Method for identifying nameplate text information of power equipment

Abstract

The invention discloses a method for identifying nameplate text information of power equipment, and relates to the technical field of image processing. The method comprises the steps of obtaining a multi-angle image of a target nameplate, and performing gray level conversion to obtain a plurality of gray level images; performing perspective transformation on the inclined grayscale image to obtain a front image; performing strong light detection on the front image to obtain a strong light area; according to the strong light area, fusing the plurality of front images to obtain a fused image; taking the fused image after binarization processing as input of a pre-trained text recognition model to obtain text information; and inputting the text information into a preset template to generate nameplate identification information. Through a series of preprocessing steps of gray level conversion, perspective transformation, strong light detection, image fusion, binarization processing and the like, the image quality is remarkably improved, the influence of strong light on the recognition effect is effectively reduced, the false recognition rate and the missed recognition rate are effectively reduced, and the recognition accuracy is improved.

Description

Technical Field

[0001] The invention relates to the technical field of image processing, and in particular to a method for recognizing nameplate text information of electric equipment. Background Art

[0002] With the rapid development of the power industry, the number of power equipment has increased dramatically, and the safety and stability of its operation are directly related to the overall performance of the power system and the quality of power supply. As the only identification of the equipment, the nameplate of the power equipment records the key information such as the equipment model, specifications, manufacturer, date of manufacture, rated capacity, rated voltage, etc., and is an important identification proof of the equipment's identity. In the daily management and maintenance of power equipment, inspection and verification is a vital task. By inspecting and verifying the equipment nameplate, the accuracy and completeness of the equipment information can be ensured, which helps the equipment management personnel to timely understand the operating status of the equipment, discover potential safety hazards, and take corresponding maintenance measures.

[0003] Patent CN113554026A discloses a method, device and electronic device for identifying nameplates of electric equipment, which realizes automatic recognition of nameplate text based on images, and solves the problems of harsh environment, high labor intensity and danger existing in manual inspections.

[0004] At present, the recognition of text information on the nameplate of power equipment often relies on a single image. However, the nameplate of power equipment is usually a metal nameplate, which is prone to strong light reflection covering the text, causing the text recognition result to be missing some content, resulting in low accuracy in the recognition of the nameplate text information. Summary of the invention

[0005] The purpose of the present invention is to solve the problem of low accuracy mentioned in the above background technology, and to propose a method for identifying text information on nameplates of electric equipment.

[0006] A first aspect of the present invention provides a method for identifying nameplate text information of an electric power device, the method comprising:

[0007] Acquire multi-angle images of the target nameplate and perform grayscale conversion to obtain multiple grayscale images;

[0008] Performing perspective transformation on a target grayscale image to obtain a first frontal image; the target grayscale image is any one of the multiple grayscale images;

[0009] Performing strong light detection on the first front image to obtain a strong light area;

[0010] According to the strong light area, a plurality of first front images are fused to obtain a first fused image;

[0011] performing binarization processing on the first fused image to obtain a second fused image;

[0012] Using the second fused image as an input of a pre-trained text recognition model to obtain text information;

[0013] The text information is entered into a preset template to generate nameplate identification information.

[0014] Optionally, the plurality of grayscale images include an image of the front side of the nameplate;

[0015] The performing perspective transformation on the target grayscale image to obtain the first frontal image comprises:

[0016] The target grayscale image is reduced by a preset multiple to obtain a thumbnail;

[0017] Converting the thumbnail into a binary image, and performing Gaussian filtering on the binary image to obtain a smoothed image;

[0018] Performing a probabilistic Hough transform on the smoothed image to obtain positions of four edge lines;

[0019] According to the positions of the four edge lines, the coordinates of the four intersection points are calculated;

[0020] Enlarging the coordinates of the four intersection points by the preset multiple value to obtain the coordinates of the first corner point;

[0021] Calculate the coordinates of the second corner point of the front image of the nameplate;

[0022] Constructing a transformation matrix according to the first corner point coordinates and the second corner point coordinates;

[0023] The target grayscale image is transformed into a first frontal image according to the transformation matrix.

[0024] Optionally, the performing strong light detection on the first front image to obtain a strong light area includes:

[0025] Establishing a square sliding window according to a preset window size, and sliding the window on the first front image with a preset step length to obtain a plurality of sub-regions;

[0026] Calculating the grayscale average and grayscale variance of a target subregion; the target subregion is any one of the multiple subregions;

[0027] If the grayscale average value is greater than a first preset threshold value, and the grayscale variance is less than a second preset threshold value, then the target sub-region is judged to be a strong photon region;

[0028] The strong photon regions with overlapping areas are merged to obtain a strong light region.

[0029] Optionally, the plurality of grayscale images include an image of the front side of the nameplate;

[0030] The fusing the plurality of first front images according to the strong light area to obtain a first fused image comprises:

[0031] Initialize a blank fused image;

[0032] Filling the non-high light area of ​​the nameplate front image into the blank fused image to obtain a temporary fused image;

[0033] Dividing the to-be-filled area of ​​the temporary fused image into a plurality of first sub-images;

[0034] For a target sub-image, extract a corresponding second sub-image from a plurality of first front images; the target sub-image is any one of the plurality of first sub-images;

[0035] Remove the second sub-image belonging to the bright light area to obtain a candidate sub-image;

[0036] Calculate the contrast of each candidate sub-image, and select the candidate sub-image with the highest contrast to fill in the target sub-image;

[0037] The filled temporary fused image is used as the first fused image.

[0038] Optionally, the binarizing the first fused image to obtain the second fused image includes:

[0039] obtaining a segmentation threshold of the first fused image according to the maximum inter-class variance method, and performing binarization processing on the first fused image according to the segmentation threshold to obtain a first enhanced image;

[0040] Using a Sobel operator to perform edge detection on the first fused image, and performing binarization processing on the first fused image according to a result of the edge detection to obtain a second enhanced image;

[0041] Performing an OR operation on the first enhanced image and the second enhanced image to obtain a second fused image.

[0042] Optionally, the text recognition model is composed of an EAST model and a CRNN model; the text information includes a plurality of text box positions and their corresponding text contents;

[0043] The step of using the second fused image as an input of a pre-trained text recognition model to obtain text information includes:

[0044] The second fused image is used as the input of the EAST model to obtain a plurality of text boxes; each text box has a position represented by four vertex coordinates;

[0045] cropping each text box area from the second fused image to obtain a plurality of text images;

[0046] Each text image is used as the input of the CRNN model to obtain the text content.

[0047] Optionally, the text information includes multiple text box positions and their corresponding text contents;

[0048] Entering the text information into a preset template to generate nameplate identification information includes:

[0049] Determine the coverage area of ​​multiple attribute content text boxes according to preset structural information;

[0050] Matching a target text box position with a plurality of coverage areas to determine the attributes of a target text content; the target text box position is any one of a plurality of text box positions; the target text content is the text content corresponding to the target text box position;

[0051] According to the attributes of the target text content, the target text content is filled into the corresponding attribute content of the preset template to obtain the nameplate identification information.

[0052] Beneficial effects of the present invention:

[0053] The present invention proposes a method for identifying text information on nameplates of electric equipment, the method comprising: acquiring multi-angle images of a target nameplate, and performing grayscale conversion to obtain multiple grayscale images; performing perspective transformation on the target grayscale image to obtain a first front image; the target grayscale image is any one of the multiple grayscale images; performing strong light detection on the first front image to obtain a strong light area; according to the strong light area, fusing multiple first front images to obtain a first fused image; performing binarization processing on the first fused image to obtain a second fused image; using the second fused image as input of a pre-trained text recognition model to obtain text information; entering the text information into a preset template to generate nameplate recognition information.

[0054] Through a series of preprocessing steps such as grayscale conversion, perspective transformation, strong light detection, image fusion and binarization processing, the image quality is significantly improved, the influence of strong light on the recognition effect is effectively reduced, the false recognition rate and missed recognition rate are effectively reduced, and the recognition accuracy is improved. BRIEF DESCRIPTION OF THE DRAWINGS

[0055] The present invention will be further described below in conjunction with the accompanying drawings.

[0056] Figure 1 A flowchart of a method for identifying nameplate text information of electric equipment is provided for an embodiment of the present invention. DETAILED DESCRIPTION

[0057] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0058] The embodiment of the present invention provides a method for identifying nameplate text information of electric equipment. Figure 1 , Figure 1 A flowchart of a method for identifying nameplate text information of an electric equipment provided by an embodiment of the present invention. The method comprises the following steps:

[0059] S101, acquiring multi-angle images of a target nameplate and performing grayscale conversion to obtain multiple grayscale images.

[0060] S102, performing perspective transformation on the target grayscale image to obtain a first frontal image.

[0061] S103, performing strong light detection on the first front image to obtain a strong light area.

[0062] S104: fuse the multiple first front images according to the strong light area to obtain a first fused image.

[0063] S105: binarize the first fused image to obtain a second fused image.

[0064] S106: Use the second fused image as an input of a pre-trained text recognition model to obtain text information.

[0065] S107, entering the text information into a preset template to generate nameplate identification information.

[0066] The target grayscale image is any one of the multiple grayscale images.

[0067] A method for identifying text information on nameplates of electric equipment provided in an embodiment of the present invention significantly improves the quality of images through a series of preprocessing steps such as grayscale conversion, perspective transformation, strong light detection, image fusion and binarization processing, effectively reduces the influence of strong light on the recognition effect, effectively reduces the false recognition rate and missed recognition rate, and improves the recognition accuracy.

[0068] In one embodiment, the plurality of grayscale images include an image of the front face of the nameplate;

[0069] Step S102 includes:

[0070] Step 1: Reduce the target grayscale image by a preset multiple to obtain a thumbnail.

[0071] Step 2: Convert the thumbnail into a binary image and perform Gaussian filtering on the binary image to obtain a smoothed image.

[0072] Step 3: Perform probabilistic Hough transform on the smoothed image to obtain the positions of the four edge lines.

[0073] Step 4: Calculate the coordinates of the four intersection points based on the positions of the four edge lines.

[0074] Step 5: Enlarge the coordinates of the four intersection points by a preset multiple to obtain the coordinates of the first corner point.

[0075] Step six, calculate the coordinates of the second corner point of the nameplate front image.

[0076] Step seven, construct a transformation matrix based on the coordinates of the first corner point and the coordinates of the second corner point.

[0077] Step eight, transforming the target grayscale image into a first frontal image according to the transformation matrix.

[0078] In one implementation, by reducing the image size to generate a thumbnail and retaining key structural information, the amount of calculation for line detection can be reduced and the processing speed can be improved. The coordinates in the thumbnail are then enlarged to the original image size to ensure the accuracy and consistency of the corner point position.

[0079] In one implementation, the image at the oblique angle is registered with the front image to provide a data basis for subsequent image fusion.

[0080] In one embodiment, step S103 includes:

[0081] Step 1: A square sliding window is established according to a preset window size, and is slid on the first front image with a preset step length to obtain a plurality of sub-regions.

[0082] Step 2: Calculate the grayscale mean and grayscale variance of the target sub-region.

[0083] Step three: if the grayscale average value is greater than the first preset threshold value and the grayscale variance is less than the second preset threshold value, the target sub-region is judged to be a strong photon region.

[0084] Step 4: merge the strong photon regions with overlapping areas to obtain a strong light region.

[0085] The target sub-region is any one of the multiple sub-regions.

[0086] In one implementation, by establishing a square sliding window and sliding it on the image with a preset step size, each part of the image can be checked step by step to effectively locate the bright light area.

[0087] In one implementation, the bright light area refers to the area where the text is blocked by reflected light. The first preset threshold is set to ensure that only areas with higher grayscale values ​​are judged as bright light areas, reducing the misjudgment of non-bright light areas. The second preset threshold is set to ensure that only areas with smaller grayscale value changes are judged as bright light areas, excluding areas with high light intensity but where the text is not covered by reflections and is still clear. The setting of dual thresholds improves the accuracy and robustness of bright light detection.

[0088] In one embodiment, the plurality of grayscale images include an image of the front face of the nameplate;

[0089] Step S104 includes:

[0090] Step 1: Initialize the blank fusion image.

[0091] Step 2: fill the non-bright area of ​​the nameplate front image into the blank fused image to obtain a temporary fused image.

[0092] Step three: divide the area to be filled of the temporary fused image into a plurality of first sub-images.

[0093] Step 4: for the target sub-image, extract the corresponding second sub-image from the multiple first front images.

[0094] Step 5: remove the second sub-image belonging to the bright light area to obtain a candidate sub-image.

[0095] Step 6: Calculate the contrast of each candidate sub-image, and select the candidate sub-image with the highest contrast to fill in the target sub-image.

[0096] Step seven: taking the filled temporary fused image as the first fused image.

[0097] The target subgraph is any one of the multiple first subgraphs.

[0098] In one implementation, the non-high light area is first filled into the blank fused image, effectively retaining the part of the original image that is not affected by the high light, reducing the interference of the high light on the image quality. The contrast of the candidate sub-images is calculated, and the sub-image with the highest contrast is selected for filling, ensuring that each part of the fused image has the best clarity and contrast, improving the visual effect and detail expression of the image.

[0099] In one embodiment, step S105 includes:

[0100] Step 1: According to the maximum inter-class variance method, a segmentation threshold of the first fused image is obtained, and the first fused image is binarized according to the segmentation threshold to obtain a first enhanced image.

[0101] Step 2: Use the Sobel operator to perform edge detection on the first fused image, and perform binarization processing on the first fused image according to the result of the edge detection to obtain a second enhanced image.

[0102] Step three, performing an OR operation on the first enhanced image and the second enhanced image to obtain a second fused image.

[0103] In one implementation, the maximum inter-class variance method automatically determines the optimal threshold, so that the method can adapt to the grayscale distribution of different images, enhancing the robustness and adaptability of the processing. Edge detection using the Sobel operator can effectively identify edge information in the image, supplement details that may be missed by grayscale threshold segmentation alone, and improve the precision of the binary processing. The comprehensively processed image can reduce noise and unnecessary detail interference, improve the contrast and clarity of the text area, and improve the accuracy of text recognition.

[0104] In one embodiment, the text recognition model is composed of an EAST (Efficient and Accurate Scene Text Detector) model and a CRNN (Convolutional Recurrent Neural Network) model; the text information includes multiple text box positions and their corresponding text contents;

[0105] Step S106 includes:

[0106] Step 1: Use the second fused image as the input of the EAST model to obtain multiple text boxes.

[0107] Step 2: crop each text box area from the second fused image to obtain multiple text images.

[0108] Step 3: Use each text image as the input of the CRNN model to obtain the text content.

[0109] Each text box has four vertex coordinates that represent the position of the text box.

[0110] In one implementation, the EAST model can quickly and efficiently detect text areas in images, generate accurate text box locations, and improve the accuracy and robustness of text detection. The CRNN model is good at processing sequence data and can accurately identify text content in detected text boxes, improving the accuracy of text recognition.

[0111] In one implementation, each text box area is cropped from the image to ensure that the input image of the CRNN model only contains the target text area, reduce background interference, and improve recognition effect.

[0112] In one embodiment, the text information includes a plurality of text box positions and their corresponding text contents;

[0113] Step S107 includes:

[0114] Step 1: Determine the coverage area of ​​multiple attribute content text boxes according to preset structural information.

[0115] Step 2: Match the target text box position with multiple coverage areas to determine the attributes of the target text content.

[0116] Step three, according to the attributes of the target text content, fill the target text content into the corresponding attribute content of the preset template to obtain the nameplate identification information.

[0117] The target text box position is any one of the multiple text box positions; the target text content is the text content corresponding to the target text box position.

[0118] In one implementation, the target text box position is matched with a preset coverage area, which can accurately determine the attributes of each text content, ensure accurate information classification, and avoid missing or incorrect information classification.

[0119] The above is a detailed description of an embodiment of the present invention, but the content is only a preferred embodiment of the present invention and cannot be considered to limit the scope of implementation of the present invention. All equivalent changes and improvements made within the scope of the present invention should still fall within the scope of the patent coverage of the present invention.

Claims

1. A method for identifying nameplate text information of electric equipment, characterized in that: The method comprises: Acquire multi-angle images of the target nameplate and perform grayscale conversion to obtain multiple grayscale images; Performing perspective transformation on a target grayscale image to obtain a first frontal image; the target grayscale image is any one of the multiple grayscale images; Performing strong light detection on the first front image to obtain a strong light area; According to the strong light area, a plurality of first front images are fused to obtain a first fused image; performing binarization processing on the first fused image to obtain a second fused image; Using the second fused image as an input of a pre-trained text recognition model to obtain text information; The text information is entered into a preset template to generate nameplate identification information.

2. A method for identifying nameplate text information of electric equipment according to claim 1, characterized in that: The multiple grayscale images include an image of the front side of the nameplate; The performing perspective transformation on the target grayscale image to obtain the first frontal image comprises: The target grayscale image is reduced by a preset multiple to obtain a thumbnail; Converting the thumbnail into a binary image, and performing Gaussian filtering on the binary image to obtain a smoothed image; Performing a probabilistic Hough transform on the smoothed image to obtain positions of four edge lines; According to the positions of the four edge lines, the coordinates of the four intersection points are calculated; Enlarging the coordinates of the four intersection points by the preset multiple value to obtain the coordinates of the first corner point; Calculate the coordinates of the second corner point of the front image of the nameplate; Constructing a transformation matrix according to the first corner point coordinates and the second corner point coordinates; The target grayscale image is transformed into a first frontal image according to the transformation matrix.

3. A method for identifying nameplate text information of electric equipment according to claim 1, characterized in that: The performing of strong light detection on the first front image to obtain the strong light area includes: Establishing a square sliding window according to a preset window size, and sliding the window on the first front image with a preset step length to obtain a plurality of sub-regions; Calculating the grayscale average and grayscale variance of a target subregion; the target subregion is any one of the multiple subregions; If the grayscale average value is greater than a first preset threshold value, and the grayscale variance is less than a second preset threshold value, then the target sub-region is judged to be a strong photon region; The strong photon regions with overlapping areas are merged to obtain a strong light region.

4. A method for identifying nameplate text information of electric equipment according to claim 1, characterized in that: The multiple grayscale images include an image of the front side of the nameplate; The fusing the plurality of first front images according to the strong light area to obtain a first fused image comprises: Initialize a blank fused image; Filling the non-high light area of ​​the nameplate front image into the blank fused image to obtain a temporary fused image; Dividing the to-be-filled area of ​​the temporary fused image into a plurality of first sub-images; For a target sub-image, extract a corresponding second sub-image from a plurality of first front images; the target sub-image is any one of the plurality of first sub-images; Remove the second sub-image belonging to the bright light area to obtain a candidate sub-image; Calculate the contrast of each candidate sub-image, and select the candidate sub-image with the highest contrast to fill in the target sub-image; The filled temporary fused image is used as the first fused image.

5. The method for identifying nameplate text information of electric equipment according to claim 1, characterized in that: The binarization processing of the first fused image to obtain the second fused image comprises: obtaining a segmentation threshold of the first fused image according to the maximum inter-class variance method, and performing binarization processing on the first fused image according to the segmentation threshold to obtain a first enhanced image; Using a Sobel operator to perform edge detection on the first fused image, and performing binarization processing on the first fused image according to a result of the edge detection to obtain a second enhanced image; Performing an OR operation on the first enhanced image and the second enhanced image to obtain a second fused image.

6. A method for identifying nameplate text information of electric equipment according to claim 1, characterized in that: The text recognition model is composed of an EAST model and a CRNN model; the text information includes a plurality of text box positions and their corresponding text contents; The step of using the second fused image as an input of a pre-trained text recognition model to obtain text information includes: The second fused image is used as the input of the EAST model to obtain a plurality of text boxes; each text box has four vertex coordinates describing the position of the text box; cropping each text box area from the second fused image to obtain a plurality of text images; Each text image is used as the input of the CRNN model to obtain the text content.

7. A method for identifying nameplate text information of electric power equipment according to claim 1, characterized in that: The text information includes multiple text box positions and their corresponding text contents; Entering the text information into a preset template to generate nameplate identification information includes: Determine the coverage area of ​​multiple attribute content text boxes according to preset structural information; Matching a target text box position with a plurality of coverage areas to determine the attributes of a target text content; the target text box position is any one of a plurality of text box positions; the target text content is the text content corresponding to the target text box position; According to the attributes of the target text content, the target text content is filled into the corresponding attribute content of the preset template to obtain the nameplate identification information.

Images