid annotation and training of APP recognition models according to claim 9, characterized in that, The model library management unit is also configured to record the model's metadata information when saving the dedicated recognition model. The metadata information includes: model type, required tag set, optional tag set, number of uses, and accuracy index.

Images