CN-121982387-A - Cloud edge cooperation-based hidden danger image detection method, system and medium for power transmission line

CN121982387ACN 121982387 ACN121982387 ACN 121982387ACN-121982387-A

Abstract

The invention relates to the technical field of image detection and discloses a cloud edge cooperation-based power transmission line hidden danger image detection method, a cloud edge cooperation-based power transmission line hidden danger image detection system and a cloud edge cooperation-based power transmission line hidden danger image detection medium. According to the cloud edge cooperative architecture and physical priori fusion mechanism, the exogenous heat entering and endogenous self-heating signals of the power transmission line target area are separated, the precision and the robustness of hidden danger detection are greatly improved, the problem of false detection and omission caused by sunlight interference is avoided, and the operation and maintenance efficiency and the power grid safety guarantee capability are improved.

Inventors

XU DUNBIN
ZHANG XIAO
GU HAO
LI CHENXI
XU ZHIPENG
HAO ZHAN
SU LINGDONG
DUAN QINGQUAN
ZHENG CHENGHAO
TAN YUYING
LI MEN
CHANG YUANYUAN
ZHOU FENG

Assignees

国网江苏省电力有限公司徐州供电分公司

Dates

Publication Date: 20260505
Application Date: 20260119

Claims (10)

1. The cloud edge cooperation-based hidden danger image detection method for the power transmission line is characterized by comprising the following steps of: Step S101, acquiring an interested region in an infrared image at the edge end, synchronously acquiring brightness data of the interested region in a visible light image, and determining a sunlight dip moment and a corresponding shading amplitude according to time change of the brightness data; step S102, acquiring a temperature sequence of the region of interest, and extracting an inert residual coefficient and a temperature residual sequence of the region of interest in an observation window according to the time of solar dip; step S103, constructing a residual temperature body space-time tensor by utilizing the infrared image and the temperature residual sequence, extracting the temperature attenuation rate from the temperature residual sequence, and carrying out priori modulation on the residual temperature body space-time tensor by utilizing the shading amplitude, the inertia residual coefficient and the temperature attenuation rate; step S104, extracting the modulated residual temperature body space-time tensor by utilizing a characteristic extraction network, and outputting a residual ratio estimator and an edge risk score for the region of interest; step S105, curve fitting is carried out on the temperature sequence at the cloud, the estimated value of the residual ratio is used as an initial parameter of fitting, the back-calculated inert residual coefficient is obtained by combining the shading amplitude, and the back-calculated result and the edge risk score are subjected to logarithmic probability fusion in the same decision channel, so that the final probability is obtained; and S106, comparing the final probability with a judgment threshold value, and outputting hidden danger candidate marks of the region of interest.
2. The cloud edge collaboration-based power transmission line hidden danger image detection method according to claim 1, wherein a target area of a wire clamp or a splicing sleeve is positioned in an infrared image, and an area with the same pixel coordinate position as the infrared image in a visible light image is synchronously acquired; Extracting brightness components of all pixel points in a target area, and calculating arithmetic average values of the brightness components of all pixel points in the target area to obtain a brightness average value sequence; Selecting a sampling point with the largest negative change value in the brightness change rate, and determining the time of solar shock when the brightness average falling ratio of the previous time and the next preset time of the sampling point is not lower than a preset amplitude threshold; and calculating the difference between the brightness average value at the moment before the daylighting dip moment and the brightness average value at the fixed offset moment after the daylighting dip moment, and dividing the difference by the brightness average value at the moment before the daylighting dip moment to obtain the shading amplitude.
3. The cloud edge collaboration-based hidden danger image detection method for the power transmission line, according to claim 1, is characterized in that gray values of all pixel points in an interested area in an infrared image are mapped to apparent temperature values, and arithmetic average values of the apparent temperature values of all the pixel points in the interested area are calculated to obtain a temperature time sequence; in the temperature time sequence, extracting a temperature value corresponding to a position two seconds after the time of solar shock, and determining the temperature value as a base line temperature; Calculating a first temperature difference between a temperature value at a sixty-second position after the solar dip time and a base line temperature in the temperature time sequence, calculating a second temperature difference between the temperature value at a sampling time before the solar dip time and the base line temperature in the temperature time sequence, and dividing the first temperature difference by the second temperature difference to obtain an inertia residual coefficient; and in an observation window from two seconds after the daylighting dip time to sixty seconds after the daylighting dip time, calculating the difference value between the temperature value corresponding to the sampling time and the baseline temperature according to each discrete sampling time, and arranging each difference value in time sequence to obtain a temperature residual sequence.
4. The cloud edge collaboration-based power transmission line hidden danger image detection method according to claim 1, wherein a pixel matrix of a plurality of discrete sampling moments after a solar dip moment in an infrared image is extracted, the pixel matrix of a base line moment is subtracted from the pixel matrix of each sampling moment to obtain pixel-level temperature difference distribution corresponding to each moment, and the pixel-level temperature difference distribution is stacked according to a time dimension to obtain a residual temperature body space-time tensor; Obtaining a first residual value of a base line moment and a second residual value of a preset time length position from a temperature residual sequence, calculating a ratio of the second residual value to the first residual value, performing natural logarithm operation on the ratio, dividing an operation result by a negative number of a duration value between the base line moment and the preset time length position, and obtaining a temperature attenuation rate; integrating the shading amplitude, the inertia residual coefficient and the temperature attenuation rate, generating a priori parameter set, mapping the priori parameter set to generate a channel scaling vector and a channel translation vector, multiplying the numerical value of each channel in the residual temperature volume space-time tensor by the corresponding component in the channel scaling vector, and adding the corresponding component in the channel translation vector to obtain the modulated residual temperature volume space-time tensor.
5. The cloud edge collaboration-based power transmission line hidden danger image detection method according to claim 1, wherein a three-dimensional convolution encoder is used for carrying out convolution operation of space dimension and time dimension on the modulated residual temperature body space-time tensor, and an intermediate characteristic tensor is obtained through extraction; performing linear weighted calculation on the prior parameter set, and inputting a calculation result into a logistic regression function to obtain a gating coefficient in a scalar form; scaling multiplication processing is carried out on the gating coefficient in a scalar form and each channel in the intermediate characteristic tensor, so that the gating characteristic tensor is obtained; global average pooling processing is carried out on the gating feature tensor, and average compression is carried out on data of the gating feature tensor in the space dimension and the time dimension to obtain a one-dimensional feature vector; And performing linear weighting calculation on the one-dimensional feature vector, inputting the calculation result into a non-negative linear rectification function to output a residual ratio estimation amount, and inputting the calculation result into a logistic regression function to output an edge risk score.
6. The cloud edge collaboration-based power transmission line hidden danger image detection method according to claim 1 is characterized in that least square fitting based on an exponential decay model is performed on a temperature time sequence in a time interval from two seconds after a solar dip time to sixty seconds after the solar dip time, wherein natural logarithm operation is performed on a residual ratio estimator, a length value of the time interval is divided by a negative value of an operation result to obtain a time constant initial value, a temperature value of two seconds after the solar dip time is subtracted from a temperature value of a sampling time before the solar dip time to obtain an amplitude initial value, the temperature value of two seconds after the solar dip time is determined to be a base line initial value, and a time constant fitting value, an amplitude fitting value and a base line fitting value are obtained through iterative calculation.
7. The cloud edge collaboration-based power transmission line hidden danger image detection method according to claim 6 is characterized by comprising the steps of subtracting a baseline fitting value from a temperature value of a temperature time sequence sixty seconds after a sunlight dip time to obtain a third temperature difference, subtracting the baseline fitting value from a temperature value of the temperature time sequence at a sampling time before the sunlight dip time to obtain a fourth temperature difference, dividing the third temperature difference by the fourth temperature difference to obtain a preliminary residual ratio, dividing the preliminary residual ratio by a shading amplitude to obtain a recalculated inert residual coefficient; Vector mapping processing is carried out on the calculated inert residual coefficient and the time constant fitting value, linear weighting calculation is carried out on the mapped result to obtain the cloud logarithm probability, the logarithm score value of the edge risk score is calculated, the cloud logarithm probability and the logarithm score value of the edge risk score are added to obtain the fusion logarithm probability, and the fusion logarithm probability is input into a logistic regression function to obtain the final probability.
8. The cloud edge collaboration-based power transmission line hidden danger image detection method according to claim 1, wherein a numerical difference between a final probability and a preset judgment threshold is calculated; Setting the hidden danger candidate mark as a first preset value when the value difference is greater than or equal to zero, setting the hidden danger candidate mark as a second preset value when the value difference is less than zero, and outputting the set hidden danger candidate mark as a detection result of the region of interest.
9. The cloud edge cooperation-based power transmission line hidden danger image detection system, characterized in that the cloud edge cooperation-based power transmission line hidden danger image detection method according to any one of claims 1 to 8 is executed, comprising: The first module is used for acquiring an interested region in the infrared image at the edge end, synchronously acquiring brightness data of the interested region in the visible light image, and determining a sunlight dip moment and a corresponding shading amplitude according to time change of the brightness data; The second module is used for acquiring a temperature sequence of the region of interest and extracting an inert residual coefficient and a temperature residual sequence of the region of interest in the observation window according to the sunlight dip time; The third module is used for constructing a residual temperature body space-time tensor by utilizing the infrared image and the temperature residual error sequence, extracting the temperature attenuation rate from the temperature residual error sequence and carrying out priori modulation on the residual temperature body space-time tensor by utilizing the shading amplitude, the inertia residual coefficient and the temperature attenuation rate; A fourth module, which is used for extracting the modulated residual temperature body space-time tensor by utilizing the characteristic extraction network and outputting the residual ratio estimation quantity and the edge risk score aiming at the region of interest; A fifth module, performing curve fitting on the temperature sequence at the cloud, using the estimated residual ratio as an initial parameter of fitting, combining the shading amplitude to obtain a recalculated inert residual coefficient, and performing logarithmic probability fusion on the recalculated result and the edge risk score in the same decision channel to obtain a final probability; and a sixth module for comparing the final probability with the judgment threshold value and outputting hidden danger candidate marks of the region of interest.
10. A computer readable medium having stored thereon a computer program, wherein the program, when executed by a processor, performs the cloud edge synergy-based power transmission line hidden danger image detection method of any one of claims 1 to 8.

Description

Cloud edge cooperation-based hidden danger image detection method, system and medium for power transmission line Technical Field The present invention relates to the field of image detection technology, and more particularly, the method, the system and the medium for detecting hidden danger images of the power transmission line based on cloud edge cooperation are provided. Background The key connection parts of the wire clamp, the connecting tube and the like of the power transmission line serving as a core infrastructure of the power system are easy to cause the problems of increased contact resistance, aging and corrosion, loosening of bolts and the like due to long-term operation, so that endogenous self-heating hidden danger is generated. If the hidden trouble is not found in time, the continuous temperature rise causes serious faults such as line fusing, interphase short circuit and the like, and the safe and stable operation of the power grid is directly threatened. The infrared image detection becomes a mainstream technology of power transmission line hidden trouble inspection by virtue of high sensitivity to temperature change, and is widely applied to scenes such as unmanned aerial vehicle inspection, manual handheld equipment detection and the like. However, in an outdoor inspection environment, the temperature of a target area is extremely easily disturbed by external heat input such as sunlight. Under natural environment, conditions such as rapid shielding of cloud layer, sudden change of sun angle and the like can cause sudden decrease of sunlight, so that the temperature of a target area greatly fluctuates in a short time, and an exogenous heat-entering temperature signal and an endogenous self-heating hidden danger signal are mutually overlapped and confused. The existing detection method lacks a signal separation mechanism based on physical characteristics, simply depends on an image gray level threshold value or simple temperature statistics, cannot accurately distinguish two types of temperature signals, is difficult to extract the real characteristics of intrinsic self-heating, causes frequent false detection of hidden danger judgment, misjudges sunlight fluctuation as hidden danger, or leaks detection of real hidden danger covered by exogenous heat. Meanwhile, the traditional detection mode has obvious cloud edge cooperative defects. When the method relies on cloud processing, temperature sequence fitting lacks effective initial parameter support, is often trapped into local optimum or slow in convergence due to random initialization, and brightness and temperature dynamic data acquired by an edge in real time are not fully fused, so that the detection efficiency and the detection precision are difficult to meet the requirement of large-scale inspection. Disclosure of Invention The invention provides a cloud edge cooperation-based hidden danger image detection method, a system and a medium for a power transmission line, solving the technical problems in the background technology. The invention provides a cloud edge cooperation-based hidden danger image detection method for a power transmission line, which comprises the following steps of: Step S101, acquiring an interested region in an infrared image at the edge end, synchronously acquiring brightness data of the interested region in a visible light image, and determining a sunlight dip moment and a corresponding shading amplitude according to time change of the brightness data; step S102, acquiring a temperature sequence of the region of interest, and extracting an inert residual coefficient and a temperature residual sequence of the region of interest in an observation window according to the time of solar dip; step S103, constructing a residual temperature body space-time tensor by utilizing the infrared image and the temperature residual sequence, extracting the temperature attenuation rate from the temperature residual sequence, and carrying out priori modulation on the residual temperature body space-time tensor by utilizing the shading amplitude, the inertia residual coefficient and the temperature attenuation rate; step S104, extracting the modulated residual temperature body space-time tensor by utilizing a characteristic extraction network, and outputting a residual ratio estimator and an edge risk score for the region of interest; step S105, curve fitting is carried out on the temperature sequence at the cloud, the estimated value of the residual ratio is used as an initial parameter of fitting, the back-calculated inert residual coefficient is obtained by combining the shading amplitude, and the back-calculated result and the edge risk score are subjected to logarithmic probability fusion in the same decision channel, so that the final probability is obtained; and S106, comparing the final probability with a judgment threshold value, and outputting hidden danger candidate marks of the region of interest. The invention provide