A bionic vision optimization method and imaging device adapting to GIS internal signal strength

Abstract

The application discloses a bionic vision optimization method and imaging device suitable for adapting to GIS internal signal strength, and belongs to the field of GIS internal discharge imaging technology.The method comprises the following steps: constructing an event pulse stream acquisition model based on bionic vision perception, collecting event pulse stream signals caused by different discharge phenomena in GIS; preprocessing the event pulse stream signals and calculating bionic vision sensor imaging accuracy; performing offline learning, optimizing the bionic vision light intensity change threshold, and outputting the optimal light intensity change threshold of all to-be-learned discharge phenomena; and utilizing the event pulse stream acquisition model to detect the light intensity change in GIS in real time and perceive different discharge phenomena in GIS.The application designs a bionic vision optimization method which can realize closed-loop feedback optimization between offline learning and online application, has high imaging accuracy and conforms to actual conditions, can accurately detect different discharge phenomena of GIS, and improves the GIS device imaging efficiency and the imaging recognition accuracy of GIS discharge phenomena.

Description

TECHNICAL FIELD

[0001] The present application belongs to the field of discharge detection, and more particularly, relates to a bionic vision optimization method and imaging device adapted to GIS internal signal strength. BACKGROUND

[0002] With the continuous development of new power systems towards digitization and large-scale, the reliability of power equipment, especially high-voltage switchgear, is becoming increasingly demanding. As a core component of the power grid, gas insulated metal enclosed switchgear (GIS) has been widely used in the control and protection of high-voltage switchgear in power systems due to its small footprint and high reliability. However, due to problems such as insulation material aging, power system overload, and equipment insulation defects, various internal discharge phenomena, including partial discharge, corona discharge, surface discharge, and breakdown discharge, may occur inside GIS, which can easily cause insulation breakdown, equipment failure, and electromagnetic interference, etc., seriously endangering the safe and stable operation of the power system. At the same time, different types of discharge phenomena produce signals of different intensities, for example, partial discharge usually produces low-intensity signals, while breakdown discharge produces high-intensity signals. Therefore, to ensure the reliable and safe operation of GIS equipment, it is urgent to develop high-precision discharge detection methods for different signal intensity discharge phenomena.

[0003] Bionic vision perception technology can capture the optical signals radiated outward by GIS partial discharge by simulating the functions and principles of the biological visual system, and analyze its characteristics and state using image processing and pattern recognition algorithms, achieving real-time detection and identification of GIS internal partial discharge phenomena. It has the characteristics of strong anti-interference, high detection sensitivity and accuracy.

[0004] Some research has already been conducted on monitoring discharge phenomena within GIS systems using image sensors at varying signal intensities. Patent No. 202211400731.6 proposes a GIS partial discharge detection device based on ultraviolet-infrared fusion imaging. This method achieves high sensitivity and reliability for GIS partial discharge detection, visualizes discharge signals, and resists electromagnetic interference. Patent No. 202311509376.0 proposes a multi-signal fusion monitoring method for partial discharge faults in GIS pot-type insulators. Using an ultraviolet imaging detector and a photomultiplier tube, this method monitors the optical signals of GIS pot-type insulators in real time without power outages, providing reliable, accurate, and comprehensive monitoring data for assessing the operating status of GIS pot-type insulators. However, this method fails to account for the differences in discharge light intensity thresholds caused by factors such as reflection and obstruction within the GIS equipment at various signal intensities. This method is unable to dynamically adjust the threshold based on the differentiated discharge phenomena within the GIS, resulting in lengthy imaging detection times and low accuracy. At the same time, the above methods often rely solely on the results of offline training and are difficult to respond to new discharge phenomena generated inside the GIS. They are unable to learn the optimal light intensity change threshold based on new discharge phenomena collected in online applications, and thus cannot achieve closed-loop feedback optimization between offline learning and online applications, affecting the imaging accuracy of discharge phenomena inside the GIS. Summary of the Invention

[0005] In order to solve the shortcomings of the existing technology, such as the inability to dynamically adjust the threshold based on the differentiated discharge phenomenon inside GIS, long imaging detection time and low accuracy, the present invention provides a bionic vision optimization method and imaging device that adapts to the internal signal strength of GIS.

[0006] The present invention adopts the following technical solutions.

[0007] The first aspect of the present invention proposes a bionic visual optimization method for adapting the internal signal strength of a GIS, comprising the following steps:

[0008] Construct an event pulse flow acquisition model based on bionic visual perception to collect event pulse flow signals caused by different discharge phenomena inside the GIS, and preset the real event pulse flow based on historical discharge phenomena;

[0009] Preprocessing the event pulse stream signal, and calculating the imaging accuracy of the bionic visual sensor to obtain a set M of discharge phenomena to be learned;

[0010] Offline learning is performed on the set M of discharge phenomena to be learned, and a reward function and a loss function are constructed. The event pulse stream in the set M of discharge phenomena to be learned is iteratively trained to optimize the threshold of bionic visual light intensity change, and the optimal light intensity change threshold of all discharge phenomena to be learned is output;

[0011] Based on the optimal light intensity change threshold set, the event pulse stream acquisition model is used to detect the light intensity changes inside the GIS in real time and perceive different discharge phenomena inside the GIS.

[0012] Preferably, the event pulse stream acquisition model includes an event camera deployed inside the GIS and including ultraviolet, infrared and visible light visual sensors. The visual sensors respond to changes in light intensity in discharge phenomena inside the GIS in the form of asynchronous sparse events. When the light intensity of a pixel at the current moment changes by more than a threshold compared to the preset standard light intensity of the pixel, the visual sensor outputs an event pulse at the pixel. All event pulses output by each visual sensor constitute the event pulse stream under the spectrum.

[0013] Preferably, the set of different types of discharge phenomena occurring inside the GIS is defined as E = {e1,…,e o ,…,e O}, e o , is the oth discharge phenomenon, O is the total number of different types of discharge phenomena;

[0014] The discharge phenomena collected by ultraviolet vision sensor, infrared vision sensor and visible light vision sensor are o The generated event pulse stream is recorded as:

[0015]

[0016] Wherein, L = {UV, IR, VIS}, UV represents ultraviolet vision sensor, IR represents infrared vision sensor, and VIS represents visible light vision sensor. is the nth event in the event pulse stream generated by the sensor represented by L, and N is the total number of event pulses generated by the sensor represented by L;

[0017] Each event pulse is represented in the form of a quaternion:

[0018]

[0019] in, for The two-dimensional pixel position of for The trigger time; for The polarity of When the value of the pixel is greater than the value of the pixel in the previous frame, it represents that the light intensity at the pixel is enhanced, and when the value of the pixel is less than the value of the pixel in the previous frame, it represents that the light intensity at the pixel is weakened. When the value of the pixel is greater than the value of the pixel in the previous frame, it represents that the light intensity at the pixel is enhanced, and when the value of the pixel is less than the value of the pixel in the previous frame, it represents that the light intensity at the pixel is weakened.

[0020] Preferably, the preprocessing refers to slicing the event pulse stream generated by the visual sensor with the real event pulse stream at a reference frame rate f to obtain a set of sliced images.

[0021] Preferably, the calculation of the biomimetic visual imaging accuracy calculation model specifically includes:

[0022] Calculating the mutual information of the normalized corresponding images in the set of acquired event pulse stream slice images and the set of event pulse stream slice images obtained after slicing the event pulse stream generated by the visual sensor and the real event pulse stream;

[0023]

[0024] Wherein, L={UV, IR, VIS}, UV represents an ultraviolet visual sensor, IR represents an infrared visual sensor, and VIS represents a visible light visual sensor; respectively, are the lth images in the set of acquired event pulse stream slice images and the set of event pulse stream slice images obtained after slicing by the sensor corresponding to L; is the mutual information of the images and the images respectively, are the entropies of the images and the images

[0025] The normalized mutual information of different visual sensors is weighted and summed to obtain the imaging accuracy The calculation formula is:

[0026]

[0027] Wherein, respectively, are the total number of images in the set of acquired event pulse stream slice images obtained after slicing by the ultraviolet, infrared, and visible light sensors, and respectively, are the weights of the ultraviolet, infrared, and visible light sensors set to collect the discharge phenomenon e o , and the value range is 0 to 1, and

[0028] Preferably, the optimization of the biomimetic visual light intensity change threshold is implemented, specifically as follows:

[0029] The optimization target P1 formula is:

[0030] ​​

[0031] Where, and are the light intensity change threshold sets of the ultraviolet, infrared and visible light bionic sensors for all discharge phenomena, They are discharge phenomena e o The light intensity change thresholds of the ultraviolet, infrared and visible light bionic sensors must be constrained between the minimum and maximum light intensity change thresholds;

[0032] The optimization objective of maximizing the average imaging accuracy is transformed into maximizing the imaging accuracy of each discharge phenomenon:

[0033]

[0034] Preferably, the reward function constructed is specifically the discharge phenomenon e o Imaging accuracy i is the number of iterations.

[0035] Preferably, the constructed loss function is specifically:

[0036]

[0037] Where,

[0038]

[0039] Where γ is the discount factor; Q(S(i),ρ(i); θ main ) is the output of the main network; Q(S(i+1),ρ(i+1); θ target ) is the output of the target network; S(i+1) and S(i) are the states at iterations i+1 and i, respectively; ρ(i+1) and ρ(i) are the actions at iterations i+1 and i, respectively; Γ(i) represents the experience replay pool; Represents some samples randomly drawn from the experience replay pool, for The number of samples in θ; main represents the main network, θ target represents the target network;

[0040] Based on this loss function, the main network is updated. Where μ represents the update step size.

[0041] Preferably, the obtained optimal light intensity change threshold set is used to detect the light intensity change inside the GIS in real time using the event pulse stream acquisition model to sense different discharge phenomena inside the GIS:

[0042] The event camera keeps detecting a state, when a light intensity change exceeding a minimum light intensity change threshold in a set of optimal light intensity change thresholds, collects an event pulse stream signal caused by the discharge signal, and pre-processes to obtain a set of perception pulse stream slice images;

[0043] The imaging accuracy of the set of perception pulse stream slice images and all known real event pulse streams of discharge phenomena is calculated, and whether it is a known discharge phenomenon is judged according to the imaging accuracy rate;

[0044] If it is a known discharge phenomenon, the occurring discharge phenomenon is the discharge phenomenon with the highest imaging accuracy rate, the optimal light intensity change threshold of the occurring discharge phenomenon is selected to collect the discharge phenomenon occurring in the current GIS, and the corresponding event pulse stream is generated, and if the light intensity change exceeding the optimal light intensity change threshold of the occurring discharge phenomenon is not detected within the set maximum waiting time The event pulse stream perception ends this time;

[0045] If it is not a known discharge phenomenon, closed-loop feedback learning is performed, the light intensity change threshold of the phenomenon with the highest imaging accuracy rate is used to collect the current discharge phenomenon, and the corresponding event pulse stream is generated, which is put into the set of discharge phenomena to be learned M, and the optimal threshold learning of the set of discharge phenomena to be learned M is performed again.

[0046] Preferably, the specific scheme for judging whether it is a known discharge phenomenon is that the imaging accuracy of the set of perception pulse stream slice images and all known real event pulse streams of discharge phenomena is compared with a set minimum matching threshold Ω min When one or more imaging accuracy rates are higher than the set minimum matching threshold Ω min It is considered that a known discharge phenomenon occurs in the GIS at this time; when all imaging accuracy rates are lower than the minimum matching threshold Ω min It is considered that a new discharge phenomenon occurs.

[0047] The second aspect of the application proposes a bionic visual imaging device for adapting the signal strength in the GIS, which uses the method of the first aspect of the application, comprising an event camera module, a preprocessing module, a calculation module and an execution module, characterized in that:

[0048] The event camera module is used to build an event pulse stream collection model based on bionic visual perception, collect event pulse stream signals caused by different discharge phenomena in the GIS, and preset real event pulse streams according to historical discharge phenomena;

[0049] The preprocessing module is used to pre-process the event pulse stream signal and calculate the imaging accuracy rate of the bionic visual sensor to obtain a set of discharge phenomena to be learned M;

[0050] A computing module is used to perform offline learning on the set M of discharge phenomena to be learned, construct a reward function and a loss function, iteratively train the event pulse stream in the set M of discharge phenomena to be learned, optimize the bionic visual light intensity change threshold, and output the optimal light intensity change threshold for all discharge phenomena to be learned;

[0051] The execution module is used to detect the light intensity changes inside the GIS in real time based on the optimal light intensity change threshold set and use the event pulse stream acquisition model to perceive different discharge phenomena inside the GIS.

[0052] The beneficial effects of the present invention are that, compared with existing technologies, by considering the differences in light signal intensity of different discharge phenomena, differentiated light intensity change thresholds are set for different discharge phenomenon characteristics. Based on the light intensity change thresholds, an event pulse stream with high imaging accuracy and consistent with actual conditions is generated, thus supporting accurate visual perception under various signal intensities, enhancing the accurate description of GIS discharge phenomena by the event camera, and improving the accuracy of GIS discharge phenomenon perception. A two-stage event-driven bionic visual perception optimization method is proposed, combining offline learning with online data to improve the algorithm's adaptability to field conditions. In the first stage, a main network and a target network are constructed to perform offline iterative learning for each known discharge phenomenon within the GIS, seeking to optimize the accuracy of discharge phenomenon imaging. The offline training results are then used in the second stage for online discharge detection applications. In the second stage, unknown discharge phenomena in the online discharge detection application are closed-loop fed back to the offline learning stage to learn the optimal light intensity change threshold. The optimal light intensity change threshold can be learned using new discharge phenomena collected in the online application, achieving closed-loop feedback optimization between offline learning and online application, further improving the bionic imaging efficiency of GIS equipment and the accuracy of GIS discharge phenomenon imaging recognition. BRIEF DESCRIPTION OF THE DRAWINGS

[0053] Figure 1 It is a bionic visual perception framework diagram that adapts to various signal strengths within GIS;

[0054] Figure 2 This is a schematic diagram of the bionic visual perception optimization method;

[0055] Figure 3 It is a flow chart of the bionic visual optimization method that adapts to various signal strengths within GIS;

[0056] Figure 4 It is a bionic visual imaging device that adapts to various signal intensities within GIS. DETAILED DESCRIPTION

[0057] To make the objectives, technical solutions, and advantages of the present invention more clear, the technical solutions of the present invention will be clearly and completely described below in conjunction with the accompanying drawings in the embodiments of the present invention. The embodiments described in this application are only part of the embodiments of the present invention, not all of them. Based on the spirit of the present invention, all other embodiments obtained by ordinary technicians in this field without making creative efforts are within the scope of protection of the present invention.

[0058] like Figure 3 As shown, embodiment 1 of the present invention proposes a bionic visual optimization method for adapting the internal signal strength of GIS, comprising the following steps:

[0059] Construct an event pulse flow acquisition model based on bionic visual perception to collect event pulse flow signals caused by different discharge phenomena inside the GIS, and preset the real event pulse flow based on historical discharge phenomena;

[0060] like Figure 1 As shown, the event pulse stream signal is preprocessed, and the imaging accuracy of the bionic visual sensor is calculated to obtain a set M of discharge phenomena to be learned;

[0061] Offline learning is performed on the set M of discharge phenomena to be learned, and a reward function and a loss function are constructed. The event pulse stream in the set M of discharge phenomena to be learned is iteratively trained to optimize the threshold of bionic visual light intensity change, and the optimal light intensity change threshold of all discharge phenomena to be learned is output;

[0062] Based on the optimal light intensity change threshold set, the event pulse stream acquisition model is used to detect the light intensity changes inside the GIS in real time and perceive different discharge phenomena inside the GIS.

[0063] Preferably, the event pulse stream acquisition model includes an event camera deployed inside the GIS and including ultraviolet, infrared and visible light visual sensors. The visual sensors respond to changes in light intensity in discharge phenomena inside the GIS in the form of asynchronous sparse events. When the light intensity of a pixel at the current moment changes by more than a threshold compared to the preset standard light intensity of the pixel, the visual sensor outputs an event pulse at the pixel. All event pulses output by each visual sensor constitute the event pulse stream under the spectrum.

[0064] Preferably, the set of different types of discharge phenomena occurring inside the GIS is defined as E = {e1,…,e o ,…,e O}, e o , is the oth discharge phenomenon, O is the total number of different types of discharge phenomena;

[0065] The discharge phenomena collected by ultraviolet vision sensor, infrared vision sensor and visible light vision sensor are oThe generated event pulse stream is recorded as:

[0066]

[0067] Wherein, L = {UV, IR, VIS}, UV represents ultraviolet vision sensor, IR represents infrared vision sensor, and VIS represents visible light vision sensor. is the nth event in the event pulse stream generated by the sensor represented by L, and N is the total number of event pulses generated by the sensor represented by L;

[0068] Each event pulse is represented in the form of a quaternion:

[0069]

[0070] in, for The two-dimensional pixel position of for The trigger time; for The polarity of When , it means that the light intensity at the pixel is enhanced, and when , it means that the light intensity at the pixel is weakened.

[0071] Preferably, the preprocessing refers to slicing the event pulse stream generated by the visual sensor and the real event pulse stream at a reference frame rate f to obtain a sliced ​​image set.

[0072] Preferably, the calculation model for calculating the accuracy of bionic visual imaging specifically includes:

[0073] Calculating the mutual information between a set of acquired event pulse stream slice images obtained by slicing the event pulse stream generated by the visual sensor and the real event pulse stream and the normalized corresponding images in the set of event pulse stream slice images;

[0074]

[0075] Wherein, L = {UV, IR, VIS}, UV represents ultraviolet vision sensor, IR represents infrared vision sensor, and VIS represents visible light vision sensor; are respectively the collection event pulse stream slice image set obtained by the sensor corresponding to L after slicing and the l-th image in the event pulse stream slice image set; For images and images The mutual information of Images and images Entropy;

[0076] The imaging accuracy is obtained by weighted summation of the normalized mutual information of different visual sensors. Calculation formula:

[0077]

[0078] in, The total number of images in the collection of event pulse stream slice images obtained by slicing the images collected by the ultraviolet, infrared and visible light sensors, and The discharge phenomena are collected by the set ultraviolet, infrared and visible light sensors respectively. o The weight ranges from 0 to 1, and

[0079] Preferably, if Figure 2 As shown, the optimization of the bionic visual light intensity change threshold is achieved, specifically:

[0080] The optimization objective P1 formula is:

[0081]

[0082] Where, and are the light intensity change threshold sets of the ultraviolet, infrared and visible light bionic sensors for all discharge phenomena, They are discharge phenomena e o The light intensity change thresholds of the ultraviolet, infrared and visible light bionic sensors must be constrained between the minimum and maximum light intensity change thresholds;

[0083] The optimization objective of maximizing the average imaging accuracy is transformed into maximizing the imaging accuracy of each discharge phenomenon:

[0084]

[0085] Preferably, the reward function constructed is specifically the discharge phenomenon e o Imaging accuracy i is the number of iterations.

[0086] Preferably, the constructed loss function is specifically:

[0087]

[0088] Where,

[0089]

[0090] Where γ is the discount factor; Q(S(i),ρ(i); θ main ) is the output of the main network; Q(S(i+1),ρ(i+1); θtarget ) is the output of the target network; S(i+1) and S(i) are the states at iterations i+1 and i, respectively; ρ(i+1) and ρ(i) are the actions at iterations i+1 and i, respectively; Γ(i) represents the experience replay pool; Represents some samples randomly drawn from the experience replay pool, for The number of samples in θ; main represents the main network, θ target represents the target network;

[0091] Based on this loss function, the main network is updated. Where μ represents the update step size.

[0092] Preferably, the obtained optimal light intensity change threshold set is used to detect the light intensity change inside the GIS in real time using the event pulse stream acquisition model to sense different discharge phenomena inside the GIS:

[0093] The event camera maintains a detection state. When the detected light intensity change exceeds the minimum light intensity change threshold in the optimal light intensity change threshold set, the event pulse stream signal caused by the discharge signal is collected and preprocessed to obtain a set of perception pulse stream slice images.

[0094] Calculate the imaging accuracy of the perceived pulse flow slice image set and the real event pulse flow of all known discharge phenomena, and determine whether it is a known discharge phenomenon based on the imaging accuracy;

[0095] If it is a known discharge phenomenon, the discharge phenomenon that occurs is the discharge phenomenon with the highest imaging accuracy. The optimal light intensity change threshold of the discharge phenomenon is selected to collect the discharge phenomenon that occurs in the current GIS and generate the corresponding event pulse stream. If the maximum waiting time is set, If no light intensity change exceeding the optimal light intensity change threshold for the discharge phenomenon is detected, the pulse flow sensing of this event ends;

[0096] If it is not a known discharge phenomenon, closed-loop feedback update learning is performed, and the light intensity change threshold of the phenomenon with the highest imaging accuracy is used to collect the current discharge phenomenon, and a corresponding event pulse stream is generated, which is placed in the set M of discharge phenomena to be learned, and the optimal threshold is learned again for the set M of discharge phenomena to be learned.

[0097] Preferably, the specific scheme for determining whether it is a known discharge phenomenon is: comparing the imaging accuracy of the perceived pulse flow slice image set with the real event pulse flow of all known discharge phenomena with a set minimum matching threshold Ω min In contrast, when one or more imaging accuracy rates are higher than the set minimum matching threshold Ω minWhen , it is considered that a known discharge phenomenon has occurred inside the GIS; all imaging accuracies are lower than the minimum matching threshold Ω min It is considered that a new discharge phenomenon occurs at this time.

[0098] like Figure 4 As shown, embodiment 2 of the present invention proposes a bionic visual imaging device for adapting the internal signal strength of GIS using the method described in embodiment 1 of the present invention, including an event camera module, a preprocessing module, a calculation module, and an execution module, characterized in that:

[0099] The event camera module is used to build an event pulse flow acquisition model based on bionic visual perception, collect event pulse flow signals caused by different discharge phenomena inside the GIS, and preset the real event pulse flow based on historical discharge phenomena;

[0100] A preprocessing module is used to preprocess the event pulse stream signal and calculate the imaging accuracy of the bionic visual sensor to obtain a set M of discharge phenomena to be learned;

[0101] A computing module is used to perform offline learning on the set M of discharge phenomena to be learned, construct a reward function and a loss function, iteratively train the event pulse stream in the set M of discharge phenomena to be learned, optimize the bionic visual light intensity change threshold, and output the optimal light intensity change threshold for all discharge phenomena to be learned;

[0102] The execution module is used to detect the light intensity changes inside the GIS in real time based on the optimal light intensity change threshold set and use the event pulse stream acquisition model to perceive different discharge phenomena inside the GIS.

[0103] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention rather than to limit it. Although the present invention has been described in detail with reference to the above embodiments, ordinary technicians in the field should understand that the specific implementation methods of the present invention can still be modified or replaced by equivalents. Any modification or equivalent replacement that does not depart from the spirit and scope of the present invention should be covered by the scope of protection of the claims of the present invention.

Claims

1. A bionic visual optimization method for adapting to the internal signal strength of GIS, characterized by: include: An event pulse stream acquisition model based on bionic visual perception is constructed to collect event pulse stream signals caused by different discharge phenomena within the GIS, and to preset real event pulse streams based on historical discharge phenomena. The event pulse stream acquisition model includes an event camera deployed within the GIS that contains ultraviolet, infrared, and visible light vision sensors. The vision sensors respond to changes in light intensity in discharge phenomena within the GIS in the form of asynchronous sparse events. When the change in light intensity at a pixel at a given moment exceeds a threshold compared to the preset standard light intensity for that pixel, the vision sensor outputs an event pulse at that pixel. All event pulses output by each vision sensor constitute the event pulse stream under that spectrum. Preprocess the event pulse flow signals and real event pulse flow caused by different discharge phenomena, calculate the imaging accuracy of the bionic visual sensor, and obtain the set M of discharge phenomena to be learned; Offline learning is performed on the set M of discharge phenomena to be learned, and a reward function and a loss function are constructed. The event pulse stream in the set M of discharge phenomena to be learned is iteratively trained to optimize the threshold of bionic visual light intensity change, and the optimal light intensity change threshold of all discharge phenomena to be learned is output; Based on the optimal light intensity change threshold set, the event pulse stream acquisition model is used to detect the light intensity changes inside the GIS in real time and perceive different discharge phenomena inside the GIS.

2. The bionic visual optimization method for adapting the internal signal strength of GIS according to claim 1 is characterized in that: The set of different types of discharge phenomena occurring inside GIS is defined as , For the o A discharge phenomenon, is the total number of different types of discharge phenomena; The discharge phenomenon collected by ultraviolet vision sensor, infrared vision sensor and visible light vision sensor The generated event pulse stream is recorded as: ； in, , Indicates ultraviolet vision sensor, Indicates infrared vision sensor, Represents a visible light vision sensor, is the nth event in the event pulse stream generated by the sensor represented by L, The total number of event pulses generated for the sensor represented by L; Each event pulse is represented in the form of a quaternion: in, for The two-dimensional pixel position of for The trigger time; for The polarity of When , it means that the light intensity at the pixel is enhanced, and when , it means that the light intensity at the pixel is weakened.

3. The bionic visual optimization method for adapting the internal signal strength of GIS according to claim 1 is characterized in that: The pre-processing refers to the base frame rate The event pulse stream generated by the visual sensor and the real event pulse stream are sliced ​​to obtain a sliced ​​image set.

4. The bionic visual optimization method for adapting to the internal signal strength of GIS according to claim 3 is characterized by: Computational bionic visual imaging accuracy calculation model, specifically including: Calculating the mutual information between a set of acquired event pulse stream slice images obtained by slicing the event pulse stream generated by the visual sensor and the real event pulse stream and the normalized corresponding images in the set of event pulse stream slice images; in, , Indicates ultraviolet vision sensor, Indicates infrared vision sensor, represents a visible light vision sensor; After slicing L The corresponding sensor obtains the collection event pulse stream slice image set and the first one in the event pulse stream slice image set l images; For images and images The mutual information of Images and images Entropy; The imaging accuracy is obtained by weighted summation of the normalized mutual information of different visual sensors. Calculation formula: in, 、 、 The total number of images in the collection of event pulse stream slice images obtained by slicing the images collected by the ultraviolet, infrared and visible light sensors, 、 and The discharge phenomena are collected by the set ultraviolet, infrared and visible light sensors respectively The weight ranges from 0 to 1, and .

5. A bionic visual optimization method for adapting the internal signal strength of a GIS according to any one of claims 1 to 4, characterized in that: Optimize the threshold of bionic visual light intensity change, specifically: Optimization goal The formula is: Where, 、 and are the light intensity change threshold sets of the ultraviolet, infrared and visible light bionic sensors for all discharge phenomena, 、 、 Discharge phenomenon The light intensity change thresholds of the ultraviolet, infrared and visible light bionic sensors must be constrained between the minimum and maximum light intensity change thresholds; The optimization objective of maximizing the average imaging accuracy is transformed into maximizing the imaging accuracy of each discharge phenomenon: 。 6. The bionic visual optimization method for adapting to the internal signal strength of GIS according to claim 5 is characterized by: The reward function constructed is specifically the discharge phenomenon Imaging accuracy , i is the number of iterations.

7. The bionic visual optimization method for adapting the internal signal strength of GIS according to claim 6, characterized in that: The constructed loss function is specifically: Where, Where, is the discount factor; is the output of the main network; is the output of the target network; The number of iterations is 、 The state when The number of iterations are 、 Actions when Represents the experience replay pool; Represents some samples randomly drawn from the experience replay pool, for The number of samples in ; represents the main network, represents the target network; Based on this loss function, the main network is updated. ,in, Indicates the update step size.

8. The bionic visual optimization method for adapting the internal signal strength of GIS according to claim 7, characterized in that: The optimal light intensity change threshold set obtained above is used to detect the light intensity change inside the GIS in real time using the event pulse stream acquisition model to sense different discharge phenomena inside the GIS: The event camera maintains a detection state. When the detected light intensity change exceeds the minimum light intensity change threshold in the optimal light intensity change threshold set, the event pulse stream signal caused by the discharge signal is collected and preprocessed to obtain a set of perception pulse stream slice images. Calculate the imaging accuracy of the perceived pulse flow slice image set and the real event pulse flow of all known discharge phenomena, and determine whether it is a known discharge phenomenon based on the imaging accuracy; If it is a known discharge phenomenon, the discharge phenomenon that occurs is the discharge phenomenon with the highest imaging accuracy. The optimal light intensity change threshold of the discharge phenomenon is selected to collect the discharge phenomenon that occurs in the current GIS and generate the corresponding event pulse stream. If the maximum waiting time is set, If no light intensity change exceeding the optimal light intensity change threshold for the discharge phenomenon is detected, the pulse flow sensing of this event ends; If it is not a known discharge phenomenon, closed-loop feedback update learning is performed, and the light intensity change threshold of the phenomenon with the highest imaging accuracy is used to collect the current discharge phenomenon, and a corresponding event pulse stream is generated, which is placed in the set M of discharge phenomena to be learned, and the optimal threshold is learned again for the set M of discharge phenomena to be learned.

9. The bionic visual optimization method for adapting the internal signal strength of GIS according to claim 8, characterized in that: The specific scheme for determining whether it is a known discharge phenomenon is: comparing the imaging accuracy of the perceived pulse flow slice image set with the real event pulse flow of all known discharge phenomena with the set minimum matching threshold In contrast, when one or more imaging accuracy rates are higher than the set minimum matching threshold When , it is considered that a known discharge phenomenon has occurred inside the GIS; all imaging accuracies are lower than the minimum matching threshold It is considered that a new discharge phenomenon occurs at this time.

10. A bionic visual imaging device adapted to the internal signal strength of a GIS using the method according to any one of claims 1 to 9, comprising an event camera module, a preprocessing module, a calculation module, and an execution module, characterized in that: The event camera module is used to build an event pulse flow acquisition model based on bionic visual perception, collect event pulse flow signals caused by different discharge phenomena inside the GIS, and preset the real event pulse flow based on historical discharge phenomena; A preprocessing module is used to preprocess the event pulse stream signal and calculate the imaging accuracy of the bionic visual sensor to obtain a set M of discharge phenomena to be learned; A computing module is used to perform offline learning on the set M of discharge phenomena to be learned, construct a reward function and a loss function, iteratively train the event pulse stream in the set M of discharge phenomena to be learned, optimize the bionic visual light intensity change threshold, and output the optimal light intensity change threshold for all discharge phenomena to be learned; The execution module is used to detect the light intensity changes inside the GIS in real time based on the optimal light intensity change threshold set and use the event pulse stream acquisition model to perceive different discharge phenomena inside the GIS.

Images