A bird-repelling device and method based on active monitoring and visual recognition

Through active monitoring and visual recognition technology, combined with SVM classification model, birds near power transmission pole towers are identified and driven away, and the problem of difficulty in identifying and driving up in the existing technology is solved, the recognition accuracy and driving up efficiency are improved, and the equipment is reduced Consumption.

CN118823689BActive Publication Date: 2024-11-19TAIYUAN LONGWAY ELECTRONICS SCI & TECH
View PDF 0 Cites 0 Cited by

Patent Information

Application Number
CN202411307914.2
Authority / Receiving Office
CN · China
Patent Type
Patents(China)
Current Assignee / Owner
Filing Date
2024-09-19
Publication Date
2024-11-19
Estimated Expiration
2044-09-19

AI Technical Summary

Technical Problem

The prior art is difficult to quickly identify and effectively drive away birds near the power transmission pole tower, resulting in insulator failure or line discharge, and there is a risk of line tripping.

Method used

Using bird repelling method based on active monitoring and visual recognition, the target scene type of the target monitoring area of ​​the target transmission pole tower is obtained, and the target image is classified using a pre-trained SVM classification model to determine whether there are birds, and the bird repelling is controlled to perform bird repelling. The agency carries out bird repelling.

Benefits of technology

It reduces the data processing volume, improves the accuracy and speed of bird identification, realizes rapid identification and effective driving of birds near the transmission pole tower, and reduces the power consumption of the bird repelling actuator.

✦ Generated by Eureka AI based on patent content.

Smart Images

  • Figure CN118823689B_ABST
    Figure CN118823689B_ABST
Patent Text Reader

Abstract

The present application relates to the technical field of bird repelling for power transmission lines, and in particular to a bird repelling device and method based on active monitoring and visual recognition. The method comprises: obtaining the target scene type of the target monitoring area of ​​the target transmission tower at the current moment; determining the target SVM classification model corresponding to the target scene type from a plurality of pre-trained SVM classification models; obtaining the target image for the target monitoring area at the current moment, and inputting the target image into the target SVM classification model to obtain the classification result of the target image; when the classification result indicates that there are birds in the target image, controlling the bird repelling actuator corresponding to the target transmission tower to repel the birds. Through the above technical scheme, the target image can be classified using the SVM classification model that matches the scene type at the current moment, so as to better repel the birds that may exist near the transmission tower.
Need to check novelty before this filing date? Find Prior Art

Description

Technical Field

[0001] The present application relates to the technical field of bird repelling on power transmission lines, and in particular to a bird repelling device and method based on active monitoring and visual recognition. Background Art

[0002] Transmission towers are tower structures used to support and suspend high-voltage transmission lines. They are used to transmit the electricity generated by power stations to substations through high-voltage transmission lines, and then distribute it to various users by substations. With the improvement of the ecological environment, birds may move around near transmission towers, and may even build nests in the gaps of transmission towers, thereby moving more frequently near transmission towers.

[0003] When birds move around near transmission towers, the bird droppings they excrete may cause failures in insulators installed at the transmission towers. Or, when the bodies of birds pass between transmission lines, they may cause discharges between the lines, posing the risk of causing line tripping.

[0004] In order to avoid the possible threat of bird activities to transmission towers, the relevant technology, for example, the Chinese patent application document with publication number CN117958245A, provides an intelligent directional bird-repelling method in response to obtaining the current time. If the current time is in the night time period, the laser bird-repelling device is instructed to work; if the current time is in the daytime period, video stream data is obtained, and the video stream data is input into the image detection model to obtain the recognition result output by the image detection model; based on the recognition result, the position information corresponding to the target bird is obtained, and according to the recognition result and the position information, the laser bird-repelling device or the sound bird-repelling device is instructed to work.

[0005] The related technology determines whether there is bird activity through video stream data. However, the amount of data processing required to determine whether there are birds through video data streams is large, which is not conducive to quickly identifying the birds near transmission towers and driving away the birds near transmission towers. Therefore, the related technology is difficult to effectively drive away the birds near transmission towers. Summary of the invention

[0006] In order to overcome the problem in the related art that it is difficult to effectively drive away birds near transmission towers, the present application provides a bird-repelling device and method based on active monitoring and visual recognition.

[0007] According to a first aspect of an embodiment of the present application, a bird-repelling method based on active monitoring and visual recognition is provided, including: obtaining a target scene type of a target monitoring area of ​​a target transmission tower at a current moment, wherein the target scene type is determined based on at least one of weather information, season information, and time period information of the target monitoring area at a current moment; determining a target SVM classification model corresponding to the target scene type from a plurality of pre-trained SVM classification models; the SVM classification model is obtained by training based on a target HOG feature corresponding to the scene type, and the target HOG feature is a HOG feature of a bird image under the scene type; obtaining a target image for the target monitoring area at a current moment, and inputting the target image into the target SVM classification model to obtain a classification result of the target image; and when the classification result indicates that there are birds in the target image, controlling a bird-repelling actuator corresponding to the target transmission tower to perform bird repelling.

[0008] In this way, by obtaining the target scene type of the target monitoring area of ​​the target transmission tower at the current moment, and classifying the target image of the target monitoring area through the target SVM classification model corresponding to the target scene type, different SVM classification models can be used for image classification under different scene types, so that the SVM classification model used for image classification can match the scene type at the current moment, thereby better determining whether there are birds in the target monitoring area, so as to drive away birds that may exist near the transmission tower.

[0009] Compared with determining whether there are birds in the target monitoring area through video stream data, when the existing birds are effectively identified, the amount of data processing can be reduced, which helps to quickly identify the birds in the target monitoring area, thereby effectively driving away the birds that may exist near the transmission towers.

[0010] Furthermore, the target scene type is determined based on at least one of the weather information, season information and time period information of the target monitoring area at the current moment. Since the characteristics of bird images may be different under different weather information, season information and time period information, targeted identification of birds can be achieved based on the current weather information, season information and time period information, thereby improving the accuracy of identifying birds that may exist near transmission towers, and helping to better control the bird-repellent actuator to carry out bird-repellent.

[0011] Optionally, the target HOG feature corresponding to the scene type is obtained in the following manner: determining the target window length corresponding to the scene type from a plurality of preset window lengths, different scene types corresponding to different preset window lengths; obtaining a target bird image under the scene type, and determining a HOG feature vector of the target bird image under the target window length; and using the HOG feature vector as the target HOG feature corresponding to the scene type.

[0012] In this way, by determining the target window length corresponding to the scene type, it is helpful to make the target window length used when obtaining the HOG feature match the scene type, and to obtain the HOG features of the bird image under the scene type in a targeted manner, thereby improving the accuracy of the HOG features of the bird image under the scene type obtained.

[0013] Optionally, determining the HOG feature vector of the target bird image at the target window length includes: dividing the target bird image using the target window length to obtain multiple sub-regions of the target bird image, and determining the gradient histogram of the sub-regions; determining the sub-feature vector of the region based on multiple gradient histograms of multiple sub-regions in the same region of the target bird image, wherein the side length of the region is greater than the target window length; determining the HOG feature vector based on multiple sub-feature vectors of multiple regions in the target bird image.

[0014] Optionally, the preset window length is obtained in the following manner: obtaining a bird image under a scene type, and determining a first HOG feature vector of the bird image under an initial preset window length; training an initial SVM classification model using multiple first HOG feature vectors corresponding to multiple bird images under the scene type to obtain a first SVM classification model; determining the classification accuracy of the first SVM classification model for classifying bird images under the scene type, and adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type.

[0015] In this way, the initial SVM classification model is trained using multiple first HOG feature vectors corresponding to multiple bird images under the scene type to obtain the first SVM classification model, which can enable the first SVM classification model to learn the characteristics of the bird images under the scene type. The classification accuracy of the bird images under the scene type classified by the first SVM classification model can verify the application effect of the first SVM classification model in the scene type.

[0016] Optionally, the initial preset window length is adjusted according to the classification accuracy to obtain the preset window length corresponding to the scene type, including: when the classification accuracy is less than a preset accuracy threshold, subtracting a preset step length from the initial preset window length to obtain a first preset window length; determining a second HOG feature vector of the bird image under the first preset window length; using multiple second HOG feature vectors corresponding to multiple bird images under the scene type to train an initial SVM classification model to obtain a second SVM classification model; when the classification accuracy of the second SVM classification model is greater than or equal to a preset accuracy threshold, using the first preset window length as the preset window length.

[0017] In this way, the first preset window length is equal to the difference between the initial preset window length and the preset step length. Compared with the initial preset window length, the first preset window length is smaller. The first preset window length can more accurately describe the characteristics of the bird image, which helps to improve the classification accuracy of bird images under different scene types.

[0018] Optionally, the method also includes: when the classification accuracy of the second SVM classification model is less than a preset accuracy threshold, executing the step of adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type.

[0019] In this way, by executing the step of adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type, a smaller preset window length can be further obtained, and an SVM classification model corresponding to the smaller preset window length can be obtained, so as to obtain an SVM classification model with a classification accuracy greater than or equal to the preset accuracy threshold.

[0020] Optionally, the preset window length is obtained in the following manner: obtaining a bird image under a scene type, and determining multiple third HOG feature vectors of the same bird image under multiple second preset window lengths; using multiple third HOG feature vectors corresponding to the same second preset window length to train an initial SVM classification model to obtain a third SVM classification model; determining an evaluation parameter of the third SVM classification model, wherein the evaluation parameter is determined based on the accuracy and classification time of the third SVM classification model for classifying the bird images under the scene type; and using the second preset window length corresponding to the third SVM classification model with the highest evaluation parameter as the preset window length corresponding to the scene type.

[0021] In this way, by selecting the third SVM classification model with the highest evaluation parameter, the classification time for bird images can be shortened as much as possible while ensuring the classification accuracy; the second preset window length corresponding to the third SVM classification model with the highest evaluation parameter can better obtain the features for accurately describing the bird images under the scene type; using the second preset window length corresponding to the third SVM classification model with the highest evaluation parameter as the preset window length corresponding to the scene type, the preset window length can be matched with the scene type, which helps to better identify and drive away the birds under the scene type.

[0022] Optionally, the initial preset window length is obtained by: obtaining a bird region of a bird image corresponding to a scene type; determining an area of ​​the bird region in the bird image, and determining an average area of ​​multiple area areas corresponding to multiple bird images; taking a square root of the average area to obtain a square root result, and taking a rounded result of the square root result as the initial preset window length.

[0023] In this way, the size of the birds under the scene type can be reflected by taking the average area of ​​multiple regions corresponding to multiple bird images, and the square root of the average area is obtained, and the rounded result of the square root is used as the initial preset window length, so that the initial preset window length can be adapted to the size of the birds under the scene type; since the initial preset window length can be more adapted to the size of the birds under the scene type, it helps to shorten the number of times the preset window length is adjusted, or it helps to shorten the adjustment time of the preset window length, thereby reducing the power consumption of the running device and reducing the running memory pressure.

[0024] Optionally, when the classification result indicates that there are birds in the target image, controlling a bird repellent actuator corresponding to a target transmission tower to repel birds includes: when the classification result indicates that there are birds in the target image, determining the bird repellent actuator corresponding to the target transmission tower; sending a control instruction to the bird repellent actuator so that the bird repellent actuator performs bird repellent, wherein the control instruction is used to control the bird repellent actuator to perform bird repellent.

[0025] In this way, compared with controlling the bird-repelling actuator to periodically perform bird-repelling, bird-repelling can be performed when there are birds in the target monitoring area of ​​the transmission tower, thereby reducing the power consumption of the bird-repelling actuator.

[0026] According to a second aspect of an embodiment of the present application, a bird-repelling device based on active monitoring and visual identification is provided, comprising: a processor and a memory, the memory storing computer program instructions, and the computer program instructions, when executed by the processor, implement the steps of the bird-repelling method based on active monitoring and visual identification provided in the first aspect of the present application.

[0027] The technical solution provided by the embodiments of the present application may include the following beneficial effects: by obtaining the target scene type of the target monitoring area of ​​the target transmission tower at the current moment, and classifying the target image of the target monitoring area by the target SVM classification model corresponding to the target scene type, different SVM classification models can be used for image classification under different scene types, so that the SVM classification model used for image classification can match the scene type at the current moment, thereby better determining whether there are birds in the target monitoring area, so as to drive away birds that may exist near the transmission tower.

[0028] Compared with determining whether there are birds in the target monitoring area through video stream data, when the existing birds are effectively identified, the amount of data processing can be reduced, which helps to quickly identify the birds in the target monitoring area, thereby effectively driving away the birds that may exist near the transmission towers.

[0029] Furthermore, the target scene type is determined based on at least one of the weather information, season information and time period information of the target monitoring area at the current moment. Since the characteristics of bird images may be different under different weather information, season information and time period information, targeted identification of birds can be achieved based on the current weather information, season information and time period information, thereby improving the accuracy of identifying birds that may exist near transmission towers, and helping to better control the bird-repellent actuator to carry out bird-repellent.

[0030] It should be understood that the foregoing general description and the following detailed description are exemplary and explanatory only and are not restrictive of the present application. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] The accompanying drawings, which are incorporated in and constitute a part of this specification, illustrate embodiments consistent with the present application and, together with the description, serve to explain the principles of the present application.

[0032] Figure 1 is a flow chart showing a bird-repelling method based on active monitoring and visual recognition according to an exemplary embodiment;

[0033] Figure 2 It is a structural schematic diagram of a bird-repelling device based on active monitoring and visual recognition according to an exemplary embodiment. DETAILED DESCRIPTION

[0034] Here, exemplary embodiments are described in detail, and examples thereof are shown in the accompanying drawings. When the following description refers to the drawings, unless otherwise indicated, the same numbers in different drawings represent the same or similar elements. The implementations described in the following exemplary embodiments do not represent all implementations consistent with the present application.

[0035] First, the application scenario of the embodiment of the present application is briefly introduced. In the application scenario of the present application, in order to avoid the adverse effects of bird activities on transmission towers, or to avoid the adverse effects of bird activities on transmission lines carried by transmission towers, it is usually necessary to drive away the birds near the transmission towers.

[0036] In the related art, the presence of birds is mainly determined through video stream data near the transmission tower. However, the storage volume, transmission volume and required processing volume of the video stream data are large, which is not conducive to the rapid identification of birds near the transmission tower. Therefore, it is difficult to effectively drive away the birds near the transmission tower.

[0037] In response to the above technical problems, the embodiments of the present application provide a bird-repelling device and method based on active monitoring and visual recognition. By obtaining the target scene type of the target monitoring area of ​​the target transmission tower at the current moment, and classifying the target image of the target monitoring area through the target SVM classification model corresponding to the target scene type, different SVM classification models can be used for image classification under different scene types, and the SVM classification model used for image classification can be matched with the scene type at the current moment, so as to better determine whether there are birds in the target monitoring area, so as to drive away birds that may exist near the transmission tower.

[0038] Compared with determining whether there are birds in the target monitoring area through video stream data, when the existing birds are effectively identified, the amount of data processing can be reduced, which helps to quickly identify the birds in the target monitoring area, thereby effectively driving away the birds that may exist near the transmission towers.

[0039] Furthermore, the target scene type is determined based on at least one of the weather information, season information and time period information of the target monitoring area at the current moment. Since the characteristics of bird images may be different under different weather information, season information and time period information, targeted identification of birds can be achieved based on the current weather information, season information and time period information, thereby improving the accuracy of identifying birds that may exist near transmission towers, and helping to better control the bird-repellent actuator to carry out bird-repellent.

[0040] Figure 1 is a flowchart of a bird-repelling method based on active monitoring and visual recognition according to an exemplary embodiment. The method provided in the embodiment of the present application can be applied to a terminal device, such as Figure 1 As shown, the method includes the following steps.

[0041] In step S101, the target scene type of the target monitoring area of ​​the target transmission tower at the current moment is obtained.

[0042] The target scene type is determined based on at least one of weather information, season information, and time period information of the target monitoring area at the current moment.

[0043] Weather information may be, for example, one of the natural weather conditions such as sunny, rainy, cloudy, and snowy days; season information may be, for example, one of the seasons such as spring, summer, autumn, and winter; time period information may be, for example, one of the time periods such as morning, noon, and evening. The natural day may be divided more finely according to actual needs. For example, a natural day may be divided into 4 time periods or 12 time periods. The embodiment of the present application does not limit the number of divided time periods.

[0044] When at least one of the weather information, season information and time period information is different, the bird images collected by the image acquisition device may be different; for example, the light on the transmission tower in different weather or time periods may be different, resulting in different bird images obtained; for another example, the birds that are active in different seasons or time periods are different, resulting in different birds active near the transmission tower in different seasons or time periods.

[0045] If the same bird identification strategy is used for different weather information, seasonal information, and time period information, the accuracy of bird identification will be low, making it difficult to effectively drive away birds.

[0046] For example, due to the low accuracy of bird recognition, it may be mistakenly believed that there are birds near the transmission tower when there are no birds near the transmission tower, and the birds are driven away when there are no birds near the transmission tower, resulting in a waste of power consumption for the bird-repelling equipment.

[0047] For another example, due to the low accuracy of bird identification, when there are birds near the transmission towers, people may mistakenly believe that there are no birds near the transmission towers. As a result, birds are not driven away when there are birds near the transmission towers. Birds move around near the transmission towers, posing a safety hazard to the transmission towers or the transmission lines carried by the transmission towers.

[0048] In this way, obtaining the target scene type of the target monitoring area of ​​the target transmission tower at the current moment helps to adopt a bird recognition strategy that matches the target scene type at the current moment, thereby improving the accuracy of driving away birds.

[0049] In step S102, a target SVM classification model corresponding to the target scene type is determined from a plurality of pre-trained SVM classification models.

[0050] The SVM classification model is obtained by training according to the corresponding target HOG features, and the target HOG features are the HOG features of the bird images under the scene type.

[0051] The bird images can be classified by a pre-trained SVM (Support Vector Machine) classifier. For example, the bird images can be classified into one of images containing birds and images not containing birds, so as to determine whether the bird images contain birds.

[0052] HOG (Histogram of Oriented Gradients) can better describe the characteristics of an image so as to determine whether there are birds in the image. Since the HOG feature can better describe the characteristics of an image, an SVM classification model is obtained by training the HOG features of bird images under the scene type, which can facilitate the use of the trained SVM classification model to obtain classification results for the image.

[0053] For example, labels can be added to bird images under scene types in advance. For example, the labels can be divided into two types: containing birds and not containing birds. The HOG features of the bird images can be used as input, and the labels of the bird images can be used as input to train the pre-built initial SVM classification model. When the classification results output by the initial SVM classification model meet the preset conditions, the training of the initial SVM classification model can be stopped, and the trained initial SVM classification model can be used as the trained SVM classification model.

[0054] In step S103, a target image for the target monitoring area at the current moment is obtained, and the target image is input into a target SVM classification model to obtain a classification result for the target image.

[0055] By setting up one or more image acquisition devices at the transmission tower, or by setting up a drone with one or more image acquisition devices and controlling the drone to fly to the target monitoring area, the target image of the target monitoring area at the current moment can be obtained; the target monitoring area can be any monitoring area within a preset range of the target transmission tower, for example, the target monitoring area can be a monitoring area within 5 meters near the target transmission tower.

[0056] The target image is input into the target SVM classification model. Since the target SVM classification model is an SVM classification model that matches the scene type at the current moment, a more accurate classification result for the target image can be obtained through the target SVM classification model.

[0057] For example, when the season represented by the seasonal information is spring, the target SVM classification model can be obtained by training based on the HOG features of bird images in spring. Since the HOG features of bird images in spring include the features of birds active in spring, or the HOG features of bird images in spring include the features of birds under spring lighting conditions, inputting the target image into the target SVM classification model can better realize bird recognition in spring and improve the accuracy of bird recognition.

[0058] For another example, when the season represented by the weather information is sunny, the target SVM classification model can be trained based on the HOG features of bird images on sunny days. Since the HOG features of bird images on sunny days include the features of birds active on sunny days, or the HOG features of bird images on sunny days include the features of birds under the lighting conditions of sunny days, inputting the target image into the target SVM classification model can better realize the recognition of birds on sunny days and improve the accuracy of bird recognition.

[0059] In step S104, when the classification result indicates that there are birds in the target image, the bird-repelling actuator corresponding to the target transmission tower is controlled to repel birds.

[0060] When the classification result indicates that there are birds in the target image, it means that there is bird activity in the target monitoring area. In order to avoid the adverse effects of bird activities on transmission towers, or to avoid the adverse effects of bird activities on transmission lines carried by transmission towers, the bird repellent actuator corresponding to the target transmission tower can be controlled to repel birds, so as to avoid bird activities near the target transmission tower.

[0061] The bird-repelling actuator may be a device capable of repelling birds. The bird-repelling actuator may be a sonic bird repellent device. The sonic bird repellent device may emit sounds that birds do not like. For example, the sonic bird repellent device may output the call of an owl or ultrasonic waves.

[0062] Alternatively, the bird-repelling actuator may be a visual bird-repelling device, which may be provided with a flashing light, a reflective strip, a windmill, a kite, or other components that can scare birds by flashing or moving; or, the bird-repelling actuator may be able to spray out bird-repelling spices or liquids to drive away birds. The specific type of the bird-repelling actuator is not limited in the embodiments of the present application.

[0063] In one embodiment, when the classification result indicates that there are birds in the target image, controlling a bird repellent actuator corresponding to a target transmission tower to repel birds comprises: when the classification result indicates that there are birds in the target image, determining the bird repellent actuator corresponding to the target transmission tower; sending a control instruction to the bird repellent actuator so that the bird repellent actuator performs bird repellent, wherein the control instruction is used to control the bird repellent actuator to perform bird repellent.

[0064] The bird-repellent actuator can be set on a transmission tower; or, the bird-repellent actuator can also be a device independent of the transmission tower, such as a low-altitude flying device such as a drone used for bird repelling. The same bird-repellent actuator can correspond to multiple transmission towers, or the same bird-repellent actuator can correspond to one transmission tower. Therefore, the terminal device can determine the bird-repellent actuator corresponding to the target transmission tower so as to repel birds from the target monitoring area of ​​the target transmission tower; the bird-repellent actuator can be connected to the terminal device for communication, and the bird-repellent actuator can repel birds from the target monitoring area of ​​the target transmission tower according to the received control instructions.

[0065] In this way, compared with controlling the bird-repelling actuator to periodically perform bird-repelling, bird-repelling can be performed when there are birds in the target monitoring area of ​​the transmission tower, thereby reducing the power consumption of the bird-repelling actuator.

[0066] For example, when the bird-repelling actuator repels birds by spraying fragrances or liquids, the amount of the sprayed fragrances or liquids can be reduced while ensuring the bird-repelling effect.

[0067] Through the bird-repelling method based on active monitoring and visual recognition provided in the embodiment of the present application, the target scene type of the target monitoring area of ​​the target transmission tower at the current moment is obtained, and the target image of the target monitoring area is classified by the target SVM classification model corresponding to the target scene type. Different SVM classification models can be used for image classification under different scene types, and the SVM classification model used for image classification can be matched with the scene type at the current moment, so as to better determine whether there are birds in the target monitoring area, so as to drive away birds that may exist near the transmission tower.

[0068] Compared with determining whether there are birds in the target monitoring area through video stream data, when the existing birds are effectively identified, the amount of data processing can be reduced, which helps to quickly identify the birds in the target monitoring area, thereby effectively driving away the birds that may exist near the transmission towers.

[0069] Furthermore, the target scene type is determined based on at least one of the weather information, season information and time period information of the target monitoring area at the current moment. Since the characteristics of bird images may be different under different weather information, season information and time period information, targeted identification of birds can be achieved based on the current weather information, season information and time period information, thereby improving the accuracy of identifying birds that may exist near transmission towers, and helping to better control the bird-repellent actuator to carry out bird-repellent.

[0070] In one embodiment, the target HOG feature corresponding to the scene type is obtained in the following manner: determining the target window length corresponding to the scene type from a plurality of preset window lengths, different scene types corresponding to different preset window lengths; obtaining a target bird image under the scene type, and determining a HOG feature vector of the target bird image under the target window length; and using the HOG feature vector as the target HOG feature corresponding to the scene type.

[0071] For example, since the characteristics of birds are different in different scene types, or the lighting conditions are different in different scene types, when obtaining the HOG features of bird images, different scene types require different window lengths.

[0072] In this way, by determining the target window length corresponding to the scene type, it is helpful to make the target window length used when obtaining the HOG feature match the scene type, and to obtain the HOG features of the bird image under the scene type in a targeted manner, thereby improving the accuracy of the HOG features of the bird image under the scene type obtained.

[0073] In one embodiment, determining the HOG feature vector of the target bird image at the target window length includes: dividing the target bird image using the target window length to obtain multiple sub-regions of the target bird image, and determining the gradient histograms of the sub-regions; determining the sub-feature vector of the region according to multiple gradient histograms of multiple sub-regions in the same region of the target bird image, wherein the side length of the region is greater than the target window length; determining the HOG feature vector according to multiple sub-feature vectors of multiple regions in the target bird image.

[0074] For example, when the target window length is 5, the size of the divided sub-region is 5×5. By determining the gradient histogram of the sub-region with a size of 5×5, the characteristics of the sub-region can be described; for example, the gradient direction of each pixel in the sub-region with a size of 5×5 can be determined respectively, and the frequency of all gradient directions of all pixels in the sub-region can be counted to obtain the gradient histogram of the corresponding sub-region, and the gradient histogram includes the frequency of occurrence of each gradient direction.

[0075] The gradient direction of a pixel point refers to the direction in which the brightness of a pixel in an image changes, for example, the direction in which the grayscale value of a pixel point at a corresponding position in a grayscale image changes.

[0076] For example, the gradient range of [0°, 180°] can be equally divided into 9 sub-gradient ranges, and the frequency of pixel points whose gradient directions fall into the sub-gradient ranges in the sub-regions can be counted. For example, the statistical result is [5, 3, 2, 0, 2, 1, 3, 4, 5], and the gradient histogram can be normalized.

[0077] The side length of the region in the target bird image can be an integer multiple of the window length. For example, the same region in the target bird image can include 4 sub-regions arranged in a square. For example, the 4 sub-regions included in the same region are arranged in 2 rows and 2 columns. The sub-feature vector corresponding to the region can be obtained through the 4 gradient histograms of the 4 sub-regions in the region; for example, the vector corresponding to the gradient histogram can be determined, and the multiple vectors corresponding to the multiple gradient histograms can be vector-joined to obtain the sub-feature vector corresponding to the region.

[0078] In one embodiment, the preset window length is obtained in the following manner: obtaining a bird image under a scene type, and determining a first HOG feature vector of the bird image under an initial preset window length; using multiple first HOG feature vectors corresponding to multiple bird images under the scene type to train an initial SVM classification model to obtain a first SVM classification model; determining the classification accuracy of the first SVM classification model for classifying the bird images under the scene type, and adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type.

[0079] In this way, the initial SVM classification model is trained using multiple first HOG feature vectors corresponding to multiple bird images under the scene type to obtain the first SVM classification model, which can enable the first SVM classification model to learn the characteristics of the bird images under the scene type. The classification accuracy of the bird images under the scene type classified by the first SVM classification model can verify the application effect of the first SVM classification model in the scene type.

[0080] The bird images pre-classified under the scene type may be identified by the first SVM classification model to obtain the classification accuracy of the bird images under the scene type classified by the first SVM classification model.

[0081] For example, when the first SVM classification model classifies a bird image that has been pre-classified as containing birds, if the recognition result indicates that the bird image contains birds, it means that the first SVM classification model has correctly classified the bird image; for another example, when the first SVM classification model classifies a bird image that has been pre-classified as not containing birds, if the recognition result indicates that the bird image does not contain birds, it means that the first SVM classification model has incorrectly classified the bird image.

[0082] For another example, when the first SVM classification model classifies a bird image that has been pre-classified as not containing birds, if the recognition result indicates that the bird image contains birds, it means that the first SVM classification model has incorrectly classified the bird image. For another example, when the first SVM classification model classifies a bird image that has been pre-classified as containing birds, if the recognition result indicates that the bird image does not contain birds, it means that the first SVM classification model has incorrectly classified the bird image.

[0083] The bird images pre-classified under the scene type can be identified, and the proportion of the bird images classified correctly can be used as the classification accuracy of the first SVM classification model for classifying the bird images under the scene type.

[0084] In an embodiment of the present application, the bird image used to determine the classification accuracy of the first SVM classification model for classifying bird images under the scene type may be an image different from the bird image corresponding to the HOG feature used to train the first SVM classification model. In this way, the accuracy of the first SVM classification model can be better tested.

[0085] Here, since the first SVM classification model is obtained by training based on the first HOG feature vector under the initial preset window length, and the window lengths corresponding to the HOG features applicable to different scene types are different, the first HOG feature vector under the initial preset window length may not be able to match the bird image under the scene type, which may cause the classification effect of the first SVM classification model obtained by training to be poor. Therefore, the initial preset window length can be adjusted according to the classification accuracy to obtain the preset window length corresponding to the scene type, so that the preset window length matches the scene type.

[0086] In one embodiment, the initial preset window length is adjusted according to the classification accuracy to obtain the preset window length corresponding to the scene type, including: when the classification accuracy is less than the preset accuracy threshold, the initial preset window length is subtracted from the preset step length to obtain the first preset window length; the second HOG feature vector of the bird image under the first preset window length is determined; the initial SVM classification model is trained using multiple second HOG feature vectors corresponding to multiple bird images under the scene type to obtain the second SVM classification model; when the classification accuracy of the second SVM classification model is greater than or equal to the preset accuracy threshold, the first preset window length is used as the preset window length.

[0087] When the classification accuracy is less than the preset accuracy threshold, it means that the SVM classification model has a poor recognition effect on the bird images under the scene type, and the HOG corresponding to a smaller window length can usually more accurately describe the characteristics of the image. Therefore, the preset window length can be reduced based on the initial preset window length to obtain a first preset window length; wherein the preset accuracy threshold can be between 75% and 90%, for example, the preset accuracy threshold can be equal to 75%, and the preset step size can be 1 or 2, for example.

[0088] In this way, the first preset window length is equal to the difference between the initial preset window length and the preset step length. Compared with the initial preset window length, the first preset window length is smaller. The first preset window length can more accurately describe the characteristics of the bird image, which helps to improve the classification accuracy of bird images under different scene types.

[0089] The initial SVM classification model is trained using multiple second HOG feature vectors corresponding to multiple bird images under scene types to obtain a second SVM classification model. Since the second HOG feature vector is the HOG feature vector of the bird image under the first preset window length, and the first preset window length is smaller than the initial preset window length, the amount of information contained in the second HOG feature vector is larger than the amount of information contained in the first HOG feature vector. The second SVM classification model can achieve more accurate recognition of bird images under scene types.

[0090] When the classification accuracy of the second SVM classification model is greater than or equal to the preset accuracy threshold, it means that after adjusting the initial preset window length, the second SVM classification model obtained has a good classification effect. Using the first preset window length as the preset window length can ensure that the preset window length can better describe the characteristics of the bird image under the scene type, while avoiding the increase in calculation amount caused by further reduction of the window length, which helps to reduce the power consumption of the device.

[0091] In another embodiment, when the classification accuracy of the second SVM classification model is less than a preset accuracy threshold, the step of adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type is performed.

[0092] When the classification accuracy of the second SVM classification model is less than the preset accuracy threshold, it means that after adjusting the initial preset window length, the second HOG feature vector corresponding to the first preset window length obtained after the adjustment is still difficult to accurately describe the features in the bird image under the scene type, so that the classification effect of the second SVM classification model obtained by training according to the second HOG feature vector is still poor.

[0093] Here, by executing the step of adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type, a smaller preset window length can be further obtained, and an SVM classification model corresponding to the smaller preset window length can be obtained, so as to obtain an SVM classification model with a classification accuracy greater than or equal to a preset accuracy threshold.

[0094] In one embodiment, the preset window length is obtained by: obtaining a bird image under a scene type, and determining multiple third HOG feature vectors of the same bird image under multiple second preset window lengths; using multiple third HOG feature vectors corresponding to the same second preset window length to train an initial SVM classification model to obtain a third SVM classification model; determining an evaluation parameter of the third SVM classification model, the evaluation parameter being determined based on the accuracy and classification time of the third SVM classification model for classifying the bird images under the scene type; and using the second preset window length corresponding to the third SVM classification model with the highest evaluation parameter as the preset window length corresponding to the scene type.

[0095] For example, for bird images under scene types, different SVM classifiers corresponding to different window lengths can be obtained, and different SVM classifiers corresponding to different window lengths can be used to identify bird images under scene types; through the evaluation parameters of the SVM classifier, the accuracy and classification time of SVM classification under scene types can be comprehensively characterized, thereby taking into account both the classification accuracy and the classification time, so as to shorten the classification time of bird images as much as possible while ensuring the classification accuracy.

[0096] In this way, by selecting the third SVM classification model with the highest evaluation parameter, the classification time for bird images can be shortened as much as possible while ensuring the classification accuracy; the second preset window length corresponding to the third SVM classification model with the highest evaluation parameter can better obtain the features for accurately describing the bird images under the scene type; using the second preset window length corresponding to the third SVM classification model with the highest evaluation parameter as the preset window length corresponding to the scene type, the preset window length can be matched with the scene type, which helps to better identify and drive away the birds under the scene type.

[0097] In one embodiment, the evaluation parameters of the third SVM classification model can be determined in the following manner: obtaining the average classification time of the third SVM classification model for classifying bird images under the scene type; obtaining the classification accuracy of the third SVM classification model for classifying bird images under the scene type; and using the ratio of the classification accuracy and the average classification time as the evaluation parameter of the third SVM classification model.

[0098] In this way, using the ratio of classification accuracy and average classification time as the evaluation parameter of the third SVM classification model can comprehensively consider the influence of classification accuracy and classification time, which is helpful to screen out a preset window length that is more suitable for the scene type.

[0099] Compared with the change in the recognition accuracy of the SVM classification model caused by the change in the preset window length, when the preset window length changes, the extraction time of the HOG features of the bird image corresponding to the preset window length will change more obviously. In order to reduce the extraction time of the HOG features, the extraction time of the HOG features can also be taken into consideration for the SVM model.

[0100] For example, the evaluation parameters of the SVM model may also be determined based on the classification accuracy, the classification time, and the extraction time of the corresponding HOG features.

[0101] In one embodiment, the initial preset window length is obtained by: obtaining a bird region of a bird image corresponding to a scene type; determining an area of ​​the bird region in the bird image, and determining an average area of ​​multiple area areas corresponding to multiple bird images; performing square root operation on the average area to obtain a square root result, and taking the rounded result of the square root result as the initial preset window length.

[0102] In this way, the size of the birds under the scene type can be reflected by taking the average area of ​​multiple regions corresponding to multiple bird images, and the square root of the average area is obtained, and the rounded result of the square root is used as the initial preset window length, so that the initial preset window length can be adapted to the size of the birds under the scene type; since the initial preset window length can be more adapted to the size of the birds under the scene type, it helps to shorten the number of times the preset window length is adjusted, or it helps to shorten the adjustment time of the preset window length, thereby reducing the power consumption of the running device and reducing the running memory pressure.

[0103] Figure 2 is a schematic diagram of a bird-repelling device based on active monitoring and visual recognition according to an exemplary embodiment. Figure 2 The bird-repelling device 1000 based on active monitoring and visual recognition includes: a processor 1100 and a memory 1200, wherein the memory 1200 stores computer program instructions, and when the computer program instructions are executed by the processor 1100, all or part of the steps of the bird-repelling method based on active monitoring and visual recognition in the present application are implemented.

[0104] It should be understood that the features of some embodiments of the various applications described herein may be combined with each other, unless specifically stated otherwise.

[0105] Although terms such as "first", "second", and "third" may be used herein to describe various components, parts, regions, layers, or sections, these components, parts, regions, layers, or sections are not limited to these terms. Instead, these terms are only used to distinguish one component, part, region, layer, or section from another component, part, region, layer, or section. Therefore, without departing from the teachings of the examples described herein, the first component, part, region, layer, or section mentioned in the examples may also be referred to as the second component, part, region, layer, or section.

[0106] In addition, the terms "first" and "second" are used for descriptive purposes only and should not be understood as indicating or implying relative importance or implicitly indicating the number of the indicated technical features. Therefore, the features defined as "first" and "second" may explicitly or implicitly include at least one of the features. In the description herein, the meaning of "plurality" is at least two, such as two, three, etc., unless otherwise clearly and specifically defined.

[0107] In addition, the word "exemplary" is used herein to mean serving as an example, instance, or diagram. Any aspect or design described herein as "exemplary" is not necessarily to be understood as being advantageous over other aspects or designs. Instead, the use of the word exemplary is intended to present concepts in a concrete manner. As used herein, the term "or" is intended to mean an inclusive "or" rather than an exclusive "or".

[0108] Likewise, although the present application has been shown and described with respect to one or more implementations, equivalent variations and modifications will occur to those skilled in the art upon reading and understanding this specification and the accompanying drawings. With particular regard to the various functions performed by the components (e.g., elements, resources, etc.) described above, unless otherwise noted, the terms used to describe such components are intended to correspond to any component (functionally equivalent) that performs the specific functions of the described components, even if not structurally equivalent to the disclosed structures.

[0109] Additionally, while particular features of the present application may have been disclosed with respect to only one of several implementations, such features may be combined with one or more other features of other implementations as may be desired and advantageous for any given or particular application.

[0110] Those skilled in the art will readily appreciate other embodiments of the present application after considering the specification and practicing the invention disclosed herein. The present application is intended to cover any modification, use or adaptation of the present application, which follows the general principles of the present application and includes common knowledge or customary technical means in the art that are not disclosed in the present application, and the specification and embodiments are only considered as exemplary.

[0111] It should be understood that the present application is not limited to the exact construction that has been described above and shown in the drawings, and that various modifications and changes may be made without departing from the scope thereof.

Claims

1. A bird-repelling method based on active monitoring and visual recognition, characterized in that: include: Obtaining a target scene type of a target monitoring area of ​​a target transmission tower at a current moment, wherein the target scene type is determined based on at least one of weather information, season information, and time period information of the target monitoring area at a current moment; Determining a target SVM classification model corresponding to the target scene type from a plurality of pre-trained SVM classification models; the SVM classification model is obtained by training according to a target HOG feature corresponding to the scene type, and the target HOG feature is a HOG feature of a bird image under the scene type; Acquire a target image for the target monitoring area at the current moment, and input the target image into the target SVM classification model to obtain a classification result for the target image; When the classification result indicates that there are birds in the target image, controlling a bird-repelling actuator corresponding to the target transmission tower to repel birds; The target HOG feature corresponding to the scene type is obtained in the following manner: determining a target window length corresponding to the scene type from a plurality of preset window lengths, where different scene types correspond to different preset window lengths; obtaining a target bird image under the scene type, and determining a HOG feature vector of the target bird image under the target window length; and using the HOG feature vector as the target HOG feature corresponding to the scene type; The preset window length is obtained in the following manner: obtaining a bird image under a scene type, and determining a first HOG feature vector of the bird image under an initial preset window length; using multiple first HOG feature vectors corresponding to multiple bird images under the scene type to train an initial SVM classification model to obtain a first SVM classification model; determining the classification accuracy of the first SVM classification model for classifying the bird images under the scene type, and adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type.

2. The bird-repelling method based on active monitoring and visual recognition according to claim 1, characterized in that: Determining the HOG feature vector of the target bird image at the target window length includes: Dividing the target bird image by using the target window length to obtain a plurality of sub-regions of the target bird image, and determining the gradient histograms of the sub-regions; Determine a sub-feature vector of a region according to a plurality of gradient histograms of a plurality of sub-regions in the same region of the target bird image, wherein the side length of the region is greater than the target window length; The HOG feature vector is determined according to a plurality of sub-feature vectors of a plurality of regions in the target bird image.

3. The bird-repelling method based on active monitoring and visual recognition according to claim 1, characterized in that: Adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type includes: When the classification accuracy is less than a preset accuracy threshold, subtracting a preset step length from the initial preset window length to obtain a first preset window length; Determine a second HOG feature vector of the bird image under the first preset window length; Using multiple second HOG feature vectors corresponding to multiple bird images under the scene type, the initial SVM classification model is trained to obtain a second SVM classification model; When the classification accuracy of the second SVM classification model is greater than or equal to the preset accuracy threshold, the first preset window length is used as the preset window length.

4. The bird-repelling method based on active monitoring and visual recognition according to claim 3, characterized in that: The method further comprises: When the classification accuracy of the second SVM classification model is less than the preset accuracy threshold, the step of adjusting the initial preset window length according to the classification accuracy to obtain the preset window length corresponding to the scene type is performed.

5. The bird-repelling method based on active monitoring and visual recognition according to any one of claims 1, 3 and 4, characterized in that: The initial preset window length is obtained by: Obtaining a bird region of a bird image corresponding to a scene type; determining a region area of ​​a bird region in the bird image, and determining an average area of ​​a plurality of region areas corresponding to a plurality of bird images; The average area is squared to obtain a square root result, and a rounded result of the square root result is used as the initial preset window length.

6. The bird-repelling method based on active monitoring and visual recognition according to claim 1, characterized in that: When the classification result indicates that there are birds in the target image, controlling the bird-repelling actuator corresponding to the target transmission tower to repel birds includes: When the classification result indicates that there are birds in the target image, determining a bird-repelling actuator corresponding to the target transmission tower; A control instruction is sent to the bird-repelling actuator to enable the bird-repelling actuator to repel birds, wherein the control instruction is used to control the bird-repelling actuator to repel birds.

7. A bird-repelling device based on active monitoring and visual recognition, characterized in that: include: A processor and a memory, wherein the memory stores computer program instructions, and when the computer program instructions are executed by the processor, the bird-repelling method based on active monitoring and visual recognition according to any one of claims 1 to 6 is implemented.