A frame title recognition method, storage medium and device based on target detection

Abstract

The present invention discloses a frame title recognition method, storage medium and device based on target detection. By classifying supported resources and sampling a part of each type of resource, a corresponding recognition model is obtained through training. The subsequent frame title can be automatically realized, which not only greatly reduces the workload of early template entry, but also reduces the workload of subsequent frame title, and improves the degree of automation of intelligent correction.

Description

Technical Field

[0001] The present invention relates to the field of intelligent education technology, and in particular to a frame title recognition method, storage medium and device based on target detection. Background Art

[0002] Currently, question recognition for teaching aids or test papers generally involves pre-entering the template question coordinates of each supported teaching aid or test paper resource into the intelligent system. The questions are then segmented and identified from the captured target image of the teaching aid or test paper based on the coordinates. While this solution is feasible, recording test paper and teaching aid templates is very cumbersome. Not only does it require marking the coordinates of each page of the teaching aid, but it also requires associating the framed questions and coordinates on each page. This work is mostly done manually, which is prone to coordinate marking errors or manual association errors. This results in high labor costs and a relatively high probability of error. Summary of the Invention

[0003] In view of the shortcomings of the existing technology, the present invention aims to provide a frame title recognition method, storage medium and device based on target detection.

[0004] In order to achieve the above object, the present invention adopts the following technical solutions:

[0005] A frame title recognition method based on target detection, the specific process is as follows:

[0006] S1. Obtain the hierarchical structure of the target resource, obtain samples of the target resource and annotate them, and use the annotated samples to train a recognition model for the target resource:

[0007] S1.1. Determine the hierarchical structure of the target resource. The hierarchical structure of the topic refers to the levels of topics included;

[0008] S1.2. Classify the resources that support identification;

[0009] S1.3. For each type of resource, select a set number of resources and collect images of each page of the selected resources as samples. Each level of the topic structure requires corresponding samples.

[0010] S1.4. Label each sample of each resource type. Using the question hierarchy structure determined in step S1.1, use a rectangle to outline the serial number and punctuation marks corresponding to each level of each question in the sample, and then label the level to which it belongs.

[0011] S1.5. For each resource type, a neural network is trained using the corresponding sample to obtain a corresponding recognition model for each resource type. Each recognition model is then associated with information about the supported resources, including resource ID, resource name, resource number, resource publisher, resource editor, applicable grade level, and applicable subject.

[0012] S2. Before performing frame title recognition, first obtain the resource ID to be recognized; then match the corresponding recognition model according to the obtained resource ID;

[0013] S3, collecting the target image to be identified and preprocessing the target image;

[0014] S4. Using the recognition model matched in step S2, the target image to be recognized is recognized, the serial number of each level in each question is identified, and then the coordinate data of the recognized serial number of each level is obtained;

[0015] S5. Select the question box of each level; the height of the rectangular box corresponding to each question is determined by the height between the upper left corner vertical coordinate of the serial number of the question and the upper left corner vertical coordinate of the serial number of the next question of the same level, and the width of the rectangular box is determined by the length of the target image.

[0016] Furthermore, in step S3, the preprocessing includes rectangle correction and curved surface correction.

[0017] Furthermore, in step S5, when there is no other question below a certain question, the height of the corresponding rectangular frame is the height of the upper left corner vertical coordinate of the serial number of the question and the lower left corner vertical coordinate of the target image.

[0018] The present invention also provides a computer-readable storage medium, wherein the computer-readable storage medium stores a computer program, and the computer program implements the above method when executed by a processor.

[0019] The present invention also provides a computer device, characterized in that it includes a processor and a memory, the memory is used to store a computer program; and the processor is used to implement the above method when executing the computer program.

[0020] The beneficial effects of the present invention are as follows: the present invention classifies the supported resources, samples a part of each type of resource, and trains a corresponding recognition model, which can be automatically implemented when performing subsequent frame questions. This not only greatly reduces the workload of early template entry, but also reduces the workload of subsequent frame questions, thereby improving the degree of automation of intelligent correction. DETAILED DESCRIPTION

[0021] The present invention will be further described below. It should be noted that this embodiment is based on the technical solution and provides a detailed implementation method and specific operation process, but the protection scope of the present invention is not limited to this embodiment.

[0022] This embodiment provides a method for identifying title frames based on target detection. The specific process is as follows:

[0023] S1. Obtain the hierarchical structure of the target resource, obtain samples of the target resource and annotate them, and use the annotated samples to train a recognition model for the target resource. The specific process is as follows:

[0024] S1.1. Determine the hierarchical structure of the target resource. The hierarchical structure of the questions refers to the levels of questions (e.g., some questions contain only a first-level question, some contain a first-level question - a second-level question, some contain a first-level question - a second-level question - a third-level question, etc.);

[0025] S1.2. Categorize resources that support identification. For example, a series of supplementary teaching materials published by the same publisher uses a similar editorial layout, so they can be categorized by publisher. Alternatively, resources from the same series of the same subject published by the same publisher can be categorized further (e.g., for a workbook named "XXX" published by a publisher for the subject of Chinese, all workbooks applicable to different grades can be classified as the same type).

[0026] S1.3. For each type of resource, select a set number of resources and collect images of each page of the selected resources as samples. Each level of the topic structure requires corresponding samples.

[0027] S1.4. Label each sample of each type of resource. According to the question hierarchy structure determined in step S1.1, use a rectangle to frame the serial number corresponding to each level of each question in the sample and the punctuation mark after the serial number, and mark the level to which it belongs.

[0028] S1.5. For each type of resource, a neural network is trained using the corresponding samples to obtain a recognition model corresponding to each type of resource, and each recognition model is associated with the information of the supported resources; the resource information includes resource ID, resource name, resource number, resource publisher, resource editor, applicable grade, applicable subject, etc.

[0029] S2. Before performing frame title recognition, first obtain the resource ID to be recognized (which can be input manually, by scanning, by image recognition, etc.); then match the corresponding recognition model according to the obtained resource ID;

[0030] S3. Capture the target image to be identified and pre-process the target image, including rectangle correction, curved surface correction, etc.

[0031] S4. Using the recognition model matched in step S2, the target image to be recognized is recognized, the serial number of each level in each question is identified, and then the coordinate data of the recognized serial number of each level is obtained;

[0032] S5. Select the question box of each level; the height of the rectangular box corresponding to each question is determined by the height between the upper left corner vertical coordinate of the serial number of the question and the upper left corner vertical coordinate of the serial number of the next question of the same level, and the width of the rectangular box is determined by the length of the target image.

[0033] When there is no other question below a certain question, the height of the corresponding rectangular box is the height of the upper left corner vertical coordinate of the question number and the lower left corner vertical coordinate of the target image.

[0034] After completing the frame questions using the above method, you can further perform intelligent correction.

[0035] Those skilled in the art can make various corresponding changes and modifications based on the above technical solutions and concepts, and all of these changes and modifications should be included in the scope of protection of the claims of the present invention.

Claims

1. A frame title recognition method based on target detection, characterized in that: The specific process is: S1. Obtain the hierarchical structure of the target resource, obtain samples of the target resource and annotate them, and use the annotated samples to train a recognition model for the target resource: S1.

1. Determine the hierarchical structure of the target resource. The hierarchical structure of the topic refers to the levels of topics included; S1.

2. Classify the resources that support identification; S1.

3. For each type of resource, select a set number of resources and collect images of each page of the selected resources as samples. Each level of the topic structure requires corresponding samples. S1.

4. Label each sample of each resource type. Using the question hierarchy structure determined in step S1.1, use a rectangle to outline the serial number and punctuation marks corresponding to each level of each question in the sample, and then label the level to which it belongs. S1.

5. For each resource type, a neural network is trained using the corresponding sample to obtain a corresponding recognition model for each resource type. Each recognition model is then associated with information about the supported resources, including resource ID, resource name, resource number, resource publisher, resource editor, applicable grade level, and applicable subject. S2. Before performing frame title recognition, first obtain the resource ID to be recognized; then match the corresponding recognition model according to the obtained resource ID; S3, collecting the target image to be identified and preprocessing the target image; S4. Using the recognition model matched in step S2, the target image to be recognized is recognized, the serial number of each level in each question is identified, and then the coordinate data of the recognized serial number of each level is obtained; S5. Select the question box of each level; the height of the rectangular box corresponding to each question is determined by the height between the upper left corner vertical coordinate of the serial number of the question and the upper left corner vertical coordinate of the serial number of the next question of the same level, and the width of the rectangular box is determined by the length of the target image.

2. The method according to claim 1, characterized in that In step S3, the preprocessing includes rectangle correction and curved surface correction.

3. The method according to claim 1, characterized in that In step S5, when there is no other question below a certain question, the height of the corresponding rectangular frame is the height of the upper left corner vertical coordinate of the serial number of the question and the lower left corner vertical coordinate of the target image.

4. A computer-readable storage medium, characterized in that The computer-readable storage medium stores a computer program, and when the computer program is executed by a processor, the method according to any one of claims 1 to 3 is implemented.

5. A computer device, characterized in that: The method comprises a processor and a memory, wherein the memory is used to store a computer program; and when the processor is used to execute the computer program, the method according to any one of claims 1 to 3 is implemented.