Insulator defect automatic identification device and method based on deep learning

Abstract

The invention discloses an insulator defect automatic identification device and method based on deep learning, and belongs to the technical field of power equipment detection. The device comprises an image acquisition module, a data preprocessing module, a deep learning model module, a result analysis module and a storage and display module. The method comprises the following steps: acquiring an insulator image through multiple view angles and preprocessing the insulator image; utilizing an improved deep learning model to automatically identify and position defects; and evaluating the defect severity based on an identification result and reversely optimizing model parameters. According to the method, the deep neural network fused with the attention mechanism is adopted, various defect types such as cracks, damage, dirt accumulation and aging stripping can be effectively identified, accurate positioning and severity grading of the defects are realized, the automation degree, identification precision and generalization ability of insulator defect detection are remarkably improved, and the method is suitable for popularization and application. And reliable technical support is provided for intelligent operation and maintenance of an insulator of a power system.

Description

Technical Field

[0001] This invention belongs to the field of insulator detection technology, and particularly relates to an automatic insulator defect identification device and method based on deep learning. Background Technology

[0002] Insulators are critical components in power systems, and their operational status directly impacts grid security. Defect detection is a crucial step in ensuring their reliability. Traditional insulator defect detection relies heavily on manual visual inspection, which suffers from low efficiency, high subjectivity, and a high risk of missed defects. Existing automated detection methods are mostly based on traditional image processing algorithms, which have poor adaptability to complex backgrounds, lighting variations, and subtle defects, resulting in limited recognition accuracy. With the expansion of power grids and the increasing complexity of insulator operating environments, traditional methods are no longer sufficient to meet the demands for efficient and accurate detection.

[0003] The current limitations of related technologies mainly include: (1) reliance on manual annotation and experience-based judgment, resulting in low intelligence; (2) weak model generalization ability, leading to significant differences in recognition performance across different scenarios; and (3) lack of a model self-optimization mechanism based on recognition results, making it difficult to continuously improve detection performance. Therefore, it is urgent to design an automatic insulator defect identification device and method based on deep learning to achieve efficient defect identification and dynamic model optimization, thereby improving the intelligence and reliability of detection. Summary of the Invention

[0004] This invention provides an automatic insulator defect identification device and method based on deep learning, which solves the technical problems of existing insulator defect detection, such as strong reliance on manual methods, low identification accuracy, poor generalization ability, and lack of self-optimization mechanism.

[0005] In a first aspect, the present invention provides an automatic insulator defect identification device based on deep learning, comprising:

[0006] Image acquisition module;

[0007] The data preprocessing module receives the output data from the image acquisition module and performs noise reduction, enhancement, and size normalization processing.

[0008] A deep learning model module, the input of which is connected to a data preprocessing module, performs defect feature extraction and classification on the preprocessed image;

[0009] The result analysis module has its input connected to the deep learning model module, outputting defect type, location coordinates, and confidence level, and generating model optimization parameters based on the recognition results.

[0010] The storage and display module is connected to the result analysis module and the deep learning model module respectively. It stores the raw data, recognition results and model parameters, and visualizes the defect information.

[0011] Furthermore, the image acquisition module includes an industrial camera, an adjustable light source, and a gimbal bracket. The industrial camera has a resolution of ≥5 million pixels and a frame rate of ≥15fps.

[0012] Furthermore, the data preprocessing module uses Gaussian filtering to remove image noise, enhances defect features through adaptive histogram equalization, and uniformly adjusts the image size to 512×512 pixels.

[0013] Furthermore, the deep learning model module includes a feature extraction network and a classification and recognition network. The feature extraction network adopts an improved ResNet-50 structure, and the classification and recognition network combines an attention mechanism. The types of defects that can be identified include cracks, damage, dirt accumulation, and aging peeling.

[0014] Furthermore, the result analysis module assesses the severity of the defect by calculating the ratio of the defect area to the total area of ​​the insulator. When the ratio is ≥5%, it is determined to be a serious defect. At the same time, it generates model parameter adjustment suggestions and inputs them back into the deep learning model module.

[0015] Secondly, the present invention provides an automatic insulator defect identification method based on deep learning, comprising the following steps:

[0016] S1: The image acquisition module acquires multi-view images of the insulator and transmits them to the data preprocessing module for denoising, enhancement and standardization to obtain training samples and samples to be tested;

[0017] S2: Input the labeled training samples into the deep learning model module, train the model through the backpropagation algorithm, optimize the network weight parameters, until the model accuracy is ≥95%;

[0018] S3: Input the preprocessed sample to be detected into the trained deep learning model module and output the defect identification result;

[0019] S4: The results analysis module performs quantitative analysis on the identification results, determines the severity of the defects, generates model optimization parameters based on the incorrectly identified samples, updates the deep learning model module in reverse, and the storage and display module synchronously records and visualizes the results.

[0020] This application discloses an automatic insulator defect identification device and method based on deep learning. It improves data quality through high-resolution image acquisition and multi-dimensional preprocessing, achieves accurate defect identification and localization using an improved deep learning model, and realizes reverse optimization of the model by combining a result analysis module. This significantly reduces reliance on manual labor, improves detection accuracy and generalization ability in complex environments, and provides effective technical support for intelligent detection of insulator defects. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the present invention, the drawings used in the following description of the embodiments will be briefly introduced. Obviously, the drawings described below are some embodiments of the present invention. For those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0022] Figure 1 This is a schematic diagram of the overall structure of an automatic insulator defect identification device based on deep learning, provided in an embodiment of the present invention. Detailed Implementation

[0023] To make the objectives, technical solutions, and advantages of the embodiments of the present invention clearer, the technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0024] like Figure 1 As shown, an automatic insulator defect identification device based on deep learning includes an image acquisition module 1, a data preprocessing module 2, a deep learning model module 3, a result analysis module 4, and a storage and display module 5. The image acquisition module 1 uses a 12-megapixel CMOS camera (model: Basler acA2440-75uc), a ring-shaped adjustable light source of 16 LEDs (color temperature 3000-6500K), and a motorized pan-tilt head supporting 0-360° rotation (accuracy ±0.5°). The data preprocessing module 2 uses an NVIDIA Jetson Xavier NX processor, running bilateral filtering and Retinex algorithms. The deep learning model module 3 is deployed on a server (configured with an Intel Xeon Gold 6348 CPU and an NVIDIA A100 GPU), with a feature extraction submodule using ResNet-50 with added attention mechanism and a recognition submodule using an improved YOLOv8. The result analysis module 4 and the storage and display module 5 are integrated into an industrial control computer (equipped with a 2TB SSD and a 27-inch touchscreen).

[0025] In this embodiment, the above-mentioned device is used to automatically identify defects in 100 porcelain insulators (including various defects). The specific steps are as follows:

[0026] S1: Image Acquisition and Preprocessing: Control the motorized pan-tilt unit to rotate once every 30°. Under the three levels of 3000K, 4500K, and 6500K of the ring-shaped adjustable light source, acquire 12 images (1200 images in total) for each insulator using an industrial camera. The images are then transmitted to the data preprocessing module 2. The noise reduction unit performs bilateral filtering with a 5×5 window, the enhancement unit improves the contrast using the Retinex algorithm, and the normalization unit adjusts the images to 512×512 pixels.

[0027] S2: Model Training: 1200 images were labeled (including 230 cracks, 180 breaks, 350 dirt accumulations, and 210 aging and peeling), and divided into a training set (840 images), a validation set (240 images), and a test set (120 images) in a 7:2:1 ratio; the deep learning model module 3 was input, the learning rate was set to 0.001, the batch size was 16, and the model was trained for 50 epochs. The validation set accuracy reached 96.3% on the 32nd epoch, training was stopped and the model was saved.

[0028] S3: Defect Identification: After preprocessing 120 images from the test set in S1, the images are input into the model. The identification submodule outputs the defect type and location. The accuracy rates for crack identification are 95.2%, breakage 97.3%, dirt accumulation 94.1%, and aging and peeling 96.7%, with an average confidence level of 0.89.

[0029] S4: Results Analysis and Model Optimization: The defect assessment unit identified 15 severe defects (≥5%), 32 moderate defects, and 68 minor defects. The model optimization unit analyzed 7 misidentified samples in the test set (3 cracks were misidentified as aging, and 4 contamination samples had a confidence level <0.8) and calculated the weight correction gradient. After accumulating 1000 misidentified samples, the model accuracy improved to 97.1%. The storage and display module 5 saves all data, and the touch screen displays defect markings and analysis results in real time.

[0030] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and not to limit them; although the present invention has been described in detail with reference to the foregoing embodiments, those skilled in the art should understand that modifications can still be made to the technical solutions described in the foregoing embodiments, or equivalent substitutions can be made to some of the technical features; and these modifications or substitutions do not cause the essence of the corresponding technical solutions to deviate from the spirit and scope of the technical solutions of the embodiments of the present invention.

Claims

1. An automatic insulator defect identification device based on deep learning, characterized in that, include: Image acquisition module (1); The data preprocessing module (2) receives the output data from the image acquisition module (1) and performs noise reduction, enhancement and size normalization processing. The deep learning model module (3) is connected to the data preprocessing module (2) at its input end, and performs defect feature extraction and classification recognition on the preprocessed image. The result analysis module (4) is connected to the deep learning model module (3) at its input end. It outputs the defect type, location coordinates and confidence level, and generates model optimization parameters based on the recognition results. The storage and display module (5) is connected to the result analysis module (4) and the deep learning model module (3) respectively, and stores the original data, recognition results and model parameters, while visually displaying the defect information.

2. The insulator defect automatic identification device based on deep learning according to claim 1, characterized in that, The image acquisition module (1) includes an industrial camera, an adjustable light source and a gimbal bracket. The industrial camera has a resolution of ≥5 million pixels and a frame rate of ≥15fps.

3. The insulator defect automatic identification device based on deep learning according to claim 1, characterized in that, The data preprocessing module (2) uses Gaussian filtering to remove image noise, enhances defect features through adaptive histogram equalization, and uniformly adjusts the image size to 512×512 pixels.

4. The insulator defect automatic identification device based on deep learning according to claim 1, characterized in that, The deep learning model module (3) includes a feature extraction network and a classification and recognition network. The feature extraction network adopts an improved ResNet-50 structure, and the classification and recognition network combines an attention mechanism. The types of defects that can be identified include cracks, damage, dirt accumulation, and aging and peeling.

5. The insulator defect automatic identification device based on deep learning according to claim 1, characterized in that, The result analysis module (4) assesses the severity of the defect by calculating the ratio of the defect area to the total area of ​​the insulator. When the ratio is ≥5%, it is judged as a serious defect. At the same time, it generates model parameter adjustment suggestions and inputs them back into the deep learning model module (3).

6. A method for automatic identification of insulator defects based on deep learning, characterized in that, Includes the following steps: S1: The image acquisition module (1) acquires multi-view images of the insulator and transmits them to the data preprocessing module (2) for denoising, enhancement and standardization to obtain training samples and samples to be tested; S2: Input the labeled training samples into the deep learning model module (3), train the model through the backpropagation algorithm, optimize the network weight parameters, until the model accuracy is ≥95%; S3: Input the preprocessed sample to be detected into the trained deep learning model module (3) and output the defect identification results (type, location, confidence level); S4: The result analysis module (4) performs quantitative analysis on the identification results, determines the severity of the defects, generates model optimization parameters based on the incorrect identification samples, updates the deep learning model module (3) in reverse, and the storage and display module (5) records and visualizes the results synchronously.

Images