Interference determination method of external damage potential point based on artificial intelligence

Abstract

In order to solve the problem of inaccurate identification of external damage hidden danger points and easy false detection and missed detection in the prior art, the present invention provides an interference determination method for external damage hidden danger points based on artificial intelligence, and its technical solution to solve the technical problem includes the following steps: 1. Image input, collecting drones and monitoring photos; 2. Feature extraction, using the pre-trained feature matrix to calculate the collected pictures to obtain feature results; 3. Feature encoding, encoding the feature results in the second; 4. Result classification, after the encoding is completed, it is compared with the samples in the knowledge base to complete the classification; 5. Output results, according to the classification category, decode and output the corresponding warning information. The present invention analyzes and processes the images or monitoring images taken during the drone inspection process through the method of artificial intelligence, realizes the accurate determination of hidden danger points and realizes the grading and evaluation, and provides a more reliable basis for the subsequent investigation and maintenance of hidden danger points.

Description

Technical Field

[0001] The present invention belongs to the technical field of identifying external failure potential points of power transmission lines, and in particular to an interference determination method for external failure potential points based on artificial intelligence. Background Art

[0002] In the current era, the safety of transmission lines is a necessary condition to ensure industrial and domestic electricity consumption. During the actual operation of transmission lines, they are easily damaged by external forces and cause accidents. The hidden dangers of external damage are mainly reflected in the following aspects: 1. Theft, 2. Prevention problems, 3. Caused by illegal buildings.

[0003] With the development of society, automatic patrol of drones is often used for inspection of power transmission lines. This operation method greatly improves the inspection speed and efficiency. However, drone inspections are not accurate in identifying external damage potential points and are prone to missed inspections. Therefore, how to improve the accuracy of identifying external damage potential points during the inspection process has always been an issue that needs attention. Summary of the invention

[0004] In order to solve the problems in the prior art of inaccurate identification of external damage potential danger points and easy false detection and missed detection, the present invention provides an interference determination method for external damage potential danger points based on artificial intelligence. The images or monitoring images taken during the drone inspection are analyzed and processed by artificial intelligence methods to achieve accurate determination of potential danger points and achieve classification and evaluation, providing a more reliable basis for subsequent inspection and maintenance of potential danger points.

[0005] The present invention provides an interference determination method for external damage potential point based on artificial intelligence, and the technical solution to solve the technical problem includes the following steps:

[0006] Step 1: Image input, collect drone and surveillance photos;

[0007] Step 2: Feature extraction: use the pre-trained feature matrix to calculate the collected images and obtain the feature results;

[0008] Step 3: feature encoding, encoding the feature results in step 2;

[0009] Step 4: Result classification. After the encoding is completed, it is compared with the samples in the knowledge base to complete the classification;

[0010] Step 5: Output the results. According to the classification category, decode and output the corresponding warning information.

[0011] Preferably, the features in step 2 include linear features and nonlinear features.

[0012] Preferably, the linear feature is the result of a linear transformation between the image bit and its two-dimensional Euclidean space neighborhood.

[0013] Preferably, a linear filter is used to extract linear features, and the specific algorithm is:

[0014]

[0015] Among them, dst(x, y) is the target image, that is, the result of linear transformation;

[0016] kernel is the filter kernel involved in the calculation.

[0017] cols represents the row length (number of columns) of the filter kernel;

[0018] rows represents the column length (number of rows) of the filter kernel;

[0019] src is the original image;

[0020] x, y are bits;

[0021] x', y' are the bits currently being calculated;

[0022] Anchor is the anchor point involved in the calculation, and anchor determines the step distance of the kernel.

[0023] Preferably, according to different filter kernels, the linear features can be extracted into edge features and shape features.

[0024] Preferably, the nonlinear features are extracted by GLCM, and the nonlinear features include contrast, correlation, entropy, and homogeneity.

[0025] Preferably, in step 3, feature encoding is performed based on the GBDT classification algorithm: for the input feature matrix and the priori result, a classifier is assumed to satisfy the following relationship:

[0026]

[0027] Where: N is the number of nodes of the classifier, i is the state of the classifier;

[0028] Assume the total number of iterations is M, and the current iteration cycle is m, then:

[0029] Solution: Get the result and update:

[0030] Get the final solution,

[0031] Finally, f i(x) is the feature encoder, and its output is the feature value.

[0032] Preferably, the comparison in step 4 is performed by comparing the characteristic value with the target value, wherein during the comparison process, different information is compared respectively using different vector values, and the value with the smallest variance is selected.

[0033] In summary, the technical solution of the present invention has at least the following beneficial effects:

[0034] 1. It can extract features from the images captured by drones during inspections using feature matrices, and can convert the captured images into feature information for subsequent analysis and processing;

[0035] 2. By extracting the linear and nonlinear features of the image respectively, the subsequent comparison is more accurate;

[0036] 3. By assigning different alarm information to the target value, different levels of settings can be achieved according to different characteristic values. BRIEF DESCRIPTION OF THE DRAWINGS

[0037] In order to more clearly illustrate the technical solutions in the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0038] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION

[0039] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0040] The present invention provides an interference determination method for external damage potential points based on artificial intelligence, see attached Figure 1 , comprising the following steps;

[0041] Step 1: Image input, collect drone and surveillance photos; during the inspection process, take photos by controlling the drone or the drone automatically cruises, or use surveillance photos in certain areas to collect the captured pictures / images for input;

[0042] Step 2: Feature extraction: use the pre-trained feature matrix to calculate the collected images and obtain the feature results;

[0043] Step 3: feature encoding, encoding the feature results in step 2;

[0044] Step 4: Result classification. After the encoding is completed, it is compared with the samples in the knowledge base to complete the classification;

[0045] Step 5: Output the results. According to the classification category, decode and output the corresponding warning information.

[0046] It should be noted that the features in step 2 of the present invention include linear features and nonlinear features, and the linear features and nonlinear features are extracted respectively by corresponding algorithms. The linear features are the result of linear transformation between the image bit and the two-dimensional Euclidean space neighborhood in which they are located. In the specific calculation, a linear filter is used to extract the linear features, and the specific algorithm is:

[0047]

[0048] Among them, dst(x, y) is the target image, that is, the result of linear transformation;

[0049] Kernel is the filter kernel involved in the calculation, that is, a two-dimensional vector. Different filter kernels extract different features.

[0050] information;

[0051] cols represents the row length (number of columns) of the filter kernel;

[0052] rows represents the column length (number of rows) of the filter kernel;

[0053] src is the original image;

[0054] x, y are bits;

[0055] x', y' are the bits currently being calculated;

[0056] Anchor is the anchor point involved in the calculation. Anchor determines the step distance of the kernel and the density of feature extraction.

[0057] It should be noted that, in the process of extracting linear features in the present invention, the linear features can be extracted into edge features and shape features according to different filter kernels.

[0058] The nonlinear features in the present invention are extracted by GLCM, wherein the nonlinear features mentioned in the present invention mainly include contrast, correlation, entropy, and homogeneity. When performing the specific calculation and extraction, the following method is used:

[0059] For a grayscale image of a color image, its gray level co-occurrence matrix (GLCM) is expressed as follows:

[0060]

[0061] Where (Δx, Δy) is a set of offsets, and the offset is a vector with length and direction;

[0062] C is the gray-level co-occurrence matrix, I is the input gray-level image, and x and y are bits;

[0063] After extracting GLCM, the following commonly used eigenvalues ​​(p(i, j) are obtained by calculating GLCM:

[0064] bits in GLCM):

[0065] Contrast ratio:

[0066]

[0067] Correlation (μ is the correlation coefficient):

[0068]

[0069] entropy:

[0070]

[0071] Homogeneity:

[0072]

[0073] The above four eigenvalues ​​are the nonlinear features to be extracted and their calculation precautions in the present invention.

[0074] It should be noted that the feature encoding in step 3 of the present invention is performed based on the GBDT classification algorithm, specifically as follows: for the input feature matrix and the priori result, a classifier is assumed to satisfy the following relationship:

[0075]

[0076] Where: N is the number of nodes of the classifier, i is the state of the classifier;

[0077] Assume the total number of iterations is M, and the current iteration cycle is m, then:

[0078] Solution:

[0079] Get the result and update:

[0080] Get the final solution,

[0081] Finally, f i (x) is the feature encoder, and its output is the feature value.

[0082] The comparison in step 4 of the present invention is to compare the obtained eigenvalue with the target value. Because different warning information is represented by different eigenvalues, the output result is compared with the priori eigenvalue, and the value with the smallest variance is selected. Different information is compared using different vector values, which can realize the recognition of specific areas (prior interest areas) in infrared images and identify their categories.

[0083] The present invention first determines that orthographic images are the only data source for identifying external damage. On the basis of intelligently identifying the hidden danger points of external damage construction, the image is divided into regions by obtaining the altitude information of the image and the tower ledger data, combined with the focal length of the drone gimbal, and the pixel value is allocated to the target position according to the difference between the similarity matrix of the target position and the similarity matrix of the neighboring position to obtain a grayscale image; the grayscale image is segmented into regions to obtain a regional segmentation image, which can be divided into a defect hidden danger area within a certain range of the line channel, a safety precaution area a little further away, and a concern area a little further away, and then the risk degree of the risk point is graded and judged.

[0084] The above is a preferred embodiment of the present invention. It should be pointed out that a person skilled in the art can make several improvements and modifications without departing from the principle of the present invention. These improvements and modifications are also considered to be within the scope of protection of the present invention.

Claims

1. The method for determining interference of external damage potential points based on artificial intelligence is characterized by: The steps include: Step 1: Image input, collect drone and surveillance photos; Step 2: Feature extraction: use the pre-trained feature matrix to calculate the collected images and obtain the feature results; Step 3: feature encoding, encoding the feature results in step 2; Step 4: Result classification. After the encoding is completed, it is compared with the samples in the knowledge base to complete the classification; Step 5: Output the results, decode and output the corresponding warning information according to the classification category; The features in step 2 include linear features and nonlinear features, and the linear features and nonlinear features are extracted respectively by corresponding algorithms; The linear feature is the result of a linear transformation between the image bit and its two-dimensional Euclidean space neighborhood; According to the different filter kernels, the linear features can be extracted into edge features and shape features; The nonlinear features are extracted by GLCM, and the nonlinear features include contrast, correlation, entropy, and homogeneity; Use linear filters to extract linear features. The specific algorithm is: , in, is the target image, i.e. the result of linear transformation; is the filter kernel involved in the calculation, cols represents the number of columns of the filter kernel; rows represents the number of rows of the filter kernel; is the original image; , It is a bit; , is the bit currently being calculated; is the anchor point involved in the calculation, Decided Step distance; In the step 3, feature encoding is performed based on the GBDT classification algorithm: for the input feature matrix and the prior results, a classifier is assumed to satisfy the following relationship: , in: is the number of nodes of the classifier, is the state of the classifier; Assume the total number of iterations is M, and the current iteration cycle is m, then: , solve: , Get the result and update: , and obtain the final solution, at last, It is the feature encoder, and its output is the feature value; The comparison in step 4 is performed by comparing the characteristic value with the target value, wherein during the comparison process, different information is compared respectively using different vector values, and the value with the smallest variance is selected.

Images