Relay protection setting value data checking method and device based on OCR (Optical Character Recognition)

Abstract

The invention discloses a relay protection setting value data checking method and device based on OCR (Optical Character Recognition), and relates to the technical field of text recognition. Carrying out text region identification on the constant value protection device image and then carrying out image enhancement to obtain an enhanced constant value protection device image; performing OCR identification on the constant value single image to obtain first constant value data, and performing OCR identification on the enhanced constant value protection device image to obtain second constant value data; performing similarity comparison on the first constant value data and the second constant value data; if the similar comparison result is that the content is consistent, storing the constant value single image and the enhanced constant value protection device image, and generating an electronic report; according to the method, text region identification and enhancement processing are firstly carried out on device images, then constant value data of the two types of images are extracted through the OCR technology, and automatic similarity comparison is carried out, so that the identification problem caused by poor image quality can be solved in a targeted manner, dependence on energy of manual checking personnel can be thoroughly eliminated, and the checking efficiency of the relay protection constant value data is improved.

Description

Technical Field

[0001] This invention belongs to the field of text recognition technology, specifically relating to a method and apparatus for verifying relay protection setting data based on OCR recognition. Background Technology

[0002] Substation setting protection verification is a key link in ensuring the safe and stable operation of the power system. According to industry standards, each substation must complete the setting protection verification twice a year. The current mainstream verification mode relies on manual operation, and the process is as follows: After the dispatching department exports the setting sheet from the system, two staff members need to work together. One person operates the setting protection device and reads the data, while the other person holds the setting sheet and checks it item by item, records any inconsistencies, and completes the stamping and confirmation procedures after all verifications are completed.

[0003] However, this model has significant drawbacks: First, the total amount of set data is huge, and the visual and operational fatigue of the verification personnel is likely to occur after long hours of work, which leads to an increase in the verification error rate and the risk of human error. Second, manual verification is inefficient, and coupled with the large number of substations, the annual labor cost remains high, resulting in low overall work efficiency.

[0004] While some existing power data verification tools exist, they are mostly limited to processing single-format data, such as supporting only electronic setting sheets or data reading from specific device models. They also lack an integrated automatic acquisition, comparison, recording, and reminder mechanism, failing to fundamentally solve the drawbacks of manual verification. Therefore, an optimized solution of acquiring images and conducting data recognition and comparison has been proposed. However, the images acquired in practice often suffer from poor quality, increasing the difficulty of data recognition and resulting in low efficiency in verifying relay protection setting data. Summary of the Invention

[0005] The purpose of this invention is to solve the problem that the acquired images are often of poor quality, which increases the difficulty of data recognition and makes the verification of relay protection setting data inefficient. Therefore, this invention proposes a method and device for verifying relay protection setting data based on OCR recognition.

[0006] In a first aspect of this invention, a method for verifying relay protection setting data based on OCR recognition is first proposed, the method comprising: Acquire the image of the setting sheet and the image of the setting protection device, respectively; The target text region recognition image is obtained by performing text region recognition on the image of the set value protection device; Image enhancement is performed on the target text region recognition image to obtain an enhanced fixed value protection device image; The first set value data is obtained by performing OCR recognition on the image of the set value sheet, and the second set value data is obtained by performing OCR recognition on the image of the enhanced set value protection device. A similarity comparison is performed between the first setpoint data and the second setpoint data; If the similarity comparison result is that the content is consistent, then the image of the setting sheet and the image of the enhanced setting protection device are saved, and an electronic report is generated.

[0007] Optionally, performing text region recognition on the image of the setpoint protection device to obtain a target text region recognition image includes: The image of the fixed-value protection device is substituted into the first depth convolution module to obtain the first depth convolution feature; the image of the fixed-value protection device is substituted into the second depth convolution module to obtain the second depth convolution feature; the scale of the second depth convolution feature is half the scale of the image of the fixed-value protection device; the scale of the first depth convolution feature is half the scale of the second depth convolution feature. The first depthwise convolutional feature is subjected to 1×1 convolution, 3×3 convolution and pooling respectively to obtain the first convolutional feature, the second convolutional feature and the first pooling feature. The first convolutional feature, the second convolutional feature and the first pooling feature are concatenated to obtain the first concatenated feature. The second depthwise convolutional feature is subjected to 1×1 convolution, 3×3 convolution and pooling respectively to obtain the third convolutional feature, the fourth convolutional feature and the second pooling feature. The third convolutional feature, the fourth convolutional feature and the second pooling feature are concatenated to obtain the second concatenated feature. Perform a 3×3 convolution operation on the first spliced ​​feature and the second spliced ​​feature respectively to obtain the first spliced ​​convolution feature and the second spliced ​​convolution feature; The first concatenated convolutional feature is upsampled and then concatenated with the second concatenated convolutional feature to obtain the third concatenated feature; After upsampling the third splicing feature, a 1×1 convolution operation is performed to obtain the target text region recognition image.

[0008] Optionally, image enhancement of the target text region recognition image to obtain the enhanced setpoint protection device image includes: The target text region recognition image is sequentially substituted into two preset first convolutional blocks to obtain the first convolutional text feature; The target text region recognition image and the first convolutional text features are concatenated and then normalized to obtain enhanced sample features; The target text region recognition image is sequentially substituted into two preset second convolutional blocks and a degradation-guided modulation block to obtain the second convolutional text features; After fusing the enhanced sample features and the second convolutional text features, the third convolutional text features are obtained by sequentially substituting them into the preset second convolutional block and the degenerate guided modulation block. Substituting the third convolutional text features and the enhanced sample features into the deep enhancement model yields the enhanced fixed-value protection device image.

[0009] Optionally, substituting the third convolutional text features and the enhanced sample features into the deep enhancement model to obtain the enhanced fixed-value protection device image includes: The first fused feature is obtained by multiplying the third convolutional text feature and the enhanced sample feature element by the first fused feature and then inputting the second convolutional block into the preset second convolutional block. The enhanced sample features are substituted into the preset second convolutional block and then multiplied element-wise with the third convolutional text features to obtain the second fused feature; The second fusion feature is normalized and then fused with the first fusion feature to obtain the third fusion feature; After fusing the third fusion feature and the third convolutional text feature, a 1×1 convolution operation is performed to obtain the enhanced fixed value protection device image.

[0010] Optionally, after performing a similarity comparison between the first setpoint data and the second setpoint data, the method further includes: If the similarity comparison result indicates that there are differences in content, then the difference region in the fixed value single image is determined; The differential areas are highlighted and an alarm is issued, and an electronic report is also generated.

[0011] In a second aspect of the present invention, a relay protection setting data verification device based on OCR recognition is provided, comprising: The image data acquisition module is used to acquire images of the setting sheet and the setting protection device, respectively. A text region determination module for a setpoint protection device is used to perform text region recognition on the image of the setpoint protection device to obtain a target text region recognition image. The text region enhancement module for the fixed value protection device is used to enhance the image of the target text region recognition image to obtain an enhanced fixed value protection device image. The fixed value data extraction module is used to perform OCR recognition on the fixed value single image to obtain first fixed value data, and to perform OCR recognition on the enhanced fixed value protection device image to obtain second fixed value data; The data intelligent comparison module is used to perform a similarity comparison between the first set value data and the second set value data; The fixed value data recording module is used to save the fixed value sheet image and the enhanced fixed value protection device image and generate an electronic report if the similarity comparison result is consistent.

[0012] Optionally, the text region determination module of the setpoint protection device includes: A depthwise convolutional feature extraction module is used to substitute the image of the fixed-value protection device into a first depthwise convolutional module to obtain a first depthwise convolutional feature; and to substitute the image of the fixed-value protection device into a second depthwise convolutional module to obtain a second depthwise convolutional feature; the scale of the second depthwise convolutional feature is half the scale of the fixed-value protection device image; and the scale of the first depthwise convolutional feature is half the scale of the second depthwise convolutional feature. The first concatenated feature generation module is used to perform 1×1 convolution, 3×3 convolution and pooling processing on the first depthwise convolutional feature to obtain the first convolutional feature, the second convolutional feature and the first pooling feature, and to concatenate the first convolutional feature, the second convolutional feature and the first pooling feature to obtain the first concatenated feature. The second concatenation feature generation module is used to perform 1×1 convolution, 3×3 convolution and pooling processing on the second depthwise convolution feature to obtain the third convolution feature, the fourth convolution feature and the second pooling feature, and to concatenate the third convolution feature, the fourth convolution feature and the second pooling feature to obtain the second concatenation feature. The concatenated convolutional feature generation module is used to perform 3×3 convolution operations on the first concatenated feature and the second concatenated feature respectively to obtain the first concatenated convolutional feature and the second concatenated convolutional feature; The third splicing feature generation module is used to upsample the first splicing convolutional feature and then splice it with the second splicing convolutional feature to obtain the third splicing feature; The target text region recognition image generation module is used to perform a 1×1 convolution operation on the third splicing feature to obtain the target text region recognition image.

[0013] Optionally, the text area enhancement module of the setpoint protection device includes: The first convolutional text feature generation module is used to sequentially substitute the target text region recognition image into two preset first convolutional blocks to obtain the first convolutional text features. An enhanced sample feature generation module is used to concatenate the target text region recognition image and the first convolutional text features and then normalize them to obtain enhanced sample features. The second convolutional text feature generation module is used to sequentially substitute the target text region recognition image into two preset second convolutional blocks and a degradation-guided modulation block to obtain the second convolutional text features. The third convolutional text feature generation module is used to fuse the enhanced sample features and the second convolutional text features, and then sequentially substitute them into the preset second convolutional block and the degenerate guided modulation block to obtain the third convolutional text features; An enhanced fixed-value protection device image generation module is used to substitute the third convolutional text features and the enhanced sample features into a deep enhancement model to obtain an enhanced fixed-value protection device image.

[0014] Optionally, the enhanced setpoint protection device image generation module includes: The first fusion feature generation module is used to multiply the third convolutional text feature and the enhanced sample feature element by the element and then input the result into the preset second convolutional block to obtain the first fusion feature; The second fusion feature generation module is used to substitute the enhanced sample features into the preset second convolution block and then multiply them element-wise with the third convolution text features to obtain the second fusion feature; The third fusion feature generation module is used to normalize the second fusion feature and then fuse it with the first fusion feature to obtain the third fusion feature; An enhanced value protection device image determination module is used to perform a 1×1 convolution operation on the third fusion feature and the third convolutional text feature to obtain an enhanced value protection device image.

[0015] Optionally, the device further includes: The difference region determination module is used to determine the difference region in the fixed value single image if the similarity comparison result shows that there is a difference in content; The alarm recording module is used to highlight the discrepancies and issue an alarm, and also generate an electronic report.

[0016] The beneficial effects of this invention are: This invention proposes a relay protection setting data verification method based on OCR recognition. First, text region recognition and enhancement processing are performed on the device image. Then, setting data of two types of images are extracted using OCR technology and automatically compared for similarity. Finally, the relevant images are automatically saved and an electronic report is generated for consistent results. This method not only specifically solves the recognition problem caused by poor image quality, but also completely eliminates the dependence on manual verification personnel's energy, reducing the risk of human error caused by visual fatigue and operational fatigue from the source. At the same time, it eliminates the need for two staff members to work together, greatly improving the verification efficiency of massive data and multi-substation scenarios, and improving the efficiency of relay protection setting data verification. Attached Figure Description

[0017] The invention will now be further described with reference to the accompanying drawings.

[0018] Figure 1 A flowchart of a relay protection setting data verification method based on OCR recognition provided in an embodiment of the present invention; Figure 2 This is a schematic diagram of a relay protection setting data verification device based on OCR recognition, provided in an embodiment of the present invention. Detailed Implementation

[0019] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments.

[0020] Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0021] This invention provides a method for verifying relay protection setting data based on OCR recognition. See also... Figure 1 , Figure 1 This is a flowchart illustrating a method for verifying relay protection setting data based on OCR recognition, provided in an embodiment of the present invention. The method includes the following steps: S101, acquire the setting sheet image and the setting protection device image respectively; S102, perform text region recognition on the image of the setpoint protection device to obtain the target text region recognition image; S103, Image enhancement is performed on the target text region recognition image to obtain the enhanced fixed value protection device image; S104, perform OCR recognition on the fixed value sheet image to obtain the first fixed value data, and perform OCR recognition on the enhanced fixed value protection device image to obtain the second fixed value data; S105, perform a similarity comparison between the first fixed value data and the second fixed value data; S106. If the similarity comparison result is that the content is consistent, the image of the setting sheet and the image of the enhanced setting protection device are saved, and an electronic report is generated.

[0022] The present invention provides a method for verifying relay protection setting data based on OCR recognition. First, text region recognition and enhancement processing are performed on the device image. Then, setting data for two types of images are extracted using OCR technology and automatically compared for similarity. Finally, the relevant images are automatically saved and an electronic report is generated for consistent results. This method not only specifically solves the recognition problem caused by poor image quality but also completely eliminates the reliance on manual verification personnel's energy, reducing the risk of human error caused by visual fatigue and operational fatigue from the source. Furthermore, it eliminates the need for two staff members to work together, significantly improving the verification efficiency for massive amounts of data and multiple substation scenarios, thereby enhancing the efficiency of relay protection setting data verification.

[0023] In one implementation, image acquisition and OCR recognition are used to replace manual reading and item-by-item comparison, eliminating the effects of visual fatigue and operational fatigue, and reducing the risk of human error and omission from the source; text region recognition and enhancement processing are first performed on the device image to specifically solve the recognition problem caused by poor image quality, making data extraction more accurate, and solving the problem of low efficiency in relay protection setting data verification due to poor image quality.

[0024] In one implementation, image enhancement of the fixed-value protection device image solves the problem that the actual acquired images often have poor quality. Through enhancement processing, the image clarity and text recognizability are improved, laying the foundation for subsequent OCR recognition.

[0025] In one implementation, by first determining the text region, the text region can be quickly enhanced, reducing the computational load.

[0026] In one implementation, the similarity comparison of the first fixed value data and the second fixed value data specifically includes: matching the parameter name and parameter value of the first fixed value data and the second fixed value data using regular expressions; after ensuring that the two are comparing the same parameter, performing a data comparison to determine whether the data are the same; if they are the same, the similarity comparison result is that the content is consistent; if they are different, the similarity comparison result is that the content is different.

[0027] In one implementation, a correlation is established between the data in the setting sheet and the data in the device based on unique identifiers such as the setting name and number, forming a one-to-one comparison mapping. For scenarios with multiple versions or sets of data, filtering and matching by timestamp or version number is supported. After the comparison is completed, the results are displayed in a table, where consistent items are marked as normal (green) and inconsistent items are marked as abnormal (red), and the specific difference values ​​are displayed. It is also possible to export an Excel format comparison report, which includes elements such as comparison time, device information, and difference details.

[0028] In one embodiment, performing text region recognition on an image of a setpoint protection device to obtain a target text region recognition image includes: The image of the fixed-value protection device is substituted into the first depth convolution module to obtain the first depth convolution feature; the image of the fixed-value protection device is substituted into the second depth convolution module to obtain the second depth convolution feature; the scale of the second depth convolution feature is half the scale of the fixed-value protection device image; the scale of the first depth convolution feature is half the scale of the second depth convolution feature. The first depthwise convolutional feature is processed by 1×1 convolution, 3×3 convolution and pooling respectively to obtain the first convolutional feature, the second convolutional feature and the first pooling feature. The first convolutional feature, the second convolutional feature and the first pooling feature are concatenated to obtain the first concatenated feature. The second depthwise convolutional features are processed by 1×1 convolution, 3×3 convolution and pooling respectively to obtain the third convolutional features, the fourth convolutional features and the second pooling features. The third convolutional features, the fourth convolutional features and the second pooling features are concatenated to obtain the second concatenated features. Perform 3×3 convolution operations on the first and second spliced ​​features respectively to obtain the first spliced ​​convolution feature and the second spliced ​​convolution feature; The first concatenated convolutional feature is upsampled and then concatenated with the second concatenated convolutional feature to obtain the third concatenated feature; After upsampling the third splicing feature, a 1×1 convolution operation is performed to obtain the target text region recognition image.

[0029] In one implementation, two deep convolutional modules are used to extract features at three different levels: the original image (original scale), the first deep convolutional feature (half the original scale), and the second deep convolutional feature (half the scale of the first deep convolutional feature). The first and second deep convolutional modules have identical structures, consisting of convolutional layers, convolutional layers, residual blocks, and residual blocks connected sequentially. The residual blocks in both modules have the same parameters: 1×1 convolution and 3×3 convolution. The first convolutional layer in the first depthwise convolutional module has the following parameters: 64 channels, stride=1, padding=1, and 1×1 convolution. The second convolutional layer has the following parameters: 3×3 convolution, 64 filters, stride=1, padding=1. The third convolutional layer has the following parameters: 3×3 convolution, 128 filters, stride=2, padding=1. The fourth convolutional layer has the following parameters: 3×3 convolution, 128 filters, stride=1, padding=1.

[0030] In one implementation, 1×1 convolution, 3×3 convolution, and pooling operations are used in parallel when processing the first and second depth convolution features. The results are then concatenated, allowing the network to capture features with different receptive fields and different levels of abstraction at the same level, thereby more comprehensively describing text regions and having better adaptability to texts with varied fonts and styles.

[0031] In one implementation, the deep, small-scale feature map (first concatenated convolutional feature) is enlarged to align with the size of the shallow feature map (second concatenated convolutional feature). The upsampled deep semantic features are then concatenated with the shallow, high-resolution detail features. The shallow features provide accurate edge and position information, while the deep features provide strong semantic guidance for the text. The resulting target text region recognition image has very clear and accurate boundaries, and the text region is displayed correctly. The text region is marked, and the pixel values ​​of other regions are assigned to 0 (black), thereby updating the final target text region recognition image.

[0032] In one embodiment, image enhancement of the target text region recognition image to obtain an enhanced fixed-value protection device image includes: The target text region recognition image is sequentially substituted into two preset first convolutional blocks to obtain the first convolutional text features; The enhanced sample features are obtained by concatenating the target text region recognition image and the first convolution text features and then normalizing them. The target text region recognition image is sequentially substituted into two preset second convolutional blocks and a degradation-guided modulation block to obtain the second convolutional text features; After fusing the enhanced sample features and the second convolutional text features, the third convolutional text features are obtained by sequentially substituting them into the preset second convolutional block and the degenerate guided modulation block. Substituting the third convolutional text features and enhanced sample features into the deep enhancement model yields the enhanced fixed-value protection device image.

[0033] In one implementation, two pre-defined first convolutional blocks extract basic text features, which are then concatenated with the original image and normalized. This preserves the original image information and enhances the distinction between text and background through feature fusion, providing high-quality basic samples for subsequent enhancement. Enhancement sample features are extracted through a first path, which focuses on extracting essential features of the text region, such as stroke structure and character shape. Second convolutional text features are extracted through a second path, which uses a Degradation Guided Modulation Block (DGM, existing technology, not described here) to perceive and model the degradation type and degree of the image. This dual-path design allows the model to apply the most appropriate inverse processing to restore the image based on the specific degradation type detected.

[0034] In one implementation, the enhanced sample features are fused with the second convolutional features, and then processed by the second convolutional block and the degenerate modulation block to obtain the third convolutional features, forming an iterative mechanism of extraction, modulation, and re-optimization, which gradually amplifies the text features and suppresses background interference.

[0035] In one implementation, the third convolutional features and enhanced sample features are ultimately fused through a deep enhancement model. This enhances text clarity while preserving the morphological features of the original text to the maximum extent, providing high-fidelity input for subsequent OCR recognition and reducing recognition errors caused by image quality.

[0036] In one embodiment, substituting the third convolutional text features and enhanced sample features into the deep enhancement model to obtain the enhanced fixed-value protection device image includes: The first fused feature is obtained by multiplying the third convolutional text feature and the enhanced sample feature element by the element and then inputting the result into the preset second convolutional block. The enhanced sample features are substituted into the preset second convolutional block and then multiplied element-wise with the third convolutional text features to obtain the second fused feature; The second fusion feature is normalized and then fused with the first fusion feature to obtain the third fusion feature; After fusing the third fusion feature and the third convolutional text feature, a 1×1 convolution operation is performed to obtain the enhanced fixed value protection device image.

[0037] In one implementation, the third convolutional text feature (containing degradation optimization information) and the enhanced sample feature (containing original text basic information) achieve deep interaction between the two features through bidirectional operations of element-wise multiplication and convolution (first and second fused features). The degradation optimization information corrects the original features, while the original features constrain the optimization direction, thus avoiding text distortion during the enhancement process.

[0038] In one implementation, normalizing the second fusion feature can eliminate the differences in numerical distribution between different features, making subsequent fusion more balanced, avoiding the over-amplification of certain features (such as noise features), and enhancing the model's adaptability to complex scenes (such as areas with weak lighting).

[0039] In one implementation, the third fusion feature and the third convolutional text feature are fused again to further integrate the global optimization trend (from the iteratively enhanced third convolutional feature) and local detail information (from the fine structure of the sample features). Finally, the enhanced result focusing on the text region is output through 1×1 convolution to compress the channel dimension, thereby enhancing the clarity of character edges and stroke details.

[0040] In one embodiment, after performing a similarity comparison on the first setpoint data and the second setpoint data, the method further includes: If the similarity comparison result indicates that there are differences in content, then the difference region in the fixed value single image is determined; It highlights and alerts areas of difference, and also generates electronic reports.

[0041] Based on the same inventive concept, this invention also provides a relay protection setting data verification device based on OCR recognition. See also Figure 2 , Figure 2 A schematic diagram of a relay protection setting data verification device based on OCR recognition provided in an embodiment of the present invention includes: The image data acquisition module is used to acquire images of the setting sheet and the setting protection device, respectively. The text region determination module for the setpoint protection device is used to perform text region recognition on the image of the setpoint protection device to obtain the target text region recognition image. The text region enhancement module for the setpoint protection device is used to enhance the image of the target text region recognition image to obtain an enhanced image of the setpoint protection device. The fixed value data extraction module is used to perform OCR recognition on the fixed value sheet image to obtain the first fixed value data, and to perform OCR recognition on the enhanced fixed value protection device image to obtain the second fixed value data. The data intelligent comparison module is used to perform similarity comparison between the first set value data and the second set value data. The fixed value data recording module is used to save the fixed value sheet image and the enhanced fixed value protection device image and generate an electronic report if the similarity comparison result shows that the content is consistent.

[0042] In one embodiment, the text region determination module of the setpoint protection device includes: The depth convolution feature extraction module is used to input the image of the fixed value protection device into the first depth convolution module to obtain the first depth convolution feature; input the image of the fixed value protection device into the second depth convolution module to obtain the second depth convolution feature; the scale of the second depth convolution feature is half the scale of the fixed value protection device image; the scale of the first depth convolution feature is half the scale of the second depth convolution feature. The first concatenated feature generation module is used to perform 1×1 convolution, 3×3 convolution and pooling processing on the first depthwise convolutional feature to obtain the first convolutional feature, the second convolutional feature and the first pooling feature, and to concatenate the first convolutional feature, the second convolutional feature and the first pooling feature to obtain the first concatenated feature. The second concatenation feature generation module is used to perform 1×1 convolution, 3×3 convolution and pooling on the second depth convolution feature to obtain the third convolution feature, the fourth convolution feature and the second pooling feature, and to concatenate the third convolution feature, the fourth convolution feature and the second pooling feature to obtain the second concatenation feature. The concatenated convolutional feature generation module is used to perform 3×3 convolution operations on the first concatenated feature and the second concatenated feature respectively to obtain the first concatenated convolutional feature and the second concatenated convolutional feature; The third concatenation feature generation module is used to upsample the first concatenation convolution feature and then concatenate it with the second concatenation convolution feature to obtain the third concatenation feature. The target text region recognition image generation module is used to perform a 1×1 convolution operation on the third splicing feature to obtain the target text region recognition image.

[0043] In one embodiment, the text area enhancement module of the setpoint protection device includes: The first convolutional text feature generation module is used to sequentially substitute the target text region recognition image into two preset first convolutional blocks to obtain the first convolutional text features. The enhanced sample feature generation module is used to concatenate the target text region recognition image and the first convolutional text features and then normalize them to obtain enhanced sample features; The second convolutional text feature generation module is used to sequentially substitute the target text region recognition image into two preset second convolutional blocks and a degradation-guided modulation block to obtain the second convolutional text features. The third convolutional text feature generation module is used to fuse the enhanced sample features and the second convolutional text features and then sequentially substitute them into the preset second convolutional block and the degenerate guided modulation block to obtain the third convolutional text features. An enhanced fixed-value protection device image generation module is used to input the third convolutional text features and enhanced sample features into a deep enhancement model to obtain an enhanced fixed-value protection device image.

[0044] In one embodiment, the enhanced setpoint protection device image generation module includes: The first fusion feature generation module is used to multiply the third convolutional text features and the enhanced sample features by their elements and then input them into a preset second convolutional block to obtain the first fusion feature. The second fusion feature generation module is used to substitute the enhanced sample features into the preset second convolution block and then multiply them element-wise with the third convolution text features to obtain the second fusion feature. The third fusion feature generation module is used to normalize the second fusion feature and then fuse it with the first fusion feature to obtain the third fusion feature; The enhanced fixed-value protection device image determination module is used to perform a 1×1 convolution operation on the third fusion feature and the third convolution text feature to obtain the enhanced fixed-value protection device image.

[0045] In one embodiment, the apparatus further includes: The difference region determination module is used to determine the difference region in the fixed value single image if the similarity comparison result shows that there is a difference in content; The alarm logging module is used to highlight and issue alarms for areas of difference, and also generates electronic reports.

[0046] The foregoing has provided a detailed description of one embodiment of the present invention, but this description is merely a preferred embodiment and should not be construed as limiting the scope of the invention. All equivalent variations and modifications made within the scope of the claims of this invention should still fall within the patent coverage of this invention.

Claims

1. A method for verifying relay protection setting data based on OCR recognition, characterized in that, The method includes: Acquire the image of the setting sheet and the image of the setting protection device, respectively; The target text region recognition image is obtained by performing text region recognition on the image of the set value protection device; Image enhancement is performed on the target text region recognition image to obtain an enhanced fixed value protection device image; The first set value data is obtained by performing OCR recognition on the image of the set value sheet, and the second set value data is obtained by performing OCR recognition on the image of the enhanced set value protection device. A similarity comparison is performed between the first setpoint data and the second setpoint data; If the similarity comparison result is that the content is consistent, then the image of the setting sheet and the image of the enhanced setting protection device are saved, and an electronic report is generated.

2. The method for verifying relay protection setting data based on OCR recognition according to claim 1, characterized in that, The text region recognition image obtained by performing text region recognition on the image of the setpoint protection device includes: The image of the fixed-value protection device is substituted into the first depth convolution module to obtain the first depth convolution feature; the image of the fixed-value protection device is substituted into the second depth convolution module to obtain the second depth convolution feature; the scale of the second depth convolution feature is half the scale of the image of the fixed-value protection device; the scale of the first depth convolution feature is half the scale of the second depth convolution feature. The first depthwise convolutional feature is subjected to 1×1 convolution, 3×3 convolution and pooling respectively to obtain the first convolutional feature, the second convolutional feature and the first pooling feature. The first convolutional feature, the second convolutional feature and the first pooling feature are concatenated to obtain the first concatenated feature. The second depthwise convolutional feature is subjected to 1×1 convolution, 3×3 convolution and pooling respectively to obtain the third convolutional feature, the fourth convolutional feature and the second pooling feature. The third convolutional feature, the fourth convolutional feature and the second pooling feature are concatenated to obtain the second concatenated feature. Perform a 3×3 convolution operation on the first spliced ​​feature and the second spliced ​​feature respectively to obtain the first spliced ​​convolution feature and the second spliced ​​convolution feature; The first concatenated convolutional feature is upsampled and then concatenated with the second concatenated convolutional feature to obtain the third concatenated feature; After upsampling the third splicing feature, a 1×1 convolution operation is performed to obtain the target text region recognition image.

3. The method for verifying relay protection setting data based on OCR recognition according to claim 1, characterized in that, Image enhancement of the target text region recognition image to obtain the enhanced fixed value protection device image includes: The target text region recognition image is sequentially substituted into two preset first convolutional blocks to obtain the first convolutional text feature; The target text region recognition image and the first convolutional text features are concatenated and then normalized to obtain enhanced sample features; The target text region recognition image is sequentially substituted into two preset second convolutional blocks and a degradation-guided modulation block to obtain the second convolutional text features; After fusing the enhanced sample features and the second convolutional text features, the third convolutional text features are obtained by sequentially substituting them into the preset second convolutional block and the degenerate guided modulation block. Substituting the third convolutional text features and the enhanced sample features into the deep enhancement model yields the enhanced fixed-value protection device image.

4. The method for verifying relay protection setting data based on OCR recognition according to claim 3, characterized in that, Substituting the third convolutional text features and the enhanced sample features into the deep enhancement model, the resulting enhanced fixed-value protection device image includes: The first fused feature is obtained by multiplying the third convolutional text feature and the enhanced sample feature element by the first fused feature and then inputting the second convolutional block into the preset second convolutional block. The enhanced sample features are substituted into the preset second convolutional block and then multiplied element-wise with the third convolutional text features to obtain the second fused feature; The second fusion feature is normalized and then fused with the first fusion feature to obtain the third fusion feature; After fusing the third fusion feature and the third convolutional text feature, a 1×1 convolution operation is performed to obtain the enhanced fixed value protection device image.

5. The method for verifying relay protection setting data based on OCR recognition according to claim 1, characterized in that, After performing a similarity comparison between the first setpoint data and the second setpoint data, the process further includes: If the similarity comparison result indicates that there are differences in content, then the difference region in the fixed value single image is determined; The differential areas are highlighted and an alarm is issued, and an electronic report is also generated.

6. A relay protection setting data verification device based on OCR recognition, characterized in that, The device includes: The image data acquisition module is used to acquire images of the setting sheet and the setting protection device, respectively. A text region determination module for a setpoint protection device is used to perform text region recognition on the image of the setpoint protection device to obtain a target text region recognition image. The text region enhancement module for the fixed value protection device is used to enhance the image of the target text region recognition image to obtain an enhanced fixed value protection device image. The fixed value data extraction module is used to perform OCR recognition on the fixed value single image to obtain first fixed value data, and to perform OCR recognition on the enhanced fixed value protection device image to obtain second fixed value data; The data intelligent comparison module is used to perform a similarity comparison between the first set value data and the second set value data; The fixed value data recording module is used to save the fixed value sheet image and the enhanced fixed value protection device image and generate an electronic report if the similarity comparison result is consistent.

7. A relay protection setting data verification device based on OCR recognition according to claim 6, characterized in that, The text region determination module of the setpoint protection device includes: A depthwise convolutional feature extraction module is used to substitute the image of the fixed-value protection device into a first depthwise convolutional module to obtain a first depthwise convolutional feature; and to substitute the image of the fixed-value protection device into a second depthwise convolutional module to obtain a second depthwise convolutional feature; the scale of the second depthwise convolutional feature is half the scale of the fixed-value protection device image; and the scale of the first depthwise convolutional feature is half the scale of the second depthwise convolutional feature. The first concatenated feature generation module is used to perform 1×1 convolution, 3×3 convolution and pooling processing on the first depthwise convolutional feature to obtain the first convolutional feature, the second convolutional feature and the first pooling feature, and to concatenate the first convolutional feature, the second convolutional feature and the first pooling feature to obtain the first concatenated feature. The second concatenation feature generation module is used to perform 1×1 convolution, 3×3 convolution and pooling processing on the second depthwise convolution feature to obtain the third convolution feature, the fourth convolution feature and the second pooling feature, and to concatenate the third convolution feature, the fourth convolution feature and the second pooling feature to obtain the second concatenation feature. The concatenated convolutional feature generation module is used to perform 3×3 convolution operations on the first concatenated feature and the second concatenated feature respectively to obtain the first concatenated convolutional feature and the second concatenated convolutional feature; The third splicing feature generation module is used to upsample the first splicing convolutional feature and then splice it with the second splicing convolutional feature to obtain the third splicing feature; The target text region recognition image generation module is used to perform a 1×1 convolution operation on the third splicing feature to obtain the target text region recognition image.

8. A relay protection setting data verification device based on OCR recognition according to claim 6, characterized in that, The text region enhancement module of the setpoint protection device includes: The first convolutional text feature generation module is used to sequentially substitute the target text region recognition image into two preset first convolutional blocks to obtain the first convolutional text features. An enhanced sample feature generation module is used to concatenate the target text region recognition image and the first convolutional text features and then normalize them to obtain enhanced sample features. The second convolutional text feature generation module is used to sequentially substitute the target text region recognition image into two preset second convolutional blocks and a degradation-guided modulation block to obtain the second convolutional text features. The third convolutional text feature generation module is used to fuse the enhanced sample features and the second convolutional text features, and then sequentially substitute them into the preset second convolutional block and the degenerate guided modulation block to obtain the third convolutional text features; An enhanced fixed-value protection device image generation module is used to substitute the third convolutional text features and the enhanced sample features into a deep enhancement model to obtain an enhanced fixed-value protection device image.

9. A relay protection setting data verification device based on OCR recognition according to claim 8, characterized in that, The enhanced setpoint protection device image generation module includes: The first fusion feature generation module is used to multiply the third convolutional text feature and the enhanced sample feature element by the element and then input the result into the preset second convolutional block to obtain the first fusion feature; The second fusion feature generation module is used to substitute the enhanced sample features into the preset second convolution block and then multiply them element-wise with the third convolution text features to obtain the second fusion feature; The third fusion feature generation module is used to normalize the second fusion feature and then fuse it with the first fusion feature to obtain the third fusion feature; An enhanced value protection device image determination module is used to perform a 1×1 convolution operation on the third fusion feature and the third convolutional text feature to obtain an enhanced value protection device image.

10. A relay protection setting data verification device based on OCR recognition according to claim 6, characterized in that, The device further includes: The difference region determination module is used to determine the difference region in the fixed value single image if the similarity comparison result shows that there is a difference in content; The alarm recording module is used to highlight the discrepancies and issue an alarm, and also generate an electronic report.

Images