Power transmission line inspection method based on multi-time-sequence image sequence analysis

Abstract

The invention discloses a power transmission line inspection method based on a multi-time-sequence image sequence. The method comprises the following steps: constructing a structure scene base model of a multi-time sequence image, and generating a base field for describing a stable structure of a power transmission line; extracting a multi-level dynamic response feature of the time sequence image on the substrate field, and constructing a time sequence feature vector chain; constructing a time sequence consistency evaluation graph according to the feature vector chain, deducing an evolution mode of the line structure, and generating a time sequence consistency index; embedding and track analysis are carried out on the consistency index in the submerged space, abnormal offset deviating from a normal evolution track is identified, and line abnormal probability scores under multiple time sequences are obtained; and executing closed-loop reasoning based on the abnormal probability score, eliminating false anomalies, locking an unstable path of a real structure, and finally outputting an inspection conclusion of the power transmission line. The accuracy and the intelligent level of power transmission line inspection in a complex environment can be improved.

Description

Technical Field

[0001] This invention relates to the field of intelligent inspection of power transmission lines and multimodal image analysis, specifically to a method for power transmission line inspection based on multi-temporal image sequence analysis. Background Technology

[0002] With the continuous expansion of the scale of cross-regional and long-distance power transmission line deployment, higher requirements are being placed on their operational safety and reliability. To reduce the cost of manual inspections and improve inspection efficiency, intelligent inspection methods based on image recognition and video analysis have been rapidly developed. However, existing technologies mainly suffer from the following problems:

[0003] It relies heavily on static features and struggles to depict the evolution of structures over time. Traditional inspection methods are mostly based on single-frame images or a small number of time-segment samples, using methods such as target detection and defect identification to determine local anomalies in conductors, tower materials, or insulators. They lack the ability to model the evolution of structures over multiple time periods and are unable to identify structural risks caused by the accumulation of subtle changes.

[0004] Feature extraction methods are sensitive to illumination, angle, and environmental disturbances, resulting in insufficient stability. Existing image-based inspection algorithms generally rely on low-level features such as edges, textures, and colors. These features are significantly affected by factors such as changes in illumination, viewpoint shifts, and occlusion at different times, leading to significant fluctuations in judgment results across time periods.

[0005] Inferences are made based on images at a single moment, lacking a complete temporal reasoning chain. Most current inspection technologies treat anomaly identification as "instantaneous decision-making," while changes in the structure of transmission lines often have gradual evolutionary characteristics, such as icing accumulation, stress deformation, and component loosening. Their abnormal signals are usually not obvious in the early stages, making it difficult to detect them in a timely manner based on single-frame identification.

[0006] The lack of verification and closed-loop reasoning mechanisms for abnormal results leads to a high false alarm rate. In existing solutions, factors such as noise, lighting, and wind disturbances can easily trigger incorrect identification, while the system lacks proactive verification and backtracking mechanisms, making it unable to effectively eliminate false anomalies and resulting in wasted maintenance resources.

[0007] In summary, existing transmission line inspection technologies still have significant shortcomings in utilizing multi-temporal information, modeling continuous changes, and reasoning about anomalies. There is an urgent need for a technical solution that can extract structurally stable features from multi-temporal image sequences, construct dynamic evolution models, and improve the accuracy of anomaly identification through latent space reasoning and closed-loop verification. Summary of the Invention

[0008] This invention provides a method for power transmission line inspection based on multi-time-series image sequence analysis, comprising:

[0009] S10. Construct a structural scene basis model for multi-temporal images. By performing brightness drift adaptive calibration, background structure decomposition and stable feature extraction on cross-temporal image sequences, a basis field is generated to describe the stable structure of transmission lines.

[0010] S20. Extract multi-level dynamic response features of temporal images on the base field, and construct a temporal feature vector chain through multi-scale differential coding, texture perturbation tracking and local deformation path analysis to characterize the dynamic change mode of the line structure.

[0011] S30. Construct a temporal consistency evaluation graph based on the feature vector chain, deduce the evolution mode of the circuit structure through indicators such as structural change amplitude, texture perturbation continuity and deformation propagation direction, and generate a temporal consistency index that reflects the rationality of evolution.

[0012] S40. Embed the consistency index and perform trajectory analysis in the latent space. By defining latent space reachability and physical consistency constraints, identify abnormal deviations from the normal evolution trajectory and obtain the line anomaly probability score under multiple time series.

[0013] S50. Based on the anomaly probability score, perform closed-loop reasoning, verify and eliminate false anomalies through reverse trajectory, lock the real structural instability path, and finally output the inspection conclusion of the transmission line.

[0014] The transmission line inspection method based on multi-temporal image sequence analysis described above includes constructing a structural scene basis model of multi-temporal images. This model generates a basis field describing the stable structure of the transmission line by performing brightness drift adaptive calibration, background structure decomposition, and stability feature extraction on the cross-temporal image sequence. This includes:

[0015] Adaptive brightness calibration is performed on cross-time image sequences based on a local brightness estimation function, dynamically adjusting the illumination difference according to the brightness trend of adjacent frames to form an input sequence with consistent brightness;

[0016] Background structure decomposition is performed on the brightness-calibrated image sequence. By separating the changing components of the scene, the main structure of the line that remains stable is extracted to generate a structural base field that describes the static morphology of the line.

[0017] The transmission line inspection method based on multi-temporal image sequence analysis described above involves extracting multi-level dynamic response features from temporal images on the base field. A temporal feature vector chain is constructed through multi-scale differential coding, texture perturbation tracking, and local deformation path analysis to characterize the dynamic change patterns of the line structure, including:

[0018] Multi-scale difference features are calculated based on the difference relationship between the base field and the corresponding time period image to characterize the degree of local change of the line structure at different scales;

[0019] Based on the texture perturbation tracking model, the stability and change trend of cross-frame textures are analyzed, and the dynamic change trajectory is derived by combining the local deformation path to construct a continuous temporal feature vector chain.

[0020] The transmission line inspection method based on multi-temporal image sequence analysis described above includes analyzing the stability and change trend of cross-frame textures through a texture perturbation tracking model, including:

[0021] The local texture energy detection operator is used to extract cross-frame texture perturbation responses, and a texture continuity curve is constructed based on the perturbation response sequence.

[0022] Texture perturbations are classified according to the steady and abrupt regions of the texture continuity curve, and stable perturbation paths that significantly contribute to structural dynamic changes are selected to provide input for deformation path analysis.

[0023] The transmission line inspection method based on multi-temporal image sequence analysis described above includes constructing a temporal consistency evaluation map based on the feature vector chain, deriving the evolution mode of the transmission line structure through indicators such as structural change amplitude, texture disturbance continuity, and deformation propagation direction, and generating a temporal consistency index reflecting the rationality of the evolution, including:

[0024] The structural change magnitude for each time period is calculated based on the feature vector chain, and an initial consistency evaluation map is constructed to identify potential structural drift regions.

[0025] Texture perturbation continuity and deformation propagation direction are used as dynamic weights in the consistency evaluation graph to deduce the overall evolution trend of the structure and generate a temporal consistency index.

[0026] The transmission line inspection method based on multi-temporal image sequence analysis described above involves embedding a consistency index and parsing the trajectory in the latent space. By defining latent space reachability and physical consistency constraints, it identifies abnormal deviations from the normal evolution trajectory and obtains a line anomaly probability score under multiple time series, including:

[0027] Latent space embedding is performed on the temporal consistency index to form an evolutionary trajectory in a low-dimensional measurable space to characterize the structural change pattern;

[0028] Based on the degree of trajectory deviation detected by latent space accessibility and physical consistency constraints, instantaneous deviation points that do not conform to normal evolutionary patterns are identified, and anomaly probability scores are calculated.

[0029] The transmission line inspection method based on multi-time-series image sequence analysis described above includes: performing closed-loop inference based on anomaly probability scoring; eliminating false anomalies and identifying the actual structural instability path through reverse trajectory verification; and finally outputting the inspection conclusions of the transmission line, specifically including:

[0030] Based on the anomaly probability score, the reverse trajectory of the anomaly points is traced back, and the repeatability of the anomaly points in multiple time periods is checked to confirm whether they belong to real structural disturbances.

[0031] False anomalies that fail the repeatability verification are removed, and structural stability inference is performed on the remaining anomaly paths to generate the final inspection conclusion.

[0032] This invention also provides a transmission line inspection system based on multi-time-series image sequence analysis, comprising:

[0033] The base field construction module is used to perform cross-time period brightness drift adaptive calibration, background structure decomposition and stable feature extraction to construct a structural scene base model of the transmission line;

[0034] The temporal feature extraction module is used to perform multi-scale differential coding, texture perturbation tracking and deformation path parsing on the basis field to generate a temporal feature vector chain.

[0035] The consistency evaluation module is used to construct a time series consistency evaluation graph based on the time series feature vector chain and generate a time series consistency index.

[0036] The latent space anomaly identification module is used to perform trajectory analysis on the consistency index based on latent space reachability and physical consistency constraints in order to calculate the anomaly probability score;

[0037] The closed-loop inference module is used to perform reverse trajectory verification based on the anomaly probability score and lock the actual structural instability path to output the final inspection conclusion.

[0038] The beneficial effects achieved by this invention are as follows: This invention constructs a structural scene basis model of multi-temporal images, establishes a temporal feature vector chain reflecting the dynamic changes of the structure, forms a temporal consistency index characterizing the rationality of structural evolution, and identifies the structural instability path based on latent space trajectory analysis and closed-loop inference mechanism. It realizes a unified framework for cross-time period stable feature extraction, fine characterization of dynamic changes, and high reliability verification of abnormal results. It can maintain high recognition accuracy under conditions of illumination change, viewpoint disturbance and environmental noise, effectively reduce the false alarm rate of inspection, and improve the accuracy and stability of transmission line anomaly identification. Attached Figure Description

[0039] To more clearly illustrate the technical solutions in the embodiments of the present invention or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments recorded in the present invention. For those skilled in the art, other drawings can be obtained based on these drawings.

[0040] Figure 1This is a flowchart of a transmission line inspection method based on multi-time-series image sequence analysis provided in Embodiment 1 of this application;

[0041] Figure 2 This is a schematic diagram of a power transmission line inspection system based on multi-time-series image sequence analysis provided in Embodiment 2 of this application. Detailed Implementation

[0042] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0043] Example 1

[0044] like Figure 1 As shown, Embodiment 1 of this application provides a method for inspecting transmission lines based on multi-time-series image sequence analysis, including the following steps:

[0045] S10. Construct a structural scene basis model for multi-temporal images. By performing brightness drift adaptive calibration, background structure decomposition and stable feature extraction on cross-temporal image sequences, a basis field is generated to describe the stable structure of transmission lines.

[0046] In multi-time-period inspection scenarios, the overall brightness and local contrast of transmission line images are highly uneven due to changes in weather, acquisition time, angle, and imaging exposure mechanisms. Furthermore, the background area, vegetation disturbance, and shadow changes are extremely significant in images from different time periods. If dynamic structural analysis is performed directly on the original image sequence, illumination disturbances will be misjudged as structural drift, and short-term background changes will interfere with the stability identification of line components, leading to unreliable analysis paths. Therefore, this step constructs a scene base with a unified illumination hierarchy and long-term static structural representation through brightness adaptive calibration, background structure stripping, and stable region selection. Specifically, it includes the following sub-steps:

[0047] S101. Perform adaptive brightness calibration on the cross-time period image sequence based on the local brightness estimation function, and dynamically adjust the illumination difference according to the brightness trend of adjacent frames to form an input sequence with consistent brightness.

[0048] Brightness calibration addresses the significant brightness drift across the same line area at different acquisition times. It constructs a brightness estimation function to estimate inter-frame deviations by statistically analyzing local grayscale values ​​around towers, conductors, and insulators. This function derives correction values ​​based on multi-frame local brightness relationships, adjusting pixel grayscale values ​​regionally to achieve consistency in overall image brightness, local contrast, and texture, thereby mitigating grayscale differences caused by variations in weather conditions, backlighting, and background light reflection. Brightness calibration is not a simple global equalization but rather an adaptive compensation for brightness trends in structural regions, ensuring uniform illumination across multiple time periods and maintaining texture stability at the spatial detail level, providing ideal input for structural decomposition and temporal tracking.

[0049] S102. Perform background structure decomposition on the brightness-calibrated image sequence, extract the stable main structure of the line by separating the changing components of the scene, and generate a structural base field describing the static morphology of the line.

[0050] Even under consistent brightness conditions, short-period perturbations still exist in multi-frame images, such as swaying leaves, cloud drift, changes in ground reflection, or occasional obstructions. If these changes are not removed, the line components will be affected by background interference, leading to spurious change identification. Background structure decomposition relies on the magnitude of pixel changes and texture persistence over time. Pixel regions that remain stable over a long period are considered the main structure, while frequently changing regions are considered background elements. Structural pixels exhibit significant characteristics in spatial connectivity, geometric linearity, and edge consistency. Region filtering can eliminate isolated texture points and irrelevant background interference, ultimately retaining stable structural regions such as towers, crossarms, conductors, and insulators. After multi-frame fusion, these structural regions form a static structural base field within a unified coordinate framework, providing a location reference for subsequent dynamic change analysis and avoiding the influence of short-term perturbations and illumination contrast on the temporal judgment process. The structural base field can be described as a set of stable pixels, whose corresponding regions have continuous texture representation and identifiable geometric features over time, providing data sources for differential operators, perturbation models, and trajectory analysis in subsequent steps.

[0051] S20. Extract multi-level dynamic response features of temporal images on the base field, and construct a temporal feature vector chain through multi-scale differential coding, texture perturbation tracking and local deformation path analysis to characterize the dynamic change mode of the line structure.

[0052] This step aims to use the basis field as a stable reference to systematically model the local differences, texture perturbation behavior, and deformation propagation trends of images across time periods, and to organize these dynamic features into a continuous vector chain in temporal order. Specifically, it includes the following sub-steps:

[0053] S201. Calculate multi-scale difference features based on the difference relationship between the base field and the corresponding time period image to characterize the degree of local change of the line structure at different scales.

[0054] Based on the structural base field, the system performs multi-scale differential encoding on images from different time periods using differential operators of different scales to identify local changes in the structural region. Small-scale differential encoding is used to magnify conductor details and subtle disturbances, while large-scale differential encoding is used to reveal the overall offset trend of local areas of towers or fittings. Through cross-scale differential fusion, the system obtains a set of differential features that reflect the magnitude of changes in the structural region.

[0055] S202. Based on the texture perturbation tracking model, analyze the stability and change trend of cross-frame textures, and combine the local deformation path to deduce the dynamic change trajectory in order to construct a continuous temporal feature vector chain.

[0056] After obtaining multi-scale differential features, the system further utilizes a texture perturbation tracking model to identify cross-frame texture change behavior. Cross-frame perturbation responses are extracted using a local texture energy detection operator, and texture continuity curves are generated based on the perturbation response sequences to characterize stable segments, abrupt change segments, and trend continuation regions of the texture over time. The system classifies perturbations according to the texture continuity curves and selects stable perturbation paths that can continuously reflect the evolution pattern of the line structure. Based on this, by analyzing the spatial displacement and direction of change of feature points in the structural region, local deformation paths are derived, thus comprehensively presenting the dynamic trajectory of structural changes.

[0057] After generating differential features, texture perturbation features, and deformation path features, the system jointly models multidimensional features from different sources to form a feature vector that can completely describe the dynamic change state of the structural region at time t. To achieve a stable and measurable dynamic fusion process, this step introduces a dynamic feature fusion model to uniformly calculate the structural change amplitude, perturbation continuity, local deformation path, and temporal neighborhood correlation. This model is defined as follows: The output of this dynamic fusion model The dynamic response intensity represents the intensity at time t, reflecting the overall degree of change of the structural region at that time. Normalization factor; These are the spatial coordinates of the pixel. It is a set of pixels representing stable structural regions extracted by S10, including pixel regions related to the main body of the line, such as towers, conductors, and insulators; S represents the specific scale number; S is the scale set. , where is the scale weight, representing the importance of scale s in the overall dynamic features; For scale s, there is a difference operator used to measure the pixel differences between the current frame and the base field in the structural region; Let p be the gray level or feature value of the image at time t; is the variance control coefficient for scale s, used to adjust the sensitivity to exponential decay; Used to adjust the contribution of texture perturbation energy to the overall dynamic characteristics; The intensity of texture perturbation at time t and pixel p is obtained from the texture perturbation tracking model and is used to characterize the stability and amplitude of texture changes across frames. This is the balance coefficient for the deformation path term; This refers to the local deformation path length or deformation intensity, describing the geometric offset of a pixel in the time dimension. This is the balance coefficient for the time-series correlation term; For neighborhood time index; The set of time windows around time t; This is a temporal correlation index of pixel p at time t in its neighborhood.

[0058] In obtaining the dynamic response scalar Then, the system constructs a feature vector by combining it with the difference features, texture perturbation features, and deformation path features according to a fixed dimension. Subsequently, the feature vectors of all time periods are arranged in chronological order to form a temporal feature vector chain used to describe the multi-time period evolution behavior of the line structure.

[0059] S30. Construct a temporal consistency evaluation graph based on the feature vector chain, deduce the evolution mode of the circuit structure through indicators such as structural change amplitude, texture perturbation continuity and deformation propagation direction, and generate a temporal consistency index that reflects the rationality of evolution.

[0060] The system takes a time-series feature vector chain as input, quantifies the degree of structural change in adjacent time periods, and combines the continuity of texture perturbation in the time dimension with the stability of deformation propagation direction to establish an evaluation map reflecting the evolution of the line structure over time. Subsequently, the system generates a consistency index based on these time-series indicators to measure whether the line structure remains continuous, stable, and conforms to physical evolution trends across multiple time periods. Specifically, it includes the following sub-steps:

[0061] S301. Calculate the structural change amplitude for each time period based on the feature vector chain and construct an initial consistency evaluation map to identify potential structural drift regions.

[0062] After obtaining the temporal feature vector chain, the system first calculates the structural change amplitude between adjacent time periods to describe the degree of structural difference between different time points. To fully reflect the morphological changes of the line structure at multiple scales and to incorporate texture perturbation and deformation direction into the change assessment, the system uses the following structural change amplitude model. : ,in, This represents the magnitude of structural change over time t; For scale indexing; For scale quantity; , These are the eigenvectors of time t and t–1 at scale s, respectively. For scale weights; It is a scaling index; For texture perturbation energy; As an indicator of texture continuity; Statistical characteristics characterizing the direction of deformation propagation; This is the direction adjustment factor; This represents the perturbation coupling coefficient. Through analysis of all time periods... By arranging the data chronologically, an initial consistency evaluation map can be generated. Regions with significantly increased variation amplitudes are used to identify structural locations that may exhibit drift, early deformation, or abnormal change trends.

[0063] S302. Apply the continuity of texture perturbation and the direction of deformation propagation as dynamic weights to the consistency evaluation graph, deduce the overall evolution trend of the structure and generate the temporal consistency index.

[0064] Based on the initial consistency evaluation map, the system further considers the temporal continuity of texture disturbances and the stability of deformation direction, making the judgment of structural evolution trends more consistent with the actual stress and environmental changes of the line. To this end, the system constructs a temporal consistency index. This is used to evaluate the rationality of structural changes from a temporal perspective. It is defined as follows: : ,in, Let be the consistency index at time t; Normalization factor; , , These are the weighting coefficients; , , These are the smoothing factors for structural changes, texture continuity, and deformation direction, respectively. As an indicator of texture continuity; Indicator of the direction of deformation propagation; and This represents the average of texture continuity and deformation direction within the temporal neighborhood, reflecting background trends. When a certain time period... When the value continues to decrease, it indicates that the structural evolution may deviate from the normal pattern, and the system marks this period as a suspected abnormal interval.

[0065] S40. Embed the consistency index and perform trajectory analysis in the latent space. By defining latent space reachability and physical consistency constraints, identify abnormal deviations from the normal evolution trajectory and obtain the line anomaly probability score under multiple time series.

[0066] This step takes the temporal consistency index as input and maps it to a low-dimensional latent space, allowing the overall trend of the line structure's changes over time to be presented as an evolutionary trajectory. By analyzing the continuity, smoothness, and reachability between adjacent points in the latent space, the system can determine whether the line structure evolves according to the expected pattern and identify potential deviation behaviors accordingly. Specifically, it includes the following sub-steps:

[0067] S401. Perform latent space embedding on the temporal consistency index to form an evolutionary trajectory in a low-dimensional measurable space to characterize the structural change pattern.

[0068] After obtaining the temporal consistency index sequence, the system projects it into a low-dimensional space using a latent space embedding method, forming a continuous trajectory for the time series in this space. The shape of the trajectory reflects the overall change law of structural evolution: when the line structure is stable and the consistency index is stable, the trajectory exhibits a smooth and uniform temporal distribution; if a sudden change occurs at a certain time period, the trajectory will show a shift or a discrete trend at that position. This trajectory provides a basic framework for judging the trend of structural changes, enabling the system to analyze the stability and shift of the structure over time at a more macroscopic scale.

[0069] S402. Based on the latent space accessibility and physical consistency constraints, detect the degree of trajectory deviation, identify instantaneous deviation points that do not conform to the normal evolution law, and calculate the anomaly probability score.

[0070] After establishing the trajectory in the latent space, the system determines whether a point in time deviates from the normal evolution path of the route by evaluating the reachability range and physical consistency constraints between trajectory points. If there is a significant distance deviation between a trajectory point and its temporal neighborhood, or if its consistency index is significantly inconsistent with the neighborhood statistics, it indicates that there may be a structural anomaly during that period.

[0071] Based on the degree of offset, the system uses the following formula to calculate the anomaly probability score. This is used to quantify the probability of an abnormal change occurring at that moment: ,in, This represents the offset of the trajectory at time t; This is the offset scale adjustment parameter, used to control the range of influence of the offset on the anomaly probability. The larger the value, the more likely the structural change is an anomalous offset. This score can reflect whether there are atypical disturbances in the line structure under multiple time-series conditions, such as sudden increases in local deformation, severe texture disturbances, and abrupt changes in suspension point direction.

[0072] S50. Based on the anomaly probability score, perform closed-loop reasoning, verify and eliminate false anomalies through reverse trajectory, lock the real structural instability path, and finally output the inspection conclusion of the transmission line.

[0073] This step, based on the previous latent space trajectory analysis, uses the anomaly probability scoring sequence as the starting point for inference. It performs dual verification of the continuity and structural correspondence of anomaly points in the time dimension to distinguish between genuine structural disturbances and false anomalies caused by lighting, noise, etc., thereby forming a three-dimensional, traceable inspection decision result. Specifically, it includes the following sub-steps:

[0074] S501. Based on the anomaly probability score, reverse the trajectory of the anomaly point and confirm whether it belongs to the real structural disturbance by checking the repeatability of the anomaly point in multiple time periods.

[0075] After obtaining the anomaly probability score, suspected anomalies exceeding the threshold are used as initial candidates, and their context frames in the time series are traced back in reverse. During the backtracking process, the system analyzes the consistency index changes of adjacent frames, the continuity of local texture disturbances, and the direction of deformation propagation to determine whether the anomaly exhibits a stable and traceable offset pattern across multiple time periods. If an anomaly can form a continuous offset chain with the preceding and following time periods and corresponds to the line structure topology, it is considered to have genuine disturbance characteristics and proceeds to the next verification stage; otherwise, it is marked as a suspected false anomaly and will be removed later.

[0076] S502. Remove pseudo-anomalies that fail the repeatability verification and perform structural stability reasoning on the remaining anomaly paths to generate the final inspection conclusion.

[0077] After completing the reverse trajectory comparison, the repetitive results of all suspected anomalies are screened, and non-persistent anomalies caused by factors such as transient noise, camera shake, and cloud shadow obstruction are eliminated. For the remaining abnormal paths, the system further analyzes their spatial propagation direction, variation amplitude, and correlation with critical structures of the line to determine whether the anomaly may cause structural instability or operational risks. After forming a stability judgment, the system integrates the severity and distribution characteristics of the abnormal paths to generate a final inspection conclusion, including levels such as normal, minor anomaly, significant deviation, or potential instability.

[0078] Example 2

[0079] like Figure 2 As shown, Embodiment 2 of this application provides a power transmission line inspection system based on multi-time-series image sequence analysis, including:

[0080] The base field construction module 21 is used to perform brightness drift adaptive calibration and background structure decomposition on multi-time period transmission line image sequences, extract long-term stable main structure regions of the line, and construct a structural scene base model for subsequent time-series analysis. Specifically, it includes the following sub-modules:

[0081] The brightness adaptive calibration submodule 211 is used to perform regional dynamic compensation of pixel grayscale in key structural areas such as towers, conductors, and insulators based on the local brightness estimation results of cross-time period image sequences. This submodule adjusts the illumination difference according to the brightness trend of adjacent frames, so that the overall brightness, local contrast, and texture performance of images across multiple time periods tend to be consistent, reducing grayscale drift caused by changes in weather, backlighting, and background reflection.

[0082] Background structure decomposition submodule 212 is used to analyze the variation amplitude and texture persistence of pixels in the time dimension based on the brightness-calibrated image sequence, distinguishing between short-period disturbance regions and long-term stable regions. By stripping away high-frequency variation regions such as leaf swaying, cloud drift, ground reflection changes, and occasional obstructions, only the main line structure regions such as towers, crossarms, conductors, and insulators, which exhibit stable performance in terms of spatial connectivity, geometric linearity, and edge consistency, are retained, forming a static structural base field under a unified coordinate framework.

[0083] The temporal feature extraction module 22 is used to perform multi-scale differential coding, texture perturbation tracking, and local deformation path analysis on multi-time period images based on the structural base field, constructing a temporal feature vector chain that can characterize the dynamic change pattern of the line structure. Specifically, it includes the following sub-modules:

[0084] The multi-scale differential feature encoding submodule 221 is used to calculate the differential response at different scales based on the differential relationship between the structural base field and images at different time periods, thereby obtaining the local variation characteristics of the line structure region. By performing small-scale and large-scale differential analysis on the conductor, fittings, and tower material regions respectively, this submodule can both amplify subtle disturbances and characterize the overall offset trend, and fuse the differential results at different scales into a set of differential features reflecting the amplitude of structural changes.

[0085] The texture perturbation and deformation path analysis submodule 222 is used to analyze the cross-frame texture energy response based on the texture perturbation tracking model, construct texture continuity curves, distinguish stable segments, abrupt change segments, and trend continuation regions, and thereby select stable perturbation paths that can continuously reflect the evolution mode of the line structure. Based on this, the submodule analyzes the displacement and directional changes of feature points in the structural region to obtain local deformation paths, and jointly models the differential features, texture perturbation features, and deformation path features to generate a feature vector Vt describing the dynamic state of the structure at time t, which is then arranged in chronological order to form a temporal feature vector chain.

[0086] The consistency evaluation module 23 is used to construct a time-series consistency evaluation diagram that reflects the evolution of the line structure over time, taking the time-series feature vector chain as input, and generating a time-series consistency index to measure the rationality of the evolution. Specifically, it includes the following sub-modules:

[0087] The structural change amplitude calculation submodule 231 is used to quantify the degree of structural change between adjacent time periods based on the temporal feature vector chain, form a structural change amplitude sequence, and construct an initial consistency evaluation map. This submodule comprehensively considers the changes in multi-scale features at different times, as well as indicators such as texture perturbation energy, texture continuity, and deformation propagation direction, and marks time periods with significantly increased change amplitudes to indicate areas where structural drift, early deformation, or abnormal change trends may exist.

[0088] The consistency index generation submodule 232 is used to introduce texture perturbation continuity and deformation propagation direction as dynamic weights into the evaluation process based on the initial consistency evaluation map, and to comprehensively evaluate the rationality of structural changes from a temporal perspective. This submodule generates a temporal consistency index sequence by fusing the deviations of structural change amplitude, texture continuity index, and deformation direction index within the time neighborhood. When the consistency index continuously decreases over a certain time period, this submodule marks it as a suspected anomaly interval, providing a key focus for subsequent latent space trajectory analysis.

[0089] The latent space anomaly identification module 24 is used to embed the structural evolution process into a low-dimensional latent space to form an evolution trajectory based on the temporal consistency index, and to identify anomalous deviations from the normal evolution trajectory under the constraints of latent space reachability and physical consistency, and to calculate the line anomaly probability score under multiple time series. Specifically, it includes the following sub-modules:

[0090] The latent space embedding submodule 241 performs nonlinear embedding on the temporal consistency index sequence, mapping the one-dimensional time series to a low-dimensional measurable space, so that the overall trend of the line structure changing over time is presented in the form of a continuous trajectory. This submodule describes the evolution pattern through trajectory morphology. When the line state is stable, the trajectory presents a smooth and uniform distribution in the latent space; when a sudden change occurs in a certain time period, the trajectory is reflected as a turning point, offset, or discrete jump point at the corresponding position, providing a spatial representation basis for anomaly detection.

[0091] The offset detection and anomaly scoring submodule 242 is used to detect instantaneous offset points deviating from the normal evolution trajectory on the latent space trajectory based on the distance changes between the trajectory points and their temporal neighborhoods, the reachability range, and the consistency with the physical evolution trend, and to calculate anomaly probability scores accordingly. This submodule assigns higher anomaly probability scores to trajectory segments with large offsets, long durations, and those that do not conform to the normal stress and environmental action patterns of the line, in order to reflect atypical disturbances such as sudden increases in local deformation, severe texture disturbances, or abrupt changes in the direction of suspension points under multi-temporal conditions.

[0092] The closed-loop inference module 25 is used to perform time-series-based closed-loop inference starting from the anomaly probability score. It eliminates false anomalies through reverse trajectory verification, identifies the actual structural instability path, and ultimately outputs the inspection conclusions of the transmission line. Specifically, it includes the following sub-modules:

[0093] The anomaly trajectory backtracking submodule 251 is used to perform reverse trajectory backtracking for time periods where the anomaly probability score exceeds a preset threshold. By examining the repeatability and continuity of anomaly points across multiple time periods, it determines whether they belong to real structural disturbances. This submodule comprehensively analyzes the changes in consistency index between adjacent time periods, the continuity of local texture disturbances, and the consistency of deformation propagation direction to identify anomaly points that can form continuous offset chains in time and correspond to the line topology in space, serving as candidate inputs for subsequent structural stability inference.

[0094] The instability path reasoning and inspection conclusion generation submodule 252 is used to eliminate pseudo-anomalies that fail repeatability verification and perform structural stability reasoning on the remaining abnormal paths. This submodule comprehensively analyzes dimensions such as spatial propagation direction, deviation amplitude changes, and the correlation between the abnormal path and key structural components such as the tower, conductors, and insulators to determine whether the anomaly may cause structural instability or operational risks. Based on this, the submodule generates a final inspection conclusion according to the severity and distribution range of the abnormal path, classifying the line status into levels such as normal, minor anomaly, significant deviation, or potential instability risk, providing a decision-making basis for maintenance personnel to formulate maintenance strategies.

[0095] Corresponding to the above embodiments, the present invention provides a computer storage medium, including: at least one memory and at least one processor;

[0096] The memory is used to store one or more program instructions;

[0097] A processor is used to run one or more program instructions to execute a transmission line inspection method based on multi-time-series image sequence analysis.

[0098] Corresponding to the above embodiments, this embodiment of the invention provides a computer-readable storage medium containing one or more program instructions, which are executed by a processor to provide a transmission line inspection method based on multi-time-series image sequence analysis.

[0099] The embodiments disclosed in this invention provide a computer-readable storage medium storing computer program instructions. When the computer program instructions are executed on a computer, the computer performs the above-described method for inspecting power transmission lines based on multi-time-series image sequence analysis.

[0100] In this embodiment of the invention, the processor can be an integrated circuit chip with signal processing capabilities. The processor can be a general-purpose processor, a digital signal processor (DSP), an application-specific integrated circuit (ASIC), a field-programmable gate array (FPGA), or other programmable logic devices, discrete gate or transistor logic devices, or discrete hardware components.

[0101] The various methods, steps, and logic diagrams disclosed in the embodiments of this invention can be implemented or executed. The general-purpose processor can be a microprocessor or any conventional processor. The steps of the methods disclosed in the embodiments of this invention can be directly implemented by a hardware decoding processor, or implemented by a combination of hardware and software modules in the decoding processor. The software modules can reside in random access memory, flash memory, read-only memory, programmable read-only memory, electrically erasable programmable memory, registers, or other mature storage media in the art. The processor reads information from the storage medium and, in conjunction with its hardware, completes the steps of the above methods.

[0102] The storage medium can be memory, such as volatile memory or non-volatile memory, or may include both volatile and non-volatile memory.

[0103] Among them, non-volatile memory can be read-only memory (ROM), programmable read-only memory (PROM), erasable programmable read-only memory (EPROM), electrically erasable programmable read-only memory (EEPROM), or flash memory.

[0104] Volatile memory can be random access memory (RAM), which is used as an external cache. By way of example, but not limitation, many forms of RAM are available, such as static random access memory (SRAM), dynamic random access memory (DRAM), synchronous dynamic random access memory (SDRAM), double data rate synchronous dynamic random access memory (DDRSDRAM), enhanced synchronous dynamic random access memory (ESDRAM), synchronous linked dynamic random access memory (Synchlink DRAM, SLDRAM), and direct memory bus RAM (DRRAM).

[0105] The storage media described in the embodiments of the present invention are intended to include, but are not limited to, these and any other suitable types of memory.

[0106] Those skilled in the art will recognize that, in one or more of the examples above, the functions described in this invention can be implemented using a combination of hardware and software. When applied as software, the corresponding functions can be stored in a computer-readable medium or transmitted as one or more instructions or code on a computer-readable medium. Computer-readable media include computer storage media and communication media, wherein communication media include any medium that facilitates the transmission of computer programs from one place to another. Storage media can be any available medium accessible to general-purpose or special-purpose computers.

[0107] The specific embodiments described above further illustrate the purpose, technical solution, and beneficial effects of the present invention. It should be understood that the above description is only a specific embodiment of the present invention and is not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc., made on the basis of the technical solution of the present invention should be included within the scope of protection of the present invention.

Claims

1. A method for inspecting transmission lines based on multi-time-series image sequence analysis, characterized in that, include: S10. Construct a structural scene basis model for multi-temporal images. By performing brightness drift adaptive calibration, background structure decomposition and stable feature extraction on cross-temporal image sequences, a basis field is generated to describe the stable structure of transmission lines. S20. Extract multi-level dynamic response features of temporal images on the base field, and construct a temporal feature vector chain through multi-scale differential coding, texture perturbation tracking and local deformation path analysis to characterize the dynamic change mode of the line structure. S30. Construct a temporal consistency evaluation graph based on the feature vector chain, deduce the evolution mode of the circuit structure through indicators such as structural change amplitude, texture perturbation continuity and deformation propagation direction, and generate a temporal consistency index that reflects the rationality of evolution. S40. Embed the consistency index and perform trajectory analysis in the latent space. By defining latent space reachability and physical consistency constraints, identify abnormal deviations from the normal evolution trajectory and obtain the line anomaly probability score under multiple time series. S50. Based on the anomaly probability score, perform closed-loop reasoning, verify and eliminate false anomalies through reverse trajectory, lock the real structural instability path, and finally output the inspection conclusion of the transmission line.

2. The transmission line inspection method based on multi-time-series image sequence analysis as described in claim 1, characterized in that, A structural scene basis model of multi-temporal images is constructed. By performing brightness drift adaptive calibration, background structure decomposition, and stability feature extraction on cross-temporal image sequences, a basis field is generated to describe the stable structure of transmission lines. The specific steps are as follows: Adaptive brightness calibration is performed on cross-time image sequences based on a local brightness estimation function, dynamically adjusting the illumination difference according to the brightness trend of adjacent frames to form an input sequence with consistent brightness; Background structure decomposition is performed on the brightness-calibrated image sequence. By separating the changing components of the scene, the main structure of the line that remains stable is extracted to generate a structural base field that describes the static morphology of the line.

3. The transmission line inspection method based on multi-time-series image sequence analysis as described in claim 1, characterized in that, Multi-level dynamic response features of temporal images are extracted from the base field. A temporal feature vector chain is constructed through multi-scale differential coding, texture perturbation tracking, and local deformation path analysis to characterize the dynamic change pattern of the line structure. The process is divided into the following sub-steps: Multi-scale difference features are calculated based on the difference relationship between the base field and the corresponding time period image to characterize the degree of local change of the line structure at different scales; Based on the texture perturbation tracking model, the stability and change trend of cross-frame textures are analyzed, and the dynamic change trajectory is derived by combining the local deformation path to construct a continuous temporal feature vector chain.

4. The transmission line inspection method based on multi-time-series image sequence analysis as described in claim 3, characterized in that, The stability and variation trend of textures across frames are analyzed using a texture perturbation tracking model, which is specifically divided into the following sub-steps: The local texture energy detection operator is used to extract cross-frame texture perturbation responses, and a texture continuity curve is constructed based on the perturbation response sequence. Texture perturbations are classified according to the steady and abrupt regions of the texture continuity curve, and stable perturbation paths that significantly contribute to structural dynamic changes are selected to provide input for deformation path analysis.

5. The transmission line inspection method based on multi-time-series image sequence analysis as described in claim 1, characterized in that, A temporal consistency evaluation graph is constructed based on the feature vector chain. The evolution mode of the circuit structure is deduced through indicators such as the magnitude of structural changes, the continuity of texture perturbation, and the direction of deformation propagation. A temporal consistency index reflecting the rationality of the evolution is generated. The process is divided into the following sub-steps: The structural change magnitude for each time period is calculated based on the feature vector chain, and an initial consistency evaluation map is constructed to identify potential structural drift regions. Texture perturbation continuity and deformation propagation direction are used as dynamic weights in the consistency evaluation graph to deduce the overall evolution trend of the structure and generate a temporal consistency index.

6. The transmission line inspection method based on multi-time-series image sequence analysis as described in claim 1, characterized in that, Embedding the consistency index and performing trajectory analysis in the latent space, and identifying anomalous deviations from the normal evolution trajectory by defining latent space reachability and physical consistency constraints, the line anomaly probability score under multiple time series is obtained. The specific steps are as follows: Latent space embedding is performed on the temporal consistency index to form an evolutionary trajectory in a low-dimensional measurable space to characterize the structural change pattern; Based on the degree of trajectory deviation detected by latent space accessibility and physical consistency constraints, instantaneous deviation points that do not conform to normal evolutionary patterns are identified, and anomaly probability scores are calculated.

7. The transmission line inspection method based on multi-time-series image sequence analysis as described in claim 1, characterized in that, Closed-loop inference is performed based on anomaly probability scoring. False anomalies are eliminated and the actual structural instability path is identified through reverse trajectory verification. Finally, the inspection conclusion of the transmission line is output. The process is divided into the following sub-steps: Based on the anomaly probability score, the reverse trajectory of the anomaly points is traced back, and the repeatability of the anomaly points in multiple time periods is checked to confirm whether they belong to real structural disturbances. False anomalies that fail the repeatability verification are removed, and structural stability inference is performed on the remaining anomaly paths to generate the final inspection conclusion.

8. A transmission line inspection system based on multi-time-series image sequence analysis, characterized in that, include: The base field construction module is used to perform cross-time period brightness drift adaptive calibration, background structure decomposition and stable feature extraction to construct a structural scene base model of the transmission line; The temporal feature extraction module is used to perform multi-scale differential coding, texture perturbation tracking and deformation path parsing on the basis field to generate a temporal feature vector chain. The consistency evaluation module is used to construct a time series consistency evaluation graph based on the time series feature vector chain and generate a time series consistency index. The latent space anomaly identification module is used to perform trajectory analysis on the consistency index based on latent space reachability and physical consistency constraints in order to calculate the anomaly probability score; The closed-loop inference module is used to perform reverse trajectory verification based on the anomaly probability score and lock the actual structural instability path to output the final inspection conclusion.

Images