Insulator string status identification method, system and device

Abstract

The present invention discloses a method, system, and device for identifying the state of an insulator string. The method involves obtaining an image of an insulator string to be detected and extracting multiple texture features of the insulator string image using a preset texture feature extraction model. The similarity between each texture feature and a plurality of tree features within a preset tree data structure is then calculated. The texture features are then classified into texture states corresponding to the respective tree data structures according to the similarity, and the number of classifications for each texture state is counted in real time. Finally, the texture state corresponding to the largest number of classifications is selected as the target insulator string state corresponding to the insulator string image. The texture features corresponding to the insulator string image are extracted using the texture feature extraction model and compared for similarity with the tree features within the preset tree data structure, thereby determining the target insulator string state, improving recognition speed, and achieving high recognition accuracy.

Description

Technical Field

[0001] The present invention relates to the technical field of insulator string state identification, and in particular to an insulator string state identification method, system and device. Background Art

[0002] In power systems, the safety and reliability of transmission lines are crucial to the smooth operation of power transmission. Insulator strings, a key component of transmission lines that provides insulation and support, can cripple the entire line if they develop defects. Therefore, identifying the status of insulator strings is a core component of power system transmission line inspections.

[0003] At present, manual inspection and drone inspection methods are used to identify whether insulator strings have defects such as dropped strings or dropped pieces. Since insulator strings are installed at high places or in places that are difficult for people to access, the use of manual inspection methods will make it difficult for inspectors to make judgments. Therefore, drone inspection methods are often used to identify the status of insulator strings.

[0004] However, the existing drone inspection method requires manual evaluation of the photos after taking them to determine whether the insulator strings have defects. Manual evaluation is not only slow but also prone to misjudgment, resulting in low accuracy of the detection results. Summary of the Invention

[0005] The present invention provides a method, system and device for identifying the status of an insulator string, which solves the technical problem that the existing drone inspection method requires manual evaluation of the photos after taking them to determine whether the insulator string has defects. Manual evaluation is not only slow but also prone to misjudgment, resulting in low accuracy of the detection results.

[0006] The present invention provides a method for identifying the state of an insulator string, comprising:

[0007] Obtain an image of the insulator string to be inspected;

[0008] Extracting multiple texture features of the insulator string image using a preset texture feature extraction model;

[0009] respectively calculating the similarity between each of the texture features and the tree features in a plurality of preset tree data structures;

[0010] Classifying the texture features into texture states corresponding to the respective tree-shaped data structures according to the similarities, and counting the number of classifications of the respective texture states in real time;

[0011] The texture state corresponding to the largest number of classifications is selected as the target insulator string state corresponding to the insulator string image.

[0012] Optionally, it also includes:

[0013] Acquire multiple initial insulator string state images, and label the initial insulator string state images according to preset insulator string state types to obtain an initial training sample set;

[0014] Performing target detection on the initial training sample set using a preset target detection model to obtain a credibility value corresponding to each initial insulator string state image in the initial training sample set;

[0015] Performing time series target detection on the initial training sample set using a preset time series detection model, and obtaining a target training sample set in combination with the credibility value;

[0016] The texture features corresponding to the target training sample set are extracted by the texture feature extraction model, clustered, and the corresponding tree data structure is constructed.

[0017] Optionally, the step of obtaining a plurality of initial insulator string state images and labeling the initial insulator string state images according to preset insulator string state types to obtain an initial training sample set includes:

[0018] Acquiring a plurality of initial insulator string state images captured by at least one insulator string image capture device;

[0019] Marking each of the initial insulator string state images according to a preset insulator string state type;

[0020] The labeled initial insulator string state image is amplified to obtain an initial training sample set.

[0021] Optionally, the preset target detection model includes a target detection layer; and the step of performing target detection on the initial training sample set using the preset target detection model to obtain a credibility value corresponding to each initial insulator string state image in the initial training sample set includes:

[0022] Obtaining a true value of an insulator string corresponding to each initial insulator string state image in the initial training sample set;

[0023] Obtaining the insulator string prediction value corresponding to the initial insulator string state image through the target detection layer, and calculating the corresponding intersection-to-parallel ratio based on the corresponding insulator string true value;

[0024] The intersection-to-parallel ratio that meets a preset credibility threshold is used as a credibility value of the initial insulator string state image.

[0025] Optionally, the step of performing time series target detection on the initial training sample set using a preset time series detection model and obtaining a target training sample set in combination with the credibility value includes:

[0026] According to the credibility value corresponding to each initial insulator string state picture in the initial training sample set, marking the substring state corresponding to the initial insulator string state picture;

[0027] Inputting the initial insulator string state images in the initial training sample set into a preset time series detection model in the order of collection, and inputting the difference between the credibility value corresponding to each initial insulator string state image and the credibility values ​​of the preceding ones;

[0028] Performing a linear regression calculation on the credibility value, the difference, and the substring state corresponding to each of the initial insulator substring state images using the time series detection model to obtain a calculation result;

[0029] Comparing the calculation result with a preset threshold to obtain a comparison result;

[0030] According to the comparison result and in combination with the insulator string state type, the insulator string state type corresponding to the initial insulator string state image is updated to obtain a target training sample set.

[0031] Optionally, the step of extracting texture features corresponding to the target training sample set by the texture feature extraction model, performing clustering, and constructing the corresponding tree data structure includes:

[0032] Extracting texture features corresponding to each target insulator string state image in the target training sample set through the convolution layer of the texture feature extraction model, thereby obtaining a texture set corresponding to the target training sample set;

[0033] Clustering the texture vectors corresponding to the texture features in the texture set according to the insulator string state type corresponding to the texture feature, and obtaining clusters corresponding to the insulator string state type in combination with a preset number of cluster centers;

[0034] Setting each texture vector in the cluster as a corresponding tree feature;

[0035] Using the insulator string state type corresponding to the cluster as the texture state corresponding to the tree feature;

[0036] According to the tree features corresponding to the clusters and the texture states corresponding to the tree features, a corresponding tree data structure is constructed.

[0037] Optionally, the step of respectively calculating the similarity between each of the texture features and the tree features in a plurality of preset tree data structures includes:

[0038] respectively calculating distance values ​​between a texture vector corresponding to each of the texture features and texture vectors corresponding to tree features in a plurality of preset tree data structures;

[0039] Each of the distance values ​​is converted into a similarity according to a preset similarity conversion formula to obtain a similarity corresponding to the distance value.

[0040] Optionally, the step of classifying the texture features into texture states corresponding to the respective tree-shaped data structures according to the similarities, and counting the number of classifications of the respective texture states in real time, comprises:

[0041] Selecting the tree data structure corresponding to the maximum similarity between the texture feature and the tree features in a plurality of preset tree data structures;

[0042] The texture state corresponding to the tree data structure is used as the texture state of the texture feature, and the classification quantity of each texture state is counted in real time.

[0043] The present invention also provides an insulator string status identification system, comprising:

[0044] An insulator string image acquisition module is used to acquire an image of the insulator string to be inspected;

[0045] A texture feature extraction module, configured to extract a plurality of texture features of the insulator string image using a preset texture feature extraction model;

[0046] A similarity calculation module, configured to respectively calculate the similarity between each of the texture features and the tree features in a plurality of preset tree data structures;

[0047] a texture state classification and classification quantity statistics module, configured to classify the texture features into texture states corresponding to the respective tree-shaped data structures according to the similarities, and to count the classification quantity of each of the texture states in real time;

[0048] The target insulator string state selection module is configured to select the texture state corresponding to the largest number of classifications as the target insulator string state corresponding to the insulator string image.

[0049] The present invention also provides an electronic device, comprising a memory and a processor, wherein the memory stores a computer program, and when the computer program is executed by the processor, the processor executes the steps of implementing any of the above-mentioned insulator string status identification methods.

[0050] It can be seen from the above technical solutions that the present invention has the following advantages:

[0051] The present invention obtains an image of an insulator string to be inspected and extracts multiple texture features of the insulator string image using a preset texture feature extraction model. The similarity between each texture feature and the tree features within a plurality of preset tree data structures is then calculated. The texture features are then classified into texture states corresponding to the respective tree data structures according to the similarity, and the number of classifications for each texture state is counted in real time. Finally, the texture state corresponding to the largest number of classifications is selected as the target insulator string state corresponding to the insulator string image. This solves the technical problem that the existing drone inspection method requires manual determination of whether the insulator string has defects after taking a photo. Manual determination is not only slow but also prone to misjudgment, resulting in low detection accuracy. By extracting the texture features corresponding to the insulator string image using a texture feature extraction model and comparing their similarity with the tree features within the preset tree data structure, the target insulator string state is determined, which not only improves recognition speed but also achieves high recognition accuracy. BRIEF DESCRIPTION OF THE DRAWINGS

[0052] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the following briefly introduces the drawings required for use in the embodiments or the description of the prior art. 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 paying any creative work.

[0053] Figure 1 A flowchart of a method for identifying an insulator string state provided in the first embodiment of the present invention;

[0054] Figure 2 A flowchart of a method for identifying an insulator string state provided in a second embodiment of the present invention;

[0055] Figure 3 This is a schematic diagram of the detection results output by the time series detection model in the second embodiment of the present invention;

[0056] Figure 4 This is a structural block diagram of an insulator string status identification system provided in Example 3 of the present invention. DETAILED DESCRIPTION

[0057] Embodiments of the present invention provide a method, system, and device for identifying the status of an insulator string, which are used to solve the technical problem that the existing drone inspection method requires manual evaluation of the photos after taking them to determine whether the insulator string has defects. Manual evaluation is not only slow but also prone to misjudgment, resulting in low accuracy of the detection results.

[0058] In order to make the purpose, features, and advantages of the present invention more obvious and easy to understand, the technical solutions in the embodiments of the present invention will be clearly and completely described below in conjunction with the drawings in the embodiments of the present invention. Obviously, the embodiments described below 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 making creative work are within the scope of protection of the present invention.

[0059] See also Figure 1 , Figure 1 A method for identifying the state of an insulator string is provided in the first embodiment of the present invention.

[0060] The present invention provides a method for identifying the state of an insulator string, comprising:

[0061] Step 101: Obtain an image of an insulator string to be inspected.

[0062] In the embodiment of the present invention, if it is necessary to detect whether an insulator string is in any state of normal, chipped or defective, a camera installed on the high-voltage line is used to obtain an insulator string image corresponding to the insulator string.

[0063] Step 102: extract multiple texture features of the insulator string image using a preset texture feature extraction model.

[0064] The default texture feature extraction model is trained on ImageNet data and is used to extract texture features from images. It is typically constructed using VGG-VD. Texture features are small, semi-periodic, or regularly arranged patterns within a certain range of an image.

[0065] In an embodiment of the present invention, an image of an insulator string to be detected is input into a texture feature extraction model, and a convolution operation is performed on the image of the insulator string to be detected using a convolution filter through the convolution layer of the texture feature extraction model to extract texture features corresponding to the insulator string image, thereby outputting multiple texture features corresponding to the insulator string image and obtaining a texture set corresponding to the insulator string image.

[0066] Step 103: Calculate the similarity between each texture feature and the tree features in the plurality of preset tree data structures.

[0067] The preset tree data structure is constructed by clustering the texture vectors of each cluster according to the insulator string status type. This tree data structure includes a normal insulator string tree data structure, a chip-dropped insulator string tree data structure, and a defective insulator string tree data structure. The tree features are the texture features corresponding to each target insulator string status image in the target training set. Similarity refers to the degree of similarity between the texture features extracted from the insulator string image to be tested and the tree features within the tree data structure.

[0068] In an embodiment of the present invention, similarity calculation is performed on each texture feature extracted from the insulator string image to be detected and the tree features in the normal tree data structure of the insulator string, the chip-drop tree data structure of the insulator string, and the defect tree data structure of the insulator string.

[0069] Step 104: Classify the texture features into texture states corresponding to the respective tree-shaped data structures according to similarity, and count the number of classifications of each texture state in real time.

[0070] The texture state refers to the insulator string status type corresponding to the tree data structure, including normal insulator string, missing insulator string, and defective insulator string. The classification quantity refers to the number of texture features in the insulator string image that belong to each texture state.

[0071] In an embodiment of the present invention, a tree data structure corresponding to the maximum similarity between each texture feature and the tree features in multiple preset tree data structures is selected, the texture state corresponding to the tree data structure is used as the texture state of the texture feature, and the classification number of each texture state is counted in real time.

[0072] Step 105: Select the texture state corresponding to the largest number of classifications as the target insulator string state corresponding to the insulator string image.

[0073] The target insulator string state refers to the state corresponding to the insulator string in the insulator string image obtained after the insulator string image is recognized.

[0074] In an embodiment of the present invention, the maximum similarity between each texture feature and the tree feature is selected for classification to obtain the texture state corresponding to each texture feature and the classification number of each texture state, and the texture state corresponding to the maximum classification number is selected from the classification number as the target insulator string state of the insulator string image.

[0075] In an embodiment of the present invention, an image of an insulator string to be inspected is obtained and multiple texture features of the insulator string image are extracted using a preset texture feature extraction model. The similarity between each texture feature and tree features within a preset tree data structure is then calculated. The texture features are then classified into texture states corresponding to each tree data structure based on the similarity, and the number of classifications for each texture state is counted in real time. Finally, the texture state corresponding to the largest number of classifications is selected as the target insulator string state corresponding to the insulator string image. This solves the technical problem of existing drone inspection methods, which require manual evaluation of the photos after taking them to determine whether the insulator string has defects. This manual evaluation is not only slow but also prone to misjudgments, resulting in low detection accuracy. By extracting texture features corresponding to the insulator string image using a texture feature extraction model and comparing their similarity with the tree features within the preset tree data structure, the target insulator string state is determined, which not only improves recognition speed but also increases recognition accuracy.

[0076] See also Figure 2 , Figure 2 A method for identifying the state of an insulator string is provided in the second embodiment of the present invention.

[0077] Step 201: Acquire multiple initial insulator string state images, and label the initial insulator string state images according to preset insulator string state types to obtain an initial training sample set.

[0078] Furthermore, step 201 may include the following sub-steps S11-S13:

[0079] S11. Acquire multiple initial insulator string state images captured by at least one insulator string image capture device.

[0080] An insulator string image acquisition device refers to an image acquisition device installed around an insulator string, such as a camera mounted on a high-voltage line. An initial insulator string state image refers to a color image of the insulator string captured by the insulator string image acquisition device under various weather conditions, lighting conditions, and at different times.

[0081] In an embodiment of the present invention, a plurality of initial insulator string state images corresponding to various weather conditions, lighting conditions and different times of the insulator string are acquired by at least one insulator string image acquisition device.

[0082] S12. Label each initial insulator string state image according to a preset insulator string state type.

[0083] The preset insulator string status type is a status type set in advance according to the status that often occurs in the insulator string. The insulator string status types include insulator string normal, insulator string piece loss and insulator string defect.

[0084] In the embodiment of the present invention, the insulator string status type corresponding to each initial insulator string status image is marked according to the insulator string normal, insulator string missing, and insulator string defect among the insulator string status types, and the position of the insulator string in the initial insulator string status image is marked.

[0085] S13. Amplify the labeled initial insulator string state image to obtain an initial training sample set.

[0086] Augmentation refers to the operation of enhancing data and increasing the number of samples by rotating, mirroring, and other methods on the initial insulator string state image.

[0087] In an embodiment of the present invention, the labeled initial insulator string state images are rotated and mirrored left and right, thereby achieving data enhancement and increasing the number of initial insulator string state images. All initial insulator string state images are used to construct an initial training sample set.

[0088] Step 202: Perform target detection on the initial training sample set using a preset target detection model to obtain a credibility value corresponding to each initial insulator string state image in the initial training sample set.

[0089] Furthermore, the preset target detection model includes a target detection layer, and step 202 may include the following sub-steps S21-S23:

[0090] S21. Obtain the true value of the insulator string corresponding to each initial insulator string state image in the initial training sample set.

[0091] The true value of the insulator string refers to the true box corresponding to the insulator string in the initial insulator string state image obtained by manual annotation.

[0092] In the embodiment of the present invention, based on the insulator string state type, each initial insulator string state picture in the initial training sample set is manually labeled with the corresponding real box, that is, the real value of the insulator string.

[0093] S22. Obtain the insulator string prediction value corresponding to the initial insulator string state image through the target detection layer, and calculate the corresponding intersection-to-parallel ratio by combining it with the corresponding insulator string true value.

[0094] The object detection model predicts a prediction box and classification probability for each object, typically using the Yolov5 algorithm. The object detection layer is the layer within the object detection model that obtains the prediction box corresponding to the insulator string state in the initial insulator string state image and calculates the intersection-over-union ratio. The insulator string prediction value is the prediction box corresponding to the insulator string in the initial insulator string state image, output by the object detection layer.

[0095] In an embodiment of the present invention, an initial insulator string state image is input into a target detection model, and a target detection layer of the target detection model performs target recognition on the initial insulator string state image. The initial insulator string state image is labeled with a corresponding prediction box of the insulator string state corresponding to the recognition result, and the ratio of the intersection between the real box corresponding to the real value of the insulator string and the corresponding prediction box to the union between the real box corresponding to the real value of the insulator string and the corresponding prediction box is calculated to obtain the intersection-to-union ratio corresponding to the initial insulator string state image.

[0096] S23. Using the intersection-to-union ratio that meets a preset credibility threshold as a credibility value of the initial insulator string state image.

[0097] The preset credible threshold refers to a threshold set according to detection requirements to determine whether the predicted intersection-over-union ratio is correct.

[0098] In an embodiment of the present invention, each intersection-and-parallel ratio output by the target detection layer of the target detection model is compared with a preset trustworthy threshold, and the intersection-and-parallel ratio that meets the preset trustworthy threshold is used as the credibility value of the initial insulator string state image. If multiple intersection-and-parallel ratios meet the preset trustworthy threshold, the largest intersection-and-parallel ratio is selected as the credibility value of the initial insulator string state image.

[0099] Step 203: Perform time series target detection on the initial training sample set using a preset time series detection model, and obtain a target training sample set in combination with the credibility value.

[0100] Furthermore, step 203 may include the following sub-steps S31-S35:

[0101] S31. Mark the substring states corresponding to the initial insulator string state pictures according to the credibility values ​​corresponding to the initial insulator string state pictures in the initial training sample set.

[0102] The substring status refers to whether there is an insulating substring in the detection box. The substring status types are divided into insulating substring and non-insulating substring. When marking, 1 is usually used to indicate the presence of an insulating substring and 0 is used to indicate the absence of an insulating substring.

[0103] In an embodiment of the present invention, the credibility value corresponding to each initial insulator string state image in the initial training sample set is obtained, and the initial insulator string state image is marked with 1 or 0 according to whether there is an insulator string in the detection area corresponding to the credibility value.

[0104] S32. Input the initial insulator string state images in the initial training sample set into a preset time series detection model in the order of collection, and input the difference between the credibility value corresponding to each initial insulator string state image and the credibility values ​​preceding it.

[0105] The acquisition sequence refers to the order in which the initial insulator string status images are acquired from the smallest to the largest time. The preset time detection model refers to a model that performs refined detection of insulator strings based on time sequence.

[0106] In an embodiment of the present invention, the substring state to which each initial insulator string state picture belongs is labeled in the initial training sample set, and the labeled initial insulator string state pictures are sequentially input into a preset time series detection model in the chronological order of picture acquisition. Furthermore, the difference between the credibility value corresponding to each initial insulator string state picture and the credibility values ​​of the preceding ones is sequentially input into the time series detection model.

[0107] S33. Perform linear regression calculation on the credibility value, difference value, and substring state corresponding to each initial insulator substring state image through a time series detection model to obtain a calculation result.

[0108] In an embodiment of the present invention, the time series detection model uses the credibility value and difference value corresponding to each initial insulator string state image as input parameters, takes the substring state as the target, performs linear regression calculation, and obtains the calculation results corresponding to each initial insulator string state image.

[0109] S34: Compare the calculation result with a preset threshold to obtain a comparison result.

[0110] The preset threshold refers to a critical value determined based on the substring states of the insulator strings in the training sample and used to determine whether the insulator string state type in the initial insulator string state picture has changed. Since the output score of the time series detection model is between 0 and 1, the threshold is usually selected as 0.5.

[0111] In the embodiment of the present invention, the calculation results corresponding to the initial insulator string state images are respectively compared with preset thresholds to obtain comparison results corresponding to the initial insulator string state images.

[0112] S35. According to the comparison result and in combination with the insulator string state type, the insulator string state type corresponding to the initial insulator string state image is updated to obtain a target training sample set.

[0113] The target training sample set refers to the sample set obtained after the initial training sample set is trained by the target detection model and the time series detection model.

[0114] In this embodiment of the present invention, if the calculation result is less than a preset threshold, the insulator string state type corresponding to the initial insulator string state image corresponding to the calculation result changes. Based on the insulator string state type of the initial insulator string state image and the detection area corresponding to its credibility value, it is determined that the initial insulator string state image is in an insulator string chip-off or insulator string defect state, and the insulator string state type corresponding to the initial insulator string state image is updated. If the calculation result is greater than the preset threshold, the insulator string state type corresponding to the initial insulator string state image corresponding to the calculation result remains unchanged and does not need to be updated. The updated initial insulator string state images and the original initial insulator string state images are used to construct a target training sample set.

[0115] like Figure 3 As shown in the figure, the horizontal axis represents the image frames corresponding to the initial insulator string state images captured by the insulator string image acquisition device, sorted in the acquisition order. The vertical axis represents the score output by the time series detection model, with a preset threshold of 0.5. The yolov5 score refers to the credibility value corresponding to each image frame output by the target detection model, and the predict score refers to the predicted value output by the time series detection model, i.e., the calculation result described above. The figure shows that after frame 235, the insulator string state type of the initial insulator string state image corresponding to the initial insulator string state image has changed, indicating that an insulator string is missing or defective. Based on the detection area corresponding to the initial insulator string state image, the updated state type of the initial insulator string state image is determined.

[0116] Step 204: extract texture features corresponding to the target training sample set through a texture feature extraction model, perform clustering, and construct a corresponding tree data structure.

[0117] Furthermore, step 204 may include the following sub-steps S41-S45:

[0118] S41. Extracting texture features corresponding to each target insulator string state image in the target training sample set through the convolution layer of the texture feature extraction model, and obtaining a texture set corresponding to the target training sample set.

[0119] A convolutional layer is a layer used to extract texture features from an image. A texture set is a collection of texture features from each image in a sample set.

[0120] In this embodiment of the present invention, each target insulator string state image within the target training sample set is input into a texture feature extraction model. The convolutional layer of the texture feature extraction model performs a convolution operation on each target insulator string state image using a convolution kernel. The 512-dimensional data output from the last convolutional layer is used as the texture features of the corresponding target insulator string state image. The texture feature extraction model is fed into the texture feature extraction model according to the corresponding insulator string state type, resulting in a normal insulator string texture set, a missing insulator string texture set, and a defective insulator string texture set. The normal, missing, and defective insulator string texture sets are combined to obtain the texture set corresponding to the target training sample set.

[0121] S42 , clustering the texture vectors corresponding to the texture features in the texture set according to the insulator string state type corresponding to the texture features, and obtaining clusters corresponding to the insulator string state types in combination with a preset number of cluster centers.

[0122] Clustering involves dividing texture vectors according to their corresponding insulator string status types. This is typically done using the k-means clustering algorithm. Cluster centers are used in the clustering process. The number of cluster centers can be set as needed, with 64 cluster centers available. Cluster clusters are the collections of texture vectors clustered according to insulator string status types.

[0123] In an embodiment of the present invention, for the normal texture set of the insulator string, the chip-drop texture set of the insulator string, and the defective texture set of the insulator string, k-means and 64 clustering centers are used to cluster the texture vectors corresponding to each texture feature in the texture set to obtain the normal clustering cluster of the insulator string, the chip-drop clustering cluster of the insulator string, and the defective clustering cluster of the insulator string.

[0124] S43. Set each texture vector in the cluster as a corresponding tree feature.

[0125] Tree features refer to the nodes in the tree data structure.

[0126] In the embodiment of the present invention, the texture vectors in each cluster are respectively set as tree features for constructing a tree data structure.

[0127] S44. Using the insulator string state type corresponding to the cluster as the texture state corresponding to the tree feature.

[0128] In the embodiment of the present invention, the insulator string state type corresponding to each cluster is used as the texture state of each tree feature corresponding to the cluster.

[0129] S45. Construct a corresponding tree data structure according to the tree features corresponding to the clusters and the texture states corresponding to the tree features.

[0130] The tree data structure, namely kd-tree, refers to a tree data structure constructed by all texture vectors in a cluster, and is used for searching tree features.

[0131] In an embodiment of the present invention, according to the texture state corresponding to the tree feature, all tree features of the same texture state are respectively used to construct a tree data structure corresponding to the texture state, thereby obtaining a normal tree data structure of an insulator string, a chip-dropped tree data structure of an insulator string, and a defective tree data structure of an insulator string.

[0132] Step 205: Obtain an image of the insulator string to be inspected.

[0133] In an embodiment of the present invention, according to detection requirements, an insulator string image corresponding to the insulator string state to be detected is obtained, and the insulator string image can be obtained from a camera installed on the high-voltage line that can capture the insulator string.

[0134] Step 206: extract multiple texture features of the insulator string image using a preset texture feature extraction model.

[0135] In an embodiment of the present invention, the obtained insulator string image to be detected is input into a texture feature extraction model, the convolution layer of the texture feature extraction model uses a convolution kernel to perform a convolution operation on the insulator string image, and the 512-dimensional data output by the last convolution layer is used as the texture feature of the corresponding insulator string image, thereby obtaining multiple texture features corresponding to the insulator string image.

[0136] Step 207: Calculate the similarity between each texture feature and the tree features in the plurality of preset tree data structures.

[0137] Furthermore, step 207 may include the following sub-steps S51-S52:

[0138] S51 , respectively calculating distance values ​​between a texture vector corresponding to each texture feature and texture vectors corresponding to tree features in a plurality of preset tree data structures.

[0139] In an embodiment of the present invention, after obtaining multiple texture features corresponding to the insulator string image, the distance values ​​between the texture vector corresponding to each texture feature and each texture vector in the normal tree data structure of the insulator string, the missing piece tree data structure of the insulator string, and the defective tree data structure of the insulator string are calculated.

[0140] S52 : performing similarity conversion on each distance value according to a preset similarity conversion formula to obtain the similarity corresponding to the distance value.

[0141] The preset similarity conversion formula refers to a formula for converting a distance value into a corresponding similarity according to the conversion principle that the smaller the distance value, the greater the similarity, such as a conversion formula used to express similarity using Euclidean distance.

[0142] In the embodiment of the present invention, each distance value is brought into a preset similarity conversion formula, and each distance value is converted into a corresponding similarity.

[0143] Step 208: Classify the texture features into texture states corresponding to each tree data structure according to similarity, and count the number of classifications of each texture state in real time.

[0144] Furthermore, step 208 may include the following sub-steps S61-S62:

[0145] S61: Select a tree data structure corresponding to the maximum similarity between a texture feature and tree features in a plurality of preset tree data structures.

[0146] In this embodiment of the present invention, the maximum similarity among the similarities corresponding to each texture feature and each tree feature in the normal insulator string tree data structure, the missing insulator string tree data structure, and the defective insulator string tree data structure is selected. The tree data structure corresponding to the maximum similarity is used as the tree data structure of the texture feature.

[0147] S62: Using the texture state corresponding to the tree data structure as the texture state of the texture feature, and counting the number of classifications of each texture state in real time.

[0148] In an embodiment of the present invention, after determining the tree data structure corresponding to each texture feature, the texture state corresponding to the tree data structure is used as the texture state of the texture feature, and the number of texture features corresponding to each texture state is used as the classification quantity corresponding to each texture state.

[0149] Step 209: Select the texture state corresponding to the largest number of classifications as the target insulator string state corresponding to the insulator string image.

[0150] In the embodiment of the present invention, the largest classification number is selected from the classification numbers corresponding to the various texture states, and the texture state corresponding to the largest classification number is used as the target insulator string state corresponding to the insulator string image.

[0151] In an embodiment of the present invention, an initial training sample set is obtained by obtaining multiple initial insulator string state images and labeling them according to preset insulator string state types. The initial training sample set is then subjected to target detection using a preset target detection model to obtain a credibility value corresponding to each initial insulator string state image within the initial training sample set. A preset time series detection model is then used to perform time series target detection on the initial training sample set, and the credibility values ​​are combined to obtain a target training sample set. Finally, a texture feature extraction model is used to extract texture features corresponding to the target training sample set, clustering them to construct a corresponding tree data structure. When insulator string state detection is required, an insulator string image to be detected is obtained, and multiple texture features of the insulator string image are extracted using the preset texture feature extraction model. The similarity between each texture feature and the tree features within the multiple preset tree data structures is calculated, and each distance value is converted to a similarity using a preset similarity conversion formula to obtain the similarity corresponding to the distance value. The texture features are then classified into texture states corresponding to various tree-like data structures based on similarity. The number of classifications for each texture state is counted in real time, and the texture state corresponding to the largest number of classifications is selected as the target insulator string state corresponding to the insulator string image. A corresponding tree-like data structure is constructed based on the insulator string state type. When the insulator string state needs to be detected, the insulator string image is simply fed into the trained texture feature extraction model to extract the multiple texture features corresponding to the insulator string image. The model then searches through the constructed multiple tree-like data structures to determine the target insulator string state for the insulator string in the insulator string image, achieving fast recognition speed and high accuracy.

[0152] See also Figure 4 , Figure 4 This is a structural block diagram of an insulator string status identification system provided in Example 3 of the present invention.

[0153] An embodiment of the present invention provides an insulator string status identification system, including:

[0154] The insulator string image acquisition module 401 is used to acquire an image of the insulator string to be detected.

[0155] The texture feature extraction module 402 is configured to extract multiple texture features of the insulator string image using a preset texture feature extraction model.

[0156] The similarity calculation module 403 is used to respectively calculate the similarity between each texture feature and the tree features in the plurality of preset tree data structures.

[0157] The texture state classification and classification quantity statistics module 404 is used to classify the texture features into the texture states corresponding to each tree data structure according to similarity, and to count the classification quantity of each texture state in real time.

[0158] The target insulator string state selection module 405 is configured to select the texture state corresponding to the largest number of classifications as the target insulator string state corresponding to the insulator string image.

[0159] Optionally, the system further includes:

[0160] The initial training sample set obtaining module is used to obtain a plurality of initial insulator string state images and label the initial insulator string state images according to preset insulator string state types to obtain an initial training sample set.

[0161] The credibility value obtaining module is used to perform target detection on the initial training sample set through a preset target detection model to obtain the credibility value corresponding to each initial insulator string state image in the initial training sample set.

[0162] The target training sample set acquisition module is used to perform time series target detection on the initial training sample set through a preset time series detection model, and obtain the target training sample set in combination with the credibility value.

[0163] The tree data structure construction module is used to extract the texture features corresponding to the target training sample set through the texture feature extraction model, perform clustering, and construct the corresponding tree data structure.

[0164] Optionally, the module for obtaining the initial training sample set includes:

[0165] The initial insulator string state image acquisition module is used to acquire multiple initial insulator string state images acquired by at least one insulator string image acquisition device.

[0166] The initial insulator string state picture labeling module is used to label each initial insulator string state picture according to a preset insulator string state type.

[0167] The initial training sample set obtaining submodule is used to amplify the labeled initial insulator string state image to obtain the initial training sample set.

[0168] Optionally, the preset target detection model includes a target detection layer, and the credibility value obtaining module includes:

[0169] The insulator string true value acquisition module is used to obtain the true value of the insulator string corresponding to each initial insulator string state image in the initial training sample set.

[0170] The intersection-to-parallel ratio obtaining module is used to obtain the insulator string prediction value corresponding to the initial insulator string state image through the target detection layer, and calculate the corresponding intersection-to-parallel ratio by combining it with the corresponding insulator string true value.

[0171] The credibility value obtaining submodule is used to use the intersection-to-union ratio that meets a preset credibility threshold as the credibility value of the initial insulator string state image.

[0172] Optionally, the target training sample set obtaining module includes:

[0173] The substring state labeling module is used to label the substring state corresponding to the initial insulator substring state picture according to the credibility value corresponding to each initial insulator substring state picture in the initial training sample set.

[0174] The module for inputting the difference between the initial insulator string state images and the credibility is used to input the initial insulator string state images in the initial training sample set into the preset time series detection model in the order of collection, and input the difference between the credibility value corresponding to each initial insulator string state image and the credibility values ​​preceding it.

[0175] The calculation result obtaining module is used to perform linear regression calculation on the credibility value, difference and substring state corresponding to each initial insulation substring state image through a time series detection model to obtain the calculation result.

[0176] A comparison result obtaining module is used to compare the calculation result with a preset threshold value to obtain a comparison result;

[0177] The target training sample set obtaining submodule is used to update the insulator string state type corresponding to the initial insulator string state image according to the comparison result and the insulator string state type to obtain the target training sample set.

[0178] Optionally, the tree data structure building module includes:

[0179] The texture set acquisition module is used to extract the texture features corresponding to each target insulator string state image in the target training sample set through the convolution layer of the texture feature extraction model, and obtain the texture set corresponding to the target training sample set.

[0180] The clustering module is used to cluster the texture vectors corresponding to each texture feature in the texture set according to the insulator string state type corresponding to the texture feature, and obtain the cluster cluster corresponding to each insulator string state type in combination with the preset number of cluster centers.

[0181] The tree feature setting module is used to set each texture vector in the cluster as a corresponding tree feature.

[0182] The texture state setting module is used to use the insulator string state type corresponding to the cluster as the texture state corresponding to the tree feature.

[0183] The tree data structure construction submodule is used to construct a corresponding tree data structure according to the tree features corresponding to the clusters and the texture states corresponding to the tree features.

[0184] Optionally, the similarity calculation module 403 includes:

[0185] The distance value calculation module is used to respectively calculate the distance value between the texture vector corresponding to each texture feature and the texture vectors corresponding to the tree features in a plurality of preset tree data structures.

[0186] The similarity calculation submodule is used to perform similarity conversion on each distance value according to a preset similarity conversion formula to obtain the similarity corresponding to the distance value.

[0187] Optionally, the texture state classification and classification quantity statistics module 404 includes:

[0188] A tree data structure selection module is used to select a tree data structure corresponding to the maximum similarity between the texture feature and the tree features in a plurality of preset tree data structures.

[0189] The texture state classification and classification quantity statistics submodule is used to use the texture state corresponding to the tree data structure as the texture state of the texture feature and to count the classification quantity of each texture state in real time.

[0190] An embodiment of the present invention further provides an electronic device, comprising: a memory and a processor, wherein a computer program is stored in the memory; when the computer program is executed by the processor, the processor executes the insulator string status identification method as described in any of the above embodiments.

[0191] The memory may be an electronic memory such as a flash memory, an EEPROM (Electrically Erasable Programmable Read-Only Memory), an EPROM, a hard disk, or a ROM. The memory has storage space for program code for executing any of the method steps described above. For example, the storage space for program code may include individual program codes for implementing various steps in the above method. These program codes may be read from or written to one or more computer program products. These computer program products include program code carriers such as a hard disk, a compact disc (CD), a memory card, or a floppy disk. The program code may be compressed, for example, in a suitable format. When executed by a computing device, these codes cause the computing device to execute the various steps in the above-described method for identifying the state of an insulator string.

[0192] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the systems, devices and units described above can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0193] In the several embodiments provided in this application, it should be understood that the disclosed systems, devices and methods can be implemented in other ways. For example, the device embodiments described above are merely schematic. For example, the division of units is only a logical function division. In actual implementation, there may be other division methods, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be an indirect coupling or communication connection through some interface, device or unit, which can be electrical, mechanical or other forms.

[0194] Units described as separate components may or may not be physically separate, and components shown as units may or may not be physical units, that is, they may be located in one place or distributed across multiple network units. Some or all of these units may be selected to achieve the purpose of this embodiment according to actual needs.

[0195] In addition, the functional units in the various embodiments of the present invention may be integrated into a single processing unit, each unit may exist physically separately, or two or more units may be integrated into a single unit. The aforementioned integrated units may be implemented in the form of hardware or software functional units.

[0196] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product. The computer software product is stored in a storage medium and includes several instructions for enabling a computer device (which can be a personal computer, server, or network device, etc.) to execute all or part of the steps of the various embodiments of the present invention. The aforementioned storage medium includes: U disk, mobile hard disk, read-only memory (ROM, Read-Only Memory), random access memory (RAM, Random Access Memory), disk or optical disk, and other media that can store program code.

[0197] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. A method for identifying the state of an insulator string, characterized in that: include: Obtain an image of the insulator string to be inspected; Extracting multiple texture features of the insulator string image using a preset texture feature extraction model; The similarity between each texture feature and a plurality of tree features in a preset tree data structure is calculated respectively; the plurality of preset tree data structures are constructed by texture vectors corresponding to the respective texture features in the cluster according to the insulator string state type; the insulator string state type includes normal insulator string, missing insulator string, and defective insulator string; Classifying the texture feature into texture states corresponding to each of the tree data structures according to the similarity, and counting the number of classifications of each of the texture states in real time, the step comprising: selecting the tree data structure corresponding to the maximum similarity between the texture feature and tree features in a plurality of preset tree data structures, using the texture state corresponding to the tree data structure as the texture state of the texture feature, and counting the number of classifications of each of the texture states in real time; Selecting the texture state corresponding to the largest number of classifications as the target insulator string state corresponding to the insulator string image; The construction process of the tree data structure includes: Acquire multiple initial insulator string state images, and label the initial insulator string state images according to preset insulator string state types to obtain an initial training sample set; Performing target detection on the initial training sample set using a preset target detection model to obtain a credibility value corresponding to each initial insulator string state image in the initial training sample set; Performing time series target detection on the initial training sample set using a preset time series detection model, and obtaining a target training sample set in combination with the credibility value; Extracting texture features corresponding to the target training sample set through the texture feature extraction model, performing clustering, and constructing the corresponding tree data structure; The step of extracting texture features corresponding to the target training sample set by the texture feature extraction model, clustering the texture features, and constructing the corresponding tree data structure includes: Extracting texture features corresponding to each target insulator string state image in the target training sample set through the convolution layer of the texture feature extraction model, thereby obtaining a texture set corresponding to the target training sample set; Clustering the texture vectors corresponding to the texture features in the texture set according to the insulator string state type corresponding to the texture feature, and obtaining clusters corresponding to the insulator string state type in combination with a preset number of cluster centers; Setting each texture vector in the cluster as a corresponding tree feature; Using the insulator string state type corresponding to the cluster as the texture state corresponding to the tree feature; According to the tree features corresponding to the clusters and the texture states corresponding to the tree features, a corresponding tree data structure is constructed.

2. The insulator string status identification method according to claim 1, characterized in that: The step of obtaining a plurality of initial insulator string state images and labeling the initial insulator string state images according to preset insulator string state types to obtain an initial training sample set includes: Acquiring a plurality of initial insulator string state images captured by at least one insulator string image capture device; Marking each of the initial insulator string state images according to a preset insulator string state type; The labeled initial insulator string state image is amplified to obtain an initial training sample set.

3. The insulator string status identification method according to claim 1, characterized in that: The preset target detection model includes a target detection layer; the step of performing target detection on the initial training sample set using the preset target detection model to obtain a credibility value corresponding to each initial insulator string state image in the initial training sample set includes: Obtaining a true value of an insulator string corresponding to each initial insulator string state image in the initial training sample set; Obtaining the insulator string prediction value corresponding to the initial insulator string state image through the target detection layer, and calculating the corresponding intersection-to-parallel ratio based on the corresponding insulator string true value; The intersection-to-parallel ratio that meets a preset credibility threshold is used as a credibility value of the initial insulator string state image.

4. The insulator string status identification method according to claim 1, characterized in that: The step of performing time series target detection on the initial training sample set by using a preset time series detection model and obtaining a target training sample set in combination with the credibility value includes: According to the credibility value corresponding to each initial insulator string state picture in the initial training sample set, marking the substring state corresponding to the initial insulator string state picture; Inputting the initial insulator string state images in the initial training sample set into a preset time series detection model in the order of collection, and inputting the difference between the credibility value corresponding to each initial insulator string state image and the credibility values ​​of the preceding ones; Performing a linear regression calculation on the credibility value, the difference, and the substring state corresponding to each of the initial insulator substring state images using the time series detection model to obtain a calculation result; Comparing the calculation result with a preset threshold to obtain a comparison result; According to the comparison result and in combination with the insulator string state type, the insulator string state type corresponding to the initial insulator string state image is updated to obtain a target training sample set.

5. The insulator string status identification method according to claim 1, characterized in that: The step of respectively calculating the similarity between each of the texture features and the tree features in a plurality of preset tree data structures comprises: respectively calculating distance values ​​between a texture vector corresponding to each of the texture features and texture vectors corresponding to tree features in a plurality of preset tree data structures; Each of the distance values ​​is converted into a similarity according to a preset similarity conversion formula to obtain a similarity corresponding to the distance value.

6. An insulator string status identification system, characterized in that: include: An insulator string image acquisition module is used to acquire an image of the insulator string to be inspected; A texture feature extraction module, configured to extract a plurality of texture features of the insulator string image using a preset texture feature extraction model; a similarity calculation module for respectively calculating the similarity between each of the texture features and tree features within a plurality of preset tree data structures; the plurality of preset tree data structures being constructed by respectively using texture vectors corresponding to the respective texture features within the clusters according to insulator string status types; the insulator string status types including normal insulator string, missing insulator string, and defective insulator string; a texture state classification and classification quantity counting module, configured to classify the texture feature into texture states corresponding to each of the tree data structures according to the similarity, and to count the classification quantity of each of the texture states in real time, the step comprising: selecting the tree data structure corresponding to the maximum similarity between the texture feature and a plurality of tree features within a preset tree data structure, using the texture state corresponding to the tree data structure as the texture state of the texture feature, and counting the classification quantity of each of the texture states in real time; a target insulator string state selection module, configured to select a texture state corresponding to the largest number of classifications as a target insulator string state corresponding to the insulator string image; The system further comprises: An initial training sample set obtaining module is used to obtain a plurality of initial insulator string state images and label the initial insulator string state images according to a preset insulator string state type to obtain an initial training sample set; a credibility value obtaining module, configured to perform target detection on the initial training sample set using a preset target detection model, and obtain a credibility value corresponding to each initial insulator string state image in the initial training sample set; A target training sample set obtaining module is used to perform time series target detection on the initial training sample set using a preset time series detection model, and obtain a target training sample set in combination with the credibility value; A tree data structure construction module, configured to extract texture features corresponding to the target training sample set through the texture feature extraction model, perform clustering, and construct the corresponding tree data structure; Wherein, the tree data structure construction module includes: A texture set obtaining module, configured to extract texture features corresponding to each target insulator string state image in the target training sample set through the convolution layer of the texture feature extraction model, and obtain a texture set corresponding to the target training sample set; a cluster obtaining module, configured to cluster the texture vectors corresponding to the texture features in the texture set according to the insulator string state type corresponding to the texture feature, and obtain clusters corresponding to the insulator string state type in combination with a preset number of cluster centers; A tree feature setting module, used for setting each texture vector in the cluster as a corresponding tree feature; A texture state setting module, configured to use the insulator string state type corresponding to the cluster as the texture state corresponding to the tree feature; The tree data structure construction submodule is used to construct a corresponding tree data structure according to the tree features corresponding to the clusters and the texture states corresponding to the tree features.

7. An electronic device, characterized in that: The method comprises a memory and a processor, wherein a computer program is stored in the memory, and when the computer program is executed by the processor, the processor executes the steps of the insulator string status identification method according to any one of claims 1 to 5.

Images