Method and System for Corrosion Assessment of Transmission Line Tower Angle Steel Based on Multi-Layer Segmentation

Abstract

The present invention belongs to the technical field of power engineering monitoring, and discloses a method and system for evaluating the corrosion of angle steels of transmission line towers based on multi-layer segmentation. The method includes performing instance segmentation on the images of transmission line towers to obtain each angle steel instance; performing semantic segmentation on a single angle steel to obtain the corrosion areas on the corresponding angle steel; calculating the corrosion proportion of each angle steel; and performing weighted averaging on the corrosion proportions of all angle steels to obtain the overall corrosion proportion of the transmission line tower, thereby evaluating the overall corrosion degree of the transmission line tower. The present invention can effectively improve the recognition accuracy of angle steels and corrosion, which is beneficial to improving the efficiency of inspection personnel in detecting the corrosion condition of transmission line towers, reducing the incidence rate of safety accidents such as tower collapse caused by corrosion, avoiding certain economic losses, having a certain driving force for the update of the power grid inspection plan, and ensuring the safe and stable operation of transmission line towers.

Description

Technical Field

[0001] The present invention belongs to the technical field of power engineering monitoring, relates to the evaluation of angle steels of transmission line towers, and particularly relates to a method and system for evaluating the corrosion of angle steels of transmission line towers based on multi-layer segmentation. Background Art

[0002] With the acceleration of the pace of power grid construction, the scale of power is continuously expanding, and more and more transmission line towers are put into construction and use. And the safe and stable operation of the power grid is the basic guarantee for people's production and life. Therefore, higher requirements are put forward for the corrosion detection of angle steels of transmission line towers. Most transmission line towers face a harsh atmospheric corrosion environment and must be subjected to timely and effective anti-corrosion maintenance. Otherwise, they may be suddenly damaged due to corrosion, which will seriously affect the safe operation of the transmission line and even cause major safety accidents such as tower collapse and wire breakage. Therefore, formulating an anti-corrosion maintenance plan for transmission line towers, regularly conducting a comprehensive detection of the corrosion failure situation of transmission line towers, and taking effective protective measures according to the detection results are of great significance for ensuring the safe operation of transmission lines and the stable operation of the power grid.

[0003] When the angle steel of a transmission line tower is corroded, the angle steel will show an obvious color change, which is darker red than the non-corroded angle steel. In the images of transmission line towers taken by drones, the angle steels will overlap with each other due to the change of the shooting angle, and the light is too dark, which brings difficulties to the corrosion detection of angle steels, and this also leads to a low accuracy rate of traditional corrosion detection methods for angle steels of transmission line towers. Summary of the Invention

[0004] The purpose of the present invention aims to overcome the problems such as low accuracy rate and efficiency of traditional transmission line tower detection methods, and provides a method for evaluating the corrosion of angle steels of transmission line towers based on multi-layer segmentation, so as to realize the efficient identification of each angle steel on the transmission line tower and evaluate the corrosion degree.

[0005] Another purpose of the present invention lies in providing a system for evaluating the corrosion of angle steels of transmission line towers based on multi-layer segmentation.

[0006] In order to achieve the above purpose, the purpose of the present invention can be realized by the following technical solutions.

[0007] The method for evaluating the corrosion of angle steels of transmission line towers based on multi-layer segmentation provided by the present invention includes the following steps:

[0008] S1 Obtain an image of a transmission line tower;

[0009] S2 Perform instance segmentation on the image of the transmission line tower to obtain each angle steel instance; this step includes the following sub-steps:

[0010] S21 Extract features from the image of the transmission line tower;

[0011] S22 Generates a sample image of the angle steel of the transmission line tower scanned by the network using the extracted feature map, determines whether the image belongs to the foreground or the background, and simultaneously obtains an accurate candidate region;

[0012] S23 Maps the generated candidate region onto the feature map to obtain a regional feature map;

[0013] S24 Performs pooling processing on the regional feature map, and determines the accurate position of the boundary polygon box of the angle steel target through the regional regression method to complete the instance segmentation of the angle steel;

[0014] S3 Performs semantic segmentation on a single angle steel to obtain the corrosion area on the corresponding angle steel;

[0015] S4 Calculates the corrosion proportion of each angle steel;

[0016] S5 Performs weighted averaging on the corrosion proportions of all angle steels to obtain the overall corrosion proportion of the transmission line tower, and further evaluates the overall corrosion degree of the transmission line tower.

[0017] In the above step S2, instance segmentation is performed on the transmission line tower image to obtain each angle steel instance. For the specific operation, see steps S21 - S24.

[0018] In the above step S21, feature extraction is performed on the transmission line tower image through a convolutional neural network, such as the FCN network.

[0019] In the above step S22, the feature map is processed through the RPN (Region Proposal Network) network. First, a sliding window process is performed on the feature map to obtain a series of anchor points; based on a series of anchor points, the softmax function is used to determine whether the anchor points belong to the foreground or the background, and at the same time, the positions of the anchor points are corrected through regional regression to obtain accurate candidate regions, that is, regions that may contain the angle steel target.

[0020] In the above step S23, the method of using the RoI Align layer to replace RoIPooling is adopted. The candidate region is mapped into the feature map obtained in step S21 in a pixel - by - pixel alignment manner to obtain a regional feature map of a unified size; in order to obtain a feature map of a fixed size, the "bilinear interpolation" algorithm is used for calculation for floating - point numbers, which can avoid quantization errors. Bilinear interpolation is an image scaling algorithm that fully utilizes the four real - existing pixel values around the virtual points in the original image to jointly determine a pixel value in the target image, that is, it can estimate the pixel value corresponding to a virtual position point such as 10.56. The formula for calculating any point (x, y) between two points (x1, y1) and (x2, y2) by bilinear interpolation is as follows:

[0021]

[0022] In the above step S24, the uniformly sized regional feature map obtained after the above mapping is averaged and pooled into a one-dimensional vector, and the exact position of the target boundary polygon box of the angle steel is obtained through regional regression calculation, that is, the pixel point of the image with the highest probability of the angle steel category, so as to realize the instance segmentation of the angle steel and obtain several single-angle steel images.

[0023] In the above step S3, semantic segmentation is performed on each single-angle steel to obtain the corrosion area on the corresponding angle steel. The specific operations are shown in steps S31-32:

[0024] S31 Use the encoder to extract features to obtain the semantic segmentation feature map;

[0025] S32 Use the decoder to decode the semantic segmentation features to determine the category of each pixel point, and realize the semantic segmentation of the corrosion area on the angle steel.

[0026] In the above step S31, for the input image of the angle steel of the transmission line tower, the encoder programs the background, angle steel, and corrosion in the image into an intermediate expression form, that is, performs feature extraction to obtain the semantic segmentation feature map. The encoder includes a convolutional layer, a normalization layer, an activation function, and a max-pooling layer arranged in sequence; the calculation process of the encoder is as follows: first, perform convolution calculation with unchanged size to obtain the convolution sum; then perform batch normalization calculation to obtain the normalized convolution sum; use the ReLU activation function to activate the normalized convolution sum, and finally use max-pooling to calculate the activated convolution sum and record the index position of the maximum value in each layer of convolution sum, programming the background, angle steel, and corrosion areas in the input image into an intermediate expression form, that is, gradually reducing the feature map and extracting high-level semantic information to obtain the semantic segmentation feature map.

[0027] In the above step S32, use the decoder to perform decoding operations on the obtained semantic segmentation feature map to obtain the probabilities of each pixel point in the input image of the transmission line tower angle steel in the three categories of background, angle steel, and corrosion. The category with the highest probability is the category to which the pixel point belongs, and the semantic segmentation of the corrosion area on the angle steel is realized. The decoder includes an upsampler, a convolutional layer, a normalization layer, and an activation function; the calculation process of the decoder is as follows: first, upsample the input semantic segmentation feature map by a factor of two, that is, fill 0 outside the feature map matrix to obtain a feature map enlarged by a factor of two; then, according to the index position after the max-pooling of the encoder, put the data of the input feature map into the blocks filled with zeros, and the other positions remain zero; then perform convolution calculation with unchanged size to extract features; then perform batch normalization calculation and use the ReLU function to activate the result of the batch normalization calculation, and output the category probability of each pixel point.

[0028] In the above step S4, the calculation formula for the corrosion proportion T of a single angle steel is: T = divide the number of pixel points in the corroded area (i.e., the corroded area) on the image of the single angle steel by the number of pixel points in the image of the angle steel (the area of the angle steel), and use it as the corrosion proportion of each angle steel. The corrosion degree of each angle steel can be evaluated through this corrosion proportion.

[0029] In the above step S5, the calculation method for weighted calculating the corrosion proportion P of the entire tower is as follows: Select the top five angle steels with the largest corrosion proportion as the independent variables for weighted calculation (T1, T2, T3, T4, and T5 are the corrosion proportions of the angle steels with the first, second, third, fourth, and fifth largest corrosion proportions respectively). Correspondingly, the weights are set as W1 = 0.4, W2 = 0.2, W3 = 0.2, W4 = 0.1, and W5 = 0.1 respectively. The calculation formula is as follows:

[0030] P = W1T1 + W2T2 + W3T3 + W4T4 + W5T5

[0031] The value range of the calculation result of P must be [0, 100%]. The corrosion degree of the entire tower can be judged through this corrosion proportion.

[0032] The evaluation criteria for the corrosion degree of the angle steels of transmission line towers can be designed based on past experience. The relevant standards within power grid enterprises include DL / T 1424 - 2015 "Grid Metal Technology Supervision Regulations", DL / T 1453 - 2015 "Anti - corrosion Protection Coating for Transmission Line Towers", DL / T 2055 - 2019 "Guidelines for Safety Assessment of Corrosion of Transmission Line Steel Structures", etc.

[0033] Based on the corrosion degree grading standard of transmission line towers, the corrosion grades of the angle steels of transmission line towers are specifically divided into 6 levels, including Grade A: Micro - corrosion (corroded area is 0, galvanized layer is bluish - gray or silver - white), Grade B: Weak - corrosion (corroded area is 0, galvanized layer is dark - gray or gray - black), Grade C: Light - corrosion (0 < corroded area < 3%), Grade D: Medium - corrosion (3% ≤ corroded area < 10%), Grade E: Heavy - corrosion (10% ≤ corroded area < 33%), Grade F: Extremely - heavy - corrosion (corroded area ≥ 33%).

[0034] In the above step S6, for the six levels, the corresponding evaluation status and operation and maintenance suggestions are as follows:

[0035] The evaluation status of Grade A is "Normal", and the operation and maintenance suggestion is "No treatment is required and it can continue to be used";

[0036] The evaluation status of Grade B is "Normal", and the operation and maintenance suggestion is "No treatment is required for the time being, and it can be paid attention to during inspections";

[0037] The C-level evaluation status is "average", and the operation and maintenance suggestions are as follows: "It should be monitored during use, and extra attention is needed during inspections; try to arrange the overhaul plan for anti-corrosion coating construction within 3 years, and if conditions permit, anti-corrosion can be carried out in advance.";

[0038] The D-level evaluation status is "abnormal", and the operation and maintenance suggestions are as follows: "Key attention is needed during inspections; try to arrange the overhaul plan for anti-corrosion coating construction within 2 years, and if conditions permit, anti-corrosion can be carried out in advance; the next corrosion assessment cycle is shortened to 1 year.";

[0039] The E-level evaluation status is "abnormal", and the operation and maintenance suggestions are as follows: "Further corrosion measurement should be carried out to evaluate corrosion safety, and mechanical tests can be carried out by sampling if conditions permit; the overhaul plan for anti-corrosion coating construction should be arranged within 1 year; the corrosion assessment cycle is shortened to 1 year.";

[0040] The F-level evaluation status is "severe", and the operation and maintenance suggestions are as follows: "Further corrosion measurement should be carried out immediately to evaluate whether replacement is needed. Replace as soon as the replacement conditions are met, and anti-corrosion coating construction should be arranged immediately even if the replacement conditions are not met; before the renovation is completed, the corrosion assessment cycle is shortened to less than half a year.";

[0041] The present invention further provides a corrosion detection system for angle steel of transmission line towers based on multi-layer segmentation, which is used to implement the above-mentioned corrosion assessment method for angle steel of transmission line towers. It includes an image acquisition device, a single angle steel acquisition device, a single angle steel corrosion area acquisition device, a single angle steel corrosion degree assessment device, and an overall corrosion degree assessment device for transmission line towers;

[0042] The image acquisition device is used to acquire images of transmission line towers;

[0043] The single angle steel acquisition device is used to perform instance segmentation on the image of the transmission line tower to obtain each angle steel instance; it includes the following units:

[0044] The first feature extraction unit is used to extract features from the image of the transmission line tower;

[0045] The candidate region acquisition unit uses the extracted feature map to generate a sample image of the angle steel of the transmission line tower scanned by the network, determines whether the image belongs to the foreground or the background, and simultaneously obtains accurate candidate regions;

[0046] The region feature map acquisition unit is used to map the generated candidate region to the feature map to obtain the region feature map;

[0047] The angle steel acquisition unit is used to perform pooling processing on the region feature map and determine the accurate position of the polygon frame of the angle steel target boundary through the region regression method to complete the instance segmentation of the angle steel;

[0048] The single-angle steel corrosion area acquisition device is used for semantic segmentation of a single-angle steel to obtain the corrosion area on the corresponding angle steel;

[0049] The single-angle steel corrosion degree evaluation device is used to calculate the corrosion ratio of each angle steel;

[0050] The overall corrosion degree evaluation device of the transmission line tower is used to perform weighted averaging on the corrosion ratios of all angle steels to obtain the overall corrosion ratio of the transmission line tower, and then evaluate the overall corrosion degree of the transmission line tower.

[0051] The above image acquisition device can be set as a shooting drone. After shooting the distant view image of the transmission line tower, it can further perform preliminary analysis on the collected image; if it is found that there are angle steels with corrosion phenomena in the image, the drone can fly closer to collect the close-up images of these angle steels to obtain the transmission line tower angle steel images that can be used for corrosion analysis of the transmission line tower angle steel. In addition to the transmission line tower angle steel images obtained, the image acquisition device can simultaneously obtain the pole tower number, the line to which the pole tower belongs, and the pole tower name corresponding to the transmission line tower angle steel image.

[0052] The above single-angle steel corrosion area acquisition device includes an encoder and a decoder, and its specific explanation is as described above.

[0053] The transmission line tower angle steel corrosion evaluation method and system based on multi-layer segmentation provided by the present invention have the following beneficial effects:

[0054] (1) The present invention uses the instance segmentation method to identify and segment the angle steel from the background, and then uses the semantic segmentation algorithm to identify and segment the corrosion area on each angle steel from the angle steel; based on the specific positions and the number of pixel points of the obtained corrosion area and the angle steel, calculate the corrosion ratio of each angle steel and calculate the corrosion ratio of the whole tower by weighting, and then give the corrosion evaluation status, corrosion grade and operation and maintenance suggestions accordingly;

[0055] (2) The present invention can effectively improve the recognition accuracy of angle steel and corrosion, which is beneficial to improving the efficiency of inspection personnel in detecting the corrosion condition of transmission line towers, reducing the incidence of safety accidents such as tower collapse caused by corrosion, avoiding certain economic losses, having a certain driving force for the update of the power grid inspection plan, and ensuring the safe and stable operation of transmission line towers;

[0056] (3) The application scenario of the present invention is relatively wide, and it can be extended to the corrosion evaluation applications of towers composed of steel components of China Tower, transportation, railway and other enterprises and steel bridges. Description of the Drawings

[0057] Figure 1 It is a schematic flow chart of the transmission line tower angle steel corrosion evaluation method based on multi-layer segmentation;

[0058] Figure 2 Schematic diagram of the instance segmentation process for transmission line tower images;

[0059] Figure 3 Schematic diagram of the semantic segmentation process for a single angle steel;

[0060] Figure 4 Original image of the transmission line tower;

[0061] Figure 5 Instance segmentation image of the transmission line tower;

[0062] Figure 6 Semantic segmentation image of the angle steel of the transmission line tower. Specific implementation mode

[0063] It is intended to clearly and completely describe the technical solutions of the embodiments of the present invention in conjunction with the accompanying drawings. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative work belong to the present invention.

[0064] Embodiment 1

[0065] The method for evaluating the corrosion of the angle steel of the transmission line tower based on multi-layer segmentation provided by the present invention, as Figures 1 - 3 shown, includes the following steps:

[0066] S1 Obtain the image of the transmission line tower.

[0067] In this embodiment, a typical transmission line tower composed of angle steel is taken as the target.

[0068] In this embodiment, a drone camera is used as the image acquisition source to obtain the image of the transmission line tower. This image acquisition method can capture the tower images at various distances and angles, which helps to improve the accuracy of corrosion evaluation in the subsequent steps.

[0069] S2 Perform instance segmentation on the image of the transmission line tower to obtain each angle steel instance; this step includes the following sub-steps:

[0070] S21 Extract features from the image of the transmission line tower.

[0071] In this step, a convolutional neural network is used to extract features from the transmission line tower image, such as the FCN network. See Dai J, He K, Li Y, et al. Instance-sensitive fully convolutional networks[C] / / European Conference on Computer Vision. Springer, Cham, 2016:534-549.

[0072] S22 uses the extracted feature map to generate a sample image of the angle steel of the transmission line tower scanned by the network, determines whether the image belongs to the foreground or the background, and simultaneously obtains accurate candidate regions.

[0073] In this embodiment, the RPN (Region Proposal Network) network is used to process the extracted feature map. The RPN network structure includes a 3*3 convolutional layer (padding number is 1) and two parallel 1*1 convolutional layers (no padding). The RPN network first performs a sliding window process on the feature map. For a sliding window, 256-dimensional features are obtained through the 3*3 convolutional layer. According to fixed sizes and ratios, a set of anchor points for this sliding window is obtained (denoted by k as the number of boxes generated by the anchor points). Then, through two fully connected layers, 2k class probabilities and 4k anchor point position coordinates are obtained. Then, the softmax function is used to determine whether the anchor point belongs to the foreground or the background. At the same time, the anchor point position is corrected through region regression to obtain accurate candidate regions, that is, regions that may contain the angle steel target. The specific implementation method of region regression is to learn a regression model to predict the small correction amount of each anchor point relative to its initial position. This correction amount is used to adjust the position of the anchor point to make it closer to the position of the real target. The Mean Squared Error is used as the loss function, and the regression target is set to minimize the distance between the corrected anchor point and the real target position. Then, optimization methods such as gradient descent are used, and the regression model is trained through the backpropagation algorithm to enable it to predict the best position correction amount. Each transmission line tower angle steel image can generate 384 candidate regions.

[0074] S23 maps the generated candidate regions to the feature map to obtain a region feature map.

[0075] A method using the RoI Align layer to replace the RoI Pooling. In the way of pixel-by-pixel alignment, the candidate regions are mapped into the feature map obtained in step S21 to obtain a region feature map of a unified size. In order to obtain a feature map of a fixed size, the "bilinear interpolation" algorithm is used for calculation of floating-point numbers, which can avoid quantization errors. Bilinear interpolation is an image scaling algorithm that fully utilizes the four real pixel values around the virtual points in the original image to jointly determine a pixel value in the target image, that is, it can estimate the pixel values corresponding to virtual position points such as 10.56. The formula for calculating any point (x, y) between two points (x1, y1) and (x2, y2) by bilinear interpolation is as follows:

[0076]

[0077] S24 performs pooling processing on the region feature map, and determines the precise position of the boundary polygon box of the angle steel target through the region regression method, completing the instance segmentation of the angle steel.

[0078] The unified-size region feature map obtained after the above mapping is averaged and pooled into a one-dimensional vector, and then the precise position of the boundary polygon box of the angle steel target is obtained through region regression calculation, that is, the pixel point of the image with the highest probability of the angle steel category, realizing the instance segmentation of the angle steel and obtaining several single-angle steel images.

[0079] S3 performs semantic segmentation on the single-angle steel to obtain the corrosion area on the corresponding angle steel. This step includes the following sub-steps:

[0080] S31 uses an encoder for feature extraction to obtain a semantic segmentation feature map.

[0081] For the input transmission line tower angle steel image, the encoder programs the background, angle steel, and corrosion in the image into an intermediate expression form, that is, performs feature extraction to obtain a semantic segmentation feature map. The encoder includes a 3*3 convolutional layer (padding number is 1), a normalization layer, a ReLU activation function, and a max pooling layer (stride is 2, convolutional kernel size is 2*2, no padding) set in sequence.

[0082] The calculation process of the encoder is as follows: The single-angle steel image obtained by instance segmentation is input into the encoder for feature extraction. First, the convolution sum is obtained through convolution calculation with unchanged size; then the batch normalization calculation is performed to obtain the normalized convolution sum; the ReLU activation function is used to activate the normalized convolution sum, and finally the max pooling calculation is used to calculate the activated convolution sum and record the index position of the maximum value in each layer of convolution sum, programming the background, angle steel, and corrosion areas in the input image into an intermediate expression form, that is, gradually reducing the feature map and extracting high-level semantic information to obtain a semantic segmentation feature map.

[0083] S32 Decodes the semantic segmentation features using a decoder to determine the category of each pixel point, realizing the semantic segmentation of the corrosion area on the angle steel.

[0084] The decoder performs a decoding operation on the obtained semantic segmentation feature map to obtain the probabilities of each pixel point in the transmission line tower angle steel image in the three categories of background, angle steel, and corrosion. The category with the highest probability is the category to which the pixel point belongs, realizing the semantic segmentation of the corrosion area on the angle steel.

[0085] The decoder includes an upsampler, a 3*3 convolutional layer (without padding), a normalization layer, and a ReLU activation function.

[0086] The calculation process of the decoder is as follows: First, the input semantic segmentation feature map is magnified by two times through upsampling, that is, 0 is filled outside the feature map matrix to obtain a feature map magnified by two times; then, according to the index position after the encoder's max pooling, the data of the input feature map is placed into the blocks filled with zeros, and the other positions remain zero; then, features are extracted through convolution calculations with unchanged dimensions; then, batch normalization calculations are performed and the results of the batch normalization calculations are activated using the ReLU function to output the category probabilities of each pixel point, that is, decoding the intermediate representation form into the output, gradually restoring the spatial information of the image and classifying each pixel to realize the segmentation of the corrosion area on the angle steel.

[0087] S4 Calculates the corrosion ratio of each angle steel.

[0088] The calculation formula for the corrosion ratio T of a single angle steel is: T = dividing the number of pixel points (i.e., the corrosion area) of the corrosion area on the single angle steel image by the number of pixel points (the area of the angle steel) of the angle steel image, which is used as the corrosion ratio of each angle steel. The corrosion degree of each angle steel can be evaluated through this corrosion ratio.

[0089] S5 Performs a weighted average on the corrosion ratios of all angle steels to obtain the overall corrosion ratio of the transmission line tower, and further evaluates the overall corrosion degree of the transmission line tower.

[0090] The calculation method for the weighted calculation of the overall tower corrosion ratio P is as follows: Select the top five angle steels with the highest corrosion ratios as the independent variables for weighted calculation (T1, T2, T3, T4, and T5 are the corrosion ratios of the angle steels with the first, second, third, fourth, and fifth highest corrosion ratios respectively). Correspondingly, the weights are set as W1 = 0.4, W2 = 0.2, W3 = 0.2, W4 = 0.1, and W5 = 0.1. The calculation formula is as follows:

[0091] P = W1T1 + W2T2 + W3T3 + W4T4 + W5T5

[0092] The value range of the calculation result of P must be [0, 100%]. The overall corrosion degree of the tower can be judged through this corrosion ratio.

[0093] The evaluation criteria for the corrosion degree of angle steel in transmission line towers can be designed based on past experience. Relevant standards within power grid enterprises include DL / T 1424-2015 "Technical Supervision Regulations for Grid Metals", DL / T 1453-2015 "Anti-corrosion Protection Coating for Transmission Line Towers", DL / T 2055-2019 "Guidelines for Safety Assessment of Corrosion of Steel Structures in Transmission Lines", etc.

[0094] Based on the corrosion degree grading standard of transmission line towers, the corrosion grades of angle steel in transmission line towers are specifically divided into 6 levels, including Level A: Micro-corrosion (corrosion area is 0, galvanized layer is bluish-gray or silver-white), Level B: Weak corrosion (corrosion area is 0, galvanized layer is dark gray or gray-black), Level C: Light corrosion (0 < corrosion area < 3%), Level D: Medium corrosion (3% ≤ corrosion area < 10%), Level E: Severe corrosion (10% ≤ corrosion area < 33%), Level F: Extremely severe corrosion (corrosion area ≥ 33%).

[0095] In the above step S6, for the six levels, the corresponding evaluation status and operation and maintenance suggestions are as follows:

[0096] The evaluation status of Level A is "Normal", and the operation and maintenance suggestion is "No treatment is required, and it can continue to be used".

[0097] The evaluation status of Level B is "Normal", and the operation and maintenance suggestion is "No treatment is required for the time being, and it can be paid attention to during inspections".

[0098] The evaluation status of Level C is "General", and the operation and maintenance suggestion is "It should be monitored during use and extra attention is required during inspections; try to arrange an overhaul plan for anti-corrosion coating construction within 3 years, and if conditions permit, anti-corrosion can be carried out in advance".

[0099] The evaluation status of Level D is "Abnormal", and the operation and maintenance suggestion is "Key attention is required during inspections; try to arrange an overhaul plan for anti-corrosion coating construction within 2 years, and if conditions permit, anti-corrosion can be carried out in advance; the next corrosion assessment cycle is shortened to 1 year".

[0100] The evaluation status of Level E is "Abnormal", and the operation and maintenance suggestion is "Further corrosion measurement should be carried out to evaluate corrosion safety, and sampling for mechanical tests can be carried out if conditions permit; an overhaul plan for anti-corrosion coating construction should be arranged within 1 year; the corrosion assessment cycle is shortened to 1 year".

[0101] The evaluation status of Level F is "Severe", and the operation and maintenance suggestion is "Further corrosion measurement should be carried out immediately to evaluate whether replacement is needed. Replace as soon as possible if the replacement conditions are met, and anti-corrosion coating construction should be arranged immediately if the replacement conditions are not met; before the renovation is completed, the corrosion assessment cycle is shortened to less than half a year".

[0102] For example,Figure 4 An original image of a transmission line tower is given. Through the above-mentioned step S2, instance segmentation is performed on the image, and the obtained result is as Figure 5 shown. Through the above-mentioned step S3, semantic segmentation is performed on some angle steels to obtain the corresponding corrosion areas, such as Figure 6 shown.

[0103] According to step S4, for the Figure 6 angle steel from the lower left to the upper right in the middle, the calculated corrosion ratio is: (613256 / 645120)*100% = 95.1%. Then, the corrosion degree of each angle steel is evaluated through the corrosion ratio, that is, extremely severe corrosion of grade F, and the operation and maintenance suggestion is given as "further corrosion measurement should be carried out immediately to evaluate whether replacement is needed. Replace as soon as the replacement conditions are met, and anti-corrosion coating construction should also be arranged immediately for those that do not meet the replacement conditions; before the renovation is completed, the corrosion assessment period should be shortened to within half a year".

[0104] According to step S5, the weighted average of the corrosion ratios of all angle steels is calculated, and the formula is as follows:

[0105] P = 0.4*0.951 + 0.2*0.942 + 0.2*0.923 + 0.1*0.871 + 0.1*0.653 = 0.906 = 90.6%.

[0106] Through this corrosion ratio, it can be judged that the corrosion degree of the whole tower is extremely severe corrosion of grade F, and the operation and maintenance suggestion is given as "further corrosion measurement should be carried out immediately to evaluate whether replacement is needed. Replace as soon as the replacement conditions are met, and anti-corrosion coating construction should also be arranged immediately for those that do not meet the replacement conditions; before the renovation is completed, the corrosion assessment period should be shortened to within half a year".

[0107] Embodiment 2

[0108] This embodiment provides a corrosion detection system for angle steels of a transmission line tower based on multi-layer segmentation, which is used to implement the above-mentioned corrosion assessment method for angle steels of a transmission line tower. It includes an image acquisition device, a single-angle steel acquisition device, a single-angle steel corrosion area acquisition device, a single-angle steel corrosion degree assessment device, and an overall corrosion degree assessment device for the transmission line tower.

[0109] The image acquisition device is used to acquire the image of the transmission line tower.

[0110] The image acquisition device can be set to photograph a drone. After taking a distant view image of a transmission line tower, it can further perform a preliminary analysis on the acquired image. If angle steels with corrosion are found in the image, the drone can fly closer to collect close-up images of these angle steels, obtaining transmission line tower angle steel images that can be used for corrosion analysis of the angle steels of the transmission line tower. In addition to the acquired transmission line tower angle steel images, the image acquisition device can simultaneously obtain the tower number, the line to which the tower belongs, and the tower name corresponding to the transmission line tower angle steel image.

[0111] The single angle steel acquisition device is used to perform instance segmentation on the transmission line tower image to obtain each angle steel instance. It includes the following units:

[0112] The first feature extraction unit is used to extract features from the transmission line tower image. The first feature extraction unit uses the FCN network given in Embodiment 1.

[0113] The candidate region acquisition unit uses the extracted feature map to generate a sample image of the transmission line tower angle steel scanned by the network, determines whether the image belongs to the foreground or the background, and simultaneously obtains accurate candidate regions. The candidate region acquisition unit uses the RPN network given in Embodiment 1.

[0114] The region feature map acquisition unit is used to map the generated candidate regions to the feature map to obtain a region feature map. The region feature map acquisition unit uses the RoI Align layer given in Embodiment 1.

[0115] The angle steel acquisition unit is used to perform pooling processing on the region feature map and determine the accurate position of the angle steel target boundary polygon frame through a region regression method to complete the instance segmentation of the angle steel. The angle steel acquisition unit includes an average pooling layer and an angle steel target boundary polygon frame acquisition subunit; the angle steel target boundary polygon frame acquisition subunit obtains the accurate position of the angle steel target boundary polygon frame through region regression calculation, that is, the pixel point of the image with the highest probability of being the angle steel category.

[0116] The single angle steel corrosion region acquisition device is used to perform semantic segmentation on a single angle steel to obtain the corrosion region on the corresponding angle steel. The single angle steel corrosion region acquisition device includes the encoder and decoder given in Embodiment 1.

[0117] The single angle steel corrosion degree evaluation device is used to calculate the corrosion ratio of each angle steel; it is calculated according to the method given in Embodiment 1.

[0118] The transmission line tower overall corrosion degree evaluation device is used to perform a weighted average on the corrosion ratios of all angle steels to obtain the overall corrosion ratio of the transmission line tower, and then evaluate the overall corrosion degree of the transmission line tower; it is determined according to the method given in Embodiment 1.

[0119] The network models in the above single-angle steel acquisition device and single-angle steel corrosion area acquisition device are trained through the historical data of transmission line towers. The two can be trained separately or together. In this embodiment, the single-angle steel acquisition device and the single-angle steel corrosion area acquisition device are trained respectively through the training dataset. The loss function used in the training is the pixel-level cross-entropy loss (Pixel-wise Cross-Entropy Loss), and the stochastic gradient descent algorithm with momentum (SGD with Momentum) is used to optimize the network parameters of the device.

[0120] To implement the training of the network model in the above single-angle steel acquisition device, it is necessary to perform target annotation on each angle steel in the transmission line tower sample image dataset, add a target boundary polygon box and an angle steel category label to obtain the annotated tower angle steel sample image, and construct a transmission line tower angle steel sample training dataset. The specific production method includes:

[0121] (1) After installing labelme through the pip command, use the labelme software to open the folder storing the enhanced dataset of transmission line tower angle steel images. Set automatic saving before annotation, and the generated json file will also be saved in the same folder as the picture. Click to create a polygon box to start annotation. Note that the image cannot be compressed or its format changed first because there will be precise calculation requirements for the pixel points of the image in subsequent processing; after drawing the polygon box, the labels of the angle steel and the corrosion area should be marked in time. After the above steps of annotation, the annotated transmission line tower angle steel image can be obtained.

[0122] (2) Perform data enhancement operations on the annotated transmission line tower angle steel image through geometric transformations such as mirror inversion, translation by 30 pixel points, reduction to 0.8 times and magnification to 1.25 times, and clockwise and counterclockwise rotation by 15 degrees to construct an enhanced dataset of transmission line tower angle steel images, which is beneficial to improving the accuracy of the subsequent feature map.

[0123] Those of ordinary skill in the art will realize that the embodiments described herein are for helping the reader understand the principles of the present invention, and it should be understood that the protection scope of the present invention is not limited to such specific statements and embodiments. Those of ordinary skill in the art can make various other specific deformations and combinations that do not depart from the essence of the present invention based on the technical revelations disclosed in the present invention, and these deformations and combinations are still within the protection scope of the present invention.

Claims

1. A method for evaluating the corrosion of angle steel of transmission line towers based on multi-layer segmentation, characterized in that, Including the following steps: S1 Obtain the image of the transmission line tower; S2 Perform instance segmentation on the image of the transmission line tower to obtain each angle steel instance; This step includes the following sub-steps: S21 Extract features from the image of the transmission line tower; S22 Use the extracted feature map to generate a sample image of the angle steel of the transmission line tower scanned by the network, determine whether the image belongs to the foreground or the background, and at the same time obtain an accurate candidate region; S23 Map the generated candidate region to the feature map to obtain a region feature map; S24 Perform pooling processing on the region feature map, and determine the accurate position of the polygon frame of the angle steel target boundary through the region regression method to complete the instance segmentation of the angle steel; In a specific implementation manner, the region feature map obtained after mapping is averaged and pooled into a one-dimensional vector, and the accurate position of the polygon frame of the angle steel target boundary is obtained through region regression calculation, that is, the pixel point of the image is the pixel point with the highest probability of the angle steel category, to achieve the instance segmentation of the angle steel and obtain several single-angle steel images; S3 Perform semantic segmentation on a single angle steel to obtain the corrosion area on the corresponding angle steel; Step S3 includes the following sub-steps: S31 Use an encoder to extract features to obtain a semantic segmentation feature map; S32 Use a decoder to decode the semantic segmentation features to determine the category of each pixel point, and achieve semantic segmentation of the corrosion area on the angle steel; S4 Calculate the corrosion ratio of each angle steel; The calculation formula for the corrosion ratio T of a single angle steel is: T = Divide the number of pixel points of the corrosion area on the single-angle steel image by the number of pixel points of the angle steel image as the corrosion ratio of each angle steel; S5 weights and averages the corrosion percentages of all angle steels to obtain the overall corrosion percentage of the transmission line tower, and then evaluates the overall corrosion degree of the transmission line tower; the calculation method for weighted calculation of the overall tower corrosion percentage P is as follows: take the top five angle steels with the highest corrosion percentages as the independent variables for weighted calculation, and and and are the corrosion percentages of the angle steels with the first, second, third, fourth, and fifth highest corrosion percentages respectively. Correspondingly, the weights are set as and and and respectively. The calculation formula is as follows: .

2. The method for evaluating the corrosion of angle steel of transmission line towers based on multi-layer segmentation according to claim 1, characterized in that, In step S21, extract features from the image of the transmission line tower through a convolutional neural network.

3. The method for evaluating the corrosion of angle steel of transmission line towers based on multi-layer segmentation according to claim 1, characterized in that, In step S22, use the RPN network to process the extracted feature map. First, perform a sliding window process on the feature map to obtain a series of anchor points; Based on a series of anchor points, use the softmax function to judge whether the anchor point belongs to the foreground or the background, and at the same time correct the position of the anchor point through region regression to obtain an accurate candidate region.

4. The method for evaluating the corrosion of angle steel of transmission line towers based on multi-layer segmentation according to claim 1, characterized in that, In step S23, use the RoI Align layer to map the candidate region to the feature map obtained in step S21.

5. The method for evaluating the corrosion of angle steel of transmission line towers based on multi-layer segmentation according to claim 4, characterized in that, In order to obtain a feature map of a fixed size, bilinear interpolation algorithm is used for calculation of floating-point numbers.

6. A corrosion detection system for angle steel of transmission line towers based on multi-layer segmentation using the method according to claim 1, characterized in that, Including an image acquisition device, a single-angle steel acquisition device, a single-angle steel corrosion area acquisition device, a single-angle steel corrosion degree evaluation device, and a transmission line tower overall corrosion degree evaluation device; The image acquisition device is used to obtain the image of the transmission line tower; The single-angle steel acquisition device is used to perform instance segmentation on the image of the transmission line tower to obtain each angle steel instance; It Includes the following units: The first feature extraction unit is used to extract features from the image of the transmission line tower; The candidate region acquisition unit uses the extracted feature map to generate a sample image of the angle steel of the transmission line tower scanned by the network, determine whether the image belongs to the foreground or the background, and at the same time obtain an accurate candidate region; The region feature map acquisition unit is used to map the generated candidate region to the feature map to obtain a region feature map; The angle steel acquisition unit is used to perform pooling processing on the regional feature map, and determine the precise position of the target boundary polygon frame of the angle steel through the regional regression method, completing the instance segmentation of the angle steel; The single-angle steel corrosion area acquisition device is used to perform semantic segmentation on a single angle steel to obtain the corrosion area on the corresponding angle steel; The single-angle steel corrosion degree evaluation device is used to calculate the corrosion ratio of each angle steel; The transmission line tower overall corrosion degree evaluation device is used to perform weighted averaging on the corrosion ratios of all angle steels to obtain the overall corrosion ratio of the transmission line tower, and further evaluate the overall corrosion degree of the transmission line tower.

Images