A method for express information extraction and distributed representation based on deep learning

By combining YOLOv5 target detection network and PaddleOCR optical character recognition technology, efficient identification of express delivery pages and target text areas is achieved, and the recognized text information is converted into unified digital encoding through information digitization, solving the efficiency of traditional express delivery information processing The problems of low and poor accuracy have achieved automated management and efficient distribution of express delivery information.

CN118799890BActive Publication Date: 2024-11-12MOPAI SPACE (TIANJIN) CO LTD
View PDF 5 Cites 0 Cited by

Patent Information

Application Number
CN202411280328.3
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-13
Publication Date
2024-11-12
Estimated Expiration
2044-09-13

AI Technical Summary

Technical Problem

Traditional express information processing methods rely on manual operations, are inefficient and prone to errors, and cannot efficiently manage and distribute express information efficiently and accurately.

Method used

The YOLOv5 target detection network and PaddleOCR optical character recognition technology are used to realize efficient identification of express delivery pages and target text areas. The identified text information is converted into unified digital encoding through information digitization methods, and automated management and efficient distribution are carried out.

Benefits of technology

It improves the efficiency and accuracy of express information processing, reduces the error rate of manual operation, and realizes the automated management and efficient distribution of express information.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118799890B_ABST
    Figure CN118799890B_ABST
Patent Text Reader

Abstract

The present invention relates to a method for extracting express information based on deep learning and its distributed representation, which comprises the following steps: data preparation, obtaining the express delivery bill data set required for training through data enhancement and preprocessing; model construction, using YOLOv5 target detection network to train the express delivery bill area and text area recognition model; express delivery bill recognition, cutting out the express delivery bill area; bill valid information, obtaining the target text area coordinates; PaddleOCR text recognition, screening and identifying text information; text screening, classifying and digitizing the identified text information; distributed representation, converting the address, telephone and name information into digital codes respectively; express matching, matching the express delivery feature information with the feature information in the customer database by comparing the digital codes, and binding the corresponding customers. The method improves the efficiency and accuracy of express information processing and its distribution, and reduces the error rate of manual operation.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present invention relates to the fields of logistics warehousing, computer vision and natural language processing, and specifically to a method for express information extraction and distributed representation based on deep learning. Background Art

[0002] With the rapid development of e-commerce, the business volume of the express delivery industry has increased dramatically. How to efficiently and accurately manage and distribute express delivery information has become an urgent problem to be solved. Traditional express delivery information processing methods usually rely on manual operations, which are not only inefficient but also prone to errors. The rapid development of deep learning-based object detection and optical character recognition (OCR) technology has provided a new technical means to solve this problem. Summary of the invention

[0003] The present invention aims to provide a method for extracting and representing express delivery information based on deep learning. By using the YOLOv5 target detection network and PaddleOCR optical character recognition technology, the express delivery bill and the target text area can be efficiently identified, and the identified text information can be digitized to form a unified digital code to achieve automatic management and efficient distribution of express delivery information. Traditional express delivery information processing methods usually rely on manual operations, which are not only inefficient but also prone to errors.

[0004] A method for extracting express information and its distributed representation based on deep learning, the method comprising the following steps:

[0005] S1. Obtain the express delivery bill dataset required for training, enrich the training set through data enhancement methods, preprocess the dataset and complete data annotation;

[0006] S2. Use the YOLOv5 target detection network to train a network for express delivery bill area recognition, predict the bill position and rotation angle; perform detection model inference, perform rotated target detection on the bill according to the detected position and angle, output the bill image and bill rotation angle, and rotate the cut bill to obtain the complete bill area;

[0007] S3. Use the YOLOv5 target detection network to train a model to detect the target text position and predict the target text position; perform detection model reasoning, calculate the text area coordinates based on the detected position, and obtain the YOLO target information box;

[0008] S4. Use PaddleOCR optical character recognition technology to recognize the express delivery label information and obtain the PaddleOCR text box. Based on the target recognition results of YOLOv5 and the text recognition results of PaddleOCR optical character recognition technology, calculate the overlap between the coordinates of the PaddleOCR text box and the coordinates of the YOLO target information box to filter the text information;

[0009] S5. Establish a digital distributed representation method for information to classify the obtained text information and store it in an Excel table;

[0010] S6. Establish a complete customer database based on the distributed characterization method of digital information, where the characteristic information of each customer in the database includes name, address, and telephone number, and is stored digitally;

[0011] S7. According to the information digital distributed representation method, the identified express feature information containing text, name, address, telephone number and other features is converted into digital feature information, and then the express feature information is compiled into a string of ordered codes; when a certain express delivery bill feature information is not identified, the encoding uses a single-byte character 0 to identify the information;

[0012] S8. The express characteristic information obtained according to the information digital distributed representation method is a string of digital codes arranged in a predetermined order. It is compared with the digital code corresponding to the customer characteristic information in the database, and the digital address code, digital telephone code and digital name code are matched in turn. If they can all be matched correctly, the express will be bound to the customer; if they cannot be matched correctly, the information will be entered into the customer database.

[0013] As a further technical solution of the present invention, the information digital distributed representation method includes:

[0014] a) Identify the address and convert each segment of the text address (including community, building and household number) into a digital address. The address is an 8-bit digital string;

[0015] b) Identify the customer's telephone number, which is an 11-digit string;

[0016] c) Identify the customer's name. The customer's name is a 12-digit string. According to the four-corner number dictionary, one Chinese character corresponds to four digits to form a coding string.

[0017] As a further technical solution of the present invention, the data annotation of the data set in S1 includes two types: annotating the express delivery bill area and the effective information area in the bill. The express delivery bill area data set collects multiple groups of express deliveries with bills shot at different angles, and converts the RGB image into a resolution of 640×480. The effective information data set in the bill collects the cropped express delivery bill, converts the grayscale image into a resolution of 640×480, and uses a data enhancement method for data enhancement. The data enhancement method includes rotation, cropping, flipping and noise addition; annotating the express delivery bill area is to use Labelme software to perform rectangular label annotating on the express delivery bill area in the RGB image of the express delivery bill shot at different angles, and the label saving format is selected in YOLO format; annotating the effective information area in the bill is to use Labelme software to perform rectangular label annotating on the express delivery number and recipient information in the grayscale image of the cropped express delivery bill, and the label saving format is selected in YOLO format.

[0018] As a further technical solution of the present invention, the model is obtained by training the YOLOv5 target detection network in S2, and the model can effectively detect and cut and extract overlapping face sheets, multiple face sheets, blurred face sheets, face sheets with incomplete information and wrinkled face sheets; face sheet extraction is to cut the complete face sheet area, firstly rotate the face sheet angle, use OpenCV's affine transformation to perform preliminary correction, then use the orientation classification model to determine the direction of the face sheet, and finally correct the face sheets that are inverted and left and right directions, and finally output the face sheets with the text facing up at a positive angle; face sheet information extraction is based on YOLOv5's rotated target detection, rotates the face sheet, and cuts the complete face sheet area according to the rotated rectangular frame predicted by the model, excluding other non-face sheet areas; the YOLOv5 target detection network training trained in S3, the samples are divided into training set, verification set and test set in a ratio of 3:1:1, and a model for detecting the target text position is obtained; target position detection is to frame out the valid text area information in the image, including the customer's name, address, telephone number and express number in the face sheet.

[0019] As a further technical solution of the present invention, the PaddleOCR optical character recognition technology used in S4 is an optical character recognition technology based on a convolutional neural network. The text is converted into image information through an optical input method, and the optical input method includes scanning. The image information is then converted into editable computer text by using text recognition technology. The text information is filtered by calculating the overlap between the PaddleOCR box coordinates and the YOLO box coordinates; the overlap between the PaddleOCR text box and the YOLOv5 text box is calculated, and the PaddleOCR text box with an overlap of 0.5 is retained to achieve the purpose of filtering text information.

[0020] As a further technical solution of the present invention, the preset order of the information digitization distributed representation method in S5 is: from left to right, it includes community, building, household number, telephone number and name; 31 single-byte characters are used to form a code to represent a specific feature information, and the code consists of 5 segments, namely, community code segment, building code segment, household number code segment, telephone code segment and name code segment, wherein the community code segment consists of 2 single-byte characters, the building code segment consists of 2 single-byte characters, the household number code segment consists of 4 single-byte characters, the telephone code segment consists of 11 single-byte characters, and the name code segment consists of 12 single-byte characters.

[0021] As a further technical solution of the present invention, the address code segments of the information digitization distributed representation method in S5 include: one-tenth address code segment, one-hundredth address code segment, one-thousandth address segment and one-ten-thousandth address code segment, the one-tenth address code segment is the first and second characters of the address code segment, the one-hundredth address code segment is the third and fourth characters of the address code segment, the one-thousandth address code segment is the fifth and sixth characters of the area code segment, and the one-ten-thousandth address code segment is the seventh and eighth characters of the address code segment.

[0022] As a further technical solution of the present invention, the complete customer database in S6 is the digital characteristic information of the customers in the community. The customers in the community enter their own information, and the characteristic information of the community customers is converted into an ordered code composed of numbers according to the information digital distributed representation method to obtain the characteristic digital information of the customers in the community.

[0023] As a further technical solution of the present invention, the express characteristic information in S7 is obtained by photographing the express delivery bill with a camera during express delivery, and using optical character recognition technology based on deep learning to obtain information on the express delivery bill. The recognized express delivery bill information is converted into digital characteristic information according to an information digitization method. The digital information is allocated in a step-by-step manner from front to back to form a string of ordered codes. If the bill information does not exist due to external factors, the non-existent information is converted into the number 0 when converted according to the information digitization method. The external factors include blurring, overlapping, stains and wrinkles on the express delivery bill.

[0024] As a further technical solution of the present invention, the express characteristic information obtained by the information digital distributed representation method in S8 is a string of digital codes arranged in a predetermined order. If the express characteristic information and the customer characteristic information match correctly, the express is bound to the corresponding customer; if they fail to match correctly, it is determined that the customer information of the customer to whom the express belongs is not in the customer database, and the express characteristic information is stored in the customer database as the customer characteristic information in the database.

[0025] Compared with the prior art, the present invention has the following beneficial effects:

[0026] The present invention realizes efficient recognition of express delivery bills and target text areas by combining the YOLOv5 target detection network and the PaddleOCR optical character recognition technology, and converts the recognized text information into a unified digital code through an information digitization method, thereby realizing the automatic management and efficient distribution of express delivery information, improving the efficiency and accuracy of express delivery information processing, and reducing the error rate of manual operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0027] Figure 1 It is a flowchart of the express information extraction and distributed representation method in a specific embodiment of the present invention.

[0028] Figure 2 It is a structural diagram of the information digital distributed representation method in a specific embodiment of the present invention. DETAILED DESCRIPTION

[0029] The technical solution of the present invention is further described in detail below in conjunction with specific implementation methods.

[0030] See also Figure 1 to Figure 2 , an embodiment of the present invention provides a method for extracting express information and its distributed representation based on deep learning, the method comprising the following steps:

[0031] S1. Obtain the express delivery bill dataset required for training, enrich the training set through data enhancement methods, preprocess the dataset and complete data annotation;

[0032] A large number of express delivery bills required for training were obtained from different express delivery companies. The data samples include bill images taken at various angles and lighting conditions, including multiple groups of express delivery with bills taken at different angles. The RGB images were converted to a resolution of 640×480, and data enhancement operations such as rotation, cropping, flipping, and noise addition were performed. The data set was preprocessed, and the Labelme software was used to annotate the express delivery bill areas taken at different angles with rectangular labels. The label saving format was selected in YOLO format, and the sample size after labeling was about 4800. The cropped express delivery bills were processed and converted into grayscale images with a resolution of 640×480. The Labelme software was used to annotate the valid information in the bill (such as the express delivery bill number and recipient information) with rectangular labels. The label saving format was selected in YOLO format, and the sample size after labeling was about 4800.

[0033] The data annotation of the data set in S1 includes two types: annotating the express delivery bill area and the valid information area in the bill. The express delivery bill area is annotated by using Labelme software to annotate the express delivery bill area in the RGB image of the express delivery bill taken at different angles with rectangular labels, and the label saving format is selected in YOLO format; the valid information area in the bill is annotated by using Labelme software to annotate the express delivery bill number and recipient information in the grayscale image of the cropped express delivery bill with rectangular labels, and the label saving format is selected in YOLO format.

[0034] S2. Use the YOLOv5 target detection network to train a network for express delivery bill area recognition, predict the bill position and rotation angle; perform detection model inference, perform rotated target detection on the bill according to the detected position and angle, output the bill image and bill rotation angle, and rotate the cut bill to obtain the complete bill area;

[0035] In S2, a model is obtained by training the YOLOv5 target detection network. The model can effectively detect and cut out overlapping bills, multiple bills, blurred bills, bills with incomplete information, and wrinkled bills. Bill extraction is to cut out the complete bill area. First, the bill angle is rotated, and the OpenCV affine transformation is used for preliminary correction. Then, the orientation classification model is used to determine the bill orientation. The bills that are inverted or sideways are finally corrected. Finally, the output bills are all positive angles with the text facing up. Bill information extraction is based on YOLOv5's rotation target detection, which performs rotation target detection on the bill. The complete delivery area is cropped according to the rotated rectangular frame predicted by the model, excluding other non-delivery areas; the YOLOv5 target detection network trained in S3 is trained, and the samples are divided into training set, validation set and test set in a ratio of 3:1:1. About 1440 samples are used for model training, about 480 samples are used for model validation, and about 480 samples are used for model testing. The labeled data set is trained to obtain a model for detecting the target text position; the target position detection is to select the valid text area information in the image, including the customer's name, address, telephone number and express number in the delivery. The model calculates the coordinates of the text area to achieve valid text area information selection.

[0036] S3. Use the YOLOv5 target detection network to train a model to detect the target text position and predict the target text position; perform detection model reasoning, calculate the text area coordinates based on the detected position, and obtain the YOLO target information box;

[0037] The original express image is input into the trained model for detecting the target text position, the detection model is inferred on the test set, the delivery bill image and rotation angle are output, and the affine transformation of OpenCV is used for preliminary correction. The inverted and left-right delivery bills are finally corrected to obtain the forward delivery bill image with the text facing upwards. The express delivery bill area is identified and cropped to obtain a separate express delivery bill image.

[0038] S4. Use PaddleOCR optical character recognition technology to recognize the express delivery label information and obtain the PaddleOCR text box. Based on the target recognition results of YOLOv5 and the text recognition results of PaddleOCR optical character recognition technology, calculate the overlap between the coordinates of the PaddleOCR text box and the coordinates of the YOLO target information box to filter the text information;

[0039] The PaddleOCR optical character recognition technology used in S4 is an optical character recognition technology based on a convolutional neural network. It converts text into image information through an optical input method, which includes scanning. The image information is then converted into editable computer text using text recognition technology. The text information is filtered by calculating the overlap between the PaddleOCR box coordinates and the YOLO box coordinates; the overlap between the PaddleOCR text box and the YOLOv5 text box is calculated, and the PaddleOCR text box with an overlap of 0.5 is retained to achieve the purpose of filtering text information.

[0040] S5. Establish a digital distributed representation method for information to classify the obtained text information and store it in an Excel table;

[0041] Information digital distributed representation methods include:

[0042] a) Identify the address and convert each segment of the text address (including community, building and household number) into a digital address. The address is an 8-bit digital string;

[0043] b) Identify the customer's telephone number, which is an 11-digit string;

[0044] c) Identify the customer's name. The customer's name is a 12-digit string. According to the four-corner number dictionary, one Chinese character corresponds to four digits to form a coding string.

[0045] The preset order of the information digital distributed representation method in S5 is: from left to right, it includes community, building, household number, telephone number and name; 31 single-byte characters are used to form a code to represent a specific feature information, and the code consists of 5 segments, namely, community code segment, building code segment, household number code segment, telephone code segment and name code segment, wherein the community code segment consists of 2 single-byte characters, the building code segment consists of 2 single-byte characters, the household number code segment consists of 4 single-byte characters, the telephone code segment consists of 11 single-byte characters, and the name code segment consists of 12 single-byte characters.

[0046] The address code segments of the information digitization distributed representation method in S5 include: one-tenth address code segment, one-hundredth address code segment, one-thousandth address segment and one-ten-thousandth address code segment, the one-tenth address code segment is the first and second characters of the address code segment, the one-hundredth address code segment is the third and fourth characters of the address code segment, the one-thousandth address code segment is the fifth and sixth characters of the area code segment, and the one-ten-thousandth address code segment is the seventh and eighth characters of the address code segment.

[0047] S6. Establish a complete customer database based on the distributed characterization method of digital information, where the characteristic information of each customer in the database includes name, address, and telephone number, and is stored digitally;

[0048] The complete customer database in S6 is the digital characteristic information of the customers in the community. The customers in the community enter their own information, and the community customer characteristic information is converted into a digital ordered code according to the information digital distributed representation method to obtain the characteristic digital information of the customers in the community.

[0049] S7. According to the information digital distributed representation method, the identified express feature information containing text, name, address, telephone number and other features is converted into digital feature information, and then the express feature information is compiled into a string of ordered codes; when a certain express delivery bill feature information is not identified, the encoding uses a single-byte character 0 to identify the information;

[0050] The express feature information in S7 is obtained by photographing the express delivery bill with a camera during express delivery, and using optical character recognition technology based on deep learning to obtain the information on the express delivery bill. The recognized express delivery bill information is converted into digital feature information according to the information digitization method. The digital information is distributed in a step-by-step manner from front to back to form a string of ordered codes. If the bill information does not exist due to external factors, the non-existent information is converted into the number 0 when converted according to the information digitization method. The external factors include blurring, overlapping, stains and wrinkles on the express delivery bill.

[0051] S8. The express characteristic information obtained according to the information digital distributed representation method is a string of digital codes arranged in a predetermined order. It is compared with the digital code corresponding to the customer characteristic information in the database, and the digital address code, digital telephone code and digital name code are matched in turn. If they can all be matched correctly, the express will be bound to the customer; if they cannot be matched correctly, the information will be entered into the customer database.

[0052] The express characteristic information obtained by the information digital distributed representation method in S8 is a string of digital codes arranged in a predetermined order. If the express characteristic information and the customer characteristic information match correctly, the express will be bound to the corresponding customer; if they fail to match correctly, it is determined that the customer information of the customer to whom the express belongs is not in the customer database, and the express characteristic information is stored in the customer database as the customer characteristic information in the database.

[0053] It should be noted that, in this article, the terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or device including a series of elements includes not only those elements, but also other elements not explicitly listed, or also includes elements inherent to such process, method, article or device. In the absence of further restrictions, an element defined by the sentence "comprises a ..." does not exclude the existence of other identical elements in the process, method, article or device including the element.

[0054] The above are only preferred embodiments of the present invention, and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention specification and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.

Claims

1. A method for express information extraction and distributed representation based on deep learning, It is characterized in that The method comprises the following steps: S1. Obtain the express delivery bill dataset required for training, enrich the training set through data enhancement methods, preprocess the dataset and complete data annotation; S2. Use the YOLOv5 target detection network to train a network for express delivery bill area recognition, predict the bill position and rotation angle; perform detection model inference, perform rotated target detection on the bill according to the detected position and angle, output the bill image and bill rotation angle, and rotate the cut bill to obtain the complete bill area; S3. Use the YOLOv5 target detection network to train a model to detect the target text position and predict the target text position; perform detection model reasoning, calculate the text area coordinates based on the detected position, and obtain the YOLO target information box; S4. Use PaddleOCR optical character recognition technology to recognize the express delivery label information and obtain the PaddleOCR text box. Based on the target recognition results of YOLOv5 and the text recognition results of PaddleOCR optical character recognition technology, calculate the overlap between the coordinates of the PaddleOCR text box and the coordinates of the YOLO target information box to filter the text information; S5. Establish a digital distributed representation method for information to classify the obtained text information and store it in an Excel table; S6. Establish a complete customer database based on the distributed characterization method of digital information, where the characteristic information of each customer in the database includes name, address, and telephone number, and is stored digitally; S7. According to the information digital distributed representation method, the identified express feature information containing text name, address and telephone number is converted into digital feature information, and then the express feature information is compiled into a string of ordered codes; when a certain express delivery label feature information is not identified, the encoding uses a single-byte character 0 to identify the information; S8. The express characteristic information obtained according to the information digital distributed representation method is a string of digital codes arranged in a predetermined order. It is compared with the digital code corresponding to the customer characteristic information in the database, and the digital address code, digital telephone code and digital name code are matched in turn. If they can all be matched correctly, the express will be bound to the customer; if they cannot be matched correctly, the information will be entered into the customer database.

2. According to claim 1, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that Information digital distributed representation methods include: a) Identify the address and convert each segment of the text address into a digital address. The address is an 8-bit digital string; b) Identify the customer's telephone number, which is an 11-digit string; c) Identify the customer's name. The customer's name is a 12-digit string. According to the four-corner number dictionary, one Chinese character corresponds to four digits to form a coding string.

3. According to claim 1, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that The data annotation of the data set in S1 includes two types: annotating the express delivery bill area and the effective information area in the bill. The express delivery bill area data set collects multiple groups of express deliveries with bills shot at different angles, and converts the RGB image into a resolution of 640×480. The effective information in the bill data set collects the cropped express delivery bill, converts the grayscale image into a resolution of 640×480, and uses data enhancement methods for data enhancement. The data enhancement methods include rotation, cropping, flipping and noise addition; annotating the express delivery bill area is to use Labelme software to perform rectangular labeling on the express delivery bill area in the RGB image of the express delivery bill shot at different angles, and the label saving format is selected in YOLO format; annotating the effective information area in the bill is to use Labelme software to perform rectangular labeling on the express delivery number and recipient information in the grayscale image of the cropped express delivery bill, and the label saving format is selected in YOLO format.

4. According to claim 1, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that In S2, a model is obtained by training the YOLOv5 target detection network, and the model can effectively detect and cut out overlapping bills, multiple bills, blurred bills, bills with incomplete information, and wrinkled bills; bill extraction is to cut out the complete bill area, first rotate the bill angle, use OpenCV's affine transformation to perform preliminary correction, then use the orientation classification model to determine the bill orientation, and finally correct the bills that are inverted or sideways, and finally output the bills with the text facing up at a positive angle; bill information extraction is based on YOLOv5's rotation target detection, and rotates the bill to detect the bill, and cuts out the complete bill area according to the rotated rectangular frame predicted by the model, excluding other non-bill areas; the YOLOv5 target detection network training trained in S3 divides the samples into training set, validation set, and test set in a ratio of 3:1:1 to obtain a model for detecting the target text position; target position detection is to frame out the valid text area information in the image, including the customer's name, address, telephone number, and courier number in the bill.

5. According to claim 1, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that The PaddleOCR optical character recognition technology used in S4 is an optical character recognition technology based on a convolutional neural network. It converts text into image information through an optical input method, which includes scanning. The image information is then converted into editable computer text using text recognition technology. The text information is filtered by calculating the overlap between the PaddleOCR box coordinates and the YOLO box coordinates; the overlap between the PaddleOCR text box and the YOLOv5 text box is calculated, and the PaddleOCR text box with an overlap of 0.5 is retained to achieve the purpose of filtering text information.

6. According to claim 2, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that The preset order of the information digital distributed representation method in S5 is: from left to right, it includes community, building, household number, telephone number and name; 31 single-byte characters are used to form a code to represent a specific feature information, and the code consists of 5 segments, namely, community code segment, building code segment, household number code segment, telephone code segment and name code segment, wherein the community code segment consists of 2 single-byte characters, the building code segment consists of 2 single-byte characters, the household number code segment consists of 4 single-byte characters, the telephone code segment consists of 11 single-byte characters, and the name code segment consists of 12 single-byte characters.

7. According to claim 6, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that The address code segments of the information digitization distributed representation method in S5 include: one-tenth address code segment, one-hundredth address code segment, one-thousandth address code segment and one-ten-thousandth address code segment, the one-tenth address code segment is the first and second characters of the address code segment, the one-hundredth address code segment is the third and fourth characters of the address code segment, the one-thousandth address code segment is the fifth and sixth characters of the area code segment, and the one-ten-thousandth address code segment is the seventh and eighth characters of the address code segment.

8. According to claim 1, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that The complete customer database in S6 is the digital characteristic information of the customers in the community. The customers in the community enter their own information, and the community customer characteristic information is converted into a digital ordered code according to the information digital distributed representation method to obtain the characteristic digital information of the customers in the community.

9. According to claim 1, a method for extracting express information based on deep learning and its distributed representation, It is characterized in that The express feature information in S7 is obtained by photographing the express delivery bill with a camera during express delivery, and using optical character recognition technology based on deep learning to obtain the information on the express delivery bill. The recognized express delivery bill information is converted into digital feature information according to the information digitization method. The digital information is distributed in a step-by-step manner from front to back to form a string of ordered codes. If the bill information does not exist due to external factors, the non-existent information is converted into the number 0 when converted according to the information digitization method. The external factors include blurring, overlapping, stains and wrinkles on the express delivery bill.

10. The method for extracting express information based on deep learning and its distributed representation according to claim 1, It is characterized in that The express characteristic information obtained by the information digital distributed representation method in S8 is a string of digital codes arranged in a predetermined order. If the express characteristic information and the customer characteristic information match correctly, the express will be bound to the corresponding customer; if they fail to match correctly, it is determined that the customer information of the customer to whom the express belongs is not in the customer database, and the express characteristic information is stored in the customer database as the customer characteristic information in the database.

Citation Information

Patent Citations

  • Intelligent information extraction method for city-wide express orders

    CN113919345A

  • Dual-system dual-mode vehicle management system based on AI algorithm

    CN116129416A

  • Electricity consumption information input method and system based on image recognition and computer equipment

    CN118155226A

  • Synchronized Identity, Document, and Transaction Management

    US20230134651A1

  • Deep neural network model-based address information feature extraction method

    WO2021000362A1