Cable label sample set processing method and cable label identification method

Abstract

The invention relates to a cable label sample set processing method and a cable label identification method. The cable label sample set processing method comprises the following steps: acquiring a cable label image; pre-processing the cable label image, wherein the pre-processing comprises inclined image correction; screening out at least one target data enhancement strategy from a plurality of preset data enhancement strategies; and based on the screened at least one target data enhancement strategy, performing data enhancement processing on the preprocessed cable tag image to obtain a cable tag sample set. By adopting the method, the cable label identification accuracy can be improved.

Description

Technical Field

[0001] The present application relates to the field of cable label recognition, and particularly to a method and device for processing a cable label sample set, a computer device, a computer-readable storage medium, and a computer program product, as well as a method and device for cable label recognition, a computer device, a computer-readable storage medium, and a computer program product. Background Art

[0002] A cable label, that is, a sign for communication wires such as cables, wires, optical fibers, and network cables, is mainly used for line management in the construction process and daily maintenance of telecommunications cabling, circuit decoration, communication machine rooms, engineering cabling, security, etc. Cable label recognition has a wide range of applications in the industrial field, such as equipment management and maintenance, and fault line location in the power system.

[0003] Traditional cable label reading is through text detection. According to the set feature extraction rules, background separation is completed through features such as texture, color, and boundary in a specified scenario, and text detection of the image is performed. However, there is a problem of low detection accuracy for cable label recognition in complex scenarios. Summary of the Invention

[0004] Based on this, in view of the above technical problems, it is necessary to provide a method and device for processing a cable label sample set, a computer device, a computer-readable storage medium, and a computer program product that are beneficial to improving the accuracy of cable label recognition, as well as a method and device for cable label recognition, a computer device, a computer-readable storage medium, and a computer program product.

[0005] In a first aspect, the present application provides a method for processing a cable label sample set, including:

[0006] Obtaining a cable label image;

[0007] Preprocessing the cable label image, where the preprocessing includes correcting the inclined image;

[0008] Selecting at least one target data augmentation strategy from multiple preset data augmentation strategies;

[0009] Based on the at least one selected target data augmentation strategy, performing data augmentation processing on the preprocessed cable label image to obtain a cable label sample set.

[0010] In one embodiment, the preset data augmentation strategies at least include a noise addition strategy, a noise elimination strategy, and an image cropping strategy.

[0011] In one embodiment, when the number of cable label images is multiple, the method further includes:

[0012] When the target data augmentation strategy is the noise addition strategy, for each cable label image, determine the noise value range based on the maximum pixel value of the cable label image;

[0013] Based on the noise value range, add Gaussian noise to the cable label image to obtain the cable label image after data augmentation.

[0014] In one embodiment, the method further includes:

[0015] When the target data augmentation strategy is the image cropping strategy, obtain the minimum horizontal cropping distance and the minimum vertical cropping distance;

[0016] Based on the minimum horizontal cropping distance and the minimum vertical cropping distance, crop the cable label image to obtain the cable label image after data augmentation.

[0017] In one embodiment, the method further includes:

[0018] When the target data augmentation strategy is the noise elimination strategy, perform Gaussian filtering on the cable label image to obtain the cable label image after data augmentation.

[0019] In one embodiment, screening at least one target data augmentation strategy from multiple preset target data augmentation strategies includes:

[0020] Randomly screen at least two target data augmentation strategies from multiple preset data augmentation strategies;

[0021] Based on the at least one screened target data augmentation strategy, perform data augmentation processing on the cable label image to obtain a cable label sample set, including:

[0022] Based on the at least two screened target data augmentation strategies, iteratively perform data augmentation processing on the cable label image to obtain the first cable label sample set.

[0023] In one embodiment, screening at least one target data augmentation strategy from multiple preset target data augmentation strategies further includes:

[0024] Randomly screen one target data augmentation strategy from multiple preset data augmentation strategies;

[0025] Based on the at least one screened target data augmentation strategy, perform data augmentation processing on the cable label image to obtain a cable label sample set, further including:

[0026] Based on the target data augmentation strategy, perform data augmentation processing on the cable label image to obtain the second cable label sample set;

[0027] Combine the first cable label sample set and the second cable label sample set to obtain a cable label sample set.

[0028] In one embodiment, the preprocessing further includes at least one of grayscale conversion and binarization.

[0029] In a second aspect, the present application further provides a method for identifying cable labels, including:

[0030] Obtain a cable label image to be detected;

[0031] Taking the cable label image to be detected as input, call a trained cable label text detection model to obtain a text detection result, and the cable label text detection model is trained based on the cable label sample set processing method of any one of the above;

[0032] Taking the text detection result as input, call a trained cable label text recognition model to obtain a text recognition result, and the cable label text recognition model is trained based on a set of historical cable label images carrying text labels.

[0033] In a third aspect, the present application further provides a device for processing a cable label sample set, including:

[0034] An image acquisition module, configured to acquire a cable label image;

[0035] An image preprocessing module, configured to preprocess the cable label image, and the preprocessing includes skew image correction;

[0036] A strategy screening module, configured to screen at least one target data augmentation strategy from a plurality of preset data augmentation strategies;

[0037] An augmentation processing module, configured to perform data augmentation processing on the preprocessed cable label image based on the at least one screened target data augmentation strategy to obtain a cable label sample set.

[0038] In a fourth aspect, the present application further provides a device for identifying cable labels, including:

[0039] A cable label image acquisition module, configured to acquire a cable label image to be detected;

[0040] A cable label text detection module, configured to take the cable label image to be detected as input, call a trained cable label text detection model to obtain a text detection result, and the cable label text detection model is trained based on the cable label sample set processing method of any one of the above;

[0041] The cable label text recognition module is used to take the text detection result as input, call the trained cable label text recognition model, and obtain the text recognition result. The cable label text recognition model is trained based on the cable label historical image set carrying text labels.

[0042] In a fifth aspect, the present application also provides a computer device, including a memory and a processor. The memory stores a computer program, and when the processor executes the computer program, it implements the steps in any one of the above-mentioned cable label sample set processing method embodiments and the steps in a cable label recognition method embodiment.

[0043] In a sixth aspect, the present application also provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps in any one of the above-mentioned cable label sample set processing method embodiments and the steps in a cable label recognition method embodiment.

[0044] In a seventh aspect, the present application also provides a computer program product, including a computer program. When the computer program is executed by a processor, it implements the steps in any one of the above-mentioned cable label sample set processing method embodiments and the steps in a cable label recognition method embodiment.

[0045] For the above-mentioned cable label sample set processing method, device, computer device, computer-readable storage medium, and computer program product, considering that in the actual scenario, the cable label may be in an inclined and bent state, resulting in low robustness of the model trained based on the cable label image. In view of this feature, preprocessing of inclined image correction is performed on the cable label image to improve the quality of the training sample set. At the same time, considering that during the training process of the cable label recognition model, the cable label images in the training sample set may be affected by various factors such as environmental illumination, debris occlusion, label position offset, and multi-label stacking, resulting in low robustness of the model trained based on the cable label image. Therefore, before training the cable label text detection model, according to the characteristics of the cable label, an image information enhancement mechanism is introduced, multiple data enhancement strategies are preset in advance, at least one target data enhancement strategy is selected from the multiple preset data enhancement strategies, and based on the selected at least one target data enhancement strategy, the cable label image is enhanced, and at the same time, the training sample set is expanded, which is beneficial to improving the quality of the cable image sample set, highlighting the characteristics of the cable label, reducing the background interference in the image, and thus, training the model based on the cable label sample set is beneficial to improving the model's recognition ability of the cable label in the cable label image, and further, is beneficial to improving the adaptability to complex recognition environments and improving the accuracy of cable label recognition.

[0046] The above-mentioned cable label recognition method, device, computer equipment, computer-readable storage medium, and computer program product. First, a cable label text detection model is trained in advance based on the above-mentioned cable label sample set processing method. The obtained cable label image to be detected is input into the cable label text detection model to obtain the text detection result of the cable label. In this way, training the cable label text detection model based on the sample set after data augmentation and expansion is beneficial to improving the adaptability of the model to the cable label text detection in complex detection environments and enhancing the accuracy of cable label text detection. Secondly, the accurate text detection result is input into the trained cable label text recognition model to obtain the text recognition result, which improves the accuracy of cable label text recognition and helps to enhance the accuracy of cable management and maintenance. Description of the Drawings

[0047] To more clearly illustrate the technical solutions in the embodiments of the present application or related technologies, the following will briefly introduce the drawings required for the description of the embodiments of the present application or related technologies. Obviously, the following drawings are only some embodiments of the present application. For those of ordinary skill in the art, other related drawings can be obtained based on these drawings without creative efforts.

[0048] Figure 1 It is an application environment diagram of the cable label sample set processing method in an embodiment;

[0049] Figure 2 It is a flowchart of the cable label sample set processing method in an embodiment;

[0050] Figure 3 It is a flowchart of the cable label sample set processing method in another embodiment;

[0051] Figure 4 It is a flowchart of the cable label recognition method in an embodiment;

[0052] Figure 5 It is a flowchart of the cable label text detection and recognition method in an embodiment;

[0053] Figure 6 It is a structural block diagram of the cable label sample set processing device in an embodiment;

[0054] Figure 7 It is a structural block diagram of the cable label recognition processing device in an embodiment;

[0055] Figure 8 It is an internal structure diagram of the computer equipment in an embodiment. Detailed Embodiments

[0056] To make the objectives, technical solutions, and advantages of this application more clear and understandable, the following further details this application in conjunction with the accompanying drawings and embodiments. It should be understood that the specific embodiments described herein are merely for explaining this application and are not used to limit this application.

[0057] The method for processing a cable label sample set provided by an embodiment of this application can be applied to an application environment as Figure 1 shown. Among them, the terminal 102 communicates with the server 104 through a network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or can be placed in the cloud or on other network servers.

[0058] Specifically, it can be that an operator uploads the collected cable label image to the server 104 through the terminal 102, and then sends a cable label sample set processing message to the server 104 through the terminal 102. The server 104 obtains the cable label image data. Secondly, at least one target data augmentation strategy is selected from multiple preset data augmentation strategies. Finally, data augmentation processing is performed on the cable label image based on the at least one selected target data augmentation strategy to obtain a cable label sample set.

[0059] Among them, the terminal 102 can be, but is not limited to, various personal computers, laptop computers, smartphones, tablet computers, Internet of Things devices, and portable wearable devices. The Internet of Things devices can be smart speakers, smart TVs, smart air conditioners, smart in-vehicle devices, projection devices, etc. The portable wearable devices can be smart watches, smart bracelets, head-mounted devices, etc. The head-mounted device can be a virtual reality (VR) device, an augmented reality (AR) device, smart glasses, etc. The server 104 can be an independent physical server, or a server cluster or distributed system composed of multiple physical servers, or a cloud server providing cloud computing services.

[0060] In an exemplary embodiment, as Figure 2 shown, a method for processing a cable label sample set is provided. Taking the method applied to the Figure 1 server 104 in it as an example for description, it includes the following steps (hereinafter simply referred to as S) S100 to S400. Among them:

[0061] S100, obtain a cable label image.

[0062] Among them, the cable label image is an image including a cable label. Depending on the cable usage, there are various different types of cable labels, such as heat-shrinkable tube cable identification, flame-retardant number tube, wound self-adhesive film label, flag / knife / F / T-shaped cable label, and 4-hole identification plate cable label, etc. The cable label content can include cable number, name, voltage, current, interface attributes, or usage, etc. It can be understood that the cable label content is determined according to the cable usage, label type, and identification purpose. The application scenarios of cable labels can include construction processes such as telecommunications cabling, circuit decoration, communication machine rooms, engineering cabling, security, etc., and daily line management.

[0063] In some embodiments, the cable label image can be obtained by an operator collecting images of different types of cable labels in the actual application scenario and uploading the collected cable label images to the server. It can also be obtained by a monitoring device or a patrol robot collecting images containing cable labels in the actual application scenario and storing them in the local database. It can also be obtained by other means such as obtaining a cable label image dataset through an open-source dataset platform. The present application does not limit the acquisition method of the cable label image.

[0064] In actual application, taking the communication machine room as an example of the application scenario, the server can obtain the cable label image through the cable historical maintenance records and patrol images stored in the local database, and perform cropping and size adjustment on the cable label image to make the image size uniform. Then, data annotation is performed on the cable label image. It can be manually annotated on the cable label image through an image dataset annotation tool. During the annotation process, mark the position of the cable label text, and use a rectangular box to mark the text area. When selecting the box, mark the upper left corner coordinates and the lower right corner coordinates , and annotate the corresponding category label for each box. The category label can include whether it is cable label text.

[0065] S200, preprocess the cable label image, and the preprocessing includes skew image correction.

[0066] In actual application, the processor can perform skew correction based on transformation algorithms such as Rando transformation or Hough transformation. Taking the server using Hough transformation to perform skew correction on the image as an example, specifically, it can use an edge detection algorithm (such as Canny edge detection) to extract the edges of the cable label to obtain an edge detection map, perform Hough transformation on the edge detection map to detect straight lines, obtain the transformation result, select the longest straight line from the transformation result, calculate the skew angle according to the slope of the longest straight line, and perform rotation correction on the image according to the skew angle.

[0067] S300. Select at least one target data augmentation strategy from multiple preset data augmentation strategies.

[0068] Among them, the preset data augmentation strategies include, but are not limited to, contrast enhancement, brightness adjustment, histogram equalization, geometric transformation, color enhancement, filtering operation, noise suppression, etc. Among them, contrast enhancement can be achieved by changing the dynamic range of image pixels to enhance the light and dark contrast of the image; brightness adjustment can be to increase or decrease the overall brightness level of the image; histogram equalization can be to apply global histogram equalization to redistribute the gray-level distribution of the image and enhance the overall contrast. Geometric transformation can include flipping, rotation, cropping, deformation, and scaling, etc. Among them, rotation can be to randomly rotate the image to simulate cable labels at different shooting angles; flipping can be to horizontally or vertically flip the image to increase the diversity of the image; scaling and cropping can be to randomly scale and crop the image to simulate the distance ratios of different cable labels. The target data augmentation strategy is a strategy for performing data augmentation processing on cable images.

[0069] In other embodiments, the preset data augmentation strategies can also be set according to the characteristics of the cable label image. For example, considering the situation where the label text on the cable label image is worn and blurred, the set data augmentation strategy also includes text blurring. Exemplarily, text blurring can be to process the image through a blur filter to simulate the situation of blurred text. Considering that there are a large number of cables in the actual application scenario and there is a situation where the cable label text is blocked, the set data augmentation strategy also includes character occlusion. Exemplarily, character occlusion can be to randomly occlude some characters in the image.

[0070] In actual applications, it can be to select at least one target data augmentation strategy from multiple preset data augmentation strategies according to the characteristics of the cable label image and the actual application scenario. Specifically, continuing with the above example, taking the cable label image in a communication machine room as an example, considering that the image may have uneven lighting conditions, a large number of cables, compact wiring, complex background, blurred text, and character occlusion, etc., the operator pre-selects candidate data augmentation strategies from the preset data augmentation strategies according to the characteristics of the cable label image and the actual application scenario. For example, the candidate data augmentation strategies can include contrast enhancement, brightness adjustment, text blurring, and character occlusion. The processor randomly selects at least one target data augmentation strategy from the candidate data augmentation strategies.

[0071] In other embodiments, the preset data augmentation strategies can be pre-classified into two categories, one is the basic data augmentation strategy, and the other is the special data augmentation strategy. Among them, the special data augmentation strategy represents a targeted data augmentation strategy set according to the characteristics of the cable label image, such as text blurring, character occlusion, and geometric transformation. According to the actual model training requirements, the server can randomly select at least one target data augmentation strategy from the basic data augmentation strategies, and then randomly select at least one target data augmentation strategy from the special data augmentation strategies.

[0072] S400. Based on the at least one target data augmentation strategy selected, perform data augmentation processing on the preprocessed cable label image to obtain a cable label sample set.

[0073] Among them, the cable label sample set includes the cable label image before data augmentation processing and is used for model training.

[0074] In practical applications, for each selected target data augmentation strategy, the processor can use the target data augmentation strategy to perform multiple data augmentation processes on the preprocessed cable label image to obtain the cable label image after data augmentation processing.

[0075] In the above cable label sample set processing method, considering that in the actual scenario, the cable label may be in a tilted and bent state, resulting in low robustness of the model trained based on the cable label image. For this characteristic, preprocessing of tilt image correction is performed on the cable label image to improve the quality of the training sample set. At the same time, considering that during the training process of the cable label recognition model, the cable label images in the training sample set may be affected by various factors such as environmental illumination, debris occlusion, label position offset, and multi-label stacking, resulting in low robustness of the model trained based on the cable label image. Therefore, before training the cable label text detection model, according to the characteristics of the cable label, an image information enhancement mechanism is introduced, multiple data augmentation strategies are preset, at least one target data augmentation strategy is selected from the multiple preset data augmentation strategies, and based on the at least one selected target data augmentation strategy, the cable label image is enhanced, and at the same time, the training sample set is expanded, which is beneficial to improving the quality of the cable image sample set, highlighting the characteristics of the cable label, reducing background interference in the image. Thus, training the model based on the cable label sample set is beneficial to improving the model's recognition ability of the cable label in the cable label image, and further, is beneficial to improving the adaptability to complex recognition environments and improving the accuracy of cable label recognition.

[0076] To improve the diversity of data augmentation processing, in an exemplary embodiment, as Figure 3 shown, S300 includes S320, and S400 includes S420. Among them:

[0077] S320. Randomly select at least two target data augmentation strategies from multiple preset data augmentation strategies.

[0078] S420. Iteratively perform data augmentation processing on the cable label image based on the at least two selected target data augmentation strategies to obtain a first cable label sample set.

[0079] Among them, the first cable label sample set includes the cable label image and the cable label image after data augmentation processing.

[0080] In practical applications, the processor may pre-assign strategy numbers to different preset data augmentation strategies, randomly select at least two non-repeating strategy numbers from within the range of strategy numbers through a random number generator, and determine the data augmentation strategies corresponding to the selected strategy numbers as the target data augmentation strategies. In other embodiments, it may also be to randomly select at least two target data augmentation strategies from multiple preset data augmentation strategies through a random number generation algorithm. The random number generation algorithm includes, but is not limited to, the linear congruential method and the Mersenne twister algorithm, etc. This embodiment does not limit the method of randomly selecting target data augmentation strategies.

[0081] After at least two target data augmentation strategies are selected, the processor can perform iterative data augmentation processing on the cable label image using the target data augmentation strategies in a random data augmentation processing order to obtain a first cable label sample set. Exemplarily, the selected target data augmentation strategies include contrast enhancement, geometric transformation, and filtering operations. A possible iterative data augmentation processing order may be geometric transformation -> contrast enhancement -> filtering operation. In other embodiments, it may also be to preset a processing priority for different data augmentation strategies, and determine the data augmentation processing order for iterative processing according to the processing priorities of the selected target data augmentation strategies.

[0082] In this embodiment, by selecting at least two target data augmentation strategies and performing iterative data augmentation processing on the cable label image, a cable label training sample set is obtained, which improves the diversity of the training sample set and is beneficial to improving the robustness and generalization ability of the model based on the diverse training sample set.

[0083] To improve the diversity of data augmentation processing, in an exemplary embodiment, selecting at least one target data augmentation strategy from multiple preset target data augmentation strategies further includes:

[0084] Randomly select one target data augmentation strategy from multiple preset data augmentation strategies.

[0085] Performing data augmentation processing on the cable label image based on the at least one selected target data augmentation strategy to obtain a cable label sample set further includes:

[0086] Perform data augmentation processing on the cable label image based on the target data augmentation strategy to obtain a second cable label sample set.

[0087] Merge the first cable label sample set and the second cable label sample set to obtain a cable label sample set.

[0088] Among them, the second cable label sample set includes the cable label image and the cable label image after data augmentation processing.

[0089] In practical applications, the processor randomly selects a target data augmentation strategy from the preset data augmentation strategies. It can be the step of randomly selecting at least two target data augmentation strategies from the preset data augmentation strategies in the above-mentioned embodiments, which will not be elaborated here. After selecting a target data augmentation strategy, perform data augmentation processing on the cable label image based on the target data augmentation strategy to obtain a second cable label sample, and merge the first cable label sample set and the second cable label sample set to obtain a diverse cable label sample set for sample expansion for each cable label image.

[0090] In other embodiments, in order to balance the effect of training the model based on the cable label sample set and reduce the possibility of the model being underfitted, it can be that the operator pre-sets the number of the expanded sample set based on the architecture of the model to be trained and the model training experience, and performs iterative data augmentation processing on the cable label image based on at least two selected target data augmentation strategies until the preset number of the expanded sample set is reached to obtain a cable label sample set.

[0091] Exemplarily, through experiments by the operator, when the number of expanded samples is between 20 - 30, the effect of enhancing the training accuracy is better, and when it exceeds 40, the model is prone to underfitting problems. Therefore, for at least two randomly selected target data augmentation strategies, set the first number of the expanded sample set, and use this target data augmentation strategy for data augmentation processing until the number of cable label images after data augmentation processing reaches the preset first number of the expanded sample set to obtain a first expanded sample set; for one randomly selected target data augmentation strategy, set the second number of the expanded sample set, and use the selected target data augmentation strategy for iterative data augmentation processing until the number of cable label images after data augmentation processing reaches the preset second number of the expanded sample set to obtain a second expanded sample set. Merge the first expanded sample set, the second expanded sample set, and the cable label images to obtain a cable label training sample set.

[0092] In this embodiment, by randomly selecting the target data augmentation strategy and performing data augmentation processing on the cable label image, it is beneficial to improve the diversity of training samples.

[0093] In order to improve the pertinence of data augmentation processing, in an exemplary embodiment, the preset data augmentation strategies at least include a noise addition strategy, a noise elimination strategy, and an image cropping strategy.

[0094] Considering the noise interference factors suffered by cable label images during transmission and storage, a noise addition strategy is set to simulate this situation for data augmentation. Specifically, the noise addition strategy can be to add salt-and-pepper noise, Poisson noise, speckle noise, exponential noise, or uniform noise, etc. to the cable label image.

[0095] Considering that there are various interference factors in the cable label image in the actual scenario, a noise elimination strategy is set for data augmentation to improve the image quality. Specifically, the noise removal strategy can include removing noise from the cable label image based on a mean filter, a median filter, or a bilateral filter, etc.

[0096] Considering that the cable label image may be collected from different shooting perspectives, an image cropping strategy is set to simulate different perspective changes of the imaging device. Specifically, the image cropping strategy can include central cropping or edge cropping, etc. Central cropping can be to crop a preset-sized area from the center of the image. Edge cropping can be to crop a preset-sized area from the edge of the image.

[0097] In this embodiment, by setting targeted data augmentation strategies, it is beneficial to improve the effect of training the model based on the training sample set after data augmentation.

[0098] In order to improve the pertinence of data augmentation processing, the cable label image is subjected to data augmentation processing according to the characteristics of the cable label image. In an exemplary embodiment, the number of cable label images is multiple, and the cable label sample set processing method further includes S442 to S444, where:

[0099] S442, when the target data augmentation strategy is the noise addition strategy, for each cable label image, based on the maximum pixel value of the cable label image, determine the noise value range.

[0100] S444, based on the noise value range, add Gaussian noise to the cable label image to obtain the cable label image after data augmentation.

[0101] In specific implementation, in order to balance data augmentation and sample expansion based on noise addition and minimize the impact on key information in the original image, the noise value range for adding noise to the cable label image is determined by combining the noise addition impact requirements and the maximum pixel value of the cable label image. For example, if the noise addition impact requirement indicates that the impact of the added noise on the original image needs to be controlled below 10%, let a and b be the upper and lower limits of the added noise value range respectively, which are determined according to the maximum pixel value of the original image. Exemplarily, let .

[0102] After determining the noise value range, the processor adds Gaussian noise to the cable label image according to the noise value range, so as to use additive Gaussian noise as the enhancement information during image training. Specifically:

[0103]

[0104]

[0105]

[0106]

[0107] In the formula, is the noise value added at the original image location, is the mean value of the noise, is the variance of the noise, and a and b are the upper and lower limits of the added Gaussian noise value range respectively; is the sample value randomly obtained within the range, is to take number of samples. In this embodiment, the number of samples m is dynamically adjusted according to the upper and lower limits of the Gaussian noise value range. The larger the range, the more the number of samples. Therefore, for each sample value, a sample is generated once, that is .

[0108] In this embodiment, by adding noise to the cable label image, the noise interference suffered by the cable label image during transmission and storage is simulated, which is beneficial to training the model with the cable label image after data augmentation, and improving the robustness, generalization ability and adaptability to noise of the model.

[0109] To improve the pertinence of data augmentation processing, in an exemplary embodiment, the cable label sample set processing method further includes S460, where:

[0110] S460, when the target data augmentation strategy is the noise elimination strategy, perform Gaussian filtering on the cable label image to obtain the cable label image after data augmentation.

[0111] In practical applications, for the random noise processor of an image, filtering is performed by using a filter kernel convolution according to the following formula:

[0112]

[0113]

[0114] In the formula, represents the original image, represents the enhanced image, is the value at the position of the image ; represents the filter convolution kernel, is the variance, which is taken as 1 in this embodiment; is the distance from the current position in the convolution kernel to the center of the convolution kernel.

[0115] Specifically, when the target data augmentation strategy is noise elimination, the size of the filter convolution kernel starts with 3x3, and the image is filtered according to the above formula. The size of the filter convolution kernel increases with the current number of expansions i.e., .

[0116] In this embodiment, by removing noise from the cable label image, it is beneficial to remove the noise in the image, improve the image quality. At the same time, the edge and detail features of the image are enhanced, which is conducive to training a model based on the cable label image after data augmentation and improving the accuracy of model recognition.

[0117] To improve the pertinence of data augmentation processing, in an exemplary embodiment, the cable label sample set processing method further includes S482 to S484, where:

[0118] S482, when the target data augmentation strategy is the image cropping strategy, obtain the minimum horizontal cropping distance and the minimum vertical cropping distance.

[0119] S484, crop the cable label image based on the minimum horizontal cropping distance and the minimum vertical cropping distance to obtain the cable label image after data augmentation.

[0120] Among them, the minimum horizontal cropping distance is the minimum cropping distance for the horizontal direction of the cable label image, and the minimum vertical cropping distance is the minimum cropping distance for the vertical direction of the cable label image. The minimum horizontal cropping distance and the minimum vertical cropping distance can be determined by analyzing the layout characteristics of the text information and the label edge in the cable label image.

[0121] In practical applications, for the cropping of cable label images, a large number of historical cable label images can be obtained in advance. By detecting the text area of the cable label in the image, and then determining the horizontal distance and vertical distance between the edge of the cable label and the detected text area, considering the interference factors in the actual application scenario, the minimum horizontal cropping distance and the minimum vertical cropping distance that minimize the impact on the text information on the cable label when cropping the edge of the original image are determined. Exemplarily, to determine the horizontal distance and vertical distance between the edge of the cable label and the detected text area, a text detection algorithm (such as CTPN (Detecting Text in Natural Image with Connectionist Text Proposal Network, connected text proposal network), EAST (Efficient and Accurate Scene Text Detector, efficient and accurate scene text detector), etc.) or optical character recognition can be used to determine the text area in the cable label image. Then, an edge detection algorithm (such as Canny edge detection) is used to identify the edge contour of the cable label. Based on the text area and edge contour of the cable label, the vertical distance and horizontal distance between the label edge and the text area are determined.

[0122] In one embodiment, the minimum horizontal cropping distance for the cable label image is determined to be approximately 1 / 8 of the cable label length, and the minimum vertical cropping distance for the cable label image is determined to be approximately 1 / 6 of the cable label width.

[0123] During specific implementation, the processor obtains the minimum horizontal cropping distance and the minimum vertical cropping distance, randomly determines the cropping size for the length of the original image according to the minimum horizontal cropping distance, randomly determines the cropping size for the width of the original image according to the minimum vertical cropping distance, and crops the image according to the determined cropping size:

[0124]

[0125]

[0126]

[0127]

[0128]

[0129] Among them, and respectively represent the length and width of the original cable label image, is the cropping size.

[0130] In this embodiment, according to the characteristics of the cable label, the minimum horizontal cropping distance and the minimum vertical cropping distance for cropping the cable label image are determined in advance, so as to crop the cable label image based on the minimum horizontal cropping distance and the minimum vertical cropping distance, which is beneficial to reducing the impact of image cropping on the text information on the cable label. At the same time, it is beneficial to train the model according to the cable label image after image cropping to improve the generalization ability of the model.

[0131] According to the characteristics of the cable label image, preprocess the cable label image. In an exemplary embodiment, the preprocessing further includes at least one of grayscale conversion and binarization.

[0132] In this embodiment, considering the problems of inclination and bending of the cable label in the cable label image, and at the same time, factors such as ambient light, debris occlusion, and multi-label stacking affect the poor effect of training the model based on the cable label sample set. According to the characteristics of these cable label images, perform at least one preprocessing of grayscale conversion, binarization, and skew image correction on the cable label image.

[0133] Among them, grayscale conversion can perform grayscale conversion operations on the cable label image by the maximum value method, the average value method, or the weighted average value method. Exemplarily, taking the grayscale conversion operation of the cable label image by the weighted average value method as an example for illustration. Specifically, according to the characteristics of the cable label image and actual requirements, weights are respectively assigned to the image in the three channels of R (Red), G (Green), and B (Blue). For example, the weights of the three channels are 0.3, 0.59, and 0.11 respectively. During specific implementation, for each pixel of the cable label image, the grayscale value of the image is determined by the method of weighted summation:

[0134]

[0135] In the formula, (x, y) is the original pixel value in the cable label image, and k1, k2, k3 (k1 + k2 + k3 = 1) are the weights of the R, G, and B channels respectively.

[0136] In one implementation manner, after the grayscale conversion operation, the cable label image can be binarized based on the global threshold method, the bimodal method, or the maximum inter-class variance method. Taking the binarization of the cable label image based on the maximum inter-class variance as an example for illustration. Specifically:

[0137] First, determine the number of pixel points in the cable label image, assumed to be N. Then, clarify the range of grayscale values in the image. Assume that there are L grayscale values, and the value range of the threshold T is from 0 to L - 1.

[0138] Secondly, select a threshold T. According to the value of T, the grayscale image is divided into two groups, V0 and V1. V0 contains pixel points with grayscale values in the range of 0 to T, and V1 contains pixel points with grayscale values in the range of T + 1 to L - 1. ni represents the number of pixel points with grayscale value i. The probability of the grayscale value i appearing in the image is Pi = ni / N. The proportions of pixel points occupied by V0 and V1 are respectively and , and the average grayscale values of V0 and V1 are respectively and , which are calculated through the following formula:

[0139]

[0140]

[0141] Thirdly, calculate the total average grayscale value of the cable label image:

[0142]

[0143] Calculate the between-class variance:

[0144]

[0145] Finally, calculate the variance based on the above information, traverse the threshold range, calculate the between-class variance corresponding to each threshold T, and determine the threshold T that maximizes the between-class variance as the optimal threshold T.

[0146]

[0147] Binarize the cable label image according to the optimal threshold. Exemplarily, set the pixel values greater than the optimal threshold to 255 (white) and the pixel values less than or equal to the optimal threshold to 0 (black).

[0148] In one embodiment, for the image after grayscale and binarization, it can be subjected to skew correction based on transformation algorithms such as Rando transformation or Hough transformation. The method for skew correction of the cable label image after grayscale and binarization refers to the steps of skew correction of the cable label image in the above-mentioned embodiment, which will not be elaborated here.

[0149] In some embodiments, the preprocessing may further include image sharpening and image smoothing, etc. Among them, image sharpening can be to perform calculus processing on the image and superimpose the operation result on the original image. Image smoothing can be to eliminate the noise in the image based on the neighborhood averaging method, median filtering method or bilateral filtering (edge-preserving filtering).

[0150] In this embodiment, according to the characteristics of the cable label image, before performing data augmentation processing on the image, preprocessing is performed on the cable label image, which is beneficial to reducing the influence of factors such as ambient light, debris occlusion, and multi-label stacking in the cable label image on the detection result.

[0151] To make a clearer description of the cable label sample set processing method provided in this application, a specific embodiment is described below. The specific embodiment includes the following steps:

[0152] S1. Obtain the cable label image.

[0153] S2. Perform preprocessing on the cable label image, and the preprocessing includes at least one of grayscale conversion, binarization, and skew image correction.

[0154] S3. Randomly select at least two target data augmentation strategies from multiple preset data augmentation strategies, and based on the at least two selected target data augmentation strategies, perform iterative data augmentation processing on the preprocessed cable label image to obtain the first cable label sample set.

[0155] S4. Randomly select one target data augmentation strategy from multiple preset data augmentation strategies, and based on the target data augmentation strategy, perform data augmentation processing on the preprocessed cable label image to obtain the second cable label sample set.

[0156] Among them, the target data augmentation strategy includes a noise addition strategy, a noise elimination strategy, and an image cropping strategy. Performing data augmentation processing on the cable label image according to at least one selected target data augmentation strategy further includes:

[0157] (1) When the target data augmentation strategy is the noise addition strategy, for each cable label image, based on the maximum pixel value of the cable label image, determine the noise value range, and based on the noise value range, add Gaussian noise to the cable label image to obtain the data-augmented cable label image.

[0158] (2) When the target data augmentation strategy is the noise elimination strategy, perform Gaussian filtering on the cable label image to obtain the data-augmented cable label image.

[0159] (3) When the target data augmentation strategy is the image cropping strategy, obtain the minimum horizontal cropping distance and the minimum vertical cropping distance, and crop the cable label image based on the minimum horizontal cropping distance and the minimum vertical cropping distance to obtain the data-augmented cable label image.

[0160] It should be understood that although the steps in the flowcharts involved in the above-described embodiments are sequentially shown as indicated by the arrows, these steps are not necessarily executed in the order indicated by the arrows. Unless there is a clear description in this article, there is no strict order restriction for the execution of these steps, and these steps can be executed in other orders. Moreover, at least a part of the steps in the flowcharts involved in the above-described embodiments may include multiple steps or multiple stages. These steps or stages are not necessarily executed at the same time, but can be executed at different times. The execution order of these steps or stages is not necessarily sequential, but can be executed alternately or in turn with at least a part of other steps or steps or stages in other steps.

[0161] The cable label recognition method provided by the embodiments of this application can be applied to an application environment as Figure 1 shown. Among them, the terminal 102 communicates with the server 104 through the network. The data storage system can store the data that the server 104 needs to process. The data storage system can be integrated on the server 104, or can be placed in the cloud or on other network servers.

[0162] Specifically, it can be that an operator sends a cable label recognition message to the server 104 through the terminal, and the server receives the cable label recognition message. Secondly, an image of the cable label to be detected is obtained. Using the image of the cable label to be detected as the input, a trained cable label text detection model is called to obtain a text detection result. The cable label text detection model is trained based on the cable label sample set processing method of any one of the above. Finally, using the text detection result as the input, a trained cable label text recognition model is called to obtain a text recognition result. The cable label text recognition model is trained based on the cable label historical image set carrying text labels.

[0163] In an exemplary embodiment, this application also provides a cable label recognition method, as Figure 4 shown, including the following S500 to S700. Among them:

[0164] S500, obtain an image of the cable label to be detected.

[0165] In practical applications, the method for obtaining an image of the cable label to be detected can refer to the steps of obtaining a cable label image in the embodiments of the above cable label sample set processing method, which will not be elaborated here.

[0166] S600, using the image of the cable label to be detected as the input, call a trained cable label text detection model to obtain a text detection result. The cable label text detection model is trained based on the cable label sample set processing method of any one of the above.

[0167] Among them, the text detection result may include the position information of the text area on the cable label in the cable label image and the cable label image of the detection box annotating the text area.

[0168] In practical applications, an initial cable label text detection model can be constructed in advance based on a deep neural network model. The deep neural network model includes, but is not limited to, the CTPN model, the EAST model, and the PixelLink model. Exemplarily, in this embodiment, an initial cable label text detection model is constructed based on the CTPN model. The processor takes the cable label training samples as input, calls the constructed initial cable label text detection model, and iteratively trains the model to detect the text area on the cable label until a preset training end condition is reached, obtaining a trained cable label text detection model. Among them, the preset training end condition may be that the loss function value is less than a preset loss threshold for a preset number of consecutive times. Specifically, in implementation, the processor takes the cable label image to be detected as input, calls the trained cable label text detection model, and outputs the position information of the text area on the cable label.

[0169] S700, taking the text detection result as input, calls the trained cable label text recognition model to obtain a text recognition result. The cable label text recognition model is trained based on a cable label historical image set carrying text labels.

[0170] Among them, the text recognition result may include the text content on the cable label.

[0171] In practical applications, a cable label historical image set is obtained in advance. The obtaining can refer to the steps of obtaining cable label images in the embodiments of the above cable label sample set processing method, which will not be elaborated here. Data annotation is performed on the images in the cable label historical image set, and text labels are annotated. The annotation method may include annotation based on an annotation tool or manual annotation.

[0172] An initial cable label text detection model is constructed in advance based on a deep neural network model, and the deep neural network model includes, but is not limited to, a CRNN (Convolutional Recurrent Neural Network) model, a TextCNN (Text Convolutional Neural Network) model, and a recurrent neural network model. Exemplarily, in this embodiment, an initial cable label text recognition model is constructed based on the CRNN model. The processor takes a cable label historical image set carrying text labels as input, calls the constructed initial cable label text recognition model, and iteratively trains the model to recognize the text content on the cable label until a preset training end condition is reached, obtaining a trained cable label text recognition model. Among them, the preset training end condition may be that the loss function value is less than a preset loss threshold for a preset number of consecutive times. Specifically in implementation, the processor takes the text detection result as input, calls the trained cable label text recognition model, and outputs the text content on the cable label.

[0173] In this embodiment, first, a cable label text detection model is trained in advance based on the above cable label sample set processing method. The obtained cable label image to be detected is input into the cable label text detection model to obtain the text detection result of the cable label. In this way, training the cable label text detection model based on the sample set after data augmentation and expansion is beneficial to improving the adaptability of the model to the cable label text detection in a complex detection environment and improving the accuracy of the cable label text detection. Secondly, the accurate text detection result is input into the trained cable label text recognition model to obtain the text recognition result, improving the accuracy of the cable label text recognition and contributing to improving the accuracy of cable management and maintenance.

[0174] In an exemplary embodiment, as Figure 5As shown, an overall process for detecting and recognizing cable label text is provided. Specifically, first, an original image set is obtained. The original image set includes multiple cable label images. For each cable label image in the original image set, preprocessing operations such as binarization, edge detection, and Hough transform are sequentially performed on the cable label image. Then, at least two target data augmentation strategies are randomly selected from a preset data augmentation strategy. According to the at least two selected target data augmentation strategies, data augmentation processing is performed on the cable label image until the number of cable label images after data augmentation processing reaches a preset first augmented sample set number, obtaining a first cable label training sample set; one target data augmentation strategy is randomly selected from the preset data augmentation strategy, and data augmentation processing is performed using this target data augmentation strategy until the number of cable label images after data augmentation processing reaches a preset second augmented sample set number, obtaining a second cable label training sample set. The first cable label training sample set and the second cable label training sample set are combined to obtain a cable label training sample set. The cable label training sample set is input into a CTPN model for training until a preset training end condition is satisfied, obtaining a trained cable label text detection model. Then, the cable label image to be recognized is input into the trained cable label text detection model to obtain a cable text detection result. The cable text detection result is input into the trained cable label text recognition model to obtain a cable label text recognition result.

[0175] In an exemplary embodiment, as Figure 6 shown, a cable label sample set processing device 600 is provided, including: an image acquisition module 610, an image preprocessing module 620, a strategy screening module 630, and an enhancement processing module 640, where:

[0176] The image acquisition module 610 is configured to acquire a cable label image;

[0177] The image preprocessing module 620 is configured to perform preprocessing on the cable label image, and the preprocessing includes skew image correction;

[0178] The strategy screening module 630 is configured to screen at least one target data augmentation strategy from a plurality of preset data augmentation strategies;

[0179] The enhancement processing module 640 is configured to perform data augmentation processing on the preprocessed cable label image based on the at least one selected target data augmentation strategy to obtain a cable label sample set.

[0180] In an exemplary embodiment, the strategy screening module 630 is further configured to randomly screen at least two target data augmentation strategies from a plurality of preset data augmentation strategies;

[0181] The enhancement processing module 640 is further configured to perform data enhancement processing on the cable label image iteratively based on at least two target data enhancement strategies screened, to obtain a first cable label sample set.

[0182] In an exemplary embodiment, the cable label sample set processing device 600 further includes a sample set merging module 650, where:

[0183] The policy screening module 630 is further configured to randomly screen out a target data enhancement policy from a plurality of preset data enhancement policies;

[0184] The enhancement processing module 640 is further configured to perform data enhancement processing on the cable label image based on the target data enhancement policy, to obtain a second cable label sample set;

[0185] The sample set merging module 650 is configured to merge the first cable label sample set and the second cable label sample set to obtain a cable label sample set.

[0186] In an exemplary embodiment, when the target data enhancement policy is a noise addition policy, the enhancement processing module 640 is further configured to, for each cable label image, determine a noise value range based on the maximum pixel value of the cable label image, and add Gaussian noise to the cable label image based on the noise value range to obtain a data-enhanced cable label image.

[0187] In an exemplary embodiment, when the target data enhancement policy is a noise cancellation policy, the enhancement processing module 640 is further configured to perform Gaussian filtering on the cable label image to obtain a data-enhanced cable label image.

[0188] In an exemplary embodiment, when the target data enhancement policy is an image cropping policy, the enhancement processing module 640 is further configured to obtain a minimum horizontal cropping distance and a minimum vertical cropping distance, and crop the cable label image based on the minimum horizontal cropping distance and the minimum vertical cropping distance to obtain a data-enhanced cable label image.

[0189] In an exemplary embodiment, as Figure 7 shown, a cable label recognition device 700 is provided, including: a cable label image acquisition module 710, a cable label text detection module 720, and a cable label text recognition module 730, where:

[0190] The cable label image acquisition module 710 is configured to acquire a cable label image to be detected;

[0191] The cable label text detection module 720 is configured to use the cable label image to be detected as input, call the trained cable label text detection model, and obtain the text detection result. The cable label text detection model is trained based on the cable label sample set processing method of any of the above.

[0192] The cable label text recognition module 730 is configured to use the text detection result as input, call the trained cable label text recognition model, and obtain the text recognition result. The cable label text recognition model is trained based on the cable label historical image set carrying text labels.

[0193] Each module in the above cable label sample set processing device 600 and the cable label recognition device 700 can be implemented in whole or in part by software, hardware, and their combination. The above modules can be embedded in the processor of the computer device in hardware form or independent of it, or stored in the memory of the computer device in software form, so that the processor can call and execute the operations corresponding to each of the above modules.

[0194] In an exemplary embodiment, a computer device is provided. The computer device can be a server, and its internal structure diagram can be as Figure 8 shown. The computer device includes a processor, a memory, an input / output interface (Input / Output, abbreviated as I / O), and a communication interface. Among them, the processor, the memory, and the input / output interface are connected through a system bus, and the communication interface is connected to the system bus through the input / output interface. Among them, the processor of the computer device is used to provide computing and control capabilities. The memory of the computer device includes a non-volatile storage medium and an internal memory. The non-volatile storage medium stores an operating system, a computer program, and a database. The internal memory provides an environment for the operation of the operating system and the computer program in the non-volatile storage medium. The database of the computer device is used to store data. The input / output interface of the computer device is used to exchange information between the processor and external devices. The communication interface of the computer device is used to communicate with external terminals through a network connection. When the computer program is executed by the processor, it implements a cable label sample set processing method and a cable label recognition method.

[0195] Those skilled in the art can understand that Figure 8 the structure shown in

[0196] In an exemplary embodiment, a computer device is provided, which includes a memory and a processor. A computer program is stored in the memory. When the processor executes the computer program, the steps in any of the above-described embodiments of the cable label sample set processing method and the steps in an embodiment of the cable label recognition method are implemented.

[0197] In an embodiment, a computer-readable storage medium is provided, on which a computer program is stored. When the computer program is executed by a processor, the steps in any of the above-described embodiments of the cable label sample set processing method and the steps in an embodiment of the cable label recognition method are implemented.

[0198] In an embodiment, a computer program product is provided, which includes a computer program. When the computer program is executed by a processor, the steps in any of the above-described embodiments of the cable label sample set processing method and the steps in an embodiment of the cable label recognition method are implemented.

[0199] It should be noted that the user information (including but not limited to user device information, user personal information, etc.) and data (including but not limited to data for analysis, stored data, displayed data, etc.) involved in this application are all information and data authorized by the user or fully authorized by all parties, and the collection, use, and processing of relevant data need to comply with relevant regulations.

[0200] Those of ordinary skill in the art can understand that all or part of the processes in the methods of the above embodiments can be completed by instructing relevant hardware through a computer program. The computer program can be stored in a non-volatile computer-readable storage medium. When the computer program is executed, it can include the processes of the embodiments of the above methods. Among them, any reference to a memory, database, or other medium used in the embodiments provided in this application can include at least one of non-volatile memory and volatile memory. Non-volatile memory can include read-only memory (ROM), magnetic tape, floppy disk, flash memory, optical memory, high-density embedded non-volatile memory, resistive random access memory (ReRAM), magnetoresistive random access memory (MRAM), ferroelectric random access memory (FRAM), phase change memory (PCM), graphene memory, etc. Volatile memory can include random access memory (RAM) or external cache memory, etc. By way of illustration and not limitation, RAM can be in various forms, such as static random access memory (SRAM) or dynamic random access memory (DRAM), etc. The databases involved in the embodiments provided in this application can include at least one of relational databases and non-relational databases. Non-relational databases can include distributed databases based on blockchain, etc., without limitation. The processors involved in the embodiments provided in this application can be general-purpose processors, central processing units, graphics processing units, digital signal processors, programmable logic devices, data processing logics based on quantum computing, artificial intelligence (AI) processors, etc., without limitation.

[0201] The technical features of the above embodiments can be combined arbitrarily. For the sake of brevity of description, not all possible combinations of the technical features in the above embodiments are described. However, as long as there is no contradiction in the combination of these technical features, it should be considered as the scope recorded in this application.

[0202] The above-described embodiments merely represent several implementation manners of the present application. The description thereof is relatively specific and detailed, but it should not be construed as a limitation on the patent scope of the present application. It should be noted that for those of ordinary skill in the art, without departing from the concept of the present application, several modifications and improvements can still be made, and these all fall within the protection scope of the present application. Therefore, the protection scope of the present application shall be subject to the appended claims.

Claims

1. A method for processing a cable label sample set, characterized in that The method includes: Obtain a cable label image; Preprocess the cable label image, where the preprocessing includes skew image correction; Select at least one target data augmentation strategy from multiple preset data augmentation strategies; Based on the at least one selected target data augmentation strategy, perform data augmentation processing on the preprocessed cable label image to obtain a cable label sample set.

2. The method according to claim 1, wherein The preset data augmentation strategies at least include a noise addition strategy, a noise cancellation strategy, and an image cropping strategy.

3. The method according to claim 2, wherein The number of the cable label images is multiple, and the method further includes: When the target data augmentation strategy is the noise addition strategy, for each cable label image, determine a noise value range based on the maximum pixel value of the cable label image; Based on the noise value range, add Gaussian noise to the cable label image to obtain a data-augmented cable label image.

4. The method according to claim 2, wherein The method further includes: When the target data augmentation strategy is the image cropping strategy, obtain a minimum horizontal cropping distance and a minimum vertical cropping distance; Based on the minimum horizontal cropping distance and the minimum vertical cropping distance, crop the cable label image to obtain a data-augmented cable label image.

5. The method according to claim 2, wherein The method further includes: When the target data augmentation strategy is the noise cancellation strategy, perform Gaussian filtering on the cable label image to obtain a data-augmented cable label image.

6. The method according to claim 1, wherein The step of selecting at least one target data augmentation strategy from multiple preset target data augmentation strategies includes: Randomly select at least two target data augmentation strategies from multiple preset data augmentation strategies; The step of performing data augmentation processing on the cable label image based on the at least one selected target data augmentation strategy to obtain a cable label sample set includes: Based on the at least two selected target data augmentation strategies, iteratively perform data augmentation processing on the cable label image to obtain a first cable label sample set.

7. The method according to claim 6, wherein The step of selecting at least one target data augmentation strategy from multiple preset target data augmentation strategies further includes: Randomly select one target data augmentation strategy from multiple preset data augmentation strategies; The step of performing data augmentation processing on the cable label image based on the at least one selected target data augmentation strategy to obtain a cable label sample set further includes: Based on the target data augmentation strategy, perform data augmentation processing on the cable label image to obtain a second cable label sample set; Merge the first cable label sample set and the second cable label sample set to obtain a cable label sample set.

8. The method according to any one of claims 1 to 7, characterized in that, The preprocessing further includes at least one of grayscale conversion and binarization.

9. A method for identifying cable labels, characterized in that, The method includes: Obtain a cable label image to be detected; Using the cable label image to be detected as input, call a trained cable label text detection model to obtain a text detection result, where the cable label text detection model is trained based on the cable label sample set processing method according to any one of claims 1 to 7. Taking the text detection result as input, a trained cable label text recognition model is called to obtain a text recognition result, and the cable label text recognition model is trained based on a historical image set of cable labels carrying text labels.

10. A cable label sample set processing device, characterized in that, The device includes: an image acquisition module for acquiring a cable label image; a policy screening module for screening at least one target data augmentation policy from a plurality of preset data augmentation policies; an augmentation processing module for performing data augmentation processing on the cable label image based on the at least one screened target data augmentation policy to obtain a cable label sample set.

Images