Insulator detection method and system based on power transmission line insulator inspection model

Abstract

The invention discloses an insulator detection method and system based on a power transmission line insulator inspection model, and relates to the technical field of insulator defect detection. Comprising the following steps: acquiring image data of a power transmission line insulator, and constructing a power transmission line insulator image data set; dividing the power transmission line insulator image data set into a training set and a test set according to a certain proportion; establishing a power transmission line insulator inspection model; converting the power transmission line insulator inspection model into a nonlinear dynamic system, and pre-training the power transmission line insulator inspection model to obtain a pre-trained power transmission line insulator inspection model; performing fine tuning training on a pre-trained power transmission line insulator inspection model by using the training set; and inputting test set data into the power transmission line insulator inspection model after fine tuning training, and carrying out defect identification and positioning on the power transmission line insulator. According to the invention, the defect problem of the insulator can be found in time to ensure the normal operation of a power system.

Description

Technical Field

[0001] The present invention relates to the technical field of insulator defect detection, and in particular to an insulator detection method and system based on a transmission line insulator inspection model. Background Art

[0002] As electricity demand continues to rise, the transmission capacity of transmission lines is increasing, placing higher demands on transmission reliability. Power line inspections play a vital role in maintaining grid stability, but this task is becoming increasingly complex and labor-intensive. Insulators, as critical components of transmission lines, often face harsh outdoor environments, increasing the risk of defects and failures. Therefore, timely inspections are crucial to ensuring the safe operation of insulators, mitigating potential safety hazards, and improving power system reliability.

[0003] Over the past few years, deep neural networks (DNNs) have demonstrated impressive performance in a variety of fields, including image analysis, natural language processing, and audio recognition. However, optimizing DNN models remains a significant challenge, particularly in situations with significant data distribution imbalances. When the majority class dominates the training process, the resulting models tend to be biased toward these classes, resulting in poor performance on underrepresented minority classes. Furthermore, the high capacity of DNNs exacerbates this problem, as they tend to overfit to the majority class, resulting in complex and shifted decision boundaries and poor generalization.

[0004] Existing methods for addressing the class imbalance problem in DNNs can be roughly categorized into data-level methods and algorithm-level methods. Data-level methods aim to balance the class distribution by oversampling the minority class, reducing the majority sample, or creating artificial instances. However, these methods often introduce additional computational overhead and may not effectively capture the intrinsic characteristics of imbalanced data. Algorithm-level methods assign different misclassification costs to different classes or adjust the learning process based on class performance. However, these methods rely on predefined cost matrices or require additional computing resources, limiting their scalability and adaptability to different datasets. Furthermore, these methods cannot be directly applied to deep neural network training, significantly increasing the complexity of their application.

[0005] Therefore, it is an urgent problem for those skilled in the art to propose an insulator detection method and system based on a transmission line insulator inspection model to solve the difficulties existing in the prior art. Summary of the Invention

[0006] In view of this, the present invention provides an insulator detection method and system based on a transmission line insulator inspection model, which can timely discover insulator defects to ensure the normal operation of the power system. The proposed model training method adopts dynamic correction parameters to effectively reduce the overfitting of the decision boundary for the majority class and promote a more balanced and generalizable solution by dynamically reweighting samples.

[0007] In order to achieve the above object, the present invention adopts the following technical solutions:

[0008] An insulator detection method based on a transmission line insulator inspection model, comprising:

[0009] S1. Acquire image data of transmission line insulators and construct a transmission line insulator image dataset;

[0010] S2, dividing the transmission line insulator image dataset into a training set and a test set according to a certain ratio;

[0011] S3. Establish a transmission line insulator inspection model;

[0012] S4. Convert the transmission line insulator inspection model into a nonlinear dynamic system, pre-train the transmission line insulator inspection model based on a unified trajectory optimization method with dynamic correction parameters, add dynamic correction parameters, and optimize the decision boundary of the transmission line insulator inspection model so that the transmission line insulator inspection model quickly converges to a local optimal solution, thereby obtaining a pre-trained transmission line insulator inspection model.

[0013] S5. Use the training set to fine-tune the pre-trained transmission line insulator inspection model;

[0014] S6. Input the test set data into the fine-tuned and trained transmission line insulator inspection model to identify and locate defects in the transmission line insulators.

[0015] In the above method, the specific content of S1 is optional: collecting sample images of insulators in transmission lines of 110kV, 220kV and 500kV voltage levels at different locations and under different weather conditions in the transmission lines, and constructing a transmission line insulator image dataset.

[0016] In the above method, optionally, the transmission line insulator inspection model in S3 is a convolutional neural network model CNN, comprising: an input feature layer, a feature extraction layer, a downsampling layer, a dense layer, and a decision layer;

[0017] The transmission line insulator inspection model is:

[0018]

[0019] Among them, c∈[0,1] is the model’s predicted probability distribution for the i-th category, t i is the actual label value corresponding to the i-th category, y i represents the predicted value of the i-th category, ω is the weight parameter used for classification in the deep learning model, and x represents the original data input into the model.

[0020] In the above method, optionally, the nonlinear dynamic system of the transmission line insulator inspection model in S3 is:

[0021] C(ω)=[c(ω,x1)c(ω,x2)...c(,x N )] T =0,ω∈R N (2),

[0022] Where ω represents the weight parameter used for optimization and learning in the transmission line insulator inspection model, x1, x2, ..., x N They represent the eigenvalues extracted from the input data, and N is the number of mini-batch samples.

[0023] Alternatively, the nonlinear dynamics system in S4 adopts a dynamic gradient system:

[0024]

[0025] in, is the Jacobian matrix of C(ω), and β is the dynamic correction parameter for adjusting the class imbalance problem.

[0026] The above method, optionally, dynamically modifies the parameters used to estimate the relative contribution of each training example to the learned decision boundary:

[0027]

[0028] Among them, ||f(x,z)-y||1 is the L1 norm of the error, and ||R||1 is the L1 norm of the FC layer input feature (R).

[0029] The above method optionally introduces dynamic correction parameters into the dynamic gradient system:

[0030]

[0031] The above method is optional. The pre-training content in S4 is as follows:

[0032] 1) Initialize the maximum random iteration step size T max , mini-batch N b , initial point ω0, data x={x1,…,x B}, and divided into B batches of equal size, nonlinear dynamic system C i and its gradient DC i , i∈[1,B]; constant γ=10 -4 ;

[0033] 2) Set ω = ω0 and calculate the initial step size

[0034] 3) Computational dynamic gradient system Calculate dynamic correction parameters Calculating step size

[0035] 4) Propose pseudo-dynamic progressive method:

[0036]

[0037] Where ω is the weight parameter, I is the identity matrix, λ is the time increment, and D is the dynamic gradient system The derivative matrix is an n×n matrix, ε n is the dynamic learning rate;

[0038] A pseudo-dynamic progressive method of the main diagonal of the block is proposed to find the search direction G:

[0039]

[0040] Among them, α is the adjustment factor, and Represents the parameter to be optimized, It's G n The dynamic gradient system at , mo represents the monitoring operator, Diag(v n ) represents a diagonal matrix whose diagonal elements are derived from v n ;

[0041] Calculate G in formula (7) n And update the weights

[0042] 5) When When is a small positive scalar, ω is a feasible solution and returns ω; when i>T max When the maximum number of iterations is reached, return ω; otherwise, go to step 3);

[0043] 6) Call the local solver in the deep neural network to obtain a high-quality local optimal solution, then reach the maximum number of iterations and return ω.

[0044] An insulator detection system based on a transmission line insulator inspection model, configured to execute any of the above-mentioned insulator detection methods based on a transmission line insulator inspection model, comprising:

[0045] A data acquisition module is used to obtain image data of transmission line insulators and construct an image data set;

[0046] Data classification module, used to divide the image dataset into training set and test set;

[0047] Model building module, used to build a transmission line insulator inspection model;

[0048] Model pre-training module, based on the unified trajectory method with dynamic parameter correction, pre-trains the transmission line insulator inspection model for class imbalanced datasets;

[0049] The model fine-tuning training module, based on a dynamic gradient system with dynamic correction parameters, fine-tunes the pre-trained transmission line insulator inspection model using training set data;

[0050] The efficient model training module uses a pseudo-dynamic progressive method and a block main diagonal pseudo-dynamic progressive method to optimize the search path, thus efficiently training the transmission line insulator inspection model.

[0051] The model testing module inputs the test set data into the fine-tuned and trained transmission line insulator inspection model to identify and locate defects in transmission line insulators.

[0052] Through the above technical solutions, it can be seen that compared with the prior art, the present invention provides an insulator detection method and system based on a transmission line insulator inspection model, which has the following beneficial effects: the inspection model of the present invention can timely detect insulator defects through regular inspection of insulators, so as to take corresponding maintenance measures to ensure the normal operation of the power system, prevent the occurrence of potential accidents, reduce the risk of power supply interruption, and ensure the reliability, safety and stability of the power system; the multi-objective problem of the deep neural network loss function is used to construct a nonlinear dynamic system, which effectively solves the problem of model overfitting caused by class imbalance; a dynamic gradient system based on dynamic correction parameters is proposed, which assigns higher weights to samples with less impact on the decision boundary and imposes a certain penalty on samples with excessive impact on the decision boundary to achieve a reasonable distribution of sample weights, prevent the dominance of the majority class, innovatively solve the class imbalance problem, and greatly improve the accuracy of small sample classification; the block main diagonal pseudo-dynamic progressive method is proposed in the model training method, which can effectively generate good initial weights for the local solver without consuming too much time, with fast training speed and high accuracy. When the sample diversity is insufficient and the deep network scale is small, a better solution can be quickly found. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] 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 merely embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on the provided drawings without paying any creative work.

[0054] Figure 1 A flow chart of an insulator detection method based on a transmission line insulator inspection model provided by the present invention;

[0055] Figure 2 This is a diagram showing the effect of an insulator detection method based on a transmission line insulator inspection model provided by the present invention. DETAILED DESCRIPTION

[0056] The following will clearly and completely describe the technical solutions in the embodiments of the present invention in conjunction with the accompanying drawings. 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 making creative efforts are within the scope of protection of the present invention.

[0057] In this application, relational terms such as first and second are used only to distinguish one entity or operation from another entity or operation, and do not necessarily require or imply any actual relationship or order between these entities or operations. The terms "include", "comprises" or any other variations thereof are intended to cover non-exclusive inclusion, so that a process, method, article or apparatus comprising a series of elements includes not only those elements, but also other elements not explicitly listed, or elements inherent to such process, method, article or apparatus. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article or apparatus comprising the element.

[0058] Reference Figure 1 As shown, the present invention discloses an insulator detection method based on a transmission line insulator inspection model, comprising:

[0059] S1. Acquire image data of transmission line insulators and construct a transmission line insulator image dataset;

[0060] S2, dividing the transmission line insulator image dataset into a training set and a test set according to a certain ratio;

[0061] S3. Establish a transmission line insulator inspection model;

[0062] S4. Convert the transmission line insulator inspection model into a nonlinear dynamic system. Based on the unified trajectory optimization method of dynamic correction parameters, pre-train the transmission line insulator inspection model. Add dynamic correction parameters and optimize the decision boundary of the transmission line insulator inspection model to improve the overall performance of the class imbalanced data set. This allows the transmission line insulator inspection model to quickly converge to the local optimal solution, thus obtaining a pre-trained transmission line insulator inspection model.

[0063] S5. Use the training set to fine-tune the pre-trained transmission line insulator inspection model;

[0064] S6. Input the test set data into the fine-tuned and trained transmission line insulator inspection model to identify and locate defects in the transmission line insulators.

[0065] Furthermore, the specific content of S1 is: collect sample images of insulators in transmission lines of 110kV, 220kV and 500kV voltage levels at different locations and under different weather conditions in the transmission lines, and build a transmission line insulator image dataset.

[0066] Furthermore, the transmission line insulator inspection model in S3 is a convolutional neural network model (CNN), which includes: an input feature layer that receives raw input data; a feature extraction layer that extracts features from the data; a downsampling layer that performs dimensionality reduction on the extracted features; a dense layer that performs classification based on the extracted features; and a decision layer that generates the final classification results.

[0067] The transmission line insulator inspection model is:

[0068]

[0069] Among them, c∈[0,1] is the model’s predicted probability distribution for the i-th category, t i is the actual label value corresponding to the i-th category, y i represents the predicted value of the i-th category, ω is the weight parameter used for classification in the deep learning model, and x represents the original data input into the model;

[0070] A deep neural network is a regularized form of a multilayer perceptron with a multi-layer network structure. Its performance is usually evaluated using a loss function between the output and the target value. The training goal of optimizing a deep neural network is to find a set of neuron weights that makes the error as close to zero as possible and reduces the gap between the training error and the test error. This process aims to achieve the global minimum of the mean squared value (RMS) loss function. Different fields can use different loss functions, such as the root mean square error (RMSE) function, the cross entropy (CE) function, and the L1 function.

[0071] Furthermore, a nonlinear dynamic system is constructed in S3 to train a given deep neural network to a local optimal solution. In some cases, the local optimal solution may be the global optimal solution. The nonlinear dynamic system of the transmission line insulator inspection model is:

[0072] C(ω)=[c(ω,x1)c(ω,x2)...c(ω,x N ] T =0,ω∈R N (2),

[0073] Where ω represents the weight parameter used for optimization and learning in the transmission line insulator inspection model, x1, x2, ..., x N They represent the eigenvalues extracted from the input data, and N is the number of mini-batch samples.

[0074] Furthermore, an effective dynamic gradient system corresponding to the nonlinear dynamic system (2) of the transmission line insulator inspection model is established in S4, and the nonlinear dynamic system (2) of the transmission line insulator inspection model is solved by the dynamic trajectory of the constructed dynamic gradient system;

[0075] Nonlinear dynamics system using dynamic gradient system:

[0076]

[0077] in, is the Jacobian matrix of C(ω), β is the dynamic correction parameter for adjusting the imbalance problem; Dynamic Gradient System (DGS) is a nonlinear, non-hyperbolic dynamic system;

[0078] Specifically, when β=1, the focused loss is used. The system fully considers the gradient information and loss information of the deep neural network model and focuses the training on a set of sparse examples.

[0079] According to the optimal solution and stable equilibrium point: the stable equilibrium point (SEP) of the dynamic gradient system (3) corresponds to the local minimum of the energy function equation of the nonlinear dynamic system (2) of the following deep learning network model:

[0080]

[0081] Therefore, the solution of the deep neural network optimization model (1) can be obtained by solving the trajectory of the dynamic gradient system.

[0082] Furthermore, we assign larger weights to data with larger errors to increase their contribution to the training gradient and enhance the network's training ability for small sample data. The following dynamic correction parameter formula is proposed:

[0083]

[0084] Among them, ||f(x,z)-y||1 is the L1 norm of the error, and ||R||1 is the L1 norm of the FC layer input feature (R).

[0085] Furthermore, dynamic correction parameters are introduced into the dynamic gradient system to ensure that the reweighting between different classes is balanced, preventing the dominance of the majority class and improving the overall performance of imbalanced datasets:

[0086]

[0087] The solution curve of the dynamic gradient system (5) with the introduction of dynamic correction parameters is set from ω0∈R N Starting from the initial state, use Represents the trajectory of ω in the time series t: if Definition ω i ∈R N The equilibrium point of the dynamic gradient system (5) with the dynamic correction parameter is introduced. For any ε>0, δ>0, satisfying ‖ω0-ω i ‖<δ, that is If these conditions are met, then ω i is stable, Ω(ω i ) is defined as the equilibrium point ω i The stable domain, and in this stable domain, all trajectories eventually converge to the equilibrium point ω i :

[0088]

[0089] If the Jacobian matrix If the real part of the eigenvalue Re(λ)>0, then the equilibrium point ω i is considered divergent, the trajectory will not converge to that point; if the Jacobian matrix If the real part of all eigenvalues Re(λ) is less than 0, then the equilibrium point ω i is attractive, all trajectories converge to this point, for a given Jacobian matrix Its edge manifold M and edgeless manifold N are defined as follows:

[0090] M(ω)={(ω1,ω2,...ω i )∈R N |ω i ≥0}

[0091] N(ω)={(ω1,ω2,...,ω n )∈R N},

[0092] If the dynamic gradient system (5) with the introduction of dynamic correction parameters satisfies the following assumptions, the trajectories of all initial conditions of its edge manifold converge to the boundary of a certain equilibrium point over time:

[0093] The real parts of the eigenvalues of all equilibrium points on the attraction domain are not all 0;

[0094] The interactive relationship of trajectories near the equilibrium point is clear;

[0095] As t→∞, the dynamic behavior of the system is stable and the trajectories of all initial conditions converge to the corresponding equilibrium points.

[0096] Furthermore, the model is pre-trained based on a unified trajectory optimization method with dynamically modified parameters. This method increases the scale of network parameters that can be trained by the unified trajectory method to hundreds of millions. The content of pre-training in S4 is as follows:

[0097] 1) Initialize the maximum random iteration step size T max , mini-batch N b , initial point ω0, data x={x1,…,x B}, and divided into B batches of equal size, nonlinear dynamic system C i and its gradient DC i , i∈[1, B]; constant γ=10 -4 ;

[0098] 2) Set ω = ω0 and calculate the initial step size

[0099] 3) Computational dynamic gradient system Calculate dynamic correction parameters Calculating step size

[0100] 4) A pseudo-dynamic progressive method is proposed. By adjusting the search direction of the dynamic gradient system, the numerical integration process can be accelerated, thereby improving the training efficiency. The method is as follows:

[0101]

[0102] Where ω is the weight parameter, I is the identity matrix, λ is the time increment, and D is the dynamic gradient system The derivative matrix is an n×n matrix, ε n is the dynamic learning rate;

[0103] Use pseudo-dynamic asymptotic methods to quickly find equilibrium solutions, and adopt block main diagonal pseudo-dynamic asymptotic methods to find search paths. Use dynamic step-size strategies to improve the training efficiency of deep neural networks, making the optimization process more efficient.

[0104] A block main diagonal pseudo-dynamic progressive method is proposed to find the search direction G. An adaptive step-size strategy is adopted to improve the scalability of the nonlinear dynamic system training method and accelerate the training process of deep neural networks. The method is as follows:

[0105]

[0106] Among them, α is the adjustment factor, and G∈R n Represents the parameter to be optimized, It's G n The dynamic gradient system at , mo represents the monitoring operator, Diag(v n ) represents a diagonal matrix whose diagonal elements are derived from v n ;

[0107] Calculate G in formula (7) n And update the weights

[0108] 5) When When is a small positive scalar, ω is a feasible solution and returns σ; when i>T max When the maximum number of iterations is reached, return ω; otherwise, go to step 3);

[0109] 6) Call the local solver in the deep neural network to obtain a high-quality local optimal solution, then reach the maximum number of iterations and return ω.

[0110] An insulator detection system based on a transmission line insulator inspection model is used to execute an insulator detection method based on a transmission line insulator inspection model, comprising:

[0111] A data acquisition module is used to obtain image data of transmission line insulators and construct an image data set;

[0112] Data classification module, used to divide the image dataset into training set and test set;

[0113] Model building module, used to build a transmission line insulator inspection model;

[0114] Model pre-training module, based on the unified trajectory method with dynamic parameter correction, pre-trains the transmission line insulator inspection model for class imbalanced datasets;

[0115] The model fine-tuning training module, based on a dynamic gradient system with dynamic correction parameters, fine-tunes the pre-trained transmission line insulator inspection model using training set data;

[0116] The efficient model training module uses a pseudo-dynamic progressive method and a block main diagonal pseudo-dynamic progressive method to optimize the search path, thus efficiently training the transmission line insulator inspection model.

[0117] The model testing module inputs the test set data into the fine-tuned and trained transmission line insulator inspection model to identify and locate defects in transmission line insulators.

[0118] In a specific embodiment, taking a typical scene feature dataset of transmission line insulator inspection as an example, the method of the present invention is compared with other common optimization methods to verify the specific contents as follows:

[0119] This experimental example uses a dataset of typical transmission line insulator inspection scenarios as an example to analyze various potential insulator defects. This dataset has four classes, including 50k training images and 10k test images. The data for this typical transmission line insulator inspection scenario exhibits significant imbalance, with the largest class accounting for 51.52% and the smallest class accounting for 5.89%, respectively. To ensure the model's generalization, this paper considers factors such as voltage and weather information.

[0120] In order to comprehensively evaluate the performance of the unified trajectory method based on dynamic correction parameters proposed in the embodiment of the present invention, the embodiment selects several commonly used optimization methods for comparison, including the adaptive moment estimation method (Adam) and the stochastic gradient descent method (SGD).

[0121] The test was conducted using a dataset of typical transmission line insulator inspection scenarios. To evaluate the model's performance, the recognition accuracy was verified on four categories. The experimental results are shown in Table 1:

[0122] Table 1 Test results of typical scene characteristics of transmission line insulator inspection

[0123]

[0124] As can be seen from Table 1, the method of the present invention has achieved varying degrees of performance improvement in the optimization of different target detection models. In particular, in the recognition of two small categories of samples, ceramic insulator edge damage and glass insulator edge damage, the method of the present invention improves the accuracy by 13.2% and 5.5% and 9.1% and 6.5% respectively, compared with the SGD and Adam methods. This shows that the method can not only more accurately identify defective insulator materials, but also improve the overall performance of class-imbalanced datasets.

[0125] In order to accurately illustrate the effect of the implementation of the present invention, the detection results of the transmission line insulator inspection targets are visualized to compare the training performance differences between the method of the present invention and the SGD and Adam methods, as shown in the following example: Figure 2 As shown in Figure 1, the Grad-CAM method is used to generate heat maps corresponding to different optimization methods. As can be seen from the figure, the highlighted area in the heat map of the method of the present invention is tightly surrounded by the insulator, while the highlighted areas in the heat maps of the SGD and Adam methods are not obvious and are relatively scattered.

[0126] Experimental results show that compared with the SGD and Adam optimization methods, the heat map generated by the method of the present invention shows stronger and clearer features, and the method of the present invention performs better in locating the position of insulator defects, providing a reliable basis for insulator defect location.

[0127] Each embodiment in this specification is described in a progressive manner. The same or similar parts between the embodiments can be referred to each other. Each embodiment focuses on the differences from other embodiments. In particular, for system or system embodiments, since they are basically similar to method embodiments, the description is relatively simple. For relevant parts, refer to the partial description of the method embodiment. The system and system embodiments described above are merely schematic, wherein the units described as separate components may or may not be physically separated, and the components displayed as units may or may not be physical units, that is, they may be located in one place, or they may be distributed on multiple network units. Some or all of the modules can be selected according to actual needs to achieve the purpose of the solution of this embodiment. A person of ordinary skill in the art can understand and implement it without expending creative work.

[0128] The above description of the disclosed embodiments is intended to enable one skilled in the art to implement or use the present invention. Various modifications to these embodiments will be readily apparent to one skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the present invention. Therefore, the present invention is not limited to the embodiments shown herein but is intended to conform to the widest scope consistent with the principles and novel features disclosed herein.

Claims

1. An insulator detection method based on a transmission line insulator inspection model, characterized in that: include: S1. Acquire image data of transmission line insulators and construct a transmission line insulator image dataset; S2, dividing the transmission line insulator image dataset into a training set and a test set according to a certain ratio; S3. Establish a transmission line insulator inspection model; S4. Convert the transmission line insulator inspection model into a nonlinear dynamic system, pre-train the transmission line insulator inspection model based on a unified trajectory optimization method with dynamic correction parameters, add dynamic correction parameters, and optimize the decision boundary of the transmission line insulator inspection model so that the transmission line insulator inspection model quickly converges to a local optimal solution, thereby obtaining a pre-trained transmission line insulator inspection model. S5. Use the training set to fine-tune the pre-trained transmission line insulator inspection model; S6. Input the test set data into the fine-tuned and trained transmission line insulator inspection model to identify and locate defects in the transmission line insulators.

2. The insulator detection method based on the transmission line insulator inspection model according to claim 1, characterized in that: The specific content of S1 is: Collect sample images of insulators in transmission lines of 110kV, 220kV and 500kV voltage levels at different locations and under different weather conditions in the transmission lines, and build a transmission line insulator image dataset.

3. The insulator detection method based on the transmission line insulator inspection model according to claim 1, characterized in that: The transmission line insulator inspection model in S3 is a convolutional neural network model (CNN), which includes: input feature layer, feature extraction layer, downsampling layer, dense layer, and decision layer; The transmission line insulator inspection model is: Among them, c∈[0,1] is the model’s predicted probability distribution for the i-th category, t i is the actual label value corresponding to the i-th category, y i represents the predicted value of the i-th category, ω is the weight parameter used for classification in the deep learning model, and x represents the original data input into the model.

4. The insulator detection method based on the transmission line insulator inspection model according to claim 3, characterized in that: The nonlinear dynamic system of the transmission line insulator inspection model in S3 is: C(ω)=[c(ω,x1)c(ω,x2)...c(ω,x N ] T =0,ω∈R N (2), Where ω represents the weight parameter used for optimization and learning in the transmission line insulator inspection model, x1, x2, ..., x N They represent the eigenvalues extracted from the input data, and N is the number of mini-batch samples.

5. The insulator detection method based on the transmission line insulator inspection model according to claim 4, characterized in that: The nonlinear dynamics system in S4 uses a dynamic gradient system: in, is the Jacobian matrix of C(ω), and β is the dynamic correction parameter for adjusting the class imbalance problem.

6. The insulator detection method based on the transmission line insulator inspection model according to claim 5, characterized in that: The dynamic correction parameters are used to estimate the relative contribution of each training data to the learned decision boundary: Among them, ||f(x,z)-y||1 is the L1 norm of the error, and ||R||1 is the L1 norm of the FC layer input feature (R).

7. The insulator detection method based on the transmission line insulator inspection model according to claim 6, characterized in that: Introducing dynamic correction parameters into the dynamic gradient system:

8. The insulator detection method based on the transmission line insulator inspection model according to claim 7, characterized in that: The content of pre-training in S4 is as follows: 1) Initialize the maximum random iteration step size T max , mini-batch N b , initial point ω0, data x={x1,…,x B }, and divided into B batches of equal size, nonlinear dynamic system C i and its gradient DC i , i∈[1, B]; constant γ=10 -4 ; 2) Set ω = ω0 and calculate the initial step size 3) Computational dynamic gradient system Calculate dynamic correction parameters Calculating step size 4) Propose pseudo-dynamic progressive method: Where ω is the weight parameter, I is the identity matrix, λ is the time increment, and D is the dynamic gradient system The derivative matrix is an n×n matrix, ε n is the dynamic learning rate; A pseudo-dynamic progressive method of the main diagonal of the block is proposed to find the search direction G: Where α is the adjustment factor, and Represents the parameter to be optimized, It's G n The dynamic gradient system at , mo represents the monitoring operator, Diag(v n ) represents a diagonal matrix whose diagonal elements are derived from v n ; Calculate G in formula (7) n And update the weights 5) When When is a small positive scalar, ω is a feasible solution and returns ω; when i>T max When the maximum number of iterations is reached, return ω; otherwise, go to step 3); 6) Call the local solver in the deep neural network to obtain a high-quality local optimal solution, then reach the maximum number of iterations and return ω.

9. An insulator detection system based on a transmission line insulator inspection model, configured to execute the insulator detection method based on a transmission line insulator inspection model according to any one of claims 1 to 8, comprising: A data acquisition module is used to obtain image data of transmission line insulators and construct an image data set; Data classification module, used to divide the image dataset into training set and test set; Model building module, used to build a transmission line insulator inspection model; Model pre-training module, based on the unified trajectory method with dynamic parameter correction, pre-trains the transmission line insulator inspection model for class imbalanced datasets; The model fine-tuning training module, based on a dynamic gradient system with dynamic correction parameters, fine-tunes the pre-trained transmission line insulator inspection model using training set data; The efficient model training module uses a pseudo-dynamic progressive method and a block main diagonal pseudo-dynamic progressive method to optimize the search path, thus efficiently training the transmission line insulator inspection model. The model testing module inputs the test set data into the fine-tuned and trained transmission line insulator inspection model to identify and locate defects in transmission line insulators.

Images