Band-pass filtering method and apparatus, chip and electronic device

Abstract

The application discloses a band-pass filtering method and device, a chip and electronic equipment. In order to solve the problem that the prior art lacks band-pass filtering based on pulse neurons, the application obtains a lower limit frequency and an upper limit frequency through double-channel high-pass filtering, and realizes band-pass filtering. According to the collocation of LIF neurons and IAF neurons, the application obtains the technical effect of low-power-consumption, low-delay and low-computing-load band-pass filtering, and can be applied to efficient capture of target points based on neuromorphic sensors. The application is suitable for the field of neuromorphic computing and computer vision.

Description

TECHNICAL FIELD

[0001] The present application relates to a band-pass filtering method and device, a chip and an electronic device, in particular to a band-pass filtering method and device using brain-like chip convolution and pulse neurons, a chip and an electronic device. BACKGROUND

[0002] High-pass filtering is a common information processing method in the field of signal processing, which only allows signals higher than a certain frequency to pass. Band-pass filtering only allows signals within a certain frequency band to pass.

[0003] The pulse neural network (SNN) is the third generation of neural networks, which transmits information through pulse events by simulating the form of brain neural network pulse excitation. Usually, the pulse neural network is used to complete the reasoning work. When the applicant faces the technical challenge of how to efficiently extract visual information (such as target points), it is independently found that the high-pass / band-pass filtering based on SNN is one of the feasible technical paths. However, for high-frequency pulses, how to realize high-pass / band-pass filtering through SNN is still unknown. SUMMARY

[0004] In order to solve or alleviate part or all of the above technical problems, the present application is realized by the following technical scheme:

[0005] A band-pass filtering method, through at least a first pulse neuron, realizing high-pass filtering of the pulse event output by a neuromorphic sensor pixel to a first channel where the first pulse neuron is located, and the cut-off frequency is the lower limit frequency; through at least a second pulse neuron, realizing high-pass filtering of the pulse event output by the neuromorphic sensor pixel to a second channel where the second pulse neuron is located, and the cut-off frequency is the upper limit frequency; the pulse events fired by the first pulse neuron and the second pulse neuron are projected to a first IAF neuron.

[0006] In some embodiments, the first pulse neuron and the second pulse neuron are both LIF neurons.

[0007] In some embodiments, the first pulse neuron uses a positive synaptic weight when projecting the pulse event to the first IAF neuron; the second pulse neuron uses a negative synaptic weight when projecting the pulse event to the first IAF neuron.

[0008] In some embodiments, by configuring the synaptic weight value, when the frequency of the external stimulus received by the neuromorphic sensor pixel is greater than the upper limit frequency, the first IAF neuron cannot always fire a pulse.

[0009] In some embodiments, the first and second spiking neurons fire at the same frequency; and the first and second spiking neurons use the same absolute value of synaptic weight when projecting spike events to the first IAF neuron.

[0010] In some embodiments, when the frequency of external stimuli received by the neuromorphic sensor pixel is greater than the upper limit frequency, the second spiking neuron always pulls down the membrane voltage of the first IAF neuron through the spike events it fires before the membrane voltage of the first IAF neuron reaches the firing threshold each time the first spiking neuron accumulates, so that the first spiking neuron cannot make the first IAF neuron fire within any time length.

[0011] In some embodiments, when implementing high-pass filtering on the first or / and second channel, the following condition is satisfied: Wxf c +bxf leak = 0, where W is the synaptic weight of the neuromorphic sensor pixel outputting spike events to the first or second spiking neuron on the first or second channel, f c is the cutoff frequency of high-pass filtering on the first or second channel, b is the leakage of the first or second spiking neuron, and f leak is the leakage frequency of the first or second spiking neuron.

[0012] In some embodiments, when implementing high-pass filtering on the first or / and second channel, the following condition is also satisfied: T min ×(Wxf e +bxf leak )= 0, where T min is the minimum dwell time, f e is the frequency of external stimuli, and 0 is the firing threshold of the first or second spiking neuron.

[0013] In some embodiments, the first or second spiking neuron leaks at least once within the minimum dwell time, i.e., T min > 1 / f leak .

[0014] In some embodiments, with a light source that flickers at a cutoff frequency f c , the membrane voltage of the first or second spiking neuron accumulates at least twice within the minimum dwell time: T min xf c > 2.

[0015] In some embodiments, for any one of the first or second channels, the synaptic weight W < 0xf leak / f c.

[0016] A band-pass filtering device, by at least a first spiking neuron, implements high-pass filtering of a neuro-morphic sensor pixel output spike event onto a first channel where the first spiking neuron is located, with a cut-off frequency being a lower limit frequency; by at least a second spiking neuron, implements high-pass filtering of the neuro-morphic sensor pixel output spike event onto a second channel where the second spiking neuron is located, with a cut-off frequency being an upper limit frequency; the spike events fired by the first spiking neuron and the second spiking neuron are projected to a first IAF neuron.

[0017] In some embodiments, the first spiking neuron and the second spiking neuron are both LIF neurons.

[0018] In some embodiments, the first spiking neuron uses positive synaptic weight when projecting spike events to the first IAF neuron; the second spiking neuron uses negative synaptic weight when projecting spike events to the first IAF neuron.

[0019] In some embodiments, by configuring the synaptic weight values, when the frequency of external stimuli received by the neuro-morphic sensor pixel is greater than the upper limit frequency, the first IAF neuron is always unable to fire a spike.

[0020] In some embodiments, when the first spiking neuron and the second spiking neuron fire spikes, the frequency of spike firing is the same; and the absolute value of the synaptic weight used by the first spiking neuron and the second spiking neuron when projecting spike events to the first IAF neuron is the same.

[0021] In some embodiments, when the frequency of external stimuli received by the neuro-morphic sensor pixel is greater than the upper limit frequency, before the membrane voltage of the first IAF neuron reaches the firing threshold each time the first spiking neuron accumulates, the second spiking neuron always pulls down the membrane voltage of the first IAF neuron through the spike events fired, so that the first spiking neuron is unable to make the first IAF neuron fire within any time length.

[0022] In some embodiments, when implementing high-pass filtering on the first channel or / and the second channel, the following condition is met: Wf c + bf leak = 0, where W is the synaptic weight of the neuro-morphic sensor pixel output spike event to the first spiking neuron or the second spiking neuron on the first channel or the second channel, f c is the cut-off frequency of high-pass filtering on the first channel or the second channel, b is the leakage of the first spiking neuron or the second spiking neuron, f leak is the leakage frequency of the first spiking neuron or the second spiking neuron.

[0023] In some embodiments, when implementing the high-pass filtering on the first channel or / and the second channel, T min × (W × f e + b × f leak ) = θ, where T min is the minimum dwell time, f e is the frequency of the external stimulus, and θ is the firing threshold of the first spiking neuron or the second spiking neuron.

[0024] In some embodiments, the first spiking neuron or the second spiking neuron leaks at least once within the minimum dwell time, i.e., T min > 1 / f leak .

[0025] In some embodiments, for a light source that flickers at a cut-off frequency f c , the membrane voltage of the first spiking neuron or the second spiking neuron accumulates at least twice within the minimum dwell time, i.e., T min × f c > 2.

[0026] In some embodiments, for either of the first channel or the second channel, the synaptic weight W ≤ θ × f leak / f c .

[0027] A chip comprising the band-pass filtering apparatus of any one of the preceding embodiments.

[0028] In some embodiments, the chip further comprises a neuromorphic sensor.

[0029] In some embodiments, the neuromorphic sensor is an event camera.

[0030] An electronic device comprising the band-pass filtering apparatus of any one of the preceding embodiments; or, the electronic device comprises the chip of any one of the preceding embodiments.

[0031] In some embodiments, the electronic device comprises a first component, wherein the first component comprises a light emitter configured to emit light at a first frequency; the electronic device further comprises a second component, wherein the second component comprises a neuromorphic sensor configured to capture the light emitter; and the band-pass filtering apparatus of any one of the preceding embodiments is configured to filter the spiking events output by the neuromorphic sensor.

[0032] In some embodiments, the first frequency is between a lower limit frequency and an upper limit frequency of the band-pass filtering apparatus.

[0033] In certain embodiments, the light emitter comprises a damped light emission pattern or / and a sustained flickering light emission pattern.

[0034] In certain embodiments, the electronic device is an air acoustic drum, and the first component is a drumstick.

[0035] The embodiments of the present application have the following beneficial technical effects:

[0036] 1) The target point can be processed with low delay, low data volume, low calculation, low power consumption and low cost.

[0037] 2) Only neuromorphic hardware is used, without other complex and redundant processing components and processes.

[0038] More beneficial effects will be further described in the preferred embodiments.

[0039] The above disclosed technical solutions / features are intended to summarize the technical solutions and features described in the detailed description section, and therefore the scope of the description may not be completely the same. However, these new technical solutions disclosed in this section also belong to part of the numerous technical solutions disclosed in the present application file, and the technical features disclosed in this section, the technical features disclosed in the subsequent detailed description section, and the parts not explicitly described in the drawings are disclosed in a reasonable combination to disclose more technical solutions.

[0040] All the technical features disclosed in any position of the present application are combined to form technical solutions, which are used to support the summary of the technical solutions, the modification of the patent file, and the disclosure of the technical solutions. BRIEF DESCRIPTION OF DRAWINGS

[0041] Figure 1 is a schematic diagram of the relationship between the event camera and the target neuron;

[0042] Figure 2 is a schematic diagram of a double channel;

[0043] Figure 3 is a schematic diagram of filtering under different frequency bands. DETAILED DESCRIPTION

[0044] Since it is impossible to exhaustively describe various alternatives, the key points of the technical solutions in the embodiments of the present application will be described clearly and completely below in combination with the drawings in the embodiments of the present application. For other technical solutions and details not disclosed in detail below, generally belong to technical objects or technical features that can be achieved by conventional means in the art, and due to the limited length, the present application does not introduce them in detail.

[0045] Unless the context clearly requires otherwise, throughout the description, the words "comprise", "comprising", and the like are to be construed in an inclusive sense, as opposed to an exclusive or exhaustive sense; that is to say, in the sense of "including, but not limited to". As used herein, the terms "have", "has", "having", "with" or "comprise" or any variation thereof are used in their on-board meanings applicable to the context. For example, a device "has" a component, which means that a device includes the component or a device "comprises" a component, which means that a device includes the component.

[0046] Various aspects will be presented in terms of systems, other components, methods, and / or other implementations of the present disclosure. These and other aspects shall not be construed or interpreted as being limiting, but rather as being illustrative of the many aspects of the present disclosure. For example, aspects can be presented in one or more claim(s), which are not necessarily all of the aspects of the present disclosure. It will be appreciated that the aspects of the present disclosure can be implemented in a variety of ways, and that the terminology is used for the purpose of description rather than as a limitation on the present disclosure.

[0047] It is explicitly noted that the mere fact that a step, module, feature, etc. is described in various embodiments of the present disclosure does not indicate that such a step, module, feature, etc. is essential to the practice of the present disclosure. The description of various embodiments of the present disclosure is not meant to limit the present disclosure to only those embodiments. The present disclosure is intended to embrace all alternatives, modifications, and variations that fall within the scope of the present disclosure. Based on the description of various embodiments of the present disclosure, a person skilled in the art will be able to affect the present disclosure using alternative elements and / or changes without departing from the scope of the present disclosure. The embodiments of the present disclosure disclosed herein are for illustrative purposes only and are not meant to limit the present disclosure to only the embodiments described herein. One skilled in the art will readily recognize from the disclosure herein, that alternative embodiments of the present disclosure can be constructed without departing from the scope of the present disclosure. The use of any and all examples, or exemplary language (e.g., "may", "can", "will", "should", "about", "approximately", "substantially", "less than", "less than or equal to", "greater than", "greater than or equal to", "between", "within", "peer to peer", or the like) introduced herein, is to be construed as open-ended, and is not intended to be limiting unless otherwise specifically indicated in the claims. Accordingly, the disclosure of variations of alternative embodiments is intended to be included within the scope of the present disclosure. It is also possible to modify or replace some of the constituent elements of the embodiments of the present disclosure in a variety of ways while still retaining the essential characteristics of the present disclosure. The essential characteristics of the present disclosure are particularly pointed out and distinctly claimed in the claims.

[0048] An event camera, also known as a dynamic vision sensor (DVS), is a new kind of vision sensor that sends asynchronous pulses (referred to as pulses or events) to the back-end system by capturing light intensity changes. When the light intensity becomes stronger, an ON event is sent, and when the light intensity becomes weaker, an OFF event is sent. For example, for a light source driven by 50Hz AC power, since the light intensity change does not distinguish the phase, it is actually equivalent to a light intensity change frequency of 100Hz, and in one period, the light intensity becomes stronger and weaker once each. Based on the assumption that one pulse is generated for each of the light intensity becoming stronger and weaker, the event generation amount is 200 events / second. In some scenarios of the present disclosure, for an external light source with a light intensity change frequency of f e , due to the ON / OFF polarity, the event generation frequency (including ON events and OFF events) of the event camera is 2f e .

[0049] As shown in FIG. 1, for a pulse neural network SNN, the pulse events output by the event camera (or other types of neuromorphic sensors, and the present disclosure is only an example of the event camera) can be received. In some embodiments of the present disclosure, the SNN is designed as a high-pass filter, which is used to only pass events with a frequency higher than a set cutoff frequency f Figure 1 .c In addition, the high-pass filter can be only a part or several layers of the SNN or used in any other place where it is applicable, and the present application does not limit this.

[0050] In some embodiments of the present application, the neuron is implemented as a leaky integrate-and-fire neuron. Preferably, the leak is a linear leak of membrane potential with a single leak amount b and a leak (clock) frequency f leak .

[0051] For a neuromorphic sensor, a pixel with coordinates (x, y) receives an input stimulus with a frequency f e , which outputs a spike event and projects to a target neuron (for the target neuron, the output spike event here is its input spike event), which is the first pulse neuron U x,y with membrane voltage V x,y .

[0052] Then the change (increment) of the membrane voltage of the target neuron from t0to t T is:

[0053]

[0054] If at some time t, the membrane voltage V x,y (t) is greater than the firing threshold θ, then the target neuron U x,y fires a spike event, denoted as U x,y (t) = 1, and resets the membrane voltage to the resting potential, such as V x,y (t) = 0; where W is the synaptic weight.

[0055] A stimulus source, such as a light source that flashes at a certain frequency pattern, if it is desired to image in the event camera only when its frequency is greater than the cutoff frequency f c , then the following condition should be satisfied: Wxf c + bxf leak = 0. In other words, if the frequency of the stimulus is not higher than the cutoff frequency f c , no matter how long the time is, there will be no accumulation of membrane voltage, and it is impossible to excite the target neuron to fire a spike.

[0056] If the actual received frequency f e > f c , Wf e + bf leak > 0, and after the positive accumulation of time, the membrane voltage V x,y of the target neuron (the sum of the initial membrane voltage and the increment of the membrane voltage) will inevitably exceed the firing threshold θ at some time, and the event camera will image the stimulus.e The target neuron of the frequency stimulation will fire a pulse, which embodies the "high-pass" feature.

[0057] Further, for a pixel and an input stimulation with frequency f e , there is a corresponding minimum dwell time T min : if the dwell time of the light imaging on a certain pixel of the event camera is less than the minimum dwell time, it means that the stimulation source cannot image on the pixel of the event camera, and thus cannot trigger the pixel to fire a pulse event.

[0058] In other words, within the minimum dwell time T min , under the input stimulation with frequency f e , the increment of the membrane voltage is equal to the firing threshold θ:

[0059]

[0060] That is: T min ×(W×f e +b×f leak )=θ. The implication for practical applications is that the higher the frequency f e of the light emitter as a stimulation source, the shorter the minimum dwell time allowed by the event camera. For example, the frequency of the light emitter designed to mark the target point can be 250Hz, 270Hz or 300Hz, which can avoid the common alternating current frequency on the one hand, and reduce the minimum dwell time length on the other hand, allowing the target point or light emitter to move faster.

[0061] For some common scenarios, such as human activities, gestures, background noise, etc. in the field of view, they are low-frequency signals for a single pixel. If a cut-off frequency f c is designed so that only stimulation above the cut-off frequency can make the target neuron fire a pulse event, then high-pass filtering in the pulse domain is achieved.

[0062] As some practical application constraints, preferably, it can be further limited that at least once leak within the minimum dwell time: T min >1 / f leak . With the light source flickering at the cut-off frequency f c , the membrane voltage accumulates at least twice within the minimum dwell time: T min ×f c >2. And the synaptic weight W≤θ×f leak / f c .

[0063] Preferably, f leak =f cIn other words, when the frequency of the external stimulus is the cut-off frequency, the membrane voltage leaks once for each accumulation, and at this time the sum of the synaptic weight W and the leakage amount b is W+b=0.

[0064] Thus, by setting the high-pass filter parameters (such as synaptic weight, leakage amount, leakage frequency, etc.) that satisfy the above conditions, a high-pass filter based on SNN can be realized, and the cut-off frequency is f c It is worth mentioning that often these parameters are not unique, but there are many combinations.

[0065] For the case of band-pass filtering, in some embodiments, the aforementioned high-pass filtering scheme can be used. For example, a double-channel filtering scheme is designed, and by configuring different high-pass filtering parameters, two high-pass filtering channels with cut-off frequencies of lower limit frequency f min and upper limit frequency f max are realized respectively.

[0066] As shown in Figure 2 , based on the filtering of target neuron 1 (also called first pulse neuron) and target neuron 2 (also called second pulse neuron) double channels, it is connected to the IAF neuron of the latter layer to realize the gating purpose. The target neuron 1 and the target neuron 2 are both LIF neurons, and without loss of generality, the IAF neuron in the figure is called the first IAF neuron.

[0067] For example, for channel 1 (also called first channel), the synaptic weight W=θ / 2 of target neuron 1 to the IAF neuron is set, and for channel 2, the synaptic weight W=-θ / 2 of target neuron 2 to the IAF neuron is set. For some embodiments, other parameters not higher than the ignition threshold θ can also be designed, such as ±θ / 3, θ / 3, -θ / 4, etc. The absolute values of the synaptic weights of channel 1 and channel 2 can be equal or not equal, which can be set according to the actual situation.

[0068] Based on these settings, a band-pass filter with a frequency band of (f min , f max ) is designed, and reference is made to Figure 3 , which shows the output of the LIF neuron and the accumulation relationship diagram of the membrane voltage V mem of the IAF neuron in three different cases (out-of-band low frequency, in-band, out-of-band high frequency) as an example.

[0069] When the frequency f e >f maxAt that time, according to the aforementioned high-pass filter parameter design, both channel 1 and channel 2 output pulse events with the same frequency. Then, the membrane voltage of the IAF neuron (without membrane voltage leakage) will be repeatedly pulled up and down by pulses of the same frequency. Due to the aforementioned synaptic weight design that does not exceed the ignition threshold θ, the IAF neuron will not be activated, thereby achieving out-of-band high-frequency blocking.

[0070] When the frequency f e <f min At this time, according to the aforementioned high-pass filter parameter design, neither channel 1 nor channel 2 outputs pulse events. Therefore, the membrane voltage of the IAF neuron (without membrane voltage leakage) will not increase, and thus no pulse will be generated, thereby achieving out-of-band low-frequency blocking.

[0071] And if the frequency of the stimulus satisfies f min <f e <f max Based on the aforementioned high-pass filter parameter design, only channel 1 outputs pulse events. Therefore, with the aforementioned ±θ / 2 parameter setting, only two pulses are needed for the IAF neuron membrane voltage to exceed the ignition threshold, thereby exciting the IAF neuron to output pulse events to the subsequent stage, thus achieving in-band information transmission. However, with a ±θ / 3 parameter setting, three pulse inputs are required to trigger pulse output. Compared to the pulse frequency output by the LIF neuron, different parameters achieve different frequency division effects.

[0072] In another embodiment of the invention, channel 2 (also called the second channel) is designed to be an empty channel, that is: in f e >f max In the corresponding case, before channel 1 can trigger the output pulse of the IAF neuron, the accumulated membrane voltage V of the IAF neuron is cleared. mem This is used to block the input corresponding to low-frequency out-of-band conditions. For example, for channel 1, the synaptic weight W from target neuron 1 to the IAF neuron is θ / 10, while when the IAF membrane voltage V... mem Before accumulating to the ignition threshold θ, for example, V mem =0.8θ, and the IAF membrane voltage V can be obtained through a single pulse from channel 2. mem The voltage is cleared to a lower membrane voltage, such as 0. In other words, in this type of embodiment, channel 2, with a lower frequency (and higher absolute value) negative synaptic weight compared to channel 1, suppresses the firing of IAF neurons. At out-of-band high frequencies, the combination of lower frequency firing and a higher absolute value of negative synaptic weight prevents channel 2 from accumulating to a membrane voltage at or above the firing threshold. Furthermore, this parameter design does not affect out-of-band low frequencies or in-band conditions, because channel 2 cannot fire pulses to IAF neurons in either of these cases.

[0073] Optionally, the SNN is designed as a band-pass filter; or, the band-pass filter of the present application is only a part or several layers of the SNN, or is used in any other place where it can be used, and the present application does not limit this.

[0074] The aforementioned neuromorphic sensor and SNN can be designed in the same chip to form an edge AI chip with sensing and computing integrated, and can be applied to various electronic devices (such as interactive devices) to realize low-latency, low-data, low-computing, low-power, and low-cost intelligent visual information processing of the target point.

[0075] Through the aforementioned high-pass / band-pass design scheme, the neuromorphic sensor can only output (sensing) a specific frequency of light emitter stimulation, which can efficiently capture dynamic information. For example, on various interactive devices such as the top end of an air shelf drum, the top end of a smart pointer, and the top end of a smart sand table, at least one light emitter that emits light in a specific frequency pattern is designed. Through the design of the aforementioned neuromorphic sensor and SNN, static vision and dynamic vision information other than the light emitter can be shielded, and finally the target point can be captured in a very sparse data form. This information processing scheme can be widely applied to AR / VR somatosensory interactive devices.

[0076] For the band-pass filtering scheme, different light emitters and their target points can also be directly identified according to the different frequency bands in which the light emitters are located. The light emitter can be designed in a burst light emission mode (detecting that the light emitter / target point is in a burst motion state, especially in a specific direction) or / and a continuous flicker light emission mode. The former can be a more minimalist information transmission and processing mode, and the combination of the two can realize a more rich interactive mode.

[0077] Although the present application has been described with reference to specific features and embodiments thereof, various modifications, combinations, replacements can be made thereto without departing from the scope of the present application. The scope of protection of the present application is not limited to the specific embodiments of the processes, machines, manufacture, material composition, devices, methods and steps described in the specification, and these methods and modules can also be implemented in one or more products, methods associated with, dependent on, coordinated with, and subsequent to each other.

[0078] Therefore, the specification and drawings should simply be regarded as an introduction to part of the embodiments of the technical solutions defined by the appended claims, and the appended claims should be interpreted according to the principle of maximum reasonable interpretation, and should be intended to cover all modifications, changes, combinations or equivalents within the scope of the present application as much as possible, while avoiding unreasonable interpretation.

[0079] For achieving better technical effects or for the needs of certain applications, the person skilled in the art can make further improvements on the basis of the present application. However, even if the part of improvements / designs is creative or / and progressive, as long as it depends on the technical concept of the present application and covers the technical features defined in the claims, the technical solution should also fall within the protection scope of the present application.

[0080] There can be alternative technical features mentioned in the claims, or the order of certain technical processes or the order of material organization can be reorganized. The person skilled in the art can easily think of these alternative means or change the order of technical processes or the order of material organization after knowing the present application, and then adopt basically the same means to solve basically the same technical problem and achieve basically the same technical effect. Therefore, even if the claims explicitly limit the above-mentioned means or / and order, however, these modifications, changes, and replacements should all fall within the protection scope of the claims according to the doctrine of equivalents.

[0081] In combination with the method steps or modules described in the embodiments disclosed herein, the implementation can be realized in hardware, software, or a combination of both. In order to clearly illustrate the interchangeability of hardware and software, the steps and components of the embodiments have been described in the above description in general terms. Whether the functions are executed in hardware or software depends on the specific application or design constraints of the technical solution. The person skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of the present application.

Claims

1. A band-pass filtering method, characterized in that: a first pulse neuron is used to implement high-pass filtering of pulse events output by a neuromorphic sensor pixel to a first channel where the first pulse neuron is located, with a lower cutoff frequency; a second pulse neuron is used to implement high-pass filtering of pulse events output by the neuromorphic sensor pixel to a second channel where the second pulse neuron is located, with an upper cutoff frequency; pulse events fired by the first pulse neuron and the second pulse neuron are projected to a first IAF neuron; a positive synaptic weight is used when the first pulse neuron projects pulse events to the first IAF neuron; a negative synaptic weight is used when the second pulse neuron projects pulse events to the first IAF neuron. 2.The band-pass filtering method of claim 1, characterized in that: the first pulse neuron is an LIF neuron. 3.The band-pass filtering method of claim 2, characterized in that: the second pulse neuron is an LIF neuron. 4.The band-pass filtering method of claim 3, characterized in that: synaptic weight values are configured such that the first IAF neuron cannot fire a pulse at all when the frequency of external stimuli received by the neuromorphic sensor pixel is greater than the upper cutoff frequency. 5.The band-pass filtering method of claim 3 or 4, characterized in that: the first pulse neuron and the second pulse neuron fire pulses at the same frequency; and the absolute values of synaptic weights used by the first pulse neuron and the second pulse neuron to project pulse events to the first IAF neuron are the same. 6.The band-pass filtering method of claim 4, characterized in that: when the frequency of external stimuli received by the neuromorphic sensor pixel is greater than the upper cutoff frequency, the second pulse neuron pulls down the membrane voltage of the first IAF neuron through a pulse event fired by the second pulse neuron before the membrane voltage of the first IAF neuron reaches a firing threshold each time the first pulse neuron accumulates, so that the first pulse neuron cannot make the first IAF neuron fire at all within any time length. 7.The band-pass filtering method of any one of claims 1-4 and 6, characterized in that: In implementing the high-pass filtering on the first channel or / and the second channel, Wxf c + bxf leak = 0, where W is the synaptic weight of the neuromorphic sensor pixel output pulse event on the first channel or the second channel to the first pulse neuron or the second pulse neuron, f c is the cut-off frequency of the high-pass filtering on the first channel or the second channel, b is the leakage amount of the first pulse neuron or the second pulse neuron, f leak is the leakage frequency of the first pulse neuron or the second pulse neuron. 8.The band-pass filtering method of claim 7, characterized in that: In implementing the high-pass filtering on the first channel or / and the second channel, T min × (W × f e + b × f leak ) = θ, wherein T min is the minimum residence time, f e is the external stimulus frequency, and θ is the firing threshold of the first spiking neuron or the second spiking neuron. 9.The band-pass filtering method of claim 8, characterized in that: In a minimum dwell time, the first spiking neuron or the second spiking neuron leaks at least once, i.e. T min >1 / f leak . 10.The band-pass filtering method of claim 9, characterized in that: at a cutoff frequency f c The flickering light source, within the minimum dwell time, the membrane voltage of the first or second pulse neuron accumulates at least 2 times: T min x f c > 2. 11.The band-pass filtering method of claim 10, characterized in that: for either of the first channel or the second channel, the synaptic weight W < θ x f leak / f c . 12.A band-pass filtering device, characterized in that: a first pulse neuron is used to implement high-pass filtering of pulse events output by a neuromorphic sensor pixel to a first channel where the first pulse neuron is located, with a lower cutoff frequency; a second pulse neuron is used to implement high-pass filtering of pulse events output by the neuromorphic sensor pixel to a second channel where the second pulse neuron is located, with an upper cutoff frequency; the pulse events emitted by the first spiking neuron and the second spiking neuron are projected to a first IAF neuron; a positive synaptic weight is used when the first spiking neuron projects a pulse event to the first IAF neuron; a negative synaptic weight is used when the second spiking neuron projects a pulse event to the first IAF neuron.

13. The band-pass filtering apparatus of claim 12, wherein: the first spiking neuron is a LIF neuron.

14. The band-pass filtering apparatus of claim 13, wherein: the second spiking neuron is a LIF neuron.

15. The band-pass filtering apparatus of claim 14, wherein: synaptic weight values are configured such that the first IAF neuron is always unable to fire a pulse when the frequency at which the neuromorphic sensor pixel receives external stimuli is greater than the upper limit frequency.

16. The band-pass filtering apparatus of claim 14 or 15, wherein: the frequency at which the first spiking neuron and the second spiking neuron fire a pulse is the same; and the absolute value of the synaptic weight used by the first spiking neuron and the second spiking neuron to project a pulse event to the first IAF neuron is the same.

17. The band-pass filtering apparatus of claim 15, wherein: when the frequency at which the neuromorphic sensor pixel receives external stimuli is greater than the upper limit frequency, the second spiking neuron always pulls down the membrane voltage of the first IAF neuron via a pulse event emitted by the second spiking neuron before the membrane voltage of the first IAF neuron reaches a firing threshold each time the first spiking neuron accumulates, such that the first spiking neuron is unable to make the first IAF neuron fire within any time duration.

18. The band-pass filtering apparatus of any one of claims 12-15, 17, wherein: In implementing the high-pass filtering on the first channel or / and the second channel, Wf c + bf leak = 0, where W is the synaptic weight of the neuromorphic sensor pixel output pulse event on the first channel or the second channel to the first pulse neuron or the second pulse neuron, f c is the cut-off frequency of the high-pass filtering on the first channel or the second channel, b is the leakage amount of the first pulse neuron or the second pulse neuron, f leak is the leakage frequency of the first pulse neuron or the second pulse neuron.

19. The band-pass filtering apparatus of claim 18, wherein: In implementing the high-pass filtering on the first channel or / and the second channel, T min × (W × f e + b × f leak ) = θ, where T min is the minimum dwell time, f e is the external stimulus frequency, and θ is the firing threshold of the first spiking neuron or the second spiking neuron.

20. The band-pass filtering apparatus of claim 19, wherein: In a minimum dwell time, the first spiking neuron or the second spiking neuron leaks at least once, i.e. T min >1 / f leak .

21. The band-pass filtering apparatus of claim 20, wherein: at a cutoff frequency f c The flickering light source, within a minimum dwell time, the membrane voltage of the first or second pulse neuron accumulates at least 2 times: T min × f c > 2.

22. The band-pass filtering apparatus of claim 21, wherein: for either of the first channel or the second channel, the synaptic weight W < θ x f leak / f c .

23. A chip, comprising: the chip comprises the band-pass filtering apparatus of any one of claims 12-22; and 24. The chip of claim 23, wherein : the chip further comprises a neuromorphic sensor.

25. The chip of claim 24, wherein: the neuromorphic sensor is an event camera.

26. An electronic device, comprising: the electronic device comprises the band-pass filtering apparatus of any one of claims 12-22; or the electronic device comprises the chip of any one of claims 23-25.

27. The electronic device of claim 26, wherein: the electronic device comprises a first component, wherein the first component comprises a light emitter configured to blink light at a first frequency; and the electronic device further comprises a second component, wherein the second component comprises a neuromorphic sensor configured to capture the light emitter. The band pass filtering device of any one of claims 12 to 22, filtering pulse events output by the neuromorphic sensor pixel.

28. The electronic device of claim 27, wherein: the first frequency is between a lower and an upper frequency of the band pass filtering device.

29. The electronic device of any one of claims 27-28, wherein: the light emitter comprises a cut-off light emission pattern or / and a sustained flickering light emission pattern.

30. The electronic device of any one of claims 27-28, wherein: the electronic device is an air acoustic drum, and the first component is a drumstick.

Images