Insulator rapid identification method based on insulator local curved surface features
The method for rapid insulator recognition using local surface features addresses environmental interference, enhancing efficiency and accuracy by leveraging color and glossiness with a neural network model adaptable to different conditions.
Patent Information
- Application Number
- CN202411026188.7
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-07-30
- Publication Date
- 2024-12-13
AI Technical Summary
Existing insulator image recognition methods are interfered by complex environments, leading to slower efficiency, longer recognition times, and reduced accuracy, especially for small or unclear defects.
A method for rapid insulator recognition based on local curved surface features, utilizing color and glossiness as distinctive features, with a feature matching algorithm and a neural network model that adjusts weights for different environments.
Improves recognition efficiency and accuracy by quickly locating insulators using local surface features, reducing processing time and computing resources while maintaining robustness across varying environments.
Smart Images
Figure CN119130902A_ABST
Abstract
Description
Technical Field
[0001] The invention relates to the technical field of image recognition, and in particular to a method for quickly identifying an insulator based on local curved surface features of the insulator. Background Art
[0002] Insulators are important supporting devices and insulation protection devices for power systems. In the digital measurement of insulator creepage distance, it is first necessary to perform image recognition on the insulators to determine their specific locations. The existing insulator layout environment is often complex. In the process of image recognition of insulators, they are often interfered by the environment, resulting in slower efficiency of insulator recognition, longer recognition time and lower recognition accuracy. At the same time, when there are small or unclear defects in the insulators, such as fine cracks or slight stains, traditional image recognition may also be difficult to detect accurately, or it may take longer to identify the details, resulting in a decrease in overall recognition efficiency. Summary of the invention
[0003] The purpose of the present invention is to overcome the shortcomings of the prior art that the process of image recognition of insulators is often interfered by the environment, resulting in slower efficiency of insulator recognition, longer recognition time and reduced recognition accuracy, and to provide a method for rapid insulator recognition based on the local surface features of the insulators.
[0004] The purpose of the present invention is achieved through the following technical solutions: A method for quickly identifying insulators based on local curved surface features of insulators comprises the following steps: Obtain typical features of the local surface of the insulator and construct a feature matching algorithm based on the typical features; Scanning a scene with insulators by an imaging device, extracting image features of the scanned image, matching the image features with typical features, and determining whether the image features match the typical features according to a feature matching algorithm; If the image feature matches the typical feature, it is determined that the scanned image corresponding to the image feature is a local surface of the insulator; The imaging device scans the image around the local curved surface of the insulator according to the local curved surface of the insulator to complete the image recognition of the insulator.
[0005] In this scheme, the local surface of the insulator is first identified, which can quickly locate the approximate position of the local surface of the insulator, and then the position of the insulator is located according to the position of the local surface of the insulator, and the image of the insulator can be recognized at this position. This scheme can quickly locate the existence of the insulator without the need for a detailed analysis of the entire image containing the insulator, which can significantly reduce processing time and computing resources and improve recognition efficiency. At the same time, since the local surface features of the insulator are usually unique, the recognition process is more accurate in this scheme. Compared with global features, local features (i.e., local surface features of the insulator) are not easily interfered by other elements in the image, so the insulator can be identified more accurately. Moreover, the recognition method of the local surface features of the insulator is easier to expand to the recognition of different types of insulators. As long as the insulators have similar local surface features, the feature matching algorithm can be adjusted to adapt to the recognition of different types of insulators.
[0006] Preferably, the typical features of the local surface of the insulator include color and glossiness, and a color feature matching algorithm is constructed based on color, and a glossiness feature matching algorithm is constructed based on glossiness. This solution selects color and glossiness as the features with greater differences among the local surface features for identification, and the identification accuracy is high.
[0007] Preferably, the color feature matching algorithm is specifically: Select HSV space to represent the color space, divide the color space into several small intervals, count the number of pixels in each interval, and get the color histogram. Use Bhattacharyya distance method to compare the similarity between image features and typical features. The Bhattacharyya distance formula is as follows: Among them, H1 is the color histogram of the typical feature, H2 is the color histogram of the image feature, N is the number of intervals of the histogram, and the smaller the Bhattacharyya distance is, the higher the similarity between the image feature and the typical feature is.
[0008] Preferably, the gloss map feature matching algorithm is specifically: The grayscale mean and variance of the image area are calculated as quantitative representations of glossiness, and the similarity between the image features and the typical features is judged by whether the values of the quantitative representations of glossiness between the image features and the typical features are within the threshold range.
[0009] Preferably, the matching of the image features with the typical features and judging whether the image features match the typical features according to the feature matching algorithm are specifically as follows: First, the Bhattacharyya distance method is used to compare the similarity of the color histogram between the image feature and the typical feature, and then the similarity of the gloss between the image feature and the typical feature is calculated. If the similarity of both color and gloss exceeds the set threshold, the image feature is considered to match the typical feature.
[0010] Preferably, the insulator fast identification method based on the local surface features of the insulator further makes the result of judging whether the image features match the typical features more accurate by adjusting the contribution of color and glossiness, specifically: Assign different initial weights to color and glossiness, which are randomly assigned or given by expert experience; Obtain image data with local surfaces of insulators, divide the image data into training sets and test sets, optimize the neural network algorithm model through the training set, the neural network algorithm model outputs the adjusted weight value, and the test set is used to verify whether the neural network algorithm model is optimized successfully; The weight values output by the neural network algorithm model are allocated to the color and glossiness, and the weight values represent the contribution of the color and glossiness; In the process of matching image features and typical features, the total similarity value is obtained by combining the similarity of color and glossiness with the corresponding contribution, and whether the image features and typical features match is determined based on the total similarity value.
[0011] Preferably, the neural network algorithm model is dynamically updated, and the neural network algorithm model is trained and updated based on the images of the surroundings of the insulator or similar environments that need to be identified as image data with local surfaces of the insulator, or the neural network algorithm model is trained and updated based on images of the environments with insulators taken at different times as image data with local surfaces of the insulator.
[0012] The design of this solution is aimed at the recognition of local surfaces of insulators under different external environments. A neural network algorithm model with different parameters is designed to enhance the recognition accuracy of local surfaces of insulators. The weight values of color and glossiness in different environments will be adjusted according to actual conditions.
[0013] The beneficial effect of the present invention is that the recognition of the local curved surface of the insulator by the present invention can quickly locate the approximate position of the local curved surface of the insulator, and then locate the position of the insulator according to the position of the local curved surface of the insulator, and the image recognition of the insulator can be performed at this position. The rapid recognition method based on the local curved surface features of the insulator has obvious advantages in improving recognition efficiency, accuracy and robustness, and has good scalability and real-time performance. BRIEF DESCRIPTION OF THE DRAWINGS
[0014] Figure 1 It is a flow chart of the present invention. DETAILED DESCRIPTION
[0015] Example embodiments will now be described more fully with reference to the accompanying drawings. However, example embodiments can be implemented in a variety of forms and should not be construed as limited to the examples set forth herein; rather, these embodiments are provided so that this application will be more comprehensive and complete and fully convey the concept of the example embodiments to those skilled in the art.
[0016] In addition, described feature, structure or characteristic can be combined in one or more embodiments in any suitable manner. In the following description, many specific details are provided to provide a full understanding of the embodiments of the present application. However, those skilled in the art will appreciate that the technical scheme of the present application can be put into practice without one or more of the specific details, or other methods, components, devices, steps, etc. can be adopted. In other cases, known methods, devices, realizations or operations are not shown or described in detail to avoid blurring the various aspects of the application.
[0017] The flowcharts shown in the accompanying drawings are only exemplary and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps can be decomposed, and some operations / steps can be combined or partially combined, so the actual execution order may change according to actual conditions.
[0018] Example: A fast insulator identification method based on the local surface features of the insulator, such as Figure 1 As shown, the following steps are included: Obtain typical features of the local surface of the insulator and construct a feature matching algorithm based on the typical features; Scanning a scene with insulators by an imaging device, extracting image features of the scanned image, matching the image features with typical features, and determining whether the image features match the typical features according to a feature matching algorithm; If the image feature matches the typical feature, it is determined that the scanned image corresponding to the image feature is a local surface of the insulator; The imaging device scans the image around the local curved surface of the insulator according to the local curved surface of the insulator to complete the image recognition of the insulator.
[0019] In this scheme, the local surface of the insulator is first identified, which can quickly locate the approximate position of the local surface of the insulator, and then the position of the insulator is located according to the position of the local surface of the insulator, and the image of the insulator can be recognized at this position. This scheme can quickly locate the existence of the insulator without the need for a detailed analysis of the entire image containing the insulator, which can significantly reduce processing time and computing resources and improve recognition efficiency. At the same time, since the local surface features of the insulator are usually unique, the recognition process is more accurate in this scheme. Compared with global features, local features (i.e., local surface features of the insulator) are not easily interfered by other elements in the image, so the insulator can be identified more accurately. Moreover, the recognition method of the local surface features of the insulator is easier to expand to the recognition of different types of insulators. As long as the insulators have similar local surface features, the feature matching algorithm can be adjusted to adapt to the recognition of different types of insulators.
[0020] The typical features of the local surface of the insulator include color and glossiness. A color feature matching algorithm is constructed based on color, and a glossiness feature matching algorithm is constructed based on glossiness. This solution selects color and glossiness as the most distinctive features of the local surface features for recognition, and the recognition accuracy is high.
[0021] Preferably, the color feature matching algorithm is specifically: Select HSV space to represent the color space, divide the color space into several small intervals, count the number of pixels in each interval, and get the color histogram. Use Bhattacharyya distance method to compare the similarity between image features and typical features. The Bhattacharyya distance formula is as follows: Among them, H1 is the color histogram of the typical feature, H2 is the color histogram of the image feature, N is the number of intervals of the histogram, and the smaller the Bhattacharyya distance is, the higher the similarity between the image feature and the typical feature is.
[0022] The gloss map feature matching algorithm is specifically as follows: The grayscale mean and variance of the image area are calculated as quantitative representations of glossiness, and the similarity between the image features and the typical features is judged by whether the values of the quantitative representations of glossiness between the image features and the typical features are within the threshold range.
[0023] The matching of the image features with the typical features and judging whether the image features match the typical features according to the feature matching algorithm are specifically as follows: First, the Bhattacharyya distance method is used to compare the similarity of the color histogram between the image feature and the typical feature, and then the similarity of the gloss between the image feature and the typical feature is calculated. If the similarity of both color and gloss exceeds the set threshold, the image feature is considered to match the typical feature.
[0024] The insulator fast identification method based on the local surface features of the insulator also makes the result of judging whether the image features match the typical features more accurate by adjusting the contribution of color and glossiness, specifically: Assign different initial weights to color and glossiness, which are randomly assigned or given by expert experience; Obtain image data with local surfaces of insulators, divide the image data into training sets and test sets, optimize the neural network algorithm model through the training set, the neural network algorithm model outputs the adjusted weight value, and the test set is used to verify whether the neural network algorithm model is optimized successfully; The weight values output by the neural network algorithm model are allocated to the color and glossiness, and the weight values represent the contribution of the color and glossiness; In the process of matching image features and typical features, the total similarity value is obtained by combining the similarity of color and glossiness with the corresponding contribution, and whether the image features and typical features match is determined based on the total similarity value.
[0025] In this embodiment, the process of building and optimizing the neural network algorithm model is as follows: Initialize network parameters: Before training begins, the network's weights and bias parameters need to be initialized. These parameters are set randomly.
[0026] Training phase: Forward propagation: Pass the input data through the network, calculate the output of each neuron according to the current weights and biases, until the final output of the network is obtained.
[0027] Calculate the loss function: Compare the output of the network with the true label and calculate the loss function (such as cross entropy loss function, mean square error loss function, etc.).
[0028] Back propagation: Based on the error calculated by the loss function, the error is propagated back layer by layer through the back propagation algorithm to update the weights and biases in the network.
[0029] Parameter update: Use optimization algorithms (such as gradient descent, Adam, etc.) to update network parameters according to the gradient information obtained by back propagation to minimize the loss function.
[0030] Iterative Optimization: Repeat the above training process and continuously optimize the network parameters through multiple epochs (training rounds) until the loss function converges or reaches the preset training rounds.
[0031] Reasoning stage: After training is complete, the trained network is used to make predictions on new input data.
[0032] During the inference phase, the network parameters remain unchanged and only forward propagation is performed to obtain the prediction results.
[0033] The neural network algorithm model is dynamically updated, and the neural network algorithm model is trained and updated based on the environmental images around the insulator or similar environments that need to be identified as image data with local surfaces of the insulator, or the neural network algorithm model is trained and updated based on environmental images with insulators taken at different times as image data with local surfaces of the insulator.
[0034] The design of this solution is aimed at the recognition of the local curved surface of the insulator under different external environments. The neural network algorithm model with different parameters is designed to enhance the recognition accuracy of the local curved surface of the insulator. The weight values of color and glossiness in different environments will be adjusted according to the actual situation. In this embodiment, different environments include situations where the background is mostly mountainous, the background is mostly plain, or the background is mostly houses. In different situations, the difference between the color of the local curved surface of the insulator and the background color is different. Therefore, the color weight setting in different environments is obviously different. If it is necessary to recognize the local curved surface of the insulator under a specific external environment, it is necessary to input image data of similar external environment to train and optimize the neural network algorithm model and adjust the weight. Similarly, when shooting at different times, due to the influence of sunlight or other factors, the change of glossiness will also vary greatly over time. Therefore, the neural network algorithm model can also be optimized through image data at different times.
[0035] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the embodiments disclosed herein. The present application is intended to cover any variations, uses or adaptations of the present application, which follow the general principles of the present application and include common knowledge or customary technical means in the art that are not disclosed in the present application.
[0036] It should be understood that the present application is not limited to the precise structures that have been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof. The scope of the present application is limited only by the appended claims.
Claims
1. A fast insulator identification method based on the local surface features of insulators. Its characteristics are: The following steps are involved: Obtain typical features of the local surface of the insulator and construct a feature matching algorithm based on the typical features; Scanning a scene with insulators by an imaging device, extracting image features of the scanned image, matching the image features with typical features, and determining whether the image features match the typical features according to a feature matching algorithm; If the image feature matches the typical feature, it is determined that the scanned image corresponding to the image feature is a local surface of the insulator; The imaging device scans the image around the local curved surface of the insulator according to the local curved surface of the insulator to complete the image recognition of the insulator.
2. A method for rapid identification of insulators based on local curved surface features of insulators according to claim 1, Its characteristics are: The typical features of the local curved surface of the insulator include color and glossiness. A color feature matching algorithm is constructed according to the color, and a glossiness feature matching algorithm is constructed according to the glossiness.
3. A method for rapid identification of insulators based on local curved surface features of insulators according to claim 2, Its characteristics are: The color feature matching algorithm is specifically as follows: Select HSV space to represent the color space, divide the color space into several small intervals, count the number of pixels in each interval, and get the color histogram. Use Bhattacharyya distance method to compare the similarity between image features and typical features. The Bhattacharyya distance formula is as follows: Among them, H 1 is the color histogram of typical features, H 2 is the color histogram of the image feature, N is the number of intervals of the histogram, and the smaller the Bhattacharyya distance is, the higher the similarity between the image feature and the typical feature is.
4. A method for quickly identifying insulators based on local curved surface features of insulators according to claim 3, Its characteristics are: The gloss map feature matching algorithm is specifically as follows: The grayscale mean and variance of the image area are calculated as quantitative representations of glossiness, and the similarity between the image features and the typical features is judged by whether the values of the quantitative representations of glossiness between the image features and the typical features are within the threshold range.
5. The method for quickly identifying insulators based on local curved surface features of insulators according to claim 4, Its characteristics are: The matching of the image features with the typical features and judging whether the image features match the typical features according to the feature matching algorithm are specifically as follows: First, the Bhattacharyya distance method is used to compare the similarity of the color histogram between the image feature and the typical feature, and then the similarity of the gloss between the image feature and the typical feature is calculated. If the similarity of both color and gloss exceeds the set threshold, the image feature is considered to match the typical feature.
6. The method for rapid identification of insulators based on local curved surface features of insulators according to claim 2, Its characteristics are: The contribution of color and glossiness is also adjusted to make the result of judging whether the image features match the typical features more accurate, specifically: Assign different initial weights to color and glossiness, which are randomly assigned or given by expert experience; Obtain image data with local surfaces of insulators, divide the image data into training sets and test sets, optimize the neural network algorithm model through the training set, the neural network algorithm model outputs the adjusted weight value, and the test set is used to verify whether the neural network algorithm model is optimized successfully; The weight values output by the neural network algorithm model are allocated to the color and glossiness, and the weight values represent the contribution of the color and glossiness; In the process of matching image features and typical features, the total similarity value is obtained by combining the similarity of color and glossiness with the corresponding contribution, and whether the image features and typical features match is determined based on the total similarity value.
7. The method for quickly identifying insulators based on local curved surface features of insulators according to claim 6, Its characteristics are: The neural network algorithm model is dynamically updated, and the neural network algorithm model is trained and updated based on the environmental images around the insulator or similar environments that need to be identified as image data with local surfaces of the insulator, or the neural network algorithm model is trained and updated based on environmental images with insulators taken at different times as image data with local surfaces of the insulator.