Power material defect screening method and system based on image and voiceprint technology

Abstract

The present application relates to the technical field of image processing, in particular to a power material defect screening method and system based on image and voiceprint technology. The method comprises the following steps: constructing a training set; training a neural network model using the training set to obtain a defect screening model; classifying the candidate boxes and establishing a Gaussian model for the features of the running sound corresponding to the images in which the candidate boxes are located; collecting target images and running sound of the power material to obtain the probability of each Gaussian model; screening the target Gaussian model, taking the quotient of the candidate box size with the highest recognition accuracy and the average size corresponding to the target Gaussian model as a proportion factor, and multiplying the target image by the proportion factor to obtain a transformed image; inputting the transformed image into the defect screening model to obtain a defect detection result. The method of the present application can greatly improve the accuracy and recognition efficiency of the power material defect recognition result and improve the automation degree of the power material defect detection.

Description

Technical Field

[0001] This invention relates to the field of image processing technology. More specifically, this invention relates to a method and system for screening defects in power materials based on image and voiceprint technology. Background Technology

[0002] The power system is the infrastructure of modern society, and its reliability and security directly affect the normal operation of society and economic development. Defects in any electrical equipment can lead to serious power outages, even large-scale blackouts. Traditional methods of detecting defects in electrical equipment mainly rely on manual inspection and simple instrument testing. These methods are often inefficient, inaccurate, and susceptible to human error, making it difficult to meet the high reliability and high efficiency requirements of modern power systems. With the development of image processing and voiceprint analysis technologies, their application in the detection of defects in electrical equipment has become possible.

[0003] In existing technologies, defect detection of power equipment typically involves acquiring images of the equipment and using image processing techniques combined with defect detection models to identify defects on its outer surface. However, because there are many defect areas on the outer surface of power equipment, and these defect areas vary in size, the accuracy of defect detection models differs for defects of different sizes. Errors may occur when identifying defects of smaller sizes with lower accuracy. For example, the electronic transformer defect detection method disclosed in Chinese patent application CN115950887A uses acquired image data of the electronic transformer combined with a defect detection model to detect the type of defect. However, this method does not consider the difference in accuracy of the defect detection model for defects of different sizes, leading to certain errors in transformer defect detection. Summary of the Invention

[0004] To address the technical problem of errors in the existing technology of using image processing technology and neural network models for defect detection of power materials, the present invention provides solutions in the following aspects.

[0005] In a first aspect, the present invention provides a method for screening defects in power materials based on image and voiceprint technology, comprising:

[0006] Historical data, including images of electrical equipment, is acquired. Candidate boxes are used to label the location and type of defects in the images to form a training set. A neural network model is trained using the training set to obtain a defect screening model. The electrical equipment includes transformers.

[0007] The historical data also includes the running sound corresponding to the image; the candidate boxes are classified, candidate boxes with similar sizes are grouped into the same category, the average size of the candidate boxes in each category is calculated, and Gaussian models are established for the features of the running sound corresponding to the image of each type of candidate box.

[0008] Collect target images and operating sounds of power materials, extract features of the operating sounds, and input the features into each Gaussian model to obtain the probability of belonging to each Gaussian model;

[0009] Target Gaussian models with probabilities greater than a preset probability threshold are selected. The quotient of the candidate box size with the highest recognition accuracy of the defect screening model and the average size corresponding to the target Gaussian model is used as a scaling factor. The target image is multiplied by the scaling factor to obtain a transformed image. If the size of the transformed image is larger than the target image, the transformed image is input into the defect screening model to obtain the defect detection result.

[0010] Its beneficial effects are as follows: image processing technology can efficiently identify external defects, while acoustic signature analysis technology can detect internal mechanical and electrical anomalies in equipment. The power equipment defect screening method based on image and acoustic signature technology of this invention, when identifying defects in power equipment, not only identifies defects based on the acquired images, but also further considers the difference between the operating sound of power equipment with defects and the operating sound of power equipment under normal conditions. By identifying the approximate size of the candidate box corresponding to the defect location through the characteristics of the operating sound of the power equipment, the acquired image is scaled so that the size of the defect location in the image is close to the optimal size for the defect screening model, thereby greatly improving the accuracy of the defect identification results. By using a neural network model to automatically identify defects in power equipment instead of manual identification, the defect detection speed is improved, and labor costs and time consumption are reduced. Combining high-resolution image processing technology and accurate acoustic signature analysis, potential defects in power equipment can be detected earlier and more accurately, thus preventing small problems from escalating into major failures. Because a neural network model is used, the defect detection algorithm can be continuously optimized, improving the intelligence level of the detection system.

[0011] In one embodiment, a clustering algorithm is used to classify the candidate boxes, including: selecting several candidate boxes as initial cluster centers during clustering, and using the difference between the size of a candidate box and the size of the candidate box at the cluster center as the distance from the candidate box to the cluster center, the calculation expression of which is:

[0012]

[0013] Where L represents the distance between the candidate box and the j-th cluster center, h i w represents the height of the i-th candidate box. ih represents the width of the i-th candidate box. j w represents the height of the candidate box for the j-th cluster center. j This represents the width of the candidate box for the j-th cluster center.

[0014] Its beneficial effects are: by using clustering algorithms to classify candidate boxes, the classification efficiency and accuracy can be greatly improved, thereby improving the efficiency and accuracy of screening for defects in power materials.

[0015] In one embodiment, the characteristics of the sound include the vibration frequency and vibration amplitude.

[0016] Its beneficial effects are as follows: For different defects, the vibration frequency and vibration amplitude of the operating sound of electrical materials are different. When constructing the Gaussian model, the vibration frequency and vibration amplitude of the operating sound are taken into account, so that the candidate box category of the defect corresponding to the operating sound can be identified more accurately.

[0017] In one embodiment, building a Gaussian model for the features of the running sound corresponding to the image containing a certain type of candidate box includes:

[0018] The image set is defined as the entire set of images containing each candidate box in this category.

[0019] Fourier transform or wavelet transform is performed on the sound of the electrical materials corresponding to each image in the image set to obtain the vibration frequency and vibration amplitude of the operating sound of each image.

[0020] Gaussian models were established for all obtained vibration frequencies and amplitudes.

[0021] In one embodiment, the probability threshold is set to 0.4.

[0022] In one embodiment, the method for obtaining the candidate box size with the highest recognition accuracy of the defect screening model is as follows: input the image of each candidate box in each type of candidate box into the defect screening model to obtain the defect recognition result, calculate the recognition accuracy of each type of candidate box based on the recognition result, and select the average size of the candidate box with the highest recognition accuracy as the candidate box size with the highest recognition accuracy of the defect screening model.

[0023] Its beneficial effects are as follows: by calculating the defect recognition accuracy corresponding to each type of candidate box separately, and taking the average size of the candidate box of the category with the highest recognition accuracy as the candidate box size with the highest recognition accuracy of the power material defect screening model, the candidate box size with the highest recognition accuracy can be obtained quickly and accurately.

[0024] In one embodiment, the method further includes: in response to the size of the transformed image being smaller than the target image, cropping the target image into multiple sub-images, wherein the size of each sub-image is equal to the size of the candidate box with the highest recognition accuracy of the filtering model, and inputting the cropped sub-images into the filtering model to obtain a defect detection result.

[0025] In one embodiment, cropping the target image into multiple sub-images includes:

[0026] A sliding window of a preset size is set, and it slides from the top left of the image according to a preset step size to obtain the image corresponding to the sliding window at each position. The size of the sliding window is equal to the size of the candidate box with the highest recognition accuracy of the filtering model.

[0027] In one embodiment, multiple cropped sub-images are input into the filtering model, which outputs the defect type corresponding to each sub-image. The defect type with the highest frequency among all the defect types output by the filtering model is taken as the defect detection result.

[0028] Its beneficial effects are as follows: different sizes of images input into the defect screening model may lead to different defect detection results, and the detection results corresponding to the same image may include multiple defect types; by taking the most frequently occurring defect as the defect detection result of power materials, the accuracy of the detection results can be improved.

[0029] In a second aspect, the present invention provides a power material defect screening system based on image and voiceprint technology, comprising a memory and a processor, wherein the memory stores computer program instructions, and the computer program instructions, when executed by the processor, implement the power material defect screening method based on image and voiceprint technology of the present invention. Attached Figure Description

[0030] The above and other objects, features, and advantages of exemplary embodiments of the present invention will become readily apparent upon reading the following detailed description with reference to the accompanying drawings. In the drawings, several embodiments of the invention are illustrated by way of example and not limitation, and like or corresponding reference numerals denote like or corresponding parts, wherein:

[0031] Figure 1 This is a flowchart illustrating an embodiment of the power material defect screening method based on image and voiceprint technology according to an embodiment of the present invention;

[0032] Figure 2 This is a flowchart illustrating an embodiment of the method for establishing a Gaussian model according to the present invention;

[0033] Figure 3 This is a schematic diagram illustrating the target image size of an embodiment of the present invention;

[0034] Figure 4 This is a schematic diagram illustrating the structure of a power material defect screening system based on image and voiceprint technology, according to an embodiment of the present invention. Detailed Implementation

[0035] 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, not all, of the embodiments of the present invention. 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.

[0036] The specific embodiments of the present invention will now be described in detail with reference to the accompanying drawings.

[0037] Example of a method for screening defects in power equipment based on image and voiceprint technology:

[0038] like Figure 1 As shown, the power material defect screening model training method of the present invention includes:

[0039] S101. Training the power material defect screening model, specifically: acquiring historical data, including images of power materials, using candidate boxes to label the defect locations and types in the images, thus forming a training set; using the training set to train a neural network model to obtain the power material defect screening model; power materials include transformers.

[0040] Images of electrical equipment from multiple historical moments include images of equipment with various defects and images of equipment without defects. Images of equipment without defects can be labeled as defect-free, and there is no need to use candidate boxes to mark the defect location. An image of a piece of electrical equipment with defects indicates that only one type of defect exists. Electrical equipment can be transformers, relay protection devices, or other electrical equipment. If the equipment is a transformer, the defect types include bushing faults and insulator faults. All images of electrical equipment were taken using the same camera, and the shooting angle and distance between the camera and the equipment were the same at the time of shooting.

[0041] For insulators, after long-term operation of the transformer, their surface may become dirty, and the insulation performance of the insulators will decrease, which will lead to abnormal operation of the transformer. For bushings, defects such as deformation, scratches, bubbles and cracks may occur, which will affect their insulation and protection performance and lead to abnormal operation of the transformer.

[0042] In one embodiment, when using candidate boxes to mark the location of defects in an image, the candidate boxes are used to select the defective parts of the transformer and text content corresponding to the defect type is added to the candidate boxes, such as the sleeve failure of the transformer. This makes it easier to select the entire sleeve area in the image with the candidate boxes, and the size of the candidate boxes is close to the size of the sleeve area.

[0043] During training, the model uses cross-entropy loss as its loss function and accuracy as its evaluation metric. The optimal model is selected based on the evaluation metric.

[0044] S102. Establish a Gaussian model, specifically: historical data also includes the running sound corresponding to the image; classify the candidate boxes, group candidate boxes with similar sizes into the same category, calculate the average size of the candidate boxes in each category, and establish a Gaussian model for the features of the running sound corresponding to the image of each category of candidate boxes.

[0045] The operating sound corresponding to the image refers to the sound emitted by the corresponding power equipment during operation when the image was captured.

[0046] The input to the Gaussian model is the features of the sound, and the output is the probability that the features of the sound belong to the Gaussian model.

[0047] There are various ways to classify candidate boxes. In this embodiment, the minimum width, maximum width, minimum height, and maximum height of all candidate boxes can be counted, and the candidate box sizes can be divided into several size ranges. For example, assuming the minimum and maximum widths are 5 and 10 respectively, and the minimum and maximum heights are 3 and 5 respectively, then three size ranges can be defined: a width range of 5 to 6 and a height of 3; a width range of 7 to 8 and a height of 4; and a width range of 9 to 10 and a height of 5. In other embodiments, clustering algorithms can also be used to classify candidate boxes.

[0048] S103. Obtain the probability of the operating sound belonging to each Gaussian model. Specifically, collect the target image and operating sound of the power materials, extract the features of the operating sound, and input the features into each Gaussian model to obtain the probability of it belonging to each Gaussian model.

[0049] S104. Obtain the defect detection results, specifically: select target Gaussian models with a probability greater than a preset probability threshold, use the quotient of the candidate box size with the highest recognition accuracy of the defect screening model and the average size corresponding to the target Gaussian model as a scaling factor, multiply the target image by the scaling factor to obtain the transformed image; if the size of the transformed image is larger than the target image, input the transformed image into the power material defect screening model to obtain the defect detection results.

[0050] In this embodiment, the probability threshold is set to 0.4, but other suitable values ​​may be used in other embodiments.

[0051] If there are multiple Gaussian models with a probability greater than the preset probability threshold, the acquired images are resized according to the average size of each Gaussian model to obtain resized images of different sizes. The resized images are then input into the defect screening model, and the defect with the highest frequency in the detection results is taken as the defect detection result of the power materials.

[0052] Different image sizes input into the power material defect screening model may lead to different defect detection results, and the detection results corresponding to the same image may include multiple defect types; by using the most frequently occurring defect as the power material defect detection result, the accuracy of the detection results can be improved.

[0053] The expression for calculating the size of a transformed image is:

[0054]

[0055] Among them, new scale Indicates the size of the transformed image, pre scale The scale represents the size of the target image. maxp This represents the size of the candidate box that achieves the highest recognition accuracy in the power equipment defect screening model. This represents the average size corresponding to the target Gaussian model.

[0056] In this embodiment, if the size of the transformed image is smaller than the size of the corresponding target image, the transformed image can be directly input into the defect screening model to obtain the defect detection result. In other embodiments, if the size of the transformed image is smaller than the size of the corresponding target image, the target image can be cropped into multiple sub-images, each sub-image having a size equal to the size of the candidate box with the highest recognition accuracy in the defect screening model. The cropped sub-images are then input into the power material defect screening model to obtain the defect detection result. The combined image information of all sub-images must cover all image information in the image before cropping.

[0057] Image processing technology can efficiently identify external defects, while acoustic signature analysis can detect internal mechanical and electrical anomalies in equipment. This invention's power equipment defect screening method, based on image and acoustic signature technologies, not only identifies defects based on acquired images but also considers the difference between the operating sound of defective power equipment and its normal operating sound. By identifying the characteristics of the operating sound, the approximate size of the candidate box corresponding to the defect location is determined. Then, the acquired image is scaled to make the size of the defect location in the image close to the optimal size for the defect screening model, thus greatly improving the accuracy of defect identification results. By using a neural network model to automatically identify defects in power equipment instead of manual identification, the defect detection speed is increased, reducing labor costs and time consumption. Combining high-resolution image processing technology and precise acoustic signature analysis, potential defects in power equipment can be detected earlier and more accurately, thus preventing small problems from escalating into major failures. Because a neural network model is used, the defect detection algorithm can be continuously optimized, improving the intelligence level of the detection system.

[0058] As can be seen from the above embodiments, clustering algorithms or other methods can be used to classify candidate boxes. In one embodiment, a clustering algorithm is used to classify candidate boxes, including: during clustering, selecting several candidate boxes as initial cluster centers, and using the difference between the size of a candidate box and the size of the candidate box at the cluster center as the distance from the candidate box to the cluster center, the calculation expression of which is:

[0059]

[0060] Where L represents the distance between the candidate box and the j-th cluster center, h i w represents the height of the i-th candidate box. i h represents the width of the i-th candidate box. j w represents the height of the candidate box for the j-th cluster center. j This represents the width of the candidate box for the j-th cluster center.

[0061] In this embodiment, the K-Means clustering algorithm can be used, and other clustering algorithms can also be used in other embodiments.

[0062] By using clustering algorithms to classify candidate boxes, classification efficiency and accuracy can be greatly improved, thereby increasing the efficiency and accuracy of defect screening for power materials.

[0063] In one embodiment, the characteristics of sound include the vibration frequency and vibration amplitude.

[0064] For different defects, the vibration frequency and amplitude of the operating sound of electrical materials are different. When constructing the Gaussian model, the vibration frequency and amplitude of the operating sound are taken into account, so that the candidate box category of the defect corresponding to the operating sound can be identified more accurately.

[0065] like Figure 2 As shown, in one embodiment, establishing a Gaussian model for the features of the running sound corresponding to the image containing a certain type of candidate box includes:

[0066] S201. The whole set of images containing each candidate box in this type of candidate box is called the image set.

[0067] S202. Process the running sound to obtain the vibration frequency and vibration amplitude of the running sound. Specifically, perform Fourier transform or wavelet transform on the running sound corresponding to each image in the image set to obtain the vibration frequency and vibration amplitude of the running sound of each image.

[0068] S203. Establish a Gaussian model for all obtained vibration frequencies and vibration amplitudes.

[0069] For example, if this type of candidate box includes three candidate boxes, and the images containing these three candidate boxes are image A, image B, and image C, then the image set is a set composed of image A, image B, and image C. Fourier transforms or wavelet transforms are performed on the running sounds corresponding to image A, image B, and image C respectively to obtain the vibration frequencies and amplitudes of the running sounds in image A, image B, and image C; a Gaussian model is then established for all the obtained vibration frequencies and amplitudes.

[0070] In one embodiment, the method for obtaining the candidate box size with the highest recognition accuracy of the defect screening model is as follows: input the image of each candidate box in each type of candidate box into the defect screening model to obtain the defect recognition result, calculate the recognition accuracy of each type of candidate box based on the recognition result, and select the average size of the candidate box with the highest recognition accuracy as the candidate box size with the highest recognition accuracy of the defect screening model.

[0071] Assume there are three types of candidate boxes: candidate boxes I, candidate boxes II, and candidate boxes III. Candidate boxes I include candidate boxes D and E; candidate boxes II include candidate boxes F and G; and candidate boxes III include candidate boxes H and I. The images corresponding to candidate boxes D, E, F, G, H, and I are input into the defect screening model. If the defect identification result for the image corresponding to candidate box D is incorrect, the defect identification result for the image corresponding to candidate box G is incorrect, and the defect identification results for the images corresponding to the other candidate boxes are correct, then the identification accuracy for defects corresponding to candidate boxes I is 50%, the identification accuracy for defects corresponding to candidate boxes II is 50%, and the identification accuracy for defects corresponding to candidate boxes III is 100%. The average size of candidate boxes H and I is then calculated as the candidate box size with the highest identification accuracy in the power material defect screening model.

[0072] By calculating the defect recognition accuracy for each type of candidate box separately, and taking the average size of the candidate boxes of the category with the highest recognition accuracy as the candidate box size with the highest recognition accuracy in the power material defect screening model, the candidate box size with the highest recognition accuracy can be obtained quickly and accurately.

[0073] In one embodiment, cropping a target image into multiple sub-images includes: setting a sliding window of a preset size, sliding it from the top left of the image according to a preset step size, and acquiring the image corresponding to the sliding window at each position. The size of the sliding window is equal to the size of the candidate box with the highest recognition accuracy of the power material defect screening model. For example, assuming the preset step size is 10 pixels and the size of the sliding window is 20×20, the initial position of the sliding window is an area with a width and height of 20 pixels, including the top left corner of the image. The sliding window first slides to the right according to the preset step size, then slides down one step size when it reaches the rightmost side of the image, and then slides to the left; when it reaches the leftmost side of the image, it slides down one step size, and then slides to the right; whenever the sliding window slides to the left or right edge of the image, it slides down one step size, and then slides horizontally; until the sliding window has traversed all pixels of the image. The sliding path of the sliding window is an S-shaped path.

[0074] By using a sliding window to crop the target image, the cropped sub-image can be guaranteed to contain all the image information of the target image while maintaining the optimal size of the sub-image.

[0075] In another embodiment, if the target image can be segmented into several sub-images whose size is equal to the size of the candidate bounding box with the highest recognition accuracy of the defect screening model, then there is no need to set a sliding window, and the target image can be segmented directly. For example... Figure 3As shown, if the width and height of the target image are twice the width and height of the candidate box with the highest recognition accuracy of the defect screening model, then the target image can be cropped from the center point along the horizontal and vertical directions to obtain four sub-images: region a, region b, region c, and region d.

[0076] In one embodiment, multiple cropped sub-images are input into a power material defect screening model. The screening model outputs the defect type corresponding to each sub-image, and the defect type with the highest frequency among all defect types output by the screening model is taken as the defect detection result. The calculation expression for the defect detection result is:

[0077]

[0078] Where pre represents the defect detection result, Q t Q represents the number of times the cropped image of electrical materials is input into the model to predict a defect type t. 总 This indicates the total number of cropped images of electrical materials. express The value of t when the maximum value is reached.

[0079] Different image sizes input into the power material defect screening model may lead to different defect detection results, and the detection results corresponding to the same image may include multiple defect types; by using the most frequently occurring defect as the power material defect detection result, the accuracy of the detection results can be improved.

[0080] Example of a power equipment defect screening system based on image and voiceprint technology:

[0081] This invention also provides a power material defect screening system based on image and voiceprint technology. For example... Figure 4 As shown, the power material defect screening system based on image and voiceprint technology includes a processor and a memory. The memory stores computer program instructions, which, when executed by the processor, implement the power material defect screening method based on image and voiceprint technology according to the first aspect of the present invention.

[0082] The power material defect screening system based on image and voiceprint technology also includes other components well known to those skilled in the art, such as communication buses and communication interfaces. Their settings and functions are known in the art and will not be described in detail here.

[0083] In this invention, the aforementioned memory can be any tangible medium containing or storing a program that can be used or combined with an instruction execution system, apparatus, or device. For example, a computer-readable storage medium can be any suitable magnetic or magneto-optical storage medium, such as Resistive Random Access Memory (RRAM), Dynamic Random Access Memory (DRAM), Static Random Access Memory (SRAM), Enhanced Dynamic Random Access Memory (EDRAM), High-Bandwidth Memory (HBM), Hybrid Memory Cube (HMC), etc., or any other medium that can be used to store desired information and can be accessed by an application, module, or both. Any such computer storage medium can be part of a device or accessible to or connected to a device. Any application or module described in this invention can be implemented using computer-readable / executable instructions that can be stored or otherwise maintained by such a computer-readable medium.

[0084] In the description of this specification, "multiple" or "several" means at least two, such as two, three or more, unless otherwise explicitly specified.

[0085] While this specification has shown and described numerous embodiments of the invention, it will be apparent to those skilled in the art that such embodiments are provided by way of example only. Many modifications, alterations, and alternatives will occur to those skilled in the art without departing from the spirit and essence of the invention. It should be understood that various alternatives to the embodiments of the invention described herein may be employed in the practice of this invention.

Claims

1. A method for screening defects in power materials based on image and voiceprint technology, characterized in that, include: Historical data, including images of power equipment, was acquired, and candidate boxes were used to label the location and type of defects in the images, thus forming a training set. A neural network model is trained using the training set to obtain a defect screening model; the power materials include transformers; The historical data also includes the running sound corresponding to the image; the candidate boxes are classified, candidate boxes with similar sizes are grouped into the same category, the average size of the candidate boxes in each category is calculated, and Gaussian models are established for the features of the running sound corresponding to the image of each type of candidate box. Collect target images and operating sounds of power materials, extract features of the operating sounds, and input the features into each Gaussian model to obtain the probability of belonging to each Gaussian model; Target Gaussian models with probabilities greater than a preset probability threshold are selected. The quotient of the candidate box size with the highest recognition accuracy of the defect screening model and the average size corresponding to the target Gaussian model is used as a scaling factor. The target image is multiplied by the scaling factor to obtain the transformed image. If the size of the transformed image is larger than the target image, the transformed image is input into the defect screening model to obtain the defect detection result; The method for obtaining the candidate box size with the highest recognition accuracy of the defect screening model is as follows: input the image of each candidate box in each type of candidate box into the defect screening model to obtain the defect recognition result, calculate the recognition accuracy of each type of candidate box based on the recognition result, and select the average size of the candidate box with the highest recognition accuracy as the candidate box size with the highest recognition accuracy of the defect screening model.

2. The method for screening defects in power materials based on image and voiceprint technology as described in claim 1, characterized in that, The candidate boxes are classified using a clustering algorithm, including: selecting several candidate boxes as initial cluster centers, and using the difference between the size of a candidate box and the size of the candidate box at the cluster center as the distance from the candidate box to the cluster center, the calculation expression of which is: ； in, This represents the distance between the candidate box and the j-th cluster center. Indicates the height of the i-th candidate box. This represents the width of the i-th candidate box. This represents the height of the candidate box for the j-th cluster center. This represents the width of the candidate box for the j-th cluster center.

3. The method for screening defects in power materials based on image and voiceprint technology as described in claim 1, characterized in that, The characteristics of the sound include the vibration frequency and vibration amplitude.

4. The method for screening defects in power materials based on image and voiceprint technology as described in claim 2, characterized in that, Building a Gaussian model for the features of the running sound corresponding to the image containing a certain type of candidate box includes: The image set is defined as the entire set of images containing each candidate box in this category. Fourier transform or wavelet transform is performed on the sound of the electrical materials corresponding to each image in the image set to obtain the vibration frequency and vibration amplitude of the operating sound of each image. Gaussian models were established for all obtained vibration frequencies and amplitudes.

5. The method for screening defects in power materials based on image and voiceprint technology as described in claim 1, characterized in that, The probability threshold is set to 0.

4.

6. The method for screening defects in power materials based on image and voiceprint technology as described in any one of claims 1 to 5, characterized in that, Also includes: In response to the fact that the size of the transformed image is smaller than the target image, the target image is cropped into multiple sub-images. The size of each sub-image is equal to the size of the candidate box with the highest recognition accuracy of the screening model. The cropped sub-images are then input into the screening model to obtain the defect detection result.

7. The method for screening defects in power materials based on image and voiceprint technology as described in claim 6, characterized in that, The step of cutting the target image into multiple sub-images includes: A sliding window of a preset size is set, and it slides from the top left of the image according to a preset step size to obtain the image corresponding to the sliding window at each position. The size of the sliding window is equal to the size of the candidate box with the highest recognition accuracy of the filtering model.

8. The method for screening defects in power materials based on image and voiceprint technology as described in claim 6, characterized in that, Multiple cropped sub-images are input into the filtering model, which outputs the defect type corresponding to each sub-image. The defect type with the highest frequency among all the defect types output by the filtering model is taken as the defect detection result.

9. A power material defect screening system based on image and voiceprint technology, comprising a memory and a processor, wherein the memory stores computer program instructions, characterized in that, When the computer program instructions are executed by the processor, they implement the power material defect screening method based on image and voiceprint technology as described in any one of claims 1 to 8.

Images