An image processing device and restoration method based on VCSEL neurons

Abstract

The application provides a kind of image processing device and recovery method based on VCSEL neuron, including picture acquisition module, picture processing module and terminal recovery and display module, camera and Wi-Fi wireless communication module are equipped in picture acquisition module, picture processing module includes arbitrary waveform generator AWG, Mach-Zehnder modulator MZM and vertical cavity surface emitting laser VCSEL, second polarization controller PC2, variable attenuator VA and optical fiber ring OC are sequentially arranged between Mach-Zehnder modulator MZM and vertical cavity surface emitting laser VCSEL, the whole system of complete image information acquisition, processing, recovery and display is constructed in the application, information acquisition and display are carried out using electronic technology, optical information is processed based on the photonic technology of semiconductor laser to realize high-speed image coding and decoding, the output optical signal can be recovered into binary coded signal, to realize the successful recovery of image information.

Description

Technical Field

[0001] This invention relates to the field of encoding and restoration technology between image binary code and spike sequences, and particularly to an image processing device and restoration method based on VCSEL neurons. Background Technology

[0002] From the first computer, the Electronic Digital Integrator (ENIAC), to the present day, in just over half a century, computers have evolved from simple mathematical calculations to capable of performing much more complex learning tasks. Especially since AlphaGo defeated professional Go players in human-computer interaction, neural networks have experienced tremendous growth. However, artificial neural networks based on the traditional von Neumann architecture suffer from the drawback of tidal data access and struggle to meet the demands of rapid transmission of large amounts of data. Based on this, photonic neural computing has emerged. Compared to traditional technologies, neural networks can simulate the processing and transmission of information in biological neural networks. A neuron consists of four parts: axon, dendrites, synapse, and cell body. Multiple pieces of information are collected by the axon, processed, and then transmitted from one axon to another via synapses. Since neuronal signal transmission propagates through specific threshold signals, we encode information based on the spike signals output by the VCSEL. This technology offers advantages such as high information processing speed, low loss, and easy compatibility with existing fiber optic systems.

[0003] In recent years, research on neuromorphic photonic devices has attracted widespread attention from researchers, including photonic crystal structures, resonant tunneling diode photodetectors, fiber lasers, semiconductor optical amplifiers, optical modulators, and semiconductor lasers. Among these neuromorphic photonic devices, semiconductor lasers exhibit a variety of behaviors similar to biological neurons and can generate impulse responses up to eight orders of magnitude faster than biological neurons, thus serving as an ideal artificial neuron. Currently, different types of semiconductor lasers have been widely studied and applied in neuromorphic computing systems, such as microdisk lasers, micropillar lasers, microring lasers, quantum dot lasers, lasers with saturable absorbers, and vertical-cavity surface-emitting lasers (VCSELs). In particular, VCSELs have advantages such as small size, low power consumption, low cost, and high coupling efficiency with optical fibers. Therefore, exploring the nonlinear dynamics of VCSEL devices and their applications in neuromorphic computing is expected to promote innovative development in the field of artificial intelligence.

[0004] Currently, image communication, real-time image encoding, and reconstruction are mostly implemented using electrical methods and algorithms. This approach is limited by the processing speed and energy consumption of electronic devices. Research on image processing and reconstruction applications based on VCSEL neuromorphic devices is still in its early stages. To date, research on VCSEL photonic neurons has primarily focused on the basic characteristics of neurons, with limited research on relevant practical applications. This clearly does not meet practical needs. Therefore, this invention proposes an image processing device and reconstruction method based on VCSEL neurons to address the problems existing in the prior art, aiming to promote the application and development of VCSEL neuromorphic devices in future high-speed neuromorphic computing fields. Summary of the Invention

[0005] To address the aforementioned problems, the present invention aims to propose an image processing device and restoration method based on VCSEL neurons. This image processing device and restoration method based on VCSEL neurons has advantages such as fast response and high restoration accuracy, thus solving the problems existing in the prior art.

[0006] To achieve the objectives of this invention, the invention is implemented through the following technical solution: an image processing device based on VCSEL neurons, comprising an image acquisition module, an image processing module, and a terminal recovery and display module. The image acquisition module includes a camera and a Wi-Fi wireless communication module. The image processing module includes an arbitrary waveform generator (AWG), a Mach-Zehnder modulator (MZM), and a vertical-cavity surface-emitting laser (VCSEL). An RF amplifier (RF) is provided between the AWG and the MZM. A second polarization controller (PC2), a variable attenuator (VA), and an optical fiber circulator (OC) are sequentially provided between the MZM and the VCSEL. The terminal recovery and display module includes a field-programmable gate array (FPGA) and a photodetector (PD). The photodetector (PD) is connected to the output of the optical fiber circulator (OC).

[0007] A further improvement is that the image acquisition module is also equipped with a continuously adjustable laser TL and a first polarization controller PC1. The continuously adjustable laser TL is connected to a Mach-Zehnder modulator MZM through the first polarization controller PC1.

[0008] A further improvement is that the vertical cavity surface-emitting laser (VCSEL) is equipped with a temperature and current controller TEMP&I.

[0009] A further improvement is that the pixel value obtained by the camera is within the range of 0-255, and each pixel is represented by 8 bits of binary, that is, a binary sequence containing image information is obtained.

[0010] A further improvement is that the sampling rate of the vertical cavity surface-emitting laser (VCSEL) and the field-programmable gate array (FPGA) is in the GHz range.

[0011] An image restoration method based on VCSEL neurons includes the following steps:

[0012] Step 1: Capture image information pixel values ​​on-site using the camera.

[0013] Step 2: Use the Wi-Fi wireless communication module to transmit the pixel value signal obtained in Step 1 to the arbitrary waveform generator (AWG).

[0014] Step 3: Use a continuously tunable laser source (TL) and a Mach-Zehnder modulator (MZM) to convert the electrical signal into an optical signal, giving it phase and amplitude information;

[0015] Step 4: Inject the optical signal from Step 3 into the Vertical Cavity Surface Emitting Laser (VCSEL);

[0016] Step 5: Acquire the signals output by the neurons, perform discrimination processing using a field-programmable gate array (FPGA), then restore them to a binary sequence, and adjust the stimulation signals of the input neurons according to the displayed results to achieve image encoding and processing.

[0017] A further improvement lies in the following: In step three, the specific steps for converting the electrical signal of the image into an optical signal are as follows:

[0018] The continuous light from the continuously tunable laser source TL is modulated by the first polarizer PC1 to the input port of the Mach-Zehnder modulator MZM. An amplified signal from the arbitrary waveform generator AWG is connected to the radio frequency input port of the Mach-Zehnder modulator MZM. A light signal with phase and intensity can be obtained at the output port of the Mach-Zehnder modulator MZM.

[0019] A further improvement is that the process of restoring the binary sequence in step five includes:

[0020] S1: Under "0" level injection, the VCSEL operates in the injection-locked region, stably outputting P. locking It is available.

[0021] S2: Under negative pulse injection, with fixed injection intensity and frequency detuning, the VCSEL is in the peak dynamics region, and the average value of the pulse peak value is defined as P. a With each "1" level signal injected, the VCSEL correspondingly excites a spike mode.

[0022] S3: Based on steps S1 and S2, the peak pulse sequence output by VCSEL is initially recovered into a binary sequence T.

[0023] S4: Set the jitter range threshold for stable laser output to P. v When the amplitude change of the pulse sequence is less than P v At that time, it is determined as the starting point of the injection locking region.

[0024] S5: The compensation time Δt for each "1" level and "0" level mode is determined by calculating the time difference between the last pulse in the first pulse mode of VCSEL excitation and the starting point of the injection lock region.

[0025] S6: Set the time T for each "1" level or "0" level mode. p1 or T p0 Divide by minimum mode time T m The bits are rounded to the nearest integer m, and finally, based on the number of bits m obtained in each mode and the initially recovered binary sequence T, the encoded pulse information is successfully recovered as a binary sequence.

[0026] The beneficial effects of this invention are as follows:

[0027] (1) It can realize the acquisition, transmission, calculation, processing and recovery of image information;

[0028] (2) This technology has the advantages of high speed, low loss and integrability.

[0029] (3) It has good compatibility and can be compatible with existing fiber optic systems;

[0030] (4) Using VCSEL as a photonic nerve can not only simulate the impulse response characteristics of biological nerves, but its response rate is also up to 8 orders of magnitude faster than that of biological nerves.

[0031] (5) Neuron weight adjustment can be accomplished using a common adjustable attenuator.

[0032] (6) The method for recovering optical signals from electrical signals is intuitive and easy to operate. Attached Figure Description

[0033] Figure 1 This is a schematic diagram of the device structure of the present invention.

[0034] Figure 2 This is a schematic diagram of the original image input for this invention.

[0035] Figure 3 This is a schematic diagram of the image encoding, processing, and restoration process of the present invention.

[0036] Figure 4 This is a schematic diagram of the restored image of the present invention. Detailed Implementation

[0037] To enhance understanding of the present invention, the present invention will be further described in detail below with reference to embodiments. These embodiments are only used to explain the present invention and do not constitute a limitation on the scope of protection of the present invention.

[0038] Example 1

[0039] according to Figure 1 As shown, this embodiment proposes an image processing device based on VCSEL neurons, including an image acquisition module, an image processing module, and a terminal recovery and display module. The image acquisition module is equipped with a camera and a Wi-Fi wireless communication module. The camera converts the analog signal of the image into a digital signal, which is then processed by the Wi-Fi wireless communication module to obtain an electrical signal sequence containing image information.

[0040] The image processing module includes an arbitrary waveform generator (AWG), a Mach-Zehnder modulator (MZM), and a vertical-cavity surface-emitting laser (VCSEL). The AWG is connected to the output of the Wi-Fi wireless communication module. Therefore, the frequency of the signal is adjusted to GHz by the AWG. The information to be processed is loaded onto the optical carrier after passing through the MZM. At the same time, an RF amplifier (RF) is provided between the AWG and the MZM. A second polarization controller (PC2), a variable attenuator (VA), and an optical fiber circulator (OC) are sequentially provided between the MZM and the VCSEL. Then, the optical signal loaded with information (the aforementioned optical carrier) sequentially passes through the second polarization controller (PC2), the variable attenuator (VA), and the optical fiber circulator (OC) into the VCSEL.

[0041] The terminal recovery and display module includes a field-programmable gate array (FPGA) and a photodetector (PD). The photodetector (PD) is connected to the output of the fiber optic circulator (OC). The signal from the fiber optic circulator (OC) is converted into an electrical signal by the photodetector (PD) and then input into the FPGA for processing and recovery into image information.

[0042] The image acquisition module also includes a continuously adjustable laser TL and a first polarization controller PC1. The continuously adjustable laser TL is connected to the Mach-Zehnder modulator MZM through the first polarization controller PC1.

[0043] The vertical cavity surface-emitting laser (VCSEL) is equipped with a temperature and current controller, TEMP&I.

[0044] The pixel values ​​obtained by the camera are within the range of 0-255. Each pixel is represented by 8 bits of binary, which can obtain a binary sequence containing image information.

[0045] The sampling rate of the vertical cavity surface-emitting laser (VCSEL) and the field-programmable gate array (FPGA) is in the GHz range.

[0046] Specifically, in this embodiment, image information is acquired by a camera, processed by a Wi-Fi wireless communication module, and then transmitted to an arbitrary waveform generator (AWG). The signal output from the AWG is modulated by a Mach-Zehnder modulator (MZM) to a continuous wave generated by a continuously tunable laser (TL) to obtain an optical binary sequence containing image information. The optical signal output from the MZM is injected into a vertical-cavity surface-emitting laser (VCSEL) for computation. The signal output from the VCSEL is converted into an electrical signal by a photodetector (PD) and then input into a field-programmable gate array (FPGA). According to the set recovery method, the signal is recovered from the spikes into a binary sequence, and the frequency of the signal input to the AWG is adjusted in real time according to the FPGA rate to realize the entire process of image acquisition, encoding, processing, recovery, and display.

[0047] like Figure 1 As shown, in this embodiment, the signal acquisition is as follows: the image analog signal is acquired by the camera, converted into pixel value signal, and the obtained pixel value signal is transmitted to the transmitting end Wi-Fi wireless communication module for remote wireless communication.

[0048] The signal is transmitted as follows: the aforementioned transmitting Wi-Fi wireless communication module transmits the information containing pixel values ​​to the receiving Wi-Fi wireless communication module via remote transmission.

[0049] The optical encoding of the image is as follows: the signal is transmitted to the arbitrary waveform generator (AWG) via the Wi-Fi wireless communication module of the receiver. The frequency of the pixel value is set by the arbitrary waveform generator (AWG) to encode the image value into time domain information, and the electrical signal is amplified by the radio frequency amplifier (RF).

[0050] The steps for converting the electrical signal of an image into an optical signal are as follows: the continuous light from the continuously tunable laser TL is adjusted to the input port of the Mach-Zehnder modulator MZM by the first polarization controller PC1, the amplified signal from the arbitrary waveform generator AWG is connected to the radio frequency input port of the Mach-Zehnder modulator MZM, and an optical signal with phase and intensity can be obtained at the output port of the Mach-Zehnder modulator MZM.

[0051] The image processing steps are as follows: After the optical signal passes through the second polarization controller PC2, it enters the vertical cavity surface emitter laser (VCSEL) through the fiber optic circulator OC;

[0052] The image restoration and display section involves transmitting the output signal processed by the vertical cavity surface-emitting laser (VCSEL) through a photodetector (PD) to a field-programmable gate array (FPGA).

[0053] An image restoration method based on VCSEL neurons includes the following steps:

[0054] Step 1: Capture image information pixel values ​​on-site using the camera.

[0055] Step 2: Use the Wi-Fi wireless communication module to transmit the pixel value signal obtained in Step 1 to the arbitrary waveform generator (AWG).

[0056] Step 3: The electrical signal is converted into an optical signal using a continuously adjustable laser source TL and a Mach-Zehnder modulator MZM, giving it phase and amplitude information. The continuous light from the continuously adjustable laser source TL is adjusted to the input port of the Mach-Zehnder modulator MZM through the first polarizer PC1. An amplified signal from an arbitrary waveform generator AWG is connected to the radio frequency input port of the Mach-Zehnder modulator MZM. An optical signal with phase and intensity can be obtained at the output port of the Mach-Zehnder modulator MZM.

[0057] Step 4: Inject the optical signal from Step 3 into the Vertical Cavity Surface Emitting Laser (VCSEL);

[0058] Step 5: Acquire the signal output by the neuron, perform discrimination processing using a field-programmable gate array (FPGA), then restore it to a binary sequence, and adjust the stimulation signal of the input neuron according to the displayed result to achieve image encoding and restoration.

[0059] The specific process of restoring the binary sequence includes:

[0060] S1: Under "0" level injection, the VCSEL operates in the injection-locked region, stably outputting P. locking It is available.

[0061] S2: Under negative pulse injection, with a fixed injection intensity (corresponding to "1" level) and frequency detuning, the VCSEL is in the peak dynamics region, and the average value of the pulse peak value is defined as P. a With each "1" level signal injected, the VCSEL correspondingly excites a spike mode.

[0062] S3: Based on steps S1 and S2, the peak pulse sequence output by VCSEL is initially recovered into a binary sequence T.

[0063] S4: Set the jitter range threshold for stable laser output to P. v When the amplitude change of the pulse sequence is less than Pv At that time, it is determined as the starting point of the injection locking region.

[0064] S5: The compensation time Δt for each "1" level and "0" level mode is determined by calculating the time difference between the last pulse in the first pulse mode of VCSEL excitation and the starting point of the injection lock region.

[0065] S6: Set the time T for each "1" level or "0" level mode. p1 or T p0 Divide by minimum mode time T m (The time length of the corresponding unit bit signal) is rounded to the nearest integer m. Finally, based on the number of bits m obtained in each mode and the initially recovered binary sequence T, the encoded pulse information is successfully recovered as a binary sequence.

[0066] Example 2

[0067] according to Figures 1-4 As shown, this embodiment proposes an image processing device based on VCSEL neurons, including an image acquisition module, an image processing module, and a terminal recovery and display module. The image acquisition module includes a camera and a Wi-Fi wireless communication module. The pixel values ​​obtained by the camera are within the range of 0-255, and each pixel is represented by 8 bits of binary, thus obtaining a binary sequence containing image information. The image processing module includes an arbitrary waveform generator (AWG), a Mach-Zehnder modulator (MZM), and a vertical-cavity surface-emitting laser (VCSEL). An RF amplifier is provided between the AWG and the MZM. The MZM and the VCSEL are connected... Between the VCSELs, there is a second polarization controller PC2, a variable attenuator VA, and a fiber optic circulator OC. The image acquisition module also includes a continuously adjustable laser TL and a first polarization controller PC1. The continuously adjustable laser TL is connected to a Mach-Zehnder modulator MZM through the first polarization controller PC1. The terminal recovery and display module includes a field-programmable gate array (FPGA) and a photodetector PD. The photodetector PD is connected to the output of the fiber optic circulator OC. Specifically, the vertical-cavity surface-emitting laser (VCSEL) includes a temperature and current controller TEMP&I. The sampling rate of the VCSEL and the FPGA is in the GHz range.

[0068] In this embodiment, a photonic neural network capable of image encoding and processing is constructed using a camera, a Wi-Fi wireless communication module, a vertical-cavity surface-emitting laser (VCSEL), and a field-programmable gate array (FPGA). Figure 1As shown, the camera acquires image information from the outside and converts it into pixel value information, realizing the conversion from analog to digital information. The camera's output port is connected to the GPIO port of the Wi-Fi wireless communication module, and the pixel value information is sent wirelessly to the Wi-Fi wireless communication module receiver. The Wi-Fi wireless communication module is a short-range wireless communication technology developed based on the IEEE 802.15.4 protocol. It has low power consumption and is considered by the industry to be the most likely wireless method for industrial control applications. It is a wireless data transmission network platform composed of up to 65,000 wireless data transmission modules. Within the entire network range, each ZigBee network data transmission module can communicate with each other. The Wi-Fi communication modules operate in the 2.4GHz and 5GHz frequency bands and use the IEEE 802.11 standard to implement data transmission. The module contains radio frequency (RF) circuits, antennas, modems, microcontrollers, and network protocol stacks, which work together to establish and maintain network connections. Throughout the network, these modules allow devices to communicate and transmit data wirelessly over short distances, which can be achieved using two Wi-Fi networks. For longer transmission distances, Wi-Fi networking or 5G base stations can be used to ensure that the pixel value information of the image can be transmitted to the Wi-Fi wireless communication module receiver. At the receiver, the information is transmitted to the Arbitrary Waveform Generator (AWG). The signal with pixel information is sampled by setting the sampling rate through the AWG, which can give the resulting digital signal timing information. The signal output from the AWG is modulated by the Mach-Zehnder modulator (MZM) onto a continuous wave generated by a tunable laser. The signal output from the MZM modulator (the amplitude of the output optical signal can be adjusted by adjusting the operating point) is injected into the Vertical Cavity Surface Emitting Laser (VCSEL) (neuron).

[0069] The signal output by the VCSEL neuron is converted into an electrical signal by a photodetector (PD) and then input into a field-programmable gate array (FPGA) for processing and display. Based on the calculation results, the stimulation signal of the input neuron is adjusted to enable accurate encoding, recognition, transmission, and recovery of images.

[0070] Specifically, the signal collected by the camera is converted into the laser field for calculation and processing, and finally the result is displayed through a field-programmable gate array (FPGA). This can significantly improve the signal processing speed and greatly reduce the system's energy consumption.

[0071] In addition, this device can process image information in real time, and the use of a field-programmable gate array (FPGA) for real-time display can change the intensity of the input signal and the frequency detuning of the injected vertical cavity surface-emitting laser (VCSEL) at any time, so that the signal can be processed accurately.

[0072] The image information flow includes end-to-end transmission from the camera to the Wi-Fi wireless communication module, wireless transmission from the Wi-Fi wireless communication module transmitter to the Wi-Fi wireless communication module, end-to-end transmission from the Wi-Fi wireless communication module receiver to the arbitrary waveform generator (AWG), and transmission from the output of the vertical cavity surface emitter laser (VCSEL) to the field programmable gate array (FPGA).

[0073] End-to-end transmission from camera to Wi-Fi wireless communication module: The camera acquires image analog signals, converts them into pixel value signals, and then transmits the pixel value signals output by the camera sensor module to the transmitting Wi-Fi wireless communication module for processing.

[0074] Wi-Fi transmitter-to-Wi-Fi wireless transmission: The digital signal containing image pixel values ​​received by the Wi-Fi wireless communication module is remotely transmitted from the transmitter to a remote Wi-Fi wireless communication module.

[0075] End-to-end transmission from the Wi-Fi receiver to the arbitrary waveform generator (AWG): The signal from the Wi-Fi wireless communication module at the receiver is directly connected to the arbitrary waveform generator (AWG) to convert the image information into time-domain information. The frequency of the pixel value is set by the arbitrary waveform generator (AWG) to encode the image value, and then the electrical signal is amplified by the radio frequency amplifier (RF) to increase the power of the electrical signal.

[0076] The output of the VCSEL neuron is as follows: the continuous light from the continuously tunable laser TL is adjusted to the input port of the Mach-Zehnder modulator MZM by the first polarization controller PC1. The amplified signal from the arbitrary waveform generator AWG is connected to the radio frequency input port of the Mach-Zehnder modulator MZM. The output port of the Mach-Zehnder modulator MZM can obtain a light signal with phase and intensity. After passing through the second polarization controller PC2, the light signal enters the VCSEL neuron through the fiber optic circulator OC.

[0077] Communication from the VCSEL neuron end to the FPGA end: The output signal processed by the vertical cavity surface emitter laser (VCSEL) is transmitted to the FPGA after passing through the photodetector (PD).

[0078] An image restoration method based on VCSEL neurons includes the following steps:

[0079] Step 1: Capture image information pixel values ​​on-site using the camera.

[0080] Step 2: Use the Wi-Fi wireless communication module to transmit the pixel value signal obtained in Step 1 to the arbitrary waveform generator (AWG).

[0081] Step 3: Using a continuously tunable laser source (TL) and a Mach-Zehnder modulator (MZM), the electrical signal is converted into an optical signal, carrying phase and amplitude information. The specific steps for converting the image's electrical signal into an optical signal are as follows:

[0082] The continuous light from the continuously tunable laser source TL is adjusted to the input port of the Mach-Zehnder modulator MZM through the first polarizer PC1. An amplified signal from the arbitrary waveform generator AWG is connected to the radio frequency input port of the Mach-Zehnder modulator MZM. A light signal with phase and intensity can be obtained at the output port of the Mach-Zehnder modulator MZM.

[0083] Step 4: Inject the optical signal from Step 3 into the Vertical Cavity Surface Emitting Laser (VCSEL);

[0084] Step 5: Acquire the signal output by the neuron, perform discrimination processing using a field-programmable gate array (FPGA), then restore it to a binary sequence, and adjust the stimulation signal of the input neuron according to the displayed result to achieve image encoding and restoration.

[0085] The specific process of restoring the binary sequence includes:

[0086] S1: Preliminary recovery of the spike pulse sequence to a binary sequence: Under fixed injection intensity and frequency detuning, the vertical-cavity surface-emitting laser (VCSEL) is in a region capable of handling spike dynamics. Here, we define the average value of the pulse timing peaks as P. a Therefore, we can determine the answer based on P. a The start and end positions of each response pulse pattern are roughly determined, corresponding to a "1" level, while other regions correspond to a "0" level. That is, binary data is represented using the pulse patterns of the VCSEL, where the pulse duration corresponds to the encoded "1" and "0" level durations. We use the pulse peak value P. a Half of the threshold is used to distinguish between "0" and "1" levels. Pulses exceeding the threshold are considered to be at the "1" level, while pulses below the threshold are considered to be at the "0" level.

[0087] S2: Determine the timing position from the injection lock region to the peak dynamics start point: Under the injection condition corresponding to "0" level, the vertical-cavity surface-emitting laser (VCSEL) will operate in the injection lock region, and due to the small vibration caused by noise, a stable intensity output P can be obtained. locking The jitter range threshold P when the amplitude change is less than the stable output of the laser. v In this case, the corresponding point is determined as the starting point of the injection locking region.

[0088] S3: Determine the compensation time for each response mode. Based on the time difference between the last pulse in the first pulse mode and the starting point of the injected locking region, the compensation time Δt for each "1" level mode and "0" level mode can be determined. Then, by considering the compensation time on the timing obtained in step a, the preliminary recovered binary sequence T can be obtained.

[0089] S4: Determine the length of the binary sequence for each mode, and assign the time Tp1 to each "1" level mode or the time T to each "0" level mode. p0 Divide by minimum mode time T m (Corresponding to the duration of one bit), then rounded to the nearest integer m, described by the formula:

[0090] m = round(T p0 / T m )

[0091] Finally, based on the number of bits m obtained in each mode and the initially recovered binary sequence T, the encoded pulse information is successfully recovered as a binary sequence.

[0092] like Figure 2 , Figure 3 and Figure 4 As shown, (a) is the original image, (b1)-(b3) are the image encoding, processing and restoration processes respectively (b1 and b3 are reversed because of the working principle of the laser, the amplitude corresponding to the injection lock is higher than the dynamic state (spiking dynamics) replication, but the position where the spiking signal is generated should be "1" in the logical restoration), and (c) is the restored image.

[0093] Specifically, the VCSEL neuron is mainly affected by temperature, current, frequency detuning, and injected light intensity. To achieve a specific task, only these four parameters need to be fixed to determine the laser's operating state. Temperature and current are controlled by the TEMP&I thermostat. Changing the frequency of the injected light controls the coupling state with the VCSEL, keeping these parameters constant. During image encoding and processing, a fixed light intensity is first set, and the image information is transmitted to the VCSEL neuron for processing. The photodetector (PD) transmits the information to the field-programmable gate array (FPGA) for reconstruction, display, and effect observation. The results displayed by the FPGA are compared with the target image. By changing the operating point of the variable light attenuator and the MZM (Mechanical Zyme Array), the weight of the injected light is altered, thus achieving real-time encoding and reconstruction of image information. Furthermore, by comparing the signal obtained after processing by the FPGA with the information obtained by the camera, the image recognition error information can be calculated, thereby evaluating the system's real-time image performance.

[0094] The foregoing has shown and described the basic principles, main features, and advantages of the present invention. Those skilled in the art should understand that the present invention is not limited to the above embodiments. The embodiments and descriptions in the specification are merely illustrative of the principles of the invention. Various changes and modifications can be made to the invention without departing from its spirit and scope, and all such changes and modifications fall within the scope of the present invention as claimed. The scope of protection of the present invention is defined by the appended claims and their equivalents.

Claims

1. A VCSEL neuron-based image processing device, comprising a picture acquisition module, a picture processing module and a terminal recovery and display module, characterized in that: The picture acquisition module is internally provided with a camera and a Wi-Fi wireless communication module, the picture processing module comprises an arbitrary waveform generator AWG, a Mach-Zehnder modulator MZM and a vertical cavity surface emitting laser VCSEL, a radio frequency amplifier RF is arranged between the arbitrary waveform generator AWG and the Mach-Zehnder modulator MZM, a second polarization controller PC2, a variable attenuator VA and an optical fiber ring OC are sequentially arranged between the Mach-Zehnder modulator MZM and the vertical cavity surface emitting laser VCSEL, and the terminal recovery and display module comprises a field programmable logic gate array FPGA and a photodetector PD, the photodetector PD is connected with the output end of the optical fiber ring OC; The image recovery method of the VCSEL neuron-based image processing device comprises the following steps: Step one: the camera acquires pixel values of picture information on site Step two: the Wi-Fi wireless communication module is used for wireless communication, and the pixel value signal obtained in step one is transmitted to the arbitrary waveform generator AWG; Step three: the continuous tunable laser source TL and the Mach-Zehnder modulator MZM are used to convert the electrical signal into an optical signal, so that the optical signal has phase and amplitude information; Step four: the optical signal in step three is injected into the vertical cavity surface emitting laser VCSEL; Step five: the signal output by the neuron is collected, and the field programmable logic gate array FPGA is used for discrimination processing, then the binary sequence is recovered, and the stimulation signal of the input neuron is adjusted and controlled according to the display result, so that the encoding and processing of the picture are realized, and the process of recovering the binary sequence in step five comprises: S1 : In the case of "0" level injection, the VCSEL works in the injection locking region, and the stable output P locking can be obtained; S2: In the case of negative pulse injection, when the injection intensity and frequency detuning are fixed, the VCSEL is in the spiking dynamics region, and the average value of the pulse peak is defined as P a Each "1" level signal injection, the VCSEL corresponding to a spiking mode; S3: Based on steps S1 and S2, the sequence of spike pulses output by the VCSEL is preliminarily recovered into a binary sequence T ; S4: set the jitter range threshold value when the laser is stable output as P v When the amplitude variation of the pulse sequence is less than P v The starting point of the injection locking region is determined. S5: Determine the compensation time Δ for each "1" level and "0" level pattern by calculating the time difference between the last pulse of the first pulse pattern excited by the VCSEL and the start point of the injection locking region t ; S6: divide the time of each "1" level or "0" level pattern by the minimum pattern time T p1 or T p0 round up to the nearest integer m, and finally recover the encoded pulse information as a binary sequence from the number of bits m obtained within each pattern and the preliminary recovered binary sequence T m T , successively.​ 2. The VCSEL neuron based image processing apparatus according to claim 1, wherein: The picture acquisition module is internally provided with a continuous tunable laser TL and a first polarization controller PC1, and the continuous tunable laser TL is connected with the Mach-Zehnder modulator MZM through the first polarization controller PC1.

3. The VCSEL neuron based image processing apparatus according to claim 1, wherein: The vertical cavity surface emitting laser VCSEL is internally provided with a temperature current controller TEMP&I.

4. The VCSEL neuron-based image processing apparatus according to claim 1, wherein: The pixel value obtained by the camera is 0-255, and each pixel point is represented by 8-bit binary, that is, a binary sequence with image information is obtained.

5. The VCSEL neuron based image processing apparatus according to claim 1, wherein: The sampling rate of the vertical cavity surface emitting laser VCSEL and the field programmable logic gate array FPGA is GHz.

6. The VCSEL neuron based image processing apparatus according to claim 1, wherein: In step three, the specific steps of converting the electrical signal of the image into an optical signal are as follows: The continuous light of the continuous tunable laser source TL is adjusted to the input port of the Mach-Zehnder modulator MZM through the first polarization controller PC1, the amplified signal from the arbitrary waveform generator AWG is connected to the radio frequency input port of the Mach-Zehnder modulator MZM, and the optical signal with phase and intensity can be obtained at the output port of the Mach-Zehnder modulator MZM.

Images