A mine visible light communication system and method based on pulse network demodulation mechanism

Abstract

The application discloses a mine visible light communication system and method based on a pulse network demodulation mechanism, comprising a transmitting system and a receiving system, wherein the transmitting system and the receiving system are respectively arranged in a mine, and modulation and demodulation equipment is arranged between the transmitting system and the receiving system; a visible light communication carrier modulation technology is applied; when the transmitting system transmits a light signal, the light signal is modulated by using an OFDM method, and the modulated light signal is transmitted; after the receiving system receives the light signal, the light signal is subjected to preliminary processing of filtering and denoising, a transform block extracts features of the light signal, and the extracted light pulse sequence is input into a trained pulse neuron potential dynamics model for processing, and then is decoded and recovered into readable data; the application can adapt to the structural characteristics of the mine, effectively improve the quality of mine light communication, and guarantee the communication requirements of safe production in the mine.

Description

TECHNICAL FIELD

[0001] The present application relates to a mine visible light communication system and method, in particular to a mine visible light communication system and method based on a pulse network demodulation mechanism, and belongs to the mine communication field. BACKGROUND

[0002] At present, coal is still the most important part of human fuel resource source. Most of the coal mines in the world are underground coal mines. The safety problem of coal mining has been an important problem in mine production activities, and how to strengthen safety production, improve early warning and post-accident processing is the top priority of many underground coal mines. In the mining process, with the continuous advancement of the mining degree, the depth of the mine is increasing, and the danger of mining is also increasing. Due to the special environment of underground coal mine, most of them are in the form of roadway, and many coal mines are associated with other metal mines. The communication transmission through the traditional radio form is easy to be disturbed, and cannot achieve the required communication effect. In addition, the mine belongs to the explosion-proof environment, and the wireless radio transmits in the form of electromagnetic wave, which is easy to gather energy at the antenna position and cause danger, and the construction cost is high. In addition, since the mine is located underground, lighting is needed at each position of the mine for 24 hours. If the light communication transmission mode using LED lamp on the ground as lighting facilities is introduced into the mine, the mine light communication system can be realized without adding too many additional equipment. However, due to the structural characteristics of the mine, if the ground light communication mode is directly used, it will cause much noise and low precision in the communication process, and finally cannot meet the requirements of mine communication. Therefore, how to provide a visible light communication system and method for mine, which can adapt to the structural characteristics of the mine, so as to realize light communication transmission in the mine, and meet the requirements of mine communication, is a technical problem to be solved. SUMMARY

[0003] In view of the problems existing in the prior art, the present application provides a mine visible light communication system and method based on a pulse network demodulation mechanism, which can adapt to the structural characteristics of the mine, so as to realize light communication transmission in the mine, and meet the requirements of mine communication.

[0004] In order to achieve the above purpose, the technical scheme adopted by the present application is: a mine visible light communication system based on a pulse network demodulation mechanism, comprising a transmitting system and a receiving system;

[0005] The transmitting system comprises a plurality of explosion-proof LED lamps and a controller, the explosion-proof LED lamps are respectively arranged at various positions under the mine, the controller is connected with the explosion-proof LED lamps and is used for controlling the explosion-proof LED lamps to illuminate and communicate; the controller is internally provided with an LED lamp identity identification module and an LED lamp modulation communication module, the LED lamp identity identification module is used for identifying and numbering each explosion-proof LED lamp and assigning an independent transmitting frequency band to each explosion-proof LED lamp, and the LED lamp modulation communication module is used for modulating a signal required to be transmitted by each explosion-proof LED lamp and then transmitting the signal to the explosion-proof LED lamp for visible light signal transmission;

[0006] The receiving system comprises a plurality of unmanned vehicles under the mine, the unmanned vehicles are distributed at various positions under the mine, each unmanned vehicle is provided with a light signal demodulation device and a photodiode array, the light signal demodulation device is connected with the photodiode array, and the photodiode array is used for receiving the visible light signal transmitted by the explosion-proof LED lamp and transmitting the visible light signal to the light signal demodulation device; the light signal demodulation device is internally provided with a light signal denoising module, a feature extraction module and an automatic decoding module.

[0007] The light signal denoising module is used for performing preliminary filtering processing on the received light signal; the feature extraction module is used for performing signal segment extraction on the denoised light signal; and the automatic decoding module is used for decoding the extracted signal segment to recover the signal segment into readable data.

[0008] Further, the explosion-proof LED lamps are respectively arranged in a coal mine transportation roadway, a shaft bottom yard, a mechanical and electrical chamber, an air return roadway and a coal preparation plant under the mine.

[0009] Further, each unmanned vehicle under the mine is provided with two photodiode arrays, the two photodiode arrays are respectively arranged at the front end and the rear end of the unmanned vehicle and are used for receiving the visible light signals from the front and rear of the vehicle.

[0010] Further, the photodiode array is composed of 2*5 photodiodes.

[0011] Further, the light signal demodulation device is a microprocessor.

[0012] The working method of the mine visible light communication system based on the pulse network demodulation mechanism comprises the following specific steps:

[0013] Step one, filter algorithm is used as a signal denoising module in the optical signal demodulation device to preliminarily denoise the received optical signal; the network structure of the pulse neural network in series with the transform block is used as a feature extraction module and an automatic decoding module to extract features and automatically decode the optical signal and restore it into readable data; the transmitting system is installed at various positions in the mine, and multiple unmanned vehicles are distributed at various positions in the mine to complete the layout of the mine visible light communication system;

[0014] Step two, before use, the pulse neuron potential dynamics model is established by the pulse neural network, the transform block is used to extract features of the optical signal to obtain multiple optical pulse sequences, then the extracted optical pulse sequences are input into the pulse neuron potential dynamics model and the model is trained by using the pulse neural network synaptic weight learning method, and the required pulse neuron potential dynamics model is obtained after completion;

[0015] Step three, when starting visible light communication, first, the LED lamp identity module is used to identify and number each explosion-proof LED lamp, and each explosion-proof LED lamp is assigned an independent transmission frequency band; then, the LED lamp modulation communication module uses the OFDM modulation method to modulate the optical signal required to be transmitted by each explosion-proof LED lamp according to each number and the corresponding explosion-proof LED lamp of the transmission frequency band, and completes the modulation, and then performs visible light transmission through each explosion-proof LED lamp;

[0016] Step four, the photodiode array of each underground unmanned vehicle can receive the optical signal transmitted by the explosion-proof LED lamp nearby, and the number of the explosion-proof LED lamp sending the optical signal can be determined by the frequency band of the received optical signal, the photodiode array transmits the received optical signal to the microprocessor, the microprocessor preliminarily denoises the received optical signal by using the filter algorithm, then extracts the processed optical signal by the transform block to obtain multiple optical pulse sequences, finally, the multiple optical pulse sequences are input into the pulse neuron potential dynamics model trained in step two for processing, and then decoded and restored into readable data, thereby completing the mine visible light communication process.

[0017] Further, the training process of the pulse neuron potential dynamics model is as follows:

[0018] The pulse neuron potential dynamics model is a leaky integrate-and-fire model, i.e. a LIF model, which is similar to the charging, leaking and firing process of a biological neuron. The LIF is a simplified mathematical model based on the dynamic characteristics of a biological neuron. The pulse neural network is connected by pulse neurons. When no light pulse is received, the internal voltage v of each pulse neuron will be exponentially stabilized to an equilibrium voltage a with time. This process is described by the LIF model as follows:

[0019]

[0020] Solving the differential equation, we get:

[0021]

[0022] where c is an arbitrary constant, and τ controls the exponential decline rate. The smaller τ is, the faster v(t) exponentially changes to a. According to the above equation, v = a - c at t = 0, and v = a at t = ∞. The equation controls the exponential stabilization of the voltage v to the equilibrium voltage a with time. Since the above equation is a continuous voltage v(t) change equation, however, a computer can only simulate a discrete process. When the discrete time interval is dt, the discrete form of the differential equation is:

[0023] v(t + dt) = β(v(t) - a) + a, which

[0024] In addition, when a pulse neuron receives a light pulse at a certain time, the pulse will be accumulated in the voltage, and the current voltage will be added to a certain value related to the synaptic weight of the input pulse. The voltage update process is as follows:

[0025] v = v + w

[0026] A firing threshold v t is set in the neuron. When the pulse neuron voltage v > v t , the pulse neuron will fire a pulse, and then the pulse neuron voltage will be immediately set to the resting potential:

[0027] v = v

[0028] The synaptic weight learning method (STDP) of the pulse neural network is a time-sequential asymmetric Hebb learning rule, which is affected by the close time correlation between the pre-synaptic and post-synaptic neuron peaks. The specific formula is as follows:

[0029]

[0030] where, is the time of the post-synaptic spike firing, is the time of the pre-synaptic spike firing; W(x) is the STDP function

[0031] W(x) = A + exp(-x / τ + ) for x>0

[0032]

[0033] The spiking neural network finds a suitable synaptic weight matrix of the optical pulse neural network for a plurality of input optical pulse sequences and a plurality of target optical pulse sequences, so that the output optical pulse sequence of the spiking neuron is as close as possible to the corresponding target optical pulse sequence, that is, the error evaluation function of the two is minimized, at this time the training is completed, and the required spiking neuron potential dynamics model is obtained.

[0034] Further, the required spiking neuron potential dynamics model is verified:

[0035] The BCELoss loss function is used to determine the degree of inconsistency between the predicted value f(x) of the model and the true value Y, the smaller the loss function, the better the robustness of the model, the loss function is the core part of the empirical risk function, and is also an important part of the structural risk function. The structural risk function of the model includes the empirical risk term and the regularization term.

[0036] BCELoss is a loss function for binary classification, and the formula of BCELoss is:

[0037] LOSS = -(ylog(p(x) + (1-y)log(1-p(x))

[0038] Where p(x) is the model output, and y is the true label.

[0039] BCELoss function derivation process:

[0040]

[0041]

[0042]

[0043]

[0044] Therefore:

[0045] Since the demodulation of the optical signal belongs to a multi-label classification problem, there are multiple categories in multi-label classification, therefore the output of the BCELoss function is not a value, but a vector, and the output data cannot be further normalized to a probability value of [0, 1] by using Softmax, and the probabilities of various categories add up to 1. Because the categories are not mutually exclusive and can appear simultaneously, the sigmoid activation function is used to convert each element of the output vector into a probability value, and the degree of inconsistency between the predicted value f(x) of the model and the true value Y is determined according to the probability value, if the threshold is not exceeded, it is determined that the model meets the requirements, if the threshold is exceeded, the pulse neuron potential dynamics model is retrained until the requirements are met.

[0046] Compared with the prior art, the present application comprises a transmitting system and a receiving system, which are respectively arranged in a mine, and there are modulation and demodulation devices between the transmitting system and the receiving system, and the visible light communication carrier modulation technology is applied, when the transmitting system transmits an optical signal, the OFDM method is used to modulate the optical signal to be transmitted, and the modulated optical signal is transmitted, after the receiving system receives the optical signal and performs preliminary processing such as filtering and denoising, the transform block extracts features from the optical signal, and the extracted optical pulse sequence is input into the trained pulse neuron potential dynamics model for processing, and then decoded and restored into readable data; in the model training process, the pulse neural network synaptic weight learning method, the BCELoss loss function and the sigmoid activation function are combined, and finally the model formed by training meets the requirements of decoding, so that the system and method of the present application can adapt to the structural characteristics of the mine, effectively improve the quality of mine optical communication, ensure the communication requirements of safe production in the mine, and realize the method without the need for additional devices, thereby promoting the popularization of mine optical communication. BRIEF DESCRIPTION OF DRAWINGS

[0047] Figure 1 is a layout schematic diagram of the visible light communication system in the present application;

[0048] Figure 2 is a flow principle diagram of the visible light communication method in the present application. DETAILED DESCRIPTION

[0049] The present application will be further described below.

[0050] As Figure 1 shown, a mine visible light communication system based on a pulse network demodulation mechanism comprises a transmitting system and a receiving system;

[0051] The transmitting system comprises a plurality of explosion-proof LED lamps and a controller, and the explosion-proof LED lamps are respectively installed in the coal mine transportation roadway, the shaft bottom yard, the electromechanical chamber, the air return roadway and the coal preparation plant under the mine. The controller is connected with each explosion-proof LED lamp and is used for controlling the explosion-proof LED lamp to perform lighting and communication. An LED lamp identity identification module and an LED lamp modulation communication module are arranged in the controller. The LED lamp identity identification module is used for identifying and numbering each explosion-proof LED lamp and assigning an independent transmitting frequency band to each explosion-proof LED lamp. The LED lamp modulation communication module is used for modulating the signal required to be transmitted by each explosion-proof LED lamp and then transmitting the modulated signal to the explosion-proof LED lamp for visible light signal transmission.

[0052] The receiving system comprises a plurality of underground unmanned vehicles, and each underground unmanned vehicle is distributed at each position under the mine. A light signal demodulation device and a photodiode array are arranged on each underground unmanned vehicle. The light signal demodulation device is a microprocessor. Two photodiode arrays are arranged on each underground unmanned vehicle and are respectively arranged at the front end and the rear end of the underground unmanned vehicle and are used for receiving the visible light signals from the front and rear of the vehicle. The photodiode array is composed of 2*5 photodiodes. The light signal demodulation device is connected with the photodiode array. The photodiode array is used for receiving the visible light signals transmitted by the explosion-proof LED lamp and transmitting the visible light signals to the light signal demodulation device. A light signal denoising module, a feature extraction module and an automatic decoding module are arranged in the light signal demodulation device.

[0053] The light signal denoising module is used for performing preliminary filtering processing on the received light signal. The feature extraction module is used for performing signal segment extraction on the denoised light signal. The automatic decoding module is used for decoding the extracted signal segment to recover the readable data.

[0054] As shown in Figure 2 The working method of the mine visible light communication system based on the pulse network demodulation mechanism comprises the following steps:

[0055] Step one, a filtering algorithm is used as a signal denoising module in the light signal demodulation device to perform preliminary denoising processing on the received light signal. A network structure of a pulse neural network in series with a transform block is used as a feature extraction module and an automatic decoding module to perform feature extraction and automatic decoding processing on the light signal and then recover the readable data. The transmitting system is installed at each position under the mine, and a plurality of underground unmanned vehicles are distributed at each position under the mine to complete the mine visible light communication system layout work.

[0056] Step 2: Before use, a spiking neuron potential dynamics model is established using a spiking neural network. A transform block is used to extract features from the light signal, obtaining multiple light pulse sequences. These extracted light pulse sequences are then input into the spiking neuron potential dynamics model, and the model is trained using a spiking neural network synaptic weight learning method. After completion, the required spiking neuron potential dynamics model is obtained. The training process is as follows:

[0057] The spiking neuron potential dynamics model is the leakage integral-discharge model, or LIF model. This model's operation is similar to the charging, leakage, and discharging process of biological neurons. LIF is a simplified mathematical model based on the dynamic characteristics of biological neurons. A spiking neural network is composed of connected spiking neurons. The input to a spiking neuron is a pulse, and its output is also a pulse. A spiking neuron contains an electromotive force (EMF). When no input is received, the EMF decays exponentially over time to a stable EMF (equilibrium voltage). When an input pulse is received at a certain moment, the EMF increases by a certain value. When the rate of increase of the EMF exceeds the rate of decay (e.g., with frequent pulse inputs), the EMF inside the neuron becomes increasingly larger until it reaches a certain firing threshold, at which point the spiking neuron fires a pulse. After this, the spiking neuron's EMF quickly returns to its resting EMF.

[0058] Let the internal voltage of each spiking neuron be v. When no light pulse input is received, the voltage v will stabilize exponentially over time to an equilibrium voltage. This process can be described by the LIF model as follows:

[0059]

[0060] Solving this differential equation yields:

[0061]

[0062] Where c is an arbitrary constant, τ controls the rate of exponential descent; the smaller τ is, the faster v(t) exponentially changes to a. From the above equation, we derive that at the initial time t = 0, v = ac, where an appropriate value for c ensures that ac equals the initial voltage of the spiking neuron. When t = ∞, v = a. This equation controls the exponential stabilization of voltage v to equilibrium voltage a over time. Since the above equation is a continuous voltage v(t) equation, however, computers can only simulate discrete processes. Taking the discrete time interval dt, the discrete form of the differential equation is:

[0063] v(t+dt)=β(v(t)-a)+a, its

[0064] In addition, when a certain moment the pulse neuron receives a light pulse input, then to accumulate the pulse to the voltage, the current voltage plus a certain value, the value is related to the input pulse synaptic weight, voltage update process is:

[0065] v = v + w

[0066] The neuron is provided with a firing threshold v t , when the pulse neuron voltage v > v t , the pulse neuron will fire a pulse, and then the pulse neuron voltage will be immediately set to the resting potential:

[0067] v = v rest

[0068] The synaptic weight learning method of the pulse neural network (STDP) is a time sequence asymmetric form of Hebb learning rule, which is affected by the close time correlation between the peak values of the presynaptic and postsynaptic neurons, and the specific formula is:

[0069]

[0070] Where, is the time of postsynaptic pulse firing, is the time of presynaptic pulse firing; W(x) is the STDP function

[0071] W(x) = A + exp(-x / τ + ) for x > 0

[0072]

[0073] The pulse neural network finds the appropriate synaptic weight matrix of the light pulse neural network for a given multiple input light pulse sequence and multiple target light pulse sequence, so that the output light pulse sequence of the pulse neuron is as close as possible to the corresponding target light pulse sequence, that is, the error evaluation function of the two is minimized, at which time the training is completed, and the required pulse neuron potential dynamics model is obtained;

[0074] Then the required pulse neuron potential dynamics model is obtained to verify:

[0075] The BCELoss (binary cross-entropy loss) loss function is used to determine the degree of inconsistency between the predicted value f(x) of the model and the true value Y, the smaller the loss function, the better the robustness of the model, and the loss function is the core part of the empirical risk function and an important part of the structural risk function. The structural risk function of the model includes the empirical risk term and the regularization term.

[0076] BCELoss is a binary loss function, and the formula of BCELoss is:

[0077] LOSS = -(ylog(p(x) + (1 - y)log(1 - p(x))

[0078] where p(x) is the model output, y is the real label;

[0079] BCELoss function derivation process:

[0080]

[0081]

[0082]

[0083]

[0084] So:

[0085] Since the demodulation of the optical signal belongs to the multi-label classification problem, there are multiple categories in multi-label classification, therefore the output of the BCELoss function is not a value, but a vector, and the output data cannot be further normalized to a probability value of [0, 1] by using Softmax, and the probabilities of each category add up to 1. Because each category is not mutually exclusive, it allows simultaneous occurrence, so the sigmoid activation function is used to convert each element of the output vector into a probability value respectively, and the Sigmoid function is:

[0086]

[0087] When x→∞, S(x)→1; when x→-∞, S(x)→0.

[0088] According to the probability value, the inconsistency degree between the predicted value f(x) of the model and the real value Y is determined, if it does not exceed the set threshold, it is determined that the model meets the requirements, if it exceeds the set threshold, the pulse neuron potential dynamics model is retrained until the requirements are met.

[0089] Step three, when starting visible light communication, first identify each explosion-proof LED lamp through the LED lamp identity module, and assign each explosion-proof LED lamp an independent transmission frequency band; then the LED lamp modulation communication module modulates the light signal required to be transmitted by each explosion-proof LED lamp according to each number and the corresponding explosion-proof LED lamp of the transmission frequency band, and completes the modulation through each explosion-proof LED lamp for visible light emission work; wherein the OFDM modulation method (Orthogonal Frequency Division Multiplexing) is the existing method, since the OFDM modulation method can be expressed as follows, the amplitude of sin(t) is a, the amplitude of sin(2t) is b, that is, a is modulated on sin(t), b is modulated on sin(2t), and the two modulated sine waves (subcarriers) are transmitted at the same time: a x sin(t) + b x sin(2t), and when receiving, the two subcarriers are integrated respectively, that is

[0090]

[0091]

[0092] In this way, the original information a and b can be demodulated, and the two subcarriers do not interfere with each other. OFDM transmits different information through multiple subcarriers that do not interfere with each other.

[0093] The baseband signal expression of OFDM is:

[0094] s(t) = b0sin(2πf0t) + b1sin(2πf1t) + … + b N-1 sin(2πf N-1 t)

[0095] + a0cos(2πf0t) + a1cos(2πf1t) + … + a N-1 cos(2πf N-1 t)

[0096] Therefore, the pulse neuron potential dynamics model of the present application can automatically decode the modulated light signal;

[0097] Step four, the photodiode array of each unmanned downhole vehicle can receive the light signal emitted by the explosion-proof LED lamp nearby, and the frequency band of the received light signal can determine the number of the explosion-proof LED lamp emitting the light signal, the photodiode array transmits the received light signal to the microprocessor, the microprocessor first uses the existing filtering algorithm to preliminarily denoise the received light signal, then the processed light signal is extracted through the transform block to obtain a plurality of light pulse sequences, the transform block includes 19 sizes in total, which can be the same as or smaller than the size of the block, the maximum can be 64x64, and the minimum can be 4x4, finally the plurality of light pulse sequences are input into the pulse neuron potential dynamics model trained in step two for processing, and then decoded and restored into readable data, so as to complete the mine visible light communication process.

[0098] The above only describes the preferred embodiments of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, and these improvements and refinements should be considered as the protection scope of the present application.

Claims

1. A working method for a visible light communication system in a mine based on a pulse network demodulation mechanism, characterized in that, The specific steps are as follows: Step 1: In the optical signal demodulation device, a filtering algorithm is used as the signal denoising module to perform preliminary denoising processing on the received optical signal; a network structure of pulse neural network with cascaded transform blocks is used as the feature extraction module and automatic decoding module to extract features from the optical signal and restore it into readable data after automatic decoding processing; the transmitting system is installed at various locations in the mine, and multiple underground unmanned vehicles are distributed at various locations in the mine to complete the deployment of the mine visible light communication system; Step 2: Before use, a spiking neuron potential dynamics model is established using a spiking neural network. The transform block is used to extract features from the light signal to obtain multiple light pulse sequences. Then, the extracted light pulse sequences are input into the spiking neuron potential dynamics model and the spiking neural network synaptic weight learning method is used to train the model. After completion, the required spiking neuron potential dynamics model is obtained. Step 3: When starting visible light communication, firstly, each explosion-proof LED is identified and numbered by the LED identification module, and assigned an independent transmission frequency band to each explosion-proof LED. Then, the LED modulation communication module modulates the light signal to be emitted by each explosion-proof LED according to its number and corresponding transmission frequency band using the OFDM modulation method. After modulation is completed, visible light is emitted through each explosion-proof LED. Step 4: The photodiode arrays of each underground unmanned vehicle can receive light signals emitted by nearby explosion-proof LED lights. By analyzing the frequency band of the received light signal, the number of the explosion-proof LED light emitting the light signal can be determined. The photodiode array transmits the received light signal to the microprocessor. The microprocessor first uses a filtering algorithm to perform preliminary noise reduction on the received light signal, and then extracts the processed light signal through a transform block to obtain multiple light pulse sequences. Finally, the multiple light pulse sequences are input into the spiking neuron potential dynamics model trained in Step 2 for processing, and then decoded to recover readable data, thereby completing the visible light communication process in the mine.

2. The working method according to claim 1, characterized in that, The mine visible light communication system includes a transmitting system and a receiving system. The transmitting system comprises multiple explosion-proof LED lights and a controller. The explosion-proof LED lights are installed at various locations underground. The controller is connected to each explosion-proof LED light for controlling its illumination and communication. The controller includes an LED light identification module and an LED light modulation communication module. The identification module identifies and assigns a unique transmission frequency band to each explosion-proof LED light. The modulation communication module modulates the signal to be transmitted by each explosion-proof LED light before transmitting it for visible light signal transmission. The receiving system... The receiving system includes multiple unmanned underground vehicles (UAVs), each distributed at various locations within the mine. Each UAV is equipped with an optical signal demodulation device and a photodiode array. The optical signal demodulation device is connected to the photodiode array, which receives the visible light signal emitted by the explosion-proof LED lights and transmits it to the optical signal demodulation device. The optical signal demodulation device includes an optical signal denoising module, a feature extraction module, and an automatic decoding module. The optical signal denoising module performs preliminary filtering on the received optical signal. The feature extraction module extracts signal segments from the denoised optical signal. The automatic decoding module decodes the extracted signal segments to recover readable data.

3. The working method according to claim 2, characterized in that, The explosion-proof LED lights are installed in the coal mine transport roadway, the bottom yard, the electromechanical chamber, the return air roadway, and the coal preparation plant.

4. The working method according to claim 2, characterized in that, Each of the underground unmanned vehicles is equipped with two photodiode arrays, which are respectively located at the front and rear of the underground unmanned vehicle to receive visible light signals from the front and rear of the vehicle.

5. The working method according to claim 2 or 4, characterized in that, The photodiode array consists of 2*5 photodiodes.

6. The working method according to claim 2, characterized in that, The optical signal demodulation device is a microprocessor.

7. The working method according to claim 1, characterized in that, The training process of the spiking neuron potential dynamics model in step two is as follows: The potential dynamics model of the spiking neuron is a leakage integral-discharge model, i.e., the LIF model. The spiking neural network is composed of connected spiking neurons. Let the internal voltage of each spiking neuron be v. When no light pulse input is received, the voltage v will stabilize exponentially with time to the equilibrium voltage. This process is described by the LIF model as follows: Solving this differential equation yields: in, It is an arbitrary constant. Control the rate of decline of the index. smaller Exponential change to The faster, as derived from the above equation, initially... time ,in Taking the appropriate value will make Equal to the initial voltage of the spiking neuron, when hour This equation controls the voltage. Exponentially stabilizes to equilibrium voltage over time Since the above equation is for continuous voltage The equation for the change of , however, computers can only simulate discrete processes, taking the discrete time interval as . When the differential equation is discrete, the form is: Furthermore, when a spiking neuron receives a light pulse input at a certain moment, it accumulates that pulse into the voltage, adding a certain value to the current voltage. This value is related to the synaptic weight of the input pulse. The voltage update process is as follows: There is a firing threshold inside the neuron. When the voltage of the spiking neuron At this time, the spiking neuron fires a pulse, after which the voltage of the spiking neuron immediately returns to its resting potential. The synaptic weight learning method for spiking neural networks is a temporally asymmetric form of the Hebb learning rule, influenced by the close temporal correlation between the peak values ​​of presynaptic and postsynaptic neurons. The specific formula is as follows: in, It is the timing of the postsynaptic pulse firing. It is the timing of the presynaptic pulse firing; This is the STDP function; For a given sequence of multiple input light pulses and multiple target light pulses, the spiking neural network seeks a suitable synaptic weight matrix to make the output light pulse sequence of the spiking neuron as close as possible to the corresponding target light pulse sequence, i.e., to minimize the error evaluation function between the two. At this point, the training is complete, and the desired spiking neuron potential dynamics model is obtained.

8. The working method according to claim 7, characterized in that, The desired spiking neuron potential dynamics model was validated. Adopting BCE Loss The loss function determines the degree of inconsistency between the model's predicted value f(x) and the true value Y. The smaller the loss function, the better the robustness of the model. BCE Loss It is the loss function for binary classification, BCE. Loss The formula is: BCE Loss =-(ylog(p(x))+(1-y)log(1-p(x))) Where p(x) is the model output and y is the true label; BCE Loss Function derivation process: so: Since the demodulation of optical signals is a multi-label classification problem, and multi-label classification involves multiple categories, therefore BCE Loss The function outputs not a single value, but a vector. Finally, the sigmoid activation function is used to convert each element of the output vector into a probability value. Based on the probability value, the degree of inconsistency between the model's predicted value f(x) and the true value Y is determined. If it does not exceed the set threshold, the model is considered to have met the requirements. If it exceeds the set threshold, the spiking neuron potential dynamics model is retrained until it meets the requirements.

Images