A DVS data encryption method, decryption method, device and system

Abstract

This invention relates to a DVS data encryption method, decryption method, apparatus, and system, comprising: acquiring an original event stream E; using an initial key parameter K to drive a one-dimensional chaotic map and LIF neurons, extracting the chaotic pulse intervals corresponding to adjacent firing pulses of LIF neurons, and performing least significant bit purification processing to generate a binary key stream KS; based on the binary key stream KS, using Fisher-Yates shuffling traversal to generate a time rearrangement mapping table and a spatial permutation mapping table; sequentially extracting events from the original event stream E, using the binary key stream KS as the overall driver, relying on the time rearrangement mapping table and the spatial permutation mapping table to complete global scrambling of timestamps, one-dimensional flattening and permutation of spatial coordinates, and combining the binary key stream KS to generate a key mask, completing event polarity reversal, and obtaining an encrypted event stream E'. This invention can reduce system latency and reduce computational resources.

Description

Technical Field

[0001] This invention relates to the field of data security and information processing technology, specifically to a DVS data encryption method, decryption method, apparatus, and system. Background Technology

[0002] In the security protection of DVS event streams, existing computer systems and algorithms typically employ the following two types of technical solutions:

[0003] Block encryption systems based on traditional symmetric block ciphers (such as the Advanced Encryption Standard). When processing streaming data from peripheral inputs, these systems typically allocate a buffer in memory, forcing asynchronously arriving events to be aggregated and filled into frame-by-frame images (containing numerous zero-pixel values). Subsequently, the encryption engine performs multiple rounds of algebraic operations on the data using the same key, ultimately outputting fixed-length ciphertext.

[0004] A streaming permutation system based on spatial dimension chaotic sorting. To adapt to the sparsity of DVS event streams, some existing systems generate continuous pseudo-random sequences through iterative two-dimensional or high-dimensional chaotic mapping. Its core processing logic is as follows: within each key update cycle, the underlying computer sorting algorithm is invoked to fully sort the generated continuous pseudo-random sequence; subsequently, using the trajectories of the sorted element positions as a spatial dimension mapping table, the two-dimensional spatial coordinates of the event stream are rearranged, supplemented by conditional polarity reversal.

[0005] The above two methods have the following drawbacks when processing DVS event streams with asynchronous and highly sparse characteristics:

[0006] (1) In the process of processing DVS event streams, standard symmetric block ciphers need to cache the DVS event streams and wait for them to accumulate to the expected length before triggering the execution logic of the encryption and decryption modules. This causes the DVS event streams to lose their asynchronous response advantage and introduces huge system latency. Due to the highly sparse nature of DVS event streams, the encryption of each frame of the image by standard symmetric block ciphers involves a large number of meaningless 0 pixel value operations, which leads to the consumption of computing resources.

[0007] (2) In a permutation system based on chaotic sorting, full sorting leads to a nonlinear computing bottleneck. The system must perform a sorting operation on the sequence each time the keystream is refreshed. For long-term DVS event streams, to prevent key reuse, the system must frequently refresh the sequence, drastically amplifying the sorting overhead and squeezing out limited computing resources. Furthermore, the permutation system based on chaotic sorting performs permutations on spatial coordinates, exposing the event stream's timeline in plaintext. Attackers can exploit the unprotected temporal continuity to reconstruct the network and reverse-engineer the motion picture, posing a serious security risk.

[0008] Therefore, it is necessary to develop a new DVS data encryption method, decryption method, device, and system. Summary of the Invention

[0009] The purpose of this invention is to provide a DVS data encryption method, decryption method, apparatus and system that can reduce system latency and reduce computing resources.

[0010] In a first aspect, the DVS data encryption method of the present invention is applied to a data encryption terminal, and the method includes the following steps:

[0011] S1: Data Acquisition: Acquire raw event stream E, which is generated and output by the DVS sensor deployed at the edge asynchronously capturing pixel-level light intensity changes. The raw event stream E is a sparse binary event stream, and each event contains a two-dimensional spatial horizontal coordinate, a two-dimensional spatial vertical coordinate, a time index, and a binary polarity representing the brightness change.

[0012] S2: Key stream generation: Using the initial key parameter K to drive the one-dimensional chaotic mapping and LIF neuron, extract the chaotic pulse interval corresponding to the adjacent firing pulses of the LIF neuron, and perform least significant bit purification processing on the chaotic pulse interval to generate a binary key stream KS.

[0013] S3: Mapping table generation: Based on the binary key stream KS, Fisher-Yates shuffling traversal is used to generate a time rearrangement mapping table and a spatial permutation mapping table;

[0014] S4: Multidimensional encryption: sequentially extract the events in the original event stream E, drive the binary key stream KS as the whole, rely on the time rearrangement mapping table and the spatial permutation mapping table to complete the global scrambling of timestamps, one-dimensional flattening and permutation of spatial coordinates, and combine the binary key stream KS to generate a key mask, complete the event polarity reversal, and obtain the encrypted event stream E'.

[0015] S5: Data transmission: Send the encrypted event stream E' outward through the IoT channel to complete the encrypted transmission of the DVS raw event stream.

[0016] Optionally, in step S2, the one-dimensional chaotic mapping is a Logistic mapping, and its system control parameter r is 3.99; the membrane potential decay factor α of the LIF neuron is 0.5, and the hardware pulse triggering threshold Vth is 2.0.

[0017] Optionally, in step S2, the one-dimensional chaotic mapping is a Tent mapping or a Chebyshev mapping.

[0018] Optionally, during the encryption process, the LIF neuron membrane potential, binary key stream KS, time rearrangement mapping table, and spatial permutation mapping table are all stored in local volatile main memory; after encryption is completed, they are cleaned up and not persistently written to the outside.

[0019] Secondly, the DVS data encryption device of the present invention includes:

[0020] The DVS event acquisition module is configured to acquire the raw event stream E, which is generated and output by the DVS sensor deployed at the edge asynchronously capturing pixel-level light intensity changes. The raw event stream E is a sparse binary event stream, and each event contains a two-dimensional spatial horizontal coordinate, a two-dimensional spatial vertical coordinate, a time index, and a binary polarity representing the brightness change.

[0021] The encryption key generation module is configured to call the initial key parameter K to drive the one-dimensional chaotic mapping and LIF neuron, extract the chaotic pulse interval corresponding to the adjacent firing pulses of the LIF neuron, and perform least significant bit purification processing on the chaotic pulse interval to generate a binary key stream KS.

[0022] The mapping table generation module is configured to generate time rearrangement mapping tables and spatial permutation mapping tables based on the binary key stream KS through Fisher-Yates shuffling traversal;

[0023] The multidimensional encryption module is configured to sequentially extract events from the original event stream E, drive the binary key stream KS as a whole, rely on the time rearrangement mapping table and the spatial permutation mapping table to complete the timestamp scrambling and spatial coordinate permutation, and combine the binary key stream KS to generate a key mask to complete the polarity reversal, thereby obtaining the encrypted event stream E' and sending it outward.

[0024] Optionally, the one-dimensional chaotic mapping in the encryption key generation module is any one of the Logistic mapping, Tent mapping, or Chebyshev mapping, and the system is deployed at the edge node of the neuromorphic industrial Internet of Things.

[0025] Thirdly, the DVS data decryption method based on chaotic pulse interval driving described in this invention is applied to a data decryption terminal, and the method includes the following steps:

[0026] S1: Data reception: Receive the encrypted event stream E' sent by the data encryption terminal, wherein the encrypted event stream E' is obtained by the DVS data encryption method as described in this invention;

[0027] S2: Key replication: Obtain the same initial key parameter K as the data encryption end, drive the same one-dimensional chaotic mapping and LIF neuron as the data encryption end, obtain the chaotic pulse interval corresponding to the neuron firing, and combine the least significant bit purification method to replicate and generate the binary key stream KS that is the same as the data encryption end.

[0028] S3: Inverse mapping table construction: Based on the binary key stream KS, Fisher-Yates shuffling traversal is used to reverse derive and generate time rearrangement inverse mapping table and space permutation inverse mapping table;

[0029] S4: Reverse decryption: Driven by the binary key stream KS, it completes the reverse timestamp rearrangement and reverse spatial coordinate permutation based on the corresponding time rearrangement reverse mapping table and spatial permutation reverse mapping table, and completes the polarity reversal in combination with the binary key stream KS. All decryption operations depend on the same source binary key stream KS.

[0030] S5: Data Restoration: After decryption, the original DVS sparse binary event stream containing two-dimensional horizontal coordinates, two-dimensional vertical coordinates, time index, and binary polarity is restored.

[0031] Fourthly, the DVS data decryption device of the present invention includes:

[0032] The encrypted receiving module is configured to receive an encrypted event stream E' sent by the data encryption end, wherein the encrypted event stream E' is obtained using the DVS data encryption method as described in this invention.

[0033] The decryption key generation module is configured to obtain the same initial key parameter K as the data encryption end, drive the same one-dimensional chaotic mapping and LIF neuron as the data encryption end, obtain the chaotic pulse interval corresponding to the neuron firing, and combine the least significant bit purification method to replicate and generate the binary key stream KS that is the same as the data encryption end.

[0034] The inverse mapping building module is configured to generate time rearrangement inverse mapping tables and spatial permutation inverse mapping tables based on the binary key stream KS.

[0035] The multidimensional reverse decryption module is configured to be driven by the binary key stream KS, and to complete the time inverse rearrangement and spatial inverse permutation by relying on the corresponding inverse mapping table. It also works with the binary key stream KS to achieve polarity inversion, and finally outputs the complete original DVS event stream to the neuromorphic processor.

[0036] Fifthly, the DVS data encryption system of the present invention includes a DVS event stream acquisition device and a DVS data encryption device as described in the present invention;

[0037] The DVS event stream acquisition module is used to acquire the raw event stream E;

[0038] The DVS data encryption device is used to encrypt the original event stream E and output the encrypted event stream E'.

[0039] Sixthly, the DVS data decryption system of the present invention includes a data receiving device and a DVS data decryption device as described in the present invention;

[0040] The data receiving device is used to receive encrypted event stream E' emitted by the DVS data encryption system as described in this invention;

[0041] The DVS data decryption device is used to decrypt the encrypted event stream E' and output the complete original DVS event stream to the neuromorphic processor.

[0042] Compared with existing technologies, the DVS data encryption method and system based on chaotic pulse interval driving proposed in this invention utilizes chaotic pulse intervals to complete key generation and combines spatial, temporal, and polarity dimensions to collaboratively complete encryption and decryption processing. This effectively solves the practical problems commonly found in existing symmetric block encryption schemes and traditional chaotic permutation schemes when processing asynchronous sparse event streams in DVS, such as high encryption latency, large amount of invalid computation, high edge computing power consumption, and susceptibility to reverse engineering. Specific beneficial effects are as follows:

[0043] (1) Reduce processing latency, minimize invalid computation, and avoid memory overflow issues. Existing block encryption algorithms require accumulating a certain number of events and padding them with zeros before encryption can be completed. This not only disrupts the asynchronous output characteristics of DVS data and generates additional latency, but also involves repeated computations on a large number of invalid zero pixels, resulting in significant waste of overall computational resources. This invention eliminates the fixed buffer window, eliminates the need to pad sparse events into frames, and directly uses the original events as the processing object to complete the encryption operation, fully preserving the asynchronous response capability of DVS data. The overall computational load of this invention is only related to the actual number of valid events generated, with a computational complexity of O(n log n). This avoids invalid calculations caused by redundant pixels; at the same time, it eliminates the need to load all long-term data at once, reducing memory usage and effectively preventing memory overflow issues during long-term data processing. It is suitable for deployment in edge industrial equipment with limited memory and computing resources.

[0044] (2) Reduce data redundancy and lower the storage and transmission pressure on edge devices. This invention is a lossless encryption method oriented towards the event itself. The number of events in the DVS event stream does not change before and after encryption and decryption, and the data volume will not increase due to framing, zero padding, or other operations. Compared with the data expansion problem caused by traditional frame encryption methods, this invention can effectively reduce the data storage cost of edge nodes, alleviate the data transmission pressure of IoT communication links, and adapt to industrial sensing application scenarios with multiple nodes, high-frequency acquisition, and real-time transmission.

[0045] (3) Achieving full-dimensional encryption, addressing the security shortcomings of existing technologies, and enhancing data resistance to attacks. Currently, most mainstream chaotic permutation encryption schemes only scramble spatial coordinates, while time-dimensional data is always transmitted in plaintext. Attackers can rely on the continuity of the time dimension to reconstruct the original motion information, posing a significant security risk. This invention relies on the binary key stream generated by the chaotic pulse interval to simultaneously encrypt the event's spatial coordinates, timestamps, and polarity information, eliminating the plaintext exposure dimension. Actual testing has verified that the absolute Phi correlation coefficient of the encrypted event data is below 0.00016, and the normalized Hamming distance reaches the theoretical optimal value for permutation encryption schemes, fully dispersing the spatiotemporal correlation features within the event. This encryption method can effectively resist image reconstruction attacks and spiking neural network semantic recognition attacks, significantly increasing the difficulty for attackers to reconstruct the original image and recognize perceived content.

[0046] (4) Optimize the key generation mechanism and improve the overall security level of the key. This invention changes the traditional method of directly generating keys based on chaotic numerical sequences, and uses the neuronal pulse interval under chaotic mapping as the original entropy source for key generation. Due to the chaotic characteristics, the relevant control parameters are highly sensitive, and even small deviations in the parameters can lead to significant changes in the key stream, exhibiting a significant avalanche effect. This key generation method can construct a large-capacity key space, with higher key randomness, which can effectively improve the ability of edge-side DVS sensing data to resist brute-force enumeration and key cracking during transmission through public channels. Attached Figure Description

[0047] Figure 1 This is a flowchart of the DVS data encryption method in the embodiments of this application;

[0048] Figure 2 This is a flowchart of the DVS data decryption method in the embodiments of this application;

[0049] Figure 3 This is a block diagram of the DVS data encryption device in the embodiments of this application;

[0050] Figure 4 This is a block diagram of the DVS data decryption device in an embodiment of this application;

[0051] Figure 5 This is a block diagram of the DVS data encryption system in the embodiments of this application;

[0052] Figure 6 This is a block diagram of the DVS data decryption system in an embodiment of this application;

[0053] Figure 7 This is a schematic diagram illustrating the encryption and decryption principles in the embodiments of this application;

[0054] Figure 8 This is a flowchart of key stream production in an embodiment of this application;

[0055] Figure 9 This is a flowchart of the multidimensional encryption process in the embodiments of this application;

[0056] Figure 10 This is a flowchart of the decryption process in an embodiment of this application. Detailed Implementation

[0057] The embodiments of the present invention will be described below with reference to the accompanying drawings and preferred embodiments. Those skilled in the art can understand other advantages and effects of the present invention from the content disclosed in this specification. The present invention can also be implemented or applied through other different specific embodiments, and various details in this specification can also be modified or changed based on different viewpoints and applications without departing from the spirit of the present invention. It should be understood that the preferred embodiments are only for illustrating the present invention and not for limiting the scope of protection of the present invention.

[0058] like Figure 1 and Figure 7 As shown, a DVS data encryption method is applied to the data encryption end, and the method includes the following steps:

[0059] S1: Data Acquisition: Acquire the raw event stream E. The raw event stream E is generated and output by the DVS sensor deployed at the edge, which asynchronously captures pixel-level light intensity changes. The raw event stream E is an asynchronous, sparse binary event stream. , where N e This represents the total number of active events in the data stream. Each event... The horizontal coordinate of the two-dimensional space containing the event Two-dimensional spatial ordinate Time Index And binary polarity representing changes in brightness ,when =1 indicates an increase in brightness, when =-1 indicates a decrease in brightness.

[0060] The DVS sensor captures pixel-level light intensity changes in real time through an event-driven mechanism. Once the contrast threshold is exceeded (adaptively adjusted according to the light intensity environment of the industrial scene), it asynchronously triggers and outputs the raw event stream E. The edge computing node receives the data and temporarily stores it.

[0061] In this application, the discrete light intensity changes transmitted by the DVS sensor are stored as a four-dimensional structure queue arranged in a time series. Each basic data unit consists of parameters of real-occurring active events, defined as follows: Invalid background zero values ​​were removed.

[0062] S2: Keystream Generation: Using the initial key parameter K to drive a one-dimensional chaotic map and LIF neurons, the inter-spike interval (ISI) corresponding to adjacent firing pulses of the LIF neurons is extracted. The Least Significant Bit (LSB) purification process is then performed on the inter-spike interval to generate a binary keystream KS. Here, ISI is the time interval between two adjacent firing pulses of the LIF neuron. LSB is a post-processing method that extracts low-order bits of the signal and improves the randomness of the sequence.

[0063] In one possible embodiment, the one-dimensional chaotic map is a Logistic map used to generate pseudo-random sequences. Its system control parameter r is set to 3.99; the membrane potential decay factor α of the LIF neuron is set to 0.5; and the hardware pulse triggering threshold V... th The value is 2.0. One-dimensional chaotic mapping can also use Tent mapping or Chebyshev mapping.

[0064] S3: Mapping table generation: Based on the binary key stream KS, a time-reordered mapping table is generated using Fisher-Yates shuffling traversal. ) and spatial permutation mapping table ( );

[0065] S4: Multidimensional Encryption: Events are sequentially extracted from the original event stream E. Driven by the binary key stream KS, the timestamps are globally shuffled, and the spatial coordinates are flattened and permuted in one dimension, relying on the time rearrangement mapping table and the spatial permutation mapping table. A key mask is generated by combining the binary key stream KS to complete the event polarity reversal, resulting in the encrypted event stream. .

[0066] S5: Data transmission: Send the encrypted event stream E' outward through the IoT channel to complete the encrypted transmission of the DVS raw event stream.

[0067] like Figure 8 As shown, the key stream generation process is as follows:

[0068] Step 1: Begin;

[0069] Step 2: Parameter and variable initialization

[0070] Input initial key parameters And initialize the variable: chaotic state. membrane potential Pulse counting Time step count Last pulse trigger time .

[0071] Step 3: Loop condition check

[0072] Determine the number of pulses generated Is it less than the total number of events? If no, proceed to step 9; if yes, proceed to step 4.

[0073] Step 4: Increment the time step

[0074] Discrete time step count update: .

[0075] Step 5: Chaotic Mapping Iteration

[0076] Perform a Logistic chaotic mapping to update the current chaotic state: .

[0077] Step 6: LIF neuron membrane potential update

[0078] Perform membrane potential decay and integration calculations: .

[0079] Step 7: Pulse trigger condition determination

[0080] Determine the current membrane potential Is it greater than or equal to the trigger threshold? If no, go back to step 4; if yes, proceed to step 8.

[0081] Step 8: Pulse processing and ISI extraction

[0082] Pulse count update: ;

[0083] Extract the first pulse interval (ISI): ;

[0084] Update the last pulse trigger time: ;

[0085] Reset membrane potential: ;

[0086] Jump back to step 3 and continue the loop.

[0087] Step 9: LSB Bit Purification

[0088] Iterate through all ISI sequences, for each Extract its least significant bit (LSB) to generate a binary key stream. .

[0089] Step 10: Output key stream

[0090] Output the final binary key stream ;

[0091] Step 11: End

[0092] Pseudocode for key stream generation

[0093] Input: Total number of active events Initial key parameters ;

[0094] Output: Binary key stream ;

[0095] initialization: , , , , ;

[0096] 1: while do

[0097] 2:

[0098] 3: / / Perform Logistic mapping

[0099] 4: / / Perform LIF neuron membrane potential update

[0100] 5: if then / / Meets the trigger threshold

[0101] 6:

[0102] 7: / / Extract ISI

[0103] 8:

[0104] 9: / / Reset membrane potential

[0105] 10: end if

[0106] 11: end while 12:

[0108] 13: for j = 1 to Ne do / / LSB purification process

[0109] 14: / / Extract the least significant bit

[0110] 15: end for

[0111] 17: Output binary key stream KS

[0112] like Figure 9 As shown, the multidimensional encryption process is as follows:

[0113] Step 1: Begin;

[0114] Step 2: Input the raw event stream E and the binary key stream KS;

[0115] Step 3: Generate the temporal rearrangement mapping table and the spatial permutation mapping table;

[0116] Step 4: Determine whether all events in the event stream have been traversed. If yes, proceed to step 10; otherwise, proceed to step 5.

[0117] Step 5: Retrieve the next event and proceed to the encryption pipeline;

[0118] Step 6: Based on the spatial permutation mapping table, perform one-dimensional flattening permutation and two-dimensional restoration on the spatial coordinates to obtain the new coordinates.

[0119] Step 7: Remap the time index according to the time rearrangement mapping table;

[0120] Step 8: Extract the key bit corresponding to the current event and determine if it is equal to 1. If yes, proceed to step 9; otherwise, return to step 4 and continue processing the next event.

[0121] Step 9: Invert the polarity of the event, return to step 4, and continue the loop to process the next event;

[0122] Step 10: Output the encrypted event stream;

[0123] Step 11: End.

[0124] Multidimensional encryption pseudocode:

[0125] Input: Raw event stream Binary key stream KS;

[0126] Output: Encrypted event stream ;

[0127] 1: Using the binary key stream KS, a time rearrangement mapping table is generated through Fisher-Yates shuffling.

[0128] 2: for t = 0 to T-1 do

[0129] 3: Using the binary key stream KS, a spatial permutation mapping table is generated through Fisher-Yates shuffling.

[0130] 4:end for 5:

[0132] 6: for each event in E do

[0133] 7: / / Space permutation

[0134] 8:

[0135] 9:

[0136] 10: 11:

[0138] 12: / / Time Reordering

[0139] 13: 14:

[0141] 15: / / Polarity conditional reversal

[0142] 16: if KS[i]==1 then

[0143] 17:

[0144] 18: else

[0145] 19:

[0146] 20: end if

[0147] 21: end for

[0148] 23: Output encrypted event stream

[0149] Explanation of the technical significance of parameters

[0150] 1) K: Initial key parameter set.

[0151] 2) KS: Binary key stream.

[0152] 3)N e The total number of active events in the data stream.

[0153] 4) z0: The initial chaotic state value of the Logistic map.

[0154] 5) r: System control parameters for Logistic mapping.

[0155] 6) α: LIF neuron membrane potential decay factor.

[0156] 7)V th The hardware spiking threshold of LIF neurons.

[0157] 8) n: Global count of discrete time steps of the system.

[0158] 9) u: The current LIF neuron membrane potential accumulation state in the accumulator register in computer memory.

[0159] 10)z: The Logistic chaotic state value at the current discrete time step.

[0160] 11)k: The number of pulses that have been triggered and generated.

[0161] 12)δ k : The kth ISI extracted.

[0162] 13)t k-1 : The absolute time step record when the (k-1)th pulse is triggered.

[0163] 14)j: ISI count during LSB purification process

[0164] 15) KS[j]: The j-th bit in the binary key stream sequence KS

[0165] 16) E: The original set of asynchronous event streams before encryption in the input system, i.e., the original event stream.

[0166] 17) The encrypted event stream is the set of events output by the system.

[0167] 18) Time rearrangement mapping table generated based on binary key stream KS

[0168] 19) : Spatial permutation mapping table generated based on binary key stream KS.

[0169] 20)t: The time step count during the generation of the time rearrangement mapping table.

[0170] 21) i: Global index when pipelined traverses event set.

[0171] 22)e i : The i-th discrete event in the event stream.

[0172] 23)x i : The original two-dimensional horizontal coordinate of the i-th discrete event.

[0173] 24)yi : The original two-dimensional spatial ordinate of the i-th discrete event.

[0174] 25)t i : The original microsecond-level timestamp or discrete-time index of the i-th discrete event.

[0175] 26)p i : The original binary polarity state of the i-th discrete event.

[0176] 27) The new coordinates, new time index, and new polarity state generated after the i-th discrete event is encrypted.

[0177] 28) idx: The addressing index that reduces two-dimensional spatial coordinates to one-dimensional memory addresses.

[0178] 29)T: The number of time frames in the event stream.

[0179] 30)W: Physical space width resolution of DVS.

[0180] 31)H: Physical space height resolution of DVS.

[0181] 32) Asymptotic time complexity is used in algorithm analysis to measure how the computational cost of a system grows with the size of the input data.

[0182] 33) The purely linear computational complexity of the algorithm of this invention.

[0183] 34) The computational complexity of traditional block cipher algorithms.

[0184] In one possible embodiment, intermediate states during the encryption process (such as the membrane potential u of LIF neurons, the generated binary key stream KS, and the time rearrangement map) Spatial permutation mapping table These intermediate states are stored locally in the compute node's volatile main memory as local variables or one-dimensional arrays. After encryption, these intermediate states are cleaned up and not persisted externally. The final encrypted event stream... It enters the underlying network's transmission queue for transmission.

[0185] In one possible embodiment, the data encryption endpoint outputs: an encrypted event stream generated after the edge computing node completes encryption. The data is submitted to the system's underlying network protocol stack, encapsulated into network data packets, and sent through the industrial IoT communication channel.

[0186] like Figure 2 and Figure 7As shown in the embodiments of this application, a DVS data decryption method based on chaotic pulse interval driving is applied to the data decryption end, and the method includes the following steps:

[0187] S1: Data reception: Receive the encrypted event stream E' sent by the data encryption end. The encrypted event stream E' is obtained using the DVS data encryption method as described in the embodiments of this application.

[0188] S2: Key Replication: Obtain the same initial key parameter K as the data encryption end, drive the same one-dimensional chaotic mapping and LIF neuron as the data encryption end, obtain the chaotic pulse interval corresponding to the neuron firing, and combine the least significant bit purification method to replicate and generate the binary key stream KS that is the same as the data encryption end.

[0189] S3: Inverse Mapping Table Construction: Based on the binary key stream KS, Fisher-Yates shuffling traversal is used to reverse derive and generate the time rearrangement inverse mapping table and the space permutation inverse mapping table.

[0190] S4: Reverse decryption: Driven by the binary key stream KS, it completes the reverse timestamp rearrangement and reverse spatial coordinate permutation based on the corresponding time rearrangement reverse mapping table and spatial permutation reverse mapping table, and completes the polarity reversal in combination with the binary key stream KS. All decryption operations depend on the same source binary key stream KS.

[0191] S5: Data Restoration: After decryption, the original DVS sparse binary event stream containing two-dimensional horizontal coordinates, two-dimensional vertical coordinates, time index, and binary polarity is restored.

[0192] Data decryption output: After the receiving node completes the reverse decryption, the restored original event stream E is read by the neuromorphic processor to drive inference tasks such as feature recognition.

[0193] like Figure 10 As shown, the decryption process is as follows:

[0194] Step 1: Begin;

[0195] Step 2: Input the encrypted event stream E' and the initial key parameter K;

[0196] Step 3: Generate a binary key stream KS based on the initial key parameter K;

[0197] Step 4: Derive the inverse time rearrangement mapping table and the inverse space permutation mapping table using the binary key stream KS;

[0198] Step 5: Determine whether all events in the event stream have been traversed. If yes, proceed to step 12; otherwise, proceed to step 6.

[0199] Step 6: Retrieve the next event and proceed to the decryption pipeline;

[0200] Step 7: Extract the key bit corresponding to the current event and determine if it is equal to 1. If not, proceed directly to step 9; if yes, proceed to step 8.

[0201] Step 8: Invert the polarity of the current event;

[0202] Step 9: Restore the encrypted time index to the original time index based on the time-rearranged inverse mapping table;

[0203] Step 10: Calculate and reconstruct the original two-dimensional spatial coordinates based on the spatial permutation inverse mapping table;

[0204] Step 11: Return to step 5 and continue traversing the event stream;

[0205] Step 12: Output the decrypted event stream, i.e., the original DVS sparse binary event stream;

[0206] Step 13: End.

[0207] like Figure 3 As shown in the embodiment of this application, a DVS data encryption device includes a DVS event acquisition module, an encryption key generation module, a mapping table generation module, and a multidimensional encryption module. The DVS event acquisition module is configured to acquire a raw event stream E, which is generated and output by a DVS sensor deployed at the edge after asynchronously capturing pixel-level light intensity changes. The raw event stream E is a sparse binary event stream, where each event contains a two-dimensional spatial horizontal coordinate, a two-dimensional spatial vertical coordinate, a time index, and a binary polarity representing the brightness change. The encryption key generation module is configured to call an initial key parameter K to drive a one-dimensional chaotic mapping and LIF neurons, extract the chaotic pulse intervals corresponding to adjacent firing pulses of the LIF neurons, and perform least significant bit purification processing on the chaotic pulse intervals to generate a binary key stream KS. The mapping table generation module is configured to generate a time rearrangement mapping table and a spatial permutation mapping table based on the binary key stream KS through Fisher-Yates shuffling traversal. The multidimensional encryption module is configured to sequentially extract events from the original event stream E, drive the binary key stream KS as a whole, rely on the time rearrangement mapping table and the spatial permutation mapping table to complete the timestamp scrambling and spatial coordinate permutation, and combine the binary key stream KS to generate a key mask to complete the polarity reversal, thereby obtaining the encrypted event stream E' and sending it outward.

[0208] like Figure 4As shown in the embodiments of this application, a DVS data decryption device includes a ciphertext receiving module, a decryption key generation module, an inverse mapping construction module, and a multidimensional inverse decryption module. The ciphertext receiving module is configured to receive an encrypted event stream E' from a data encryption end, where the encrypted event stream E' is obtained using the DVS data encryption method described in the embodiments of this application. The decryption key generation module is configured to obtain the same initial key parameter K as the data encryption end, drive the same one-dimensional chaotic mapping and LIF neurons as the data encryption end, obtain the chaotic pulse interval corresponding to the neuron firing, and, combined with the least significant bit purification method, replicate and generate a binary key stream KS that is identical to that of the data encryption end. The inverse mapping construction module is configured to generate a time rearrangement inverse mapping table and a spatial permutation inverse mapping table based on the binary key stream KS. The multidimensional inverse decryption module is configured to be driven by the binary key stream KS as a whole, relying on the corresponding inverse mapping table to complete time inverse rearrangement and spatial inverse permutation, and, in conjunction with the binary key stream KS, achieve polarity inversion, ultimately outputting the complete original DVS event stream to the neuromorphic processor.

[0209] like Figure 5 As shown in the embodiments of this application, a DVS data encryption system includes a DVS event stream acquisition device and a DVS data encryption device as described in the embodiments of this application; the DVS event stream acquisition module is used to acquire the original event stream E. The DVS data encryption device is used to encrypt the original event stream E and output the encrypted event stream E'.

[0210] like Figure 6 As shown in the embodiments of this application, a DVS data decryption system includes a data receiving device and a DVS data decryption device as described in the embodiments of this application; the data receiving device is used to send an encrypted event stream E' as described in the DVS data encryption system of this invention; the DVS data decryption device is used to decrypt the encrypted event stream E' and output the complete original DVS event stream to the neuromorphic processor.

[0211] For neuromorphic industrial IoT edge node scenarios, this application constructs an end-to-end lightweight encryption technology architecture that integrates chaotic neuron pulse driving and multi-dimensional encryption mechanisms, upgrading the traditional block cipher architecture. Traditional solutions rely on fixed block buffer padding and multiple rounds of matrix operations to achieve encryption, while this architecture can maintain the asynchronous latency advantage of DVS while ensuring data security with minimal computing power overhead.

[0212] This application designs a native data processing pipeline fully oriented towards sparse asynchronous event streams. The encryption / decryption pipeline directly performs mapping and flipping on the decoupled event attribute structure (spatial coordinates, time index, polarity), ensuring consistency in the number of events before and after encryption. This process reduces the system's computational complexity from that of traditional dense tensors. Dimensional reduction based on the number of active events Purely linear scaling This eliminates the unnecessary computational power consumption on a large number of blank backgrounds.

[0213] This application combines the underlying hardware characteristics of neuromorphic computing with the ISI dynamics of LIF neurons. The algorithm execution path consists of accumulation, threshold comparison, and LSB extraction operations, avoiding the overhead of high-power multiplier (MAC) and full sorting (Sort) operations, while ensuring that the output key stream passes cryptographic standard randomness verification.

[0214] This application resolves the technical contradiction between the characteristics of high-frequency sparse data and traditional encryption architectures. It overcomes the latency surge caused by the forced buffering mechanism and the OOM problem under long-term verification. Through multi-dimensional encryption, it compensates for the system-level vulnerability of existing single-dimensional permutation schemes to deep network semantic reconstruction, and achieves a balance between lightweight edge computing and theoretical security.

[0215] Table 1 is a comparison table of key quality:

[0216]

[0217] Table 1

[0218] For both 8-bit and 1-bit key widths, key quality tests were conducted using three metrics: uniformity, Shannon entropy, and chi-square test. The test results show that compared to the scheme using Logistic mapping alone, this application has better uniformity and Shannon entropy values ​​for both key types, and the chi-square test results are also within a reasonable range. This demonstrates that the key distribution uniformity, information randomness, and statistical properties of this application are significantly improved, resulting in better overall key quality.

[0219] Table 2 is a comparison table of safety assessments:

[0220]

[0221] Table 2

[0222] Using multiple publicly available DVS datasets (N-MNIST, DVS-Gesture, DVS-CIFAR10, Gen1), this application's encryption scheme is compared with three mainstream encryption schemes: block encryption systems, 1D logistic mapping, and streaming permutation systems based on spatial dimension chaotic sorting, through a multi-dimensional security evaluation. From the perspective of DVS encryption metrics, the proposed scheme's Phi correlation coefficient, Hamming distance, and Jaccard similarity values ​​are generally at a low level, far superior to traditional block encryption systems. Furthermore, it exhibits better data obfuscation effects compared to the other two chaotic encryption schemes. Visual reconstruction attack tests on the Gen1 dataset demonstrate that this application... With lower peak signal-to-noise ratio and structural similarity index, and higher pixel change rate and uniform average change intensity, it can effectively resist image reconstruction attacks. In semantic recognition attack experiments, the recognition accuracy of the unencrypted dataset is at a high benchmark level. Various encryption methods can reduce the data recognition accuracy to varying degrees. However, the recognition accuracy of this application is the lowest overall in the three datasets of N-MNIST, DVS-Gesture, and DVS-CIFAR10, with small data fluctuation and strong stability. The comprehensive test indicators can confirm that the encryption scheme of this application has better resistance to visual reconstruction attacks and semantic recognition attacks than existing encryption technologies, and has significant advantages in encryption security and algorithm stability.

[0223] Table 3 shows the key stream parameters and key sensitivity test results.

[0224]

[0225] Table 3

[0226] In this embodiment, small perturbations of different magnitudes are applied to z0, r, α, and Vth, and the keystream parameter sensitivity test is completed based on the Hamming distance between the original ciphertext and the perturbed ciphertext corresponding to the Gen1 and DVS-Gesture datasets. The chaotic parameter is within 10... -10 The order of magnitude, LIF neuron parameters are in the range of 10 -2 At the order of magnitude, the Hamming distance approaches the saturation at the upper limit of the permutation ciphers in the dataset. In summary, the keystream of this encryption scheme exhibits extremely high sensitivity to all key parameters; even minute parameter deviations can generate completely different key sequences. This effectively evades attacks such as brute-force attacks and parameter fine-tuning attacks, further enhancing the overall security redundancy of the encryption system.

[0227] The above embodiments are preferred embodiments of the present invention, but the embodiments of the present invention are not limited to the above embodiments. Any changes, modifications, substitutions, combinations, or simplifications made without departing from the spirit and principle of the present invention shall be considered equivalent substitutions and shall be included within the protection scope of the present invention.

Claims

1. A DVS data encryption method, characterized by, When applied to data encryption, the method includes the following steps: S1: Data Acquisition: Acquire raw event stream E, which is generated and output by the DVS sensor deployed at the edge asynchronously capturing pixel-level light intensity changes. The raw event stream E is a sparse binary event stream, and each event contains a two-dimensional spatial horizontal coordinate, a two-dimensional spatial vertical coordinate, a time index, and a binary polarity representing the brightness change. S2: Key stream generation: Using the initial key parameter K to drive the one-dimensional chaotic mapping and LIF neuron, extract the chaotic pulse interval corresponding to the adjacent firing pulses of the LIF neuron, and perform least significant bit purification processing on the chaotic pulse interval to generate a binary key stream KS. S3: Mapping table generation: Based on the binary key stream KS, Fisher-Yates shuffling traversal is used to generate a time rearrangement mapping table and a spatial permutation mapping table; S4: Multidimensional encryption: sequentially extract the events in the original event stream E, drive the binary key stream KS as the whole, rely on the time rearrangement mapping table and the spatial permutation mapping table to complete the global scrambling of timestamps, one-dimensional flattening and permutation of spatial coordinates, and combine the binary key stream KS to generate a key mask, complete the event polarity reversal, and obtain the encrypted event stream E'. S5: Data transmission: Send the encrypted event stream E' outward through the IoT channel to complete the encrypted transmission of the DVS raw event stream.

2. The DVS data encryption method of claim 1, wherein: In step S2, the one-dimensional chaotic mapping is a Logistic mapping, and its system control parameter r is 3.99; the membrane potential decay factor α of the LIF neuron is 0.5, and the hardware pulse triggering threshold Vth is 2.

0.

3. The DVS data encryption method of claim 1, wherein: In step S2, the one-dimensional chaotic mapping is either a Tent mapping or a Chebyshev mapping.

4. The DVS data encryption method according to claim 1, characterized in that: During the encryption process, the LIF neuron membrane potential, binary key stream KS, time rearrangement mapping table, and spatial permutation mapping table are all stored in local volatile main memory; they are cleaned up after encryption is completed and are not persistently written to the outside.

5. A DVS data encryption device, characterized in that, include: The DVS event acquisition module is configured to acquire the raw event stream E, which is generated and output by the DVS sensor deployed at the edge asynchronously capturing pixel-level light intensity changes. The raw event stream E is a sparse binary event stream, and each event contains a two-dimensional spatial horizontal coordinate, a two-dimensional spatial vertical coordinate, a time index, and a binary polarity representing the brightness change. The encryption key generation module is configured to call the initial key parameter K to drive the one-dimensional chaotic mapping and LIF neuron, extract the chaotic pulse interval corresponding to the adjacent firing pulses of the LIF neuron, and perform least significant bit purification processing on the chaotic pulse interval to generate a binary key stream KS. The mapping table generation module is configured to generate time rearrangement mapping tables and spatial permutation mapping tables based on the binary key stream KS through Fisher-Yates shuffling traversal; The multidimensional encryption module is configured to sequentially extract events from the original event stream E, drive the binary key stream KS as a whole, rely on the time rearrangement mapping table and the spatial permutation mapping table to complete the timestamp scrambling and spatial coordinate permutation, and combine the binary key stream KS to generate a key mask to complete the polarity reversal, thereby obtaining the encrypted event stream E' and sending it outward.

6. The DVS data encryption device according to claim 5, characterized in that: The one-dimensional chaotic mapping in the encryption key generation module is any one of the Logistic mapping, Tent mapping, or Chebyshev mapping, and the system is deployed at the edge node of the neuromorphic industrial Internet of Things.

7. A DVS data decryption method based on chaotic pulse interval driving, characterized in that, When applied to the data decryption end, the method includes the following steps: S1: Data reception: Receive the encrypted event stream E' sent by the data encryption terminal, wherein the encrypted event stream E' is obtained by the DVS data encryption method as described in any one of claims 1 to 4; S2: Key replication: Obtain the same initial key parameter K as the data encryption end, drive the same one-dimensional chaotic mapping and LIF neuron as the data encryption end, obtain the chaotic pulse interval corresponding to the neuron firing, and combine the least significant bit purification method to replicate and generate the binary key stream KS that is the same as the data encryption end. S3: Inverse mapping table construction: Based on the binary key stream KS, Fisher-Yates shuffling traversal is used to reverse derive and generate time rearrangement inverse mapping table and space permutation inverse mapping table; S4: Reverse decryption: Driven by the binary key stream KS, it completes the reverse timestamp rearrangement and reverse spatial coordinate permutation based on the corresponding time rearrangement reverse mapping table and spatial permutation reverse mapping table, and completes the polarity reversal in combination with the binary key stream KS. All decryption operations depend on the same source binary key stream KS. S5: Data Restoration: After decryption, the original DVS sparse binary event stream containing two-dimensional horizontal coordinates, two-dimensional vertical coordinates, time index, and binary polarity is restored.

8. A DVS data decryption device, characterized in that, include: The ciphertext receiving module is configured to receive the encrypted event stream E' sent by the data encryption end, wherein the encrypted event stream E' is obtained by the DVS data encryption method as described in any one of claims 1 to 4; The decryption key generation module is configured to obtain the same initial key parameter K as the data encryption end, drive the same one-dimensional chaotic mapping and LIF neuron as the data encryption end, obtain the chaotic pulse interval corresponding to the neuron firing, and combine the least significant bit purification method to replicate and generate the binary key stream KS that is the same as the data encryption end. The inverse mapping building module is configured to generate time rearrangement inverse mapping tables and spatial permutation inverse mapping tables based on the binary key stream KS. The multidimensional reverse decryption module is configured to be driven by the binary key stream KS, and to complete the time inverse rearrangement and spatial inverse permutation by relying on the corresponding inverse mapping table. It also works with the binary key stream KS to achieve polarity inversion, and finally outputs the complete original DVS event stream to the neuromorphic processor.

9. A DVS data encryption system, characterized in that, Includes a DVS event stream acquisition device and a DVS data encryption device as described in claim 5 or 6; The DVS event stream acquisition module is used to acquire the raw event stream E; The DVS data encryption device is used to encrypt the original event stream E and output the encrypted event stream E'.

10. A DVS data decryption system, characterized in that, Includes a data receiving device and a DVS data decryption device as described in claim 8; The data receiving device is used to receive the encrypted event stream E' emitted by the DVS data encryption system as described in claim 9; The DVS data decryption device is used to decrypt the encrypted event stream E' and output the complete original DVS event stream to the neuromorphic processor.

Images