Method and system for monitoring and early warning icing thickness of catenary dropper

Abstract

The invention relates to the technical field of thickness detection, and discloses a catenary dropper icing thickness monitoring and early warning method and system, and the method comprises the steps: obtaining a to-be-detected image of a catenary dropper, converting the to-be-detected image to an HSV and Lab color space, and generating a fusion edge feature map; segmenting a dropper icing area, and extracting a pixel width sample set; fitting the pixel width sample set to obtain initial icing thickness probability distribution, and extracting a measurement uncertainty factor; historical icing thickness prior distribution is constructed; determining a current measurement confidence coefficient, and calculating a fusion coefficient; performing conjugate updating on the initial icing thickness probability distribution and the historical icing thickness prior distribution to obtain icing thickness posterior probability distribution and distribution variance; acquiring a real-time environment wind speed, and calculating a safety thickness threshold value and an early warning probability threshold value; and when the cumulative probability of the icing thickness posterior probability distribution exceeding the safety thickness threshold is greater than the adjusted early warning probability threshold, outputting an early warning signal. According to the invention, the precision of icing thickness monitoring results is improved.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of thickness detection, and particularly relates to a catenary dropper icing thickness monitoring and early warning method and system. BACKGROUND

[0002] In winter and high-cold areas, power transmission cables, especially catenary droppers, are prone to icing. The continuous accumulation of ice significantly increases the mechanical load of the catenary dropper. Therefore, real-time and accurate monitoring and early warning of the icing thickness is one of the key technologies to ensure the safe operation of railways.

[0003] Current monitoring methods include manual inspection, sensor monitoring and visual monitoring. However, these monitoring methods have their own limitations. For example, manual inspection is low in efficiency, high in labor intensity and has safety risks. Although the sensor monitoring method can monitor at a fixed point, it is high in cost and difficult to achieve full-line coverage, and the sensor itself is also susceptible to adverse weather.

[0004] In addition, the visual monitoring method also faces the following challenges in application: first, the icing is often semi-transparent or transparent, with weak edge features, and the scene background is complex and the light is variable. The traditional gray scale gradient detection operator cannot accurately and completely extract the icing contour, resulting in low segmentation accuracy and affecting the accuracy of thickness calculation. Second, when encountering adverse weather such as rain, snow and thick fog or light changes, the decline in the quality of a single image will cause the measurement result to fluctuate sharply or even be wrong.

[0005] Finally, at the early warning decision-making level, the existing methods generally ignore the dynamic influence of environmental factors such as wind speed on the actual harm degree of icing, and do not consider the inherent uncertainty of the measurement process itself, resulting in insufficient scientificity of early warning decision-making and easy false positives or false negatives. SUMMARY

[0006] The present application provides a catenary dropper icing thickness monitoring and early warning method and system to solve the problem of low accuracy of icing thickness monitoring results and easy false positives or false negatives in the prior art.

[0007] In a first aspect, the catenary dropper icing thickness monitoring and early warning method of the present application comprises the following steps: An image to be measured of a catenary dropper is obtained, the image to be measured is converted to HSV and Lab color spaces, the saturation gradient of the S channel and the a *The chroma gradient of the channel is generated to generate a fusion edge feature map; an icing area of a suspender is segmented based on the fusion edge feature map, and a pixel width sample set is extracted along a radial direction of the icing area of the suspender; a bimodal Gaussian mixture model is used to fit the pixel width sample set to obtain an initial icing thickness probability distribution representing a uniform icing layer and an irregular ice ridge, and a variance of a Gaussian component corresponding to the uniform icing layer is extracted as a measurement uncertainty factor; a historical icing thickness prior distribution with an exponentially decaying weight over time is constructed; a current measurement confidence is determined by combining gradient information entropy of the to-be-measured image and the measurement uncertainty factor, and a fusion coefficient is calculated based on the current measurement confidence; a conjugate update is performed on the initial icing thickness probability distribution and the historical icing thickness prior distribution by using the fusion coefficient to obtain an icing thickness posterior probability distribution and a distribution variance; a real-time environmental wind speed is obtained, and a safe thickness threshold is calculated based on the real-time environmental wind speed; a warning probability threshold is calculated according to the distribution variance of the icing thickness posterior probability distribution, and the greater the distribution variance, the lower the warning probability threshold; when a cumulative probability of the icing thickness posterior probability distribution in a part exceeding the safe thickness threshold is greater than an adjusted warning probability threshold, a warning signal is output.

[0008] Preferably, the saturation gradient of the S channel image is calculated and weighted to fuse with the a * The chroma gradient of the channel is generated to generate a fusion edge feature map, including: The saturation gradient map of the S channel image is calculated The a * The chroma gradient map of the channel image is calculated ; The fusion edge feature map is calculated by the following formula : ; wherein, and are preset weight coefficients.

[0009] Preferably, the bimodal Gaussian mixture model is used to fit the pixel width sample set, including: The bimodal Gaussian mixture model is set as ; wherein, the first Gaussian component corresponds to the uniform icing layer, and the second Gaussian component corresponds to the irregular ice ridge; an expectation maximization (EM) algorithm is used to iteratively solve the model parameters , until the gain of the model log-likelihood function value is less than , and the iteration is stopped.

[0010] Preferably, the historical icing thickness prior distribution with an exponentially decaying weight over time is constructed, including: The icing thickness posterior probability distributions of the last N time points are stored; the weight of the posterior probability distribution at the t-k time point is Where k = 1, 2, ..., N, The preset attenuation coefficient; The weights of all posterior probability distributions are normalized; the posterior probability distributions of ice thickness at N time points are weighted and summed according to their corresponding normalized weights to generate the historical prior distribution of ice thickness at the current time.

[0011] Preferably, determining the current measurement confidence level by combining the gradient information entropy of the image to be measured with the measurement uncertainty factor includes: Calculate the gray-level histogram of the fused edge feature map, and calculate the gradient information entropy H based on the gray-level histogram; obtain the Gaussian component variance corresponding to the uniform ice layer. As a measurement uncertainty factor, the current measurement confidence level C is calculated using the following formula: ,in, This is the preset basic uncertainty term.

[0012] Preferably, the calculation of the safe thickness threshold based on real-time environmental wind speed includes: The safety thickness threshold is calculated using the following formula. : ; in, The preset baseline safety thickness threshold, The preset wind speed influence coefficient, Real-time ambient wind speed; The safety thickness threshold is the lower limit of the preset safety thickness threshold. When the calculated result is lower than the lower limit of the safety thickness threshold, the safety thickness threshold... Pick .

[0013] Preferably, the step of calculating the warning probability threshold based on the variance of the posterior probability distribution of icing thickness includes: The warning probability threshold is calculated using the following formula. : ; in, The preset baseline warning probability threshold, This is the preset variance adjustment factor. Let V be the variance of the posterior probability distribution of icing thickness; The lower limit of the preset warning probability threshold is set. When the calculated result is lower than the lower limit of the warning probability threshold, the warning probability threshold is adjusted. Pick .

[0014] Preferably, adopt The Sobel operator calculates the saturation gradient map of the S-channel image. With a * Chroma gradient map of channel image .

[0015] Preferably, acquiring the image to be measured of the overhead contact line dropper includes: An industrial camera is used to periodically photograph the overhead contact line droppers to obtain RGB format images of the droppers.

[0016] Secondly, the contact wire suspension cable icing thickness monitoring and early warning system of the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the above-mentioned contact wire suspension cable icing thickness monitoring and early warning method.

[0017] The beneficial effects of this invention are as follows: By fusing gradient information from multiple color spaces, this invention can extract the edge contours of translucent icing more completely, thus improving the accuracy of icing region segmentation. By inputting a bimodal Gaussian mixture model, it can not only distinguish between uniform icing and irregular ice ridges and provide a refined probability distribution representation of icing thickness, but also represent uncertainties in the measurement process. A temporal fusion update mechanism is established, combining historical data with the confidence level of the current measurement, smoothing out abrupt changes in measurement results caused by poor image quality in a single frame, and enhancing the stability and reliability of the monitoring system. In the early warning stage, real-time wind speed is combined with measurement uncertainties, making the early warning criteria more scientific and rigorous, reducing the risk of false alarms and missed alarms, and comprehensively improving the overall performance of overhead contact line icing monitoring and early warning. Attached Figure Description

[0018] Figure 1 A flowchart illustrating the method for monitoring and early warning of ice thickness on overhead contact line droppers provided in this embodiment of the invention. Detailed Implementation

[0019] Embodiments of the present invention are described in detail below, examples of which are illustrated in the accompanying drawings. The embodiments described below with reference to the accompanying drawings are exemplary and intended to explain the present invention, and should not be construed as limiting the present invention.

[0020] like Figure 1 As shown, an embodiment of the contact wire suspension cable icing thickness monitoring and early warning method provided by the present invention includes the following steps: S1. Obtain the image of the overhead contact line dropper, convert the image to HSV and Lab color spaces, and calculate and weightedly fuse the saturation gradient of the S channel with the a channel. * The chromaticity gradient of the channel is used to generate a fused edge feature map.

[0021] Specifically, industrial cameras deployed on inspection vehicles or along the lines periodically capture RGB images of the target image. Standard color space conversion algorithms are then used to convert the RGB images to both the HSV and Lab color space models. In the HSV space, the S-channel component image is extracted, and the saturation gradient map is calculated using operators such as the Sobel operator. In the Lab space, the a-channel component image is extracted. * The channel component map is used to calculate the chroma gradient map. Weighting coefficients are set, for example, 0.6 and 0.4. The saturation gradient map and the chroma gradient map are then summed at the pixel level to obtain a fused edge feature map that highlights the subtle edges of the semi-transparent icing.

[0022] In an optional embodiment, the calculated and weighted fusion of the saturation gradient of the S channel and a * The chromaticity gradient of the channels is used to generate a fused edge feature map, including: use The Sobel operator calculates the saturation gradient map of the S-channel image. With a * Chroma gradient map of channel image ; The fused edge feature map is calculated using the following formula. : ;in, and These are the preset weighting coefficients.

[0023] Specifically, the acquired RGB images were converted to HSV and CIELab color spaces, respectively, and the grayscale image of the saturation S channel and the chromaticity a channel were extracted from them. * The grayscale image has multiple channels. The S channel has high discrimination between snow and background, while the a channel... * The channel is not sensitive to color changes under different lighting conditions, and the combination of the two can enhance the icing edge and suppress noise.

[0024] Use one Sobel convolution kernels, respectively applied to the S-channel image and the a-channel image. * The channel image is convolved to calculate the gradient magnitude of each pixel, thereby generating a saturation gradient map. and chromaticity gradient map The two gradient maps are weighted and merged, with the weights of the saturation gradient map being... Set to 0.6, the weight of the chromaticity gradient map Set it to 0.4 to form an edge feature map that can clearly represent the outline of the ice.

[0025] S2. The icing region of the suspension wire is segmented based on the fused edge feature map, and the pixel width sample set is extracted radially along the icing region of the suspension wire. The pixel width sample set is fitted with a bimodal Gaussian mixture model to obtain the initial ice thickness probability distribution representing the uniform ice layer and irregular ice ridges, and the variance of the Gaussian component corresponding to the uniform ice layer is extracted as the measurement uncertainty factor.

[0026] Specifically, the fused edge feature map is binarized, morphological closing operations are performed to connect broken edges, and hole filling is performed to obtain a complete binary mask identifying the icy area of ​​the suspension cable. The skeleton of this area is extracted to obtain the centerline of the icy area. Each pixel on the centerline is traversed, and the width of the binary mask is measured along the normal direction of that point. All measured pixel width values ​​are collected to form a pixel width sample set. The pixel widths are converted to actual physical dimensions in mm using pre-calibrated camera parameters. An expectation-maximization algorithm is used, employing a mixture model with two Gaussian components to fit this sample set. In the two Gaussian distributions obtained after fitting, the component with the smaller mean and smaller variance is considered to represent a uniform icing layer, while the component with the larger mean and larger variance represents irregular ice ridges. The variance of the Gaussian component representing the uniform icing layer is taken as the measurement uncertainty factor for this measurement.

[0027] In an optional embodiment, fitting the pixel width sample set using a bimodal Gaussian mixture model includes: Set the bimodal Gaussian mixture model as Among them, the first Gaussian component For a uniform ice layer, the second Gaussian component For irregular icicles, the Expectation-Maximization (EM) algorithm is used to iteratively solve the model parameters. until the gain of the model's log-likelihood function is less than Stop iterating when the time comes.

[0028] Specifically, from pixel width samples containing two different icing patterns, the portion representing a uniform icing layer is precisely separated. A mixture model consisting of two Gaussian distributions is established. The first Gaussian component is assumed to represent the small, concentrated pixel width data generated by the edge of the uniform icing layer, with a mean of... This is the target thickness feature to be determined. The second Gaussian component is used to represent the large and scattered pixel width data generated by irregular ice spikes or noise.

[0029] The model is trained using the Expectation-Maximization (EM) algorithm. This algorithm iteratively updates the six model parameters (i.e., the weights of two components) by alternately executing the Expectation (E) step and the Maximization (M) step. and mean and and variance and The iterative process continues until the increase in the log-likelihood function value of the entire model with respect to the sample data between two consecutive iterations is less than one part per million, indicating that the model parameters have converged to a steady state. At this point, the mean of the first Gaussian component... Extracted as the most reliable estimate of the pixel width representing a uniform ice layer.

[0030] S3. Construct a prior distribution of historical icing thickness with weights decaying exponentially over time; determine the current measurement confidence by combining the gradient information entropy of the image to be measured and the measurement uncertainty factor; calculate the fusion coefficient based on the current measurement confidence; use the fusion coefficient to perform a conjugate update on the initial icing thickness probability distribution and the historical icing thickness prior distribution to obtain the posterior probability distribution and distribution variance of icing thickness.

[0031] Specifically, the prior distribution of historical icing thickness is modeled as a Gaussian distribution. The mean and variance are updated from the posterior distribution of the previous time step using a decay coefficient less than 1, ensuring that data from further back has less influence. The gradient information entropy of the current image to be measured is calculated; this value reflects the image's sharpness. This gradient information entropy is combined with the measurement uncertainty factor obtained in the previous step. For example, the information entropy is divided by the uncertainty factor and then normalized to obtain a current measurement confidence score between 0 and 1. This score is the fusion coefficient. The higher the confidence, the larger the fusion coefficient. The historical prior distribution and the initial icing thickness probability distribution represented by the Gaussian component of the uniform icing layer are weighted and fused according to the fusion coefficient. The new Gaussian distribution parameters, i.e., the mean and variance of the final posterior probability distribution of icing thickness, are calculated according to the Bayesian conjugate update rule.

[0032] In an optional embodiment, the construction of the historical icing thickness prior distribution with weights decaying exponentially over time includes: Store the posterior probability distribution of icing thickness for the most recent N time steps; the weight of the posterior probability distribution at time tk is... Where k = 1, 2, ..., N, The preset attenuation coefficient; The weights of all posterior probability distributions are normalized; the posterior probability distributions of ice thickness at N time points are weighted and summed according to their corresponding normalized weights to generate the historical prior distribution of ice thickness at the current time.

[0033] For example, to allow the estimation of ice thickness to account for trends over time, the posterior probability distribution of ice thickness at the past N time points is cached; for instance, N could be set to 20. Historical data constitutes a time series reflecting the growth or melting of ice. To integrate historical information, a time-weighted exponential decay strategy is employed. For a historical distribution k time units from the current time, the weights are... The weights are calculated using a formula, where the attenuation coefficient λ can be set to 0.2. This weighting formula assigns higher weights to more recent historical data and lower weights to more distant data. After calculating all weights from k=1 to N, they are normalized by dividing each weight by the sum of all weights, ensuring the new weight sum is 1. Multiplying these N posterior probability distributions by their corresponding normalized weights and then summing them yields the prior probability distribution for the current moment. This distribution integrates historical information and emphasizes recent trends.

[0034] In an optional embodiment, determining the current measurement confidence by combining the gradient information entropy of the image to be measured with the measurement uncertainty factor includes: Calculate the gray-level histogram of the fused edge feature map, and calculate the gradient information entropy H based on the gray-level histogram; obtain the Gaussian component variance corresponding to the uniform ice layer. As a measurement uncertainty factor, the current measurement confidence level C is calculated using the following formula: ,in, This is the preset basic uncertainty term.

[0035] The reliability of the current measurement is represented by evaluating image quality and the stability of the measurement results. For the generated fused edge feature map, a gray-level histogram is calculated, and the frequency of each gray level is statistically analyzed. Based on the frequency distribution, the gradient information entropy H is calculated using the Shannon entropy formula from information theory. A high information entropy value generally indicates clear image edges, rich details, and good image quality.

[0036] From the fitting results of the bimodal Gaussian mixture model, extract the variance of the Gaussian component representing the uniform ice layer. It reflects the dispersion of pixel width samples of a uniform ice layer. Variance is used as a measurement uncertainty factor; the larger the variance, the greater the fluctuation of the measured value and the higher the uncertainty. A small constant is input as the basic uncertainty term. For example, 0.2 is used to avoid a denominator of zero. The measurement confidence level C is calculated using a formula. This measurement confidence level C will be used in subsequent Bayesian updates to adjust the influence of the current measurement data.

[0037] S4. Obtain the real-time ambient wind speed and calculate the safe thickness threshold based on the real-time ambient wind speed; calculate the warning probability threshold according to the distribution variance of the posterior probability distribution of the icing thickness, where the larger the distribution variance, the lower the warning probability threshold; when the cumulative probability of the posterior probability distribution of the icing thickness in the part exceeding the safe thickness threshold is greater than the adjusted warning probability threshold, output a warning signal.

[0038] Specifically, real-time wind speed data at the location of the overhead contact line droppers is obtained by connecting to the meteorological monitoring system along the railway line. A safe icing thickness model is established, for example, by subtracting an adjustment amount proportional to the wind speed from a basic safe icing thickness value, thus obtaining a safe icing thickness threshold that varies with wind speed; the higher the wind speed, the lower the permissible safe icing thickness. A functional relationship is established between the warning probability threshold and the posterior distribution variance, for example, by multiplying a basic probability threshold by an exponential decay function that decreases as the variance increases, so that the more uncertain the measurement result, i.e., the larger the variance, the more lenient the probability condition required to trigger a warning. The probability of the icing thickness exceeding the safe icing thickness threshold is calculated using the Gaussian cumulative distribution function of the final icing thickness posterior probability distribution. This calculated cumulative probability is compared with the adjusted warning probability threshold; if the former is greater than the latter, an icing over-limit warning signal is issued to the monitoring center.

[0039] In an optional embodiment, the calculation of the safe thickness threshold based on real-time environmental wind speed includes: calculating the safe thickness threshold using the following formula. : ; in, The preset baseline safety thickness threshold, The preset wind speed influence coefficient, Real-time ambient wind speed; The safety thickness threshold is the lower limit of the preset safety thickness threshold. When the calculated result is lower than the lower limit of the safety thickness threshold, the safety thickness threshold... Pick .

[0040] Specifically, in order to adapt the icing warning threshold to changes in the external environment, a baseline safe thickness threshold is set under windless or light wind conditions. For example, 25mm. A wind speed influence coefficient is determined based on the structure's characteristics and safety requirements. For example, 0.4 mm / (m / s) represents the amount by which the allowable thickness should be reduced for every 1 m / s increase in wind speed. The current ambient wind speed is obtained in real time via sensors. The unit is meters per second. A linear formula is used to calculate the safe thickness threshold. For example, when the wind speed is 15 m / s, the calculated threshold is 19 mm. A lower limit for the safe thickness threshold is set. For example, 12mm. If calculated according to the formula... For thicknesses less than 12mm, 12mm is used as the safe thickness threshold.

[0041] In an optional embodiment, calculating the warning probability threshold based on the variance of the posterior probability distribution of the icing thickness includes: calculating the warning probability threshold using the following formula. : ; in, The preset baseline warning probability threshold, This is the preset variance adjustment factor. Let V be the variance of the posterior probability distribution of icing thickness; The lower limit of the preset warning probability threshold is set. When the calculated result is lower than the lower limit of the warning probability threshold, the warning probability threshold is adjusted. Pick .

[0042] For example, setting A value of 0.95 indicates that an alarm should be issued when the probability of exceeding the icing thickness limit reaches 95%. (Setting) The value is set to 0.3 to adjust for the influence of uncertainty on the threshold. After each update, the variance of the posterior probability distribution of the icing thickness is calculated. The variance of this distribution represents the uncertainty in thickness estimation: the larger the variance, the higher the uncertainty. When the measurement is very accurate and the variance is small, the warning probability threshold will remain high, close to 0.95; however, when the measurement uncertainty increases and the variance becomes larger, the warning probability threshold will automatically decrease, making the system more likely to trigger warnings under less certain conditions, reflecting the principle of conservative decision-making. To prevent the warning probability threshold from being too low and causing frequent false alarms, a lower limit for the warning probability threshold is set. For example, 0.75. When the calculated result is lower than this value, then... As a threshold for the probability of early warning.

[0043] The implementation principle of the contact wire suspension cable icing thickness monitoring and early warning method of this invention is as follows: Firstly, by fusing gradient information from multiple color spaces, this invention can more completely extract the edge contours of translucent icing, significantly improving the segmentation accuracy of the icing area. Secondly, this invention uses a bimodal Gaussian mixture model to fit the icing thickness data, which not only effectively distinguishes between uniform icing layers and irregular ice ridges, achieving a refined probability distribution description of icing thickness, but also quantifies the uncertainty in the measurement process. Thirdly, this invention establishes a time-series fusion update mechanism that combines historical data with the confidence level of the current measurement, effectively smoothing abrupt changes in measurement results caused by poor single-frame image quality, significantly enhancing the stability and reliability of the monitoring system. Finally, in the early warning stage, this invention innovatively incorporates real-time wind speed and measurement uncertainty into consideration simultaneously, making the early warning criteria more scientific and rigorous, thereby effectively reducing the risk of false alarms and missed alarms, and comprehensively improving the overall performance of contact wire icing monitoring and early warning.

[0044] An embodiment of the contact wire suspension cable icing thickness monitoring and early warning system provided by the present invention includes a memory and a processor. The memory stores computer instructions, and when the processor executes the computer instructions, it implements the contact wire suspension cable icing thickness monitoring and early warning method in the above embodiment.

[0045] Although embodiments of the present invention have been shown and described above, it is understood that the above embodiments are exemplary and should not be construed as limiting the present invention. Those skilled in the art can make changes, modifications, substitutions and variations to the above embodiments within the scope of the present invention.

Claims

1. A catenary dropper icing thickness monitoring and early warning method, characterized in that, The method comprises the following steps: acquiring a to-be-tested image of a catenary dropper, converting the to-be-tested image to HSV and Lab color spaces, calculating and weightedly fusing a saturation gradient of an S channel and a chroma gradient of an a * channel, generating a fused edge feature map; segmenting an icing area of the dropper based on the fused edge feature map, extracting a pixel width sample set along a radial direction of the icing area of the dropper; fitting the pixel width sample set by using a double-peak Gaussian mixture model to obtain an initial icing thickness probability distribution representing a uniform icing layer and an irregular ice ridge, and extracting a Gaussian component variance corresponding to the uniform icing layer as a measurement uncertainty factor; constructing a historical icing thickness prior distribution with an exponentially decaying weight over time; determining a current measurement confidence by combining gradient information entropy of the to-be-tested image and the measurement uncertainty factor, and calculating a fusion coefficient based on the current measurement confidence; performing a conjugate update on the initial icing thickness probability distribution and the historical icing thickness prior distribution by using the fusion coefficient to obtain an icing thickness posterior probability distribution and a distribution variance; acquiring a real-time environmental wind speed, calculating a safety thickness threshold based on the real-time environmental wind speed; calculating a warning probability threshold according to the distribution variance of the icing thickness posterior probability distribution, wherein the larger the distribution variance is, the lower the warning probability threshold is; when a cumulative probability of the icing thickness posterior probability distribution that exceeds the safety thickness threshold is greater than an adjusted warning probability threshold, outputting a warning signal.

2. The catenary dropper ice thickness monitoring and early warning method according to claim 1, characterized in that, The calculated and weighted fusion of the saturation gradient of the S channel and a * The chromaticity gradient of the channels is used to generate a fused edge feature map, including: Computing a saturation gradient map of an s channel image with a * Computing a chroma gradient map of a channel image ; The fused edge feature map is calculated by the following formula : ; wherein, and are preset weight coefficients.

3. The catenary dropper ice thickness monitoring and early warning method of claim 1, wherein, The fitting of the bimodal Gaussian mixture model to the pixel width sample set comprises: The bimodal Gaussian mixture model is set as ; wherein the first Gaussian component corresponds to a uniform ice-coating layer, and the second Gaussian component corresponds to an irregular ice ridge; and the model parameters are solved iteratively by using an expectation maximization (EM) algorithm , until the gain of the model log-likelihood function value is less than , and the iteration is stopped.

4. The catenary dropper ice thickness monitoring and early warning method of claim 1, wherein, The constructing of the history icing thickness prior distribution with exponentially decaying weights over time comprises: Store the icing thickness posterior probability distribution of the last N time points; the weight of the posterior probability distribution of the t-k time point is wherein k=1, 2,..., N, is a preset attenuation coefficient; The weights of all posterior probability distributions are normalized, and the icing thickness posterior probability distributions at N time points are weighted and summed according to the corresponding normalized weights to generate the history icing thickness prior distribution at the current time.

5. The catenary dropper ice thickness monitoring and early warning method of claim 1, wherein, The current measurement confidence is determined by combining the gradient information entropy of the to-be-measured image and the measurement uncertainty factor. A gray scale histogram of the fused edge feature map is calculated, and gradient information entropy H is calculated based on the gray scale histogram; a Gaussian component variance corresponding to the uniform ice cover is obtained as a measurement uncertainty factor; calculate the current measurement confidence C by the following formula: wherein, is a preset base uncertainty term.

6. The catenary dropper ice thickness monitoring and early warning method of claim 1, wherein, The safety thickness threshold is calculated based on the real-time environmental wind speed. The safety thickness threshold is calculated by the following equation : ； wherein, is a preset reference safety thickness threshold, is a preset wind speed influence coefficient, is a real-time environmental wind speed; a lower limit of a preset safety thickness threshold, when the calculation result is lower than the lower limit of the safety thickness threshold, the safety thickness threshold take .

7. The catenary dropper ice thickness monitoring and early warning method of claim 1, wherein, The warning probability threshold is calculated according to the distribution variance of the icing thickness posterior probability distribution. The early warning probability threshold is calculated by the following formula : ； wherein, is a preset reference early warning probability threshold, is a preset variance adjustment coefficient, is a distribution variance of the icing thickness posterior probability distribution; a lower limit of a preset warning probability threshold, when the calculation result is lower than the lower limit of the warning probability threshold, the warning probability threshold take .

8. The catenary dropper ice thickness monitoring and early warning method of claim 2, wherein, Sobel operator of saturated gradient map of the S channel image is computed with a * chrominance gradient map of the channel image .

9. The catenary dropper ice thickness monitoring and early warning method of claim 1, wherein, The to-be-measured image of the catenary dropper is acquired. An industrial camera is used to regularly capture the catenary dropper to acquire the to-be-measured image of the catenary dropper in RGB format.

10. A catenary dropper icing thickness monitoring and early warning system, characterized in that, The method comprises a memory and a processor, the memory stores computer instructions, and the processor executes the computer instructions to realize the catenary dropper icing thickness monitoring and warning method in any one of claims 1-9.

Images