Substation meter type and measuring range identification method based on yov8 classification

Abstract

The invention discloses a transformer substation meter type and range identification method based on yolov8 classification. The method comprises the following steps: obtaining a transformer substation site picture and inputting the transformer substation site picture into yolov8 for instrument detection to obtain an instrument type; when the instrument is detected to be the concerned instrument type, intercepting a rectangular area of the instrument and obtaining a large class name to which the instrument belongs; inputting an instrument area screenshot into the trained yov8m-cls classification model for instrument classification, and obtaining names of instrument subdivision types and corresponding measuring range information; when the name of the instrument subdivision type classified and identified by the yolov8m-cls belongs to the subclass of the instrument large class name detected by the yolov8, outputting a classification result and corresponding range information; otherwise, the classification result of the yolov8m-cls belongs to a false alarm, and the classification result of the yolov8m-cls is eliminated. According to the invention, an image classification algorithm is combined with priori knowledge to finely classify different types of instruments according to different appearance and range information, and the subdivision category and range information of the instruments are obtained at the same time.

Description

Technical Field

[0001] The present invention relates to the technical field of instrument identification, and in particular to a method for identifying types and ranges of substation meters based on yolov8 classification. Background Art

[0002] Substations typically use a large number of various instruments to record the operating status of other related equipment. To ensure timely and accurate understanding of the substation's overall operational status, it is necessary to regularly record the data from these instruments, including their types. However, due to the complex environment within substations and the large number and variety of instruments, manual recording of instrument data was previously time-consuming and dangerous. Deep learning-based object detection or optical character recognition (OCR) methods can now automatically identify instrument readings. However, subtle differences in appearance and range information between instruments of the same type can lead to a high incidence of misidentification using deep learning-based instrument reading methods. Summary of the Invention

[0003] The purpose of the present invention is to provide a method for identifying substation meter types and ranges based on yolov8 classification, thereby achieving both instrument classification and range identification.

[0004] The technical solution adopted in the present invention is: The method for identifying substation meter types and ranges based on yolov8 classification includes the following steps: Step 1: Get a picture of the substation site and input yolov8 to perform instrument detection; Step 2: Determine whether Yolov8 detects an instrument; if so, output the type of instrument detected and execute step 3; otherwise, execute step 1; Step 3: Determine whether the detected instrument is of the instrument type of interest; if so, intercept the rectangular area of ​​the instrument and obtain the name of the major category to which the instrument belongs and execute step 4; otherwise, execute step 1; Step 4: Take a screenshot of the instrument area and input the trained yolov8m-cls classification model to classify the instrument and obtain the name of the instrument subdivision type and the corresponding range information; Step 5, determine whether the name of the instrument subdivision type identified by yolov8m-cls classification belongs to the subcategory of the instrument category name detected by yolov8; if so, output the classification result and the corresponding range information; otherwise, it is a false alarm and the classification result of yolov8m-cls is eliminated.

[0005] Furthermore, the substation site image data in step 1 is preprocessed to obtain a preprocessed image. Specifically, the acquired instrument image data is brightness enhanced to obtain enhanced image data; the obtained enhanced image data is adjusted to a fixed format size, wherein the enhanced image data is cropped during the adjustment process; the brightness enhancement is not mechanical brightness enhancement, but rather is to increase the brightness to the equivalent brightness of the image under normal lighting, avoiding the problem of over-brightening the image.

[0006] Furthermore, in step 2, the type of the corresponding instrument is obtained according to the position data of the instrument pointer, the dial center, the range zero value, and the maximum range value; Specifically, based on the pre-processed image, feature acquisition is performed on the instrument, where the feature acquisition includes position data of the instrument pointer, dial center, range zero value, and maximum range value; Based on the collected feature candidate frames, the frames are screened to obtain instrument frames; the obtained instrument frames are integrated with the feature data to obtain pending image areas and classification results corresponding to each pending area; the image area corresponding to the classification result that meets the preset conditions is determined as the instrument image, and the dial type corresponding to the determined instrument image is the target dial type of the instrument image; wherein, due to the accuracy of instrument reading recognition and the safety of factory instruments, the reading operation of pointer-type instruments is often performed by a large number of professional staff. In order to improve work efficiency and reduce work intensity, the present invention cooperates with the fixed-point high-definition shooting of the industrial inspection robot to achieve the function of automatically identifying the instrument reading. The captured pointer-type instrument image contains image information of the instrument pointer, dial center, zero value, and maximum range value.

[0007] Furthermore, the training method of the yolov8m-cls classification model specifically includes the following steps: Step 4-1: Acquire instrument image data, organize and annotate the acquired instrument image data, and divide it into a training set, a test set, and a validation set; Step 4-2, build the YOLOv8M-CLS network, and use the training set and test set to obtain the YOLOv8M-CLS classification model for instrument classification.

[0008] Preferably, the training parameters of the classification model are set, including: the batch size is set to 8 and the number of epochs is set to 3000 times.

[0009] In step 4-3, the recognition performance of the yolov8m-cls classification model is verified using the images in the validation set. When the validation performance meets the customer's requirements, the final trained yolov8m-cls classification model is obtained.

[0010] Furthermore, in step 5, the corresponding instrument subdivision type table is queried to determine whether the name of the table subdivision type belongs to a subcategory of the instrument category name detected by yolov8.

[0011] This invention adopts the above technical solution. By collecting visual images from the substation environment, the yolov8 classification algorithm is used to classify several common instrument categories in the substation scene in more detail based on their appearance and range. The range information of the corresponding instrument is further obtained based on prior knowledge, assisting other instrument reading recognition methods to complete the reading recognition of substation field instruments. This method can be applied to fields such as substation automation monitoring and industrial automation. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] The present invention will be further described in detail below with reference to the accompanying drawings and specific embodiments; Figure 1 Schematic diagram of the structure of the substation meter type and range identification method based on yolov8 classification of the present invention; Figure 2 A schematic diagram of a table showing the names of 157 types of instruments involved in the present invention; Figure 3 The following is a schematic diagram of the appearance of some instruments of the present invention. DETAILED DESCRIPTION

[0013] In order to make the purpose, technical solutions and advantages of the embodiments of the present application clearer, the technical solutions in the embodiments of the present application will be clearly and completely described below in conjunction with the drawings in the embodiments of the present application.

[0014] like Figures 1 to 3 As shown in FIG1 , the present invention discloses a method for identifying the type and range of substation meters based on YOLOv8 classification, which is mainly divided into three parts: image information acquisition and training of the YOLOv8M-CLS classification model, cropping the meter area based on the meter position output by YOLOv8, and using the YOLOv8M-CLS model to classify the cropped meter image and provide the corresponding meter range information. The present invention discloses a method for identifying the type and range of substation meters based on YOLOv8 classification, which specifically includes the following steps: Step 1: Get a picture of the substation site and input it into yolov8 for instrument detection. All yolov8 models in the present invention are yolov8 models for instrument detection obtained by training based on existing mature technologies. They are mature applications of existing technologies and are not the innovation of the present invention. They will not be described here.

[0015] Step 2: Determine whether Yolov8 detects an instrument; if so, output the type of instrument detected and execute step 3; otherwise, execute step 1; Step 3: Determine whether the detected instrument is of the instrument type of interest; if so, intercept the rectangular area of ​​the instrument and obtain the name of the major category to which the instrument belongs and execute step 4; otherwise, execute step 1; Step 4: Take a screenshot of the instrument area and input the trained yolov8m-cls classification model to classify the instrument and obtain the name of the instrument subdivision type and the corresponding range information; Step 5, determine whether the name of the instrument subdivision type identified by yolov8m-cls classification belongs to the subcategory of the instrument category name detected by yolov8; if so, output the classification result and the corresponding range information; otherwise, it is a false alarm and the classification result of yolov8m-cls is eliminated.

[0016] Furthermore, the substation site image data in step 1 is preprocessed to obtain a preprocessed image. Specifically, the acquired instrument image data is brightness enhanced to obtain enhanced image data; the obtained enhanced image data is adjusted to a fixed format size, wherein the enhanced image data is cropped during the adjustment process; the brightness enhancement is not mechanical brightness enhancement, but rather is to increase the brightness to the equivalent brightness of the image under normal lighting, avoiding the problem of over-brightening the image.

[0017] Furthermore, in step 2, the type of the corresponding instrument is obtained according to the position data of the instrument pointer, the dial center, the range zero value, and the maximum range value; Specifically, based on the pre-processed image, feature acquisition is performed on the instrument, where the feature acquisition includes position data of the instrument pointer, dial center, range zero value, and maximum range value; Based on the collected feature candidate frames, the frames are screened to obtain instrument frames; the obtained instrument frames are integrated with the feature data to obtain pending image areas and classification results corresponding to each pending area; the image area corresponding to the classification result that meets the preset conditions is determined as the instrument image, and the dial type corresponding to the determined instrument image is the target dial type of the instrument image; wherein, due to the accuracy of instrument reading recognition and the safety of factory instruments, the reading operation of pointer-type instruments is often performed by a large number of professional staff. In order to improve work efficiency and reduce work intensity, the present invention cooperates with the fixed-point high-definition shooting of the industrial inspection robot to achieve the function of automatically identifying the instrument reading. The captured pointer-type instrument image contains image information of the instrument pointer, dial center, zero value, and maximum range value.

[0018] Furthermore, the training method of the yolov8m-cls classification model specifically includes the following steps: Step 4-1: Acquire instrument image data, organize and annotate the acquired instrument image data, and divide it into a training set, a test set, and a validation set; Step 4-2, build the YOLOv8M-CLS network, and use the training set and test set to obtain the YOLOv8M-CLS classification model for instrument classification.

[0019] Preferably, the training parameters of the classification model are set, including: the batch size is set to 8 and the number of epochs is set to 3000 times.

[0020] In step 4-3, the recognition performance of the yolov8m-cls classification model is verified using the images in the validation set. When the validation performance meets the customer's requirements, the final trained yolov8m-cls classification model is obtained.

[0021] Furthermore, in step 5, the corresponding instrument subdivision type table is queried to determine whether the name of the table subdivision type belongs to a subcategory of the instrument category name detected by yolov8.

[0022] The present invention can simultaneously identify the subdivision type and range information of the instrument; using the yolov8m-cls deep learning classification model, the algorithm has high running speed and classification accuracy; it can cover most common instrument types in substation scenarios, involving a total of 5 common instrument categories and 157 subcategories. The subcategory names of all instruments are as follows Figure 2 The codes for obtaining the corresponding range information of some instruments and the screenshots of the appearance of some instruments are shown in Figure 3 shown.

[0023] The present invention has the following specific technical features: (1) Unlike the method that only realizes instrument classification or range extraction, the present invention uses image classification algorithm combined with prior knowledge to classify different types of instruments according to their appearance and range information, and obtains the subdivision category and range information of the instruments at the same time; (2) Unlike the existing method that realizes classification by instrument categories based on target detection, the present invention uses yolov8m-cls classification algorithm to classify instruments more accurately; (3) Although the existing method of classifying instruments based on fine-grained networks also realizes instrument classification, it does not realize outputting the range information of the corresponding instruments at the same time, and the types of meters involved are relatively few. The present invention realizes accurate classification of 157 subcategories of five common meter categories (leakage current meter, oil temperature meter, oil level meter, three-phase current meter, SF6 leakage current meter), and can output the range information corresponding to each subcategory; (4) The present invention first uses yolov8 target detection algorithm to detect the position of the instrument according to the instrument categories mentioned in (3) from the picture of the substation scene, and then crops the rectangular frame area of ​​the instrument to obtain a screenshot of the instrument. Due to the characteristics of the yolov8-cls algorithm, the present invention does not need to resize the intercepted instrument image and can directly use yolov8-cls for classification.

[0024] Obviously, the described embodiments are part of the embodiments of the present application, rather than all of the embodiments. In the absence of conflict, the embodiments in the present application and the features in the embodiments can be combined with each other. The components of the embodiments of the present application generally described and shown in the drawings here can be arranged and designed in various different configurations. Therefore, the detailed description of the embodiments of the present application is not intended to limit the scope of the application for protection, but merely represents the selected embodiments of the present application. Based on the embodiments in the present application, all other embodiments obtained by ordinary technicians in this field without making creative work are within the scope of protection of this application.

Claims

1. A method for identifying substation meter types and ranges based on yolov8 classification, characterized by: It includes the following steps: Step 1: Get a picture of the substation site and input yolov8 to perform instrument detection; Step 2: Determine whether Yolov8 detects the instrument; If yes, output the detected instrument type and execute step 3; Otherwise, go to step 1; Step 3: Determine whether the detected instrument is of the instrument type of interest; If yes, then intercept the rectangular area of ​​the instrument and obtain the name of the major category to which the instrument belongs and execute step 4; Otherwise, go to step 1; Step 4: Take a screenshot of the instrument area and input the trained yolov8m-cls classification model to classify the instrument and obtain the name of the instrument subdivision type and the corresponding range information; Step 5, judging whether the name of the instrument subdivision type identified by yolov8m-cls classification belongs to the subcategory of the instrument category name detected by yolov8; If yes, the classification result and the corresponding range information are output; otherwise, it is a false alarm and the classification result of yolov8m-cls is eliminated.

2. The method for identifying substation meter types and ranges based on yolov8 classification according to claim 1 is characterized in that: In step 1, the substation scene picture is preprocessed to obtain a preprocessed image, that is, the brightness of the acquired instrument image data is enhanced to obtain enhanced image data.

3. The method for identifying substation meter types and ranges based on yolov8 classification according to claim 1 is characterized in that: In step 2, the type of the corresponding instrument is obtained based on the position data of the instrument pointer, the dial center, the range zero value, and the maximum range value.

4. The method for identifying substation meter types and ranges based on yolov8 classification according to claim 1 is characterized in that: The training method of the yolov8m-cls classification model specifically includes the following steps: Step 4-1: Acquire instrument image data, organize and annotate the acquired instrument image data, and divide it into a training set, a test set, and a validation set; Step 4-2, construct the YOLOv8M-CLS network, and use the training set and test set to obtain the YOLOv8M-CLS classification model for instrument classification. In step 4-3, the recognition performance of the yolov8m-cls classification model is verified using the images in the validation set. When the validation performance meets the customer's requirements, the final trained yolov8m-cls classification model is obtained.

5. The method for identifying substation meter types and ranges based on yolov8 classification according to claim 4, characterized in that: In step 4-2, the training parameters of the classification model are set, including: batch size is set to 8 and the number of epochs is set to 3000.

6. The method for identifying substation meter types and ranges based on yolov8 classification according to claim 1 is characterized in that: In step 5, the corresponding instrument subdivision type table is queried to determine whether the name of the table subdivision type belongs to the subcategory of the instrument category name detected by yolov8.

Images