Operation equipment intelligent identification method and system for power transmission unplanned operation

Abstract

The invention relates to an intelligent identification method and system for operation equipment for power transmission unplanned operation, and relates to the field of image processing, and the method comprises the steps: receiving an image of to-be-identified operation equipment and the world coordinate information of a positioning patch disposed on the equipment; according to the coordinate information, sorting is carried out through an operation equipment first-level identification model to obtain model sets: when the number of the first-level selected operation equipment model sets is larger than 1, the first-level selected operation equipment model sets are traversed to obtain a reference image set, meeting the world coordinate information, of the first-level selected operation equipment model sets; performing similarity clustering analysis on the reference image set to obtain a reference image clustering result; and traversing the clustering results of the reference images, respectively carrying out image similarity analysis on the clustering results and the to-be-identified equipment images, and extracting and outputting the operation equipment model corresponding to the reference image with the maximum image similarity, thereby solving the technical problems that the operation equipment is difficult to accurately identify and the efficient scheduling requirement of the power transmission unplanned operation cannot be met.

Description

Technical Field

[0001] The present invention relates to the field of image processing, and in particular to an intelligent identification method and system for operating equipment used in unplanned power transmission operations. Background Art

[0002] In the field of unplanned power transmission operations, rapid and accurate identification of operating equipment is crucial to ensuring operational safety and improving operational efficiency. However, traditional operating equipment identification and scheduling mainly rely on manual guidance, which is not only time-consuming and labor-intensive, but also easily affected by human factors, resulting in low accuracy. With the rapid development of computer vision and machine learning technologies, intelligent recognition methods have gradually been applied to various fields. However, in the scenario of unplanned power transmission operations, there are many types of operating equipment, such as spanning frames, tower materials, cranes, excavators, pickup trucks, and boom trucks. The specific operations also present multimodal actions, and are often in a complex and changeable operating environment. This undoubtedly brings great challenges to intelligent recognition, making it difficult to accurately identify operating equipment and unable to meet the efficient scheduling requirements of unplanned power transmission operations. Moreover, most of the existing intelligent recognition methods use a single recognition model, which is often difficult to cope with situations with a wide variety of operating equipment and complex features. At the same time, these methods are often more general in feature extraction, lack of refined processing for the characteristics of different operating equipment, and fail to fully consider the importance of different features, resulting in the accuracy and reliability of the recognition results to be improved. Summary of the invention

[0003] The present invention aims to solve the technical problem that operating equipment in the prior art is difficult to accurately identify and cannot meet the efficient scheduling requirements of unplanned power transmission operations, and provides an intelligent identification method and system for operating equipment for unplanned power transmission operations to solve the problem.

[0004] The technical solution of the present invention to solve the above technical problems is as follows: In a first aspect, the present invention provides an intelligent identification method for operating equipment for unplanned power transmission operations, comprising: receiving an image of an operating equipment to be identified and world coordinate information of a positioning patch deployed on the operating equipment to be identified; sorting the image through a first-level identification model of the operating equipment according to the world coordinate information to obtain a first-level selected operating equipment model set, wherein the first-level identification model of the operating equipment includes multiple sub-channels, and any one of the sub-channels is machine-learned based on multiple groups of data training, and any one of the multiple groups of data includes: the world record coordinates of the positioning patch of a preset operating equipment model, and a label indicating whether it belongs to the preset operating equipment model; when the number of the first-level selected operating equipment model set is greater than 1, traversing the first-level selected operating equipment model set to obtain a reference image set of the first-level selected operating equipment model set that satisfies the world coordinate information; performing similarity clustering analysis on the reference image set to obtain a reference image clustering result; traversing the reference image clustering result, performing image similarity analysis with the image of the equipment to be identified respectively, and extracting the operating equipment model corresponding to the reference image with the maximum image similarity value for output.

[0005] In a second aspect, the present invention provides an intelligent identification system for operating equipment for unplanned power transmission operations, comprising: an information receiving module, for receiving an image of an operating equipment to be identified, and world coordinate information of a positioning patch deployed on the operating equipment to be identified; an intelligent sorting module, for sorting, according to the world coordinate information, through a first-level identification model of the operating equipment, to obtain a first-level selected operating equipment model set, wherein the first-level identification model of the operating equipment includes multiple sub-channels, and any one of the sub-channels is obtained by machine learning based on multiple groups of data training, and any one of the multiple groups of data includes: the world record coordinates of the positioning patch of a preset operating equipment model, and a label indicating whether it belongs to the preset operating equipment model; an information screening module, for traversing the first-level selected operating equipment model set when the number of the first-level selected operating equipment model set is greater than 1, and obtaining a reference image set of the first-level selected operating equipment model set that satisfies the world coordinate information; a clustering analysis module, for performing similarity clustering analysis on the reference image set to obtain a reference image clustering result; a result output module, for traversing the reference image clustering result, performing image similarity analysis with the image of the equipment to be identified, respectively, and extracting the operating equipment model corresponding to the reference image with the maximum image similarity value for output.

[0006] The beneficial effects of the present invention are as follows: by receiving the image of the operating equipment to be identified and the world coordinate information of the positioning patch, a primary recognition model including multiple sub-channels trained based on multiple groups of data is used for preliminary sorting to obtain a set of possible operating equipment models; when there are multiple possible models, the most matching operating equipment model is further determined and output through image similarity analysis, thereby achieving the technical effect of rapid and accurate identification of the operating equipment and improving the efficiency and safety of unplanned power transmission operations. BRIEF DESCRIPTION OF THE DRAWINGS

[0007] Figure 1 A schematic flow chart of a method for intelligently identifying operating equipment for unplanned power transmission operations provided by the present invention.

[0008] Figure 2 A schematic diagram of the structure of the intelligent identification system for operating equipment for unplanned power transmission operations provided by the present invention.

[0009] Explanation of the accompanying drawings: information receiving module 11, intelligent sorting module 12, information screening module 13, clustering analysis module 14, result output module 15. DETAILED DESCRIPTION

[0010] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative work are within the scope of protection of the present invention.

[0011] In the description of the present invention, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Thus, the features defined as "first" and "second" may explicitly or implicitly include one or more of the features. In the description of the present invention, the meaning of "plurality" is two or more, unless otherwise clearly and specifically defined.

[0012] In the description of the present invention, the term "for example" is used to mean "used as an example, illustration or explanation". Any embodiment described as "for example" in the present invention is not necessarily to be construed as being more preferred or advantageous than other embodiments. The following description is given to enable any technician in the field to implement and use the present invention. In the following description, details are listed for the purpose of explanation. It should be understood that a person of ordinary skill in the art can recognize that the present invention can be implemented without using these specific details. In other examples, well-known structures and processes will not be elaborated in detail to avoid obscuring the description of the present invention with unnecessary details. Therefore, the present invention is not intended to be limited to the embodiments shown, but is consistent with the widest scope consistent with the principles and features disclosed in the present invention.

[0013] Embodiment 1

[0014] like Figure 1 As shown, an embodiment of the present invention provides an intelligent identification method for operating equipment for unplanned power transmission operations, comprising: S10: Receive an image of the working device to be identified and world coordinate information of a positioning patch deployed on the working device to be identified.

[0015] S20: According to the world coordinate information, sorting is performed through the first-level identification model of the operating equipment to obtain a first-level selected operating equipment model set, wherein the first-level identification model of the operating equipment includes multiple sub-channels, and any sub-channel machine learning is obtained based on multiple groups of data training, and any group of the multiple groups of data includes: the world record coordinates of the positioning patch of the preset operating equipment model, and a label indicating whether it belongs to the preset operating equipment model.

[0016] S30: When the number of the first-level selected operating equipment model sets is greater than 1, traverse the first-level selected operating equipment model sets to obtain a reference image set of the first-level selected operating equipment model sets that satisfies the world coordinate information.

[0017] S40: performing similarity clustering analysis on the reference image set to obtain reference image clustering results.

[0018] S50: Traversing the clustering results of the reference images, performing image similarity analysis with the device images to be identified, extracting the operating device model corresponding to the reference image with the maximum image similarity and outputting it.

[0019] For example, unplanned transmission operations usually refer to transmission operations that are not planned or arranged in advance in the power system. These operations may be carried out temporarily due to emergencies, equipment failures or other urgent needs. Since they are not planned and arranged in detail in advance, they may bring certain challenges to the stable operation of the power system. In actual situations, due to the existence of various unforeseen factors, unplanned transmission operations are difficult to completely avoid. Intelligent identification of operating equipment refers to the use of advanced technical means, such as computer vision and machine learning, to automatically identify equipment at the operation site. This identification technology can quickly obtain information such as equipment type, model, and status, thereby providing strong support for subsequent operations. In unplanned transmission operations, intelligent identification of operating equipment can help workers quickly and accurately identify the equipment that needs to be operated, thereby improving work efficiency and safety.

[0020] In this embodiment, when an unplanned power transmission operation occurs, it is first necessary to obtain an image of the operating equipment to be identified at the operation site by some means (such as camera shooting). This image will be used as the basic data for the subsequent identification process, including key information such as the appearance, shape, and color of the equipment. At the same time, some positioning patches have been deployed in advance on the operating equipment. The function of these patches is to provide a fixed reference point so that the system can accurately know the position of the equipment in three-dimensional space. Each positioning patch has a corresponding world coordinate information, which describes the exact position of the patch in space. Obtaining the world coordinate information of these positioning patches is usually achieved through some positioning technology, such as GPS positioning, laser radar positioning, or measurement through a pre-set fixed reference object. These technologies can accurately give the coordinate value of each positioning patch in three-dimensional space. In the actual intelligent recognition process, the system first receives the image of the operating equipment to be identified, and then obtains the world coordinate information of the positioning patches deployed on these devices. This information will be sent together to the subsequent processing module. The processing module will sort the equipment through the first-level recognition model of the operating equipment according to the world coordinate information of the positioning patch, quickly narrow the range of the equipment to be identified, and obtain a set of possible matching operating equipment models.

[0021] Further, when the image of the equipment to be identified at the work site is captured and the world coordinate information of the positioning patch deployed on the equipment is obtained, this information is input into the first-level identification model of the working equipment. This model is a machine learning model that has been designed and trained, which is specifically used to quickly and accurately identify the model of the working equipment. The first-level identification model of the working equipment contains multiple sub-channels, which are not independent of each other, but work together to improve the accuracy and efficiency of identification. Each sub-channel is trained based on a large amount of data, which contains the world record coordinates of the positioning patch of the preset working equipment model, and a label that identifies whether these coordinates belong to a specific preset working equipment model. Specifically, each set of data in the training data set includes the world coordinate information of a positioning patch and a corresponding label. Among them, the label is a simple indicator to indicate whether the coordinate matches a specific working equipment model. For example, if the coordinates in a set of data match the coordinates of the positioning patch on a specific model of transformer, then its label will identify that the coordinate belongs to the transformer of that model. During the training process, each sub-channel will learn how to extract useful features from the input world coordinate information and predict the equipment model based on these features. Since each sub-channel is trained based on a different data set, they are able to capture subtle differences between equipment models, thereby improving recognition accuracy. When actually used, the input world coordinate information will be fed into the first-level recognition model for operating equipment, and multiple sub-channels in the model will process this information in parallel and predict possible equipment models based on their respective learned features. Ultimately, the outputs of these sub-channels will be integrated together to form a first-level selected operating equipment model set. This set contains all equipment models that the model believes may match the input world coordinate information. Through this process, the first-level recognition model for operating equipment can use world coordinate information to quickly narrow the range of equipment to be identified, providing strong support for subsequent more accurate identification.

[0022] Next, when the first-level recognition model of the operating equipment preliminarily screens out a group of possible operating equipment models (i.e., the first-level selected operating equipment model set) according to the world coordinate information, if the number of equipment models in this set is greater than 1, it means that there are multiple possible matches. In order to further determine the most accurate equipment model, the system will then perform a more refined matching process. The core of this process is to traverse the first-level selected operating equipment model set and find a corresponding reference image set that meets the world coordinate information for each equipment model in the set. The reference image set consists of a series of images of known equipment models, which are taken under different conditions, but each image contains the accurate position information of the positioning patch on the equipment model. Specifically, the system will take the first equipment model from the first-level selected operating equipment model set and then search it in its corresponding reference image set. The basis for the search is whether the positioning patch position information in the reference image matches the input world coordinate information. If the match is successful, it means that the equipment model represented by the reference image is consistent with the spatial layout of the on-site equipment, so the reference image is added to the reference image set that meets the conditions. This process will be repeated until all equipment models in the first-level selected operating equipment model set have been traversed. Finally, the system will obtain one or more reference image sets that meet the world coordinate information, each of which corresponds to a possible device model. This step not only further narrows the range of possible device models, but also provides a more accurate and targeted reference image set for subsequent image similarity analysis, which will help improve recognition accuracy and efficiency, especially in scenarios such as unplanned power transmission operations that require rapid response.

[0023] After that, similarity cluster analysis is performed on all the reference image sets that meet the world coordinate information. The purpose of this step is to classify visually similar or close reference images together to form clustering results. Cluster analysis is based on image features, such as color, texture, shape, etc., and divides different clusters by calculating the similarity between images. The result of similarity cluster analysis is one or more clusters, each of which contains visually similar reference images. These clusters provide a more accurate reference set for subsequent image similarity analysis, which helps to reduce the amount of calculation and improve the accuracy of recognition. Next, these clustering results are traversed, and image similarity analysis is performed on the reference images in each cluster with the device image to be identified. Image similarity analysis is a technology that quantitatively compares the similarity between two images. It can be achieved by calculating the distance or similarity score between image features. During the traversal process, the system will take out the reference images in each cluster one by one and calculate the similarity score between them and the device image to be identified. This score reflects the visual proximity between the reference image and the device image to be identified. The higher the score, the more similar the two images are. After the system traverses the entire clustering result, it will get a series of similarity scores. Then, find the maximum value among these scores. The reference image corresponding to this maximum value is the image that is most similar to the image of the device to be identified. Then, extract the operating equipment model represented by the reference image corresponding to this maximum similarity value and output it as the final recognition result. This output result is the operating equipment model determined by the system through a series of intelligent recognition processes based on the input image of the device to be identified and the world coordinate information. Through this process, the image similarity analysis technology can be used to accurately identify the specific model of the on-site operating equipment from multiple possible equipment models, providing key information for the subsequent processing of unplanned power transmission operations, ensuring the efficient scheduling requirements of unplanned power transmission operations, and achieving the technical effect of improving the efficiency and safety of unplanned power transmission operations.

[0024] In a preferred embodiment, the world coordinate information of the positioning patch deployed on the operating equipment to be identified includes: splitting the preset equipment model into equipment components to obtain an equipment component set; based on the equipment component set, collecting multiple equipment operation motion monitoring logs of the preset equipment model through the Internet, wherein any one of the multiple equipment operation motion monitoring logs includes an equipment component motion position timing information set; based on the equipment component motion position timing information set, traversing the multiple equipment operation motion monitoring logs to perform motion consistency clustering on the equipment component set respectively to obtain multiple equipment component clustering results; performing frequency grouping based on the multiple equipment component clustering results to obtain multiple groups of equipment components, wherein the frequency of any group of equipment components of the multiple groups of equipment components being divided into the same category in the multiple equipment component clustering results exceeds a selected frequency threshold, and the selected frequency threshold is greater than or equal to 0.9 of the number of equipment operation motion monitoring logs; randomly selecting components from the multiple groups of equipment components respectively to obtain multiple equipment components as positioning patch deployment components of the preset equipment model.

[0025] Optionally, for the preset equipment models, the equipment components are split, that is, the complex equipment is disassembled into several independent equipment components to form a set of equipment components. These components are the basic components of the equipment and the basis for subsequent analysis. Next, network technology is used to collect multiple motion monitoring logs of these preset equipment models during the operation process, each of which records in detail the motion position timing information of the equipment components in a specific time period, that is, the position change of the components at different time points. These log data provide rich materials for subsequent cluster analysis. With these log data, motion consistency cluster analysis is started, and each component in the equipment component set is clustered by traversing each equipment operation motion monitoring log. The basis for clustering is whether the motion position timing information of the components in different logs shows consistency, that is, whether the components show similar motion patterns under similar operating conditions. Through such cluster analysis, components with similar motion characteristics can be classified into one category. After clustering is completed, these clustering results are further grouped by frequency, the purpose of which is to find component combinations that show consistency in all or most logs. Specifically, if the frequency at which a group of equipment components are classified as the same type in multiple clustering results exceeds the selected frequency threshold (this threshold is usually set to 0.9 or higher of the number of equipment operation motion monitoring logs), this group of components is considered to be highly stable and representative. Finally, a number of components are randomly selected from these frequently occurring component combinations as deployment components for positioning patches. The location information of these components will be used as the basis for the deployment of positioning patches to obtain key world coordinate information. Since these components are strictly screened, their location information is not only representative, but also relatively small in number, thereby reducing the number of positioning patches and reducing the consumption of subsequent computing resources. In summary, through this series of steps, the deployment position of the positioning patch can be determined efficiently and accurately, while reducing the number of patches and improving efficiency, providing strong support for the subsequent intelligent recognition process.

[0026] In a preferred embodiment, according to the equipment component motion position timing information set, the multiple equipment operation motion monitoring logs are traversed to perform motion consistency clustering on the equipment component set respectively to obtain multiple equipment component clustering results, including: according to the multiple equipment operation motion monitoring logs, extracting the equipment component motion position timing information set of the first equipment operation motion monitoring log; performing motion similarity calculation on the equipment component motion position timing information set to obtain similarities of several component motion trajectories; based on a motion trajectory similarity threshold, clustering analysis is performed on the equipment component set in combination with the several component motion trajectory similarities to obtain a first equipment component clustering result, and adding it to the multiple equipment component clustering results.

[0027] Furthermore, key information is first extracted from multiple equipment operation motion monitoring logs. This information constitutes a set of equipment component motion position timing information, which records in detail the position changes and time series of each equipment component during the operation process. First, focus on the first equipment operation motion monitoring log, and extract the corresponding equipment component motion position timing information set from this log. This set contains the position data of all equipment components in a specific operation cycle, as well as the changes of these data over time. Next, the motion similarity of this timing information set is calculated. The purpose of this step is to quantify the similarity of different equipment components in motion trajectories. Appropriate algorithms or methods, such as dynamic time warping (DTW), Euclidean distance, etc., can be used to calculate the motion trajectory similarity between each pair of components. Here, since the timing information of the motion trajectory may not be aligned, dynamic time warping time calculation is preferred to improve the accuracy of the comparison. The main output of DTW is a distance value, which represents the minimum cumulative distance between two time series. The smaller this distance value is, the more similar the two time series are after time warping. The specific form can be more intuitively represented by converting the distance into similarity. The calculation formula is: ;in, is a positive scaling parameter. In this way, the distance is converted to similarity, and the closer the value is to 1, the more similar it is. Set The reason is that by adjusting You can control the sensitivity of the similarity score to changes in the DTW distance. Makes the similarity score more sensitive to distance changes, while smaller This makes the similarity score less sensitive to distance changes, facilitating efficient calculations. For example, assuming that the DTW distance between two time series X and Y is D, their similarity can be calculated as ,if =0.1, distance D=10, then after calculation = ≈0.3679. Through these calculations, a set of similarities of motion trajectories of several components can be obtained, where each similarity value represents the degree of proximity of a pair of components in motion trajectories. With these similarity values, cluster analysis can be performed. A motion trajectory similarity threshold is set, which is used to determine which components should be classified into the same category. Specifically, if the motion trajectory similarity between two components is higher than this threshold, they are considered to be consistent in motion and should be classified into one category. Based on this principle, the device component set is clustered in combination with the similarities of several component motion trajectories, thereby obtaining the device component clustering result corresponding to the first device operation motion monitoring log. This clustering result contains several categories, each of which contains a group of device components that are consistent in motion. This result is added to multiple device component clustering results as a preliminary analysis result. Subsequently, the above process is repeated to perform the same processing on the remaining device operation motion monitoring logs. Each log contributes a device component clustering result, and these results are added to multiple device component clustering results. Finally, we get a set of clustering results corresponding to all logs, which provides comprehensive information about the motion consistency of device components. This process can not only effectively identify which device components have consistent motion, but also provide strong support for subsequent positioning patch deployment and intelligent recognition processes.

[0028] In a preferred embodiment, any one of the sub-channel machine learning is obtained based on multiple sets of data training, and any one of the multiple sets of data includes: the world record coordinates of the positioning patch of the preset operation equipment model and the label indicating whether it belongs to the preset operation equipment model, including: constructing the equipment action recognition loss function: , ,in, Characterizes the device action recognition loss value, A label indicating whether the device belongs to the preset operation device model. 1 indicates yes, and 0 indicates no. Characterize the judgment results of the training output, is the dynamic weight, Characterize the probability of predicting the equipment model of the preset operation, Represents the average value of the predicted probability of all samples, A hyperparameter that characterizes the speed at which the weights are controlled to change; configuring the same number of input nodes according to the number of world record coordinates of the positioning patch of the preset operating equipment model, constructing a convolutional layer, a pooling layer, and a fully connected layer that are connected in sequence, and obtaining a sub-channel topology, wherein the fully connected layer includes an output layer, and the output layer includes two output nodes, 1 and 0; using the label indicating whether the identifier belongs to the preset operating equipment model as supervision, using the world record coordinates of the positioning patch of the preset operating equipment model as input, training the sub-channel topology, and generating the sub-channel.

[0029] For example, in machine learning, in order to train a subchannel that can identify the preset operating equipment model, a training method based on multiple sets of data is used. Each set of data contains two key pieces of information: one is the recorded coordinates of the positioning patch of the preset operating equipment model in the world coordinate system, which are used as input features; the other is the label that identifies whether these coordinates belong to the preset operating equipment model. Furthermore, in order to evaluate the accuracy of the model in identifying equipment actions, a device action recognition loss function is constructed: , In this function Characterizes the device action recognition loss value, A label indicating whether the device belongs to the preset operation device model. 1 indicates yes, and 0 indicates no. Characterize the judgment results of the training output, is the dynamic weight, Characterize the probability of predicting the equipment model of the preset operation, Represents the average value of the predicted probability of all samples, Characterizes the hyperparameter that controls the speed of weight change. Several key elements in the function are: label value (true value), judgment result of training output (predicted value), dynamic weight, and hyperparameter k that controls the speed of weight change. The dynamic weight is calculated based on the probability of predicting that it belongs to the preset operating device model and the average of the predicted probabilities of all samples, and its change speed is affected by the k value. Larger k values ​​will make the weight change more drastic, and the model will pay more attention to samples that are difficult to classify; while smaller k values ​​will make the weight change more gradually. For practical applications, a default k value is usually set, such as k=2, to balance the sensitivity and stability of the model. Subsequently, the same number of input nodes is configured according to the number of world record coordinates of the positioning patch of the preset operating device model. Then, a neural network structure is constructed as a subchannel topology, which connects the convolution layer, the pooling layer, and the fully connected layer in sequence. The number of nodes in the input layer matches the feature dimension of the operating equipment action image, that is, the number of coordinates of the positioning patch is consistent; the convolution layer contains multiple layers of convolution operations, each layer has a different number of convolution kernels, usually 3x3 in size, with a stride of 1 and a padding of 1, to extract features from the image. For example, the first layer has 64 convolution kernels, the second layer has 128, and so on, until the fourth layer has 512 convolution kernels. Each convolution layer is followed by a 2x2 maximum pooling layer with a stride of 2, which is used to reduce the dimension of the feature map and reduce the amount of calculation while retaining important features. Finally, the fully connected layer contains two layers, each with 4096 neurons, which are used to integrate the features extracted by the convolution layer and the pooling layer and classify them. The last layer is the output layer, which contains 2 neurons, corresponding to the two classification results of "yes" and "no". At the same time, the ReLU activation function is used in the training process of the neural network to increase the nonlinear expression ability of the model, and the Dropout layer is added between the fully connected layers to reduce the risk of overfitting. During training, the label that identifies whether it belongs to the preset working equipment model is used as the supervision information, and the world record coordinates of the positioning patch of the preset working equipment model are used as the input data. The network parameters are continuously adjusted through the back propagation algorithm, and finally a sub-channel that can accurately identify the preset working equipment model is generated.

[0030] In a preferred embodiment, the traversal of the reference image set, performing image similarity analysis with the device image to be identified respectively, extracting the operating equipment model corresponding to the reference image with the maximum image similarity for output, includes: obtaining a first reference image of the reference image set; activating an associated feature extraction channel to perform feature extraction according to the operating equipment model corresponding to the first reference image, and obtaining a first reference image feature extraction result; extracting features of the device image to be identified according to the associated feature extraction channel, and obtaining a feature extraction result of the device image to be identified; calculating the first reference image similarity between the feature extraction result of the first reference image and the feature extraction result of the device image to be identified, and adding them to the image similarity set; extracting the operating equipment model corresponding to the reference image with the maximum value of the image similarity set for output.

[0031] Optionally, a first reference image is obtained from the reference image set, and this image represents a certain specific operating equipment model. Then, according to the operating equipment model corresponding to this reference image, a feature extraction channel associated with it is activated. This channel has been pre-trained and optimized to efficiently extract key feature information from the image. The first reference image is feature extracted using this channel to obtain the feature extraction result of the first reference image. Then, the same feature extraction channel is used to extract features from the image of the device to be identified to obtain the feature extraction result of the image of the device to be identified. These two feature extraction results contain key information or feature vectors in the image. These feature vectors are mathematical descriptions of the image in a certain specific representation space. They can capture key visual elements in the image, such as edges, textures, shapes, colors, etc., as well as the spatial relationship and hierarchy between these elements, and can be used for subsequent similarity calculations. Next, the similarity between the feature extraction result of the first reference image and the feature extraction result of the image of the device to be identified is calculated. This similarity value reflects the distance or similarity between the two images in the feature space. This similarity value is added to the image similarity set for subsequent comparison and analysis. Repeat the above process to traverse all the reference images in the reference image set, and perform feature extraction and similarity calculation with the image of the device to be identified. The similarity value obtained each time will be added to the image similarity set. Finally, the maximum value is extracted from the image similarity set. The reference image corresponding to this maximum value is the image most similar to the image of the device to be identified. According to the operating equipment model corresponding to this reference image, it is output as the recognition result of the image of the device to be identified. The entire process realizes accurate recognition of the image of the device to be identified through feature extraction and similarity calculation, providing strong support for subsequent automated processing and decision-making.

[0032] In a preferred embodiment, according to the working equipment model corresponding to the first reference image, activating the associated feature extraction channel to perform feature extraction to obtain the first reference image feature extraction result, the previous step includes: training the associated feature extraction channel according to the preset working equipment model; wherein, when the preset working equipment model belongs to a crossing frame, the feature extraction dimensions include frame structure features, height features, and connection node features; when the preset working equipment model belongs to a tower material, the feature extraction dimensions include cross-sectional shape features, surface texture features, length features, and curvature features; when the preset working equipment model belongs to a crane, the feature extraction dimensions include boom features, body features, and counterweight features; when the preset working equipment model belongs to an excavator, the feature extraction dimensions include bucket features, arm features, and track features; when the preset working equipment model belongs to a pickup truck, the feature extraction dimensions include vehicle front features, cargo box features, and tire features; when the preset working equipment model belongs to a boom truck, the feature extraction dimensions include boom features, working bucket features, and chassis features.

[0033] Furthermore, in order to ensure accuracy, different feature extraction dimensions are defined for different types of operating equipment to ensure that the model can capture the most discriminative information. For example, when the preset operating equipment model is a spanning frame, the feature extraction dimensions include frame structure features (such as frame profile, number and spacing of vertical and horizontal poles, etc.), height features (the overall height of the spanning frame), and connection node features (shape, position, and other information of the connection node). For tower materials, the feature extraction dimensions focus on cross-sectional shape features (cross-sectional shape of tower materials), surface texture features (texture pattern on the surface of tower materials), length features, and curvature features (length of tower materials and whether they are curved). In the case of cranes, the feature extraction dimensions include boom features (length, shape, and inclination angle of the boom), body features (shape, size, and color of the body), and counterweight features (shape, position, and number of counterweights). For excavators, the focus is on bucket features (bucket shape, size, and capacity), boom features (boom length, shape, and range of motion), and track features (track shape, size, and width). When the preset working equipment model is a pickup truck, the feature extraction dimensions include front features (front shape, size and brand logo), cargo box features (cargo box shape, size and loading capacity) and tire features (tires size, number and pattern). For a boom truck, the feature extraction dimensions include arm features (arm length, shape and reach), bucket features (bucket size, shape and load capacity) and chassis features (chassis shape, size and suspension system). During the training process, a large amount of annotated image data is used, which contains different instances of the various working equipment models mentioned above. Through supervised learning, the model learns how to map the input image to these predefined feature dimensions, so that the key information related to the equipment model can be accurately extracted. Once the training is completed, these associated feature extraction channels can be used to process new equipment images to be identified. The corresponding feature extraction channel is activated according to the working equipment model corresponding to the first reference image, and the feature vector of the first reference image is extracted. This feature vector is then used to compare with the feature vector of the equipment image to be identified to calculate the similarity between them. The whole process not only improves the accuracy and efficiency of equipment identification, but also provides strong support for subsequent automated processing and decision-making. By defining specific feature extraction dimensions for different types of operating equipment, it is possible to ensure that the model can still maintain excellent performance when processing complex and changing on-site images.

[0034] In a preferred embodiment, the calculation of the first reference image similarity between the first reference image feature extraction result and the image feature extraction result of the device to be identified includes: when the operating equipment type is a spanning frame, the first reference image similarity is a weighted mean calculation result of frame structure similarity, height similarity, and connection node similarity, and the weighted weight is assigned by the Delphi method; when the operating equipment type is a tower material, the first reference image similarity is a weighted mean calculation result of cross-sectional shape similarity, surface texture similarity, length and curvature similarity, and the weighted weight is assigned by the Delphi method; when the operating equipment type is a crane, the first reference image similarity is a weighted mean calculation result of boom similarity, The weighted mean calculation result of the vehicle body similarity and the counterweight block similarity, and the weighted weight is assigned by the Delphi method; when the operating equipment type is an excavator, the first benchmark image similarity is the weighted mean calculation result of the bucket similarity, the arm similarity, and the track similarity, and the weighted weight is assigned by the Delphi method; when the operating equipment type is a pickup truck, the first benchmark image similarity is the weighted mean calculation result of the front similarity, the cargo box similarity, and the tire similarity, and the weighted weight is assigned by the Delphi method; when the operating equipment type is a boom truck, the first benchmark image similarity is the weighted mean calculation result of the boom similarity, the working bucket similarity, and the chassis similarity, and the weighted weight is assigned by the Delphi method.

[0035] Exemplarily, in order to calculate the similarity between the feature extraction results of the first reference image and the feature extraction results of the image of the device to be identified, different weighted mean calculation methods are used according to the type of operating equipment. A series of feature similarity indicators are defined for each type of operating equipment, which reflect the degree of proximity between the two images in specific feature dimensions. When the operating equipment type is a spanning frame, the frame structure similarity (measuring the similarity of the frame outline, the number and spacing of vertical and horizontal poles), the height similarity (measuring the similarity of the spanning frame height) and the connection node similarity (measuring the similarity of the shape and position of the connection node) are calculated. These similarity indicators are weighted by the weighted weights determined by the Delphi method, and then the weighted mean is calculated as the first reference image similarity. The specific calculation formula for the frame structure similarity is: ;in, and are the lengths of the i-th vertical pole or horizontal pole in the real-time image and the reference image, respectively; n is the number of vertical poles or horizontal poles, is the average length. The specific calculation formula for high similarity is: ;in, and are the heights of the brackets in the real-time image and the reference image, is the preset maximum height. The specific calculation formula for the connection node similarity is: ;in, is the number of connected nodes that match between the live image and the reference image, is the total number of connected nodes in the reference image. For tower materials, the cross-sectional shape similarity, surface texture similarity, length and curvature similarity are calculated, and the weighted weights determined by the Delphi method are also used for weighting to obtain the weighted mean as the similarity result. The specific calculation formula for cross-sectional shape similarity is: ,in It is the contour distance between the cross-sectional shape of the tower material in the real-time image and the reference image, measured by Frechet distance. Frechet distance is often used to measure the similarity between two curves or shapes. The smaller the value, the more similar the cross-sectional shapes are. The closer it is to 1. The specific calculation formula for surface texture similarity is: ,here and They are the grayscale co-occurrence matrices of the tower material surface in the real-time image and the reference image, respectively. The grayscale co-occurrence matrix is ​​a statistical method used to describe the texture features in an image. The formula measures the similarity of surface textures by calculating the ratio of the dot product of two matrices to the product of their norms. The closer the value is to 1, the more similar the textures are. The specific calculation formulas for length and curvature similarity are: ,in and is the length of the tower material, and is the curvature. This formula takes into account the difference between the tower material length and curvature and the reference value to calculate the similarity. The smaller the difference, the The closer it is to 1. In the case of cranes, consider the boom similarity (including similarity in length, shape and tilt angle), body similarity (similarity in shape, size and color), and counterweight similarity (similarity in shape, position and quantity), and use the corresponding weighted weights to calculate the weighted mean. The specific calculation formula for boom similarity is: ,in, and are the actual angle and reference angle of the boom respectively; and are the actual length of the boom and the reference length respectively. This formula takes into account the cosine value of the boom angle difference and the relative value of the length difference to calculate the boom similarity. The specific calculation formula for the body similarity is: ,in, and are the width and height of the vehicle body respectively, the subscript "real" indicates the actual value, and "bench" indicates the benchmark value. This formula measures the similarity of the vehicle body by calculating the sum of the relevant ratios of the width and height of the vehicle body respectively. The specific calculation formula for the similarity of the counterweight block is: ,in, and They are the actual distance between the counterweight and the vehicle body and the reference distance. is the average distance. This formula calculates the counterweight similarity based on the ratio of the difference between the counterweight and the vehicle body to the average distance. For excavators, the bucket similarity, arm similarity and track similarity are calculated using the weighted mean method, and the weights are determined by the Delphi method. The specific calculation formula for bucket similarity is: ,in, and They are the actual area of ​​the bucket and the reference area. This formula measures the similarity of buckets by calculating the product of the actual bucket area and the reference bucket area and the ratio of the sum of the squares of the two. The closer the value is to 0.5, the higher the similarity. The specific calculation formula for boom similarity is: ,in is the boom length, is the boom angle. This formula takes into account the relative difference between the actual value and the reference value of the boom length and angle. The smaller the difference, the The closer it is to 1, the higher the boom similarity. The specific calculation formula for track similarity is: ,in and are the actual width and reference width of the track, is the average width. This formula calculates the track similarity based on the ratio of the difference between the actual track width and the reference width to the average width. The smaller the difference, the The closer it is to 1, the higher the track similarity. When the operating equipment is a pickup truck, the similarity of the head, cargo box and tires is evaluated, and the similarity is calculated using the weighted mean and the weight determined by the Delphi method. The specific calculation formula for the head similarity is: ,The similarity is measured by calculating the ratio of the difference in the coordinates of the points on the front contour to the area of ​​the front. The smaller the difference in coordinates, The closer it is to 1, the higher the similarity of the front of the vehicle. The specific calculation formula for the similarity of the cargo box is: , calculated based on the relationship between the actual volume of the cargo box and the reference volume. The closer the value is to 0.5, the higher the cargo box similarity. The specific calculation formula for tire similarity is: , calculated based on the ratio of the difference between the actual tire radius and the reference radius to the average radius. The smaller the radius difference, The closer it is to 1, the higher the tire similarity. Finally, for the boom truck, the similarity of the boom, the similarity of the working bucket, and the similarity of the chassis are calculated. The weighted mean and the Delphi method weight are also used to determine the similarity. The specific calculation formula for the boom similarity is: , combined with the cosine value product of the bucket arm angle and the relative value of the length difference to calculate, the closer the angle and length are to the reference value, The closer it is to 1, the higher the bucket arm similarity. The specific calculation formula for the bucket similarity is: , calculated based on the relationship between the actual capacity of the working bucket and the benchmark capacity, the closer the value is to 0.5, the higher the similarity of the working bucket. The specific calculation formula for chassis similarity is: , measured by calculating the ratio of the difference between the width of each part of the chassis and the reference width relative to the average width. The smaller the width difference, The closer it is to 1, the higher the chassis similarity. In summary, the Delphi method plays a key role in this process. It can reasonably assign weights to each feature similarity indicator based on expert opinions. This method takes into account the importance of different features in identifying specific operating equipment models, thereby improving the accuracy and reliability of similarity calculations. By calculating the weighted mean, a comprehensive similarity score can be obtained, which reflects the degree of proximity between the image of the device to be identified and the first reference image in the feature space. This score can then be used to determine the model of the device to be identified, and select the device model with the highest similarity to the first reference image as the identification result. In general, this method combines feature extraction and weighted mean similarity calculation to provide an efficient and accurate method for automatic identification of operating equipment models.

[0036] In a preferred embodiment, the benchmark image clustering results are traversed, image similarity analysis is performed with the device images to be identified respectively, and the operating equipment model corresponding to the benchmark image with the maximum image similarity is extracted for output, including: extracting the first centroid benchmark image of the first category benchmark image set of the benchmark image clustering results until the Lth centroid benchmark image of the Lth category benchmark image set is extracted; traversing the first centroid benchmark image until the Lth centroid benchmark image, performing image similarity analysis with the device images to be identified respectively, and extracting a selected category benchmark image set of the category to which the centroid benchmark image with the maximum image similarity belongs; traversing the selected category benchmark image set, performing image similarity analysis with the device images to be identified respectively, and extracting the operating equipment model corresponding to the benchmark image with the maximum image similarity for output.

[0037] Specifically, obtain the benchmark images for clustering processing, which represent the models of different working equipment and are divided into L categories. Each category contains a series of benchmark images similar to the equipment models of the category. In order to simplify the calculation and improve efficiency, a representative image, namely the centroid benchmark image, is extracted from the benchmark images of each category. The centroid benchmark image can be regarded as the average or center of all the benchmark images of the category, and it can represent the characteristics of the entire category to a certain extent. Next, traverse these centroid benchmark images, starting from the first centroid benchmark image in the first category benchmark image set to the Lth centroid benchmark image in the Lth category benchmark image set, and perform image similarity analysis on each centroid benchmark image and the equipment image to be identified one by one. This process is achieved by comparing the features of the two images (such as color, texture, shape, etc.), with the aim of finding a centroid benchmark image that is most similar to the equipment image to be identified. Once the centroid benchmark image with the highest similarity to the equipment image to be identified is found, the category to which it belongs is determined, that is, the selected class benchmark image set. This set contains all benchmark images that are similar to the centroid benchmark image (i.e., belong to the same category). Then, the traversal is performed again, but this time within the selected class reference image set. Each reference image in the selected class is subjected to image similarity analysis with the image of the device to be identified to find the reference image with the highest similarity. The purpose of this step is to further refine the matching results to ensure that the final output operating equipment model is the most consistent with the image of the device to be identified. Finally, the operating equipment model corresponding to the reference image with the highest similarity to the image of the device to be identified is extracted and output as the recognition result. This model is the most likely model of the device to be identified. The entire process achieves accurate identification of the device model by gradually narrowing the search range and improving the matching accuracy.

[0038] In a preferred embodiment, extracting a first centroid reference image of a first category of reference image set of the reference image clustering result includes: extracting the first reference image of the first category of reference image set, and a first group of image similarity sets between images within the category and the first reference image, performing mean calculation to obtain a first similarity mean; until a Qth similarity mean is calculated; extracting the reference image with the maximum value from the first similarity mean to the Qth similarity mean, and setting it as the first centroid reference image.

[0039] Furthermore, the clustering result of a reference image includes multiple image categories. Here, the process focuses on the first category reference image set. Specifically, all reference images belonging to the first category are identified and extracted from the clustering results to form a set. In this set, a reference image is arbitrarily selected as the starting point, which is called the first reference image. Then, the similarity between the first reference image and all other images in the set is calculated to obtain a similarity set, which contains the similarity values ​​between the first reference image and each image in the set. After that, the mean of the above similarity set is calculated to obtain a similarity mean, which is the first similarity mean. This mean reflects the average similarity between the first reference image and other images in the set. At this time, the above steps are not repeated, but this time another reference image in the set that has not been calculated as a starting point is selected as a new starting point, and its similarity with other images in the set is repeatedly calculated to obtain a new similarity mean. This process continues until each reference image in the set is selected as a starting point at least once and the corresponding similarity mean is calculated. Assuming that Q such calculations are performed, Q similarity means are obtained, and then the similarity means from the first to the Qth are obtained. After all the similarity means are obtained, the sizes of these means are compared to find the maximum value. The reference image corresponding to the maximum similarity mean is the image with the highest average similarity with other images in the set among all the reference images. This reference image is set as the centroid reference image of the first category of reference image set. In summary, by calculating the similarity mean between each reference image in the set and other images and finding the reference image with the largest similarity mean, a centroid reference image that can best represent this type of image is determined, which improves the accurate baseline of the comparison and ensures the accuracy of subsequent results.

[0040] In a preferred embodiment, the traversal of the benchmark image clustering results, performing image similarity analysis with the equipment images to be identified respectively, extracting the operating equipment model corresponding to the benchmark image with the maximum image similarity and outputting it, also includes: when the number of operating equipment models corresponding to the benchmark image with the maximum image similarity is greater than 1, analyzing multiple groups of screened operating equipment models of the equipment images to be identified at multiple moments; taking the operating equipment model row with the highest frequency of occurrence among the multiple groups of screened operating equipment models and outputting it.

[0041] Specifically, when the reference image with the greatest similarity to the image of the device to be identified is not unique, but there are multiple ones, and these images correspond to different operating equipment models. This means that the features of the image of the device to be identified have a high degree of similarity with multiple reference images, making it difficult for the system to make a unique judgment. In order to solve this problem, the system will further analyze the images of the device to be identified at multiple times, and perform similarity analysis and screening of the operating equipment models for the images at each time. In this way, multiple groups of results for screening operating equipment models can be obtained. Finally, the operating equipment model with the highest frequency in these multiple groups of screening results is counted. Since this model has been identified in the images of the device to be identified at multiple times, it can be considered to be the most likely correct answer. Therefore, the operating equipment model with the highest frequency is output as the final recognition result. The entire process achieves accurate recognition of the image of the device to be identified by traversing the clustering results of the reference images, performing image similarity analysis, handling the situation where the maximum similarity value is not unique, and counting the operating equipment model with the highest frequency.

[0042] The intelligent identification method for operating equipment for unplanned power transmission operations provided by the embodiment of the present invention has at least the following technical effects: 1. By combining technologies such as feature extraction, image similarity analysis, and weighted mean calculation, key features can be efficiently extracted from the image of the device to be identified and accurately matched with the images in the reference image set. This method not only improves the accuracy of recognition, but also greatly shortens the recognition time, allowing the system to quickly respond and output the correct operating equipment model.

[0043] 2. For different types of operating equipment, the system defines specific feature extraction dimensions and similarity calculation indicators, and determines reasonable weightings through the Delphi method. This flexible design enables the system to adapt to complex and changing on-site environments and accurately identify operating equipment of various shapes, sizes and colors. At the same time, when the maximum similarity value is not unique, the system can further analyze the image data at multiple times to improve the robustness and reliability of recognition.

[0044] 3. Through automatic identification and output of operating equipment models, manual intervention and error rates can be significantly reduced, and overall automated processing and decision-making efficiency can be improved. This is especially important for operating equipment identification tasks that require rapid response and efficient management, which helps reduce operating costs and improve production efficiency and safety.

[0045] Embodiment 2

[0046] like Figure 2As shown, based on the same inventive concept as the method for intelligently identifying operating equipment for unplanned power transmission operations provided in Embodiment 1, an embodiment of the present invention also provides an intelligent identification system for operating equipment for unplanned power transmission operations, the system comprising: The information receiving module 11 is used to receive the image of the working device to be identified and the world coordinate information of the positioning patch deployed on the working device to be identified.

[0047] The intelligent sorting module 12 performs sorting through the first-level identification model of the operating equipment according to the world coordinate information to obtain a first-level selected operating equipment model set, wherein the first-level identification model of the operating equipment includes multiple sub-channels, and any sub-channel machine learning is obtained based on multiple groups of data training, and any group of the multiple groups of data includes: the world record coordinates of the positioning patch of the preset operating equipment model, and a label indicating whether it belongs to the preset operating equipment model.

[0048] The information screening module 13 is used to traverse the first-level selected operating equipment model set when the number of the first-level selected operating equipment model set is greater than 1, and obtain a reference image set of the first-level selected operating equipment model set that satisfies the world coordinate information.

[0049] The cluster analysis module 14 is used to perform similarity cluster analysis on the reference image set to obtain a reference image clustering result; The result output module 15 is used to traverse the reference image clustering results, perform image similarity analysis with the device images to be identified, extract the operating device model corresponding to the reference image with the maximum image similarity, and output it.

[0050] Furthermore, the information receiving module 11 is further configured to perform the following steps: The preset equipment model is split into equipment components to obtain an equipment component set; based on the equipment component set, multiple equipment operation motion monitoring logs of the preset equipment model are collected online, wherein any one of the multiple equipment operation motion monitoring logs includes a set of equipment component motion position timing information; based on the set of equipment component motion position timing information, the multiple equipment operation motion monitoring logs are traversed to perform motion consistency clustering on the equipment component set respectively to obtain multiple equipment component clustering results; frequency grouping is performed based on the multiple equipment component clustering results to obtain multiple groups of equipment components, wherein the frequency of any one group of equipment components of the multiple groups of equipment components being divided into the same category in the multiple equipment component clustering results exceeds a selected frequency threshold, and the selected frequency threshold is greater than or equal to 0.9 of the number of equipment operation motion monitoring logs; random components are selected from the multiple groups of equipment components respectively to obtain multiple equipment components as positioning patch deployment components of the preset equipment model.

[0051] Furthermore, the information receiving module 11 is further configured to perform the following steps: According to the multiple equipment operation motion monitoring logs, extract the equipment component motion position timing information set of the first equipment operation motion monitoring log; perform motion similarity calculation on the equipment component motion position timing information set to obtain the similarities of several component motion trajectories; based on the motion trajectory similarity threshold, perform cluster analysis on the equipment component set in combination with the several component motion trajectory similarities to obtain a first equipment component clustering result, and add it to the multiple equipment component clustering results.

[0052] Furthermore, the intelligent sorting module 12 is further configured to perform the following steps: Construct the device action recognition loss function: , ,in, Characterizes the device action recognition loss value, A label indicating whether the device belongs to the preset operation device model. 1 indicates yes, and 0 indicates no. Characterize the judgment results of the training output, is the dynamic weight, Characterize the probability of predicting the equipment model of the preset operation, Represents the average value of the predicted probability of all samples, A hyperparameter that characterizes the speed at which the weights are controlled to change; configuring the same number of input nodes according to the number of world record coordinates of the positioning patch of the preset operating equipment model, constructing a convolutional layer, a pooling layer, and a fully connected layer that are connected in sequence, and obtaining a sub-channel topology, wherein the fully connected layer includes an output layer, and the output layer includes two output nodes, 1 and 0; using the label indicating whether the identifier belongs to the preset operating equipment model as supervision, using the world record coordinates of the positioning patch of the preset operating equipment model as input, training the sub-channel topology, and generating the sub-channel.

[0053] Furthermore, the result output module 14 is further configured to perform the following steps: Obtain a first reference image of the reference image set; according to the operating equipment model corresponding to the first reference image, activate the associated feature extraction channel to perform feature extraction, and obtain a first reference image feature extraction result; extract the features of the image of the device to be identified according to the associated feature extraction channel, and obtain a feature extraction result of the image of the device to be identified; calculate the first reference image similarity between the feature extraction result of the first reference image and the feature extraction result of the image of the device to be identified, and add them to the image similarity set; extract the operating equipment model corresponding to the reference image with the maximum value of the image similarity set and output it.

[0054] Furthermore, the result output module 15 is further configured to perform the following steps: Extract the first centroid reference image of the first category of reference image set of the reference image clustering result, until the Lth centroid reference image of the Lth category of reference image set is extracted; traverse the first centroid reference image until the Lth centroid reference image, perform image similarity analysis with the image of the device to be identified, and extract the selected category reference image set of the category to which the centroid reference image with the maximum image similarity belongs; traverse the selected category reference image set, perform image similarity analysis with the image of the device to be identified, and extract the operating equipment model corresponding to the reference image with the maximum image similarity for output.

[0055] Furthermore, the result output module 15 is further configured to perform the following steps: Extract the first reference image of the first class of reference image sets, and the first group of image similarity sets between the images in the class and the first reference image, perform mean calculation, and obtain a first similarity mean; until the Qth similarity mean is calculated; extract the reference image with the maximum value from the first similarity mean to the Qth similarity mean, and set it as the first centroid reference image. Further, the result output module 15 is also used to perform the following steps: When the number of operating equipment models corresponding to the reference image with the maximum image similarity value is greater than 1, multiple groups of screened operating equipment models of the equipment images to be identified at multiple moments are analyzed; and the operating equipment model row with the highest frequency of appearance among the multiple groups of screened operating equipment models is output.

[0056] Through the above-mentioned detailed description of the method for intelligent identification of operating equipment for unplanned power transmission operations in this specification, those skilled in the art can clearly understand the intelligent identification system for operating equipment for unplanned power transmission operations in this embodiment. For the system disclosed in the embodiment, since it corresponds to the method disclosed in the embodiment, the description is relatively simple, and the relevant parts can be referred to the description of the method part.

[0057] The above description of the disclosed embodiments enables those skilled in the art to implement or use the present application. Various modifications to these embodiments will be apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present application. Therefore, the present application will not be limited to the embodiments shown herein, but will conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An intelligent identification method for operating equipment in unplanned power transmission operations, characterized in that: include: Receiving an image of the working device to be identified and world coordinate information of a positioning patch deployed on the working device to be identified; According to the world coordinate information, sorting is performed through the first-level identification model of the operating equipment to obtain a first-level selected operating equipment model set, wherein the first-level identification model of the operating equipment includes multiple sub-channels, and any sub-channel machine learning is obtained based on multiple groups of data training, and any group of the multiple groups of data includes: the world record coordinates of the positioning patch of the preset operating equipment model and a label indicating whether it belongs to the preset operating equipment model; When the number of the first-level selected operating equipment model sets is greater than 1, traverse the first-level selected operating equipment model sets to obtain a reference image set of the first-level selected operating equipment model sets that satisfies the world coordinate information; The reference image set is traversed, image similarity analysis is performed with the device image to be identified respectively, and the operating device model corresponding to the reference image with the maximum image similarity is extracted and outputted.

2. The method according to claim 1, characterized in that And the world coordinate information of the positioning patch deployed on the working equipment to be identified, including: Decompose the preset device model into device components to obtain a device component set; According to the equipment component set, a plurality of equipment operation motion monitoring logs of the preset equipment model are collected online, wherein any one of the plurality of equipment operation motion monitoring logs includes a set of equipment component motion position timing information; According to the equipment component motion position timing information set, traverse the multiple equipment operation motion monitoring logs to perform motion consistency clustering on the equipment component set respectively to obtain multiple equipment component clustering results; Performing frequency grouping according to the clustering results of the multiple equipment components to obtain multiple groups of equipment components, wherein the frequency of any group of equipment components of the multiple groups of equipment components being classified as the same type in the clustering results of the multiple equipment components exceeds a selected frequency threshold, and the selected frequency threshold is greater than or equal to 0.9 of the number of equipment operation motion monitoring logs; Randomly select components from the multiple groups of device components respectively to obtain multiple device components as positioning patch deployment components of the preset device model.

3. The method according to claim 2, characterized in that According to the equipment component motion position timing information set, the plurality of equipment operation motion monitoring logs are traversed to perform motion consistency clustering on the equipment component set respectively, and a plurality of equipment component clustering results are obtained, including: Extracting a set of equipment component motion position timing information of a first equipment operation motion monitoring log according to the plurality of equipment operation motion monitoring logs; Performing motion similarity calculation on the set of motion position sequence information of the device components to obtain similarities of motion trajectories of several components; Based on the motion trajectory similarity threshold, cluster analysis is performed on the device component set in combination with the motion trajectory similarities of the plurality of components to obtain a first device component clustering result, which is added to the plurality of device component clustering results.

4. The method according to claim 1, characterized in that Any sub-channel machine learning is obtained based on multiple sets of data training, and any set of the multiple sets of data includes: the world record coordinates of the positioning patch of the preset operation equipment model and the label indicating whether it belongs to the preset operation equipment model, including: Construct the device action recognition loss function: ， ， in, Characterizes the device action recognition loss value, A label indicating whether the device belongs to the preset operation device model. 1 indicates yes, and 0 indicates no. Characterize the judgment results of the training output, is the dynamic weight, Characterize the probability of predicting the equipment model of the preset operation, Represents the average value of the predicted probability of all samples, Characterizes the hyperparameters that control how fast the weights change; According to the number of world record coordinates of the positioning patch of the preset operation equipment model, the same number of input nodes are configured, and a convolution layer, a pooling layer and a fully connected layer connected in sequence are constructed to obtain a sub-channel topology, wherein the fully connected layer includes an output layer, and the output layer includes two output nodes of 1 and 0; The sub-channel topology is trained by taking a label indicating whether the identification belongs to a preset operation device model as supervision and taking the world record coordinates of the positioning patch of the preset operation device model as input to generate the sub-channel.

5. The method according to claim 1, characterized in that Traversing the reference image set, performing image similarity analysis with the device image to be identified respectively, extracting the operating device model corresponding to the reference image with the maximum image similarity and outputting it, including: Obtaining a first reference image of the reference image set; According to the operating equipment model corresponding to the first reference image, activating the associated feature extraction channel to perform feature extraction, and obtaining a feature extraction result of the first reference image; Extracting features of the image of the device to be identified according to the associated feature extraction channel to obtain a feature extraction result of the image of the device to be identified; Calculate the first reference image similarity between the first reference image feature extraction result and the first reference image feature extraction result of the device to be identified, and add them into an image similarity set; The operating equipment model corresponding to the reference image with the maximum value of the image similarity set is extracted and outputted.

6. The method according to claim 5, characterized in that According to the operating equipment model corresponding to the first reference image, activating the associated feature extraction channel to perform feature extraction to obtain the first reference image feature extraction result, which includes: According to the preset operation equipment model, training the associated feature extraction channel; Wherein, when the preset operation equipment model belongs to a spanning frame, the feature extraction dimensions include frame structure features, height features, and connection node features; When the preset operation equipment model belongs to tower material, the feature extraction dimensions include cross-sectional shape features, surface texture features, length features and curvature features; When the preset working equipment model is a crane, the feature extraction dimensions include boom features, vehicle body features, and counterweight features; When the preset working equipment model is an excavator, the feature extraction dimensions include bucket features, boom features, and crawler features; When the preset operating equipment model is a pickup truck, the feature extraction dimensions include vehicle head features, cargo box features, and tire features; When the preset working equipment model belongs to a boom truck, the feature extraction dimensions include boom features, working bucket features, and chassis features.

7. The method according to claim 6, characterized in that Calculating the first reference image similarity between the first reference image feature extraction result and the to-be-identified device image feature extraction result, comprising: When the operating equipment type is a spanning frame, the first reference image similarity is a weighted average calculation result of the frame structure similarity, height similarity, and connection node similarity, and the weighted weight is assigned by the Delphi method; When the operating equipment type is tower material, the first reference image similarity is a weighted mean calculation result of cross-sectional shape similarity, surface texture similarity, length and curvature similarity, and the weighted weight is assigned by the Delphi method; When the operating equipment type is a crane, the first reference image similarity is a weighted mean calculation result of the boom similarity, the vehicle body similarity, and the counterweight block similarity, and the weighted weight is assigned by the Delphi method; When the working equipment type is an excavator, the first reference image similarity is a weighted mean calculation result of bucket similarity, arm similarity, and track similarity, and the weighted weight is assigned by the Delphi method; When the operating equipment type is a pickup truck, the first reference image similarity is a weighted average calculation result of the vehicle head similarity, the cargo box similarity, and the tire similarity, and the weighted weight is assigned by the Delphi method; When the operating equipment type is a boom truck, the first reference image similarity is a weighted mean calculation result of the boom similarity, the operating bucket similarity, and the chassis similarity, and the weighted weight is assigned by the Delphi method.

8. The method according to claim 1, characterized in that Traversing the reference image set, performing image similarity analysis with the device image to be identified respectively, extracting the operating device model corresponding to the reference image with the maximum image similarity and outputting it, further comprising: When the number of operating equipment models corresponding to the reference image with the maximum image similarity is greater than 1, analyzing multiple groups of images of equipment to be identified at multiple moments to select operating equipment models; Take the row of the operating equipment model with the highest frequency of appearance among the multiple groups of screened operating equipment models and output it.

9. Intelligent identification system for operating equipment in unplanned power transmission operations, characterized in that: A method for intelligently identifying operating equipment for unplanned power transmission operations according to any one of claims 1 to 8, the system comprising: An information receiving module, used for receiving an image of the working device to be identified and world coordinate information of a positioning patch deployed on the working device to be identified; The intelligent sorting module performs sorting through the first-level identification model of the operating equipment according to the world coordinate information to obtain a first-level selected operating equipment model set, wherein the first-level identification model of the operating equipment includes multiple sub-channels, and any sub-channel machine learning is obtained based on multiple groups of data training, and any group of the multiple groups of data includes: the world record coordinates of the positioning patch of the preset operating equipment model and a label indicating whether the identification belongs to the preset operating equipment model; An information screening module, configured to, when the number of the first-level selected operating equipment model sets is greater than 1, traverse the first-level selected operating equipment model sets to obtain a reference image set of the first-level selected operating equipment model sets that satisfies the world coordinate information; The result output module is used to traverse the reference image set, perform image similarity analysis with the device image to be identified, extract the operating device model corresponding to the reference image with the maximum image similarity, and output it.

Images