A method for text feature recognition of customs documents based on deep learning

Abstract

The present invention relates to the field of customs document text feature recognition, and in particular to a customs document text feature recognition method based on deep learning, comprising: using real-time customs document text for preprocessing to obtain customs document text features; establishing a customs document text classification and analysis model based on deep learning according to the customs document text features; using the customs document text classification and analysis model for feedback adjustment processing to obtain customs document text feature recognition results, by simultaneously considering multi-dimensional features such as data type, data capacity, text content, timestamp and historical data, a dynamic threshold judgment system is established to improve the accuracy of abnormal document recognition, and at the same time, a cross-validation mechanism of the data content classification model and the data content analysis model reduces the missed detection rate of high-risk documents in testing, effectively solving the problem of cold start of training data.

Description

Technical Field

[0001] The present invention relates to the field of customs document text feature recognition, and in particular to a customs document text feature recognition method based on deep learning. Background Art

[0002] With the continued growth of global trade, customs document processing is experiencing a surge in data volume and increasing complexity. Traditional rule-based text recognition methods face challenges such as high data processing pressure, interference from abnormal text, and inaccurate and timely customs data processing. Summary of the Invention

[0003] In response to the shortcomings of the existing technology, the present invention provides a customs document text feature recognition method based on deep learning. By extracting large quantities of historical customs document text data, combining the text and content features therein, and establishing a classification analysis model based on deep learning, rapid evaluation and screening of customs document texts can be achieved.

[0004] To achieve the above objectives, the present invention provides a method for recognizing text features of customs documents based on deep learning, comprising:

[0005] S1. Preprocessing the real-time customs document text to obtain customs document text features;

[0006] S2. Establishing a customs document text classification analysis model based on deep learning according to the customs document text features;

[0007] S3. Using the customs document text classification analysis model to perform feedback adjustment processing to obtain customs document text feature recognition results.

[0008] Preferably, the preprocessing of the real-time customs document text to obtain customs document text features includes:

[0009] S1-1. Collect real-time customs document text;

[0010] S1-2. Obtaining the data type and data capacity of the customs document text according to the real-time customs document text;

[0011] S1-3. Obtaining the text content and upload time of the customs document text according to the real-time customs document text;

[0012] S1-4. Obtaining corresponding historical customs document text according to the real-time customs document text;

[0013] S1-5. Preprocess the data type, data capacity, text content, upload time, and historical customs document texts to obtain customs document text features.

[0014] Furthermore, the customs document text features obtained by preprocessing the data type, data capacity, text content, upload time, and historical customs document texts include:

[0015] S1-5-1. Obtain the text content of the historical customs document text according to the historical customs document text;

[0016] S1-5-2. Obtain the data capacity of the historical customs document text using the data type of the customs document text;

[0017] S1-5-3. Obtain the upload time of the corresponding historical customs document text according to the text content of the historical customs document text using the data type of the customs document text;

[0018] S1-5-4. Obtain the data capacity of the historical customs document text and establish a data capacity threshold corresponding to the maximum and minimum values;

[0019] S1-5-5. Determine whether the data capacity of the customs document text is within the data capacity threshold. If so, execute S1-5-6; otherwise, directly execute S1-5-7;

[0020] S1-5-6. Determine whether the text content of the customs document text corresponds to the upload time of the historical customs document text. If so, use the data type, data capacity, text content, upload time of the customs document text and the historical customs document text as customs document text features. Otherwise, execute S1-5-7.

[0021] S1-5-7. Output the data type, data capacity, text content, and upload time of the customs document text as abnormal customs document text characteristics;

[0022] The historical customs document text includes the data type, data capacity, text content and upload time of the historical customs document text.

[0023] Furthermore, establishing a customs document text classification analysis model based on deep learning according to the customs document text features includes:

[0024] S2-1. Establishing a data content classification model for the customs document text based on deep learning using the customs document text features;

[0025] S2-2. Establishing a data content analysis model for the customs document text based on deep learning using the customs document text features;

[0026] S2-3. Perform cross-optimization processing using the data content classification model of the customs document text and the data content analysis model of the customs document text to obtain a customs document text classification analysis model.

[0027] Furthermore, establishing a data content classification model for customs document text based on deep learning using the customs document text features includes:

[0028] S2-1-1. Using the data type and data capacity corresponding to the historical customs document text features of the customs document text features as input, and the text content of the historical customs document text features of the customs document text features as output, perform training based on deep learning to establish an initial data content classification model;

[0029] S2-1-2. Using the customs document text features, inputting the real-time customs document text into an initial data content classification model to obtain an initial data content classification result of the real-time customs document text;

[0030] S2-1-3. Determine whether the initial data content classification result of the real-time customs document text corresponds to the text content of the customs document text feature. If so, execute S2-1-4. Otherwise, update the historical customs document text of the customs document text feature using the real-time customs document text corresponding to the customs document text feature, and return to S2-1-1.

[0031] S2-1-4. Determine whether the data capacity corresponding to the initial data content classification result of the real-time customs document text is consistent with the data capacity of the customs document text feature. If so, execute S2-1-5; otherwise, return to S2-1-1;

[0032] S2-1-5. Determine whether the data type corresponding to the initial data content classification result of the real-time customs document text is consistent with the data type of the customs document text feature. If so, use the initial data content classification model as the data content classification model of the customs document text. Otherwise, re-acquire the historical customs document text and return to S2-1-1.

[0033] Furthermore, establishing a data content analysis model for the customs document text based on deep learning using the customs document text features includes:

[0034] S2-2-1. Perform keyword extraction processing using the text content of the customs document text features to obtain a first keyword data set;

[0035] S2-2-2. Perform keyword extraction processing on the historical customs document text using the customs document text features to obtain a second keyword data set;

[0036] S2-2-3. Using the text content of the customs document text feature as input and the first keyword dataset corresponding to the text content of the customs document text feature as output, establish an initial data content analysis model through training based on deep learning;

[0037] S2-2-4. Using the text content of the historical customs document text as input and the second keyword dataset corresponding to the text content of the historical customs document text as output, establish an auxiliary data content analysis model through training based on deep learning;

[0038] S2-2-5. Use the initial data content analysis model and the auxiliary data content analysis model to sequentially establish a data content analysis model for the customs document text;

[0039] The keyword extraction process is to delete punctuation marks, auxiliary words and modal particles in the text.

[0040] Furthermore, the customs document text classification and analysis model is obtained by cross-optimizing the data content classification model of the customs document text and the data content analysis model of the customs document text, including:

[0041] S2-3-1. Using the data type and data capacity of the real-time customs document text, inputting the data content classification model of the customs document text to obtain a classification result of the data content to be processed;

[0042] S2-3-2. Using the classification result of the data content to be processed, the data content analysis model of the customs document text is input into the corresponding initial data content analysis model to obtain the analysis result of the initial data content to be processed;

[0043] S2-3-3. Using the classification result of the data content to be processed, inputting the data content analysis model of the customs document text into the corresponding auxiliary data content analysis model to obtain the analysis result of the auxiliary data content to be processed;

[0044] S2-3-4. Determine whether the content analysis results of the initial data to be processed and the content analysis results of the auxiliary data to be processed are duplicated. If so, delete the initial data content analysis model and use the data content classification model of the customs document text and the auxiliary data content analysis model to establish a customs document text classification analysis model in sequence. Otherwise, execute S2-3-5.

[0045] S2-3-5. Determine whether the number of executions of S2-3-4 at the current moment is 1. If so, use the data type and data capacity of the real-time customs document text to update the data type and data capacity of the historical customs document text corresponding to the customs document text feature, and return to S2-1-1. Otherwise, output the analysis results of the content of the initial data to be processed and the analysis results of the content of the auxiliary data to be processed, and end the processing.

[0046] Furthermore, the customs document text classification analysis model is used to perform feedback adjustment processing to obtain customs document text feature recognition results including:

[0047] S3-1. Obtaining initial customs document text classification analysis results using the customs document text classification analysis model;

[0048] S3-2. Verify the initial customs document text classification analysis result to obtain a customs document text feature recognition result.

[0049] Furthermore, obtaining initial customs document text classification analysis results using the customs document text classification analysis model includes:

[0050] Using the data type and data capacity corresponding to the real-time customs document text to input the customs document text classification analysis model to obtain the classification result of the data content to be analyzed and the analysis result of the data content to be analyzed of the real-time customs document text;

[0051] Obtaining corresponding risk text content according to the classification result of the data content to be analyzed;

[0052] Obtain corresponding risk keywords based on the content analysis results of the data to be analyzed;

[0053] The classification results of the data content to be analyzed, the analysis results of the data content to be analyzed, the risk keywords and the risk text content are used as the initial customs document text classification analysis results.

[0054] Furthermore, verification processing is performed based on the initial customs document text classification analysis result to obtain customs document text feature recognition results including:

[0055] S3-2-1. Determine whether the classification result of the data content to be analyzed in the initial customs document text classification analysis result and the risk text content are duplicated. If so, execute S3-2-2. Otherwise, the customs document text feature recognition result is normal, and output the classification result of the data content to be analyzed in the initial customs document text classification analysis result and the analysis result of the data content to be analyzed.

[0056] S3-2-2. Determine whether the content analysis result of the data to be analyzed of the initial customs document text classification analysis result is consistent with the risk keyword. If so, the customs document text feature recognition result is abnormal, and the content analysis result of the data to be analyzed is output. Otherwise, execute S3-2-3.

[0057] S3-2-3. Determine whether the content analysis result of the data to be analyzed of the initial customs document text classification analysis result is completely inconsistent with the risk keyword. If so, the customs document text feature recognition result is normal, and the content classification result of the data to be analyzed and the content analysis result of the data to be analyzed of the initial customs document text classification analysis result are output; otherwise, the customs document text feature recognition result is risk, and the content classification result of the data to be analyzed and the content analysis result of the data to be analyzed of the initial customs document text classification analysis result are output.

[0058] Compared with the closest prior art, the present invention has the following beneficial effects:

[0059] By simultaneously considering multi-dimensional features such as data type, data capacity, text content, timestamp and historical data, a dynamic threshold judgment system is established to improve the accuracy of identifying abnormal documents. At the same time, the cross-validation mechanism of the data content classification model and the data content analysis model reduces the missed detection rate of high-risk documents in testing. The historical data comparison function of the auxiliary model effectively solves the problem of cold start of training data. The three-level verification process brings significant results. Through the real-time feedback loop, the delay time of incremental model updates is significantly reduced, and the response speed to data drift problems is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0060] Figure 1 This is a flowchart of a method for recognizing text features of customs documents based on deep learning provided by the present invention. DETAILED DESCRIPTION

[0061] The specific embodiments of the present invention will be further described in detail below with reference to the accompanying drawings.

[0062] To make the objectives, technical solutions, and advantages of the embodiments of the present invention more clear, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts shall fall within the scope of protection of the present invention. Example

[0063] The present invention provides a method for recognizing customs document text features based on deep learning, such as Figure 1 Shown, including:

[0064] S1. Preprocessing the real-time customs document text to obtain customs document text features;

[0065] S2. Establishing a customs document text classification analysis model based on deep learning according to the customs document text features;

[0066] S3. Using the customs document text classification analysis model to perform feedback adjustment processing to obtain customs document text feature recognition results.

[0067] S1 specifically includes:

[0068] S1-1. Collect real-time customs document text;

[0069] S1-2. Obtaining the data type and data capacity of the customs document text according to the real-time customs document text;

[0070] S1-3. Obtaining the text content and upload time of the customs document text according to the real-time customs document text;

[0071] S1-4. Obtaining corresponding historical customs document text according to the real-time customs document text;

[0072] S1-5. Preprocess the data type, data capacity, text content, upload time, and historical customs document texts to obtain customs document text features.

[0073] S1-5 specifically includes:

[0074] S1-5-1. Obtain the text content of the historical customs document text according to the historical customs document text;

[0075] S1-5-2. Obtain the data capacity of the historical customs document text using the data type of the customs document text;

[0076] S1-5-3. Obtain the upload time of the corresponding historical customs document text according to the text content of the historical customs document text using the data type of the customs document text;

[0077] S1-5-4. Obtain the data capacity of the historical customs document text and establish a data capacity threshold corresponding to the maximum and minimum values;

[0078] S1-5-5. Determine whether the data capacity of the customs document text is within the data capacity threshold. If so, execute S1-5-6; otherwise, directly execute S1-5-7;

[0079] S1-5-6. Determine whether the text content of the customs document text corresponds to the upload time of the historical customs document text. If so, use the data type, data capacity, text content, upload time of the customs document text and the historical customs document text as customs document text features. Otherwise, execute S1-5-7.

[0080] S1-5-7. Output the data type, data capacity, text content, and upload time of the customs document text as abnormal customs document text characteristics;

[0081] The historical customs document text includes the data type, data capacity, text content and upload time of the historical customs document text.

[0082] S2 specifically includes:

[0083] S2-1. Establishing a data content classification model for the customs document text based on deep learning using the customs document text features;

[0084] S2-2. Establishing a data content analysis model for the customs document text based on deep learning using the customs document text features;

[0085] S2-3. Perform cross-optimization processing using the data content classification model of the customs document text and the data content analysis model of the customs document text to obtain a customs document text classification analysis model.

[0086] S2-1 specifically includes:

[0087] S2-1-1. Using the data type and data capacity corresponding to the historical customs document text features of the customs document text features as input, and the text content of the historical customs document text features of the customs document text features as output, perform training based on deep learning to establish an initial data content classification model;

[0088] S2-1-2. Using the customs document text features, inputting the real-time customs document text into an initial data content classification model to obtain an initial data content classification result of the real-time customs document text;

[0089] S2-1-3. Determine whether the initial data content classification result of the real-time customs document text corresponds to the text content of the customs document text feature. If so, execute S2-1-4. Otherwise, update the historical customs document text of the customs document text feature using the real-time customs document text corresponding to the customs document text feature, and return to S2-1-1.

[0090] S2-1-4. Determine whether the data capacity corresponding to the initial data content classification result of the real-time customs document text is consistent with the data capacity of the customs document text feature. If so, execute S2-1-5; otherwise, return to S2-1-1;

[0091] S2-1-5. Determine whether the data type corresponding to the initial data content classification result of the real-time customs document text is consistent with the data type of the customs document text feature. If so, use the initial data content classification model as the data content classification model of the customs document text. Otherwise, re-acquire the historical customs document text and return to S2-1-1.

[0092] S2-2 specifically includes:

[0093] S2-2-1. Perform keyword extraction processing using the text content of the customs document text features to obtain a first keyword data set;

[0094] S2-2-2. Perform keyword extraction processing on the historical customs document text using the customs document text features to obtain a second keyword data set;

[0095] S2-2-3. Using the text content of the customs document text feature as input and the first keyword dataset corresponding to the text content of the customs document text feature as output, establish an initial data content analysis model through training based on deep learning;

[0096] S2-2-4. Using the text content of the historical customs document text as input and the second keyword dataset corresponding to the text content of the historical customs document text as output, establish an auxiliary data content analysis model through training based on deep learning;

[0097] S2-2-5. Use the initial data content analysis model and the auxiliary data content analysis model to sequentially establish a data content analysis model for the customs document text;

[0098] The keyword extraction process is to delete punctuation marks, auxiliary words and modal particles in the text.

[0099] S2-3 specifically includes:

[0100] S2-3-1. Using the data type and data capacity of the real-time customs document text, inputting the data content classification model of the customs document text to obtain a classification result of the data content to be processed;

[0101] S2-3-2. Using the classification result of the data content to be processed, the data content analysis model of the customs document text is input into the corresponding initial data content analysis model to obtain the analysis result of the initial data content to be processed;

[0102] S2-3-3. Using the classification result of the data content to be processed, inputting the data content analysis model of the customs document text into the corresponding auxiliary data content analysis model to obtain the analysis result of the auxiliary data content to be processed;

[0103] S2-3-4. Determine whether the content analysis results of the initial data to be processed and the content analysis results of the auxiliary data to be processed are duplicated. If so, delete the initial data content analysis model and use the data content classification model of the customs document text and the auxiliary data content analysis model to establish a customs document text classification analysis model in sequence. Otherwise, execute S2-3-5.

[0104] S2-3-5. Determine whether the number of executions of S2-3-4 at the current moment is 1. If so, use the data type and data capacity of the real-time customs document text to update the data type and data capacity of the historical customs document text corresponding to the customs document text feature, and return to S2-1-1. Otherwise, output the analysis results of the content of the initial data to be processed and the analysis results of the content of the auxiliary data to be processed, and end the processing.

[0105] S3 specifically includes:

[0106] S3-1. Obtaining initial customs document text classification analysis results using the customs document text classification analysis model;

[0107] S3-2. Verify the initial customs document text classification analysis result to obtain a customs document text feature recognition result.

[0108] S3-1 specifically includes:

[0109] S3-1-1. Input the data type and data capacity corresponding to the real-time customs document text into the customs document text classification analysis model to obtain a classification result and an analysis result of the data content to be analyzed of the real-time customs document text;

[0110] S3-1-2. Obtain corresponding risk text content based on the classification results of the data to be analyzed;

[0111] S3-1-3. Obtain corresponding risk keywords based on the content analysis results of the data to be analyzed;

[0112] S3-1-4. Utilize the classification results of the data content to be analyzed, the analysis results of the data content to be analyzed, the risk keywords and the risk text content as the initial customs document text classification analysis results.

[0113] S3-2 specifically includes:

[0114] S3-2-1. Determine whether the classification result of the data content to be analyzed in the initial customs document text classification analysis result and the risk text content are duplicated. If so, execute S3-2-2. Otherwise, the customs document text feature recognition result is normal, and output the classification result of the data content to be analyzed in the initial customs document text classification analysis result and the analysis result of the data content to be analyzed.

[0115] S3-2-2. Determine whether the content analysis result of the data to be analyzed of the initial customs document text classification analysis result is consistent with the risk keyword. If so, the customs document text feature recognition result is abnormal, and the content analysis result of the data to be analyzed is output. Otherwise, execute S3-2-3.

[0116] S3-2-3. Determine whether the content analysis result of the data to be analyzed of the initial customs document text classification analysis result is completely inconsistent with the risk keyword. If so, the customs document text feature recognition result is normal, and the content classification result of the data to be analyzed and the content analysis result of the data to be analyzed of the initial customs document text classification analysis result are output; otherwise, the customs document text feature recognition result is risk, and the content classification result of the data to be analyzed and the content analysis result of the data to be analyzed of the initial customs document text classification analysis result are output.

[0117] In this embodiment, a method for recognizing text features of customs documents based on deep learning is described. The specific real-time process is as follows:

[0118] A customs port receives approximately 5,000 declarations per day. The system performs the following pre-processing procedures:

[0119] Data collection: Scan paper documents through the OCR system and receive electronic EDI messages to generate structured text data (XML / JSON format)

[0120] Feature extraction:

[0121] Data type recognition: Automatically distinguish 12 types of documents including "Import Declaration Form", "Export Declaration Form", "Transfer Form", etc.

[0122] Data capacity analysis: Count the number of fields in each document (average 200 ± 50 fields)

[0123] Text content analysis: extract core fields such as product code, origin, and declared value

[0124] Timestamp record: reporting time accurate to milliseconds

[0125] Historical data comparison:

[0126] Call the database of similar documents from the past three years (about 3 million records)

[0127] Establish dynamic thresholds: For example, the typical number of fields in a customs declaration for a certain type of chemical product is 180-220.

[0128] Anomaly Detection:

[0129] It was found that a customs declaration form for "lithium batteries" contained only 80 fields.

[0130] The system automatically marks it as an abnormal feature (triggering subsequent manual review).

[0131] The process of data content classification model training is as follows:

[0132] Training data:

[0133] Input: data type encoding (one-hot vector) + data capacity (normalized value)

[0134] Output: TF-IDF vector of text content (dimension 5000)

[0135] Network structure:

[0136] Python

[0137] model = Sequential([

[0138] Dense(512, input_dim=13, activation='relu'), # 12 document types + 1 capacity feature

[0139] Dropout(0.2),

[0140] Dense(256, activation='relu'),

[0141] Dense(5000, activation='sigmoid') # Multi-label classification ])

[0143] The training results are as follows:

[0144] Achieved 92.3% field content prediction accuracy on the test set;

[0145] Forecast accuracy for unusual products, such as new electronics, dropped to 78%.

[0146] Data content analysis model training:

[0147] Keyword extraction:

[0148] Use HanLP tool to remove particles / punctuation

[0149] Reserved: "Polyethylene", "CIF", "HSCODE-3901" and other professional terms

[0150] Dual-model architecture:

[0151] Initial model: Processing the current document (89% accuracy)

[0152] Auxiliary model: Processing historical documents (accuracy rate 93%)

[0153] Optimization strategy:

[0154] Using adversarial training, the hidden layers of the two models share weights.

[0155] Those skilled in the art will appreciate that embodiments of the present invention may be provided as methods, systems, or computer program products. Thus, the present invention may take the form of an entirely hardware embodiment, an entirely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention may take the form of a computer program product implemented on one or more computer-usable storage media (including but not limited to magnetic disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code.

[0156] The present invention is described with reference to the flowcharts and / or block diagrams of the methods, devices (systems), and computer program products according to the embodiments of the present application. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor, or other programmable data processing device to produce a machine, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 a process or multiple processes and / or boxes Figure 1 A device that provides the functions specified in a block or multiple blocks.

[0157] These computer program instructions may also be stored in a computer readable memory that can direct a computer or other programmable data processing device to work in a specific manner, so that the instructions stored in the computer readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 a process or multiple processes and / or boxes Figure 1 The function specified in one or more boxes.

[0158] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operational steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing the instructions executed on the computer or other programmable device for implementing the process. Figure 1 a process or multiple processes and / or boxes Figure 1 A step that specifies a function in one or more boxes.

[0159] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention and not to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A method for recognizing text features of customs documents based on deep learning, characterized in that: include: S1. Preprocessing the real-time customs document text to obtain customs document text features; S2. Establishing a customs document text classification analysis model based on deep learning according to the customs document text features; S2-1. Establishing a data content classification model for the customs document text based on deep learning using the customs document text features; S2-1-1. Using the data type and data capacity corresponding to the historical customs document text features of the customs document text features as input, and the text content of the historical customs document text features of the customs document text features as output, perform training based on deep learning to establish an initial data content classification model; S2-2. Establishing a data content analysis model for the customs document text based on deep learning using the customs document text features; S2-2-3. Using the text content of the customs document text feature as input and the first keyword dataset corresponding to the text content of the customs document text feature as output, establish an initial data content analysis model through training based on deep learning; S2-2-4. Using the text content of the historical customs document text as input and the second keyword dataset corresponding to the text content of the historical customs document text as output, establish an auxiliary data content analysis model through training based on deep learning; S2-2-5. Use the initial data content analysis model and the auxiliary data content analysis model to sequentially establish a data content analysis model for the customs document text; S2-3. Performing cross-optimization processing using the data content classification model of the customs document text and the data content analysis model of the customs document text to obtain a customs document text classification analysis model; S2-3-1. Using the data type and data capacity of the real-time customs document text, inputting the data content classification model of the customs document text to obtain a classification result of the data content to be processed; S2-3-2. Using the classification result of the data content to be processed, the data content analysis model of the customs document text is input into the corresponding initial data content analysis model to obtain the analysis result of the initial data content to be processed; S2-3-3. Using the classification result of the data content to be processed, inputting the data content analysis model of the customs document text into the corresponding auxiliary data content analysis model to obtain the analysis result of the auxiliary data content to be processed; S2-3-4. Determine whether the content analysis results of the initial data to be processed and the content analysis results of the auxiliary data to be processed are duplicated. If so, delete the initial data content analysis model and use the data content classification model of the customs document text and the auxiliary data content analysis model to establish a customs document text classification analysis model in sequence. Otherwise, execute S2-3-5. S2-3-5. Determine whether the number of executions of S2-3-4 at the current moment is 1. If so, update the corresponding data type and data capacity of the historical customs document text in the customs document text feature using the data type and data capacity of the real-time customs document text, and return to S2-1-1. Otherwise, output the analysis results of the content of the pending initial data and the content analysis results of the pending auxiliary data, and terminate the processing. S3. Using the customs document text classification analysis model to perform feedback adjustment processing to obtain customs document text feature recognition results; S3-1. Obtaining initial customs document text classification analysis results using the customs document text classification analysis model; S3-2. Verify the initial customs document text classification analysis result to obtain a customs document text feature recognition result.

2. A method for recognizing customs document text features based on deep learning according to claim 1, characterized in that: The customs document text features obtained by preprocessing the real-time customs document text include: S1-1. Collect real-time customs document text; S1-2. Obtaining the data type and data capacity of the customs document text according to the real-time customs document text; S1-3. Obtaining the text content and upload time of the customs document text according to the real-time customs document text; S1-4. Obtaining corresponding historical customs document text according to the real-time customs document text; S1-5. Preprocess the data type, data capacity, text content, upload time, and historical customs document texts to obtain customs document text features.

3. A method for recognizing customs document text features based on deep learning as claimed in claim 2, characterized in that: The customs document text features obtained by preprocessing the data type, data capacity, text content, upload time and historical customs document text include: S1-5-1. Obtain the text content of the historical customs document text according to the historical customs document text; S1-5-2. Obtain the data capacity of the corresponding historical customs document text using the data type of the customs document text; S1-5-3. Obtain the upload time of the corresponding historical customs document text according to the text content of the historical customs document text using the data type of the customs document text; S1-5-4. Obtain the data capacity of the historical customs document text and establish a data capacity threshold corresponding to the maximum and minimum values; S1-5-5. Determine whether the data capacity of the customs document text is within the data capacity threshold. If so, execute S1-5-6; otherwise, directly execute S1-5-7; S1-5-6. Determine whether the text content of the customs document text corresponds to the upload time of the historical customs document text. If so, use the data type, data capacity, text content, upload time of the customs document text and the historical customs document text as customs document text features. Otherwise, execute S1-5-7. S1-5-7. Output the data type, data capacity, text content, and upload time of the customs document text as abnormal customs document text characteristics; The historical customs document text includes the data type, data capacity, text content and upload time of the historical customs document text.

4. A method for recognizing customs document text features based on deep learning according to claim 1, characterized in that: Establishing a data content classification model for customs document text based on deep learning using the customs document text features includes: S2-1-2. Using the customs document text features, inputting the real-time customs document text into an initial data content classification model to obtain an initial data content classification result of the real-time customs document text; S2-1-3. Determine whether the initial data content classification result of the real-time customs document text corresponds to the text content of the customs document text feature. If so, execute S2-1-4. Otherwise, update the historical customs document text of the customs document text feature using the real-time customs document text corresponding to the customs document text feature, and return to S2-1-1. S2-1-4. Determine whether the data capacity corresponding to the initial data content classification result of the real-time customs document text is consistent with the data capacity of the customs document text feature. If so, execute S2-1-5; otherwise, return to S2-1-1; S2-1-5. Determine whether the data type corresponding to the initial data content classification result of the real-time customs document text is consistent with the data type of the customs document text feature. If so, use the initial data content classification model as the data content classification model of the customs document text. Otherwise, re-acquire the historical customs document text and return to S2-1-1.

5. A method for recognizing customs document text features based on deep learning as claimed in claim 4, characterized in that: Establishing a data content analysis model for customs document text based on deep learning using the customs document text features includes: S2-2-1. Perform keyword extraction processing using the text content of the customs document text features to obtain a first keyword data set; S2-2-2. Perform keyword extraction processing on the historical customs document text using the customs document text features to obtain a second keyword data set; The keyword extraction process is to delete punctuation marks, auxiliary words and modal particles in the text.

6. A method for recognizing customs document text features based on deep learning as claimed in claim 1, characterized in that: Obtaining initial customs document text classification analysis results using the customs document text classification analysis model includes: Using the data type and data capacity corresponding to the real-time customs document text to input the customs document text classification analysis model to obtain the classification result of the data content to be analyzed and the analysis result of the data content to be analyzed of the real-time customs document text; Obtaining corresponding risk text content according to the classification result of the data content to be analyzed; Obtain corresponding risk keywords based on the content analysis results of the data to be analyzed; The classification results of the data content to be analyzed, the analysis results of the data content to be analyzed, the risk keywords and the risk text content are used as the initial customs document text classification analysis results.

7. A method for recognizing customs document text features based on deep learning as claimed in claim 6, characterized in that: The customs document text feature recognition results obtained by performing verification processing based on the initial customs document text classification analysis results include: S3-2-1. Determine whether the classification result of the data content to be analyzed in the initial customs document text classification analysis result and the risk text content are duplicated. If so, execute S3-2-2. Otherwise, the customs document text feature recognition result is normal, and output the classification result of the data content to be analyzed in the initial customs document text classification analysis result and the analysis result of the data content to be analyzed. S3-2-2. Determine whether the content analysis result of the data to be analyzed of the initial customs document text classification analysis result is consistent with the risk keyword. If so, the customs document text feature recognition result is abnormal, and the content analysis result of the data to be analyzed is output. Otherwise, execute S3-2-3. S3-2-3. Determine whether the content analysis result of the data to be analyzed of the initial customs document text classification analysis result is completely inconsistent with the risk keyword. If so, the customs document text feature recognition result is normal, and the content classification result of the data to be analyzed and the content analysis result of the data to be analyzed of the initial customs document text classification analysis result are output; otherwise, the customs document text feature recognition result is risk, and the content classification result of the data to be analyzed and the content analysis result of the data to be analyzed of the initial customs document text classification analysis result are output.

Images