Meter identification intelligent analysis method, system, medium, program product and device

Abstract

The present application belongs to the field of meter identification, and proposes a meter identification intelligent analysis method, system, medium, program product and equipment. The local parameters of a large model are fine-tuned based on a low-rank adaptive algorithm, so that it is suitable for the meter field. The fine-tuned large model is jointly trained using instruction data in the power field to guide the fine-tuned large model to learn each scene task of the meter. Corresponding tool modules are created according to each scene task. User input information is obtained, and the large model after joint training extracts user intent according to the input information. The created tool module corresponding to the corresponding scene task is called, and the tool module is used to perform scene task execution on the target meter image or the target meter image and text information to obtain the analysis result of meter identification. The present application effectively improves the performance of the meter identification analysis method.

Description

Technical Field

[0001] This invention belongs to the field of meter identification, specifically relating to an intelligent analysis method, system, medium, program product, and equipment for meter identification. Background Technology

[0002] The statements in this section are merely background information related to the present invention and do not necessarily constitute prior art.

[0003] Substation meters are a crucial component of power systems, used to monitor the operation of electrical equipment. Defect identification and automatic reading of these meters are essential tasks for remote intelligent substation inspections. There are various types of meters, including sulfur hexafluoride density relays, surge arrester leakage current meters, transformer thermometers, oil level gauges, gas relays, and instrument transformer oil pressure indicators, among others. Each type is manufactured by multiple companies, resulting in diverse appearances and specifications, lacking unified standards and specifications. Furthermore, defects such as damaged or corroded casings, cracked or blurred dials may exist, and obtaining defect data is extremely difficult, all of which pose challenges to intelligent meter analysis.

[0004] Existing intelligent meter identification and analysis methods primarily focus on meter reading recognition, rarely addressing the defect analysis of the meter's appearance before the reading process. Defects such as damaged casings, blurred dials, or broken dials can severely impact subsequent reading recognition and analysis, leading to unnecessary misidentifications. Therefore, intelligent meter identification and analysis should not only focus on downstream automatic meter reading recognition but also prioritize the analysis of the meter's overall appearance to determine if it meets the prerequisites for reading recognition. Current solutions lack a more comprehensive and intelligent analysis process for intelligent meter identification and analysis. Summary of the Invention

[0005] To address the aforementioned problems, this invention proposes a method, system, medium, program product, and equipment for intelligent meter identification analysis. By leveraging advanced graphic multimodal large model technology, this invention provides a reliable solution for intelligent meter identification analysis tasks.

[0006] According to some embodiments, the present invention adopts the following technical solution: In a first aspect, the present invention provides a smart analysis method for meter identification.

[0007] A smart analysis method for meter identification includes the following steps: Obtain relevant meter manuals in the power sector, integrate them, establish a meter vector knowledge base, and enhance the generation of a large model embedded in the knowledge base through retrieval. Acquire image data covering various types of meters, as well as the annotation information of the image data, and form instruction data for corresponding scene tasks by combining the image data and the corresponding annotation information. The scene tasks include meter target detection, meter status classification, key point detection, and meter reading recognition. The local parameters of the large model are fine-tuned based on the low-rank adaptation algorithm. The fine-tuned large model is then jointly trained using the instruction data to guide it in learning various metering tasks. Each tool module is created according to the task of each scenario. Each tool module is used to execute the corresponding task based on the acquired meter images or meter images and text information. The system acquires user input information, uses a jointly trained large model to extract user intent based on the input information, calls the corresponding tool modules for the scene task, and uses the tool modules to perform scene tasks on the target meter image or the target meter image and text information to obtain the analysis results of meter recognition.

[0008] The above solution first utilizes strategies such as prompt word engineering, knowledge base embedding, and fine-tuning training to achieve a private customization of the general large model within the meter / instrument professional field, resulting in a more specialized large model suitable for automatic meter analysis. Then, various dedicated tool modules are created, including state classification, target detection, key point detection, and meter reading recognition, which can process different aspects of the meter, fully acquiring elemental information from various aspects of the meter target, facilitating the large model's invocation and analysis. This invention combines the large model with an external knowledge base, enabling the large model to reference this external knowledge when generating answers; this increases the model's ability to handle complex and specific problems.

[0009] In summary, the above solution can systematically call various meter detection and analysis tool modules. These various tool modules complement the large model, effectively improving the performance of the meter identification and analysis method.

[0010] As an alternative implementation method, the large model is a graphic-text multimodal large model.

[0011] As an alternative implementation, the process of determining prompt words in the field of meter recognition includes defining the professional roles of the large model, constructing prompt words using existing prompt examples, and using an approach of adding assumptions to enable the large model to give a negative response under uncertain conditions.

[0012] As an alternative implementation method, the specific process of generating a large model embedded knowledge base through retrieval enhancement includes: We obtained instruction manuals for various specifications from publicly available online databases and meter manufacturers, and organized the documents according to a unified standard. The embedded model is used to convert the organized instruction manual content into vectors and store them in a vector knowledge base; Convert user queries into vectors and retrieve relevant knowledge from a vector knowledge base; An enhanced suggestion template is constructed based on the search results, and the large model generates a response by combining the enhanced suggestion template.

[0013] As an alternative implementation, the process of forming instruction data for a corresponding scenario task from image data and corresponding annotation information includes: acquiring on-site collected meter image data or historical meter image data from the power scenario, covering various types of meters, forming a dataset, and ensuring that the dataset contains images under different lighting conditions, shooting angles, and background environments, and increasing the diversity of data through rotation, scaling, flipping, and cropping operations. Each image is annotated in detail, including the instrument type, instrument status, instrument reading, and annotation information of key point locations; The aforementioned images, annotation information, and textual descriptions of the images are integrated into instruction data, which is related to the tasks of each scenario.

[0014] As a further implementation, the instrument status includes normal dial, damaged dial, blurred dial, and damaged casing.

[0015] As an alternative implementation method, the process of fine-tuning the local parameters of a large model based on the low-rank adaptation algorithm includes: Let the pre-trained parameter weights be... During the fine-tuning phase, the weights are updated: ; in, It is the size of the weights updated during the fine-tuning training phase; Based on the low-rank adaptation algorithm Perform matrix transformations: ; Where BA are matrices. , , , ; In fine-tuning training, only the right side needs to be calculated. , The matrix can be modified without changing the pre-trained weights on the left side, thus reducing the total number of weight parameters. for: ; Furthermore, the rank of the low-rank adaptation algorithm is set to 64, and the Dropout probability value is 0.05.

[0016] As an optional implementation, the tool module includes a meter status classification module, which is used to determine the meter status of the image, obtain the specific status of the meter in the current scene, and issue a warning message if the current meter has an appearance defect, indicating the specific defect type. The meter status classification module uses a multi-head classifier, which includes multiple attribute classification heads. Each attribute classification head is configured with a loss function. The total loss function of the multi-head classifier is the sum of the loss functions of each attribute classification head multiplied by their respective weight values.

[0017] As a further step, the multi-head classifier includes four attribute classification heads. The first attribute classification head is used to distinguish dial attributes, specifically including dial normal, dial damaged, and dial blurry. The second attribute classification head is used to classify casing attributes, specifically including casing normal, casing damaged, casing dirty, and casing corroded. The third attribute classification head is used to classify target attributes, specifically including pointer-type targets, digital display-type targets, and other target types that do not require attention. The fourth attribute classification head is used to distinguish pointer attributes, specifically including pointer detachment, pointer exceeding limits, pointer pointing to warning range, and pointer pointing to minimum or maximum range value.

[0018] As an alternative implementation, the tool module further includes: The meter target detection module uses a target detection network model to achieve meter positioning and type identification. The meter key point detection module is used to detect the key point positions on the pointer meter, including the pointer tip position, pointer tail position, the start point of the range, and the end point of the range. The meter text detection and recognition module is used to detect and recognize text information on the meter, including scale values, meter number, and several units of measurement.

[0019] As an alternative implementation, the process of obtaining user input information and extracting user intent based on the input information using a jointly trained large model includes obtaining the user input statement, determining the intent or purpose expressed in the user input statement, and identifying the intent by classifying the input statement into a predefined intent category.

[0020] As an alternative implementation, when the extracted user intent is to identify the meter reading, the starting scale point A, the ending scale point B, the pointer tip point C, and the pointer tail point D are extracted. The specific steps for calculating the meter reading include: If the line segment AB connecting the starting point A and the ending point B is found, and the perpendicular bisector L of the line segment AB is found, then the center of the current dial circle must exist on the line L. By connecting the pointer tip C and the pointer tail D to obtain line segment CD and extending CD, find the intersection point of the perpendicular bisector L and the line CD. When the perpendicular bisector L is not collinear with line CD, find the unique intersection point O of L with lines AB and CD. Point O is the current center position of the dial. Calculate the ratio of ∠AOC to ∠AOB to obtain the proportion of the current pointer's pointing angle in the entire dial range. When line AB and line CD are collinear, the current pointer position angle is half of the entire dial range. After calculating the percentage of the current reading based on the coordinates of the key points, locate the scale frame and corresponding scale value closest to the starting point A and the ending point B, respectively, to obtain the current range of the meter and the actual reading of the current pointer instrument. ; in, This represents the maximum scale value of the current meter. This is the minimum scale value of the current meter.

[0021] As a further implementation, when positioning the pointer and the start and end positions of the range, the quadrant is used for determination: when the minimum range is greater than the maximum range at an angle, the angle should be determined to be on the side with a value greater than 180 degrees.

[0022] in, This refers to the starting angle of the measuring range. This refers to the angle at the end of the measurement range, and the unit is degrees.

[0023] As a further implementation method, for pointer instruments with non-uniform ranges, the identified scale values ​​are sorted in ascending order, and the center points of the scale target frame are connected in pairs in sequence. It is determined which scale range's line segment intersects with the line where the pointer is located, and the intersection point is on the line segment connecting the scales. The scale range pointed to by the pointer is then calculated, and this part of the scale range is taken as the start and end range of the entire range. The dial range value is calculated based on the included angle.

[0024] A second aspect of the present invention provides a meter identification intelligent analysis system.

[0025] A meter identification intelligent analysis system, comprising: The knowledge base construction module is configured to acquire relevant meter manuals in the power field, integrate them, establish a meter vector knowledge base, and generate a large model embedded knowledge base through retrieval enhancement. The instruction generation module is configured to acquire image data covering various types of meters, as well as annotation information of the image data, and to form instruction data for the corresponding scene task by combining the image data and the corresponding annotation information. The scene task includes meter target detection, meter status classification, key point detection, and meter reading recognition. The large model fine-tuning module is configured to fine-tune the local parameters of the large model based on the low-rank adaptation algorithm, and use the instruction data to jointly train the fine-tuned large model to guide the fine-tuned large model to learn various scenario tasks of metering. The tool module creation module is configured to create corresponding tool modules based on each scenario task. Each tool module is used to execute the corresponding scenario task based on the acquired meter images or meter images and text information. The meter recognition and analysis module is configured to acquire user input information, extract user intent based on the input information using a jointly trained large model, call the tool module corresponding to the scene task, and use the tool module to perform scene task execution on the target meter image or the target meter image and text information to obtain the meter recognition analysis result.

[0026] A third aspect of the present invention provides a computer-readable storage medium.

[0027] A computer-readable storage medium for storing computer instructions, which, when executed by a processor, perform the steps in the above method.

[0028] A fourth aspect of the present invention provides a computer program product.

[0029] A computer program product comprising a computer program that, when executed by a processor, implements the steps of the above-described method.

[0030] A fifth aspect of the present invention provides an electronic device.

[0031] An electronic device includes a memory and a processor, as well as computer instructions stored in the memory and running on the processor, wherein the computer instructions, when executed by the processor, perform the steps in the method described above.

[0032] Compared with the prior art, the beneficial effects of the present invention are as follows: 1. This invention innovatively proposes an intelligent analysis method for meter identification. It utilizes strategies such as prompt word engineering, knowledge base embedding, and fine-tuning training to achieve a customized, specialized meter / instrument model tailored to the specific needs of meter analysis. Then, it creates various dedicated tool modules, including state classification, target detection, key point detection, and meter reading recognition, which can process different aspects of the meter, fully acquiring elemental information from various aspects of the meter target for easy access and analysis by the large model. This invention combines the large model with an external knowledge base, allowing the large model to reference this external knowledge when generating answers; this increases the model's ability to handle complex and specific problems.

[0033] 2. This invention is not only applicable to the analysis and processing of pointer-type meters, but also to the analysis and processing of digital display meters. By leveraging the powerful image understanding capabilities of multimodal large models and calling various analysis tool modules, this invention achieves more universal intelligent automatic meter analysis.

[0034] 3. This invention utilizes prompt word engineering, knowledge base embedding, and fine-tuning training strategies to make it more suitable for meter recognition scenarios, enabling maintenance personnel to more easily and professionally understand and analyze scene images. Inputting an image automatically analyzes the meter feedback recognition results, making it convenient to use and greatly improving maintenance efficiency.

[0035] 4. This invention embeds a knowledge base into a large model by using retrieval-enhanced generation technology. By retrieving relevant information from an external knowledge base and providing this information as additional context to the large model, the model's ability to generate text is enhanced, which can help the model reduce illusions and improve the accuracy of content generation.

[0036] 5. This invention uses a large model as its foundation and systematically calls upon various meter detection tool modules to fully acquire information on all aspects of the meter target. These various tool modules complement the large model, effectively improving the performance of the meter identification and analysis method.

[0037] To make the above-mentioned objects, features and advantages of the present invention more apparent and understandable, preferred embodiments are described below in detail with reference to the accompanying drawings. Attached Figure Description

[0038] The accompanying drawings, which form part of this invention, are used to provide a further understanding of the invention. The illustrative embodiments of the invention and their descriptions are used to explain the invention and do not constitute an improper limitation of the invention.

[0039] Figure 1 This is a flowchart of an embodiment of a smart analysis method for meter identification; Figure 2This is a schematic diagram of the network structure of a meter status classification model according to one embodiment; Figure 3 A schematic diagram illustrating the angle calculation process in one embodiment; Figure 4 This is a diagram illustrating the implementation effect of a meter identification intelligent agent in one embodiment. Figure 5 This is a physical image of a pressure gauge for which instruction data is provided in one embodiment. Figure 6 This is a schematic diagram of an embodiment of a meter identification intelligent analysis system; Figure 7 This is a schematic diagram of an electronic device according to one embodiment. Detailed Implementation

[0040] The present invention will be further described below with reference to the accompanying drawings and embodiments.

[0041] It should be noted that the following detailed descriptions are exemplary and intended to provide further illustration of the invention. Unless otherwise specified, all technical and scientific terms used herein have the same meaning as commonly understood by one of ordinary skill in the art to which this invention pertains.

[0042] It should be noted that the terminology used herein is for the purpose of describing particular embodiments only and is not intended to limit the scope of exemplary embodiments according to the invention. As used herein, the singular form is intended to include the plural form as well, unless the context clearly indicates otherwise. Furthermore, it should be understood that when the terms "comprising" and / or "including" are used in this specification, they indicate the presence of features, steps, operations, devices, components, and / or combinations thereof.

[0043] Where there is no conflict, the embodiments and features described in this application may be combined with each other.

[0044] Example 1 As described in the background section, existing intelligent meter identification and analysis processes primarily focus on meter reading recognition, rarely addressing defect analysis of the meter's appearance before the reading process. This embodiment provides an intelligent meter identification and analysis method that improves data processing efficiency by utilizing a large model, combined with a professional knowledge base and tools, and has broad application prospects.

[0045] This embodiment describes the solution using intelligent meter identification and analysis in substation instruments as the application scenario.

[0046] Of course, this does not mean that the method provided in this embodiment can only be used in this scenario.

[0047] A smart analysis method for meter identification, such as Figure 1 As shown, it includes the following steps: S1: Obtain relevant meter manuals in the power field, integrate them, establish a meter vector knowledge base, and generate a large model embedded in the knowledge base through retrieval enhancement. S2: Acquire image data covering various types of meters, as well as the annotation information of the image data, and form instruction data for the corresponding scene task by combining the image data and the corresponding annotation information. The scene task includes meter target detection, meter status classification, key point detection, and meter reading recognition. S3: Fine-tune the local parameters of the large model based on the low-rank adaptation algorithm, and use the instruction data to jointly train the fine-tuned large model to guide the fine-tuned large model to learn various scenario tasks of metering. S4: Create corresponding tool modules according to each scenario task. Each tool module is used to execute the corresponding scenario task based on the acquired meter images or meter images and text information. S5: Obtain user input information, use the jointly trained large model to extract user intent based on the input information, call the tool module corresponding to the scene task, and use the tool module to perform scene task execution on the target meter image or the target meter image and text information to obtain the meter recognition analysis results.

[0048] In step S1, the large model can be selected from existing large-scale multimodal model series. Regardless of the type of multimodal model, it is pre-trained and then fine-tuned using high-quality data to better reflect human preferences.

[0049] In step S1, to better suit the task requirements of intelligent meter identification, this embodiment prioritizes private customization for the substation instrumentation field based on a general large model. The main methods include: (1) Prompt word engineering To address the requirements of intelligent meter recognition tasks, this embodiment guides the model to generate the desired output by skillfully designing input prompts.

[0050] First, you can define professional roles for the large model, such as "You are a professional substation meter identification assistant" or "You are a substation instrument identification expert". Assigning professional roles can bring very good results.

[0051] Secondly, providing a small number of prompt examples will also allow the model to output better results. Existing prompt examples can be used to construct prompt words.

[0052] Finally, in order to obtain the most accurate results possible, this embodiment uses the method of adding assumptions to enable the large model to give a negative response under uncertain conditions.

[0053] (2) Knowledge base embedding This embodiment establishes a vector knowledge base for substation instruments by integrating instruction manuals for hundreds of common substation specifications. The knowledge base is continuously updated and supplemented to maintain its accuracy and timeliness.

[0054] This embodiment uses a retrieval-enhanced generation technique to embed a knowledge base into a large model. By retrieving relevant information from an external knowledge base and providing this information as additional context to the large model, the model's ability to generate text is enhanced, which can help the model reduce illusions and improve the accuracy of content generation.

[0055] In this embodiment, the retrieval enhancement generation process includes: Step S101, Knowledge Preparation: Collect instruction manuals for common substation meter specifications from online public information databases, instrument sales manufacturers, and other channels, and organize the documents according to unified standards.

[0056] Step S102, Embedding and Indexing: Use the embedding model to convert the instruction manual content into vectors and store them in the vector database.

[0057] Step S103, Query and Retrieval: Convert the user query into a vector and retrieve relevant knowledge from the database.

[0058] Step S104, Enhanced prompts: Construct an enhanced prompt template based on the search results, and generate a response based on the enhanced prompt template.

[0059] The specific process of step S2 includes: This embodiment uses a publicly available large-scale image-text multimodal model as the base model, which has powerful multimodal processing capabilities and pre-trained weights. To optimize model performance and make it more professional and adaptable to the intelligent identification task of substation instruments, we perform instruction fine-tuning on the large-scale model using the dataset.

[0060] To make it clearer to those skilled in the art, the following implementation steps are provided: The first step is data preparation. For example, if a user needs to know whether an image contains meter targets of interest, an example of the instruction data would be as follows: { "id": "identity_1", "conversations": [ { "from": "user", "value": "<|vision_start|> / rootpath / biaoji.jpg<|vision_end|>" }, The above instructions are mainly used to obtain user input and determine whether the image contains meters that need to be monitored. Please describe this objective.

[0061] { "from": "assistant", "value": "Yes, this image shows a pressure gauge used to measure the pressure of SF6 gas. The gauge's range is (-0.1~0.9) MPa, the red scale range is (-0.1~0.5) MPa, the orange scale range is (0.5~0.55) MPa, and the green scale range is (0.55~0.9) MPa." } ] } The image corresponding to / rootpath / biaoji.jpg in this instruction data is as follows: Figure 5 As shown.

[0062] The above demonstrates an example of instruction data for image scene analysis of an input image. In practice, instructions for various task types can be compiled, mainly including meter target detection, meter status classification, key point detection, and meter reading recognition. When preparing data, tasks such as meter status classification will involve target images in various states, such as normal dial, damaged dial, blurred dial, and damaged casing. Integrating these different status types and descriptive texts and including them in the training data ensures the richness and diversity of the training data.

[0063] Of course, in other embodiments, a large-scale graphic and textual multimodal model, including but not limited to Qwen2-VL-2B-Instruct and Qwen2-VL-7B-Instruct, can be selected as the basic large-scale model.

[0064] Similarly, in other embodiments, existing large-scale multi-modal graphics models such as MiniGPT-4 and Shusheng 2.5 can be selected, and they will not be listed exhaustively here.

[0065] In step S3, full parameter fine-tuning requires adjusting all parameters of the model. As the size of the pre-trained model continues to increase, the resource pressure of full fine-tuning also increases exponentially. In order to achieve model parameter fine-tuning more efficiently, this embodiment adopts low-rank adaptation (Lora) for fine-tuning. Its core idea is to simulate the amount of parameter change through low-rank decomposition, thereby achieving indirect training of a large model with a very small number of parameters.

[0066] Assume the pre-trained parameter weights are During the fine-tuning phase, the weights are updated: ; in, It refers to the weights updated during the fine-tuning training phase.

[0067] Lora will Perform matrix transformations: ; in, , , , , The values ​​are generally 1, 2, 4, 8, etc.

[0068] In fine-tuning training, only the right side needs to be calculated. , A matrix, without needing to modify the left side. (Pre-trained weights). This results in the total number of weight parameters. : ; This greatly reduces the number of weight parameters that need to be adjusted.

[0069] In the specific implementation, the layer for which Lora is applied is specified as ["q_proj", "k_proj", "v_proj", "o_proj", "gate_proj", "up_proj", "down_proj"]. The rank of Lora is set to 64, and the Dropout probability value of the Lora layer is 0.05.

[0070] In summary, to improve the overall performance of the model, this embodiment first designs specific instruction data for various instrument identification and analysis tasks, and then conducts joint training with multiple tasks to guide the model to learn the relevant content of various instrument identification tasks. During fine-tuning, low-rank adaptation (Lora) is used to more efficiently achieve domain-specific customization of the model.

[0071] Of course, in other embodiments, depending on the different requirements of the meter identification intelligent analysis scenario, such as accuracy, precision, or speed, other methods can be used to fine-tune the large model.

[0072] In step S4, the specific process of creating the tool module includes: To obtain more accurate results, this embodiment creates various tool modules that can be used by large models. These modules are specifically designed to process different aspects of meters, thereby providing comprehensive and accurate results.

[0073] The main tool modules are as follows: (1) Meter status classification module The meter status classification module is used to determine the meter status of an image, obtaining the specific status of the meter in the current scene (i.e., whether there are any appearance defects). If the current meter has appearance defects (focusing on three types of defects: damaged casing, blurred dial, and broken dial), it can directly issue an alert for the specific defect type. It is assumed that the current meter may have one or more appearance defects, and these types are not necessarily mutually exclusive. For example, if the current meter has a broken dial, it may also have other defects such as a broken casing. Therefore, a single label cannot represent an entire image. Thus, the meter status classification task is considered a multi-label, multi-classification problem.

[0074] For meter status classification, this embodiment designs a multi-head classifier, which differs from common single-label multi-classification models. Because multiple attributes need to be predicted, this embodiment designs multiple attribute classifiers instead of using a single classifier. Each attribute classifier can be trained with its own multi-classification loss function or single-classification loss function. For mutually exclusive categories of the same attribute, a multi-label classification loss function is used; for categories of the same attribute that are not necessarily mutually exclusive, a single-label classification loss function is used. During training, the number of loss function branches corresponds to the number of attribute classifiers, and each is assigned a specific weight. Adding the loss values, in this embodiment, the optimized design of the meter status classification model network structure is as follows: Figure 2 As shown, including four attribute classification heads, the total loss value is: ; like Figure 2 As shown, category header 1 is used to distinguish dial attributes, specifically categorized as normal dial, damaged dial, and blurry dial; category header 2 is used to classify casing attributes, specifically categorized as normal casing, damaged casing, dirty casing, and rusted casing; category header 3 is used to classify target type attributes, specifically categorized as pointer-type target type, digital display target type, and other target types that do not require attention; category header 4 is used to distinguish pointer attributes, specifically categorized as pointer detachment, pointer exceeding limits, pointer pointing to warning range, and pointer pointing to minimum or maximum range value. Here, the data processing categorizes each attribute and category of the meter target in considerable detail, with a focus on the three types of defects: damaged casing, blurry dial, and damaged dial.

[0075] Of course, the above structure can also be extended to other classification tasks with multiple attributes and multiple labels.

[0076] (2) Meter target detection module To avoid excessive background information and interference from other instruments on the large model analysis, the meter target detection network model here, including but not limited to common target detection network models such as YOLOv8 and Faster R-CNN, can realize meter location and type identification.

[0077] (3) Meter key point detection module The key point detection module is used to detect the key point positions on pointer-type meters. Generally, for single-pointer meters, four points need to be monitored: the tip of the pointer, the tail of the pointer, the start of the range, and the end of the range.

[0078] (4) Meter text detection and recognition module The meter text detection and recognition module is used to detect and recognize text information on meters, such as scale values, meter numbers, units of measurement, etc., as well as to recognize meter readings.

[0079] In this embodiment, to more accurately identify meter readings, an angle calculation process is provided, such as... Figure 3 As shown, it specifically includes: First, the key points of the dial are defined here as the starting mark point A, the ending mark point B, the pointer tip point C, and the pointer tail point D.

[0080] The specific calculation steps are as follows: If the line segment AB connecting the starting point A and the ending point B is found, and the perpendicular bisector L of the line segment AB is found, then the center of the current dial circle must exist on the line L. By connecting the pointer tip C and the pointer tail D to obtain line segment CD and extending CD, find the intersection point of the perpendicular bisector L and the line CD. When the perpendicular bisector L is not collinear with line CD, the unique intersection point O of line AB and line CD can be obtained. Point O is the current center position of the dial. Calculating the ratio of ∠AOC to ∠AOB will give the percentage of the current pointer's position angle across the entire dial range. When line AB and line CD are collinear, the current pointer position angle is half of the entire dial range.

[0081]

[0082] After calculating the percentage of the current reading based on the coordinates of the key points, locate the scale frame and corresponding scale value closest to the starting point A and the ending point B, respectively. This will give you the current measurement range of the meter. Then, use the following calculation formula: ; This allows you to calculate the actual reading of the current pointer instrument.

[0083] Because there are different dial appearances with various orientations, and some dial scales have an angle range greater than 180 degrees while others have an angle range less than 180 degrees, this embodiment has made effective improvements in calculating angle values, so that the above angle calculation method can be universally applied.

[0084] First, define a fixed direction as the 0-degree position. Using a Cartesian coordinate system as an example, assuming the negative y-axis is the 0-degree angle, a full clockwise rotation of the pointer will result in angles in the four quadrants: (180°, 270°), (90°, 180°), (0°, 90°), and (270°, 360°). These quadrants can be used to determine the starting and ending positions of the pointer and range. When the minimum range is greater than the maximum range, the angle should be set to the side with a value greater than 180 degrees.

[0085] ; Here This refers to the starting angle of the measuring range. This refers to the angle at the end of the measuring range.

[0086] Specifically, for pointer-type instruments with non-uniform ranges, the detected scale values ​​are sorted in ascending order, and the center points of the scale target frame are connected in pairs sequentially. The intersection point of the pointer's line with a line segment within a specific scale range is determined, and this intersection point lies on the line segment connecting the scale segments. This allows the calculation of the scale range the pointer is pointing to. This portion of the scale range is then used as the start and end range of the entire range, and the dial range value can be calculated based on the included angle. This calculation method also applies to pointer-type instruments with uniform ranges, providing more accurate readings.

[0087] The design and implementation of these modules enable the acquisition of comprehensive information on various aspects of the metering target. These various tool modules complement the large model, effectively improving the recognition performance of the metering identification intelligent agent analysis method described in this embodiment. This embodiment uses the large model as a foundation and collaboratively calls various tool modules (including but not limited to the four mentioned above) to achieve comprehensive and accurate analysis of various types of metering images.

[0088] In step S5, the workflow of the intelligent agent in this embodiment can be summarized as follows: input - intent recognition - tool invocation - image understanding - output.

[0089] The following is a detailed description: Intent recognition aims to determine the intent or purpose expressed in user input. It can be viewed as a classification problem, identifying intent by categorizing input statements into predefined intent categories. These categories can include various tasks, queries, requests, etc. Given the current capabilities of large-scale models, intent recognition can actually be implemented using prompt word engineering.

[0090] In the large model fine-tuning section of step S3 above, the instruction design for the training data was proposed, and the specific implementation of intent recognition in this embodiment is also reflected therein. We design different instruction design methods for different intents. For example, if the user intent is scene classification, the instruction description can have multiple forms of expression, such as: "Does the current scene contain any meters?" "You are a substation instrumentation engineer. Please analyze whether the diagram contains any targets that you are interested in?" "Are there any metering targets to be analyzed in the current scenario?"

[0091] The above is only one example; other expressions may be used in other embodiments.

[0092] Although these questions are expressed in various ways, they all point to the same task: image scene analysis. This is taken into account in our instruction fine-tuning data; therefore, we design the same response pattern for different instruction descriptions of the same task. In this way, during the implementation process, only appropriate prompts need to be designed to meet the requirements of intent recognition.

[0093] In summary, the prompt words constructed in this embodiment are as follows: "You are an intelligent meter analysis expert. Please refer to the instrument instruction manual knowledge base to accurately identify the user's intent based on the user's input and take different actions for different situations:" Scenario 1: If the user's intent is scene classification, i.e., the user wants to determine whether the image contains meter targets to decide whether to continue with automated intelligent meter analysis, please describe the image scene and determine whether there are meter targets to be analyzed in the image. The meter target detection module can be called to locate the meters.

[0094] Case 2: If the user's intention is object detection, that is, the user expects to obtain the meter target area in the input image scene, then the meter target detection module is called to detect the meter targets in the image.

[0095] Scenario 3: If the user's intent is state classification, that is, the user wants to obtain the working status of the meter in the input image scene and analyze whether the meter target in the image has appearance defects such as dial damage, dial blurring, or casing damage, then the meter state classification module is called for classification.

[0096] Scenario 4: When the user's intent is reading recognition, that is, when the user wants to obtain the meter reading results in the image, the meter key point detection module and the meter text detection and recognition module can be called or the automated analysis process can be executed.

[0097] Scenario 5: If the user's intention is for a different question, that is, not one of the above scenarios, and they do not understand the question, please reply that the question is unclear.

[0098] The automated analysis process defined in this embodiment is as follows: (1) Preliminary inspection: Target detection: First, check if there are meter targets in the input image.

[0099] Appearance defect inspection: If a meter is present, further inspect the meter for appearance defects, such as a broken dial, a blurred dial, or a damaged casing.

[0100] (2) Digital display type meter processing: Direct reading: If the image only contains meters with digital displays, the reading is directly obtained and returned.

[0101] End the conversation: Analyze whether the reading results conform to the specifications based on the instrument instruction document knowledge base, and then end the current conversation.

[0102] (3) Handling pointer-type meters: Object detection: For pointer-type meters, call the object detection function to obtain the target bounding box of the dial.

[0103] Key point detection: Call the key point detection function to obtain key coordinate values ​​such as the pointer tip, the start point of the range, the end point of the range, and the center of the dial.

[0104] (4) Visual markers: Draw the target bounding box: Use a box to identify the target bounding box of the dial in the image.

[0105] Mark key points: Use dots of different colors to mark key points, such as the pointer tip, the start and end of the range, and the center of the dial.

[0106] Display Results: Show the resulting graph so that users can view it intuitively.

[0107] (5) Reading calculation: Text detection and recognition: Obtain the scale values ​​on the dial through text detection and recognition technology.

[0108] Angle calculation: Combining the coordinates of key points, the angle calculation method provided above is used to accurately calculate the reading of the pointer meter.

[0109] Return Reading: Returns the calculated reading result and analyzes whether the reading result conforms to the specifications based on the instrument instruction manual knowledge base.

[0110] In this embodiment, the intelligent agent, after receiving an input image, will understand the input content, analyze it step by step, and selectively call various tool modules. After completing the analysis, it will output the return result desired by the user. The specific implementation effect is shown in the figure.

[0111] Of course, in other embodiments, the various tool modules created can be flexibly invoked as needed.

[0112] Similarly, in other embodiments, different key points can be set depending on the type of meter.

[0113] Example 2 A meter identification intelligent analysis system, such as Figure 6 As shown, it includes: The knowledge base construction module is configured to acquire relevant meter manuals in the power field, integrate them, establish a meter vector knowledge base, and generate a large model embedded knowledge base through retrieval enhancement. The specific execution process of this module can be referred to step S1 of Implementation Example 1, which will not be repeated here. The instruction generation module is configured to acquire image data covering various types of meters, as well as the annotation information of the image data, and to form instruction data for the corresponding scene task by combining the image data and the corresponding annotation information. The scene task includes meter target detection, meter status classification, key point detection, and meter reading recognition. The execution process of this module can be referred to step S2 of Embodiment 1, which will not be repeated here. The large model fine-tuning module is configured to fine-tune the local parameters of the large model based on the low-rank adaptation algorithm. Using the instruction data, the fine-tuned large model is jointly trained to guide the fine-tuned large model to learn various metering tasks. The execution process of this module can be referred to step S3 of Embodiment 1, which will not be repeated here. The tool module creation module is configured to create corresponding tool modules based on each scenario task. Each tool module is used to execute the corresponding scenario task based on the acquired meter images or meter images and text information. The execution process of this module can be referred to step S4 of Embodiment 1, which will not be repeated here. The meter recognition and analysis module is configured to acquire user input information, extract user intent based on the input information using a jointly trained large model, call the corresponding tool module for the scene task, and use the tool module to perform scene task execution on the target meter image or the target meter image and text information to obtain the meter recognition analysis result. The specific execution process of this module can be referred to step S5 of Embodiment 1, and will not be repeated here. It is understood that the above-mentioned units / modules can be individually or entirely merged into one or more other units / modules, or one or more of the units can be further divided into multiple functionally smaller units to achieve the same operation without affecting the technical effect of the embodiments of this application.

[0114] The modules described above in this system are based on logical function division. In practical applications, the function of one module can be implemented by multiple modules, or the function of multiple modules can be implemented by one module.

[0115] Similarly, in other embodiments of this application, the system may also include other units / modules. In practical applications, these functions may also be implemented with the assistance of other units, and may be implemented by multiple units working together.

[0116] According to another embodiment of this application, the system described in this embodiment, and the method of embodiment one, can be constructed and implemented by running a computer program (including program code) capable of performing the steps involved in the corresponding method described in embodiment one on a general-purpose computing device, such as a computer, which includes processing elements and storage elements such as a central processing unit (CPU), random access memory (RAM), and read-only memory (ROM). The computer program can be recorded on, for example, a computer-readable recording medium, loaded into the aforementioned computing device through the computer-readable recording medium, and run therein.

[0117] Example 3 This implementation provides an electronic device, such as... Figure 7 As shown, the electronic device includes a processor 1001, a communication interface 1002, and a computer-readable storage medium 1003. The processor 1001, communication interface 1002, and computer-readable storage medium 1003 can be connected via a bus or other means.

[0118] The communication interface 1002 is used to receive and send data. The computer-readable storage medium 1003 can be stored in the memory of the electronic device. The computer-readable storage medium 1003 is used to store computer programs, which include program instructions. The processor 1001 is used to execute the program instructions stored in the computer-readable storage medium 1003.

[0119] The processor 1001 (or CPU (Central Processing Unit)) is the computing and control core of electronic devices. It is suitable for implementing one or more instructions, specifically for loading and executing one or more instructions to achieve corresponding methods or functions.

[0120] The processor 1001 is configured to perform the following process: Obtain relevant meter manuals in the power sector, integrate them, establish a meter vector knowledge base, and enhance the generation of a large model embedded in the knowledge base through retrieval. Acquire image data covering various types of meters, as well as the annotation information of the image data, and form instruction data for corresponding scene tasks by combining the image data and the corresponding annotation information. The scene tasks include meter target detection, meter status classification, key point detection, and meter reading recognition. The local parameters of the large model are fine-tuned based on the low-rank adaptation algorithm. The fine-tuned large model is then jointly trained using the instruction data to guide it in learning various metering tasks. Each tool module is created according to the task of each scenario. Each tool module is used to execute the corresponding task based on the acquired meter images or meter images and text information. The system acquires user input information, uses a jointly trained large model to extract user intent based on the input information, calls the corresponding tool modules for the scene task, and uses the tool modules to perform scene task execution on the target meter image or the target meter image and text information to obtain the analysis results of meter recognition.

[0121] The process of steps S1-S5 in Example 1 will not be repeated here.

[0122] Example 4: This implementation provides a computer-readable storage medium (Memory), which is a memory device in an electronic device used to store programs and data. It is understood that the computer-readable storage medium here can include both built-in storage media in the electronic device and extended storage media supported by the electronic device. The computer-readable storage medium provides storage space that stores the processing system of the electronic device.

[0123] Furthermore, this storage space also contains one or more instructions suitable for loading and execution by the processor. These instructions can be one or more computer programs (including program code). It should be noted that the computer-readable storage medium here can be high-speed RAM memory or unstable memory, such as at least one disk storage device; optionally, it can also be at least one computer-readable storage medium located remotely from the aforementioned processor.

[0124] In one embodiment, the computer-readable storage medium stores one or more instructions; the processor loads and executes the one or more instructions stored in the computer-readable storage medium to perform the following process: A smart analysis method for meter identification includes the following steps: Obtain relevant meter manuals in the power sector, integrate them, establish a meter vector knowledge base, and enhance the generation of a large model embedded in the knowledge base through retrieval. Acquire image data covering various types of meters, as well as the annotation information of the image data, and form instruction data for corresponding scene tasks by combining the image data and the corresponding annotation information. The scene tasks include meter target detection, meter status classification, key point detection, and meter reading recognition. The local parameters of the large model are fine-tuned based on the low-rank adaptation algorithm. The fine-tuned large model is then jointly trained using the instruction data to guide it in learning various metering tasks. Each tool module is created according to the task of each scenario. Each tool module is used to execute the corresponding task based on the acquired meter images or meter images and text information. The system acquires user input information, uses a jointly trained large model to extract user intent based on the input information, calls the tool module corresponding to the scene task, and uses the tool module to perform scene task execution on the target meter image or the target meter image and text information to obtain the analysis results of meter recognition. The process of steps S1-S5 in Example 1 will not be repeated here.

[0125] Example 5: This implementation provides a computer program product or computer program, which includes computer instructions stored in a computer-readable storage medium. The processor of an electronic device reads the computer instructions from the computer-readable storage medium and executes the computer instructions, causing the electronic device to perform the following process: A smart analysis method for meter identification includes the following steps: Obtain relevant meter manuals in the power sector, integrate them, establish a meter vector knowledge base, and enhance the generation of a large model embedded in the knowledge base through retrieval. Acquire image data covering various types of meters, as well as the annotation information of the image data, and form instruction data for corresponding scene tasks by combining the image data and the corresponding annotation information. The scene tasks include meter target detection, meter status classification, key point detection, and meter reading recognition. The local parameters of the large model are fine-tuned based on the low-rank adaptation algorithm. The fine-tuned large model is then jointly trained using the instruction data to guide it in learning various metering tasks. Each tool module is created according to the task of each scenario. Each tool module is used to execute the corresponding task based on the acquired meter images or meter images and text information. The system acquires user input information, uses a jointly trained large model to extract user intent based on the input information, calls the tool module corresponding to the scene task, and uses the tool module to perform scene task execution on the target meter image or the target meter image and text information to obtain the analysis results of meter recognition. The process of steps S1-S5 in Example 1 will not be repeated here.

[0126] Those skilled in the art will understand that embodiments of the present invention can be provided as methods, systems, or computer program products. Therefore, the present invention can take the form of a completely hardware embodiment, a completely software embodiment, or an embodiment combining software and hardware aspects. Furthermore, the present invention can take the form of one or more computer-usable storage media (including, but not limited to, disk storage, etc.) containing computer-usable program code. CD - ROM It takes the form of a computer program product implemented on (such as optical memory, etc.).

[0127] This invention is described with reference to flowchart illustrations and / or block diagrams of methods, apparatus (systems), and computer program products according to embodiments of the invention. It will be understood that each block of the flowchart illustrations and / or block diagrams, and combinations of blocks in the flowchart illustrations and / or block diagrams, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, special-purpose computer, embedded processor, or other programmable data processing apparatus to produce a machine, such that the instructions, which execute via the processor of the computer or other programmable data processing apparatus, generate instructions for implementing the flowchart illustrations and / or block diagrams. Figure 1 One or more processes and / or boxes Figure 1 A device that provides the functions specified in one or more boxes.

[0128] The above description is merely a preferred embodiment of the present invention and is not intended to limit the invention. Various modifications and variations can be made to the present invention by those skilled in the art. Any modifications, equivalent substitutions, improvements, etc., made by those skilled in the art without creative effort within the spirit and principles of the present invention should be included within the scope of protection of the present invention.

Claims

1. A smart analysis method for meter identification, characterized in that, Includes the following steps: Obtain relevant meter manuals in the power sector, integrate them, establish a meter vector knowledge base, and enhance the generation of a large model embedded in the knowledge base through retrieval. Acquire image data covering various types of meters, as well as the annotation information of the image data, and form instruction data for corresponding scene tasks by combining the image data and the corresponding annotation information. The scene tasks include meter target detection, meter status classification, key point detection, and meter reading recognition. The local parameters of the large model are fine-tuned based on the low-rank adaptation algorithm. The fine-tuned large model is then jointly trained using the instruction data to guide it in learning various metering tasks. Each tool module is created according to the task of each scenario. Each tool module is used to execute the corresponding task based on the acquired meter images or meter images and text information. The system acquires user input information, uses a jointly trained large model to extract user intent based on the input information, calls the corresponding tool modules for the scene task, and uses the tool modules to perform scene task execution on the target meter image or the target meter image and text information to obtain the analysis results of meter recognition.

2. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The large model selected is a large model of text and image multimodality.

3. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The process of establishing the meter vector knowledge base includes: determining the prompt words in the meter recognition field, defining the professional roles of the large model, constructing prompt words using existing prompt examples, and using the method of adding assumptions to enable the large model to give a negative response under uncertain conditions.

4. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The specific process of generating a large model embedded in the knowledge base through retrieval enhancement includes: We obtained instruction manuals for various specifications from publicly available online databases and meter manufacturers, and organized the documents according to a unified standard. The embedded model is used to convert the organized instruction manual content into vectors and store them in a vector knowledge base; Convert user queries into vectors and retrieve relevant knowledge from a vector knowledge base; An enhanced suggestion template is constructed based on the search results, and the large model generates a response by combining the enhanced suggestion template.

5. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The process of forming instruction data for a corresponding scene task from image data and corresponding annotation information includes: acquiring on-site collected meter image data or historical meter image data from the power scene, covering various types of meters, forming a dataset, and ensuring that the dataset contains images under different lighting conditions, shooting angles and / or background environments, and increasing the diversity of data through rotation, scaling, flipping and cropping operations. Each image is annotated in detail, including the instrument type, instrument status, instrument reading, and annotation information of key point locations; The aforementioned images, annotation information, and textual descriptions of the images are integrated into instruction data, which is related to the tasks of each scenario.

6. The intelligent analysis method for meter identification as described in claim 5, characterized in that, The instrument status includes normal dial, damaged dial, blurred dial, and damaged casing.

7. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The process of fine-tuning the local parameters of a large model based on the low-rank adaptation algorithm includes: Let the pre-trained parameter weights be... During the fine-tuning phase, the weights are updated: ； in, It is the size of the weights updated during the fine-tuning training phase; Based on the low-rank adaptation algorithm Perform matrix transformations: ； Where BA are matrices. , , , ; In fine-tuning training, only the right side needs to be calculated. , The matrix can be modified without changing the pre-trained weights on the left side, thus reducing the total number of weight parameters. for: ； Furthermore, the rank of the low-rank adaptation algorithm is set to 64, and the Dropout probability value is 0.

05.

8. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The tool module includes a meter status classification module, which is used to judge the meter status of the image, obtain the specific status of the meter in the current scene, and issue a warning message if the current meter has an appearance defect, indicating the specific defect type. The meter status classification module uses a multi-head classifier, which includes multiple attribute classification heads. Each attribute classification head is configured with a loss function. The total loss function of the multi-head classifier is the sum of the loss functions of each attribute classification head multiplied by their respective weight values.

9. The intelligent analysis method for meter identification as described in claim 8, characterized in that, The multi-head classifier includes four attribute classification heads. The first attribute classification head is used to distinguish dial attributes, specifically including dial normal, dial damaged, and dial blurry. The second attribute classification head is used to classify casing attributes, specifically including casing normal, casing damaged, casing dirty, and casing rusted. The third attribute classification head is used to classify target class attributes, specifically including pointer-type target type, digital display-type target type, and other target types that do not require attention. The fourth attribute category header is used to distinguish pointer attributes. Specific categories include pointer detachment, pointer exceeding limits, pointer pointing to a warning range, and pointer pointing to the minimum or maximum value of the range.

10. A meter identification intelligent analysis method as described in claim 1, 8, or 9, characterized in that, The tool module includes: The meter target detection module uses a target detection network model to achieve meter positioning and type identification. The meter key point detection module is used to detect the key point positions on the pointer meter, including the pointer tip position, pointer tail position, the start point of the range, and the end point of the range. The meter text detection and recognition module is used to detect and recognize text information on the meter, including scale values, meter number, and several units of measurement.

11. The intelligent analysis method for meter identification as described in claim 1, characterized in that, The process of obtaining user input information and extracting user intent based on the input information using a jointly trained large model includes obtaining the user input statement, determining the intent or purpose expressed in the user input statement, and identifying the intent by classifying the input statement into a predefined intent category.

12. The intelligent analysis method for meter identification as described in claim 1 or 11, characterized in that, When the user's intent is to identify the meter reading, the starting scale point A, the ending scale point B, the pointer tip point C, and the pointer tail point D are extracted. The specific steps for calculating the meter reading include: If the line segment AB connecting the starting point A and the ending point B is found, and the perpendicular bisector L of the line segment AB is found, then the center of the current dial circle must exist on the line L. By connecting the pointer tip C and the pointer tail D to obtain line segment CD and extending CD, find the intersection point of the perpendicular bisector L and the line CD. When the perpendicular bisector L is not collinear with line CD, find the unique intersection point O of L with lines AB and CD. Point O is the current center position of the dial. Calculate the ratio of ∠AOC to ∠AOB to obtain the proportion of the current pointer's pointing angle in the entire dial range. When line AB and line CD are collinear, the current pointer position angle is half of the entire dial range. After calculating the percentage of the current reading based on the coordinates of the key points, locate the scale frame and corresponding scale value closest to the starting point A and the ending point B, respectively, to obtain the current range of the meter and the actual reading of the current pointer instrument. ； in, This represents the maximum scale value of the current meter. This is the minimum scale value of the current meter.

13. The intelligent analysis method for meter identification as described in claim 12, characterized in that, When determining the starting and ending positions of the pointer and range, the quadrant should be used for judgment: when the minimum range is greater than the maximum range at an angle, the angle should be determined to be on the side with a value greater than 180 degrees. in, This refers to the starting angle of the measuring range. This refers to the angle at the end of the measurement range, and the unit is degrees.

14. The intelligent analysis method for meter identification as described in claim 12, characterized in that, For pointer-type instruments with non-uniform ranges, the identified scale values ​​are sorted in ascending order, and the center points of the scale target frame are connected in pairs in sequence. It is determined which scale range's line segment intersects with the line where the pointer is located, and the intersection point is on the line segment connecting the scales. The scale range pointed to by the pointer is then calculated, and this part of the scale range is taken as the start and end range of the entire range. The dial range value is calculated based on the included angle.

15. A meter identification intelligent analysis system, characterized in that, include: The knowledge base construction module is configured to acquire relevant meter manuals in the power field, integrate them, establish a meter vector knowledge base, and generate a large model embedded knowledge base through retrieval enhancement. The instruction generation module is configured to acquire image data covering various types of meters, as well as annotation information of the image data, and to form instruction data for the corresponding scene task by combining the image data and the corresponding annotation information. The scene task includes meter target detection, meter status classification, key point detection, and meter reading recognition. The large model fine-tuning module is configured to fine-tune the local parameters of the large model based on the low-rank adaptation algorithm, and use the instruction data to jointly train the fine-tuned large model to guide the fine-tuned large model to learn various scenario tasks of metering. The tool module creation module is configured to create corresponding tool modules based on each scenario task. Each tool module is used to execute the corresponding scenario task based on the acquired meter images or meter images and text information. The meter recognition and analysis module is configured to acquire user input information, extract user intent based on the input information using a jointly trained large model, call the tool module corresponding to the scene task, and use the tool module to perform scene task execution on the target meter image or the target meter image and text information to obtain the meter recognition analysis result.

16. A computer-readable storage medium for storing computer instructions, wherein when the computer instructions are executed by a processor, the instructions are characterized in that... Complete the steps in the intelligent analysis method for meter identification as described in any one of claims 1-14.

17. A computer program product comprising a computer program, wherein when the computer program is executed by a processor, it is characterized in that, Complete the steps in the intelligent analysis method for meter identification as described in any one of claims 1-14.

18. An electronic device comprising a memory and a processor, and computer instructions stored in the memory and running on the processor, characterized in that, When the computer instructions are executed by the processor, they complete the steps in the intelligent analysis method for meter identification as described in any one of claims 1-14.

Images