Method for constructing discharge channel identification model and method for identifying discharge channel

Abstract

The embodiment of the application provides a kind of for the construction method of discharge channel identification model and the method for identifying discharge channel, belong to electric power system monitoring field.The construction method includes: initial discharge channel image identification model and initial discharge channel voiceprint identification model are trained;Determine image verification result and audio verification result;According to image recognition judgment weight, image recognition accuracy, voiceprint recognition judgment weight and voiceprint recognition accuracy, determine fusion identification accuracy;In the case where fusion identification accuracy is less than preset fusion identification accuracy threshold, initial discharge channel image identification model and initial discharge channel voiceprint identification model are retrained, until the model corresponding to the fusion identification accuracy reached by retraining reaches preset fusion identification accuracy threshold, to obtain discharge channel image identification model and discharge channel voiceprint identification model.The application can realize the technical effect of improving the accuracy of identifying discharge channel.

Description

Technical Field

[0001] This invention relates to the field of power system monitoring, and more specifically to a method for constructing a discharge channel identification model and a method for identifying discharge channels. Background Technology

[0002] In practical power system monitoring, determining whether a discharge phenomenon has occurred in the power system is crucial for detecting potential power system safety hazards. Current technology relies on manual identification of images of localized areas within the power system—specifically, manually judging the presence of discharge channels in the images—to determine whether a discharge phenomenon has occurred. However, manual identification suffers from limitations in meeting accuracy requirements. Therefore, existing technologies exhibit low accuracy in identifying discharge channels. Summary of the Invention

[0003] The purpose of this application is to provide a method for constructing a discharge channel identification model, as well as a method, apparatus, and storage medium for identifying discharge channels, in order to solve the problem of low accuracy in identifying discharge channels in the prior art.

[0004] To achieve the above objectives, the first aspect of this application provides a method for constructing a discharge channel image recognition model and a discharge channel voiceprint recognition model, the method comprising: Obtain regional image training sets, regional image verification sets, regional audio training sets, and regional audio verification sets corresponding to different regions, wherein a discharge channel may or may not exist within the region. Based on the regional image training set and the regional audio training set, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are trained respectively to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified. Determine the image verification results and audio verification results of the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified with respect to the regional image verification set and the regional audio verification set, respectively; Based on the image verification results and audio verification results, determine the image recognition judgment weight and image recognition accuracy corresponding to the discharge channel image recognition model to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the discharge channel voiceprint recognition model to be verified. Based on the image recognition judgment weight, image recognition accuracy, voiceprint recognition judgment weight, and voiceprint recognition accuracy, the fusion recognition accuracy of the fusion recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified is determined. If the fusion recognition accuracy is less than the preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, so as to obtain the discharge channel image recognition model and the discharge channel voiceprint recognition model.

[0005] In this embodiment, training the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model based on the regional image training set and the regional audio training set respectively to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified includes: training the initial discharge channel image recognition model for a preset number of training rounds based on the regional image training set to obtain the image recognition accuracy of the intermediate discharge channel image recognition model with respect to the regional image training set obtained in each training round; if the deviation of the image recognition accuracy corresponding to the last two rounds is greater than a preset image recognition accuracy threshold, retraining the initial discharge channel image recognition model for a preset number of training rounds. The initial discharge channel voiceprint recognition model is trained for a preset number of training rounds based on the regional audio training set. The audio recognition accuracy of the intermediate discharge channel voiceprint recognition model obtained in each training round is obtained with respect to the regional audio training set. If the deviation of the audio recognition accuracy corresponding to the last two rounds is greater than the preset audio recognition accuracy threshold, the initial discharge channel voiceprint recognition model is trained again for a preset number of training rounds until the audio recognition accuracy is less than or equal to the preset audio recognition accuracy threshold. This process yields the trained discharge channel voiceprint recognition model.

[0006] In this embodiment, the image verification result includes the number of correctly identified samples of a first discharge channel, the number of incorrectly identified samples of a first discharge channel, the number of correctly identified samples of a first non-discharge channel, and the number of incorrectly identified samples of a first non-discharge channel. The audio verification result includes the number of correctly identified samples of a second discharge channel, the number of incorrectly identified samples of a second discharge channel, the number of correctly identified samples of a second non-discharge channel, and the number of incorrectly identified samples of a second non-discharge channel. Based on the image verification result and the audio verification result, the image recognition judgment weight and image recognition accuracy corresponding to the discharge channel image recognition model to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the discharge channel voiceprint recognition model to be verified are determined. This includes determining the sum of the number of correctly identified samples of a first discharge channel, the number of incorrectly identified samples of a first discharge channel, the number of correctly identified samples of a first non-discharge channel, and the number of incorrectly identified samples of a first non-discharge channel, to obtain the first total number of samples. The following steps are taken to determine the total number of samples: First, the number of samples correctly identified as having a second discharge channel, the number of samples incorrectly identified as having a second discharge channel, the number of samples correctly identified as not having a second discharge channel, and the number of samples incorrectly identified as not having a second discharge channel. Second, the ratio of the sum of the first and second correct discharge channel identifications to the total number of samples is determined to obtain a first ratio. Third, the ratio of the sum of the first and second correct discharge channel identifications to the total number of samples is determined to obtain a second ratio. Fourth, the ratio of the first ratio to the sum of the first and second ratios is determined to obtain image recognition weights. Fifth, the ratio of the second ratio to the sum of the first and second ratios is determined to obtain voiceprint recognition weights. Sixth, the ratio of the first correct discharge channel identification to the sum of the first and second correct discharge channel identifications to obtain image recognition accuracy. Finally, the ratio of the second correct discharge channel identification to the sum of the second correct and second incorrect discharge channel identifications is determined to obtain voiceprint recognition accuracy.

[0007] In this embodiment of the application, the fusion recognition accuracy of the fusion recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified is determined according to the image recognition judgment weight, image recognition accuracy, voiceprint recognition judgment weight and voiceprint recognition accuracy. This includes determining the sum of the product of the image recognition judgment weight and the image recognition accuracy and the product of the voiceprint recognition judgment weight and the voiceprint recognition accuracy to obtain the fusion recognition accuracy.

[0008] A second aspect of this application provides a method for identifying discharge channels. The method includes: acquiring an image and audio corresponding to a region to be identified; performing image recognition on the image based on a pre-built discharge channel image recognition model to obtain a first probability that a discharge channel exists in the region to be identified, wherein the discharge channel image recognition model is constructed using the aforementioned method for constructing a discharge channel identification model; performing voiceprint recognition on the audio based on a pre-built discharge channel voiceprint recognition model to obtain a second probability that a discharge channel exists in the region to be identified, wherein the discharge channel voiceprint recognition model is constructed using the aforementioned method for constructing a discharge channel identification model; and identifying whether a discharge channel exists in the region to be identified based on the first probability and the second probability.

[0009] In this embodiment of the application, identifying whether a discharge channel exists in the region to be identified based on a first probability and a second probability includes: obtaining image recognition judgment weights corresponding to a pre-constructed discharge channel image recognition model and voiceprint recognition judgment weights corresponding to a pre-constructed discharge channel voiceprint recognition model; determining a first product value of the first probability and the image recognition judgment weights, and a second product value of the second probability and the voiceprint recognition judgment weights; obtaining a comprehensive probability based on the sum of the determined first product value and the second product value; and determining whether a discharge channel exists in the region to be identified based on the comprehensive probability.

[0010] In this embodiment of the application, determining whether a discharge channel exists in the area to be identified based on the comprehensive probability includes: determining whether the comprehensive probability is greater than or equal to a preset comprehensive probability threshold; if the comprehensive probability is greater than or equal to the preset comprehensive probability threshold, determining that a discharge channel exists in the area to be identified; if the comprehensive probability is less than the preset comprehensive probability threshold, determining that no discharge channel exists in the area to be identified.

[0011] A third aspect of this application provides an apparatus for constructing a discharge channel identification model, comprising: a memory configured to store instructions; and a processor configured to retrieve instructions from the memory and, when executing the instructions, to implement the above-described method for constructing a discharge channel identification model.

[0012] A fourth aspect of this application provides an apparatus for identifying a discharge channel, comprising: a memory configured to store instructions; and a processor configured to retrieve instructions from the memory and, when executing the instructions, to implement the method described above for identifying a discharge channel.

[0013] A fifth aspect of this application provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described method for constructing a discharge channel identification model or the above-described method for identifying a discharge channel.

[0014] The above technical solution, based on the completed training of the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model, verifies the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified. Regional image verification sets and regional audio verification sets are input into the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified to obtain image recognition judgment weights, image recognition accuracy, voiceprint recognition judgment weights, and voiceprint recognition accuracy. Then, the fusion recognition accuracy is obtained based on the image recognition judgment weights, image recognition accuracy, voiceprint recognition judgment weights, and voiceprint recognition accuracy. This application can determine whether to retrain the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model by judging the fusion recognition accuracy. If the fusion recognition accuracy is less than a preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, thus obtaining the discharge channel image recognition model and the discharge channel voiceprint recognition model. Therefore, this application can achieve high-precision training of the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model to obtain a high-precision discharge channel image recognition model and discharge channel voiceprint recognition model, thereby achieving the technical effect of improving the accuracy of discharge channel recognition.

[0015] Other features and advantages of the embodiments of the present invention will be described in detail in the following detailed description section. Attached Figure Description

[0016] The accompanying drawings are provided to further illustrate embodiments of the present invention and form part of the specification. They are used together with the following detailed description to explain the embodiments of the present invention, but do not constitute a limitation thereof. In the drawings: Figure 1 The illustration shows a flowchart of a method for constructing a discharge channel image recognition model and a discharge channel voiceprint recognition model according to an embodiment of this application. Detailed Implementation

[0017] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. It should be understood that the specific embodiments described herein are only for illustration and explanation of the embodiments of this application and are not intended to limit the embodiments of this application. All other embodiments obtained by those skilled in the art based on the embodiments of this application without creative effort are within the scope of protection of this application.

[0018] It should be noted that the acquisition, transmission, storage, use, and processing of data in the technical solution of this application all comply with the relevant provisions of national laws and regulations. In the embodiments of this application, certain existing industry solutions such as software, components, and models may be mentioned. These should be considered exemplary, intended only to illustrate the feasibility of implementing the technical solution of this application, and do not imply that the applicant has already used or necessarily used such solutions.

[0019] It should be noted that if the embodiments of this application involve directional indicators (such as up, down, left, right, front, back, etc.), the directional indicators are only used to explain the relative positional relationship and movement of each component in a certain specific posture (as shown in the figure). If the specific posture changes, the directional indicators will also change accordingly.

[0020] Furthermore, if the embodiments of this application involve descriptions such as "first" or "second," these descriptions are for descriptive purposes only and should not be construed as indicating or implying their relative importance or implicitly specifying the number of technical features indicated. Therefore, features defined with "first" or "second" may explicitly or implicitly include at least one of those features. Additionally, the technical solutions of various embodiments can be combined with each other, but this must be based on the ability of those skilled in the art to implement them. If the combination of technical solutions is contradictory or impossible to implement, it should be considered that such a combination of technical solutions does not exist and is not within the scope of protection claimed in this application.

[0021] Figure 1 The illustration schematically shows a flowchart of a method for constructing a discharge channel identification model according to an embodiment of this application. For example... Figure 1 As shown in the figure, this application provides a method for constructing a discharge channel recognition model. The discharge channel recognition model includes a discharge channel image recognition model and a discharge channel voiceprint recognition model. Taking the application of this method to a processor as an example, the method may include the following steps: Step S101: Obtain the region image training set, region image verification set, region audio training set, and region audio verification set corresponding to different regions, wherein a discharge channel may or may not exist in the region.

[0022] Step S102: Based on the regional image training set and the regional audio training set, train the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model respectively to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified.

[0023] Step S103: Determine the image verification results and audio verification results of the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified with respect to the regional image verification set and the regional audio verification set, respectively.

[0024] Step S104: Based on the image verification results and audio verification results, determine the image recognition judgment weight and image recognition accuracy corresponding to the discharge channel image recognition model to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the discharge channel voiceprint recognition model to be verified.

[0025] Step S105: Based on the image recognition judgment weight, image recognition accuracy, voiceprint recognition judgment weight, and voiceprint recognition accuracy, determine the fusion recognition accuracy of the fusion recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified.

[0026] Step S106: If the fusion recognition accuracy is less than the preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, so as to obtain the discharge channel image recognition model and the discharge channel voiceprint recognition model.

[0027] It can be understood that the region image training set consists of image sets from different regions used for training. The region image verification set consists of image sets from different regions used for verification. The region audio training set consists of audio sets from different regions used for training. The region audio verification set consists of audio sets from different regions used for verification. The initial discharge channel image recognition model is an untrained discharge channel image recognition model. The initial discharge channel voiceprint recognition model is an untrained discharge channel voiceprint recognition model. The discharge channel image recognition model to be verified is a trained discharge channel image recognition model that can be verified. The discharge channel voiceprint recognition model to be verified is a trained discharge channel voiceprint recognition model that can be verified. The image verification result is the verification result output by the discharge channel image recognition model to be verified. The audio verification result is the verification result output by the discharge channel audio recognition model to be verified. The image recognition judgment weight is the weight of the image recognition accuracy output by the discharge channel image recognition model to be verified. The voiceprint recognition judgment weight is the weight of the voiceprint recognition accuracy output by the discharge channel audio recognition model to be verified. The image recognition accuracy is the accuracy of the result output by the discharge channel image recognition model to be verified. Voiceprint recognition accuracy refers to the accuracy of the output result of the voiceprint recognition model for the discharge channel to be verified. Fusion recognition accuracy refers to the accuracy of determining whether a discharge channel exists in the recognition area after fusing image recognition accuracy and voiceprint recognition accuracy. The preset fusion recognition accuracy threshold is a pre-defined threshold for fusion recognition accuracy, which can be 70%.

[0028] Specifically, the processor first acquires regional image training sets, regional image verification sets, regional audio training sets, and regional audio verification sets corresponding to different regions. Based on the regional image training sets and regional audio training sets, it trains the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model, respectively, to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified. After the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model have been trained, they are verified to obtain image verification results and audio verification results for the regional image verification sets and regional audio verification sets. Then, based on the image verification results and audio verification results, the image recognition judgment weights and image recognition accuracy corresponding to the discharge channel image recognition model to be verified, and the voiceprint recognition judgment weights and voiceprint recognition accuracy corresponding to the discharge channel voiceprint recognition model to be verified are determined. The processor can determine the fusion recognition accuracy based on the image recognition judgment weights, image recognition accuracy, voiceprint recognition judgment weights, and voiceprint recognition accuracy. It then determines whether the fusion recognition accuracy is less than the preset fusion recognition accuracy threshold. If the fusion recognition accuracy is less than the preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, so as to obtain the discharge channel image recognition model and the discharge channel voiceprint recognition model.

[0029] The above technical solution, based on the completed training of the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model, verifies the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified. Regional image verification sets and regional audio verification sets are input into the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified to obtain image recognition judgment weights, image recognition accuracy, voiceprint recognition judgment weights, and voiceprint recognition accuracy. Then, the fusion recognition accuracy is obtained based on the image recognition judgment weights, image recognition accuracy, voiceprint recognition judgment weights, and voiceprint recognition accuracy. This application can determine whether to retrain the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model by judging the fusion recognition accuracy. If the fusion recognition accuracy is less than a preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, thus obtaining the discharge channel image recognition model and the discharge channel voiceprint recognition model. Therefore, this application can achieve high-precision training of the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model to obtain a high-precision discharge channel image recognition model and discharge channel voiceprint recognition model, thereby achieving the technical effect of improving the accuracy of discharge channel recognition.

[0030] In one embodiment, training the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model based on the regional image training set and the regional audio training set respectively to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified includes: training the initial discharge channel image recognition model for a preset number of training rounds based on the regional image training set to obtain the image recognition accuracy of the intermediate discharge channel image recognition model with respect to the regional image training set obtained in each training round; if the deviation of the image recognition accuracy corresponding to the last two rounds is greater than a preset image recognition accuracy threshold, retraining the initial discharge channel image recognition model for a preset number of training rounds. The training continues until the image recognition accuracy is less than or equal to a preset image recognition accuracy threshold, thus obtaining the trained discharge channel image recognition model to be verified. Based on the regional audio training set, the initial discharge channel voiceprint recognition model is trained for a preset number of training rounds to obtain the audio recognition accuracy of the intermediate discharge channel voiceprint recognition model with respect to the regional audio training set obtained in each training round. If the deviation of the audio recognition accuracy corresponding to the last two rounds is greater than the preset audio recognition accuracy threshold, the initial discharge channel voiceprint recognition model is retrained for a preset number of training rounds until the audio recognition accuracy is less than or equal to the preset audio recognition accuracy threshold, thus obtaining the trained discharge channel voiceprint recognition model to be verified.

[0031] It is understood that the preset training rounds are pre-defined training rounds, which can be 200 rounds. The intermediate discharge channel image recognition model is the discharge channel image recognition model obtained by training the initial discharge channel image recognition model through each training round. The image recognition accuracy is the recognition accuracy corresponding to the intermediate discharge channel image recognition model. The intermediate discharge channel voiceprint recognition model is the discharge channel voiceprint recognition model obtained by training the initial discharge channel voiceprint recognition model through each training round. The voiceprint recognition accuracy is the recognition accuracy corresponding to the intermediate discharge channel voiceprint recognition model. The preset image recognition accuracy threshold is a pre-defined threshold for image recognition accuracy. The preset audio recognition accuracy threshold is a pre-defined threshold for voiceprint recognition accuracy. The preset image recognition accuracy threshold and the preset audio recognition accuracy threshold can be the same or different; for example, both the preset image recognition accuracy threshold and the preset audio recognition accuracy threshold can be 0.01.

[0032] Specifically, the processor trains the initial discharge channel image recognition model for a preset number of training rounds based on the regional image training set, obtaining the image recognition accuracy corresponding to each training round. If the deviation of the image recognition accuracy corresponding to the last two rounds exceeds a preset image recognition accuracy threshold, the initial discharge channel image recognition model is retrained for the preset number of training rounds until the image recognition accuracy is less than or equal to the preset image recognition accuracy threshold, thus obtaining the trained discharge channel image recognition model to be verified. Similarly, the processor trains the initial discharge channel voiceprint recognition model for a preset number of training rounds based on the regional audio training set, obtaining the audio recognition accuracy corresponding to each training round. If the deviation of the audio recognition accuracy corresponding to the last two rounds exceeds a preset audio recognition accuracy threshold, the initial discharge channel voiceprint recognition model is retrained for the preset number of training rounds until the audio recognition accuracy is less than or equal to the preset audio recognition accuracy threshold, thus obtaining the trained discharge channel voiceprint recognition model to be verified.

[0033] In this embodiment, the initial discharge channel image recognition model is trained for a preset number of training rounds until the image recognition accuracy in the last two rounds is less than or equal to a preset image recognition accuracy threshold, thus obtaining the discharge channel image recognition model to be verified. Similarly, the initial discharge channel voiceprint recognition model is trained for a preset number of training rounds until the voiceprint recognition accuracy in the last two rounds is less than or equal to a preset voiceprint recognition accuracy threshold, thus obtaining the discharge channel voiceprint recognition model to be verified. This method allows the success of model training to be measured by the accuracy achieved in the last two rounds, resulting in trained discharge channel image recognition models and trained discharge channel voiceprint recognition models. This ensures that the recognition accuracy of both models reaches an ideal level, enabling accurate recognition of images and audio.

[0034] In one embodiment, the image verification result includes the number of correctly identified samples indicating the presence of a first discharge channel, the number of incorrectly identified samples indicating the presence of a first discharge channel, the number of correctly identified samples indicating the absence of a first discharge channel, and the number of incorrectly identified samples indicating the absence of a first discharge channel. The audio verification result includes the number of correctly identified samples indicating the presence of a second discharge channel, the number of incorrectly identified samples indicating the presence of a second discharge channel, the number of correctly identified samples indicating the absence of a second discharge channel, and the number of incorrectly identified samples indicating the absence of a second discharge channel. Based on the image verification result and the audio verification result, the image recognition judgment weight and image recognition accuracy corresponding to the discharge channel image recognition model to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the discharge channel voiceprint recognition model to be verified are determined. This includes: determining the sum of the number of correctly identified samples indicating the presence of a first discharge channel, the number of incorrectly identified samples indicating the presence of a first discharge channel, the number of correctly identified samples indicating the absence of a first discharge channel, and the number of incorrectly identified samples indicating the absence of a first discharge channel, to obtain the first total number of samples; and determining the number of correctly identified samples indicating the presence of a second discharge channel and the number of incorrectly identified samples indicating the presence of a second discharge channel. The following steps are performed to determine the second total number of samples: First, the sum of the first and second correct discharge channel identifications (both correctly and incorrectly) is used to determine the first total number of samples. Second, the ratio of the sum of the first and second correct discharge channel identifications to the first total number of samples is determined to determine the first ratio. Third, the ratio of the sum of the second and third correct discharge channel identifications to the second total number of samples is determined to determine the second ratio. Fourth, the ratio of the first ratio to the sum of the first and second ratios is determined to obtain the image recognition weight. Fifth, the ratio of the second ratio to the sum of the first and second ratios is determined to obtain the voiceprint recognition weight. Sixth, the ratio of the first correct discharge channel identifications to the sum of the first and second correct discharge channel identifications to determine the image recognition accuracy. Finally, the ratio of the second correct discharge channel identifications to the sum of the second correct discharge channel identifications to the sum of the second correct discharge channel identifications to determine the voiceprint recognition accuracy.

[0035] It can be understood that the first number of correctly identified samples indicating the existence of a discharge channel is the number of correctly identified samples out of the samples indicating the existence of a discharge channel output by the discharge channel image recognition model to be verified. The first number of incorrectly identified samples indicating the existence of a discharge channel is the number of incorrectly identified samples out of the samples indicating the existence of a discharge channel output by the discharge channel image recognition model to be verified. The first number of correctly identified samples indicating the absence of a discharge channel is the number of correctly identified samples out of the samples indicating the absence of a discharge channel output by the discharge channel image recognition model to be verified. The first number of incorrectly identified samples indicating the absence of a discharge channel is the number of incorrectly identified samples out of the samples indicating the absence of a discharge channel output by the discharge channel image recognition model to be verified. The second number of correctly identified samples indicating the existence of a discharge channel is the number of correctly identified samples out of the samples indicating the existence of a discharge channel output by the discharge channel voiceprint recognition model to be verified. The second number of incorrectly identified samples indicating the existence of a discharge channel is the number of incorrectly identified samples out of the samples indicating the existence of a discharge channel output by the discharge channel voiceprint recognition model to be verified. The second number of correctly identified samples indicating the absence of a discharge channel is the number of correctly identified samples out of the samples indicating the absence of a discharge channel output by the discharge channel voiceprint recognition model to be verified. The second number of incorrectly identified samples indicating the absence of a discharge channel is the number of incorrectly identified samples out of the samples indicating the absence of a discharge channel output by the discharge channel voiceprint recognition model to be verified. It is important to note that the image recognition model for the discharge channel to be verified outputs the probability of whether a discharge channel exists in each image. The processor determines whether a discharge channel exists in each image based on this probability and a preset probability threshold. Similarly, the voiceprint recognition model for the discharge channel to be verified outputs the probability of whether a discharge channel exists in each audio segment. The processor determines whether a discharge channel exists in each audio segment based on this probability and a preset probability threshold. The first ratio is the ratio of the sum of the first number of correctly identified samples indicating the presence of a discharge channel and the first number of correctly identified samples indicating the absence of a discharge channel to the first total number of samples. The second ratio is the ratio of the sum of the second number of correctly identified samples indicating the presence of a discharge channel and the second number of correctly identified samples indicating the absence of a discharge channel to the second total number of samples.

[0036] Specifically, the processor determines a first ratio by the ratio of the sum of the number of correctly identified samples with a first discharge channel and the number of correctly identified samples without a first discharge channel to the first total number of samples. It then determines a second ratio by the ratio of the sum of the number of correctly identified samples with a second discharge channel and the number of correctly identified samples without a second discharge channel to the second total number of samples. The processor also determines the ratio of the first ratio to the sum of the first and second ratios to obtain the image recognition judgment weight. Finally, it determines the ratio of the second ratio to the sum of the first and second ratios to obtain the voiceprint recognition judgment weight. The processor determines the image recognition accuracy by the ratio of the sum of the number of correctly identified samples with a first discharge channel and the number of incorrectly identified samples with a first discharge channel. Finally, it determines the voiceprint recognition accuracy by the ratio of the sum of the number of correctly identified samples with a second discharge channel and the number of incorrectly identified samples with a second discharge channel.

[0037] In this embodiment, the processor calculates the image recognition judgment weight, voiceprint recognition judgment weight, image recognition accuracy, and voiceprint recognition accuracy based on the number of correctly judged samples of the first presence of a discharge channel, the number of incorrectly judged samples of the first presence of a discharge channel, the number of correctly judged samples of the first absence of a discharge channel, the number of incorrectly judged samples of the first absence of a discharge channel, the number of correctly judged samples of the second presence of a discharge channel, the number of incorrectly judged samples of the second presence of a discharge channel, the number of correctly judged samples of the second absence of a discharge channel, and the number of incorrectly judged samples of the second absence of a discharge channel. This not only determines the fusion accuracy to decide whether to update the model, but also obtains the proportion of the judgment results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified. The judgment results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified are then combined according to the proportion to obtain the final result of judging whether a discharge channel exists, thereby improving the accuracy of the result of judging whether a discharge channel exists.

[0038] In one embodiment, determining the fusion recognition accuracy of the fusion recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified, based on the image recognition judgment weight, image recognition accuracy, voiceprint recognition judgment weight, and voiceprint recognition accuracy, includes: determining the sum of the product of the image recognition judgment weight and the image recognition accuracy and the product of the voiceprint recognition judgment weight and the voiceprint recognition accuracy, so as to obtain the fusion recognition accuracy.

[0039] Specifically, the processor first determines the product of image recognition judgment weights and image recognition accuracy, then determines the product of voiceprint recognition judgment weights and voiceprint recognition accuracy, and adds the two products together to obtain the fusion recognition accuracy. Determining the fusion recognition accuracy prepares the processor for deciding whether to update the model based on the fusion accuracy.

[0040] In one embodiment, the method includes: acquiring an image and audio corresponding to a region to be identified; performing image recognition on the image based on a pre-built discharge channel image recognition model to obtain a first probability that a discharge channel exists in the region to be identified, wherein the discharge channel image recognition model is constructed using the aforementioned method for constructing a discharge channel recognition model; performing voiceprint recognition on the audio based on a pre-built discharge channel voiceprint recognition model to obtain a second probability that a discharge channel exists in the region to be identified, wherein the discharge channel voiceprint recognition model is constructed using the aforementioned method for constructing a discharge channel recognition model; and identifying whether a discharge channel exists in the region to be identified based on the first probability and the second probability.

[0041] It can be understood that the first probability is the probability that a discharge channel exists within the area to be identified. The second probability is the probability that a discharge channel exists within the area to be identified.

[0042] Specifically, the processor determines a first probability that a discharge channel exists in the image of the region to be identified based on the discharge channel image recognition model. The processor determines a second probability that a discharge channel exists in the audio of the region to be identified based on the discharge channel voiceprint recognition model. Finally, the processor identifies whether a discharge channel exists in the region to be identified based on the first and second probabilities, thus determining whether a discharge channel exists in the region to be identified.

[0043] In this embodiment, the processor, by fusing the first and second probabilities, integrates the judgment results of both the discharge channel image recognition model and the discharge channel voiceprint recognition model. This makes the determination of the existence of a discharge channel more reliable, reducing missed detections and false detections. Since image recognition may be affected by lighting, occlusion, and resolution, and voiceprint recognition may be affected by environmental noise, their complementarity reduces misjudgments caused by reliance on a single model. Furthermore, in complex environments (such as dimly lit scenes, long-distance monitoring, and scenes with multiple interfering noises), a single model may fail; dual-model judgment can compensate for the shortcomings of the other by using effective data from one model, covering more application scenarios. Therefore, integrating the judgment results of both the discharge channel image recognition model and the discharge channel voiceprint recognition model can improve the accuracy of determining the existence of a discharge channel.

[0044] In one embodiment, identifying whether a discharge channel exists in the region to be identified based on a first probability and a second probability includes: obtaining image recognition judgment weights corresponding to a pre-constructed discharge channel image recognition model and voiceprint recognition judgment weights corresponding to a pre-constructed discharge channel voiceprint recognition model; determining a first product value of the first probability and the image recognition judgment weights, and a second product value of the second probability and the voiceprint recognition judgment weights; obtaining a comprehensive probability based on the sum of the determined first and second product values; and determining whether a discharge channel exists in the region to be identified based on the comprehensive probability.

[0045] Specifically, the processor determines the sum of the first product of the first probability and the image recognition judgment weight, and the second product of the second probability and the voiceprint recognition judgment weight, to obtain a comprehensive probability. Based on the comprehensive probability, it determines whether a discharge channel exists in the area to be identified. This clarifies the proportion of the recognition results of the discharge channel image recognition model and the proportion of the recognition results of the discharge channel voiceprint recognition model, thereby improving the accuracy of the judgment results of the comprehensive discharge channel image recognition model and the discharge channel voiceprint recognition model.

[0046] In one embodiment, determining whether a discharge channel exists in the region to be identified based on the comprehensive probability includes: determining whether the comprehensive probability is greater than or equal to a preset comprehensive probability threshold; if the comprehensive probability is greater than or equal to the preset comprehensive probability threshold, determining that a discharge channel exists in the region to be identified; if the comprehensive probability is less than the preset comprehensive probability threshold, determining that no discharge channel exists in the region to be identified.

[0047] It is understandable that the preset comprehensive probability threshold is a pre-set threshold for comprehensive probability, which can be 50%.

[0048] Specifically, the processor determines whether the overall probability is greater than or equal to a preset overall probability threshold. If the overall probability is greater than or equal to the preset overall probability threshold, it determines that a discharge channel exists in the area to be identified. If the overall probability is less than the preset overall probability threshold, it determines that no discharge channel exists in the area to be identified. For example, it determines whether the overall probability is greater than or equal to 50%. If the overall probability is greater than or equal to 50%, it determines that a discharge channel exists in the area to be identified. If the overall probability is less than 50%, it determines that no discharge channel exists in the area to be identified.

[0049] In this embodiment of the application, it is more accurate to determine whether there is a discharge channel in the area to be identified based on a specific preset comprehensive probability threshold.

[0050] When the electric field strength inside a dielectric exceeds a certain value, a discharge channel is formed within it. Three-dimensional reconstruction of the discharge channel is crucial for analyzing its three-dimensional discharge characteristics. Accurate reconstruction helps in understanding the characteristics of different stages, such as discharge initiation, development, and breakdown. Analyzing the three-dimensional discharge characteristics of the discharge channel is essential for ensuring the safe and stable operation of power systems and is an indispensable part of lightning protection and gas insulation research.

[0051] However, the high speed of long air gap discharges makes them difficult to capture accurately in real time manually. In actual power system monitoring, because discharges occur instantaneously and last for an extremely short time, manual observation is almost impossible to accurately obtain key information about the discharge process, making it difficult to detect potential power system safety hazards in a timely manner. Manual image processing after photography is too slow, and manual methods cannot meet the requirements of real-time performance and accuracy, limiting in-depth research on long air gap discharges and rapid diagnosis of power system faults.

[0052] Therefore, a specific embodiment of this application provides a method for constructing a discharge channel identification model and a method for identifying discharge channels. This method directly learns and identifies discharge characteristics from the collected data to overcome the aforementioned technical limitations and improve the efficiency and accuracy of long air gap discharge channel identification. The specific steps are as follows: Step 1: Extract the images and audio from the discharge channels in the video captured by the camera, and divide them into image training sets, image verification sets, and image test sets, as well as audioprint training sets, audioprint verification sets, and audioprint test sets. The image datasets and audioprint datasets correspond to the respective discharge channels. Process the image datasets and assign labels to indicate the presence or absence of discharge channels based on the discharge path.

[0053] Furthermore, the training dataset is specifically obtained by capturing images and sounds of the long-gap air discharge process using a DSLR camera. Multi-angle shooting is conducted based on the actual discharge channel environment, taking into account the 3D reconstruction of the discharge channel. To ensure image clarity, the camera is configured for maximum resolution. Simultaneously, to facilitate subsequent image processing and analysis, the acquired images are saved in a lossless compression format, reducing file size without sacrificing image quality.

[0054] Step 2: Train the YOLOv11 discharge channel image recognition model using the aforementioned image training set. The model learns the data features from the training set to build a prediction model. Adjusting hyperparameters requires using the aforementioned image validation set to prevent overfitting on the training set and to evaluate the model's performance during training. Finally, validate the model's effectiveness using the image test set, save the trained image model, and use it to identify new discharge images outside the dataset.

[0055] Furthermore, regional image training sets, regional image verification sets, regional audio training sets, and regional audio verification sets are obtained for different regions, wherein a discharge channel may or may not exist within the region. Based on the regional image training set and the regional audio training set, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are trained respectively to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified. Determine the image verification results and audio verification results of the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified with respect to the regional image verification set and the regional audio verification set, respectively; Based on the image verification results and audio verification results, determine the image recognition judgment weight and image recognition accuracy corresponding to the discharge channel image recognition model to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the discharge channel voiceprint recognition model to be verified. Based on the image recognition judgment weight, image recognition accuracy, voiceprint recognition judgment weight, and voiceprint recognition accuracy, the fusion recognition accuracy of the fusion recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified is determined. If the fusion recognition accuracy is less than the preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, so as to obtain the discharge channel image recognition model and the discharge channel voiceprint recognition model.

[0056] Furthermore, hyperparameters are parameters set before machine learning, which need to be predefined and affect the generalization ability of the final model.

[0057] Step 3: Input the image to be identified into the pre-set YOLOv11 discharge channel image recognition model. The backbone network of the YOLOv11 discharge channel recognition model extracts multi-scale features from the image to be identified. Then, the detection head of the YOLOv11 discharge channel recognition model combines these multi-scale features to perform target detection on the image to be identified, thereby obtaining the discharge channel detection result. The YOLOv11 discharge channel image recognition model consists of a backbone network, a neck, and a head. When using YOLOv11 for discharge channel target detection, the input data undergoes a layer-by-layer feature extraction and spatial transformation process to ensure the integrity of the information. Input the dataset image into the YOLOv11 model. The model extracts multi-scale features from the input image dataset through its backbone network. The modules that make up the backbone network include Conv, C3k2, SPPF, and C2PSA. Conv (convolutional layer) extracts image features through convolutional kernels. The C3k2 module, comprising C3k, Bottleneck, and CBS, is an accelerated implementation of CSP Bottleneck, combining the features of convolution and grouped convolution. C3k allows users to select the kernel size, enabling optimization for different tasks and computational resources. Bottleneck consists of 1x1 and 3x3 convolutional layers; the 1x1 layers reduce the number of channels (lowering computational complexity), while the 3x3 layers extract local features. CBS uses two 3x3 convolutions, reducing resolution by a factor of 16. The SPPF module performs pooling operations on feature maps at different scales to obtain multi-scale feature information. The C2PSA module is a convolutional module based on the PSA (PyramidSqueeze Attention) mechanism, responsible for processing input tensors and enhancing feature representations through attention. This module involves convolution operations, segmentation, and multi-head attention mechanisms. The extracted features are fused through the Feature Pyramid Network (FPN) in the neck to improve the detection capability for targets of different sizes. The main modules of the detection head include Conv, Conv2d, DWConv, CloU, and CLSLoss. Finally, the model outputs the category, location, and confidence weight of the detected targets for subsequent image analysis and processing.

[0058] Further, the sum of the following parameters is determined: the number of correctly identified samples indicating the presence of a first discharge channel, the number of incorrectly identified samples indicating the presence of a first discharge channel, the number of correctly identified samples indicating the absence of a first discharge channel, and the number of incorrectly identified samples indicating the absence of a first discharge channel, to obtain a first total sample number; the sum of the following parameters is determined: the number of correctly identified samples indicating the presence of a second discharge channel, the number of incorrectly identified samples indicating the presence of a second discharge channel, the number of correctly identified samples indicating the absence of a second discharge channel, and the number of incorrectly identified samples indicating the absence of a second discharge channel, to obtain a second total sample number; the ratio of the sum ... The following ratios are used to determine the image recognition judgment weight: First, the sum of the number of correctly identified samples with no discharge channel is determined to the second total number of samples. Second, the ratio of the first ratio to the sum of the first and second ratios is determined to the voiceprint recognition judgment weight. Third, the ratio of the number of correctly identified samples with a first discharge channel to the sum of the number of correctly identified samples with a first discharge channel and the number of incorrectly identified samples with a first discharge channel is determined to the image recognition accuracy. Fourth, the ratio of the number of correctly identified samples with a second discharge channel to the sum of the number of correctly identified samples with a second discharge channel and the number of incorrectly identified samples with a second discharge channel is determined to the voiceprint recognition accuracy.

[0059] In this embodiment, the average precision mAP50 of the discharge channel image recognition model detection accuracy P and recall rate of 50% is shown in Table 1.

[0060] Table 1. YOLOv11 model detection results

[0061] As shown in Table 1, the detection accuracy P of the discharge channel image recognition model reaches 90.6%, and the present invention has the characteristics of high accuracy and accurate identification of positive samples.

[0062] Step 4: Based on the discharge channel detection results, extract the image to be identified, match it with the voiceprint dataset, and generate the audio to be detected.

[0063] Step 5: Noise is added to the audio to be detected. After adding noisy speech, batch training is performed. The front-end convolutional module encodes acoustic features through two-dimensional convolutional blocks to extract high-resolution time-frequency Mel filter bank coefficients (FBank) features. The obtained Mel filter bank coefficients (FBank) features are then input into the CAM++ speaker vector extraction model for training to obtain speaker vectors. The model is trained using the AAM-Softmax loss function and stochastic gradient descent (SGD) optimizer until the algorithm converges.

[0064] Furthermore, the noise reduction process includes adding echo noise and natural noise to the training corpus.

[0065] Furthermore, the CAM++ speaker vector extraction model mainly consists of three parts: a front-end convolutional module (FCM), a D-TDNN backbone network, and a CAM module. The front-end convolutional module (FCM) uses two-dimensional convolutional blocks, which can encode acoustic features in the time-frequency domain and extract high-resolution time-frequency details. The resulting Mel filter bank coefficients (FBank) features are flattened along the channel and frequency dimensions and used as input to the D-TDNN backbone network. The D-TDNN backbone network includes three modules, each containing a series of densely connected time-delay neural network (D-TDNN) layers. The improved CAM module is inserted into the densely connected time-delay neural network (D-TDNN) layers, assigning different attention weights to the output features of the internal time-delay neural network (TDNN) layers. By masking the feature map, it focuses on the target speaker and suppresses noise.

[0066] Furthermore, four residual blocks are added to the FCM system, with the number of channels in all residual blocks set to 32, and the last three blocks having a frequency dimension step size of 27.

[0067] Furthermore, the basic unit of the densely connected time-delay neural network (D-TDNN) layer consists of a feedforward neural network (FNN) and a TDNN layer, with the inputs of two consecutive D-TDNN layers directly connected.

[0068] Furthermore, by aggregating contextual information at different levels through multi-granularity pooling (including global average pooling and segmented average pooling), contextual embeddings are generated to predict ratio masks, and then the representation is calibrated to obtain accurate voiceprint features. Step 6: Save the trained voiceprint model for recognizing new discharge audio outside the dataset. Obtain the voiceprint feature data to be recognized, calculate the diagonal cosine value of the voiceprint vector, and determine whether the voice to be recognized is a discharge channel breakdown voiceprint based on the similarity threshold, and generate the audio to be detected.

[0069] In this embodiment, the features and accuracy P used in the discharge channel voiceprint recognition model detection are shown in Table 2.

[0070] Table 2 CAM++ model detection results

[0071] As shown in Table 2, the detection accuracy P of the discharge channel voiceprint recognition model reaches 72.0%. The voiceprint recognition model can accurately identify special voiceprints when the discharge channel breaks down by utilizing FBank features.

[0072] Step 7: Compare and verify the predicted results of the image to be identified with the predicted audio to be detected, and determine the discharge channel.

[0073] Furthermore, the image model and the voiceprint model are fused at the decision layer. The output of each single-modal model is normalized by the acousto-optic model network, and a weighted average is used as the final evaluation of the detection result.

[0074] In this embodiment, the detection accuracy P of the fusion model is shown in Table 3.

[0075] Table 3 Model fusion detection results

[0076] As shown in Table 3, the detection accuracy P of the fusion model reaches 81.3%. The fusion model combines the accuracy of the YOLOv11 model and the CAM++ model, and the fusion complements each other, making the discharge channel identification model more accurate.

[0077] Furthermore, the specific method for determining whether a discharge channel exists in the area to be identified is as follows: 1. Obtain the image and audio corresponding to the region to be identified; perform image recognition on the image based on the pre-built discharge channel image recognition model to obtain the first probability that a discharge channel exists in the region to be identified, wherein the discharge channel image recognition model is constructed using the above-mentioned construction method for the discharge channel image recognition model and the discharge channel voiceprint recognition model; perform voiceprint recognition on the audio based on the pre-built discharge channel voiceprint recognition model to obtain the second probability that a discharge channel exists in the region to be identified, wherein the discharge channel voiceprint recognition model is constructed using the above-mentioned construction method for the discharge channel image recognition model and the discharge channel voiceprint recognition model; identify whether a discharge channel exists in the region to be identified based on the first probability and the second probability.

[0078] 2. Obtain the image recognition judgment weights corresponding to the pre-constructed discharge channel image recognition model and the voiceprint recognition judgment weights corresponding to the pre-constructed discharge channel voiceprint recognition model; determine the first product value of the first probability and the image recognition judgment weights, and the second product value of the second probability and the voiceprint recognition judgment weights; obtain the comprehensive probability based on the sum of the first and second product values; determine whether the comprehensive probability is greater than or equal to a preset comprehensive probability threshold; if the comprehensive probability is greater than or equal to the preset comprehensive probability threshold, determine that a discharge channel exists in the area to be identified; if the comprehensive probability is less than the preset comprehensive probability threshold, determine that no discharge channel exists in the area to be identified.

[0079] A discharge image and discharge acoustic print are input into the YOLO-CAM fusion model. The model prediction result is calculated. If the prediction probability is greater than or equal to 0.5, it is determined to be a discharge channel.

[0080] In one embodiment, this application provides an apparatus for constructing a discharge channel identification model, comprising: a memory configured to store instructions; and a processor configured to retrieve instructions from the memory and, when executing the instructions, to implement the above-described method for constructing a discharge channel identification model.

[0081] In one embodiment, this application provides an apparatus for identifying a discharge channel, comprising: a memory configured to store instructions; and a processor configured to retrieve instructions from the memory and, when executing the instructions, to implement the method described above for identifying a discharge channel.

[0082] In one embodiment, this application provides a machine-readable storage medium storing instructions that cause a machine to execute the above-described method for constructing a discharge channel identification model, or the above-described method for identifying a discharge channel.

Claims

1. A method for constructing a discharge channel identification model, characterized in that, The discharge channel recognition model includes a discharge channel image recognition model and a discharge channel voiceprint recognition model, and the construction method includes: Obtain regional image training sets, regional image verification sets, regional audio training sets, and regional audio verification sets corresponding to different regions, wherein a discharge channel may or may not exist within the region. Based on the regional image training set and the regional audio training set, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are trained respectively to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified. Determine the image verification results and audio verification results of the discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified with respect to the regional image verification set and the regional audio verification set, respectively; Based on the image verification results and the audio verification results, determine the image recognition judgment weight and image recognition accuracy corresponding to the image recognition model of the discharge channel to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the voiceprint recognition model of the discharge channel to be verified. Based on the image recognition judgment weight, the image recognition accuracy, the voiceprint recognition judgment weight, and the voiceprint recognition accuracy, the fusion recognition accuracy of the fusion recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified is determined. If the fusion recognition accuracy is less than the preset fusion recognition accuracy threshold, the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model are retrained until the fusion recognition accuracy of the retrained model reaches the preset fusion recognition accuracy threshold, so as to obtain the discharge channel image recognition model and the discharge channel voiceprint recognition model.

2. The construction method according to claim 1, characterized in that, The step of training the initial discharge channel image recognition model and the initial discharge channel voiceprint recognition model based on the regional image training set and the regional audio training set, respectively, to obtain the trained discharge channel image recognition model and the discharge channel voiceprint recognition model to be verified, includes: Based on the regional image training set, the initial discharge channel image recognition model is trained for a preset number of training rounds to obtain the image recognition accuracy of the intermediate discharge channel image recognition model with respect to the regional image training set obtained in each training round. If the deviation of the image recognition accuracy in the last two rounds is greater than the preset image recognition accuracy threshold, the initial discharge channel image recognition model is retrained for a preset number of rounds until the image recognition accuracy is less than or equal to the preset image recognition accuracy threshold, so as to obtain the trained discharge channel image recognition model to be verified. Based on the regional audio training set, the initial discharge channel voiceprint recognition model is trained for a preset number of training rounds to obtain the audio recognition accuracy of the intermediate discharge channel voiceprint recognition model with respect to the regional audio training set obtained in each training round. If the deviation of the audio recognition accuracy in the last two rounds is greater than the preset audio recognition accuracy threshold, the initial discharge channel voiceprint recognition model is retrained for a preset number of rounds until the audio recognition accuracy is less than or equal to the preset audio recognition accuracy threshold, so as to obtain the trained discharge channel voiceprint recognition model to be verified.

3. The construction method according to claim 1, characterized in that, The image verification results include the number of samples correctly identified as having a first discharge channel, the number of samples incorrectly identified as having a first discharge channel, the number of samples correctly identified as not having a first discharge channel, and the number of samples incorrectly identified as not having a first discharge channel. The audio verification results include the number of samples correctly identified as having a second discharge channel, the number of samples incorrectly identified as having a second discharge channel, the number of samples correctly identified as not having a second discharge channel, and the number of samples incorrectly identified as not having a second discharge channel. The step of determining the image recognition judgment weight and image recognition accuracy corresponding to the image recognition model of the discharge channel to be verified, and the voiceprint recognition judgment weight and voiceprint recognition accuracy corresponding to the voiceprint recognition model of the discharge channel to be verified, based on the image verification result and the audio verification result, includes: The sum of the number of samples correctly identified as having a discharge channel, the number of samples incorrectly identified as having a discharge channel, the number of samples correctly identified as not having a discharge channel, and the number of samples incorrectly identified as not having a discharge channel is determined to obtain the first total number of samples. The number of samples correctly identified as having a second discharge channel, the number of samples incorrectly identified as having a second discharge channel, the number of samples correctly identified as not having a second discharge channel, and the number of samples incorrectly identified as not having a second discharge channel are determined to obtain the second total number of samples. The first ratio is determined by the ratio of the sum of the number of samples correctly identified as having a discharge channel and the number of samples correctly identified as not having a discharge channel to the first total number of samples. The second ratio is determined by the ratio of the sum of the number of correctly identified samples indicating the presence of a second discharge channel and the number of correctly identified samples indicating the absence of a second discharge channel to the total number of samples. Determine the ratio of the first ratio to the sum of the first and second ratios to obtain the image recognition judgment weight; Determine the ratio of the second ratio to the sum of the first and second ratios to obtain the voiceprint recognition judgment weight; The ratio of the number of correctly identified samples with the presence of the first discharge channel to the sum of the number of correctly identified samples with the presence of the first discharge channel and the number of incorrectly identified samples with the presence of the first discharge channel is determined to determine the image recognition accuracy. The ratio of the number of correctly identified samples of the second discharge channel to the sum of the number of correctly identified samples of the second discharge channel and the number of incorrectly identified samples of the second discharge channel is determined to determine the accuracy of voiceprint recognition.

4. The construction method according to claim 1, characterized in that, The determination of the fusion recognition accuracy of the fused recognition result obtained by fusing the recognition results of the image recognition model and the voiceprint recognition model of the discharge channel to be verified based on the image recognition judgment weight, the image recognition accuracy, the voiceprint recognition judgment weight, and the voiceprint recognition accuracy includes: The sum of the product of the image recognition judgment weight and the image recognition accuracy and the product of the voiceprint recognition judgment weight and the voiceprint recognition accuracy is determined to obtain the fusion recognition accuracy.

5. A method for identifying discharge channels, characterized in that, The method includes: Obtain the image and audio corresponding to the region to be identified; Based on a pre-built discharge channel image recognition model, image recognition is performed on the image to obtain a first probability that a discharge channel exists in the region to be recognized, wherein the discharge channel image recognition model is constructed by the method for constructing a discharge channel recognition model as described in any one of claims 1 to 4; Based on a pre-built discharge channel voiceprint recognition model, voiceprint recognition is performed on the audio to obtain a second probability that a discharge channel exists in the region to be identified, wherein the discharge channel voiceprint recognition model is constructed by the method for constructing a discharge channel recognition model as described in any one of claims 1 to 4; Based on the first probability and the second probability, identify whether there is a discharge channel in the region to be identified.

6. The method according to claim 5, characterized in that, The step of identifying whether a discharge channel exists in the region to be identified based on the first probability and the second probability includes: Obtain the image recognition judgment weights corresponding to the pre-constructed discharge channel image recognition model and the voiceprint recognition judgment weights corresponding to the pre-constructed discharge channel voiceprint recognition model; Determine the first product value of the first probability and the image recognition judgment weight, and the second product value of the second probability and the voiceprint recognition judgment weight; The combined probability is obtained by determining the sum of the first and second product values; Based on the comprehensive probability, it is determined whether a discharge channel exists within the region to be identified.

7. The method according to claim 6, characterized in that, The step of determining whether a discharge channel exists within the region to be identified based on the comprehensive probability includes: Determine whether the overall probability is greater than or equal to a preset overall probability threshold; If the overall probability is greater than or equal to the preset overall probability threshold, it is determined that a discharge channel exists in the region to be identified. If the overall probability is less than the preset overall probability threshold, it is determined that there is no discharge channel in the area to be identified.

8. An apparatus for constructing a discharge channel identification model, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the method for constructing a discharge channel identification model according to any one of claims 1 to 4.

9. A device for identifying discharge channels, characterized in that, include: The memory is configured to store instructions; as well as The processor is configured to retrieve the instructions from the memory and, when executing the instructions, to implement the method for identifying a discharge channel according to any one of claims 5 to 7.

10. A machine-readable storage medium, characterized in that, The machine-readable storage medium stores instructions for causing the machine to execute the method for constructing a discharge channel identification model according to any one of claims 1 to 4, or the method for identifying a discharge channel according to any one of claims 5 to 7.

Images