Industrial text detection method and device, electronic equipment and storage medium

Abstract

The invention provides an industrial text detection method and device, electronic equipment and a storage medium, and relates to the technical field of computers. The method comprises the following steps: acquiring a to-be-processed image; n stages of feature extraction are carried out on the to-be-processed image based on a feature pyramid network, pyramid features of N sizes are correspondingly obtained, the feature pyramid network takes a separable network as a main network, the pyramid features of one size are obtained through feature extraction of any stage, and N is a positive integer; determining a target feature map according to the pyramid features of N sizes; and determining an industrial text detection result according to the target feature map. According to the invention, feature extraction can be carried out based on the feature pyramid network taking the separable network as the main network, so that the parameter quantity and the calculation quantity are effectively reduced, the efficiency is improved, the accuracy and reliability of feature extraction can be ensured, and the accuracy of a detection result and the detection efficiency are effectively improved.

Description

Background Art

[0002] With the development of computer technology, the use of electronic products has become more frequent, which results in the generation of a large number of pictures and video files every moment. Among them, text is the most intuitive and valuable information in pictures and videos, so extracting text information from images has a wide range of application backgrounds and great research significance.

[0003] In the related art, text with a single background and regular shape can be detected, but for more complex industrial scenes, such as engraved characters, embossed characters, and steel-printed characters, they are the same color as the mother body and are three-dimensional characters. Therefore, text detection based on the related art will reduce the accuracy of the detection results and cannot ensure the detection efficiency.

[0004] It should be noted that the information disclosed in the above background technology section is only used to enhance the understanding of the background of the present disclosure, and therefore may include information that does not constitute the prior art known to ordinary technicians in the field. Summary of the invention

[0005] The present disclosure provides an industrial text detection method, device, electronic device and storage medium, which at least to a certain extent overcome the problem that the accuracy of detection results in related technologies is reduced and the detection efficiency cannot be ensured.

[0006] Other features and advantages of the present disclosure will become apparent from the following detailed description, or may be learned in part by the practice of the present disclosure.

[0007] According to one aspect of the present disclosure, an industrial text detection method is provided, including: acquiring an image to be processed; performing N-stage feature extraction on the image to be processed based on a feature pyramid network, and correspondingly obtaining pyramid features of N sizes, wherein the feature pyramid network uses a separable network as a main network, and feature extraction at any stage obtains a pyramid feature of one size, where N is a positive integer; determining a target feature map based on the pyramid features of N sizes; and determining an industrial text detection result based on the target feature map.

[0008] In some exemplary embodiments, performing N-stage feature extraction on the image to be processed based on the feature pyramid network to obtain pyramid features of N sizes, including: performing N-stage feature extraction on the image to be processed based on the feature pyramid network to obtain pyramid features of N sizes, wherein the size of the pyramid feature obtained by the feature extraction of the nth stage is 1 / 2 of the original size. n , n is a positive integer less than or equal to N.

[0009] In some exemplary embodiments, a target feature map is determined based on pyramid features of N sizes, including: based on the order of each stage from back to front, performing the following operations stage by stage until a fused feature map corresponding to each stage is obtained: upsampling the fused feature map corresponding to the nth stage by a factor of 2, and adding element by element to the pyramid feature obtained in the n-1th stage to obtain a fused feature map corresponding to the n-1th stage, wherein the fused feature map corresponding to the last stage is the pyramid feature obtained by feature extraction in the last stage; and determining the target feature map based on the fused feature maps corresponding to each stage.

[0010] In some exemplary embodiments, a target feature map is determined based on the fused feature maps corresponding to each stage, including: performing upsampling processing and dimensionality reduction processing of corresponding multiples on the fused feature maps corresponding to each stage to obtain each feature map to be spliced; and splicing each feature map to be spliced ​​to obtain the target feature map.

[0011] In some exemplary embodiments, the backbone network of the feature pyramid network includes a first separable network unit with a step size of 1 and a second separable network unit with a step size of 2; wherein, feature extraction at any stage in the feature pyramid network corresponds to a convolutional layer, and any convolutional layer contains the first separable network unit and the second separable network unit.

[0012] In some exemplary embodiments, when N=5, the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the second stage is 1, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the third stage is 3, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fourth stage is 7, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fifth stage is 1, and the number of repetitions of the second separable network unit is 1.

[0013] According to another aspect of the present disclosure, an industrial text detection device is also provided, including: a module for acquiring an image to be processed, used to acquire an image to be processed; a pyramid feature determination module, used to perform N-stage feature extraction on the image to be processed based on a feature pyramid network, and obtain pyramid features of N sizes accordingly, wherein the feature pyramid network uses a separable network as a backbone network, and feature extraction at any stage obtains a pyramid feature of one size, and N is a positive integer; a target feature map determination module, used to determine a target feature map based on the pyramid features of N sizes; and an industrial text detection result determination module, used to determine an industrial text detection result based on the target feature map.

[0014] In some exemplary embodiments, the pyramid feature determination module is used to perform N-stage feature extraction on the image to be processed based on the feature pyramid network, and obtain pyramid features of N sizes, wherein the size of the pyramid feature obtained by the feature extraction of the nth stage is 1 / 2 of the original size. n , n is a positive integer less than or equal to N.

[0015] In some exemplary embodiments, the target feature map determination module is used to perform the following operations stage by stage based on the order of each stage from back to front until a fused feature map corresponding to each stage is obtained: the fused feature map corresponding to the nth stage is upsampled by a factor of 2, and is added element by element to the pyramid feature obtained in the n-1th stage to obtain a fused feature map corresponding to the n-1th stage, wherein the fused feature map corresponding to the last stage is the pyramid feature obtained by feature extraction in the last stage; and the target feature map is determined based on the fused feature maps corresponding to each stage.

[0016] In some exemplary embodiments, the target feature map determination module is used to perform upsampling processing and dimensionality reduction processing of corresponding multiples according to the fusion feature maps corresponding to each stage to obtain each feature map to be spliced; and splice each feature map to be spliced ​​to obtain the target feature map.

[0017] In some exemplary embodiments, the backbone network of the feature pyramid network includes a first separable network unit with a step size of 1 and a second separable network unit with a step size of 2; wherein, feature extraction at any stage in the feature pyramid network corresponds to a convolutional layer, and any convolutional layer contains the first separable network unit and the second separable network unit.

[0018] In some exemplary embodiments, when N=5, the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the second stage is 1, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the third stage is 3, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fourth stage is 7, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fifth stage is 1, and the number of repetitions of the second separable network unit is 1.

[0019] According to another aspect of the present disclosure, an electronic device is also provided, which includes: a processor; and a memory for storing executable instructions of the processor; wherein the processor is configured to execute any one of the above-mentioned industrial text detection methods by executing the executable instructions.

[0020] According to another aspect of the present disclosure, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, any one of the above industrial text detection methods is implemented.

[0021] According to another aspect of the present disclosure, a computer program product or a computer program is provided, the computer program product or the computer program including computer instructions, the computer instructions being stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the industrial text detection method provided in various optional ways in any embodiment of the present disclosure.

[0022] The technical solution provided in the embodiments of the present disclosure can perform N-stage feature extraction on the processed image based on a feature pyramid network with a separable network as the main network, thereby determining the industrial text detection result according to the target feature map determined by the pyramid feature. The embodiments of the present disclosure can perform feature extraction based on a feature pyramid network with a separable network as the main network, which can reduce the number of parameters and the amount of calculation, thereby improving efficiency and ensuring the accuracy and reliability of feature extraction, thereby effectively improving the accuracy of the detection result and improving the detection efficiency.

[0023] It is to be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present disclosure. BRIEF DESCRIPTION OF THE DRAWINGS

[0024] The accompanying drawings herein are incorporated into the specification and constitute a part of the specification, illustrate embodiments consistent with the present disclosure, and together with the specification are used to explain the principles of the present disclosure. Obviously, the accompanying drawings described below are only some embodiments of the present disclosure, and for ordinary technicians in this field, other accompanying drawings can be obtained based on these accompanying drawings without creative work.

[0025] Figure 1 A schematic diagram showing a system architecture in an embodiment of the present disclosure;

[0026] Figure 2 A flow chart of an industrial text detection method in an embodiment of the present disclosure is shown;

[0027] Figure 3 A schematic diagram showing a first detachable network unit in an embodiment of the present disclosure;

[0028] Figure 4 A schematic diagram showing a second detachable network unit in an embodiment of the present disclosure;

[0029] Figure 5A schematic diagram showing a process of feature extraction and feature fusion in an embodiment of the present disclosure is shown;

[0030] Figure 6 A schematic diagram showing the structure of a DBNet network in an embodiment of the present disclosure is shown;

[0031] Figure 7 A schematic diagram showing an lr curve in an embodiment of the present disclosure;

[0032] Figure 8 A schematic diagram of industrial text detection effect in an embodiment of the present disclosure is shown;

[0033] Fig. 9 A schematic diagram of an industrial text detection device in an embodiment of the present disclosure is shown;

[0034] Fig.10 A structural block diagram of an electronic device according to an embodiment of the present disclosure is shown;

[0035] Fig.11 A schematic diagram of a computer-readable storage medium in an embodiment of the present disclosure is shown. DETAILED DESCRIPTION

[0036] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that the disclosure will be more comprehensive and complete and to fully convey the concepts of the example embodiments to those skilled in the art. The described features, structures, or characteristics may be combined in any suitable manner in one or more embodiments.

[0037] In addition, the accompanying drawings are only schematic illustrations of the present disclosure and are not necessarily drawn to scale. The same reference numerals in the figures represent the same or similar parts, and their repeated description will be omitted. Some of the block diagrams shown in the accompanying drawings are functional entities and do not necessarily correspond to physically or logically independent entities. These functional entities can be implemented in software form, or implemented in one or more hardware modules or integrated circuits, or implemented in different networks and / or processor devices and / or microcontroller devices.

[0038] The specific implementation of the embodiment of the present disclosure is described in detail below with reference to the accompanying drawings.

[0039] Figure 1 FIG. 2 shows an exemplary application system architecture diagram to which the industrial text detection method in the embodiment of the present disclosure can be applied. Figure 1 As shown, the system architecture may include a terminal device 101 , a network 102 and a server 103 .

[0040] Among them, the terminal device 101 can send the image to be processed to the server 103. The server 103 can perform N stages of feature extraction on the image to be processed based on the feature pyramid network, and obtain pyramid features of N sizes accordingly, wherein the feature pyramid network uses a separable network as the backbone network, and the feature extraction of any stage obtains a pyramid feature of a size, and N is a positive integer. Then, the server 103 can determine the target feature map based on the pyramid features of N sizes. Therefore, the server 103 can determine the industrial text detection result based on the target feature map. Optionally, the server 103 can send the industrial text detection result to the terminal device 101.

[0041] Alternatively, the terminal device 101 can obtain an image to be processed. The terminal device 101 can perform N stages of feature extraction on the image to be processed based on a feature pyramid network, and obtain pyramid features of N sizes, wherein the feature pyramid network uses a separable network as a main network, and feature extraction at any stage obtains a pyramid feature of a size, and N is a positive integer. Then, the terminal device 101 can determine a target feature map based on the pyramid features of N sizes. Thus, the terminal device 101 can determine the industrial text detection result based on the target feature map. Optionally, the terminal device 101 can also send the industrial text detection result to the server 103 for storage.

[0042] The network 102 is a medium for providing a communication link between the terminal device 101 and the server 103, and can be a wired network or a wireless network.

[0043] Optionally, the wireless network or wired network described above uses standard communication technology and / or protocol. The network is usually the Internet, but it can also be any network, including but not limited to a local area network (LAN), a metropolitan area network (MAN), a wide area network (WAN), a mobile, wired or wireless network, a dedicated network or any combination of a virtual private network). In some embodiments, the data exchanged through the network is represented by technologies and / or formats including Hyper Text Mark-up Language (HTML), Extensible Markup Language (XML), etc. In addition, conventional encryption technologies such as Secure Socket Layer (SSL), Transport Layer Security (TLS), Virtual Private Network (VPN), Internet Protocol Security (IPSec) can also be used to encrypt all or some links. In other embodiments, customized and / or dedicated data communication technologies can also be used to replace or supplement the above data communication technologies.

[0044] The terminal device 101 can be various electronic devices, including but not limited to smart phones, tablet computers, laptop computers, desktop computers, smart speakers, smart watches, wearable devices, augmented reality devices, virtual reality devices, etc.

[0045] Optionally, the client of the application installed in different terminal devices 101 is the same, or the client of the same type of application based on different operating systems. Based on the different terminal platforms, the specific form of the client of the application can also be different, for example, the application client can be a mobile client, a PC client, etc.

[0046] The server 103 may be a server that provides various services, such as a background management server that provides support for the device operated by the user using the terminal device 101. The background management server may analyze and process the received request and other data, and feed back the processing results to the terminal device.

[0047] Optionally, the server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides basic cloud computing services such as cloud services, cloud databases, cloud computing, cloud functions, cloud storage, network services, cloud communications, middleware services, domain name services, security services, CDN (Content Delivery Network), as well as big data and artificial intelligence platforms.

[0048] Those skilled in the art will know that Figure 1 The number of terminal devices, networks and servers in the embodiment is only for illustration, and any number of terminal devices, networks and servers may be provided according to actual needs, and the embodiments of the present disclosure do not limit this.

[0049] Under the above system architecture, an industrial text detection method is provided in an embodiment of the present disclosure, and the method can be executed by any electronic device with computing and processing capabilities.

[0050] In some embodiments, the industrial text detection method provided in the embodiments of the present disclosure can be executed by a terminal device of the above-mentioned system architecture; in other embodiments, the industrial text detection method provided in the embodiments of the present disclosure can be executed by a server in the above-mentioned system architecture; in other embodiments, the industrial text detection method provided in the embodiments of the present disclosure can be implemented by the terminal device and the server in the above-mentioned system architecture through interaction.

[0051] Figure 2 A flow chart of an industrial text detection method in an embodiment of the present disclosure is shown as follows: Figure 2 As shown, the industrial text detection method provided in the embodiment of the present disclosure includes the following steps S202 to S208.

[0052] S202, obtaining an image to be processed.

[0053] The embodiments of the present disclosure do not limit the image to be processed. Exemplarily, the image to be processed may be an image in an industrial scene, such as a tire image, a packaging bag image, an LCD screen image, and the like.

[0054] Furthermore, the image to be processed may contain text information, which may be in the form of plane characters or three-dimensional characters, such as engraved characters, embossed characters, and steel-printed characters.

[0055] S204, performing N-stage feature extraction on the image to be processed based on a feature pyramid network, and obtaining pyramid features of N sizes accordingly, wherein the feature pyramid network uses a separable network as a backbone network, and feature extraction at any stage obtains a pyramid feature of one size, and N is a positive integer.

[0056] It should be noted that the feature pyramid network can be expressed as FPN (Feature Pyramid Networks). FPN is a deep learning network structure that can fuse feature maps of different scales to improve multi-scale object detection performance.

[0057] In some exemplary embodiments, performing N stages of feature extraction on the image to be processed based on the feature pyramid network to obtain pyramid features of N sizes, including: performing N stages of feature extraction on the image to be processed based on the feature pyramid network to obtain pyramid features of N sizes, wherein the size of the pyramid feature obtained by the feature extraction of the nth stage is 1 / 2 of the original size n , n is a positive integer less than or equal to N.

[0058] In some possible implementations, the feature pyramid network may include convolutional layers of multiple stages, and the convolutional layers corresponding to each stage may output pyramid features of a certain size.

[0059] It should be noted that the original size may be the size of the image to be processed.

[0060] For example, taking the feature pyramid network performing 5 stages of feature extraction on the image to be processed as an example, the feature pyramid network may include convolutional layers corresponding to the 5 stages. After the image to be processed is input into the feature pyramid network, the step size of the convolutional layer corresponding to the first stage may be 1 / 2, and the size of the pyramid feature outputted by it is 1 / 2 of the original size.

[0061] Then, the step size of the convolution layer corresponding to the second stage can be 1 / 4, and the size of the pyramid feature output is 1 / 4 of the original size. The step size of the convolution layer corresponding to the third stage can be 1 / 8, and the size of the pyramid feature output is 1 / 8 of the original size. The step size of the convolution layer corresponding to the fourth stage can be 1 / 16, and the size of the pyramid feature output is 1 / 16 of the original size. The step size of the convolution layer corresponding to the fifth stage can be 1 / 32, and the size of the pyramid feature output is 1 / 32 of the original size.

[0062] It should be noted that the feature pyramid network can be a bottom-up path, in which the feature extraction at each stage can output a pyramid feature with a smaller size but higher level semantic information than the previous stage.

[0063] In some exemplary embodiments, the backbone network of the feature pyramid network includes a first separable network unit with a step size of 1 and a second separable network unit with a step size of 2.

[0064] Among them, the feature extraction at any stage in the feature pyramid network corresponds to a convolution layer, and any convolution layer includes the first separable network unit and the second separable network unit.

[0065] It should be noted that the first separable network unit and the second separable network unit can be two neural network architectures based on depthwise separable convolution as their core building blocks. The first separable network unit and the second separable network unit can also be called SeparableNet (separable network).

[0066] In an exemplary embodiment, since the feature pyramid network uses a separable network as the backbone network, it means that in the bottom-up path, all convolutional layers are replaced by depthwise separable convolutions. Therefore, the disclosed embodiment constructs the feature hierarchy of the feature pyramid network by using depthwise separable convolutions instead of traditional convolutional layers. Due to the high efficiency of depthwise separable convolutions, the disclosed embodiment can significantly reduce the computational complexity while maintaining similar feature extraction performance.

[0067] In an exemplary embodiment, a schematic diagram of a first detachable network unit may be as follows: Figure 3 shown.

[0068] In this Figure 3 In the example, the number of input channels can be equal to c, where c is a positive integer. Then, the input channels can be split into two parts, one of which enters the subsequent processing and the other part can be directly skipped.

[0069] like Figure 3 As shown, the input channel can be divided into two parts, c1 and c-c1. Among them, the c-c1 part can perform 1x1 convolution (Conv) and perform batch normalization (BN, Batch Normalization) and ReLU (Rectified Linear Unit, Rectified Linear Unit) activation function processing on the result after 1x1 convolution. Then, 3x3 depthwise convolution (DWConv, Depthwise Convolutio) can be performed and batch normalization can be performed on the depthwise convolution result. Then, 1x1 convolution can be performed and batch normalization and ReLU activation function processing can be performed on the result after 1x1 convolution, and finally the processing result of the c-c1 part can be obtained.

[0070] For example, the processing result of part c-c1 can be concatenated with part c1, and the number of channels after the concatenation is c. Finally, the concatenated channels can be shuffled to obtain the structure of the first separable network unit with a stride of 1. The channel shuffle can be used to ensure information mixing between channels.

[0071] In an exemplary embodiment, a schematic diagram of a second detachable network unit may be as follows: Figure 4 shown.

[0072] In this Figure 4 In the example, the number of input channels may be c, and the c channels are processed by two paths respectively.

[0073] Among them, in the first path, a 3x3 depth convolution with a stride of 2 can be performed on c channels, and the depth convolution results can be batch normalized. Then, a 1x1 convolution can be performed and the results after the 1x1 convolution can be batch normalized and processed with the ReLU activation function, and finally the processing result of the first path can be obtained.

[0074] In addition, in the second path, 1x1 convolution can be performed on c channels and the result of 1x1 convolution can be batch normalized and ReLU activation function processed. Then, 3x3 depth convolution with a step size of 2 can be performed, and the depth convolution result can be batch normalized. Then, 1x1 convolution can be performed and the result of 1x1 convolution can be batch normalized and ReLU activation function processed, and finally the processing result of the second path can be obtained. Finally, the processing results of the two paths can be combined and the channels can be rearranged.

[0075] In some possible implementations, the convolutional layer corresponding to any stage in the feature pyramid network includes the first separable network unit and / or the second separable network unit.

[0076] In some exemplary embodiments, when N=5, the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the second stage is 1, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the third stage is 3, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fourth stage is 7, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fifth stage is 1, and the number of repetitions of the second separable network unit is 1.

[0077] The embodiment of the present disclosure does not limit the number of repetitions of the first separable network unit and the second separable network unit in each level. Exemplarily, the number of repetitions can be determined based on experience or the effect of model training.

[0078] S206: Determine a target feature map according to the pyramid features of N sizes.

[0079] In some exemplary embodiments, a target feature map is determined based on pyramid features of N sizes, including: based on the order of each stage from back to front, performing the following operations stage by stage until a fused feature map corresponding to each stage is obtained: performing a 2-fold upsampling process on the fused feature map corresponding to the nth stage, and adding element by element to the pyramid feature obtained in the n-1th stage to obtain a fused feature map corresponding to the n-1th stage, wherein the fused feature map corresponding to the last stage is the pyramid feature obtained by feature extraction in the last stage; and determining the target feature map based on the fused feature maps corresponding to each stage.

[0080] Exemplarily, taking N equal to 5 as an example, at this time, the size of the pyramid feature output in the 5th stage is 1 / 32 of the original size. Afterwards, the pyramid feature output in the 5th stage can be upsampled by 2 times, and added element by element with the pyramid feature output in the 4th stage whose size is 1 / 16 of the original size, so as to obtain the fused feature map corresponding to the 4th stage. Then, the fused feature map corresponding to the 4th stage can be upsampled by 2 times, and added element by element with the pyramid feature output in the 3rd stage whose size is 1 / 8 of the original size, so as to obtain the fused feature map corresponding to the 3rd stage.

[0081] By analogy, the fused feature map corresponding to the third stage can be upsampled by a factor of 2 and added element by element to the pyramid feature output from the second stage whose size is 1 / 4 of the original size, so as to obtain the fused feature map corresponding to the second stage.

[0082] Exemplarily, after obtaining the fused feature map corresponding to the second stage, the fused feature map corresponding to the first stage can be obtained based on the above method, or only the fused feature maps corresponding to stages 2 to 5 can be obtained, that is, only the fused feature maps corresponding to the second to the last stage can be obtained. And according to the fused feature maps corresponding to the second to the last stage, the target feature map is determined.

[0083] It should be noted that since the pyramid feature size corresponding to the first stage is 1 / 2 of the original size, which is relatively large, if the memory occupied by the pyramid feature size corresponding to the first stage exceeds the preset threshold, only the fused feature map corresponding to the second to the last stage can be obtained. And according to the fused feature map corresponding to the second to the last stage, the target feature map is determined.

[0084] In some exemplary embodiments, a target feature map is determined based on the fused feature maps corresponding to each stage, including: performing upsampling and dimensionality reduction processing of corresponding multiples on the fused feature maps corresponding to each stage to obtain feature maps to be spliced; and splicing the feature maps to be spliced ​​to obtain the target feature map.

[0085] Exemplarily, taking N equal to 5 as an example, at this time, the fused feature map of the 5th stage can be subjected to 3x3 convolution and 8 times upsampling to obtain the corresponding feature map to be spliced. The fused feature map of the 4th stage can be subjected to 3x3 convolution and 4 times upsampling to obtain the corresponding feature map to be spliced. The fused feature map of the 3rd stage can be subjected to 3x3 convolution and 4 times upsampling to obtain the corresponding feature map to be spliced. The fused feature map of the 2nd stage can be subjected to 3x3 convolution to obtain the corresponding feature map to be spliced. Among them, the 3x3 convolution is used to reduce the dimension of the corresponding fused feature map.

[0086] In an exemplary embodiment, after obtaining each feature map to be spliced, splicing and 3x3 convolution can be performed to obtain a target feature map. In this embodiment, the size of the target feature map is 1 / 4 of the original size.

[0087] In an exemplary embodiment, a schematic diagram of a process of feature extraction and feature fusion can be as follows: Figure 5 shown.

[0088] In the feature extraction part, the image to be processed can be input as the input image to the convolution layer (Conv1), and the step size of Conv1 can be 1 / 2. The step size of the convolution layer corresponding to the second stage (Conv stage2) can be 1 / 4, and the step size of the convolution layer corresponding to the third stage (Conv stage3) can be 1 / 8. The step size of the convolution layer corresponding to the fourth stage (Conv stage4) can be 1 / 16. The step size of the convolution layer corresponding to the fifth stage (Conv stage5) can be 1 / 32.

[0089] Furthermore, Conv stage2, Conv stage3, Conv stage4, and Conv stage5 can each include a 1×1 convolution layer to output pyramid features M2, M3, M4, and M5 of corresponding sizes. Then, the pyramid features can be fused to obtain corresponding fused feature maps P2, P3, P4, and P5. The method of obtaining the fused feature maps corresponding to each stage is as described above and will not be repeated here.

[0090] Finally, the fused feature maps corresponding to each stage can be subjected to 3x3 convolution (3x3Conv) respectively, and up-sampled by corresponding multiples to obtain the feature maps to be spliced. After obtaining the feature maps to be spliced, they can be spliced ​​and subjected to 3x3 convolution to obtain the target feature map F.

[0091] S208: Determine an industrial text detection result according to the target feature map.

[0092] In some possible embodiments, a probability map P (Probability Map) and a threshold map T (Threshold Map) can be predicted according to the target feature map, and finally an approximate binary map can be obtained according to P and T. The industrial text detection result can be obtained according to the approximate binary map.

[0093] It should be noted that the probability map P is a feature map of the same size as the input image, where the value of each pixel represents the probability that the pixel belongs to the text area. These values ​​can be between [0, 1], 0 means that it does not belong to the text area at all, and 1 means that it belongs to the text area completely. The threshold map T can be a feature map of the same size as the input image, where the value of each pixel represents the threshold at that position. The threshold is used to decide whether to convert the corresponding probability value P to foreground or background.

[0094] Exemplarily, an approximate binary map can be obtained by comparing the probability map P and the threshold map T. For example, if for any pixel (x, y), if P(x, y)>T(x, y), then the pixel is marked as foreground (the value of the pixel in the approximate binary map is 1), indicating that the position belongs to the text area. If P(x, y)≤T(x, y), then the pixel is marked as background (the value of the pixel in the approximate binary map is 0), indicating that the position does not belong to the text area.

[0095] It should be noted that the embodiment of the present disclosure can detect texts with irregular shapes such as tilt and arc based on the adaptive binarization threshold method, thereby further improving the accuracy and reliability of text detection.

[0096] In an exemplary embodiment, the target feature map may be processed according to a Vatti clipping algorithm (a polygon clipping algorithm) to obtain a probability map P and a threshold map T.

[0097] The method provided by the embodiment of the present disclosure can perform N-stage feature extraction on the processed image based on the feature pyramid network with a separable network as the main network, thereby determining the industrial text detection result according to the target feature map determined by the pyramid feature. The embodiment of the present disclosure can perform feature extraction based on the feature pyramid network with a separable network as the main network, which can reduce the number of parameters and the amount of calculation, thereby improving efficiency and ensuring the accuracy and reliability of feature extraction, thereby effectively improving the accuracy of the detection result and improving the detection efficiency.

[0098] Exemplarily, the method provided by the embodiments of the present disclosure can be experimentally compared with DBNet (Differentiable Binarization Network) with ResNet18 (residual network 18) and ResNet50 (residual network 50) as the backbone network and DBNet with MobileNet v1 (mobile network version 1), MobileNet v2 (mobile network version 1), and ShuffleNet v1 (shuffle network version 1) as the backbone network on industrial text datasets.

[0099] Among them, a structural diagram of a DBNet network provided by an embodiment of the present disclosure can be as follows: Figure 6 As shown. Figure 6 In the example, the backbone network is used to extract text features, and then the pyramid features are obtained using FPN. The pyramid features are then upsampled by different multiples to obtain the feature maps to be spliced, where the size of each feature map to be spliced ​​is 1 / 4 of the original size. Then, the feature maps to be spliced ​​can be spliced ​​and dimensionality reduced based on 3×3 convolution to obtain a target feature map with a size of 1 / 4 of the original size.

[0100] Then, the probability map and threshold map can be predicted based on the target feature map, and finally an approximate binary map can be obtained based on the probability map and the threshold map, and finally a text prediction box can be obtained.

[0101] In a possible implementation, the images to be processed can be preprocessed during the training process, and their sizes can be uniformly scaled to 640*480, and 20% of the images in the training set can be selected as the validation set. The weights of the backbone network are randomly initialized using the Gaussian distribution function, and the RMSProp (Root Mean Square Propagation Optimizer) optimizer is selected for network optimization. The batch reading size is set to 64, and the initial value of the learning rate is lr init For 10 -3 , the learning rate lr will decay as the number of iterations increases. The calculation formula of lr can be shown as the following formula (1).

[0102]

[0103] Wherein, decay can be used to represent the decay value. The embodiment of the present disclosure does not limit the value of the decay value. The decay value can be set to 5×10 -4 . step can be used to indicate the number of parameter updates for a batch.

[0104] For example, a schematic diagram of an lr curve in a network training process provided by an embodiment of the present disclosure can be as follows: Figure 7 As shown. Figure 7 In the example, since the parameters are updated faster at the beginning, the change of lr is larger, and its decay rate decreases faster. After that, the loss decreases more slowly, the change of lr is smaller, and the decay rate gradually slows down.

[0105] In an exemplary embodiment, the method provided by the embodiment of the present disclosure is experimentally compared with DBNet with ResNet18 and ResNet50 as the backbone network and DBNet with MobileNet v1, MobileNet v2 and ShuffleNet v1 as the backbone network on an industrial text dataset, and the results can be shown in Table 1 below.

[0106] Table 1

[0107] Backbone network Precision(%) Recall(%) F-measure (%) FPS ResNet18 86.8 78.4 82.4 32 ResNet50 88.3 81.7 84.9 17 MobileNet v1 85.2 77.6 81.2 34.2 MobileNet v2 86.7 78.3 82.3 36 ShuffleNet v1 86.5 78.2 82.1 39.8

[0108] Among them, Precision, Recall, F-measure (F1 score), and FPS (frames per second) are four evaluation indicators. It can be seen that when these algorithms are trained or tested on the same training platform or test platform. When the backbone network is ResNet50, the precision and F1 score are the highest, but the detection speed (FPS) is the slowest. If ResNet18 is used, although the precision is relatively reduced by 1.5%, the corresponding F1 score is reduced by 2.5%, but the FPS is increased by nearly 1 times, which is consistent with the detection results of the original DBNet network model. If the backbone network is replaced with MobileNet, compared with ResNet18, the F1 scores and FPS of MobileNet v1 and MobileNet v2 change little. Compared with ResNet50, the FPS of MobileNet v1 and MobileNet v2 are greatly improved, but the F1 scores are reduced by 3.7% and 2.6% respectively. Compared with ResNet18, and SeparableNet, the backbone network is replaced by ShuffleNet v1, and the precision and F1 score are almost unchanged, but the FPS is greatly improved. In particular, compared with ResNet50, which has the highest precision and F1 score, the FPS of the detection with the separable network as the backbone network provided by the embodiment of the present disclosure is about 3 times that of the original. As can be seen from the above, the method provided by the embodiment of the present disclosure effectively improves the performance of industrial text detection, and ensures the accuracy of detection while improving the detection speed, which meets the requirements of text detection in industrial scenarios.

[0109] For example, a schematic diagram of an industrial text detection effect provided by an embodiment of the present disclosure can be as follows: Figure 8 shown.

[0110] In this Figure 8 In FIG. 4 , four detected text prediction boxes are shown. It can be seen that the industrial text detection method provided by the embodiment of the present disclosure can effectively identify industrial texts in various scenes.

[0111] It should be noted that the embodiments of the present disclosure can be based on the architecture of the DBNet network, and use an adaptive binarization threshold method to detect irregularly shaped texts such as tilt and arc. A new lightweight network SeparableNet is used to extract text features. Deep separable convolution and shortcut connections are used to reduce the amount of parameters and calculations, reduce running memory, and improve the feature extraction speed. It can also ensure the accuracy and reliability of text detection and has strong robustness.

[0112] It should be noted that the acquisition, storage, use, and processing of data in the technical solution disclosed in this disclosure are in compliance with the relevant provisions of national laws and regulations, and various types of data such as personal identity data, operation data, behavioral data, etc. related to individuals, customers, and groups obtained in the embodiments of this disclosure have all been authorized.

[0113] Based on the same inventive concept, an industrial text detection device is also provided in the embodiment of the present disclosure, as described in the following embodiment. Since the principle of solving the problem in the embodiment of the device is similar to that in the above method embodiment, the implementation of the embodiment of the device can refer to the implementation of the above method embodiment, and the repeated parts will not be repeated.

[0114] Fig. 9 A schematic diagram of an industrial text detection device in an embodiment of the present disclosure is shown. Fig. 9 As shown, the device comprises:

[0115] The image to be processed acquisition module 901 is used to acquire the image to be processed;

[0116] The pyramid feature determination module 902 is used to perform N-stage feature extraction on the image to be processed based on the feature pyramid network, and obtain pyramid features of N sizes accordingly, wherein the feature pyramid network uses a separable network as a backbone network, and feature extraction at any stage obtains a pyramid feature of one size, and N is a positive integer;

[0117] A target feature map determining module 903 is used to determine a target feature map according to pyramid features of N sizes;

[0118] The industrial text detection result determination module 904 is used to determine the industrial text detection result according to the target feature map.

[0119] In some exemplary embodiments, the pyramid feature determination module 902 is used to perform N stages of feature extraction on the image to be processed based on the feature pyramid network, and obtain pyramid features of N sizes, wherein the size of the pyramid feature obtained by the feature extraction of the nth stage is 1 / 2 of the original size. n , n is a positive integer less than or equal to N.

[0120] In some exemplary embodiments, the target feature map determination module 903 is used to perform the following operations stage by stage based on the order of each stage from back to front until a fused feature map corresponding to each stage is obtained: the fused feature map corresponding to the nth stage is upsampled by a factor of 2, and is element-by-element added to the pyramid feature obtained in the n-1th stage to obtain a fused feature map corresponding to the n-1th stage, wherein the fused feature map corresponding to the last stage is the pyramid feature obtained by feature extraction in the last stage; and the target feature map is determined based on the fused feature maps corresponding to each stage.

[0121] In some exemplary embodiments, the target feature map determination module 903 is used to perform upsampling processing and dimensionality reduction processing of corresponding multiples according to the fusion feature maps corresponding to each stage to obtain each feature map to be spliced; and splice each feature map to be spliced ​​to obtain the target feature map.

[0122] In some exemplary embodiments, the backbone network of the feature pyramid network includes a first separable network unit with a step size of 1 and a second separable network unit with a step size of 2; wherein, feature extraction at any stage in the feature pyramid network corresponds to a convolutional layer, and any convolutional layer contains the first separable network unit and the second separable network unit.

[0123] In some exemplary embodiments, when N=5, the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the second stage is 1, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the third stage is 3, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fourth stage is 7, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fifth stage is 1, and the number of repetitions of the second separable network unit is 1.

[0124] The device provided by the embodiment of the present disclosure can perform N-stage feature extraction on the processed image based on the feature pyramid network with a separable network as the main network, thereby determining the industrial text detection result according to the target feature map determined by the pyramid feature. The embodiment of the present disclosure can perform feature extraction based on the feature pyramid network with a separable network as the main network, which can reduce the number of parameters and the amount of calculation, thereby improving efficiency and ensuring the accuracy and reliability of feature extraction, thereby effectively improving the accuracy of the detection result and improving the detection efficiency.

[0125] It should be noted that the above-mentioned to-be-processed image acquisition module 901, pyramid feature determination module 902, target feature map determination module 903, and industrial text detection result determination module 904 correspond to S202 to S208 in the method embodiment, and the examples and application scenarios implemented by the above-mentioned modules and corresponding steps are the same, but are not limited to the contents disclosed in the above-mentioned method embodiment. It should be noted that the above-mentioned modules as part of the device can be executed in a computer system such as a set of computer executable instructions.

[0126] Those skilled in the art will appreciate that various aspects of the present disclosure may be implemented as systems, methods or program products. Therefore, various aspects of the present disclosure may be specifically implemented in the following forms, namely: complete hardware implementation, complete software implementation (including firmware, microcode, etc.), or a combination of hardware and software, which may be collectively referred to herein as "circuits", "modules" or "systems".

[0127] The embodiment of the present disclosure provides an electronic device, which exemplarily includes: a processor and a memory. The memory can be used to store executable instructions of the processor. The processor is configured to perform the industrial text detection method provided by the embodiment of the present disclosure by executing the above executable instructions.

[0128] Refer to the following Fig.10 The electronic device 1000 according to this embodiment of the present disclosure is described. Fig.10 The electronic device 1000 shown is merely an example and should not bring any limitation to the functions and scope of use of the embodiments of the present disclosure.

[0129] like Fig.10 As shown, the electronic device 1000 is in the form of a general computing device. The components of the electronic device 1000 may include but are not limited to: at least one processing unit 1010, at least one storage unit 1020, and a bus 1030 connecting different system components (including the storage unit 1020 and the processing unit 1010).

[0130] The storage unit stores a program code, which can be executed by the processing unit 1010, so that the processing unit 1010 performs the steps of various exemplary embodiments of the present disclosure described in the above “Exemplary Method” section of this specification. For example, the processing unit 1010 can perform the following steps of the above method embodiment:

[0131] Acquire an image to be processed; perform N-stage feature extraction on the image to be processed based on a feature pyramid network, and obtain pyramid features of N sizes accordingly, wherein the feature pyramid network uses a separable network as a main network, and feature extraction at any stage obtains a pyramid feature of a size, where N is a positive integer; determine a target feature map based on the pyramid features of N sizes; and determine an industrial text detection result based on the target feature map.

[0132] The storage unit 1020 may include a readable medium in the form of a volatile storage unit, such as a random access storage unit (RAM) 10201 and / or a cache storage unit 10202 , and may further include a read-only storage unit (ROM) 10203 .

[0133] The storage unit 1020 may also include a program / utility 10204 having a set (at least one) of program modules 10205, such program modules 10205 including but not limited to: an operating system, one or more application programs, other program modules, and program data, each of which or some combination may include an implementation of a network environment.

[0134] Bus 1030 may represent one or more of several types of bus structures, including a memory unit bus or memory unit controller, a peripheral bus, an accelerated graphics port, a processing unit, or a local bus using any of a variety of bus architectures.

[0135] The electronic device 1000 may also communicate with one or more external devices 1040 (e.g., keyboards, pointing devices, Bluetooth devices, etc.), may also communicate with one or more devices that enable a user to interact with the electronic device 1000, and / or communicate with any device that enables the electronic device 1000 to communicate with one or more other computing devices (e.g., routers, modems, etc.). Such communication may be performed via an input / output (I / O) interface 1050. Furthermore, the electronic device 1000 may also communicate with one or more networks (e.g., local area networks (LANs), wide area networks (WANs), and / or public networks, such as the Internet) via a network adapter 1060. As shown, the network adapter 1060 communicates with other modules of the electronic device 1000 via a bus 1030. It should be understood that, although not shown in the figure, other hardware and / or software modules may be used in conjunction with the electronic device 1000, including but not limited to: microcode, device drivers, redundant processing units, external disk drive arrays, RAID systems, tape drives, and data backup storage systems.

[0136] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a terminal device, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0137] In particular, according to an embodiment of the present disclosure, the process described above with reference to the flowchart may be implemented as a computer program product, which includes: a computer program, which implements the above industrial text detection method when executed by a processor.

[0138] In an exemplary embodiment of the present disclosure, a computer-readable storage medium is also provided, on which a computer program is stored, and when the computer program is executed by a processor, the industrial text detection method provided in the embodiment of the present disclosure can be implemented. The computer-readable storage medium can be a readable signal medium or a readable storage medium.

[0139] Fig.11 A schematic diagram of a computer-readable storage medium in an embodiment of the present disclosure is shown. Fig.11 As shown, a program product capable of implementing the above method of the present disclosure is stored on the computer-readable storage medium 1100. In some possible implementations, various aspects of the present disclosure may also be implemented in the form of a program product, which includes a program code. When the program product is run on a terminal device, the program code is used to enable the terminal device to execute the steps according to various exemplary implementations of the present disclosure described in the above “Exemplary Method” section of this specification.

[0140] More specific examples of computer-readable storage media in the present disclosure may include, but are not limited to, an electrical connection having one or more conductors, a portable computer disk, a hard disk, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), optical fiber, a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination of the foregoing.

[0141] In the present disclosure, a computer readable storage medium may include a data signal propagated in baseband or as part of a carrier wave, wherein a readable program code is carried. Such propagated data signals may take a variety of forms, including but not limited to electromagnetic signals, optical signals, or any suitable combination of the above. A readable signal medium may also be any readable medium other than a readable storage medium, which may send, propagate, or transmit a program for use by or in conjunction with an instruction execution system, apparatus, or device.

[0142] Alternatively, the program code contained on the computer-readable storage medium may be transmitted using any appropriate medium, including but not limited to wireless, wired, optical cable, RF, etc., or any suitable combination of the foregoing.

[0143] In a specific implementation, the program code for performing the operation of the present disclosure may be written in any combination of one or more programming languages, including object-oriented programming languages ​​such as Java, C++, etc., and conventional procedural programming languages ​​such as "C" or similar programming languages. The program code may be executed entirely on the user computing device, partially on the user device, as a separate software package, partially on the user computing device and partially on a remote computing device, or entirely on a remote computing device or server. In the case of a remote computing device, the remote computing device may be connected to the user computing device through any type of network, including a local area network (LAN) or a wide area network (WAN), or may be connected to an external computing device (e.g., using an Internet service provider to connect through the Internet).

[0144] It should be noted that, although several modules or units of the device for action execution are mentioned in the above detailed description, this division is not mandatory. In fact, according to the embodiments of the present disclosure, the features and functions of two or more modules or units described above can be embodied in one module or unit. On the contrary, the features and functions of one module or unit described above can be further divided into multiple modules or units to be embodied.

[0145] In addition, although the steps of the method in the present disclosure are described in a specific order in the drawings, this does not require or imply that the steps must be performed in this specific order, or that all the steps shown must be performed to achieve the desired results. Additionally or alternatively, some steps may be omitted, multiple steps may be combined into one step, and / or one step may be decomposed into multiple steps, etc.

[0146] Through the description of the above implementation, it is easy for those skilled in the art to understand that the example implementation described here can be implemented by software, or by software combined with necessary hardware. Therefore, the technical solution according to the implementation of the present disclosure can be embodied in the form of a software product, which can be stored in a non-volatile storage medium (which can be a CD-ROM, a USB flash drive, a mobile hard disk, etc.) or on a network, including several instructions to enable a computing device (which can be a personal computer, a server, a mobile terminal, or a network device, etc.) to execute the method according to the implementation of the present disclosure.

[0147] Those skilled in the art will readily appreciate other embodiments of the present disclosure after considering the specification and practicing the invention disclosed herein. The present disclosure is intended to cover any variations, uses or adaptations of the present disclosure, which follow the general principles of the present disclosure and include common knowledge or customary technical means in the art that are not disclosed in the present disclosure. The description and examples are to be regarded as exemplary only, and the true scope of the present disclosure is indicated by the appended claims.

Claims

1. An industrial text detection method, characterized in that: include: Get the image to be processed; Based on the feature pyramid network, N stages of feature extraction are performed on the image to be processed, and pyramid features of N sizes are obtained accordingly, wherein the feature pyramid network uses a separable network as a main network, and feature extraction at any stage obtains a pyramid feature of one size, and N is a positive integer; Determine the target feature map based on the pyramid features of N sizes; An industrial text detection result is determined according to the target feature map.

2. The industrial text detection method according to claim 1, characterized in that: The feature extraction of the image to be processed in N stages based on the feature pyramid network, and correspondingly obtaining pyramid features of N sizes, includes: Based on the feature pyramid network, N stages of feature extraction are performed on the image to be processed, and pyramid features of N sizes are obtained accordingly, wherein the size of the pyramid feature obtained by the feature extraction of the nth stage is 1 / 2 of the original size n , n is a positive integer less than or equal to N.

3. The industrial text detection method according to claim 1 or 2, characterized in that: The step of determining a target feature map according to pyramid features of N sizes includes: Based on the order of each stage from back to front, the following operations are performed stage by stage until the fused feature map corresponding to each stage is obtained: The fused feature map corresponding to the nth stage is upsampled by a factor of 2, and is added element by element to the pyramid feature obtained in the n-1th stage to obtain the fused feature map corresponding to the n-1th stage, wherein the fused feature map corresponding to the last stage is the pyramid feature obtained by feature extraction in the last stage; The target feature map is determined according to the fused feature maps corresponding to each stage.

4. The industrial text detection method according to claim 3, characterized in that: Determining the target feature map according to the fused feature maps corresponding to each stage includes: According to the fusion feature maps corresponding to each stage, upsampling and dimensionality reduction processing of corresponding multiples are performed to obtain each feature map to be spliced; Each feature map to be spliced ​​is spliced ​​to obtain the target feature map.

5. The industrial text detection method according to claim 1, characterized in that: The backbone network of the feature pyramid network includes a first separable network unit with a step size of 1 and a second separable network unit with a step size of 2; Among them, the feature extraction at any stage in the feature pyramid network corresponds to a convolution layer, and any convolution layer includes the first separable network unit and the second separable network unit.

6. The industrial text detection method according to claim 5, characterized in that: When N=5, the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the second stage is 1, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the third stage is 3, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fourth stage is 7, and the number of repetitions of the second separable network unit is 1; the number of repetitions of the first separable network unit contained in the convolutional layer corresponding to the fifth stage is 1, and the number of repetitions of the second separable network unit is 1.

7. An industrial text detection device, characterized in that: include: The module for acquiring the image to be processed is used to acquire the image to be processed; A pyramid feature determination module is used to perform N-stage feature extraction on the image to be processed based on a feature pyramid network, and obtain pyramid features of N sizes accordingly, wherein the feature pyramid network uses a separable network as a backbone network, and feature extraction at any stage obtains a pyramid feature of one size, and N is a positive integer; A target feature map determination module is used to determine a target feature map based on pyramid features of N sizes; The industrial text detection result determination module is used to determine the industrial text detection result according to the target feature map.

8. An electronic device, characterized in that: include: processor; as well as A memory, configured to store executable instructions of the processor; The processor is configured to execute the industrial text detection method according to any one of claims 1 to 6 by executing the executable instructions.

9. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the industrial text detection method according to any one of claims 1 to 6 is implemented.

10. A computer program product, characterized in that The computer program product includes computer instructions, which are stored in a computer-readable storage medium. A processor of an electronic device reads the computer instructions from the computer-readable storage medium, and the processor executes the computer instructions, so that the electronic device executes the industrial text detection method as described in any one of claims 1 to 6.

Images