A deep learning-based power metering field target detection method and system

Abstract

The application discloses a kind of based on deep learning's electric power measurement field target detection method and system, belong to electric power measurement field target detection technical field.The method of the present application, comprising: obtaining the field image information of electric power operation field, generates the image data of the electric power operation field;Based on the representation evolution mechanism of measurement object, according to the image data, generate the visual structure response result for representing the existence state and the form distribution of the measurement target of the electric power operation field;Based on the constraint mapping mechanism of measurement scene, the saliency intensity of candidate measurement target in the visual structure response result is judged;Based on the determination result, the human interaction behavior of electric power operation field is identified.The implementation of the present application not only makes it necessary to proceed in order according to steps in the process of electric power operation, but also can improve the stability and accuracy of target detection under complex electric power measurement field conditions, with good engineering applicability.

Description

Technical Field

[0001] This invention relates to the field of target detection technology in power metering, and more specifically, to a deep learning-based method and system for target detection in power metering. Background Technology

[0002] With the continuous expansion of electricity metering services, the requirements for on-site monitoring of operations, identification of equipment status, and perception of personnel behavior are increasing. To improve the efficiency and management level of metering operations, existing technologies are gradually incorporating intelligent sensing methods based on images or videos to identify and analyze personnel, metering devices, and related work objects at the electricity metering site.

[0003] Existing methods for target recognition in power metering sites largely rely on general image target detection techniques. These techniques extract features and determine targets from collected on-site image data to identify them. However, power metering sites typically present complex environments, variable lighting conditions, significant differences in target size, and high similarity in target appearance. In such scenarios, general target detection methods are prone to high false positive rates, high false negative rates, and insufficient stability of detection results, making it difficult to meet the needs of practical engineering applications.

[0004] Furthermore, existing technologies typically focus on identifying single targets during target detection, lacking effective semantic association with the operational semantics of power metering. This makes it difficult for detection results to directly support metering workflow management or subsequent system integration, thus limiting the practical value of target detection results. Especially in application scenarios requiring compliance verification or status analysis of metering operations, existing methods struggle to achieve semantic expression of detection results and business rule constraints.

[0005] Existing composite insulator fault detection methods based on YOLOv8 achieve pixel-level identification and diagnostic analysis of insulator fault regions through dual-modal image fusion and an improved target detection network. This method is mainly aimed at fault detection scenarios of single power equipment, relies on a specific model structure, and lacks a unified detection mechanism for multiple targets in the power metering field and the association processing of detection results with metering business semantics, thus limiting its engineering applicability. Summary of the Invention

[0006] To address the above problems, this invention proposes a deep learning-based on-site target detection method for power metering, comprising: Based on the metering semantics-driven triggering of image acquisition control, on-site image information of the power operation site is obtained, and the image information is preprocessed to generate image data of the power operation site. Based on the representation evolution mechanism of the measurement object, a visual structural response result is generated according to the image data to represent the existence state and morphological distribution of the measurement target at the power operation site. Based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined; Based on the judgment results, human interaction behaviors at the power operation site were identified.

[0007] Optional, metrological semantics driven, including: image acquisition trigger function; Constructing the image acquisition trigger function includes: Based on the semantic state set corresponding to the metering operation stage at the power operation site: The image acquisition trigger function is constructed as follows: in, A set of valid job states related to metering operations; Among them, when At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters, as follows: Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; in, The position parameters of the acquisition viewpoint; To collect the attitude angle parameters of the viewpoint; This refers to the field of view parameter of the lens.

[0008] Optionally, the image information is preprocessed to generate image data of the power operation site, including: Image structure rearrangement processing is performed to address the visibility of measurement targets in image information, including: For the image information, a regional structure weight mapping relationship is constructed to characterize the visibility of the metrological target. A structure weight matrix is ​​generated by modeling the structural response intensity and metrological correlation at different spatial locations in the image information. The formula is as follows: in, This refers to the structural response intensity in image information that is related to the shape and boundary of the measurement target. This represents the relevance weight of the corresponding region within the measurement operation view. and An adjustment coefficient used to balance the contributions of different structural responses; Based on the structural weight matrix The image information is then structurally rearranged to obtain an intermediate image representation: in, The pixel value of the image information at pixel coordinates (x, y).

[0009] For the intermediate image, adaptive image stabilization processing based on environmental perception feedback is performed, including: The set of environmental sensing parameters based on the metering site is as follows: in, To measure the on-site illumination sensing parameters, To measure the perceived emotional state parameters of on-site personnel. To measure the occlusion of the target sensing parameters.

[0010] Constructing the image steady-state adjustment function: in, This is the environmental adaptive adjustment coefficient. This is a steady-state compensation term; Through the The feedback adjustment of the changes ensures that the intermediate image maintains structural consistency under different lighting, occlusion and imaging disturbance conditions, and is output as image data.

[0011] Optionally, based on the object representation evolution mechanism, a visual structural response result is generated from the image data to characterize the existence state and morphological distribution of the metering targets at the power operation site, including: Calculate the structural difference intensity value of local regions in the image data. The calculation formula is as follows: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of gray-level dispersion within a local area. The weighting coefficient is used to balance the contribution of the pixel gradient magnitude statistics G and the grayscale dispersion statistics V within the local region.

[0012] Structural difference strength value Greater than the first threshold The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; The candidate target response regions are aggregated according to a preset adjacency rule. Isolated regions with an area smaller than a preset pixel are removed from the aggregated regions to generate an initial target response region set. For the initial target response region set, a corresponding local image block is extracted for each response region, and a morphological perturbation process with a limited amplitude is applied to the local image block to generate no less than 6 morphological variants. The morphological variants are then subjected to consistency alignment processing to obtain a regional response sequence oriented towards the morphological changes of the measurement object. The regional response sequence is input into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; Statistical aggregation of feature representations of different morphological variants within the same region is performed, and the response stability of each feature dimension is calculated. The top 70% of feature dimensions in terms of stability are retained, while the remaining feature dimensions are suppressed. The retained feature dimensions are concatenated to obtain a concatenated feature vector. The concatenated feature vector is then compressed into a regional visual structure response vector of a preset dimension through feature mapping compression processing. This vector serves as the visual structure response result of the existence state and morphological distribution of metering targets at the power operation site.

[0013] Optionally, based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined, including: Based on the visual structural response results, a saliency response scoring function is constructed for each candidate region in the power operation site. Based on the saliency response scoring function, the saliency score value corresponding to each candidate region is calculated. The significance score value Higher than the lower threshold Candidate regions are marked as valid candidate regions, and a preliminary target candidate set is generated based on the valid candidate regions; For the preliminary target candidate set, a spatial constraint mapping rule for power operation sites is introduced to verify the consistency of the relative positional relationship of candidate regions in image data; The spatial constraint mapping rules include: distance constraints between the center point of the candidate region and the typical distribution area of ​​the metering device, proportional constraints between the size of the candidate region and the spatial scale of the region in which it is located, and minimum interval constraints between candidate regions. When a candidate region violates any spatial constraint rule, the corresponding saliency score is adjusted. After attenuation processing and consistency verification, the saliency score of each candidate region is updated. The updated saliency scores of the candidate regions are sorted, and the regions with the highest saliency scores are selected. The candidate regions are selected as the final effective target regions. The corresponding regional location description information and target attribute identification information are output for the final effective target regions to form the detection result of the metering target at the power operation site, i.e. the judgment result. in, The value range is set to 40%–60%.

[0014] Optionally, based on the judgment results, human interaction behaviors at the power operation site are identified, including: Based on the judgment results, a mapping relationship table between target detection results and metrological business semantics is constructed. For each detection target, the corresponding metrological semantic identifier is retrieved from the mapping relationship table according to the corresponding target attribute identifier, regional location information and corresponding detection timestamp information, forming a target-level metrological semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time; For the target-level metering semantic mapping results, a metering business rule verification mechanism is introduced to verify the consistency and validity of the mapping results. The metering business rules include at least: spatial adjacency rules between metering devices and operators, metering operation sequence rules, and target duration rules. The duration of the target's existence is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results; When the metering semantic mapping result of any target violates the metering business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain a verified metering semantic mapping result. Based on the verified metrological semantic mapping results, structured output information is generated according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, a linkage instruction or status notification is sent to the metering business system or management platform to output the target detection results at the metering site, that is, to output the human interaction behavior identified at the power operation site.

[0015] Furthermore, this invention also proposes a deep learning-based on-site target detection system for power metering, comprising: The data acquisition unit is used to trigger image acquisition control based on metering semantics to obtain on-site image information of the power operation site, and to preprocess the image information to generate image data of the power operation site. The data processing unit is used to generate visual structural response results that characterize the existence state and morphological distribution of the metering targets at the power operation site based on the image data, according to the metering object characterization evolution mechanism. The determination unit is used to determine the salience intensity of candidate measurement targets in the visual structure response results based on the measurement scene constraint mapping mechanism. The output unit is used to identify human interaction behaviors at the power operation site based on the judgment results.

[0016] Optional, metrological semantics driven, including: image acquisition trigger function; Constructing the image acquisition trigger function includes: Based on the semantic state set corresponding to the metering operation stage at the power operation site: The image acquisition trigger function is constructed as follows: in, A set of valid job states related to metering operations; Among them, when At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters, as follows: Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; in, The position parameters of the acquisition viewpoint; To collect the attitude angle parameters of the viewpoint; This refers to the field of view parameter of the lens.

[0017] Optionally, the image information is preprocessed to generate image data of the power operation site, including: Image structure rearrangement processing is performed to address the visibility of measurement targets in image information, including: For the image information, a regional structure weight mapping relationship is constructed to characterize the visibility of the metrological target. A structure weight matrix is ​​generated by modeling the structural response intensity and metrological correlation at different spatial locations in the image information. The formula is as follows: in, This refers to the structural response intensity in image information that is related to the shape and boundary of the measurement target. This represents the relevance weight of the corresponding region within the measurement operation view. and An adjustment coefficient used to balance the contributions of different structural responses; Based on the structural weight matrix The image information is then structurally rearranged to obtain an intermediate image representation: in, The pixel value of the image information at pixel coordinates (x, y).

[0018] For the intermediate image, adaptive image stabilization processing based on environmental perception feedback is performed, including: The set of environmental sensing parameters based on the metering site is as follows: in, To measure the on-site illumination sensing parameters, To measure the perceived emotional state parameters of on-site personnel. To measure the occlusion of the target sensing parameters.

[0019] Constructing the image steady-state adjustment function: in, This is the environmental adaptive adjustment coefficient. This is a steady-state compensation term; Through the The feedback adjustment of the changes ensures that the intermediate image maintains structural consistency under different lighting, occlusion and imaging disturbance conditions, and is output as image data.

[0020] Optionally, based on the object representation evolution mechanism, a visual structural response result is generated from the image data to characterize the existence state and morphological distribution of the metering targets at the power operation site, including: Calculate the structural difference intensity value of local regions in the image data. The calculation formula is as follows: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of gray-level dispersion within a local area. The weighting coefficient is used to balance the contribution of the pixel gradient magnitude statistics G and the grayscale dispersion statistics V within the local region.

[0021] Structural difference strength value Greater than the first threshold The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; The candidate target response regions are aggregated according to a preset adjacency rule. Isolated regions with an area smaller than a preset pixel are removed from the aggregated regions to generate an initial target response region set. For the initial target response region set, a corresponding local image block is extracted for each response region, and a morphological perturbation process with a limited amplitude is applied to the local image block to generate no less than 6 morphological variants. The morphological variants are then subjected to consistency alignment processing to obtain a regional response sequence oriented towards the morphological changes of the measurement object. The regional response sequence is input into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; Statistical aggregation of feature representations of different morphological variants within the same region is performed, and the response stability of each feature dimension is calculated. The top 70% of feature dimensions in terms of stability are retained, while the remaining feature dimensions are suppressed. The retained feature dimensions are concatenated to obtain a concatenated feature vector. The concatenated feature vector is then compressed into a regional visual structure response vector of a preset dimension through feature mapping compression processing. This vector serves as the visual structure response result of the existence state and morphological distribution of metering targets at the power operation site.

[0022] Optionally, based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined, including: Based on the visual structural response results, a saliency response scoring function is constructed for each candidate region in the power operation site. Based on the saliency response scoring function, the saliency score value corresponding to each candidate region is calculated. The significance score value Higher than the lower threshold Candidate regions are marked as valid candidate regions, and a preliminary target candidate set is generated based on the valid candidate regions; For the preliminary target candidate set, a spatial constraint mapping rule for power operation sites is introduced to verify the consistency of the relative positional relationship of candidate regions in image data; The spatial constraint mapping rules include: distance constraints between the center point of the candidate region and the typical distribution area of ​​the metering device, proportional constraints between the size of the candidate region and the spatial scale of the region in which it is located, and minimum interval constraints between candidate regions. When a candidate region violates any spatial constraint rule, the corresponding saliency score is adjusted. After attenuation processing and consistency verification, the saliency score of each candidate region is updated. The updated saliency scores of the candidate regions are sorted, and the regions with the highest saliency scores are selected. The candidate regions are selected as the final effective target regions. The corresponding regional location description information and target attribute identification information are output for the final effective target regions to form the detection result of the metering target at the power operation site, i.e. the judgment result. in, The value range is set to 40%–60%.

[0023] Optionally, based on the judgment results, human interaction behaviors at the power operation site are identified, including: Based on the judgment results, a mapping relationship table between target detection results and metrological business semantics is constructed. For each detection target, the corresponding metrological semantic identifier is retrieved from the mapping relationship table according to the corresponding target attribute identifier, regional location information and corresponding detection timestamp information, forming a target-level metrological semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time; For the target-level metering semantic mapping results, a metering business rule verification mechanism is introduced to verify the consistency and validity of the mapping results. The metering business rules include at least: spatial adjacency rules between metering devices and operators, metering operation sequence rules, and target duration rules. The duration of the target's existence is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results; When the metering semantic mapping result of any target violates the metering business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain a verified metering semantic mapping result. Based on the verified metrological semantic mapping results, structured output information is generated according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, a linkage instruction or status notification is sent to the metering business system or management platform to output the target detection results at the metering site, that is, to output the human interaction behavior identified at the power operation site.

[0024] In another aspect, the present invention also provides a computing device, comprising: one or more processors; A processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the method described above is implemented.

[0025] In another aspect, the present invention also provides a computer-readable storage medium having a computer program stored thereon, which, when executed, implements the method described above.

[0026] Compared with the prior art, the beneficial effects of the present invention are as follows: This invention provides a deep learning-based method for target detection at power metering sites, comprising: triggering image acquisition control based on metering semantics to acquire on-site image information of the power operation site, and preprocessing the image information to generate image data of the power operation site; generating visual structural response results representing the existence state and morphological distribution of metering targets at the power operation site based on the image data, according to a metering object representation evolution mechanism; determining the saliency intensity of candidate metering targets in the visual structural response results based on a metering scene constraint mapping mechanism; and identifying human interaction behaviors at the power operation site based on the determination results. The implementation of this invention not only ensures that power operations are carried out in an orderly manner according to steps, but also improves the stability and accuracy of target detection under complex power metering site conditions, exhibiting good engineering applicability. Attached Figure Description

[0027] Figure 1 This is a flowchart of the method of the present invention; Figure 2 A flowchart of an embodiment of the method of the present invention. Figure 3 This is a schematic diagram illustrating the evolution of embodiments of the method of the present invention; Figure 4 This is a schematic diagram illustrating the discrimination process in an embodiment of the method of the present invention; Figure 5 This is a diagram illustrating the linkage relationships in an embodiment of the method of the present invention. Figure 6 This is a structural diagram of the system of the present invention. Detailed Implementation

[0028] Exemplary embodiments of the invention will now be described with reference to the accompanying drawings. However, the invention may be embodied in many different forms and is not limited to the embodiments described herein. These embodiments are provided to fully and completely disclose the invention and to fully convey its scope to those skilled in the art. The terminology used in the exemplary embodiments illustrated in the drawings is not intended to limit the invention. In the drawings, the same units / elements are referred to by the same reference numerals.

[0029] Unless otherwise stated, the terms used herein (including technical terms) have their common meaning as understood by one of ordinary skill in the art. Furthermore, it is understood that terms defined in commonly used dictionaries should be understood to have a meaning consistent with the context of their relevant field, and not to be interpreted as having an idealized or overly formal meaning.

[0030] Example 1: This invention proposes a deep learning-based on-site target detection method S100 for power metering, as follows: Figure 1 As shown, it includes: S101, based on metering semantics to drive triggering image acquisition control, to obtain on-site image information of the power operation site, and to preprocess the image information to generate image data of the power operation site; S102, Based on the representation evolution mechanism of the measurement object, a visual structural response result is generated according to the image data to represent the existence state and morphological distribution of the measurement target at the power operation site; S103, Based on the metrology scene constraint mapping mechanism, determine the salience intensity of the candidate metrology targets in the visual structure response results; S104, based on the judgment result, identifies human interaction behavior at the power operation site.

[0031] Among them, the metrological semantic drive includes: image acquisition trigger function; Constructing the image acquisition trigger function includes: Based on the semantic state set corresponding to the metering operation stage at the power operation site: The image acquisition trigger function is constructed as follows: in, A set of valid job states related to metering operations; Among them, when At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters, as follows: Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; in, The position parameters of the acquisition viewpoint; To collect the attitude angle parameters of the viewpoint; This refers to the field of view parameter of the lens.

[0032] The process of preprocessing the image information to generate image data of the power operation site includes: Image structure rearrangement processing is performed to address the visibility of measurement targets in image information, including: For the image information, a regional structure weight mapping relationship is constructed to characterize the visibility of the metrological target. A structure weight matrix is ​​generated by modeling the structural response intensity and metrological correlation at different spatial locations in the image information. The formula is as follows: in, This refers to the structural response intensity in image information that is related to the shape and boundary of the measurement target. This represents the relevance weight of the corresponding region within the measurement operation view. and An adjustment coefficient used to balance the contributions of different structural responses; Based on the structural weight matrix The image information is then structurally rearranged to obtain an intermediate image representation: in, The pixel value of the image information at pixel coordinates (x, y).

[0033] For the intermediate image, adaptive image stabilization processing based on environmental perception feedback is performed, including: The set of environmental sensing parameters based on the metering site is as follows: in, To measure the on-site illumination sensing parameters, To measure the perceived emotional state parameters of on-site personnel. To measure the occlusion of the target sensing parameters.

[0034] Constructing the image steady-state adjustment function: in, This is the environmental adaptive adjustment coefficient. This is a steady-state compensation term; Through the The feedback adjustment of the changes ensures that the intermediate image maintains structural consistency under different lighting, occlusion and imaging disturbance conditions, and is output as image data.

[0035] Specifically, based on the object representation evolution mechanism, visual structural response results are generated from the image data to characterize the existence state and morphological distribution of the metering targets at the power operation site, including: Calculate the structural difference intensity value of local regions in the image data. The calculation formula is as follows: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of gray-level dispersion within a local area. The weighting coefficient is used to balance the contribution of the pixel gradient magnitude statistics G and the grayscale dispersion statistics V within the local region.

[0036] Structural difference strength value Greater than the first threshold The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; The candidate target response regions are aggregated according to a preset adjacency rule. Isolated regions with an area smaller than a preset pixel are removed from the aggregated regions to generate an initial target response region set. For the initial target response region set, a corresponding local image block is extracted for each response region, and a morphological perturbation process with a limited amplitude is applied to the local image block to generate no less than 6 morphological variants. The morphological variants are then subjected to consistency alignment processing to obtain a regional response sequence oriented towards the morphological changes of the measurement object. The regional response sequence is input into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; Statistical aggregation of feature representations of different morphological variants within the same region is performed, and the response stability of each feature dimension is calculated. The top 70% of feature dimensions in terms of stability are retained, while the remaining feature dimensions are suppressed. The retained feature dimensions are concatenated to obtain a concatenated feature vector. The concatenated feature vector is then compressed into a regional visual structure response vector of a preset dimension through feature mapping compression processing. This vector serves as the visual structure response result of the existence state and morphological distribution of metering targets at the power operation site.

[0037] Specifically, based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined, including: Based on the visual structural response results, a saliency response scoring function is constructed for each candidate region in the power operation site. Based on the saliency response scoring function, the saliency score value corresponding to each candidate region is calculated. The significance score value Higher than the lower threshold Candidate regions are marked as valid candidate regions, and a preliminary target candidate set is generated based on the valid candidate regions; For the preliminary target candidate set, a spatial constraint mapping rule for power operation sites is introduced to verify the consistency of the relative positional relationship of candidate regions in image data; The spatial constraint mapping rules include: distance constraints between the center point of the candidate region and the typical distribution area of ​​the metering device, proportional constraints between the size of the candidate region and the spatial scale of the region in which it is located, and minimum interval constraints between candidate regions. When a candidate region violates any spatial constraint rule, the corresponding saliency score is adjusted. After attenuation processing and consistency verification, the saliency score of each candidate region is updated. The updated saliency scores of the candidate regions are sorted, and the regions with the highest saliency scores are selected. The candidate regions are selected as the final effective target regions. The corresponding regional location description information and target attribute identification information are output for the final effective target regions to form the detection result of the metering target at the power operation site, i.e. the judgment result. in, The value range is set to 40%–60%.

[0038] Based on the judgment results, the human interaction behaviors at the power operation site were identified, including: Based on the judgment results, a mapping relationship table between target detection results and metrological business semantics is constructed. For each detection target, the corresponding metrological semantic identifier is retrieved from the mapping relationship table according to the corresponding target attribute identifier, regional location information and corresponding detection timestamp information, forming a target-level metrological semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time; For the target-level metering semantic mapping results, a metering business rule verification mechanism is introduced to verify the consistency and validity of the mapping results. The metering business rules include at least: spatial adjacency rules between metering devices and operators, metering operation sequence rules, and target duration rules. The duration of the target's existence is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results; When the metering semantic mapping result of any target violates the metering business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain a verified metering semantic mapping result. Based on the verified metrological semantic mapping results, structured output information is generated according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, a linkage instruction or status notification is sent to the metering business system or management platform to output the target detection results at the metering site, that is, to output the human interaction behavior identified at the power operation site.

[0039] The present invention will be further described below with reference to another embodiment: The specific implementation process includes: S201: Semantic-driven on-site image acquisition and adaptive preprocessing to acquire image or video data from the power metering site; S202: Based on the representation evolution mechanism of measurement objects, generate visual structural response results to characterize the existence state and morphological distribution of targets in the measurement field; S203: Based on the metrological scene constraint mapping mechanism, the salience intensity of candidate targets in the image is determined; S204: Based on the detection results, obtain the metrological semantic mapping and business linkage output.

[0040] Step S201 specifically includes: S011: Image acquisition control based on semantic triggering of metering operations. During the power metering operation, the image acquisition control is based on the semantic state set corresponding to the metering operation stage. Construct the image acquisition trigger function: in, This represents the set of valid job states related to metering operations. When At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters... Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; S012: Image structure rearrangement processing for improving the visibility of metering targets, for the acquired power metering field image data. A regional structure weight mapping relationship is constructed to characterize the visibility of metrological targets. By jointly modeling the structural response intensity and metrological correlation at different spatial locations in the image, the corresponding structure weight matrix is ​​generated. Its expression is: in, This represents the structural response intensity in the image that is related to the shape and boundary of the measurement target. This indicates the relevance weight of the corresponding region within the measurement operation view. and These are adjustment coefficients used to balance the contributions of different structural responses. Based on the aforementioned structural weight matrix, the original image data undergoes structural rearrangement processing to obtain an intermediate image representation: By performing structural rearrangement, the region where the metrology target is located can obtain higher structural saliency in the intermediate image representation, while suppressing the interference of non-metrology-related regions on the subsequent target detection process, thereby providing input image data optimized for metrology targets for subsequent steps. S2013: Image adaptive stabilization processing based on environmental perception feedback, combined with the set of environmental perception parameters from the measurement site. Constructing the image steady-state adjustment function: in, This represents the environmental adaptive adjustment coefficient. This represents the steady-state compensation term. Through feedback adjustment based on changes in environmental parameters, the output image... Maintaining structural consistency under different lighting, occlusion, and imaging perturbation conditions serves as standardized input data for subsequent target detection.

[0041] Step S202 specifically includes: S2021: Based on the standardized image data output in step S201, calculate the structural difference intensity value of a local region in the image. The structural difference intensity value is determined jointly by the pixel gradient magnitude and the gray-level dispersion within the region. Preferably, the structural difference intensity value of the local region... Calculate using the following formula: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of grayscale dispersion within a local area; structural difference intensity greater than a first threshold is considered. The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; the candidate target response regions are aggregated according to a preset adjacency rule, preferably 8-neighbor connectivity; isolated regions with an area of ​​less than 64 pixels in the aggregated region are removed to form an initial target response region set; S2022: For the initial target response region set, extract the corresponding local image block for each response region, and apply a limited amplitude morphological perturbation to the local image block while maintaining the spatial positional relationship. The morphological perturbation includes: ±10% brightness perturbation, ±8% contrast perturbation, 1–3 pixel boundary expansion and contraction, and directional rotation not exceeding ±10°. Based on the morphological perturbation, generate no less than 6 morphological variants, and perform consistency alignment processing on each morphological variant to obtain a region response sequence oriented towards the morphological changes of the measurement object. S2023: Input the regional response sequence into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; statistically aggregate the feature representations of different morphological variants in the same region, calculate the response stability of each feature dimension, retain the top 70% of the feature dimensions in terms of stability, and suppress the remaining feature dimensions; concatenate the retained feature representations to obtain a concatenated feature vector, and compress the concatenated feature vector into a regional visual structure response vector with a dimension of 2048 through feature mapping compression processing, which serves as the visual structure response result used to characterize the existence state and morphological distribution of targets in the measurement field.

[0042] Step S3 specifically includes: S2031: Based on the regional visual structure response vector output in step S202, construct a saliency response scoring function for each candidate region in the measurement site. The saliency response score is jointly determined by regional structure strength, morphological consistency, and regional integrity. Calculate the corresponding saliency score value for each candidate region. And the significance score is lower than the preset lower threshold. Candidate regions with low confidence scores are marked as low-confidence regions, and candidate regions with salience scores higher than the threshold are marked as valid candidate regions, in order to form a preliminary target candidate set. S2032: For the preliminary target candidate set, a measurement site spatial constraint mapping rule is introduced to verify the consistency of the relative positional relationship of candidate regions in the image. The spatial constraint mapping rule includes: distance constraint between the center point of the candidate region and the typical distribution area of ​​the measurement device, proportional constraint between the size of the candidate region and the spatial scale of its region, and minimum interval constraint between candidate regions. When a candidate region violates any spatial constraint rule, its significance score is attenuated, and the attenuation coefficient is set to 0.6. After the consistency verification, the significance score results of each candidate region are updated. S2033: Sort the updated candidate region significance scores and select the top-ranked regions. The candidate regions are selected as the final effective target regions, among which The value range is set to 40%–60%; the corresponding regional location description information and target attribute identification information are output for the final effective target area to form the target detection result at the metering site, which is then used for subsequent steps to perform metering semantic mapping and business linkage processing.

[0043] Step S204 specifically includes: S2041: Based on the target detection results at the metering site output in step S203, construct a mapping table between the target detection results and the metering business semantics. The mapping table is used to describe the correspondence between different target attribute identifiers and corresponding metering business semantic tags. For each detected target, according to its target attribute identifier, regional location information, and corresponding detection timestamp information, retrieve the corresponding metering semantic identifier from the mapping table to form a target-level metering semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time. S2042: For the target-level metrological semantic mapping result, a metrological business rule verification mechanism is introduced to verify the consistency and validity of the mapping result. The metrological business rules include at least: spatial adjacency rules between the metrological device and the operator, metrological operation sequence rules, and target persistence duration rules. The target persistence duration is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results. When the metrological semantic mapping result of any target violates the metrological business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain the verified metrological semantic mapping result. S2043: Based on the verified metrological semantic mapping results, generate structured output information according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, send linkage instructions or status notifications to the metrological business system or management platform to realize the business linkage output of the target detection results at the metrological site.

[0044] The present invention will be further described below with reference to another embodiment: Specific implementation process, such as Figure 2 As shown, it includes: S301: Semantic-driven on-site image acquisition and adaptive preprocessing to acquire image or video data from the power metering site; S302: Based on the representation evolution mechanism of measurement objects, generate visual structural response results to characterize the existence state and morphological distribution of targets at the measurement site; S303: Based on the metrological scene constraint mapping mechanism, the salience intensity of candidate targets in the image is determined; S304: Based on the detection results, obtain the metrological semantic mapping and business linkage output.

[0045] Specifically, the invention first acquires image or video data from the power metering site and performs adaptive preprocessing on the image data under metering semantics. Then, it analyzes the preprocessed image data based on a metering object representation evolution mechanism to generate visual structural response results that characterize the existence state and morphological distribution of targets at the metering site. Next, it determines the saliency intensity of candidate targets in the image based on a metering scene constraint mapping mechanism to obtain corresponding target detection results. Finally, it completes metering semantic mapping based on the target detection results and outputs detection information for business linkage. This invention can improve the stability and accuracy of target detection under complex power metering site conditions and has good engineering applicability.

[0046] like Figure 3 As shown, step S302 specifically includes: S3021: Based on the standardized image data output in step S301, calculate the structural difference intensity value of a local region in the image. The structural difference intensity value is jointly determined by the pixel gradient magnitude and the gray-level dispersion within the region. Preferably, the structural difference intensity value of the local region... Calculate using the following formula: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of grayscale dispersion within a local area; structural difference intensity greater than a first threshold is considered. The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; the candidate target response regions are aggregated according to a preset adjacency rule, preferably 8-neighbor connectivity; isolated regions with an area of ​​less than 64 pixels in the aggregated region are removed to form an initial target response region set; S3022: For the initial target response region set, extract the corresponding local image block for each response region, and apply a limited amplitude morphological perturbation to the local image block while maintaining the spatial positional relationship. The morphological perturbation includes: ±10% brightness perturbation, ±8% contrast perturbation, 1–3 pixel boundary expansion and contraction, and directional rotation not exceeding ±10°. Based on the morphological perturbation, generate no less than 6 sets of morphological variants, and perform consistency alignment processing on each morphological variant to obtain a region response sequence oriented towards the morphological changes of the measurement object. S3023: Input the regional response sequence into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; statistically aggregate the feature representations of different morphological variants in the same region, calculate the response stability of each feature dimension, retain the top 70% of the feature dimensions in terms of stability, and suppress the remaining feature dimensions; concatenate the retained feature representations to obtain a concatenated feature vector, and compress the concatenated feature vector into a regional visual structure response vector with a dimension of 2048 through feature mapping compression processing, which serves as the visual structure response result used to characterize the existence state and morphological distribution of targets at the measurement site.

[0047] Specifically, the preprocessed power metering site images are first divided into a set of candidate regions with significant structural differences, and local image blocks are extracted based on the candidate regions to form an initial structural response. Then, various morphological perturbations of limited amplitude are applied to the local image blocks in sequence and consistency alignment processing is performed to construct a regional response sequence that reflects the morphological changes of the metering object. The response sequence is then subjected to feature analysis at three levels: structural strength, morphological stability, and regional consistency. Finally, the features at each level are aggregated according to a preset ratio to generate a regional visual structural response vector. After feature stability screening and compression mapping, the visual structural response result used to characterize the existence state and morphological distribution of the target at the metering site is output.

[0048] like Figure 4 As shown, step S303 specifically includes: S3031: Based on the regional visual structure response vector output in step S302, construct a saliency response scoring function for each candidate region in the measurement site. The saliency response score is jointly determined by regional structure strength, morphological consistency, and regional integrity. Calculate the corresponding saliency score value for each candidate region. And the significance score is lower than the preset lower threshold. Candidate regions with low confidence scores are marked as low-confidence regions, and candidate regions with salience scores higher than the threshold are marked as valid candidate regions, in order to form a preliminary target candidate set. S3032: For the preliminary target candidate set, a measurement site spatial constraint mapping rule is introduced to verify the consistency of the relative positional relationship of the candidate regions in the image. The spatial constraint mapping rule includes: distance constraint between the center point of the candidate region and the typical distribution area of ​​the measurement device, proportional constraint between the size of the candidate region and the spatial scale of its region, and minimum interval constraint between candidate regions. When a candidate region violates any spatial constraint rule, its significance score is attenuated, and the attenuation coefficient is set to 0.6. After the consistency verification, the significance score results of each candidate region are updated. S3033: Sort the updated candidate region significance scores and select the top-ranked regions. The candidate regions are selected as the final effective target regions, among which The value range is set to 40%–60%; the corresponding regional location description information and target attribute identification information are output for the final effective target area to form the target detection result at the metering site, which is then used for subsequent steps to perform metering semantic mapping and business linkage processing.

[0049] Specifically, the visual structure response vector of the region output in step S302 is first used as the input of candidate targets. The saliency response value of each candidate region is calculated based on the region structure strength, morphological consistency and region integrity, and candidate regions below the preset threshold are initially eliminated. Then, combined with the spatial distribution characteristics of the power metering site and the constraints of the operation scenario, the relative positional relationship and scale rationality of the remaining candidate regions are checked for consistency, and the saliency attenuation processing is implemented for candidate regions that violate the scenario constraints. Finally, the candidate regions after verification are sorted and filtered according to the saliency response value to determine the effective target regions and generate the corresponding target detection results, which serve as the input for subsequent metering semantic mapping and business linkage processing.

[0050] like Figure 5 As shown, step S304 specifically includes: S3041: Based on the target detection results at the metering site output in step S303, construct a mapping table between the target detection results and the metering business semantics. The mapping table is used to describe the correspondence between different target attribute identifiers and corresponding metering business semantic tags. For each detected target, according to its target attribute identifier, regional location information and corresponding detection timestamp information, retrieve the corresponding metering semantic identifier from the mapping table to form a target-level metering semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time. S3042: For the target-level metrological semantic mapping result, a metrological business rule verification mechanism is introduced to verify the consistency and validity of the mapping result. The metrological business rules include at least: spatial adjacency rules between the metrological device and the operator, metrological operation sequence rules, and target persistence duration rules. The target persistence duration is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results. When the metrological semantic mapping result of any target violates the metrological business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain the verified metrological semantic mapping result. S3043: Based on the verified metrological semantic mapping results, generate structured output information according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, send linkage instructions or status notifications to the metrological business system or management platform to realize the business linkage output of the target detection results at the metrological site.

[0051] Specifically, the target detection results output in step S303 are first matched with the preset metrology business semantic mapping relationship. Target-level metrology semantic identifiers are generated based on the target's attribute identifier, regional location information, and corresponding timestamp information. Then, consistency and validity checks are performed on the metrology semantic identifiers based on metrology business rules. Targets that do not conform to the spatial relationship, operation sequence, or continuous existence conditions of metrology operations are semantically corrected or marked as abnormal. Finally, structured output information is generated based on the verified metrology semantic mapping results, and a linkage instruction or status notification is sent to the metrology business system or management platform when the business linkage triggering conditions are met, thereby realizing the semantic output of target detection results at the metrology site and business linkage.

[0052] This invention not only ensures that power operations are carried out in an orderly manner according to steps, but also improves the stability and accuracy of target detection under complex power metering conditions, thus having good engineering applicability.

[0053] Compared with existing technologies, the following beneficial effects can be achieved: (1) Acquire images or video data of the power metering site, and through structural rearrangement processing, make the area where the metering target is located more structurally salient, while suppressing the interference of non-operational equipment, environment and other targets on the subsequent target detection process, and complete the adaptive preprocessing of the image data under the driving of metering semantics. Since most power metering operations are carried out outdoors, the accuracy of the acquired image or video data can be guaranteed.

[0054] (2) The preprocessed image data is analyzed using the metrology object representation evolution mechanism to generate visual structure response results that characterize the existence state and morphological distribution of targets at the metrology site. At the same time, the salience intensity of candidate targets in the image is determined by the metrology scene constraint mapping mechanism. The target detection results can be highlighted well by quantitative scoring, which helps the system to accurately identify the metrology device objects to be identified and greatly reduces the error rate.

[0055] (3) Complete the metrological semantic mapping of the target detection results, effectively transform unstructured data into structured data, and send linkage instructions or status notifications to the metering business system or management platform, so as to facilitate the power field operation platform system to recognize and understand, and can well realize the business linkage output of the target detection results at the metering site.

[0056] Example 2: Furthermore, this invention also proposes a deep learning-based on-site target detection system 200 for power metering, such as... Figure 6 As shown, it includes: The data acquisition unit 201 is used to trigger image acquisition control based on metering semantics to obtain on-site image information of the power operation site, and to preprocess the image information to generate image data of the power operation site. Data processing unit 202 is used to generate visual structural response results that characterize the existence state and morphological distribution of the metering targets at the power operation site based on the image data, according to the metering object characterization evolution mechanism. The determination unit 203 is used to determine the salience intensity of candidate measurement targets in the visual structure response results based on the measurement scene constraint mapping mechanism. The output unit 204 is used to identify human interaction behavior at the power operation site based on the judgment result.

[0057] Among them, the metrological semantic drive includes: image acquisition trigger function; Constructing the image acquisition trigger function includes: Based on the semantic state set corresponding to the metering operation stage at the power operation site: The image acquisition trigger function is constructed as follows: in, A set of valid job states related to metering operations; Among them, when At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters, as follows: Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; in, The position parameters of the acquisition viewpoint; To collect the attitude angle parameters of the viewpoint; This refers to the field of view parameter of the lens.

[0058] The process of preprocessing the image information to generate image data of the power operation site includes: Image structure rearrangement processing is performed to address the visibility of measurement targets in image information, including: For the image information, a regional structure weight mapping relationship is constructed to characterize the visibility of the metrological target. A structure weight matrix is ​​generated by modeling the structural response intensity and metrological correlation at different spatial locations in the image information. The formula is as follows: in, This refers to the structural response intensity in image information that is related to the shape and boundary of the measurement target. This represents the relevance weight of the corresponding region within the measurement operation view. and An adjustment coefficient used to balance the contributions of different structural responses; Based on the structural weight matrix The image information is then structurally rearranged to obtain an intermediate image representation: in, The pixel value of the image information at pixel coordinates (x, y).

[0059] For the intermediate image, adaptive image stabilization processing based on environmental perception feedback is performed, including: The set of environmental sensing parameters based on the metering site is as follows: in, To measure the on-site illumination sensing parameters, To measure the perceived emotional state parameters of on-site personnel. To measure the occlusion of the target sensing parameters.

[0060] Constructing the image steady-state adjustment function: in, This is the environmental adaptive adjustment coefficient. This is a steady-state compensation term; Through the The feedback adjustment of the changes ensures that the intermediate image maintains structural consistency under different lighting, occlusion and imaging disturbance conditions, and is output as image data.

[0061] Specifically, based on the object representation evolution mechanism, visual structural response results are generated from the image data to characterize the existence state and morphological distribution of the metering targets at the power operation site, including: Calculate the structural difference intensity value of local regions in the image data. The calculation formula is as follows: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of gray-level dispersion within a local area. The weighting coefficient is used to balance the contribution of the pixel gradient magnitude statistics G and the grayscale dispersion statistics V within the local region.

[0062] Structural difference strength value Greater than the first threshold The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; The candidate target response regions are aggregated according to a preset adjacency rule. Isolated regions with an area smaller than a preset pixel are removed from the aggregated regions to generate an initial target response region set. For the initial target response region set, a corresponding local image block is extracted for each response region, and a morphological perturbation process with a limited amplitude is applied to the local image block to generate no less than 6 morphological variants. The morphological variants are then subjected to consistency alignment processing to obtain a regional response sequence oriented towards the morphological changes of the measurement object. The regional response sequence is input into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; Statistical aggregation of feature representations of different morphological variants within the same region is performed, and the response stability of each feature dimension is calculated. The top 70% of feature dimensions in terms of stability are retained, while the remaining feature dimensions are suppressed. The retained feature dimensions are concatenated to obtain a concatenated feature vector. The concatenated feature vector is then compressed into a regional visual structure response vector of a preset dimension through feature mapping compression processing. This vector serves as the visual structure response result of the existence state and morphological distribution of metering targets at the power operation site.

[0063] Specifically, based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined, including: Based on the visual structural response results, a saliency response scoring function is constructed for each candidate region in the power operation site. Based on the saliency response scoring function, the saliency score value corresponding to each candidate region is calculated. The significance score value Higher than the lower threshold Candidate regions are marked as valid candidate regions, and a preliminary target candidate set is generated based on the valid candidate regions; For the preliminary target candidate set, a spatial constraint mapping rule for power operation sites is introduced to verify the consistency of the relative positional relationship of candidate regions in image data; The spatial constraint mapping rules include: distance constraints between the center point of the candidate region and the typical distribution area of ​​the metering device, proportional constraints between the size of the candidate region and the spatial scale of the region in which it is located, and minimum interval constraints between candidate regions. When a candidate region violates any spatial constraint rule, the corresponding saliency score is adjusted. After attenuation processing and consistency verification, the saliency score of each candidate region is updated. The updated saliency scores of the candidate regions are sorted, and the regions with the highest saliency scores are selected. The candidate regions are selected as the final effective target regions. The corresponding regional location description information and target attribute identification information are output for the final effective target regions to form the detection result of the metering target at the power operation site, i.e. the judgment result. in, The value range is set to 40%–60%.

[0064] Based on the judgment results, the human interaction behaviors at the power operation site were identified, including: Based on the judgment results, a mapping relationship table between target detection results and metrological business semantics is constructed. For each detection target, the corresponding metrological semantic identifier is retrieved from the mapping relationship table according to the corresponding target attribute identifier, regional location information and corresponding detection timestamp information, forming a target-level metrological semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time; For the target-level metering semantic mapping results, a metering business rule verification mechanism is introduced to verify the consistency and validity of the mapping results. The metering business rules include at least: spatial adjacency rules between metering devices and operators, metering operation sequence rules, and target duration rules. The duration of the target's existence is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results; When the metering semantic mapping result of any target violates the metering business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain a verified metering semantic mapping result. Based on the verified metrological semantic mapping results, structured output information is generated according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, a linkage instruction or status notification is sent to the metering business system or management platform to output the target detection results at the metering site, that is, to output the human interaction behavior identified at the power operation site.

[0065] The implementation of this invention not only ensures that power operations must be carried out in an orderly manner according to steps, but also improves the stability and accuracy of target detection under complex power metering site conditions, and has good engineering applicability.

[0066] Example 3: Based on the same inventive concept, this invention also provides a computer device, which includes a processor and a memory. The memory stores a computer program, which includes program instructions. The processor executes the program instructions stored in the computer storage medium. The processor may be a Central Processing Unit (CPU), or other general-purpose processors, digital signal processors (DSPs), application-specific integrated circuits (ASICs), field-programmable gate arrays (FPGAs), or other programmable logic devices, discrete gate or transistor logic devices, discrete hardware components, etc. It is the computing and control core of the terminal, suitable for implementing one or more instructions, specifically suitable for loading and executing one or more instructions in the computer storage medium to implement corresponding method flows or corresponding functions, thereby implementing the steps of the methods in the above embodiments.

[0067] Example 4: Based on the same inventive concept, this invention also provides a storage medium, specifically a computer-readable storage medium (Memory), which is a memory device in a computer device used to store programs and data. It is understood that the computer-readable storage medium here can include both the built-in storage medium in the computer device and extended storage media supported by the computer device. The computer-readable storage medium provides storage space that stores the terminal's operating system. Furthermore, this storage space also stores one or more instructions suitable for loading and execution by a 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 or non-volatile memory, such as at least one disk storage device. The processor can load and execute one or more instructions stored in the computer-readable storage medium to implement the steps of the method in the above embodiments.

[0068] 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 a computer program product implemented on one or more computer-usable storage media (including but not limited to disk storage, CD-ROM, optical storage, etc.) containing computer-usable program code. The solutions in the embodiments of the present invention can be implemented using various computer languages, such as the object-oriented programming language Java and the interpreted scripting language JavaScript.

[0069] 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.

[0070] These computer program instructions may also be stored in a computer-readable storage medium that can direct a computer or other programmable data processing device to function in a particular manner, such that the instructions stored in the computer-readable storage medium produce an article of manufacture including instruction means, which are implemented in a process Figure 1 One or more processes and / or boxes Figure 1 The function specified in one or more boxes.

[0071] These computer program instructions may also be loaded onto a computer or other programmable data processing equipment to cause a series of operational steps to be performed on the computer or other programmable equipment to produce a computer-implemented process, thereby providing instructions that execute on the computer or other programmable equipment for implementing the process. Figure 1 One or more processes and / or boxes Figure 1 The steps of the function specified in one or more boxes.

[0072] Although preferred embodiments of the invention have been described, those skilled in the art, upon learning the basic inventive concept, can make other changes and modifications to these embodiments. Therefore, the appended claims are intended to be interpreted as including both the preferred embodiments and all changes and modifications falling within the scope of the invention.

[0073] Obviously, those skilled in the art can make various modifications and variations to this invention without departing from its spirit and scope. Therefore, if these modifications and variations fall within the scope of the claims of this invention and their equivalents, this invention also intends to include these modifications and variations.

Claims

1. A deep learning-based method for on-site target detection in power metering, characterized in that, include: Based on the metering semantics-driven triggering of image acquisition control, on-site image information of the power operation site is obtained, and the image information is preprocessed to generate image data of the power operation site. Based on the representation evolution mechanism of the measurement object, a visual structural response result is generated according to the image data to represent the existence state and morphological distribution of the measurement target at the power operation site. Based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined; Based on the judgment results, human interaction behaviors at the power operation site were identified.

2. The method for recognizing human-human interaction behavior at a power operation site according to claim 1, characterized in that, The metrological semantic driver includes: an image acquisition trigger function; Constructing the image acquisition trigger function includes: Based on the semantic state set corresponding to the metering operation stage at the power operation site: The image acquisition trigger function is constructed as follows: in, A set of valid job states related to metering operations; Among them, when At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters, as follows: Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; in, The position parameters of the acquisition viewpoint; To collect the attitude angle parameters of the viewpoint; This refers to the field of view parameter of the lens.

3. The method for recognizing human-human interaction behavior at a power operation site according to claim 1, characterized in that, The image information is preprocessed to generate image data of the power operation site, including: Image structure rearrangement processing is performed to address the visibility of measurement targets in image information, including: For the image information, a regional structure weight mapping relationship is constructed to characterize the visibility of the metrological target. A structure weight matrix is ​​generated by modeling the structural response intensity and metrological correlation at different spatial locations in the image information. The formula is as follows: in, This refers to the structural response intensity in image information that is related to the shape and boundary of the measurement target. This represents the relevance weight of the corresponding region within the measurement operation view. and An adjustment coefficient used to balance the contributions of different structural responses; Based on the structural weight matrix The image information is then structurally rearranged to obtain an intermediate image representation: in, The pixel value of the image information at pixel coordinates (x, y); For the intermediate image, adaptive image stabilization processing based on environmental perception feedback is performed, including: The set of environmental sensing parameters based on the metering site is as follows: in, To measure the on-site illumination sensing parameters, To measure the perceived emotional state parameters of on-site personnel. To measure the occlusion of the target sensing parameters; Constructing the image steady-state adjustment function: in, This is the environmental adaptive adjustment coefficient. This is a steady-state compensation term; Through the The feedback adjustment of the changes ensures that the intermediate image maintains structural consistency under different lighting, occlusion and imaging disturbance conditions, and is output as image data.

4. The method for recognizing human-human interaction behavior at a power operation site according to claim 1, characterized in that, Based on the object representation evolution mechanism, and according to the image data, visual structural response results are generated to characterize the existence state and morphological distribution of the metering targets at the power operation site, including: Calculate the structural difference intensity value of local regions in the image data. The calculation formula is as follows: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of gray-level dispersion within a local area. The weighting coefficient is used to balance the contributions of the pixel gradient magnitude statistics G and the grayscale dispersion statistics V within the local region; Structural difference strength value Greater than the first threshold The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; The candidate target response regions are aggregated according to a preset adjacency rule. Isolated regions with an area smaller than a preset pixel are removed from the aggregated regions to generate an initial target response region set. For the initial target response region set, a corresponding local image block is extracted for each response region, and a morphological perturbation process with a limited amplitude is applied to the local image block to generate no less than 6 morphological variants. The morphological variants are then subjected to consistency alignment processing to obtain a regional response sequence oriented towards the morphological changes of the measurement object. The regional response sequence is input into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; Statistical aggregation of feature representations of different morphological variants within the same region is performed, and the response stability of each feature dimension is calculated. The top 70% of feature dimensions in terms of stability are retained, while the remaining feature dimensions are suppressed. The retained feature dimensions are concatenated to obtain a concatenated feature vector. The concatenated feature vector is then compressed into a regional visual structure response vector of a preset dimension through feature mapping compression processing. This vector serves as the visual structure response result of the existence state and morphological distribution of metering targets at the power operation site.

5. The method for recognizing human-human interaction behavior at a power operation site according to claim 1, characterized in that, Based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined, including: Based on the visual structural response results, a saliency response scoring function is constructed for each candidate region in the power operation site. Based on the saliency response scoring function, the saliency score value corresponding to each candidate region is calculated. The significance score value Higher than the lower threshold Candidate regions are marked as valid candidate regions, and a preliminary target candidate set is generated based on the valid candidate regions; For the preliminary target candidate set, a spatial constraint mapping rule for power operation sites is introduced to verify the consistency of the relative positional relationship of candidate regions in image data; The spatial constraint mapping rules include: distance constraints between the center point of the candidate region and the typical distribution area of ​​the metering device, proportional constraints between the size of the candidate region and the spatial scale of the region in which it is located, and minimum interval constraints between candidate regions. When a candidate region violates any spatial constraint rule, the corresponding saliency score is adjusted. After attenuation processing and consistency verification, the saliency score of each candidate region is updated. The updated saliency scores of the candidate regions are sorted, and the regions with the highest saliency scores are selected. The candidate regions are selected as the final effective target regions. The corresponding regional location description information and target attribute identification information are output for the final effective target regions to form the detection result of the metering target at the power operation site, i.e. the judgment result. in, The value range is set to 40%–60%.

6. The method for recognizing human-human interaction behavior at a power operation site according to claim 1, characterized in that, Based on the judgment results, the human interaction behaviors at the power operation site were identified, including: Based on the judgment results, a mapping relationship table between target detection results and metrological business semantics is constructed. For each detection target, the corresponding metrological semantic identifier is retrieved from the mapping relationship table according to the corresponding target attribute identifier, regional location information and corresponding detection timestamp information, forming a target-level metrological semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time; For the target-level metering semantic mapping results, a metering business rule verification mechanism is introduced to verify the consistency and validity of the mapping results. The metering business rules include at least: spatial adjacency rules between metering devices and operators, metering operation sequence rules, and target duration rules. The duration of the target's existence is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results; When the metering semantic mapping result of any target violates the metering business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain a verified metering semantic mapping result. Based on the verified metrological semantic mapping results, structured output information is generated according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, a linkage instruction or status notification is sent to the metering business system or management platform to output the target detection results at the metering site, that is, to output the human interaction behavior identified at the power operation site.

7. A deep learning-based on-site target detection system for power metering, characterized in that, include: The data acquisition unit is used to trigger image acquisition control based on metering semantics to obtain on-site image information of the power operation site, and to preprocess the image information to generate image data of the power operation site. The data processing unit is used to generate visual structural response results that characterize the existence state and morphological distribution of the metering targets at the power operation site based on the image data, according to the metering object characterization evolution mechanism. The determination unit is used to determine the salience intensity of candidate measurement targets in the visual structure response results based on the measurement scene constraint mapping mechanism. The output unit is used to identify human interaction behaviors at the power operation site based on the judgment results.

8. The power operation site human-human interaction behavior recognition system according to claim 7, characterized in that, The metrological semantic driver includes: an image acquisition trigger function; Constructing the image acquisition trigger function includes: Based on the semantic state set corresponding to the metering operation stage at the power operation site: The image acquisition trigger function is constructed as follows: in, A set of valid job states related to metering operations; Among them, when At that time, image acquisition is initiated, and based on the preset set of viewpoint parameters, as follows: Constraints are imposed on the acquisition perspective, acquisition area, and acquisition scale to obtain on-site image data that matches the measurement operation stage; in, The position parameters of the acquisition viewpoint; To collect the attitude angle parameters of the viewpoint; This refers to the field of view parameter of the lens.

9. The power operation site human-human interaction behavior recognition system according to claim 7, characterized in that, The image information is preprocessed to generate image data of the power operation site, including: Image structure rearrangement processing is performed to address the visibility of measurement targets in image information, including: For the image information, a regional structure weight mapping relationship is constructed to characterize the visibility of the metrological target. A structure weight matrix is ​​generated by modeling the structural response intensity and metrological correlation at different spatial locations in the image information. The formula is as follows: in, This refers to the structural response intensity in image information that is related to the shape and boundary of the measurement target. This represents the relevance weight of the corresponding region within the measurement operation view. and An adjustment coefficient used to balance the contributions of different structural responses; Based on the structural weight matrix The image information is then structurally rearranged to obtain an intermediate image representation: in, The pixel value of the image information at pixel coordinates (x, y); For the intermediate image, adaptive image stabilization processing based on environmental perception feedback is performed, including: The set of environmental sensing parameters based on the metering site is as follows: in, To measure the on-site illumination sensing parameters, To measure the perceived emotional state parameters of on-site personnel. To measure the occlusion of the target sensing parameters; Constructing the image steady-state adjustment function: in, This is the environmental adaptive adjustment coefficient. This is a steady-state compensation term; Through the The feedback adjustment of the changes ensures that the intermediate image maintains structural consistency under different lighting, occlusion and imaging disturbance conditions, and is output as image data.

10. The power operation site human-human interaction behavior recognition system according to claim 7, characterized in that, Based on the object representation evolution mechanism, and according to the image data, visual structural response results are generated to characterize the existence state and morphological distribution of the metering targets at the power operation site, including: Calculate the structural difference intensity value of local regions in the image data. The calculation formula is as follows: in, These are statistical values ​​of pixel gradient magnitudes within a local region. This represents the statistical value of gray-level dispersion within a local area. The weighting coefficient is used to balance the contributions of the pixel gradient magnitude statistics G and the grayscale dispersion statistics V within the local region; Structural difference strength value Greater than the first threshold The regions marked as candidate target response areas are those with structural difference intensity less than the second threshold. The region is marked as a stable background region; The candidate target response regions are aggregated according to a preset adjacency rule. Isolated regions with an area smaller than a preset pixel are removed from the aggregated regions to generate an initial target response region set. For the initial target response region set, a corresponding local image block is extracted for each response region, and a morphological perturbation process with a limited amplitude is applied to the local image block to generate no less than 6 morphological variants. The morphological variants are then subjected to consistency alignment processing to obtain a regional response sequence oriented towards the morphological changes of the measurement object. The regional response sequence is input into the shared feature mapping structure of the feature extraction network to obtain the feature representations corresponding to each morphological variant; Statistical aggregation of feature representations of different morphological variants within the same region is performed, and the response stability of each feature dimension is calculated. The top 70% of feature dimensions in terms of stability are retained, while the remaining feature dimensions are suppressed. The retained feature dimensions are concatenated to obtain a concatenated feature vector. The concatenated feature vector is then compressed into a regional visual structure response vector of a preset dimension through feature mapping compression processing. This vector serves as the visual structure response result of the existence state and morphological distribution of metering targets at the power operation site.

11. The power operation site human-human interaction behavior recognition system according to claim 7, characterized in that, Based on the metrological scene constraint mapping mechanism, the salience intensity of candidate metrological targets in the visual structure response results is determined, including: Based on the visual structural response results, a saliency response scoring function is constructed for each candidate region in the power operation site. Based on the saliency response scoring function, the saliency score value corresponding to each candidate region is calculated. The significance score value Higher than the lower threshold Candidate regions are marked as valid candidate regions, and a preliminary target candidate set is generated based on the valid candidate regions; For the preliminary target candidate set, a spatial constraint mapping rule for power operation sites is introduced to verify the consistency of the relative positional relationship of candidate regions in image data; The spatial constraint mapping rules include: distance constraints between the center point of the candidate region and the typical distribution area of ​​the metering device, proportional constraints between the size of the candidate region and the spatial scale of the region in which it is located, and minimum interval constraints between candidate regions. When a candidate region violates any spatial constraint rule, the corresponding saliency score is adjusted. After attenuation processing and consistency verification, the saliency score of each candidate region is updated. The updated saliency scores of the candidate regions are sorted, and the regions with the highest saliency scores are selected. The candidate regions are selected as the final effective target regions. The corresponding regional location description information and target attribute identification information are output for the final effective target regions to form the detection result of the metering target at the power operation site, i.e. the judgment result. in, The value range is set to 40%–60%.

12. The power operation site human-human interaction behavior recognition system according to claim 7, characterized in that, Based on the judgment results, the human interaction behaviors at the power operation site were identified, including: Based on the judgment results, a mapping relationship table between target detection results and metrological business semantics is constructed. For each detection target, the corresponding metrological semantic identifier is retrieved from the mapping relationship table according to the corresponding target attribute identifier, regional location information and corresponding detection timestamp information, forming a target-level metrological semantic mapping result. The detection timestamp information is determined by the image acquisition time or the target detection result generation time; For the target-level metering semantic mapping results, a metering business rule verification mechanism is introduced to verify the consistency and validity of the mapping results. The metering business rules include at least: spatial adjacency rules between metering devices and operators, metering operation sequence rules, and target duration rules. The duration of the target's existence is calculated by the time difference of the corresponding detection timestamp information of the same target in the continuous target detection results; When the metering semantic mapping result of any target violates the metering business rules, the semantic identifier of the target is corrected or marked as an abnormal semantic state to obtain a verified metering semantic mapping result. Based on the verified metrological semantic mapping results, structured output information is generated according to the preset business linkage triggering rules. The structured output information includes at least target identifier, metrological semantic identifier, timestamp information and location description information. When the structured output information meets the corresponding business linkage triggering conditions, a linkage instruction or status notification is sent to the metering business system or management platform to output the target detection results at the metering site, that is, to output the human interaction behavior identified at the power operation site.

13. A computer device, characterized in that, include: One or more processors; A processor is used to execute one or more programs; When the one or more programs are executed by the one or more processors, the method described in any one of claims 1-6 is implemented.

14. A computer-readable storage medium, characterized in that, It contains a computer program, which, when executed, implements the method as described in any one of claims 1-6.

Images