Method for detecting foreign body intrusion into power transmission channel based on deep semantic segmentation

Abstract

The application discloses a power transmission channel foreign matter invasion detection method based on deep semantic segmentation, which comprises the following steps: collecting a power transmission channel monitoring image sequence and performing pretreatment; performing pixel-level semantic classification; extracting semantic structure features corresponding to each semantic target in the power transmission channel; assigning semantic field parameters and calculating semantic potential distribution; determining the initial boundary of risk propagation in the power transmission channel and performing risk propagation calculation on the semantic gravitational potential field of the power transmission channel; extracting the central position coordinates of the foreign matter target at continuous time points and calculating the position change parameters; constructing the time sequence position sequence of each semantic region and calculating the phase difference value; and performing joint determination. The application adopts the deep semantic segmentation and semantic potential field modeling method, realizes intelligent detection of power transmission channel foreign matter invasion, and has the advantages of high recognition accuracy and low false alarm rate.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection of power transmission lines, and in particular to a method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation. Background Technology

[0002] Current power transmission line safety monitoring mainly relies on manual inspections, video surveillance, and foreign object identification methods based on target detection or image segmentation. By identifying floating, hanging, and obstructing objects in the area surrounding the transmission lines, early warnings can be issued for foreign objects approaching the transmission lines. As the operating environment of transmission lines becomes increasingly complex, deep learning-based image recognition technology has been gradually applied to the field of foreign object monitoring in power transmission lines.

[0003] Most existing technologies focus on the detection and classification of foreign objects themselves, lacking joint analysis of the spatial semantic relationships between transmission lines, tower structures, vegetation areas, and building areas. This makes it difficult to accurately reflect the dynamic risk status of foreign objects in transmission channels. At the same time, existing technologies do not make sufficient use of the risk propagation behavior during the continuous movement of foreign objects and the overall temporal phase changes of the scene, which easily leads to false alarms, missed alarms, and unstable intrusion judgments. This makes it difficult to meet the needs of refined foreign object intrusion detection in complex transmission channel environments. Summary of the Invention

[0004] One objective of this invention is to propose a foreign object intrusion detection method for power transmission channels based on deep semantic segmentation. This invention employs deep semantic segmentation and semantic potential field modeling to achieve intelligent detection of foreign object intrusion in power transmission channels, which has the advantages of high recognition accuracy and low false alarm rate.

[0005] The foreign object intrusion detection method for power transmission channels based on deep semantic segmentation according to an embodiment of the present invention includes the following steps:

[0006] Acquire monitoring image sequences of power transmission channels, perform preprocessing on the monitoring image sequences of power transmission channels, and generate standardized image sequences of power transmission channels;

[0007] The normalized image sequence of the power transmission channel is input into a deep semantic segmentation network to perform pixel-level semantic classification and generate a semantic segmentation result sequence of the power transmission channel;

[0008] Based on the semantic segmentation result sequence of the transmission channel, the semantic structure features corresponding to each semantic target in the transmission channel are extracted to generate a set of semantic structure parameters for the transmission channel;

[0009] Semantic field parameters are assigned to each semantic target in the semantic structure parameter set of the power transmission channel, and the semantic potential energy distribution is calculated to generate the semantic gravitational potential energy field of the power transmission channel.

[0010] The initial boundary of risk propagation in the transmission channel is determined based on the set of semantic structure parameters of the transmission channel, and risk propagation calculation is performed on the semantic gravitational potential energy field of the transmission channel to generate the risk diffusion field of the transmission channel;

[0011] Extract the center position coordinates of the foreign object target at continuous time intervals, calculate the position change parameters, and combine the semantic gravitational potential energy field and risk diffusion field of the transmission channel to generate the foreign object field response state sequence.

[0012] Construct the temporal position sequence of each semantic region, calculate the phase difference, and generate the phase synchronization state sequence of the transmission channel;

[0013] The foreign object field response state sequence and the power transmission channel phase synchronization state sequence are jointly determined to generate the foreign object intrusion determination result.

[0014] Optionally, the preprocessing includes time alignment, distortion correction, illumination normalization, noise suppression, and spatial scale unification.

[0015] Optionally, the transmission channel monitoring image sequence is a set of multi-frame monitoring image data that reflects the changes in the state of the transmission conductor and its surrounding environment, obtained continuously in chronological order by image acquisition equipment in the transmission line channel area.

[0016] Optionally, the generation of the semantic segmentation result sequence of the power transmission channel includes:

[0017] Each frame of the normalized image sequence of the power transmission channel is input into the deep semantic segmentation network in time index order, and size adaptation is performed on each frame.

[0018] A deep semantic segmentation network is used to perform multi-layer feature extraction on each frame of the standardized image sequence of the power transmission channel. In the network encoding stage, the texture, edge and structure in the image are mapped layer by layer to generate a set of feature representations of the power transmission channel image.

[0019] In the decoding stage of the deep semantic segmentation network, feature recovery and feature fusion are performed on the feature representation set of the power transmission channel image. The high-level semantic features and low-level spatial structure features are combined by a multi-scale feature fusion method to generate a set of semantic feature maps of the power transmission channel.

[0020] Based on the semantic classification output layer of the deep semantic segmentation network, semantic category determination is performed on the set of semantic feature maps of the power transmission channel to obtain the semantic category label corresponding to each pixel;

[0021] The semantic category labels of all pixels in each frame of the image are reorganized according to their spatial location in the image to generate a semantic annotation map;

[0022] Based on the temporal order of the standardized image sequence of the power transmission channel, the semantic annotation maps corresponding to each frame of the image are combined temporally to form a semantic segmentation result sequence of the power transmission channel.

[0023] Optionally, the generation of the set of semantic structure parameters for the power transmission channel includes:

[0024] Extract the semantic regions corresponding to each semantic category from the semantic annotation maps of each frame in the semantic segmentation result sequence of the power transmission channel, and divide the semantic regions into connected components according to the pixel semantic category identifier to obtain the set of semantic regions corresponding to each semantic target;

[0025] For each semantic region in the semantic region set, region contour extraction is performed, and the corresponding geometric morphological features are calculated based on the semantic region contour to form a semantic region structural feature set;

[0026] Spatial location calculations are performed on each semantic region. The corresponding spatial center position is determined based on the pixel distribution position of the semantic region in the image, and the spatial distribution range of the semantic region is extracted to obtain the set of semantic region spatial location parameters.

[0027] Based on the set of semantic region structural features and the set of semantic region spatial location parameters, a correspondence between semantic regions and the spatial structure of power transmission channels is established, and the spatial location changes of each semantic region in consecutive image frames are correlated to form a set of semantic region temporal location correlations.

[0028] Based on the set of semantic region structural features, the set of semantic region spatial location parameters, and the set of semantic region temporal location associations, a unified structural description is performed on each semantic target in the power transmission channel, and normalization processing is carried out to generate a set of semantic structural parameters for the power transmission channel.

[0029] Optionally, the generation of the semantic gravitational potential field of the power transmission channel includes:

[0030] Read the set of semantic structure parameters of the power transmission channel, and parse the semantic region structure identifier value, semantic region spatial location identifier value and semantic region location change identifier value in each semantic structure parameter value to obtain the structural state parameter value, spatial location parameter value and change state parameter value corresponding to each semantic region;

[0031] Based on the semantic category corresponding to each semantic region, its risk attribute is determined. According to the correspondence between semantic category and risk attribute, structural field strength parameter value is assigned to the structural state parameter value of each semantic region, field source position parameter value is assigned to the spatial position parameter value of each semantic region, and dynamic modulation parameter value is assigned to the changing state parameter value of each semantic region, thereby generating the semantic field parameter value of the corresponding semantic region.

[0032] Using the field source location parameter values ​​of each semantic region as the spatial field source location, each semantic region is mapped to the corresponding spatial field source unit. The initial field strength value of each spatial field source unit is determined according to the structural field strength parameter value. The dynamic field strength correction value of each spatial field source unit is determined according to the dynamic modulation parameter value. The field source action parameter value corresponding to each spatial field source unit is generated.

[0033] For each spatial source unit, its center of action in the transmission channel space is determined based on the source position parameter value. The semantic potential energy contribution value of the spatial source unit at each spatial position in the transmission channel is calculated based on the source action parameter value, and the semantic potential energy distribution value of the corresponding spatial source unit is generated.

[0034] According to the spatial location relationship of the power transmission channel, the semantic potential energy distribution values ​​corresponding to all spatial field source units are aggregated to generate the comprehensive semantic potential energy value corresponding to each spatial location, and the semantic potential energy distribution matrix of the power transmission channel is formed based on all comprehensive semantic potential energy values.

[0035] Based on the semantic potential energy distribution matrix of the power transmission channel, establish the correspondence between each spatial location and the comprehensive semantic potential energy value, and generate the semantic gravitational potential energy field of the power transmission channel.

[0036] Optionally, the generation of the transmission channel risk diffusion field includes:

[0037] Read all semantic structure parameter values ​​in the set of semantic structure parameters of the power transmission channel, and parse the semantic region spatial location identifier value in each semantic structure parameter value to obtain the region center location parameter value and spatial distribution range parameter value of the corresponding semantic region;

[0038] Based on the regional center location parameter value and the spatial distribution range parameter value, the spatial regions corresponding to the transmission lines, towers, and vegetation are determined. The initial risk propagation boundary is generated based on the spatial boundary of the spatial region corresponding to the transmission lines, and the risk propagation constraint boundary is generated based on the spatial boundaries of the spatial regions corresponding to the towers and vegetation.

[0039] A spatial location index is established in the semantic gravitational potential energy field of the power transmission channel, the comprehensive semantic potential energy value corresponding to all spatial locations is read, and the minimum spatial distance value from each spatial location in the power transmission channel to the initial boundary of risk propagation is calculated.

[0040] For spatial locations on the initial boundary and the constraint boundary of risk propagation, the corresponding comprehensive semantic potential value is determined as the risk propagation value at that spatial location;

[0041] For spatial locations in the transmission channel that are not located on the initial boundary of risk propagation, read the corresponding comprehensive semantic potential energy value and minimum spatial distance value, divide the corresponding comprehensive semantic potential energy value by the minimum spatial distance value, and generate the risk propagation value for that spatial location.

[0042] The risk propagation values ​​corresponding to all spatial locations are organized according to the spatial location index order, and the risk propagation constraint boundary is used as the risk diffusion boundary range to generate the risk diffusion field of the power transmission channel.

[0043] Optionally, the generation of the foreign object field response state sequence includes:

[0044] Read the semantic segmentation result sequence of the power transmission channel, identify the semantic region corresponding to the foreign object target in the semantic segmentation result at each time, and extract the coordinate value of the center position of the region corresponding to the foreign object target;

[0045] According to the temporal order of the semantic segmentation result sequence of the power transmission channel, the difference calculation is performed on the regional center position coordinates of the same foreign object target at adjacent time points to obtain the displacement distance value between the two time points, and then normalized to generate the foreign object target position change parameter value.

[0046] Based on the coordinates of the center of the region, the comprehensive semantic potential energy value of the corresponding spatial location in the semantic gravitational potential energy field of the power transmission channel is read and normalized to generate the foreign object potential energy response value.

[0047] Based on the coordinates of the center of the region, the risk propagation value of the corresponding spatial location in the risk diffusion field of the power transmission channel is read and normalized to generate the foreign object risk response value;

[0048] The foreign object potential energy response value and the foreign object risk response value are weighted and summed to generate the foreign object field response intensity value. The foreign object field response intensity value and the foreign object target position change parameter value are then multiplied to generate the field response value at that moment.

[0049] Based on the temporal order of the semantic segmentation results sequence of the power transmission channel, the field response values ​​corresponding to each moment are recorded sequentially to generate the foreign object field response state sequence.

[0050] Optionally, the generation of the power transmission channel phase synchronization state sequence includes:

[0051] Read the semantic segmentation result sequence of the power transmission channel, extract the coordinate values ​​of the center position of each semantic region in the semantic segmentation results at each time, and generate the semantic region position parameter values ​​corresponding to each semantic region.

[0052] Based on the semantic region location parameter values ​​corresponding to each semantic region, establish the location association relationship of consecutive moments in chronological order, and generate the temporal location sequence value corresponding to each semantic region.

[0053] Read the temporal position sequence values ​​corresponding to each semantic region, calculate the horizontal coordinate difference and vertical coordinate difference for the coordinate values ​​of the center position of the region at adjacent times, and generate the position change direction parameter values ​​corresponding to each semantic region.

[0054] Based on the position change direction parameter values ​​corresponding to each semantic region, directional encoding is performed to generate phase representation values ​​corresponding to each semantic region;

[0055] Read the phase representation values ​​corresponding to each semantic region at the same time, calculate the phase difference between each semantic region and the corresponding semantic region of the transmission line, and average all the phase differences to generate the phase synchronization feature value corresponding to that time.

[0056] The phase synchronization characteristic values ​​corresponding to each moment are organized sequentially according to the time index to generate the phase synchronization state sequence of the power transmission channel.

[0057] Optionally, the generation of the foreign object intrusion determination result includes:

[0058] Read the field response update values ​​at each time step in the foreign object field response state sequence, and extract the field response update values ​​corresponding to consecutive time steps in chronological order to generate foreign object field response trend values;

[0059] Read the phase synchronization feature values ​​at each moment in the phase synchronization state sequence of the power transmission channel, and extract the phase synchronization feature values ​​corresponding to consecutive moments in the order of time index to generate phase synchronization offset trend values;

[0060] Based on the trend value of the foreign object field response, determine whether the foreign object field response state sequence meets the condition of continuous convergence to the high-risk area corresponding to the transmission line, and generate the foreign object field response judgment value.

[0061] Based on the phase synchronization offset trend value, determine whether the phase synchronization state sequence of the transmission channel meets the preset abnormal offset condition, and generate a phase synchronization judgment value;

[0062] A joint judgment process is performed on the foreign object field response judgment value and the phase synchronization judgment value;

[0063] When the foreign object field response judgment value meets the continuous convergence condition and the phase synchronization judgment value meets the preset abnormal offset condition, a foreign object intrusion identification value is generated to indicate that the foreign object target is a foreign object intruding into the power transmission channel.

[0064] When the foreign object field response judgment value does not meet the continuous convergence condition or the phase synchronization judgment value does not meet the preset abnormal offset condition, a foreign object intrusion identification value is generated to indicate that the foreign object target is not a foreign object intruding into the power transmission channel.

[0065] Foreign object intrusion determination results are generated based on foreign object intrusion identifier values.

[0066] The beneficial effects of this invention are:

[0067] This invention provides a method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation. By performing deep semantic segmentation on a sequence of monitored images of the power transmission channel, pixel-level recognition results are obtained for various semantic targets, including transmission lines, tower structures, vegetation areas, ground areas, building areas, and foreign objects. Based on this, semantic structural features corresponding to each semantic target are extracted to construct a set of semantic structural parameters for the power transmission channel. This allows various semantic regions in the power transmission channel scene to be described using a unified structural parameter format, thereby achieving a refined expression of the spatial structural relationships of the power transmission channel. By assigning semantic field parameters to each semantic target in the semantic structural parameter set and calculating the semantic potential energy distribution, a semantic gravitational potential energy field for the power transmission channel is established. This allows the spatial influence relationship of different semantic regions in the power transmission channel on foreign objects to be quantitatively expressed through the potential energy distribution. This expands the original target detection-based recognition process into a scene understanding process that includes spatial structural relationship analysis, effectively improving the accuracy and stability of foreign object risk assessment in power transmission channel scenes.

[0068] Furthermore, this invention determines the initial boundary of risk propagation based on the set of semantic structure parameters of the transmission channel, and calculates the risk diffusion field of the transmission channel by combining the semantic gravitational potential energy field of the transmission channel. This enables the risk propagation characteristics of key semantic regions such as transmission conductors, tower structures, and vegetation areas in space to be uniformly modeled, thereby achieving a dynamic description of the risk diffusion behavior of foreign objects in the space of the transmission channel. On this basis, by extracting the center position coordinates of the foreign object at continuous time and calculating the position change parameters, a foreign object field response state sequence is generated by combining the semantic gravitational potential energy field and the risk diffusion field. At the same time, the temporal position sequence of each semantic region is constructed and the phase difference is calculated to generate a phase synchronization state sequence of the transmission channel. Finally, by performing a joint judgment on the foreign object field response state sequence and the phase synchronization state sequence, the temporal change characteristics of the overall semantic structure of the transmission channel can be analyzed while identifying the spatial movement trend of the foreign object. This effectively improves the reliability and stability of foreign object intrusion identification, reduces the occurrence of false alarms and missed alarms, and makes the foreign object intrusion detection results of the transmission channel more accurate. This is conducive to improving the intelligent level and safety assurance capability of transmission line operation monitoring. Attached Figure Description

[0069] The accompanying drawings are provided to further illustrate the invention and form part of the specification. They are used in conjunction with embodiments of the invention to explain the invention and do not constitute a limitation thereof. In the drawings:

[0070] Figure 1 This is a flowchart of the foreign object intrusion detection method for power transmission channels based on deep semantic segmentation proposed in this invention;

[0071] Figure 2 This is a schematic diagram illustrating the construction of the semantic gravitational potential energy field of the power transmission channel in the foreign object intrusion detection method based on deep semantic segmentation proposed in this invention. Detailed Implementation

[0072] The present invention will now be described in further detail with reference to the accompanying drawings. These drawings are simplified schematic diagrams, illustrating only the basic structure of the invention, and therefore only show the components relevant to the invention.

[0073] refer to Figure 1 and Figure 2 A method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation includes the following steps:

[0074] Acquire monitoring image sequences of power transmission channels, perform preprocessing on the monitoring image sequences of power transmission channels, and generate standardized image sequences of power transmission channels;

[0075] The normalized image sequence of the power transmission channel is input into a deep semantic segmentation network to perform pixel-level semantic classification and generate a semantic segmentation result sequence of the power transmission channel;

[0076] Based on the semantic segmentation result sequence of the transmission channel, the semantic structure features corresponding to each semantic target in the transmission channel are extracted to generate a set of semantic structure parameters for the transmission channel;

[0077] Semantic field parameters are assigned to each semantic target in the semantic structure parameter set of the power transmission channel, and the semantic potential energy distribution is calculated to generate the semantic gravitational potential energy field of the power transmission channel.

[0078] The initial boundary of risk propagation in the transmission channel is determined based on the set of semantic structure parameters of the transmission channel, and risk propagation calculation is performed on the semantic gravitational potential energy field of the transmission channel to generate the risk diffusion field of the transmission channel;

[0079] Extract the center position coordinates of the foreign object target at continuous time intervals, calculate the position change parameters, and combine the semantic gravitational potential energy field and risk diffusion field of the transmission channel to generate the foreign object field response state sequence.

[0080] Construct the temporal position sequence of each semantic region, calculate the phase difference, and generate the phase synchronization state sequence of the transmission channel;

[0081] The foreign object field response state sequence and the power transmission channel phase synchronization state sequence are jointly determined to generate the foreign object intrusion determination result.

[0082] In this embodiment, preprocessing includes time alignment, distortion correction, illumination normalization, noise suppression, and spatial scale unification.

[0083] In this embodiment, the transmission channel monitoring image sequence is a collection of multiple monitoring image data that reflect the changes in the state of the transmission conductor and its surrounding environment, obtained by image acquisition equipment in chronological order in the area of ​​the transmission line channel.

[0084] In this embodiment, the generation of the semantic segmentation result sequence of the power transmission channel includes:

[0085] Each frame of the normalized image sequence of the power transmission channel is input into the deep semantic segmentation network in time index order, and the size of each frame is adapted to ensure that each frame is consistent with the input structure of the deep semantic segmentation network.

[0086] A deep semantic segmentation network is used to perform multi-layer feature extraction on each frame of the standardized image sequence of the power transmission channel. In the network encoding stage, the texture, edge and structure in the image are mapped layer by layer to generate a set of feature representations of the power transmission channel image.

[0087] In the decoding stage of the deep semantic segmentation network, feature recovery and feature fusion are performed on the feature representation set of the power transmission channel image. The high-level semantic features and low-level spatial structure features are combined by a multi-scale feature fusion method to generate a set of semantic feature maps of the power transmission channel.

[0088] Based on the semantic classification output layer of the deep semantic segmentation network, semantic category determination is performed on the set of semantic feature maps of the power transmission channel to obtain the semantic category label corresponding to each pixel;

[0089] Semantic category labeling is the category code assigned by the deep semantic segmentation network after performing semantic classification on each pixel in the standardized image sequence of the power transmission channel. It is used to characterize the semantic target type to which the corresponding pixel belongs, including power transmission lines, tower structures, vegetation areas, ground areas, building areas, and foreign objects.

[0090] The semantic category labels of all pixels in each frame of the image are reorganized according to their spatial location in the image to generate a semantic annotation map;

[0091] A semantic annotation map is a two-dimensional image formed by mapping the corresponding semantic category labels point by point according to the spatial position of each pixel in the original image. It is used to characterize the spatial distribution of transmission lines, tower structures, vegetation areas, ground areas, building areas and foreign objects in a single frame of a power transmission channel image.

[0092] According to the time sequence of the standardized image sequence of the power transmission channel, the semantic annotation maps corresponding to each frame of the image are combined in a time sequence to form the semantic segmentation result sequence of the power transmission channel. The semantic segmentation result sequence of the power transmission channel is a time sequence image set formed by arranging the semantic annotation maps corresponding to each frame of the image according to the time sequence of the power transmission channel monitoring image sequence. It is used to characterize the semantic structure change process of the power transmission channel at continuous time.

[0093] In this embodiment, the generation of the semantic structure parameter set of the power transmission channel includes:

[0094] Extract the semantic regions corresponding to each semantic category from the semantic annotation maps of each frame in the semantic segmentation result sequence of the power transmission channel, and divide the semantic regions into connected components according to the pixel semantic category identifier to obtain the set of semantic regions corresponding to each semantic target;

[0095] For each semantic region in the semantic region set, region contour extraction is performed. The spatial contour of the semantic region is determined by boundary tracking. Based on the semantic region contour, the corresponding geometric morphological features are calculated to form a semantic region structural feature set.

[0096] Geometric morphological features are used to describe the feature information of the spatial structure of semantic regions, including semantic region contour shape parameters, semantic region area parameters, semantic region boundary length parameters, and semantic region aspect ratio parameters. The semantic region structural feature set is used to describe the feature data set of the spatial structure of each semantic target.

[0097] Spatial location calculations are performed on each semantic region. The corresponding spatial center position is determined based on the pixel distribution position of the semantic region in the image, and the spatial distribution range of the semantic region is extracted to obtain the set of semantic region spatial location parameters.

[0098] The semantic region spatial location parameter set consists of the regional center location and spatial distribution range of each semantic region, and is a set of parameter data used to describe the spatial location relationship of each semantic target in the power transmission channel image;

[0099] Based on the set of semantic region structural features and the set of semantic region spatial location parameters, a correspondence between semantic regions and the spatial structure of power transmission channels is established, and the spatial location changes of each semantic region in consecutive image frames are correlated to form a set of semantic region temporal location correlations.

[0100] The semantic region temporal location association set is formed by temporally associating the spatial location changes of the same semantic region in consecutive image frames. It is used to describe the spatial location relationship data of semantic regions as they change over time.

[0101] Based on the set of semantic region structural features, the set of semantic region spatial location parameters, and the set of semantic region temporal location associations, a unified structural description is performed on each semantic target in the power transmission channel, and normalization processing is carried out to generate a set of semantic structure parameters for the power transmission channel.

[0102] The set of semantic structure parameters of the transmission channel is a data set consisting of semantic structure parameter values ​​corresponding to each semantic region. Each semantic structure parameter value consists of a semantic region structure identifier value, a semantic region spatial location identifier value, and a semantic region location change identifier value, which are used to uniformly represent the structural morphology, spatial location, and spatial location change status of each semantic region in the transmission channel.

[0103] When generating the semantic structure parameter set of the transmission channel, the semantic region contour shape parameter, semantic region area parameter, semantic region boundary length parameter, and semantic region aspect ratio parameter are sequentially read for each semantic region in the semantic region structural feature set. A unified encoding is performed on these four types of structural feature parameters to generate the corresponding semantic region structural identifier value. The region center position parameter and spatial distribution range parameter corresponding to the semantic region are extracted from the semantic region spatial position parameter set. Coordinate normalization is performed on the region center position parameter, and a unified encoding is performed in conjunction with the spatial distribution range parameter to obtain the semantic region spatial position identifier value. Then, the semantic region temporal position association set is read... The spatial position change relationship of the region in consecutive image frames is analyzed. Time series sorting is performed on the continuous position change relationship, and the position change between adjacent image frames is calculated to generate the semantic region position change identifier value. After obtaining the semantic region structure identifier value, semantic region spatial position identifier value, and semantic region position change identifier value, the above three identifier values ​​are normalized and combined to generate the semantic structure parameter value of the corresponding semantic region. The above processing is repeated for all semantic regions in the semantic segmentation result sequence of the power transmission channel to obtain the semantic structure parameter value corresponding to all semantic regions. The parameters are then uniformly organized according to the semantic region category and spatial distribution order to form the power transmission channel semantic structure parameter set.

[0104] In this embodiment, the generation of the semantic gravitational potential field of the power transmission channel includes:

[0105] Read the set of semantic structure parameters of the power transmission channel, and parse the semantic region structure identifier value, semantic region spatial location identifier value and semantic region location change identifier value in each semantic structure parameter value to obtain the structural state parameter value, spatial location parameter value and change state parameter value corresponding to each semantic region;

[0106] Based on the semantic category corresponding to each semantic region, its risk attribute is determined. According to the correspondence between semantic category and risk attribute, structural field strength parameter value is assigned to the structural state parameter value of each semantic region, field source position parameter value is assigned to the spatial position parameter value of each semantic region, and dynamic modulation parameter value is assigned to the changing state parameter value of each semantic region, thereby generating the semantic field parameter value of the corresponding semantic region.

[0107] Each semantic structure parameter value in the set of semantic structure parameters for the transmission channel is processed item by item. First, the semantic region structure identifier value of the corresponding semantic region is read, and category matching is performed on the semantic region structure identifier value to obtain the semantic category identifier value corresponding to the semantic region. Then, the risk attribute is queried on the semantic category identifier value and normalized to generate the risk attribute value of the semantic region. The structural state parameter value corresponding to the semantic region is read, and the structural field strength is calculated on the structural state parameter value according to the risk attribute value to generate the structural field strength parameter value. Then, the spatial location parameter value corresponding to the semantic region is read, and the coordinates of the region center position in the spatial location parameter value are used as the field source position coordinates to generate the field source position parameter value. Next, the changing state parameter value corresponding to the semantic region is read, and dynamic modulation calculation is performed on the changing state parameter value according to the risk attribute value to generate the dynamic modulation parameter value. After obtaining the structural field strength parameter value, field source position parameter value, and dynamic modulation parameter value, the three parameter values ​​are combined to generate the semantic field parameter value of the semantic region.

[0108] Read the structural state parameter values ​​corresponding to the semantic region, and extract the semantic region contour shape parameter, semantic region area parameter, semantic region boundary length parameter, and semantic region aspect ratio parameter sequentially from the structural state parameter values; multiply the semantic region contour shape parameter with the risk attribute value to generate the contour field strength value; multiply the semantic region area parameter with the risk attribute value to generate the area field strength value; multiply the semantic region boundary length parameter with the risk attribute value to generate the boundary field strength value; multiply the semantic region aspect ratio parameter with the risk attribute value to generate the proportional field strength value; sum the contour field strength value, area field strength value, boundary field strength value, and proportional field strength value to generate the total field strength value; use the normalized total field strength value as the structural field strength parameter value corresponding to the semantic region.

[0109] The system reads the risk attribute value and change state parameter value of the corresponding semantic region, calculates the change intensity of the change state parameter value by calculating the absolute value of the change in the center position of the semantic region between consecutive image frames and summing the results, and generates a change intensity value. It then determines the risk modulation coefficient based on the risk attribute value and generates the risk modulation coefficient value by mapping the risk attribute value. Finally, it multiplies the change intensity value and the risk modulation coefficient value to generate a change modulation value. The change modulation value is then normalized to generate a normalized change modulation value. This normalized change modulation value is used as the dynamic modulation parameter value corresponding to the semantic region.

[0110] Using the field source location parameter values ​​of each semantic region as the spatial field source location, each semantic region is mapped to the corresponding spatial field source unit. The initial field strength value of each spatial field source unit is determined according to the structural field strength parameter value. The dynamic field strength correction value of each spatial field source unit is determined according to the dynamic modulation parameter value. The field source action parameter value corresponding to each spatial field source unit is generated.

[0111] Read the structural field strength parameter value corresponding to each spatial field source unit and use it as the initial field strength value of that spatial field source unit. Then, read the dynamic modulation parameter value corresponding to that spatial field source unit, multiply the initial field strength value and the dynamic modulation parameter value to generate the modulation field strength value. Next, perform a weighted sum of the initial field strength value and the modulation field strength value to generate the corrected field strength value. Then, use the corrected field strength value as the field source interaction parameter value corresponding to that spatial field source unit. Repeat the above process for all spatial field source units to obtain the field source interaction parameter values ​​corresponding to all spatial field source units.

[0112] The spatial field source unit is a spatial action node in the power transmission channel whose position is determined by the field source position parameter value of the corresponding semantic region, and whose position is used as the center of semantic potential energy generation.

[0113] For each spatial source unit, its center of action in the transmission channel space is determined based on the source position parameter value. The semantic potential energy contribution value of the spatial source unit at each spatial position in the transmission channel is calculated based on the source action parameter value, and the semantic potential energy distribution value of the corresponding spatial source unit is generated.

[0114] When generating semantic potential energy distribution values, the source position parameter value corresponding to the spatial source unit is read, and the coordinates of the region center position are extracted from the source position parameter value. The coordinates of the region center position are used as the action center position of the spatial source unit in the transmission channel space, and the action center coordinate value is generated. A set of spatial position indexes is established in the transmission channel space, so that each spatial position corresponds to a unique spatial coordinate, and spatial position coordinate values ​​are generated. Then, the spatial distance value between each spatial position coordinate value in the spatial position index set and the action center coordinate value is calculated. Subsequently, the source action parameter value corresponding to the spatial source unit is read, and the source action parameter value is divided by the corresponding spatial distance value to generate the semantic potential energy contribution value corresponding to the spatial position. The spatial distance calculation and semantic potential energy contribution value calculation are repeated for all spatial positions in the spatial position index set to generate a set of semantic potential energy contribution values ​​of the spatial source unit at all spatial positions. All semantic potential energy contribution values ​​are organized according to the spatial position index order to generate the semantic potential energy distribution value of the corresponding spatial source unit.

[0115] According to the spatial location relationship of the power transmission channel, the semantic potential energy distribution values ​​corresponding to all spatial field source units are aggregated to generate the comprehensive semantic potential energy value corresponding to each spatial location, and the semantic potential energy distribution matrix of the power transmission channel is formed based on all comprehensive semantic potential energy values.

[0116] The semantic potential energy distribution values ​​corresponding to all spatial source units are read, and a unified spatial location index is established so that each spatial location in the transmission channel space corresponds to a unique spatial location coordinate value. Then, the semantic potential energy contribution value at the corresponding spatial location coordinate value is extracted from all semantic potential energy distribution values. Semantic potential energy contribution values ​​from different spatial source units with the same spatial location coordinate value are read item by item to generate a set of semantic potential energy contribution values ​​corresponding to that spatial location. Next, the summation calculation is performed on all semantic potential energy contribution values ​​in the set of semantic potential energy contribution values ​​to generate the comprehensive semantic potential energy value corresponding to that spatial location. The process of extracting and summing semantic potential energy contribution values ​​is repeated for all spatial locations in the transmission channel space to obtain a set of comprehensive semantic potential energy values ​​corresponding to all spatial locations. All comprehensive semantic potential energy values ​​are organized sequentially according to the spatial location index order so that each spatial location coordinate value corresponds to a unique comprehensive semantic potential energy value, generating a semantic potential energy distribution matrix for the transmission channel.

[0117] The spatial source units correspond to different spatial locations, but the semantic potential energy contribution values ​​generated by multiple spatial source units to the same spatial location will be summed and superimposed at that spatial location.

[0118] Based on the semantic potential energy distribution matrix of the power transmission channel, establish the correspondence between each spatial location and the comprehensive semantic potential energy value, and generate the semantic gravitational potential energy field of the power transmission channel;

[0119] Read all matrix element values ​​in the semantic potential energy distribution matrix of the power transmission channel, and establish a spatial location index of the power transmission channel according to the matrix row index and matrix column index; determine the corresponding spatial location coordinate value based on the row index and column index of each matrix element, and generate a spatial location identifier value corresponding to each matrix element; then read the matrix element value corresponding to each spatial location identifier value, and determine the corresponding matrix element value as the comprehensive semantic potential energy value corresponding to that spatial location; bind each spatial location identifier value with the corresponding comprehensive semantic potential energy value one by one to generate a spatial location potential energy correspondence value; repeat the spatial location identifier determination and comprehensive semantic potential energy value binding process for all matrix elements in the semantic potential energy distribution matrix of the power transmission channel to obtain a set of spatial location potential energy correspondence values ​​corresponding to all spatial locations; organize the set of all spatial location potential energy correspondence values ​​in a unified manner according to the spatial location index of the power transmission channel to generate the semantic gravitational potential energy field of the power transmission channel.

[0120] In this embodiment, the generation of the risk diffusion field of the power transmission channel includes:

[0121] Read all semantic structure parameter values ​​in the set of semantic structure parameters of the power transmission channel, and parse the semantic region spatial location identifier value in each semantic structure parameter value to obtain the region center location parameter value and spatial distribution range parameter value of the corresponding semantic region;

[0122] Based on the regional center location parameter value and the spatial distribution range parameter value, the spatial regions corresponding to the transmission lines, towers, and vegetation are determined. The initial risk propagation boundary is generated based on the spatial boundary of the spatial region corresponding to the transmission lines, and the risk propagation constraint boundary is generated based on the spatial boundaries of the spatial regions corresponding to the towers and vegetation.

[0123] A spatial location index is established in the semantic gravitational potential energy field of the power transmission channel, the comprehensive semantic potential energy value corresponding to all spatial locations is read, and the minimum spatial distance value from each spatial location in the power transmission channel to the initial boundary of risk propagation is calculated.

[0124] For spatial locations on the initial boundary and the constraint boundary of risk propagation, the corresponding comprehensive semantic potential value is determined as the risk propagation value at that spatial location;

[0125] For spatial locations in the transmission channel that are not located on the initial boundary of risk propagation, read the corresponding comprehensive semantic potential energy value and minimum spatial distance value, divide the corresponding comprehensive semantic potential energy value by the minimum spatial distance value, and generate the risk propagation value for that spatial location.

[0126] The risk propagation values ​​corresponding to all spatial locations are organized according to the spatial location index order, and the risk propagation constraint boundary is used as the risk diffusion boundary range to generate the risk diffusion field of the power transmission channel.

[0127] In this embodiment, the generation of the foreign object field response state sequence includes:

[0128] Read the semantic segmentation result sequence of the power transmission channel, identify the semantic region corresponding to the foreign object target in the semantic segmentation result at each time, and extract the coordinate value of the center position of the region corresponding to the foreign object target;

[0129] According to the temporal order of the semantic segmentation result sequence of the power transmission channel, the difference calculation is performed on the regional center position coordinates of the same foreign object target at adjacent time points to obtain the displacement distance value between the two time points, and then normalized to generate the foreign object target position change parameter value.

[0130] Read the coordinates of the center of the region at two consecutive time points, calculate the difference between the horizontal and vertical coordinates between the current time and the previous time point, calculate the displacement distance between the two time points based on the difference between the horizontal and vertical coordinates, and normalize the displacement distance value to generate the parameter value of the change in the position of the foreign object target.

[0131] Based on the coordinates of the center of the region, the comprehensive semantic potential energy value of the corresponding spatial location in the semantic gravitational potential energy field of the power transmission channel is read and normalized to generate the foreign object potential energy response value.

[0132] Based on the coordinates of the center of the region, the risk propagation value of the corresponding spatial location in the risk diffusion field of the power transmission channel is read and normalized to generate the foreign object risk response value;

[0133] The foreign object potential energy response value and the foreign object risk response value are weighted and summed to generate the foreign object field response intensity value. The foreign object field response intensity value and the foreign object target position change parameter value are then multiplied to generate the field response value at that moment.

[0134] Based on the temporal order of the semantic segmentation results sequence of the power transmission channel, the field response values ​​corresponding to each moment are recorded sequentially to generate the foreign object field response state sequence.

[0135] In this embodiment, the generation of the phase synchronization state sequence of the power transmission channel includes:

[0136] Read the semantic segmentation result sequence of the power transmission channel, extract the coordinate values ​​of the center position of each semantic region in the semantic segmentation results at each time, and generate the semantic region position parameter values ​​corresponding to each semantic region.

[0137] Based on the semantic region location parameter values ​​corresponding to each semantic region, establish the location association relationship of consecutive moments in chronological order, and generate the temporal location sequence value corresponding to each semantic region.

[0138] Read the temporal position sequence values ​​corresponding to each semantic region, calculate the horizontal coordinate difference and vertical coordinate difference for the coordinate values ​​of the center position of the region at adjacent times, and generate the position change direction parameter values ​​corresponding to each semantic region.

[0139] Based on the position change direction parameter values ​​corresponding to each semantic region, directional encoding is performed to generate phase representation values ​​corresponding to each semantic region;

[0140] The system reads the position change direction parameter values ​​corresponding to each semantic region and extracts the horizontal and vertical coordinate differences from these parameters. Then, it determines the corresponding direction category identifier value based on the positive and negative relationships between the horizontal and vertical coordinate differences. Specifically, a first direction category identifier value is generated when both the horizontal and vertical coordinate differences are positive; a second direction category identifier value is generated when both are negative; a third direction category identifier value is generated when both are negative; a fourth direction category identifier value is generated when both are positive and negative; and a fifth direction category identifier value is generated when both are zero. A fifth directional category identifier value is generated when the difference is positive; a sixth directional category identifier value is generated when the difference in the horizontal coordinate is zero and the difference in the vertical coordinate is negative; a seventh directional category identifier value is generated when the difference in the horizontal coordinate is positive and the difference in the vertical coordinate is zero; an eighth directional category identifier value is generated when the difference in the horizontal coordinate is negative and the difference in the vertical coordinate is zero; and a ninth directional category identifier value is generated when both the difference in the horizontal coordinate and the difference in the vertical coordinate are zero. The directional category identifier values ​​are numerically encoded to generate the phase representation value of the corresponding semantic region. The phase representation value is a numerical value used to represent the motion direction state of the semantic region, generated by encoding the position change direction parameter values ​​of the semantic region at continuous time intervals.

[0141] Read the phase representation values ​​corresponding to each semantic region at the same time, calculate the phase difference between each semantic region and the corresponding semantic region of the transmission line, and average all the phase differences to generate the phase synchronization feature value corresponding to that time.

[0142] The phase synchronization characteristic values ​​corresponding to each moment are organized sequentially according to the time index to generate the phase synchronization state sequence of the power transmission channel.

[0143] In this embodiment, the generation of the foreign object intrusion determination result includes:

[0144] Read the field response update values ​​at each time step in the foreign object field response state sequence, and extract the field response update values ​​corresponding to consecutive time steps in chronological order to generate foreign object field response trend values;

[0145] Read the phase synchronization feature values ​​at each moment in the phase synchronization state sequence of the power transmission channel, and extract the phase synchronization feature values ​​corresponding to consecutive moments in the order of time index to generate phase synchronization offset trend values;

[0146] Based on the trend value of the foreign object field response, determine whether the foreign object field response state sequence meets the condition of continuous convergence to the high-risk area corresponding to the transmission line, and generate the foreign object field response judgment value.

[0147] The system reads the foreign object field response trend values ​​corresponding to consecutive time points and calculates the difference between the foreign object field response trend values ​​at adjacent time points in chronological order to generate response change values ​​corresponding to each consecutive time point. Then, it judges each response change value and marks the time when the response change value is less than the response change value of the previous time point as the convergence time, generating convergence flag values ​​corresponding to each consecutive time point. It performs continuous counting processing on all convergence flag values ​​to generate a continuous convergence count value. Then, it reads a preset continuous convergence count threshold and compares the continuous convergence count value with the preset continuous convergence count threshold. When the continuous convergence count value is greater than or equal to the preset continuous convergence count threshold, it generates a foreign object field response judgment value that meets the continuous convergence condition. When the continuous convergence count value is less than the preset continuous convergence count threshold, it generates a foreign object field response judgment value that does not meet the continuous convergence condition.

[0148] Based on the phase synchronization offset trend value, determine whether the phase synchronization state sequence of the transmission channel meets the preset abnormal offset condition, and generate a phase synchronization judgment value;

[0149] The system reads the phase synchronization offset trend values ​​corresponding to consecutive time points and calculates the difference between the phase synchronization offset trend values ​​of adjacent time points in chronological order to generate the offset change value corresponding to each consecutive time point. Then, it performs item-by-item comparison processing on all offset change values, and marks the time when the offset change value is greater than the offset change value of the previous time point as an abnormal offset time point, generating an abnormal offset identifier value corresponding to each consecutive time point. Next, it performs continuous counting processing on all abnormal offset identifier values ​​to generate a continuous abnormal offset count value. It reads a preset abnormal offset count threshold and compares the continuous abnormal offset count value with the preset abnormal offset count threshold. When the continuous abnormal offset count value is greater than or equal to the preset abnormal offset count threshold, a phase synchronization judgment value that meets the preset abnormal offset condition is generated. When the continuous abnormal offset count value is less than the preset abnormal offset count threshold, a phase synchronization judgment value that does not meet the preset abnormal offset condition is generated.

[0150] A joint judgment process is performed on the foreign object field response judgment value and the phase synchronization judgment value;

[0151] When the foreign object field response judgment value meets the continuous convergence condition and the phase synchronization judgment value meets the preset abnormal offset condition, a foreign object intrusion identification value is generated to indicate that the foreign object target is a foreign object intruding into the power transmission channel.

[0152] When the foreign object field response judgment value does not meet the continuous convergence condition or the phase synchronization judgment value does not meet the preset abnormal offset condition, a foreign object intrusion identification value is generated to indicate that the foreign object target is not a foreign object intruding into the power transmission channel.

[0153] Foreign object targets are candidate analysis objects identified and extracted from the scene based on the semantic segmentation result sequence of the power transmission channel. Before entering the final joint judgment step, it does not mean that it has been identified as a foreign object intruding into the power transmission channel, but rather that it participates in the risk analysis as a candidate object. Only when the foreign object field response judgment value meets the continuous convergence condition and the phase synchronization judgment value meets the preset abnormal offset condition will the foreign object target be identified as a foreign object intruding into the power transmission channel.

[0154] Foreign object intrusion determination results are generated based on foreign object intrusion identifier values.

[0155] Example 1: To verify the feasibility of this invention in practice, it was applied to the monitoring environment of a high-voltage transmission line corridor in a coastal area. The transmission lines in this area cross farmland, woodland, and residential areas, resulting in a complex surrounding environment and strong seasonal winds. Objects such as plastic film, advertising banners, and vegetation branches frequently enter the transmission corridor and approach the conductors under wind force. Traditional monitoring relies mainly on manual inspections or video surveillance observation. When there are many foreign objects or significant changes in ambient light, monitoring personnel find it difficult to promptly determine whether foreign objects are approaching the conductor area, easily leading to misjudgments or missed detections. This is especially true at night or under backlight conditions, making it even more difficult to identify the movement trend of foreign objects, posing a potential risk to the safe operation of the transmission lines.

[0156] In this environment, continuous monitoring images acquired by image acquisition devices installed along the power transmission channel are integrated into the foreign object intrusion detection system constructed using the method of this invention. The system first performs time alignment, distortion correction, illumination normalization, and noise suppression on the power transmission channel monitoring images to ensure that monitoring images from different times maintain a uniform spatial scale and image quality. Subsequently, a deep semantic segmentation network is used to perform pixel-level semantic classification on the processed images, semantically recognizing power transmission lines, tower structures, vegetation areas, ground areas, building areas, and foreign object targets in the images, and generating corresponding semantic segmentation results. The system further extracts structural features such as the contour shape, area, boundary length, and aspect ratio of each semantic region based on the semantic segmentation results, and simultaneously calculates the spatial center position and distribution range of each semantic region in the image, forming a unified set of semantic structural parameters. Based on this, the system assigns semantic field parameters according to the risk attributes of different semantic regions, and uses the spatial center position of each semantic region as the field source position. By calculating the semantic potential energy contribution value, a semantic gravitational potential energy field of the power transmission channel is established, thereby enabling the description of the influence relationship between different semantic regions on the movement behavior of foreign objects within a spatial range. Further, the system constructs an initial risk propagation boundary by combining the spatial boundary of the transmission line area and calculates the risk diffusion field of the transmission channel based on the semantic gravitational potential energy field, enabling the dynamic reflection of the potential risk propagation trend of foreign objects in space. When the system identifies a foreign object target in the image, it calculates the trajectory of the foreign object by the continuous changes in its center position, and generates a foreign object field response state sequence by combining the potential energy response and the risk diffusion response. Simultaneously, the system encodes the direction of continuous position changes in each semantic region, generates phase characterization values, and calculates phase synchronization features, forming a phase synchronization state sequence of the transmission channel. By jointly analyzing the foreign object field response state sequence and the phase synchronization state sequence, the system can determine whether the foreign object is continuously converging towards the high-risk area of ​​the transmission line and further provide the foreign object intrusion determination result.

[0157] To verify the performance of the present invention in practice, it was compared with traditional methods, and the results are shown in Table 1.

[0158] Table 1. Performance Comparison of Foreign Object Intrusion Detection Methods in Power Transmission Channels

[0159] Detection method Foreign object recognition accuracy False positive rate Average detection response time Intrusion determination accuracy Conventional target detection method 86.7% 9.8% 0.92 s 84.3% Semantic segmentation detection method 90.4% 7.1% 0.81 s 89.2% Method of the invention 96.8% 2.9% 0.63 s 94.1%

[0160] As shown in Table 1, under the same power transmission channel monitoring data conditions, the method of this invention outperforms traditional detection methods in several key indicators. Traditional target detection methods mainly rely on target morphological features to identify foreign objects. In complex background environments, they are easily affected by vegetation, building outlines, and changes in lighting, resulting in relatively low recognition accuracy and a high false alarm rate. The semantic segmentation detection method performs pixel-level semantic recognition on power transmission lines, tower structures, and surrounding environmental areas, enabling the system to distinguish different scene areas. This improves the recognition accuracy and reduces the false alarm rate to some extent. However, this method still mainly relies on the spatial information of a single frame image and lacks further analysis of the movement trend of foreign objects and the relationship of risk propagation. Therefore, there is still room for improvement in the accuracy of intrusion determination.

[0161] In contrast, the method of this invention further extracts semantic structural parameters of the transmission channel based on deep semantic segmentation, and constructs a semantic gravitational potential energy field and risk diffusion field for the transmission channel. This allows the spatial relationship between the transmission conductor, tower structure, and vegetation area to be quantitatively described through potential energy distribution, enabling the system to analyze the movement trend of foreign objects from a spatial structural perspective. Simultaneously, this invention combines the foreign object field response state sequence with the transmission channel phase synchronization state sequence for joint judgment, allowing the detection process to consider not only the positional changes of the foreign object target but also the dynamic changes in the overall semantic structure of the transmission channel. It is precisely because of the introduction of semantic potential energy field modeling, risk diffusion analysis, and temporal phase synchronization mechanisms that the system can more accurately identify whether foreign objects are approaching high-risk areas of the transmission conductor, thus demonstrating advantages in indicators such as identification accuracy, false alarm rate, and intrusion determination accuracy. Furthermore, because risk propagation analysis can identify potential dangerous trends in advance, the system's average detection response time is also shortened, further improving the real-time performance and reliability of foreign object intrusion detection in transmission channels.

[0162] The above are merely preferred embodiments of the present invention, but the scope of protection of the present invention is not limited thereto. Any equivalent substitutions or modifications made by those skilled in the art within the scope of the technology disclosed in the present invention, based on the technical solution and inventive concept of the present invention, should be covered within the scope of protection of the present invention.

Claims

1. A method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation, characterized in that, Includes the following steps: Acquire monitoring image sequences of power transmission channels, perform preprocessing on the monitoring image sequences of power transmission channels, and generate standardized image sequences of power transmission channels; The normalized image sequence of the power transmission channel is input into a deep semantic segmentation network to perform pixel-level semantic classification and generate a semantic segmentation result sequence of the power transmission channel; Based on the semantic segmentation result sequence of the transmission channel, the semantic structure features corresponding to each semantic target in the transmission channel are extracted to generate a set of semantic structure parameters for the transmission channel; Semantic field parameters are assigned to each semantic target in the semantic structure parameter set of the power transmission channel, and the semantic potential energy distribution is calculated to generate the semantic gravitational potential energy field of the power transmission channel. The initial boundary of risk propagation in the transmission channel is determined based on the set of semantic structure parameters of the transmission channel, and risk propagation calculation is performed on the semantic gravitational potential energy field of the transmission channel to generate the risk diffusion field of the transmission channel; Extract the center position coordinates of the foreign object target at continuous time intervals, calculate the position change parameters, and combine the semantic gravitational potential energy field and risk diffusion field of the transmission channel to generate the foreign object field response state sequence. Construct the temporal position sequence of each semantic region, calculate the phase difference, and generate the phase synchronization state sequence of the transmission channel; The foreign object field response state sequence and the power transmission channel phase synchronization state sequence are jointly determined to generate the foreign object intrusion determination result.

2. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The preprocessing includes time alignment, distortion correction, illumination normalization, noise suppression, and spatial scale unification.

3. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The power transmission channel monitoring image sequence is a collection of multiple monitoring image data that reflect the changes in the state of the power transmission conductor and its surrounding environment, obtained continuously in chronological order by image acquisition equipment in the power transmission line channel area.

4. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the semantic segmentation result sequence of the power transmission channel includes: Each frame of the normalized image sequence of the power transmission channel is input into the deep semantic segmentation network in time index order, and size adaptation is performed on each frame. A deep semantic segmentation network is used to perform multi-layer feature extraction on each frame of the standardized image sequence of the power transmission channel. In the network encoding stage, the texture, edge and structure in the image are mapped layer by layer to generate a set of feature representations of the power transmission channel image. In the decoding stage of the deep semantic segmentation network, feature recovery and feature fusion are performed on the feature representation set of the power transmission channel image. The high-level semantic features and low-level spatial structure features are combined by a multi-scale feature fusion method to generate a set of semantic feature maps of the power transmission channel. Based on the semantic classification output layer of the deep semantic segmentation network, semantic category determination is performed on the set of semantic feature maps of the power transmission channel to obtain the semantic category label corresponding to each pixel; The semantic category labels of all pixels in each frame of the image are reorganized according to their spatial location in the image to generate a semantic annotation map; Based on the temporal order of the standardized image sequence of the power transmission channel, the semantic annotation maps corresponding to each frame of the image are combined temporally to form a semantic segmentation result sequence of the power transmission channel.

5. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the semantic structure parameter set of the power transmission channel includes: Extract the semantic regions corresponding to each semantic category from the semantic annotation maps of each frame in the semantic segmentation result sequence of the power transmission channel, and divide the semantic regions into connected components according to the pixel semantic category identifier to obtain the set of semantic regions corresponding to each semantic target; For each semantic region in the semantic region set, region contour extraction is performed, and the corresponding geometric morphological features are calculated based on the semantic region contour to form a semantic region structural feature set; Spatial location calculations are performed on each semantic region. The corresponding spatial center position is determined based on the pixel distribution position of the semantic region in the image, and the spatial distribution range of the semantic region is extracted to obtain the set of semantic region spatial location parameters. Based on the set of semantic region structural features and the set of semantic region spatial location parameters, a correspondence between semantic regions and the spatial structure of power transmission channels is established, and the spatial location changes of each semantic region in consecutive image frames are correlated to form a set of semantic region temporal location correlations. Based on the set of semantic region structural features, the set of semantic region spatial location parameters, and the set of semantic region temporal location associations, a unified structural description is performed on each semantic target in the power transmission channel, and normalization processing is carried out to generate a set of semantic structural parameters for the power transmission channel.

6. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the semantic gravitational potential field of the power transmission channel includes: Read the set of semantic structure parameters of the power transmission channel, and parse the semantic region structure identifier value, semantic region spatial location identifier value and semantic region location change identifier value in each semantic structure parameter value to obtain the structural state parameter value, spatial location parameter value and change state parameter value corresponding to each semantic region; Based on the semantic category corresponding to each semantic region, its risk attribute is determined. According to the correspondence between semantic category and risk attribute, structural field strength parameter value is assigned to the structural state parameter value of each semantic region, field source position parameter value is assigned to the spatial position parameter value of each semantic region, and dynamic modulation parameter value is assigned to the changing state parameter value of each semantic region, thereby generating the semantic field parameter value of the corresponding semantic region. Using the field source location parameter values ​​of each semantic region as the spatial field source location, each semantic region is mapped to the corresponding spatial field source unit. The initial field strength value of each spatial field source unit is determined according to the structural field strength parameter value. The dynamic field strength correction value of each spatial field source unit is determined according to the dynamic modulation parameter value. The field source action parameter value corresponding to each spatial field source unit is generated. For each spatial source unit, its center of action in the transmission channel space is determined based on the source position parameter value. The semantic potential energy contribution value of the spatial source unit at each spatial position in the transmission channel is calculated based on the source action parameter value, and the semantic potential energy distribution value of the corresponding spatial source unit is generated. According to the spatial location relationship of the power transmission channel, the semantic potential energy distribution values ​​corresponding to all spatial field source units are aggregated to generate the comprehensive semantic potential energy value corresponding to each spatial location, and the semantic potential energy distribution matrix of the power transmission channel is formed based on all comprehensive semantic potential energy values. Based on the semantic potential energy distribution matrix of the power transmission channel, establish the correspondence between each spatial location and the comprehensive semantic potential energy value, and generate the semantic gravitational potential energy field of the power transmission channel.

7. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the risk diffusion field of the power transmission channel includes: Read all semantic structure parameter values ​​in the set of semantic structure parameters of the power transmission channel, and parse the semantic region spatial location identifier value in each semantic structure parameter value to obtain the region center location parameter value and spatial distribution range parameter value of the corresponding semantic region; Based on the regional center location parameter value and the spatial distribution range parameter value, the spatial regions corresponding to the transmission lines, towers, and vegetation are determined. The initial risk propagation boundary is generated based on the spatial boundary of the spatial region corresponding to the transmission lines, and the risk propagation constraint boundary is generated based on the spatial boundaries of the spatial regions corresponding to the towers and vegetation. A spatial location index is established in the semantic gravitational potential energy field of the power transmission channel, the comprehensive semantic potential energy value corresponding to all spatial locations is read, and the minimum spatial distance value from each spatial location in the power transmission channel to the initial boundary of risk propagation is calculated. For spatial locations on the initial boundary and the constraint boundary of risk propagation, the corresponding comprehensive semantic potential value is determined as the risk propagation value at that spatial location; For spatial locations in the transmission channel that are not located on the initial boundary of risk propagation, read the corresponding comprehensive semantic potential energy value and minimum spatial distance value, divide the corresponding comprehensive semantic potential energy value by the minimum spatial distance value, and generate the risk propagation value for that spatial location. The risk propagation values ​​corresponding to all spatial locations are organized according to the spatial location index order, and the risk propagation constraint boundary is used as the risk diffusion boundary range to generate the risk diffusion field of the power transmission channel.

8. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the foreign object field response state sequence includes: Read the semantic segmentation result sequence of the power transmission channel, identify the semantic region corresponding to the foreign object target in the semantic segmentation result at each time, and extract the coordinate value of the center position of the region corresponding to the foreign object target; According to the temporal order of the semantic segmentation result sequence of the power transmission channel, the difference calculation is performed on the regional center position coordinates of the same foreign object target at adjacent time points to obtain the displacement distance value between the two time points, and then normalized to generate the foreign object target position change parameter value. Based on the coordinates of the center of the region, the comprehensive semantic potential energy value of the corresponding spatial location in the semantic gravitational potential energy field of the power transmission channel is read and normalized to generate the foreign object potential energy response value. Based on the coordinates of the center of the region, the risk propagation value of the corresponding spatial location in the risk diffusion field of the power transmission channel is read and normalized to generate the foreign object risk response value; The foreign object potential energy response value and the foreign object risk response value are weighted and summed to generate the foreign object field response intensity value. The foreign object field response intensity value and the foreign object target position change parameter value are then multiplied to generate the field response value at that moment. Based on the temporal order of the semantic segmentation results sequence of the power transmission channel, the field response values ​​corresponding to each moment are recorded sequentially to generate the foreign object field response state sequence.

9. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the phase synchronization state sequence of the power transmission channel includes: Read the semantic segmentation result sequence of the power transmission channel, extract the coordinate values ​​of the center position of each semantic region in the semantic segmentation results at each time, and generate the semantic region position parameter values ​​corresponding to each semantic region. Based on the semantic region location parameter values ​​corresponding to each semantic region, establish the location association relationship of consecutive moments in chronological order, and generate the temporal location sequence value corresponding to each semantic region. Read the temporal position sequence values ​​corresponding to each semantic region, calculate the horizontal coordinate difference and vertical coordinate difference for the coordinate values ​​of the center position of the region at adjacent times, and generate the position change direction parameter values ​​corresponding to each semantic region. Based on the position change direction parameter values ​​corresponding to each semantic region, directional encoding is performed to generate phase representation values ​​corresponding to each semantic region; Read the phase representation values ​​corresponding to each semantic region at the same time, calculate the phase difference between each semantic region and the corresponding semantic region of the transmission line, and average all the phase differences to generate the phase synchronization feature value corresponding to that time. The phase synchronization characteristic values ​​corresponding to each moment are organized sequentially according to the time index to generate the phase synchronization state sequence of the power transmission channel.

10. The method for detecting foreign object intrusion in power transmission channels based on deep semantic segmentation according to claim 1, characterized in that, The generation of the foreign object intrusion determination result includes: Read the field response update values ​​at each time step in the foreign object field response state sequence, and extract the field response update values ​​corresponding to consecutive time steps in chronological order to generate foreign object field response trend values; Read the phase synchronization feature values ​​at each moment in the phase synchronization state sequence of the power transmission channel, and extract the phase synchronization feature values ​​corresponding to consecutive moments in the order of time index to generate phase synchronization offset trend values; Based on the trend value of the foreign object field response, determine whether the foreign object field response state sequence meets the condition of continuous convergence to the high-risk area corresponding to the transmission line, and generate the foreign object field response judgment value. Based on the phase synchronization offset trend value, determine whether the phase synchronization state sequence of the transmission channel meets the preset abnormal offset condition, and generate a phase synchronization judgment value; A joint judgment process is performed on the foreign object field response judgment value and the phase synchronization judgment value; When the foreign object field response judgment value meets the continuous convergence condition and the phase synchronization judgment value meets the preset abnormal offset condition, a foreign object intrusion identification value is generated to indicate that the foreign object target is a foreign object intruding into the power transmission channel. When the foreign object field response judgment value does not meet the continuous convergence condition or the phase synchronization judgment value does not meet the preset abnormal offset condition, a foreign object intrusion identification value is generated to indicate that the foreign object target is not a foreign object intruding into the power transmission channel. Foreign object intrusion determination results are generated based on foreign object intrusion identifier values.

Images