Oil level gauge reading recognition method and system and computer equipment

Abstract

The embodiment of the invention relates to the technical field of intelligent patrol and maintenance of transformer substations, and provides an oil level gauge reading recognition method and system and computer equipment. According to the method, on the basis of an improved lightweight yolov8 network structure, two-stage detection models are innovatively combined, oil level gauge detection and oil level detection are recognized and detected step by step, generalization is high, the oil level gauge reading recognition precision is greatly improved, and the problems that a traditional image processing method is low in generalization, prone to being affected by the environment and poor in recognition accuracy are solved. And the identification precision is low.

Description

Technical Field

[0001] The present invention relates to the technical field of intelligent patrol and maintenance of transformer substations, and in particular to an oil level gauge reading recognition method, system and computer equipment. Background Art

[0002] This section is intended to provide a background or context to the embodiments of the invention recited in the claims. No admission is made that the description herein is prior art by inclusion in this section.

[0003] In conventional substations, oil filling equipment occupies a dominant position. At present, the oil level indication devices of oil filling equipment are mainly oil level windows and mechanical oil level gauges. If an abnormal oil level problem occurs in the oil filling equipment during operation, it will seriously affect the oil level judgment and greatly affect the safe operation of the equipment and the power grid. Therefore, it is necessary to detect the oil level in time to avoid affecting the safe operation of the power grid and equipment. Summary of the invention

[0004] In order to realize oil level detection of oil level gauges in substations, the present invention proposes an oil level gauge reading recognition method, system and computer equipment, which not only have high generalization but also greatly improve the accuracy of oil level gauge reading recognition.

[0005] In order to achieve the above object, the present invention adopts the following technical solution.

[0006] First, on the one hand, the present invention provides a method for identifying an oil level gauge reading, comprising: obtaining oil level gauge device image data; inputting the oil level gauge device image data into a first model, the first model identifying the oil level gauge in the oil level gauge device image, and obtaining an oil level gauge image after cropping; inputting the oil level gauge image into a second model, the second model identifying the oil level window and oil level area in the oil level gauge image and their coordinates; calculating the oil level reading by post-processing the coordinates of the oil level window and the oil level area; and outputting the oil level detection result. The present invention innovatively combines two-stage detection models, which not only has high generalization, but also greatly improves the accuracy of oil level gauge reading recognition, solving the problems of low generalization, easy environmental impact, and low recognition accuracy of traditional image processing methods.

[0007] In some embodiments, the oil level gauge reading recognition method further includes: constructing and training the first model, obtaining a first oil level gauge device image, the first oil level gauge device image is a captured image of a device containing an oil level gauge; marking the circumscribed rectangular frame of the oil level gauge in the first oil level gauge device image; generating multiple first oil level gauge images by data enhancement as a training set for the first model, training the first model, the data enhancement including one or more of random cropping, rotation, translation, chromaticity, and brightness; and constructing and training the second model, obtaining a second oil level gauge image; using the second oil level gauge image to obtain image data of different oil levels by an enhancement algorithm as a training set for the second model, training the second model; wherein the second oil level gauge image is the first oil level gauge image or an oil level gauge image cropped from the captured oil level gauge device image. By performing data enhancement on the cropped oil level gauge image from the oil level gauge device image, the data set for the first model training is enriched, the cost of oil level gauge image data acquisition is reduced, and the problem of difficulty in data acquisition of different readings of the oil level gauge is solved. The oil level gauge image output by the first model is used as a training set for the second model, which further improves the recognition ability and accuracy of the model.

[0008] In some embodiments, the enhancement algorithm includes: reading the second oil level gauge image and / or its size; obtaining the oil level area coordinates and the oil level window coordinates of the second oil level gauge image, randomly generating a filling ratio value R, and adjusting the area of ​​the oil level area and the area of ​​the oil level window in the second oil level gauge image according to the filling ratio value R to obtain images of different oil levels. The present invention adopts the RSP-oil level data enhancement algorithm, combines the characteristics of the oil level gauge itself, identifies the oil level window and the oil level area in the oil level gauge image, and adjusts the areas of the two by different filling ratios to obtain images of different oil level heights, enriches the data of different oil level readings of the oil level gauge to the greatest extent, reduces the cost of oil level gauge image data collection, and solves the problem of data collection difficulties of different oil level gauge readings; while greatly enriching the training set of the second model, the second model can accurately identify oil levels of different heights.

[0009] Furthermore, the filling ratio value R≤0.6, and the adjusting the area of ​​the oil level region and the area of ​​the oil level window in the second oil level gauge image according to the filling ratio value R to obtain images with different oil levels includes: when 0.25≤R≤0.35, moving the oil level region in the second oil level gauge image downward, and filling the pixels of the oil level region with the pixels of the oil level window frame, thereby reducing the oil level region and obtaining an image in which the oil level is lower than the oil level in the second oil level gauge image; when R≥0.35, filling the pixels of the oil level window frame with the pixels of the oil level region to obtain an image in which the oil level is higher than the oil level in the second oil level gauge image.

[0010] In some embodiments of the present invention, the first model and / or the second model adopts a lightweight yolov8 network, and the lightweight yolov8 network uses multiple 1×1 convolutional layers and 3×3 convolutional layers to replace the original 5×5 convolutional layers.

[0011] In some embodiments, the lightweight yolov8 network includes a lightweight backbone network and a detection head part, wherein the first layer of the lightweight backbone network is a CBH structure, the third and fourth layers are 1×1 convolution layers without adding a SE attention mechanism and 3×3 depth-separable convolution layers without adding a SE attention mechanism, respectively, the second, fifth, sixth, seventh, eighth, ninth, tenth, eleventh, and twelfth layers are inverted convolution module layers with an SE attention mechanism, the thirteenth layer is an SPPF structure, the fourteenth and seventeenth layers of the detection head part are upsampling layers, the fifteenth, eighteenth, twenty-third layers are CBL structures, and the sixteenth, nineteenth, twenty-second, and twenty-fifth layers are C2f structures.

[0012] In the following embodiment of the present invention, the calculation of the oil level reading by post-processing the coordinates of the oil level window and the oil level area includes: calculating the height h of the oil level area and the height H of the oil level window respectively by the coordinates of the oil level window and the oil level area, and then obtaining the oil level reading by the ratio ρ of the two; wherein ρ=h / H.

[0013] On the other hand, the present invention also provides an oil level gauge reading recognition system for implementing the above method, the system includes an image acquisition module, an image reasoning module, a post-processing module and an output module, the image acquisition module is used to acquire image information of the oil level gauge equipment, the image reasoning module includes a first model and a second model, which are respectively used to identify the oil level gauge image in the oil level gauge equipment image and to identify the oil level window and oil level area in the oil level gauge image, the post-processing module is used to calculate the oil level reading by post-processing the coordinates of the oil level window and the oil level area, and the output module is used to output the oil level detection result according to the oil level.

[0014] Furthermore, the image acquisition module, image reasoning module, post-processing module and / or output module are deployed on a mobile robot. The mobile inspection robot can complete the inspection of substation oil level gauges and other equipment in one stop, realizing high-precision, efficient and intelligent modern intelligent substation operation and maintenance. First of all, the use of mobile robots for inspection can protect the safety of personnel, avoid personnel climbing up and down on substation equipment, and alleviate the tedious and inefficient problems of manual inspection work. Moreover, it is more flexible than installing the oil level gauge reading recognition sensor equipment. The oil level gauge reading sensor can only act on the oil level gauge; secondly, the mobile robot platform is more flexible than the fixed camera, and can complete the inspection and maintenance of the entire substation equipment in a short time.

[0015] Finally, the present invention also provides a computer device, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method when executing the computer program. BRIEF DESCRIPTION OF THE DRAWINGS

[0016] The drawings described herein are used to provide a further understanding of the present invention and constitute a part of this application. The exemplary embodiments of the present invention and their descriptions are used to explain the present invention and do not constitute an improper limitation of the present invention. In the drawings:

[0017] Figure 1 1 is a flow chart of a method for identifying an oil level gauge reading according to Embodiment 1 of the present invention;

[0018] Figure 2 This is a schematic diagram of the structure of the oil level gauge reading recognition system according to Embodiment 1 of the present invention;

[0019] Figure 3 This is a logic block diagram of the oil level gauge reading recognition system according to Embodiment 1 of the present invention. DETAILED DESCRIPTION

[0020] At present, oil level detection mainly relies on manual inspection, which will often cause problems such as "missed detection, misdetection and inability to measure". The manual transcription method requires manual photography of casing oil level gauges and oil pillow oil level gauges on high-voltage substations or use of telescopes and other equipment to observe the readings of oil level gauges for estimation, and then record them in the book. This method has high labor costs and lacks real-time performance. In addition, manual estimation has large errors and cannot accurately reflect the oil level of the oil level gauge, which will bring risks to the inspection and maintenance of substations.

[0021] Some adopt the installation of oil level monitoring devices, such as patent No. 2023100708391, which installs electronic control components and trigger components inside the oil conservator, and automatically triggers an alarm when the oil level is low. Although this method can trigger an automatic alarm when the oil level is lower than the preset oil level, it cannot report the oil level in real time, and requires modification of the internal oil level, which has high modification costs and unpredictable safety.

[0022] There are also some methods that use ultrasonic and other sensors. For example, patent No. 2023116689348 proposes to calculate the time difference of ultrasonic echoes, and deduce the height of the oil level in the oil pillow by the difference calculated by the time difference and the preset difference. Although this method does not require direct modification of the interior of the oil level tank, the ultrasonic detector needs to be close to the side wall of the oil pillow oil level, and the height of the detector needs to be adjusted multiple times. This makes working near UHV not only cumbersome but also very dangerous, which does not meet the requirements of intelligence and safety.

[0023] At present, there are some image recognition technologies based on deep learning, such as patent number 2024100336622, which constructs a key point data set of the oil level gauge, trains a model that can identify the key points of the oil level gauge, and then performs post-processing analysis on the key points of the oil level gauge detected by the model to obtain the reading of the oil level gauge. This method can only be used for a certain type of oil level gauge, is not generalizable, and requires the collection of a large amount of oil level gauge data as a data set, which is difficult to collect and the workload of marking key points is also very large.

[0024] Based on the above background, in order to protect the safety of personnel inspections, solve the problems of cumbersome manual inspection work, low efficiency and high cost of installing sensor equipment modification. After long-term research and development, the applicant has proposed a method, system, computer equipment and medium for oil level gauge reading recognition, which is not only highly generalizable, but also greatly improves the accuracy of oil level gauge reading recognition. In order to make the purpose, technical solutions and advantages of the present invention clearer, the present invention is further described in detail in combination with the embodiments and drawings. The schematic implementation mode of the present invention and its description are only used to explain the present invention and are not intended to limit the present invention.

[0025] It should be noted that, under the premise of no conflict, the various embodiments or technical features described below can be arbitrarily combined to form new embodiments, which also fall within the scope of protection of the present invention.

[0026] In the description of the present invention, the following terms need to be explained:

[0027] As for directional words, the terms "center", "lateral", "longitudinal", "length", "width", "thickness", "up", "down", "front", "back", "left", "right", "vertical", "horizontal", "top", "bottom", "inside", "outside", "clockwise", "counterclockwise" and the like indicating directions and positional relationships in the specification and claims of this application are based on the directions or positional relationships shown in the drawings and are only for the convenience of describing the present invention and simplifying the description, and do not indicate or imply that the device or element referred to must have a specific direction, be constructed and operated in a specific direction, and should not be understood as limiting the specific scope of protection of the present invention.

[0028] The terms "first", "second", etc. in the description and claims of the present application are used to distinguish similar objects and are not necessarily used to describe a particular sequence or order.

[0029] The terms "including" and "having" and any variations thereof in the specification and claims of this application are intended to cover non-exclusive inclusions. For example, a process, method, system, product or apparatus comprising a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units not explicitly listed or inherent to these processes, methods, products or apparatuses.

[0030] When an element is referred to as being "fixed to" or "disposed on" another element, it may be directly on the other element or there may be an intermediate element. When an element is referred to as being "connected to" another element, it may be directly connected to the other element or there may be an intermediate element. When an element is referred to as being "provided with" another element, it may be provided on the surface or inside of the element.

[0031] Unless explicitly stated otherwise, throughout the specification and claims, the term “comprise” or variations such as “include” or “comprising”, etc., will be understood to include the stated elements or components but not to exclude other elements or components.

[0032] The term "electrical connection" in the specification and claims of this application may refer to a circuit connection with physical contact or a communication connection, which may be a wired communication connection or a wireless communication connection.

[0033] Example 1

[0034] First, this embodiment provides a method for identifying an oil level gauge reading, such as Figure 1 As shown, including:

[0035] S1: Obtain oil level gauge device image data; the oil level gauge device image data is an image of the device on which the oil level gauge is installed, which may be a global image or a local image, as long as the image includes the oil level gauge, and may be oil level gauge device image data collected from manual, mobile robot or drone or fixed camera.

[0036] S2: Input the oil level gauge device image data into the first model, the first model identifies the oil level gauge in the oil level gauge device image, and obtains the oil level gauge image after cropping; the first model is used to identify the oil level gauge in the oil level gauge device image, and can be an oil level gauge target detector based on deep learning, which detects and identifies the oil level gauge in the oil level gauge device image to obtain the oil level gauge coordinate frame (x, y, w, h), and then crops according to this coordinate frame to obtain a picture of only the oil level gauge area, and inputs it into the second model for further processing.

[0037] S3: Input the oil level gauge image into the second model, and the second model identifies the oil level window and oil level area in the oil level gauge image and their coordinates; input the picture of only the oil level gauge area obtained in S2 (the cropped picture) into the second model, and perform more detailed detection and identification based on the oil level gauge captured by the first model. First, the second model receives the cropped picture of the oil level gauge or the oil level box detected by the first model, and then the second model infers the cropped oil level gauge picture to detect the coordinate positions of the rect oil level window and the area oil level box, which are (rect_x1, rect_y1, rect_y2, rect_y2) and (area_x1, area_y1, area_x2, area_y2) respectively.

[0038] The above scheme is to use the first model to detect and identify the oil level gauge, obtain the oil level gauge coordinate frame (x, y, w, h), and then crop the image of only the oil level gauge area according to the coordinate frame, and enter the obtained image of only the oil level gauge area (cropped image) into the second model.

[0039] S4: Calculate the oil level reading by post-processing the coordinates of the oil level window and the oil level area; specifically, calculate the height h of the oil level area and the height H of the oil level window respectively through the coordinates of the oil level window and the oil level area, and then obtain the oil level reading through the ratio ρ of the two; wherein ρ=h / H.

[0040] Specifically, according to step S3, the coordinate positions of the rect oil level window and the area oil level frame are detected, and the width rect_w and height rect_h (i.e., the height H of the oil level window) of rect are calculated respectively:

[0041] rect_w = rect_x2 - rect_x1

[0042] rect_h = rect_y2 - rect_y1

[0043] In order to reduce the detection error of the second model, the width area_w and height area_h (i.e. the height h of the oil level area) of the area are calculated as follows:

[0044] area_w=area_x2-area_x1

[0045] area_h = rect_y2 - area_y1

[0046] Then, the height ratio prop (i.e. ρ) of area to rect is calculated based on the height values ​​of the rect box and the area box:

[0047] prop = area_h / rect_h

[0048] The prop value is valid only when it is between 0 and 1.0. The obtained prop value is the height of the oil level gauge occupied by the oil level (with 0-1.0 as the oil level gauge range).

[0049] S5: Output the oil level detection result.

[0050] The output module can store an oil level threshold or range. When the oil level reading is lower than the oil level threshold or exceeds the range, it is judged as an oil level abnormality. According to different needs, the output module can display the oil level reading normally, or it can prompt an abnormality when the oil level is abnormal, so that personnel can quickly grasp the oil level situation.

[0051] In the above scheme, the two-stage detection model is combined, which not only has high generalization, but also greatly improves the accuracy of oil level gauge reading recognition, solving the problems of low generalization, easy to be affected by the environment, and low recognition accuracy of traditional image processing methods.

[0052] In this embodiment, the first model and the second model in the above method are based on an improved lightweight yolov8 network, and the lightweight yolov8 network uses a large number of 1×1 convolutional layers and 3×3 convolutional layers to replace the original 5×5 convolutional layers.

[0053] Optionally, the improved YOLOv8 consists of a lightweight backbone network and a detection head (Head) part. The first layer of the lightweight backbone network is a CBH structure (convolution layer-batch layer-Hswish activation layer), the third and fourth layers are inverted convolution modules (1 1×1 convolution layer and 1 3×3 depth-separable convolution layer) without adding the SE attention mechanism, the second, fifth, sixth, seventh, eighth, ninth, tenth, eleventh, and twelfth layers are inverted convolution module layers with added SE attention mechanism, and the thirteenth layer is an SPPF structure (composed of multiple maximum pooling layers in series). The main function of the SPPF structure is to perform multi-scale fusion of the features extracted by the backbone network, reduce the computational complexity of the backbone network, and expand the receptive field. The improved backbone network uses a large number of 1×1 and 3×3 convolutions, which greatly reduces the model. The number of parameters is reduced, and the feature extraction capability of the model is further improved by introducing residual blocks and SE attention mechanisms. The improved backbone network uses the NAS (network architecture search) algorithm to find the optimal network structure configuration for different hardware platforms and computing requirements, which greatly reduces the computational burden on the mobile robot computing module. In the head part, the 14th and 17th layers are upsampling layers (using the nearest neighbor difference method to enlarge the feature map), the 15th, 18th, 20th, and 23rd layers are CBL structures (convolutional layer-batch layer-silu activation function layer), and the 16th, 19th, 22nd, and 25th layers are C2f structures (C2f performs feature conversion on input data through two convolutional layers). C2f helps to increase the nonlinear ability and representation ability of the network, and can realize feature fusion in the channel dimension. The improved lightweight yolov8 network structure can greatly reduce the computing resources required on the edge computing end (such as the robot computing module).

[0054] The above method also includes constructing and training the best first model and second model based on the above improved lightweight yolov8 network. The details are as follows:

[0055] (1) Build and train the first model

[0056] ① Obtain the first oil level gauge device image. The oil level gauge device image data is an image of the device where the oil level gauge is installed. It can be a global image or a local image. The image only needs to include the oil level gauge. The oil level gauge image data is collected from a mobile robot, a drone, or a fixed camera.

[0057] ② Labeling: You can use LabelImg (labeling tool) to manually draw a frame (i.e., frame the oil level gauge box area). By manually drawing the outer rectangular frame of the oil level box, you can get the upper left corner coordinates of the oil level gauge box in the image and the width and height (x, y, w, h) in the labeling tool and save them in an XML file with the same name as the image.

[0058] ③ Data enhancement: Generate multiple first oil level gauge images through random cropping, rotation, translation, chromaticity, brightness and other data enhancement methods to enhance the scene richness of the oil level gauge image and use them as the training set data for the first model;

[0059] ④ Training: Use the oil level gauge data and txt labels obtained in the previous step as the oil level gauge data set to train the deep learning-based oil level gauge target detector until the iteratively optimal first model is obtained.

[0060] (2) Build and train the second model

[0061] A. Obtain a second oil level gauge image and read the image. The second oil level gauge image may be the first oil level gauge image used in the first model training. After obtaining the optimal oil level gauge (casing oil level, oil level tank) target detector through the previous training, use the optimal detector to detect and identify the oil level gauge to obtain the oil level gauge coordinate frame (x, y, w, h). Then, according to the coordinate frame, crop a picture of only the oil level gauge area. The obtained picture of only the oil level gauge area (cropped picture) is fed into the second model. The second oil level gauge image may also be an oil level gauge image cropped from a re-collected oil level gauge device image.

[0062] B. The second oil level gauge image is subjected to the RSP-oil level picture enhancement algorithm to enhance the data of different oil levels, thereby enriching the training data set of the second round of object detectors;

[0063] C. Train the second model until the iterative optimal second model is obtained. After the RSP-oil level image enhancement algorithm, very rich different oil level data are obtained, and then the improved lightweight yolov8 network of the present invention is used for training, and the optimal oil level detection model is obtained through training iteration.

[0064] The idea of ​​the RSP-oil level image enhancement algorithm is: read the second oil level gauge image and / or its size; obtain the oil level area coordinates and oil level window coordinates of the second oil level gauge image, randomly generate a filling ratio value R, and adjust the area of ​​the oil level area and the area of ​​the oil level window in the second oil level gauge image according to the filling ratio value R to obtain images with different oil levels. The details are as follows:

[0065] Use the opencv library to read the source image (src_img, which refers to the cropped image) and its size (height img_h, width img_w). Then, by processing the XML file (which saves the annotation information of src_img, where the area box represents the oil level box, and rect represents the entire oil level window, where the oil level window rect is composed of the oil level area + air), obtain the coordinate information of the area oil level box (area_x1, area_y1, area_x2, area_y2), and the coordinate information of the rect oil level window (rect_x1, rect_y1, rect_x2, rect_y2). If the information of these boxes is incomplete, skip the current loop. Define a new empty list new_area_box to store the new box coordinates. A ratio rand_ration (i.e., filling ratio value R) is generated through random numbers. The rand_ration is controlled within 0.6. The ratio will be used to calculate the filling height filling_higher. The filling height filling_higher is obtained by subtracting the height of the oil level frame from the height of the rect oil level window multiplied by rand_ration.

[0066] filling_higher=((rect_y2-rect_y1)-(area_y2-area_y1))×rand_ration

[0067] Depending on the random ratio rand_ration, the filling is divided into two branches, one for increasing the oil level and the other for decreasing the oil level: when rand_ration is between 0.25 and 0.35, the pixels in the rect oil level window frame are used to fill the pixels in the area oil level area, and the pixels in the area oil level area are moved downward, so that the oil level height can be reduced. When rand_ration is greater than or equal to 0.35, the pixels in the area oil level area frame are used to fill the pixels in the rect oil level window frame above, and a slight up and down float is performed, so that the oil level height can be increased and the oil level height is obtained higher than the original image. Filling example:

[0068] copy_src_img[rect_y1:rect_y2,rect_x1:rect_x2]=copy_src_img[area_y1:area_y2,area_x1:area_x2]

[0069] While filling, the corresponding label XML file after the oil level height is modified will also be generated synchronously. The RSP-oil level image enhancement algorithm not only enriches the different oil level height data of the oil level gauge, saves the time required for long-term data collection of different oil level heights, but also greatly saves the time for manual re-labeling.

[0070] On the other hand, this embodiment also provides an oil level gauge reading recognition system for implementing the above method, such as Figure 2 As shown, the system includes an image acquisition module, an image reasoning module, a post-processing module and an output module. The acquisition module is used to acquire image information of the oil level gauge device. The image reasoning module includes a first model and a second model, which are respectively used to identify the oil level gauge image in the oil level gauge device image and to identify the oil level window and the oil level area in the oil level gauge image. The post-processing module is used to calculate the oil level reading by post-processing the coordinates of the oil level window and the oil level area. The output module is used to output the oil level detection result according to the oil level.

[0071] In this embodiment, the image can be captured manually, or by a mobile robot, a drone, or a fixed camera, preferably a drone or a mobile robot. The image can be a picture taken directly, or it can be video data of different time periods taken by a camera device, and appropriate image data can be obtained by selecting key frames of the video data.

[0072] Optionally, in some embodiments, the image acquisition module, image reasoning module, post-processing module and / or output module are deployed on a mobile robot. Correspondingly, the first model and / or the second model are also deployed on the robot, and the mobile inspection robot completes the one-stop inspection of substation oil level gauges and other equipment, thereby realizing high-precision, efficient and intelligent modern intelligent substation operation and maintenance. For example, the mobile robot's variable focus pan-tilt head is used to collect equipment images of the oil level gauge, and then the mobile robot control unit takes the stream of the pan-tilt head, and then transmits the obtained image data to the algorithm module deployed in the computing unit of the mobile robot through the http protocol, and the algorithm module then processes the image, and obtains the reading of the oil level gauge through the algorithm module. Deployment on mobile robots can make the oil level detection technology of substation oil filling equipment a big step forward. First, 4. Use mobile robots to deploy the oil level gauge reading recognition algorithm, from shooting the oil level gauge image to outputting the oil level gauge reading, end-to-end completion, without relying on a server with strong computing power, lower cost, and better real-time performance; secondly, use mobile robots to replace manual inspections to protect the safety of inspection personnel, avoid personnel climbing up and down on substation equipment, and also reduce the tedious and inefficient problems of manual inspections. Moreover, compared with the installation of oil level gauge reading recognition sensor equipment, it is more flexible. The oil level gauge reading sensor can only act on the oil level gauge; finally, the mobile robot platform is more flexible than fixed cameras, and can complete the inspection and maintenance of the entire substation equipment in a short time.

[0073] After the first model and the second model are post-processed and deployed in the computing module of the mobile robot, the entire deployment step is completed. After the deployment is completed, the robot receives the oil level gauge image collected by the mobile robot, directly outputs the normal and abnormal readings of the current oil level gauge, and then sends it to the substation intelligent patrol maintenance system through the http protocol. Finally, the substation intelligent patrol maintenance system makes the next step to complete the closed loop of the entire data, such as Figure 3 shown.

[0074] Finally, this embodiment also provides a computer device, including a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements the above method when executing the computer program.

[0075] Finally, it should be pointed out that the above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit them. Although the present invention has been described in detail with reference to the above embodiments, a person skilled in the art should understand that the technical solutions described in the above embodiments can still be modified, or some of the technical features can be replaced by equivalents, and these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. A method for identifying oil level gauge readings, characterized in that: include: Obtain image data of oil level gauge equipment; The oil level gauge device image data is input into a first model, the first model identifies the oil level gauge in the oil level gauge device image, and obtains the oil level gauge image after cropping; Inputting the oil level gauge image into a second model, and using the second model to identify the oil level window and oil level area in the oil level gauge image and their coordinates; Calculating the oil level reading by post-processing the coordinates of the oil level window and the oil level area; Output oil level detection results.

2. The oil level gauge reading recognition method according to claim 1, characterized in that: The oil level gauge reading recognition method further comprises: constructing and training the first model, Acquire a first oil level gauge device image, where the first oil level gauge device image is a captured image of a device including an oil level gauge; Marking a circumscribed rectangular frame of the oil level gauge in the first oil level gauge device image; Generate a plurality of first oil level gauge images by data enhancement as a training set of a first model, and train the first model, wherein the data enhancement includes one or more of random cropping, rotation, translation, chromaticity, and brightness; Build and train the second model, Acquire a second oil level gauge image; Using the second oil level gauge image through an enhancement algorithm to obtain image data of different oil levels, the image data is used as a training set for the second model, and the second model is trained; The second oil level gauge image is the first oil level gauge image or an oil level gauge image cropped from a collected oil level gauge device image.

3. The oil level gauge reading recognition method according to claim 2, characterized in that: The enhancement algorithm includes: reading the second oil level gauge image and / or its size; Obtain the oil level area coordinates and oil level window coordinates of the second oil level gauge image, Randomly generate a fill ratio value R, The area of ​​the oil level region and the area of ​​the oil level window in the second oil level gauge image are adjusted according to the filling ratio value R to obtain images of different oil levels.

4. The oil level gauge reading recognition method according to claim 3, characterized in that: The filling ratio value R≤0.6, and adjusting the area of ​​the oil level region and the area of ​​the oil level window in the second oil level gauge image according to the filling ratio value R to obtain images with different oil levels includes: When 0.25≤R≤0.35, the oil level area in the second oil level gauge image is moved downward, and the pixels of the oil level area are filled with the pixels of the oil level window frame, so that the oil level area is reduced, and an image with an oil level lower than the oil level in the second oil level gauge image is obtained; When R≥0.35, the pixels of the oil level window frame are filled with the pixels of the oil level area to obtain an image in which the oil level is higher than the oil level in the second oil level gauge image.

5. The oil level gauge reading recognition method according to any one of claims 1 to 4, characterized in that: The first model and / or the second model adopts a lightweight yolov8 network, and the lightweight yolov8 network uses multiple 1×1 convolutional layers and 3×3 convolutional layers to replace the original 5×5 convolutional layers.

6. The oil level gauge reading recognition method according to claim 5, characterized in that: The lightweight yolov8 network includes a lightweight backbone network and a detection head part. The first layer of the lightweight backbone network is a CBH structure, the third and fourth layers are respectively a 1×1 convolution layer without adding a SE attention mechanism and a 3×3 depth-separable convolution layer without adding a SE attention mechanism, the second, fifth, sixth, seventh, eighth, ninth, tenth, eleventh, and twelfth layers are inverted convolution module layers with an SE attention mechanism, the thirteenth layer is an SPPF structure, the fourteenth and seventeenth layers of the detection head part are upsampling layers, the fifteenth, eighteenth, twenty-third layers are CBL structures, and the sixteenth, nineteenth, twenty-second, and twenty-fifth layers are C2f structures.

7. The oil level gauge reading recognition method according to claim 1, characterized in that: The calculating of the oil level reading by post-processing the coordinates of the oil level window and the oil level area comprises: The height h of the oil level area and the height H of the oil level window are calculated respectively by the coordinates of the oil level window and the oil level area, and then the oil level reading is obtained by the ratio ρ of the two; Wherein, ρ = h / H.

8. An oil level gauge reading recognition system, used to implement the method according to any one of claims 1 to 7, characterized in that: It includes an image acquisition module, an image reasoning module, a post-processing module and an output module. The image acquisition module is used to collect image information of the oil level gauge device. The image reasoning module includes a first model and a second model, which are respectively used to identify the oil level gauge image in the oil level gauge device image and to identify the oil level window and oil level area in the oil level gauge image. The post-processing module is used to calculate the oil level reading by post-processing the coordinates of the oil level window and the oil level area. The output module is used to output the oil level detection result according to the oil level.

9. The oil level gauge reading recognition system according to claim 8, characterized in that: The image acquisition module, image reasoning module, post-processing module and / or output module are deployed on a mobile robot.

10. A computer device comprising a memory, a processor and a computer program stored in the memory and executable on the processor, characterized in that: When the processor executes the computer program, the method according to any one of claims 1 to 7 is implemented.

Images