CN111814514A - Number identification device, method and electronic device - Google Patents

Abstract

Embodiments of the present invention provide a number identification device, method, and electronic device. The method includes: using training images to train a training model for number recognition; performing coordinate transformation on a training image, and using one or more coordinate transformed images as positive samples of training data; using the The training model performs single-digit detection on the number in the image to be detected; and combines one or more numbers obtained through the single-digit detection to obtain the number in the image to be detected. Therefore, even in the case of few training samples, the training of the classifier can be completed simply and quickly, and has high recognition accuracy.

Description

Number identification device, method and electronic device

technical field

Embodiments of the present invention relate to image recognition technology, and in particular, to a number recognition device, method, and electronic device.

Background technique

With the continuous development of information technology, the application of number identification (or number detection) is also increasingly widespread. For example, signs, athletes, staff, candidates, etc. need to detect objects (or called objects to be detected) with numbers consisting of one or more numbers. In some scenarios, these numbers need to be automatically identified. various applications.

For example, in a basketball game, by detecting and recognizing the numbers on the players' jerseys, the corresponding players can be determined according to the recognition results. In this way, each athlete's trajectory can be tracked and depicted through video of the entire game, providing better technical assistance.

In traditional number recognition methods, a classifier is generally used to classify all possible numbers. For example, for a basketball player, whose possible numbers are 0 to 99, there would need to be 100 classes of classifiers to classify such numbers. That is, a large amount of training data needs to be collected and trained for each class.

It should be noted that the above description of the technical background is only for the convenience of clearly and completely describing the technical solutions of the present invention and facilitating the understanding of those skilled in the art. It should not be assumed that the above-mentioned technical solutions are well known to those skilled in the art simply because these solutions are described in the background section of the present invention.

SUMMARY OF THE INVENTION

The inventor found that the traditional number recognition method needs to collect a large amount of training data, so it consumes time and energy; and some categories of training data are difficult to collect, for example, for some rarely used numbers, it is difficult to collect the training data of these numbers Train the classifier.

Aiming at at least one of the above technical problems, the embodiments of the present invention provide a number identification device, method, and electronic device; it is expected that the training of the classifier can be completed simply and quickly under the condition of few training samples, and it has a high performance. recognition accuracy.

According to a first aspect of the embodiments of the present invention, a number identification device is provided, including:

A training unit, which uses a training image to train a training model for number recognition; wherein a certain training image is subjected to coordinate transformation, and one or more coordinate transformed images are used as positive samples of the training data;

a detection unit, which uses the training model to perform single-digit detection on the numbers in the images to be detected; and

A merging unit, which merges one or more numbers obtained by the single-digit detection to obtain the numbers in the image to be detected.

According to a second aspect of the embodiments of the present invention, a number identification method is provided, including:

Using the training image to train the training model for number recognition; wherein a certain training image is subjected to coordinate transformation, and one or more coordinate transformed images are used as positive samples of the training data;

using the training model to perform single digit detection on the number in the image to be detected; and

One or more numbers obtained by the single-digit detection are combined to obtain the number in the image to be detected.

According to a third aspect of the embodiments of the present invention, an electronic device is provided, including the above-mentioned number identification device.

One of the beneficial effects of the embodiments of the present invention is: performing coordinate transformation on a certain training image, and using one or more images after coordinate transformation as positive samples of training data; detecting; and combining one or more numbers obtained by the single-digit detection to obtain the number in the image to be detected. Therefore, even in the case of few training samples, the training of the classifier can be completed simply and quickly, and has high recognition accuracy.

With reference to the following description and drawings, specific implementations of embodiments of the present invention are disclosed in detail, indicating the manner in which the principles of the embodiments of the present invention may be employed. It should be understood that embodiments of the present invention are not thereby limited in scope. Embodiments of the invention include many changes, modifications and equivalents within the spirit and scope of the appended claims.

Features described and/or illustrated for one embodiment may be used in the same or similar manner in one or more other embodiments, in combination with, or instead of features in other embodiments .

It should be emphasized that the term "comprising/comprising" when used herein refers to the presence of a feature, integer, step or component, but does not exclude the presence or addition of one or more other features, integers, steps or components.

Description of drawings

The accompanying drawings, which are included to provide a further understanding of the embodiments of the invention, constitute a part of the specification, are used to illustrate embodiments of the invention, and together with the written description, serve to explain the principles of the invention. Obviously, the drawings in the following description are only some embodiments of the present invention, and for those of ordinary skill in the art, other drawings can also be obtained from these drawings without creative effort. In the attached image:

1 is a schematic diagram of a number identification method according to an embodiment of the present invention;

2 is a schematic diagram of an image to be detected according to an embodiment of the present invention;

Fig. 3 is a schematic diagram of the detection frame where the object to be detected is located and the single-digit detection result in Fig. 2;

4 is an exemplary diagram of a training image according to an embodiment of the present invention;

5 is a schematic diagram of the training image coordinate transformed according to an embodiment of the present invention;

6 is another example diagram of a training image according to an embodiment of the present invention;

7 is a schematic diagram of the training image coordinate transformed according to an embodiment of the present invention;

8 is another example diagram of a training image according to an embodiment of the present invention;

9 is a schematic diagram of a number identification device according to an embodiment of the present invention;

FIG. 10 is a schematic diagram of an electronic device according to an embodiment of the present invention.

Detailed ways

The foregoing and other features of embodiments of the present invention will become apparent from the following description with reference to the accompanying drawings. In the specification and drawings, specific embodiments of the invention are disclosed in detail, which are indicative of some of the embodiments in which the principles of the embodiments of the invention may be employed, it being understood that the invention is not limited to the described embodiments, but rather , the embodiments of the present invention include all modifications, variations and equivalents falling within the scope of the appended claims.

In the embodiments of the present invention, the terms "first", "second", etc. are used to distinguish different elements in terms of appellation, but do not indicate the spatial arrangement or temporal order of these elements, and these elements should not be used by these terms restricted. The term "and/or" includes any and all combinations of one or more of the associated listed items. The terms "comprising", "including", "having", etc. refer to the presence of stated features, elements, elements or components, but do not preclude the presence or addition of one or more other features, elements, elements or components.

In the embodiments of the present invention, the singular forms "a", "the", etc. include plural forms, and should be broadly understood as "a" or "a class" rather than being limited to the meaning of "an"; in addition, the term "the" "" is understood to include both the singular and the plural, unless the context clearly dictates otherwise. In addition, the term "based on" should be understood as "at least in part based on..." and the term "based on" should be understood as "based at least in part on..." unless the context clearly dictates otherwise.

In this embodiment of the present invention, a convolutional neural network (CNN, Convolutional Neural Network) is used as an example to illustrate the training model; for example, the convolutional neural network may be Faster R-CNN, FPN (Feature Pyramid Networks for object Detection ) or YOLO (You Only Look Once: Unified, Real-Time Object Detection), etc., the present invention is not limited thereto. For basic concepts and content such as training models, training samples, and training images, you can refer to related technologies.

Example 1

The embodiment of the present invention provides a number identification method. FIG. 1 is a schematic diagram of a number identification method according to an embodiment of the present invention. As shown in FIG. 1 , the method includes:

Step 101, using training images to train a training model for number recognition; wherein a certain training image is subjected to coordinate transformation, and one or more coordinate transformed images are used as positive samples of training data;

Step 102, using the training model to perform single-digit detection on the number in the image to be detected; and

Step 103: Combine one or more numbers obtained through the single-digit detection to obtain the numbers in the image to be detected.

In one embodiment, the object to be detected can be any object that has a requirement for an identification number. For example, the object to be detected is a license plate, an athlete in a competition, a sign containing a number, a staff member participating in an event, a player participating in an exam or a competition, and the like. For example, for a basketball or soccer player participating in a game, the possible numbers are 0 to 99, and for a track and field player participating in a game, the possible numbers are 0000 to 9999.

In one embodiment, the training image and the image to be detected are images that may contain objects to be detected. For example, if the object to be detected is a basketball player, both the training image and the image to be detected may be at least one image in a video of a basketball game.

In step 101, a training model may be trained based on a convolutional neural network (CNN, Convolutional Neural Network) using multiple training images. The specific structure of the convolutional neural network and how to train it can refer to related technologies. The coordinate transformation of the training image will be described in detail in the following embodiments.

In step 102, for the number in the image to be detected, one or more single-digit detections may be performed by using the CNN-based training model, and one or more numbers may be detected. Since the convolutional neural network has powerful target recognition ability, it can simplify the complex factors from the real environment, thereby further improving the detection efficiency and detection accuracy.

It should be noted that the above Figure 1 only schematically illustrates the embodiment of the present invention, but the present invention is not limited thereto. For example, the execution order of the various steps can be adjusted appropriately, and other steps can be added or some of the steps can be reduced. Those skilled in the art can make appropriate modifications according to the above content, and are not limited to the description of the above-mentioned FIG. 1 .

The present invention is exemplified below by taking the identification of the jersey number of a player in a basketball game as an example.

FIG. 2 is a schematic diagram of an image to be detected according to an embodiment of the present invention. As shown in FIG. 2 , the image to be detected is a frame of image in a video of a basketball game, which includes a plurality of players serving as objects to be detected. As shown in FIG. 2 , each athlete in the image to be detected can be detected, and the area where each athlete is located can be obtained, which is represented by each detection frame.

FIG. 3 is a schematic diagram of the detection frame where the object to be detected is located and the single-digit detection result in FIG. 2 . As shown in Figure 3, the detected areas of each athlete are represented by each detection box, and the single-digit detection results in detection boxes 1 to 7 are "no result", "3" and "0", "2", " no result", "2" and "3", "0", "8".

In step 103, one or more numbers obtained through single-digit detection may be combined to obtain the numbers in the image to be detected. For example, as shown in Fig. 3, the single-digit detection results "3" and "0" of detection frame 2 can be combined into "30", and the single-digit detection results "2" and "3" of detection frame 5 can be combined into "twenty three".

In the traditional number recognition method, for basketball or football players participating in the game, the possible numbers are 0 to 99, then the classifier needs 100 types; Then the classifier needs 10000 types.

In the embodiment of the present invention, since only single-digit detection needs to be performed, and only 10 types of training samples from 0 to 9 need to be collected, whether it is for numbers from 0 to 99 or numbers from 0000 to 9999, it can be easily The training of the classifier is completed quickly.

The above is an exemplary description of how to perform digit detection and number recognition, and the following further describes the training samples and training models.

In this embodiment of the present invention, coordinate transformation may be performed on a certain training image, and one or more images after coordinate transformation may be used as positive samples of the training data. Therefore, even for numbers with few training samples, the training of the classifier can be completed simply and quickly, and since single numbers are detected and merged, the recognition accuracy is high.

In one embodiment, the training image can be rotated by a preset angle; wherein the bounding box of the number in the training image is also rotated by the angle; the bounding box of the number after the rotation angle is used to obtain an outer bounding box; and adjusting the outer bounding box to obtain the coordinate-transformed image.

FIG. 4 is an example diagram of a training image according to an embodiment of the present invention, the number is “10”, and its bounding box is represented as 401. FIG. 5 is a schematic diagram of the training image after coordinate transformation according to an embodiment of the present invention. As shown in FIG. 5, the training image can be rotated by a preset angle θ; the bounding box 401 of the number in the training image is also rotated by the angle θ; the bounding box 401 of the number after the rotation angle is used Obtain an outer bounding box 501 ; and adjust the outer bounding box 501 to obtain the coordinate-transformed image (shown as a positive sample in FIG. 5 ), whose numbered bounding box is denoted as 502 .

In another embodiment, the numbers in the training images may be scaled in size to obtain the resized numbers; and a background area may be added to the resized numbers to obtain the coordinate transformations post image.

FIG. 6 is another example diagram of a training image according to an embodiment of the present invention, the number included in the training image is “8”, and the bounding box is represented as 601. FIG. 7 is a schematic diagram of the training image after coordinate transformation according to an embodiment of the present invention. As shown in FIG. 7 , the number in the training image can be scaled in size to obtain the number after changing the size; for example, the number “8” in 701 to 703 in FIG. 7 has been reduced to different degrees , 704 enlarges the number "8" to a certain extent. In addition, a background area (eg, the black background shown in FIG. 7 ) may be added to the resized numbers 701 to 704 to obtain the coordinate-transformed image (4 positive samples shown in FIG. 7 ).

Therefore, by performing coordinate transformation on the training image, more positive samples for training can be obtained, the training of the classifier can be completed simply and quickly even when there are few training samples, and the recognition accuracy can be further improved.

In one embodiment, negative samples of training data may also be added to the training model, so as to eliminate the influence of misidentification of non-numbered areas.

FIG. 8 is another example diagram of a training image according to an embodiment of the present invention. As shown in FIG. 8, the portion shown in box 801 is the player's arm, but resembles the number "1" in a tilted state. The training image can be used as a negative sample of the training data, so that when similar images are detected, no number identification is performed or the number "1" is not identified, thereby eliminating the influence of misrecognition in non-number areas and further improving recognition accuracy.

In one embodiment, before step 101, a number model without a pre-trained model can be trained; and the trained convolution layer of the number model can be used as the number model for the number Recognized pretrained model.

For example, when training a CNN-based training model, the general ImageNet data set is generally used as a pre-training model, but the ImageNet data set does not specifically classify numbers, so it is not conducive to the rapid convergence of the number recognition model. Suitable for number recognition.

In the embodiment of the present invention, a number model without a pre-training model can be trained first; and the trained convolution layer of the number model can be used as a pre-training model for number recognition. As a result, the trained model can converge faster and have higher detection accuracy.

The training model and training samples in step 101 or before have been described above, and the number recognition in step 103 will be described below.

In one embodiment, after the single-digit detection is performed in step 102 to obtain one or more digits, it may also be determined whether the one or more digits are detected according to the area and confidence of the one or more number boxes. false recognition.

For example, for a single-digit number i(0-9), use the following formula to determine whether it is the misidentification:

Among them, C _s represents the recognition result of the single-digit number, A represents the frame area of the single-digit number (which can be obtained through the single-digit detection process, as shown in Figures 2 and 3), and A _s represents the area threshold (which can be preset. ), S _i represents the confidence of the single-digit number (which can be obtained by CNN output), S _s represents the confidence threshold (which can be preset), and 100 represents the background.

For another example, for a double-digit number ij (10-99), use the following formula to determine whether it is the misidentification:

Among them, C _d represents the recognition result of the double-digit number, A _ij represents the combined box area of the double-digit number (which can be obtained through the single-digit detection process, as shown in Figures 2 and 3), and A _d represents the area threshold (which can be pre- S _i and S _j represent the confidence of the double-digit number (which can be obtained by CNN output), S _d represents the confidence threshold (which can be preset), and 100 represents the background.

It is worth noting that the above only takes 0-99 as an example for description, but the present invention is not limited to this. There are four cases of double digits (10-99), three digits (100-999), and four digits (1000-9999), respectively, to judge the misrecognition. The present invention is not limited to this, as long as the misrecognition is determined based on at least the area of the number frame and the confidence.

Thereby, the misrecognition of the number is determined based on at least the area of the number frame and the degree of confidence, the influence of the misrecognition can be further eliminated or reduced, and the recognition accuracy can be improved.

The above only describes each step or process related to the present invention, but the present invention is not limited thereto. The number identification method may further include other steps or processes, and for the specific content of these steps or processes, reference may be made to the prior art. In addition, the above only takes the above formulas as examples to illustrate the embodiments of the present invention, but the present invention is not limited to these formulas, and appropriate modifications can also be made to these formulas, and these modified implementations should be included in the embodiments of the present invention. within the range.

The above embodiments are only illustrative of the embodiments of the present invention, but the present invention is not limited thereto, and appropriate modifications can also be made on the basis of the above embodiments. For example, each of the above-described embodiments may be used alone, or one or more of the above-described embodiments may be combined.

It can be known from the above-mentioned embodiments that a certain training image is subjected to coordinate transformation, and one or more images after the coordinate transformation are used as positive samples of the training data; the training model is used to perform single-digit detection on the numbers in the images to be detected; One or more numbers obtained by the single-digit detection are combined to obtain the number in the image to be detected. Therefore, even in the case of few training samples, the training of the classifier can be completed simply and quickly, and has high recognition accuracy.

Example 2

An embodiment of the present invention provides a number identification device, and the same content as that of Embodiment 1 will not be repeated.

FIG. 9 is a schematic diagram of a number identification device according to an embodiment of the present invention. As shown in FIG. 9 , the number identification device 900 includes:

A training unit 901, which uses a training image to train a training model for number recognition; wherein a certain training image is subjected to coordinate transformation, and one or more coordinate transformed images are used as positive samples of the training data;

A detection unit 902, which uses the training model to perform single-digit detection on the numbers in the images to be detected; and

A merging unit 903, which merges one or more numbers obtained by the single-digit detection to obtain the numbers in the to-be-detected image.

In one embodiment, the training unit 901 may be configured to: rotate the training image by a preset angle; wherein the bounding box of the number in the training image is also rotated by the angle; using the rotated angle obtaining an outer bounding box with the bounding box of the number; and adjusting the outer bounding box to obtain the coordinate-transformed image.

In one embodiment, the training unit 901 may be configured to: scale the number in the training image in size to obtain the number after the size change; and for the number after the size change A background area is added to obtain the coordinate transformed image.

In one embodiment, the training unit 901 may be further configured to: add negative samples of training data to the training model, so as to eliminate the influence of misidentification of non-number regions.

In one embodiment, the training unit 901 may also be used to: train a number model without a pre-trained model; and use the trained convolution layer of the number model as a pre-trained model for number recognition.

As shown in FIG. 9, the number identification device 900 may further include:

A determination unit 904, which determines whether the one or more numbers are misidentified according to the detected area and confidence of the one or more number boxes.

In one embodiment, for a single-digit number i, the determining unit 904 can use the following formula to determine whether it is the misidentification:

Among them, C _s represents the recognition result of the single-digit number, A represents the frame area of the single-digit number, A _s represents the area threshold, S _i represents the confidence of the single-digit number, S _s represents the confidence threshold, and 100 means background.

In one embodiment, for the double-digit number ij, the determining unit 904 can use the following formula to determine whether it is the misidentification:

Among them, C _d represents the recognition result of the double-digit number, A _ij represents the area of the combo box of the double-digit number, A _d represents the area threshold, S _i and S _j represent the confidence of the double-digit number, and S _d represents the confidence Degree threshold, 100 means background.

It should be noted that the above only describes the components related to the present invention, but the present invention is not limited thereto. The number identification device 900 may also include other components or modules, and for the specific content of these components or modules, reference may be made to the prior art.

In addition, for the sake of simplicity, FIG. 9 only exemplarily shows the connection relationship or signal direction between various components or modules, but it should be clear to those skilled in the art that various related technologies such as bus connection can be used. The foregoing components or modules may be implemented by hardware facilities such as a processor and a memory, which are not limited in this embodiment of the present invention.

The above embodiments are only illustrative of the embodiments of the present invention, but the present invention is not limited thereto, and appropriate modifications can also be made on the basis of the above embodiments. For example, each of the above-described embodiments may be used alone, or one or more of the above-described embodiments may be combined.

It can be known from the above-mentioned embodiments that a certain training image is subjected to coordinate transformation, and one or more images after the coordinate transformation are used as positive samples of the training data; the training model is used to perform single-digit detection on the numbers in the images to be detected; One or more numbers obtained by the single-digit detection are combined to obtain the number in the image to be detected. Therefore, even in the case of few training samples, the training of the classifier can be completed simply and quickly, and has high recognition accuracy.

Example 3

An embodiment of the present invention provides an electronic device, including the number identification device described in Embodiment 2, the content of which is incorporated herein. The electronic device may be, for example, a computer, a server, a workstation, a laptop computer, a smart phone, etc.; but the embodiment of the present invention is not limited thereto.

FIG. 10 is a schematic diagram of an electronic device according to an embodiment of the present invention. As shown in FIG. 10 , the electronic device 1000 may include: a processor (eg, a central processing unit CPU) 1010 and a memory 1020 ; the memory 1020 is coupled to the central processing unit 1010 . The memory 1020 can store various data; in addition, a program 1021 for information processing is also stored, and the program 1021 is executed under the control of the processor 1010 .

In one embodiment, the functions of the number identification device 900 may be integrated into the processor 1010 for implementation. The processor 1010 may be configured to implement the number identification method described in Embodiment 1.

In another embodiment, the number identification device 900 may be configured separately from the processor 1010 . For example, the number identification device 900 may be configured as a chip connected to the processor 1010 , and the functions of the number identification device 900 may be implemented through the control of the processor 1010 . .

In one embodiment, the processor 1010 may be configured to perform the following controls: use the training images to train a training model for number recognition; wherein coordinate transformation is performed on a training image, and one or more coordinates are transformed use the training model to perform single-digit detection on the numbers in the image to be detected; and combine one or more numbers obtained through the single-digit detection to obtain the image to be detected number in .

In one embodiment, the processor 1010 may be further configured to perform the following controls: rotate the training image by a preset angle; wherein the bounding box of the number in the training image is also rotated by the angle; use the rotation The numbered bounding box after the angle obtains an outer bounding box; and the outer bounding box is adjusted to obtain the coordinate transformed image.

In one embodiment, the processor 1010 may be further configured to perform the following controls: scaling the number in the training image in size to obtain the resized number; and for the resized number The number of increases the background area to obtain the coordinate-transformed image.

In one embodiment, the processor 1010 may be further configured to perform the following control: add negative samples of training data to the training model, so as to eliminate the influence of misidentification of non-number regions.

In one embodiment, the processor 1010 may be further configured to perform the following controls: train a number model without a pre-trained model; and use the trained convolutional layer of the number model as a pre-training for number recognition. Train the model.

In one embodiment, the processor 1010 may be further configured to perform the following control: according to the detected area and confidence level of the one or more number boxes, determine whether the one or more numbers are misidentified.

For example, for a single-digit number i, use the following formula to determine whether it is the misidentification:

Among them, C _s represents the recognition result of the single-digit number, A represents the frame area of the single-digit number, A _s represents the area threshold, S _i represents the confidence of the single-digit number, S _s represents the confidence threshold, and 100 means background.

For example, for a double-digit number ij, use the following formula to determine whether it is the misidentification:

Among them, C _d represents the recognition result of the double-digit number, A _ij represents the area of the combo box of the double-digit number, A _d represents the area threshold, S _i and S _j represent the confidence of the double-digit number, and S _d represents the confidence Degree threshold, 100 means background.

In addition, as shown in FIG. 10 , the electronic device 1000 may further include: an input/output (I/O) device 1030, a display 1040, etc.; wherein, the functions of the above components are similar to those in the prior art, and will not be repeated here. It is worth noting that the electronic device 1000 does not necessarily include all the components shown in FIG. 10 ; in addition, the electronic device 1000 may also include components not shown in FIG. 10 , and reference may be made to the related art.

An embodiment of the present invention further provides a computer-readable program, wherein when the program is executed in an electronic device, the program causes a computer to execute the number identification method described in Embodiment 1 in the electronic device.

An embodiment of the present invention further provides a storage medium storing a computer-readable program, wherein the computer-readable program causes a computer to execute the number identification method described in Embodiment 1 in an electronic device.

The above apparatus and method of the present invention may be implemented by hardware, or may be implemented by hardware combined with software. The present invention relates to a computer-readable program which, when executed by logic components, enables the logic components to implement the above-described apparatus or constituent components, or causes the logic components to implement the above-described various methods or steps. The present invention also relates to a storage medium for storing the above program, such as a hard disk, a magnetic disk, an optical disk, a DVD, a flash memory, and the like.

The method/apparatus described in conjunction with the embodiments of the present invention may be directly embodied as hardware, a software module executed by a processor, or a combination of the two. For example, one or more of the functional block diagrams and/or one or more combinations of the functional block diagrams shown in the figures may correspond to either software modules or hardware modules of the computer program flow. These software modules may respectively correspond to the various steps shown in the figure. These hardware modules can be implemented by, for example, solidifying these software modules using a Field Programmable Gate Array (FPGA).

A software module may reside in RAM memory, flash memory, ROM memory, EPROM memory, EEPROM memory, registers, hard disk, removable disk, CD-ROM, or any other form of storage medium known in the art. A storage medium can be coupled to the processor, such that the processor can read information from, and write information to, the storage medium; or the storage medium can be an integral part of the processor. The processor and storage medium may reside in an ASIC. The software module can be stored in the memory of the mobile terminal, or can be stored in a memory card that can be inserted into the mobile terminal. For example, if a device (such as a mobile terminal) adopts a larger-capacity MEGA-SIM card or a large-capacity flash memory device, the software module can be stored in the MEGA-SIM card or a large-capacity flash memory device.

For one or more of the functional blocks and/or one or more combinations of the functional blocks described in the figures, it can be implemented as a general-purpose processor, a digital signal processor (DSP) for performing the functions described in this application ), Application Specific Integrated Circuits (ASICs), Field Programmable Gate Arrays (FPGAs) or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, or any suitable combination thereof. One or more of the functional blocks and/or one or more combinations of the functional blocks described with respect to the figures can also be implemented as a combination of computing devices, eg, a combination of a DSP and a microprocessor, multiple microprocessors processor, one or more microprocessors in communication with the DSP, or any other such configuration.

The present invention has been described above with reference to the specific embodiments, but those skilled in the art should understand that these descriptions are all exemplary and do not limit the protection scope of the present invention. Various variations and modifications of the present invention can be made by those skilled in the art in accordance with the principles of the present invention, and these variations and modifications are also within the scope of the present invention.

Claims (10)

1. A number identification device, wherein the device comprises: A training unit, which uses a training image to train a training model for number recognition; wherein a certain training image is subjected to coordinate transformation, and one or more coordinate transformed images are used as positive samples of the training data; a detection unit, which uses the training model to perform single-digit detection on the numbers in the images to be detected; and A merging unit, which merges one or more numbers obtained by the single-digit detection to obtain the numbers in the image to be detected. 2. The apparatus according to claim 1, wherein the training unit is used for: rotating the training image by a preset angle; wherein the bounding box of the number in the training image is also rotated by the angle; using The numbered bounding box after the rotation angle obtains an outer bounding box; and the outer bounding box is adjusted to obtain the coordinate-transformed image. 3. The apparatus according to claim 1, wherein the training unit is used for: scaling the numbers in the training images in size to obtain the numbers after changing the size; After the number, the background area is added to obtain the coordinate-transformed image. 4 . The apparatus according to claim 1 , wherein the training unit is further configured to: add negative samples of training data to the training model, so as to eliminate the influence of misidentification of non-number regions. 5 . 5. device according to claim 1, wherein, described training unit is also used for: training does not have the number model of pre-training model; And the convolution layer of described number model after training is used for number recognition pretrained model. 6. The apparatus of claim 1, wherein the apparatus further comprises: A determination unit, which determines whether the one or more numbers are misidentified according to the detected area and confidence of the one or more number boxes. 7. The device according to claim 6, wherein, for single-digit number i, use the following formula to determine whether it is the misidentification:

Among them, C _s represents the recognition result of the single-digit number, A represents the frame area of the single-digit number, A _s represents the area threshold, S _i represents the confidence of the single-digit number, S _s represents the confidence threshold, and 100 means background. 8. The device according to claim 6, wherein, for the double-digit number ij, the following formula is used to determine whether it is the misidentification:

Among them, C _d represents the recognition result of the double-digit number, A _ij represents the area of the combo box of the double-digit number, A _d represents the area threshold, S _i and S _j represent the confidence of the double-digit number, and S _d represents the confidence Degree threshold, 100 means background. 9. A number identification method, characterized in that the method comprises: Using the training image to train the training model for number recognition; wherein a certain training image is subjected to coordinate transformation, and one or more coordinate transformed images are used as positive samples of the training data; using the training model to perform single digit detection on the number in the image to be detected; and One or more numbers obtained by the single-digit detection are combined to obtain the number in the image to be detected. 10. An electronic device comprising the number identification device according to any one of claims 1 to 8.

Images