A method of underwater backscattering communication based on visible light

Abstract

The application discloses a kind of underwater backscattering communication methods based on visible light, comprising: S1: design screen information transmission mode;S2: design event data filter;S3: construct active flickering event surface and evidence map;S4: carry out adaptive detection and positioning of region of interest;S5: carry out spatial domain signal downsampling and signal-to-noise ratio enhancement;S6: to the event data stream that has passed downsampling and signal enhancement is based on time domain characteristics and spatial density characteristics Frame clustering is carried out;S7: carry out homography perspective transformation and geometric normalization;S8: carry out gridding space demodulation and data recovery.The application can realize higher rate data transmission with the aid of optical signal, and overcome the finiteness of traditional underwater optical communication technology in underwater low-light environment and the problem of frequency mismatch between transmitting and receiving parties, and the data transmission stability is higher.

Description

Technical Field

[0001] This invention relates to the field of communications, specifically to an underwater backscattering communication method based on visible light. Background Technology

[0002] As the Internet of Things (IoT) extends into the marine environment, building long-term, low-power underwater sensor networks has become crucial for the development of underwater IoT (IoUT). For nodes requiring long-term deployment, energy constraints are the primary bottleneck, making backscatter communication an ideal solution due to its "zero-power" characteristics. However, radio frequency signals face extremely high attenuation in water, and their effective transmission distance decreases sharply with increasing frequency, making it difficult to directly adapt mature terrestrial radio frequency backscatter technology to the underwater environment.

[0003] To overcome underwater transmission losses, researchers first turned to acoustic media. Jang et al.'s Piezo-Acoustic Backscatter (PAB) utilized the impedance matching characteristics of piezoelectric materials to achieve passive reflection modulation of underwater acoustic signals for the first time. The latest EchoRider system further solved the communication challenges in mobile scenarios by introducing Doppler equalization technology. Although acoustic solutions have significant advantages in communication distance, they are limited by low bandwidth (only in the Kbps range) and high propagation delay (approximately 1500 m / s), and have consistently failed to meet the real-time requirements of video streaming or large-scale data acquisition.

[0004] Given the data rate bottleneck in acoustics, visible light backscatter communication (VLBC), operating within the low-attenuation "window" (450-570nm) underwater, has emerged. PassiveVLC, utilizing a liquid crystal (LCD) shutter and retroreflector, has verified the feasibility of passive optical communication; RetroTurbo, proposed by Wu et al., employs polarization modulation technology, effectively overcoming the physical limitations of LCD refresh rates. However, existing VLBC systems are mostly based on traditional photodiodes or frame cameras as receivers. Traditional frame cameras, limited by fixed exposure times and frame rates, often face problems such as insufficient dynamic range (prone to overexposure or underexposure) and excessive redundant data processing when encountering complex underwater lighting interference and dynamic turbulence, leading to highly unstable communication links.

[0005] Event cameras, as a type of bio-inspired visual sensor, generate asynchronous event streams only in response to changes in light intensity. They possess microsecond-level temporal resolution and extremely high dynamic range (greater than 120dB), naturally suited to the high-frequency pulse characteristics of optical communication. The extremely high dynamic range ensures that event cameras can capture even the weakest light signals, making them a promising candidate for applications in optical communication technology. For example, the Selene system successfully achieved Mbps-level ultra-high speeds in air by using high-frequency flipped digital micromirror devices (DMDs) in conjunction with event cameras. However, current research on event camera-based optical communication mainly focuses on air environments, lacking systematic research on visible light backscattering in complex underwater channels (scattering, attenuation, turbulence). Existing underwater optical communication solutions either have excessively high power consumption (active light sources) or limited receiver processing capabilities (traditional cameras), making it difficult to simultaneously meet the comprehensive requirements of low power consumption, high dynamic range, and high speed for underwater nodes. Therefore, researching underwater backscattering systems based on visible light, especially by introducing event cameras as receivers, is an effective way to solve these problems. Summary of the Invention

[0006] To overcome the technical shortcomings of existing underwater communication methods, such as low data transmission rate and slow signal propagation speed in acoustic communication, and low signal-to-noise ratio and high energy consumption in low-light underwater environments, this invention proposes an underwater backscattering communication method based on visible light.

[0007] To achieve the above objectives, the present invention provides the following technical solution:

[0008] An underwater backscattering communication method based on visible light includes the following steps:

[0009] S1: Design the screen information transmission method;

[0010] S2: Design an event data filter;

[0011] S3: Construct a surface and evidence map of active flickering events;

[0012] S4: Perform adaptive detection and localization of the region of interest;

[0013] S5: Perform spatial domain signal downsampling and signal-to-noise ratio enhancement;

[0014] S6: Perform frame-based clustering on the downsampled and signal-enhanced event data stream based on temporal features and spatial density characteristics;

[0015] S7: Perform homography perspective transformation and geometric normalization;

[0016] S8: Perform gridded spatial demodulation and data recovery.

[0017] Further, S1 includes:

[0018] Each character in the information is encoded as a byte of binary data, with each bit having only two states: 0 and 1. The screen is divided into several grids of the same size, with each grid having only two states: a first color and a second color, corresponding to 0 and 1. The brightness difference between the first color and the second color is obvious. The screen displays a two-color grid image frame, and the data image frame containing N grids contains N / 8 bytes of binary data.

[0019] Each pixel of the event camera only responds to changes in brightness. An event is triggered when the logarithmic change in light intensity of a pixel exceeds a threshold. Full first color frames are interspersed between data frames. Each pixel is fully flipped to generate a sufficient number of events when a new data frame is displayed. A fixed number of two-color frames are alternately transmitted before transmitting data frames, followed by a fixed preamble Barker frame containing a large number of identical sequences.

[0020] Further, S2 includes:

[0021] A refractory period filter is used to set a refractory period at each pixel, discarding events generated by the refractory period device after a trigger event until the refractory period ends; let the event stream before filtering at a certain pixel be... The k-th event e k Includes position coordinates (x) k ,y k ) and timestamp t k Let τ be the refractory period length, for e k The rule for determining the refractory period is as follows:

[0022]

[0023] Where, f(e) k The ) represents the result of the refractory period filtering, where 1 indicates retention and 0 indicates rejection. k-1 (x,y) is the timestamp of the last event that pixel (x,y) was preserved, and the update rule is:

[0024]

[0025] Spatiotemporal domain filtering is used to remove random noise.

[0026] Further, S3 includes:

[0027] A blinking event is defined as a triplet of consecutive polarity flip events within the same pixel, specifically a series of positive-negative-positive or negative-positive-negative events.

[0028]

[0029] BlinkingEvents represents blinking events, where t and p are the timestamps and polarities of the blinking events; span is the difference between the timestamps of the first and last events.

[0030] The active flicker event surface is defined as a dynamic matrix with the same size as the camera resolution, and the value at each position is the timestamp of the previous flicker event at that position coordinate.

[0031] Furthermore, S3 also includes:

[0032] An evidence map is constructed based on the active flicker event surface, with the same size as the active flicker event surface. When the active flicker event surface at each location coordinate is updated, a certain value is added at the same location in the evidence map, and an exponential time decay is applied to the evidence map. The specific update principle is as follows:

[0033]

[0034] Among them, EM t (x, y) represents the value of the evidence graph at coordinate (x, y) at time t. The old value decays over time, and e is the natural constant; Let I be the base evidence value for triggering any flickering event, where I is a switch variable (1 for a flickering event, 0 otherwise), and G(r) is the spatial compensation gain. The reward evidence value is the amount of time required for a flashing event to trigger.

[0035] Further, S4 includes:

[0036] A distortion correction field is constructed using a pre-calibrated intrinsic parameter matrix and distortion coefficients to correct the nonlinear geometric distortion introduced by the camera lens; inverse distortion mapping is performed on the original event flow coordinates to correct them to ideal Euclidean geometric space; and percentile-based normalization is performed on the original evidence map EM(x,y) to obtain the normalized evidence map EM. norm (x,y);

[0037] Gaussian smoothing filtering is applied to the enhanced evidence map to obtain the EM. smooth The optimal segmentation threshold T is adaptively calculated using the Otsu method. opt Generate the target binarization mask B(x,y):

[0038] A geometric reconstruction algorithm based on edge linear regression inference is used to solve the problem of screen corner occlusion or incomplete outline;

[0039] To address the jitter problem in event camera detection, a finite state machine (FSM) tracking model is constructed. This FSM introduces the intersection-over-union (IoU) ratio and centroid drift vector as observation metrics for spatiotemporal consistency. The region of interest (ROI) detection algorithm enters a "locked state" only when multiple consecutive frames satisfy the stability criterion. A temporal moving average filter is then applied to smooth the ROI parameters, ultimately yielding the point set P containing the coordinates of the four vertices of the sub-pixel-level ROI. src .

[0040] Further, S5 includes:

[0041] Construct a polygonal binary mask M(x,y) based on the vertex coordinates of the region of interest, and traverse the original event stream S. raw Perform spatial logic and operations, retaining only the valid event set S inside the mask. ROI ;

[0042] To address the modulation characteristics of on / off keying modulation, polarity-selective filtering is implemented; a time-domain circular buffer queue is constructed to reassemble the cleaned event stream in an orderly manner according to timestamps, and the effective event density is monitored in real time.

[0043] Further, S6 includes:

[0044] Construct a time-domain event density histogram H(t), and select the low quantile P. 20 As a basis estimate of background noise N floor Based on this, an adaptive segmentation threshold is set. This achieves dynamic boundary delimitation of the signal interval, where α is the dynamic scaling factor set by the spatiotemporal density clustering algorithm; the density peak clustering algorithm is executed to identify those that satisfy... High-density active regions; introduce a temporal neighborhood merging mechanism for intervals Δt gap Adjacent intervals smaller than the set tolerance are integrated for connectivity; based on the merged time interval [t] start ,t end Perform indexed slicing on the original event stream, t start t end Let C be the start and end points of the merged time interval, and extract the set of event clusters C that independently correspond to a single frame image. k This completes the conversion from discrete event streams to discrete digital frames.

[0045] Further, S7 includes:

[0046] The point set P of the sub-pixel level region of interest coordinates based on S4 solution. src Construct the reference vertex P in the target orthogonal projection space. dst The two-dimensional homography matrix H is solved using the direct linear transformation algorithm, and the projective mapping relationship from the source plane to the target plane is established.

[0047] Using this projective mapping relationship, for each event cluster set C k All discrete event coordinates (x, y) within the system are subjected to homography transformation, and the original discrete event coordinates (x, y) are remapped to the corrected coordinate system (x', y'). The transformed sparse event point cloud is resampled and rasterized using bilinear interpolation or nearest neighbor interpolation strategies to generate a standardized event distribution map with a uniform scale and orthogonal grid structure.

[0048] Further, S8 includes:

[0049] Based on the standardized event distribution map generated by S7, a topology consistent with the LCD screen layout at the transmitter is constructed. Spatial demodulation mesh, traversing mesh cells (r, c), where r and c represent the r-th row and c-th column of the mesh array respectively, and calculating the effective event density D within each mesh cell. r,c And set a decision threshold. Area cell This represents the total number of pixels in each grid cell, where β is the duty cycle coefficient. Binarization and quantization are performed when... When the grid is set to a value of logic '1', it is determined to be logic '0' otherwise, thus generating a transient binary data matrix M. k The sliding window algorithm is used to scan the continuous binary sequence, and the frame start boundary is established by matching the autocorrelation peak characteristics of the preset Barker code sequence; then the payload data is reconstructed, and finally the recovered digital information stream is output.

[0050] Compared with the prior art, the beneficial effects of the present invention are: the present invention can achieve higher data transmission rate by means of optical signals, and overcomes the limitations of traditional underwater optical communication technology in low-light underwater environment and the problem of frequency mismatch between the transmitting and receiving parties, resulting in higher data transmission stability. Attached Figure Description

[0051] Figure 1 This is a flowchart of the present invention.

[0052] Figure 2 This is a diagram illustrating the encoding of information into binary data.

[0053] Figure 3 A diagram illustrating how information is delivered during screen refresh.

[0054] Figure 4 This is a schematic diagram of a screen area detection method.

[0055] Figure 5 This is a schematic diagram of a screen area correction method.

[0056] Figure 6A schematic diagram illustrating the method of restoring event data into information.

[0057] Figure 7 This is a schematic diagram for verifying the accuracy of the invention in real-world experiments. Detailed Implementation

[0058] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. Based on the embodiments of the present invention, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of the present invention.

[0059] Please see Figure 1 A visible light-based underwater backscattering communication method includes the following steps:

[0060] S1: Design the screen information transmission method, including:

[0061] Each character in the information can be encoded as one byte (8 bits) of binary data, with each bit having only two states: 0 and 1. The screen is divided into several equally sized grids, each grid having only two states: black and white (or two other colors with a significant difference in brightness, such as red and green), corresponding to 0 and 1, as shown below. Figure 2 As shown; the screen displays a black and white grid image frame, and the data image frame containing N grids contains N / 8 bytes of binary data.

[0062] In an event camera, each pixel responds only to changes in brightness, not the absolute value of brightness. An event is triggered only when the logarithmic change in light intensity of a pixel exceeds a threshold, i.e., when a condition is met. In the formula, I(t) is the light intensity, ΔlnI(t) is the light intensity difference, t is the timestamp, Δt is the time difference, and C is a threshold related to the event camera sensitivity. This invention intersperses completely black frames between data frames, ensuring each pixel fully flips to generate sufficient events when a new data frame is displayed. A fixed number of black and white frames are alternately transmitted before the data frame is transmitted, followed by a fixed preamble Barker frame containing a large number of identical sequences. The screen refresh timing is as follows: Figure 3 As shown.

[0063] S2: Design event data filters, including:

[0064] A single strong brightness change can trigger a series of events, resulting in event redundancy. A refractory period filter is used to set a refractory period at each pixel, discarding events generated by the refractory period device after the triggered event, until the refractory period ends. Let the event stream before filtering at a certain pixel be... The k-th event ek Includes position coordinates (x) k ,y k ) and timestamp t k Let τ be the refractory period length, for e k The rule for determining the refractory period is as follows:

[0065]

[0066] Where, f(e) k The ) represents the result of the refractory period filtering, where 1 indicates retention and 0 indicates rejection. k-1 (x,y) is the timestamp of the last event that pixel (x,y) was preserved, and the update rule is:

[0067]

[0068] The characteristics of the event camera circuitry and the physical properties of the lens itself determine that the sensitivity of the event camera decreases from the center of the field of view to the edges, and random noise is easily generated at the edges. Spatiotemporal domain filtering is used to filter out random noise, and the rules are as follows:

[0069]

[0070] Where l(·) is the indicator function, 1 if the condition is met, and 0 if not, min represents the minimum function, and u k It is the event location coordinate vector (x k ,y k ), that is, event e k =(u k ,t k ), N(u k ) for u k The 8-spatial neighborhood set, T(u) is the last trigger time of pixel u, and δ is the time threshold parameter of the spatiotemporal neighborhood filter.

[0071] S3: Construct the active flickering event surface and evidence map, including:

[0072] A blinking event is defined as a triplet of consecutive polarity flip events within the same pixel, specifically a series of positive-negative-positive or negative-positive-negative events.

[0073]

[0074] BlinkingEvents represents blinking events, where t and p are the timestamps and polarities of the blinking events, respectively, and the polarity is the same as the last event. span is the difference between the timestamps of the first and last events, i.e., the blinking event period. A fixed-frequency screen refresh causes periodic brightness changes, resulting in periodically stable blinking events. Events from ambient light sources and moving objects are less likely to constitute blinking events, and these events rarely share the same period as those generated by screen refresh. The Active Blinking Event Surface (SABE) is defined as a dynamic matrix with the same size as the camera resolution. The value at each position of the matrix is ​​the timestamp of the previous blinking event at that coordinate. Specifically:

[0075]

[0076] SABE(x,y) is the timestamp of the latest blinking event at position (x,y), and SABE(x,y) is the value of the active blinking event surface at position (x,y).

[0077] An evidence map is constructed based on the active flicker event surface, with the same size as the active flicker event surface. When the active flicker event surface at each cell is updated, a certain value is added at the same position in the evidence map. This value is related to the pixel position. An exponential time decay is applied to the evidence map. The specific update principle is as follows:

[0078]

[0079] Among them, EM t (x, y) represents the value of the evidence graph at coordinates (x, y) at time t. The first term... The old value decays over time, and e is the natural constant; the second term This is the base evidence value triggered by any flashing event, where i is a switch variable: 1 for a flashing event and 0 otherwise. For spatial compensation gain, r is the distance of the pixel position from the center. The larger r is, the stronger the gain. max The maximum distance from the edge of the camera's field of view to the center of the field of view, where k is a variable gain parameter; the third term △EM is the reward evidence value for penalties imposed on flickering events that meet the periodic requirements. Blink As an intermediate quantity, W base Based on the score, F dur F is the periodicity factor, used to calculate the difference between the flickering event period and the target period. The smaller the difference, the higher the value. stab As a stability factor, the difference between this flickering event and the previous period is calculated. The smaller the difference, the more stable the event, and the higher the value.

[0080] It's important to note that pixels are independent of each other. Each pixel generates an event due to changes in brightness. This event has location coordinates (x, y), which are the same as the coordinates of the pixel that generated the event in the field of view. For example, the event generated by a pixel at position (3, 1) will also have coordinates (3, 1), only the timestamp and polarity will be different. Event data includes three attributes: position, timestamp, and memory, namely (x, y, t, p).

[0081] S4: Perform adaptive detection and localization of the region of interest (ROI), including:

[0082] A distortion correction field is constructed using a pre-calibrated intrinsic parameter matrix and distortion coefficients to correct the nonlinear geometric distortion introduced by the camera lens. Inverse distortion mapping is performed on the original event flow coordinates to correct them to an ideal Euclidean geometric space, ensuring the linearity of subsequent geometric features. Based on this, an adaptive signal enhancement method based on statistical features is proposed to address the problem of large signal-to-noise ratio fluctuations in optical communication scenarios. Unlike fixed threshold processing, the adaptive signal enhancement method uses the gray-level probability distribution of the statistical evidence map within the current time window, utilizing the tail feature of its distribution—the high quantile P. 95 (Representing the effective signal peak value) and low quantile P5 (representing the background noise), construct a dynamic normalized mapping function:

[0083]

[0084] EM(x,y) is the value of the original evidence map at (x,y). norm (x,y) represents the value of the evidence map at (x,y) after normalization.

[0085] This mapping stretches the weak effective signal to a high dynamic range, significantly suppressing background noise. Subsequently, a Gaussian smoothing filter is applied to the enhanced evidence map to improve spatial continuity, resulting in the EM... norm The result EM after Gaussian smoothing filtering of (x,y) smooth The optimal segmentation threshold T is adaptively calculated using Otsu's method. opt Generate the target binarization mask B(x,y):

[0086]

[0087] Secondly, a geometric reconstruction algorithm based on edge linear regression inference is proposed to solve the problem of screen corner occlusion or contour defects. This algorithm does not directly rely on the vulnerable corner features, but utilizes the linear constraint characteristics of the screen's physical edges in the correction space. First, the convex hull of the target connected component is extracted, and edge point sets belonging to the four borders of the screen are identified and clustered. For each set of edge points, the least squares method is used to fit a linear equation. Then, by calculating the intersection of two adjacent fitted lines, the coordinates of the occluded or damaged virtual corner points are calculated, thereby achieving sub-pixel level accurate restoration.

[0088] Finally, to address the jitter problem in event camera detection, a finite state machine (FSM) tracking model is constructed. This FSM tracking model introduces the intersection-over-union (IoU) ratio and the centroid drift vector as observation metrics for spatiotemporal consistency; the stability criterion is defined as: the current region R t With historical locked area R t-1 The centroid Euclidean distance offset Δd must satisfy the convergence constraint:

[0089]

[0090] Among them, C t and C t-1 These are the centroid vectors of the region of interest at the current and previous time steps, respectively. For vector two-dimensional norm, As a stability threshold, the region of interest (ROI) detection algorithm enters a "locked state" only when multiple consecutive frames satisfy the stability criteria (intersection over union ratio higher than a certain threshold, and centroid drift vector 2D norm less than a certain threshold). A temporal moving average filter is then applied to smooth the ROI parameters, ensuring spatial alignment accuracy. Finally, the position coordinates of the ROI are obtained, specifically a point set P containing the coordinates of the four vertices of the sub-pixel ROI. src .

[0091] The process of generating, normalizing, and detecting regions of interest in the evidence map is as follows: Figure 4 As shown.

[0092] S5: Perform spatial domain signal downsampling and signal-to-noise ratio enhancement, including:

[0093] Construct a polygonal binary mask M(x,y) based on the vertex coordinates of the region of interest, and traverse the original event stream S. raw Perform spatial logic and operations, retaining only the valid set of events within the mask, specifically:

[0094]

[0095] In the formula, S ROIM(x) is the event stream after the original event stream has been filtered through a region of interest mask. k ,y k ) is the mask for the region of interest in (x k ,y k The value at the given location.

[0096] To address the modulation characteristics of on-off keying, polarity-selective filtering is implemented. Based on the sensor response characteristics, a subset of positive polarity (ON) events with high signal-to-noise ratio is selected, or a positive-negative polarity differential verification mechanism is used to eliminate isolated noise points. Finally, a time-domain circular buffer queue is constructed to reassemble the cleaned event stream in an orderly manner according to timestamps, and the effective event density is monitored in real time to provide high-quality data input for subsequent time-domain frame clustering.

[0097] S6: Perform frame-based clustering on the downsampled and signal-enhanced event data stream based on temporal features and spatial density characteristics, including:

[0098] Based on the S5 filter-based downsampled event stream, a time-domain event density histogram H(t) is constructed to characterize the instantaneous changes in signal intensity, taking into account the non-uniform time-sparse characteristics of the event stream. Utilizing the statistical distribution characteristics of the amplitude of the time-domain event density histogram, a low quantile P is selected. 20 As a basis estimate of background noise N floor Based on this, an adaptive segmentation threshold is set. This achieves dynamic boundary delimitation of the signal interval, where α is the dynamic scaling factor set by the spatiotemporal density clustering algorithm; based on this, the density peak clustering algorithm is executed to identify those that satisfy... The high-density active region; to address brief periods of silence or random disturbances during transmission, a time-domain neighborhood merging mechanism is introduced for intervals Δt. gap Adjacent intervals smaller than the set tolerance are integrated for connectivity to ensure the topological integrity of the frame structure; finally, based on the merged time interval [t], start ,t end Perform indexed slicing on the original event stream, t start t end Let C be the start and end points of the merged time interval, and extract the set of event clusters C that independently correspond to a single frame image. k This completes the conversion from discrete event streams to discrete digital frames. The process of clustering and segmenting the event stream and restoring it to digital frames is as follows: Figure 5 As shown.

[0099] S7: Perform homography perspective transformation and geometric normalization, including:

[0100] The point set P with sub-pixel level ROI coordinates based on S4 solution. srcConstruct the reference vertex P in the target orthogonal projection space. dst The two-dimensional homography matrix H is solved using the Direct Linear Transformation (DLT) algorithm to establish the projective mapping relationship from the source plane to the target plane, specifically as follows:

[0101]

[0102] Where H is a 3x3 matrix, and the elements at each position are... Let (x, y) be the value of the i-th row and j-th column of the matrix, and (x, y) and (x', y') be the position coordinates of the event in the event stream before and after the transformation, respectively.

[0103] Using this projective mapping relationship, for each event cluster set C k All original discrete event coordinates (x, y) within the system are subjected to homography transformation, remapping the original discrete event coordinates (x, y) to the corrected coordinate system (x', y'), thereby eliminating trapezoidal distortion and rotation components introduced by the shooting perspective deviation. The effect of homography perspective transformation is as follows: Figure 6 As shown, further, bilinear interpolation or nearest neighbor interpolation strategies are used to resample and rasterize the transformed sparse event point cloud to generate a standardized event distribution map with a uniform scale and orthogonal grid structure, providing a geometrically consistent data benchmark for subsequent spatial demodulation.

[0104] S8: Performs gridded spatial demodulation and data recovery, including:

[0105] Based on the standardized event distribution map generated by S7, a topology consistent with the LCD screen layout at the transmitter is constructed. Spatial demodulation mesh, traversing mesh cells (r, c), where r and c represent the r-th row and c-th column of the mesh array respectively, and calculating the effective event density D within each mesh cell. r,c And set a decision threshold. , This represents the total number of pixels in each grid cell, where β is the duty cycle coefficient. Binarization and quantization are performed when... When the grid is set to a value of logic '1', it is determined to be logic '0' otherwise, thus generating a transient binary data matrix M. k Based on this, considering the characteristics of asynchronous data stream transmission, a sliding window algorithm is used to scan continuous binary sequences. The frame start boundary is established by matching the autocorrelation peak characteristics of a preset Barker code sequence. Then, redundant parity bits are removed according to the frame structure defined by the communication protocol, the payload data is reassembled, and finally, the recovered digital information stream is output, such as... Figure 7 As shown, the restored information stream is highly consistent with the real information, and the complete accuracy of information reading can be guaranteed through a specific forward error correction coding mechanism.

[0106] Example: This invention is based on a real-world experiment using the proposed underwater backscatter communication method. The screen is divided into 1000 grids of 40x25 pixels, with a screen refresh rate of 12Hz-15Hz; the preamble Barker frame contains 200 11110 sequences; the refractory period is set to 1ms, and the spatiotemporal neighborhood of the spatiotemporal neighborhood filter is 3x3, 1ms; the percentile P is selected. 95 Evidence graph normalization was performed on P5; the cross-union threshold was set to 0.95.

[0107] A DVXplorer event camera was selected as the receiver, and a 4.2-inch 16Hz full-emission LCD screen was used as the transmitter, driven by an STM32L432KCU6 microcontroller. The transmitter and receiver were placed in water at a distance of 50cm and illuminated in a dark environment using a blue light source to complete data transmission. A total of 15 data frames and one Barker preamble frame were transmitted in the experiment.

[0108] The collected data was fed into the adaptive detection and localization algorithm for regions of interest and the temporal frame-segmentation clustering algorithm based on density estimation proposed in this invention. Ultimately, 16 clusters and the recovered image frames were successfully identified. The real image frames and the recovered image frames were compared grid by grid. The overall accuracy of the recovery remained above 95%, with the lowest accuracy rate at 94.20% and the highest accuracy rate at 97.20%.

[0109] Although embodiments of the invention have been shown and described, it will be understood by those skilled in the art that various changes, modifications, substitutions and alterations can be made to these embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A method for underwater backscattering communication based on visible light, characterized in that, Includes the following steps: S1: Design the screen information transmission method; S2: Design an event data filter; S3: Construct the active flickering event surface and evidence map, including: A blinking event is defined as a triplet of consecutive polarity flip events within the same pixel, specifically a series of positive-negative-positive or negative-positive-negative events. ， BlinkingEvents represents blinking events, where t and p are the timestamps and polarities of the blinking events; span is the difference between the timestamps of the first and last events. The active flickering event surface is defined as a dynamic matrix with the same size as the camera resolution, and the value at each position of the matrix is ​​the timestamp value of the previous flickering event at that position coordinate. An evidence map is constructed based on the active flicker event surface, with the same size as the active flicker event surface. When the active flicker event surface at each location coordinate is updated, a certain value is added at the same location in the evidence map, and an exponential time decay is applied to the evidence map. The specific update principle is as follows: ， Among them, EM t (x, y) represents the value of the evidence graph at coordinate (x, y) at time t. The old value decays over time, and e is the natural constant; Let I be the base evidence value for triggering any flickering event, where I is a switch variable (1 for a flickering event, 0 otherwise), and G(r) is the spatial compensation gain. The reward evidence value is the value triggered by the flashing event that meets the periodic requirements. S4: Perform adaptive detection and localization of the region of interest; S5: Perform spatial domain signal downsampling and signal-to-noise ratio enhancement; S6: Perform frame-based clustering on the downsampled and signal-enhanced event data stream based on temporal features and spatial density characteristics; S7: Perform homography perspective transformation and geometric normalization; S8: Perform gridded spatial demodulation and data recovery.

2. The underwater backscattering communication method based on visible light according to claim 1, characterized in that, S1 includes: Each character in the information is encoded as a byte of binary data, with each bit having only two states: 0 and 1. The screen is divided into several grids of the same size, with each grid having only two states: a first color and a second color, corresponding to 0 and 1. The brightness difference between the first color and the second color is obvious. The screen displays a two-color grid image frame, and the data image frame containing N grids contains N / 8 bytes of binary data. Each pixel of the event camera only responds to changes in brightness. An event is triggered when the logarithmic change in light intensity of a pixel exceeds a threshold. Full first color frames are interspersed between data frames. Each pixel is fully flipped to generate a sufficient number of events when a new data frame is displayed. A fixed number of two-color frames are alternately transmitted before transmitting data frames, followed by a fixed preamble Barker frame containing a large number of identical sequences.

3. The underwater backscattering communication method based on visible light according to claim 1, characterized in that, S2 includes: A refractory period filter is used to set a refractory period at each pixel, discarding events generated by the refractory period device after a trigger event until the refractory period ends; let the event stream before filtering at a certain pixel be... The k-th event e k Includes position coordinates (x) k ,y k ) and timestamp t k Let τ be the refractory period length, for e k The rule for determining the refractory period is as follows: ， Where, f(e) k The ) represents the result of the refractory period filtering, where 1 indicates retention and 0 indicates rejection. k-1 (x,y) is the timestamp of the last event that pixel (x,y) was preserved, and the update rule is: ， Spatiotemporal domain filtering is used to remove random noise.

4. The underwater backscattering communication method based on visible light according to claim 1, characterized in that, S4 includes: A distortion correction field is constructed using a pre-calibrated intrinsic parameter matrix and distortion coefficients to correct the nonlinear geometric distortion introduced by the camera lens; inverse distortion mapping is performed on the original event flow coordinates to correct them to ideal Euclidean geometric space; and percentile-based normalization is performed on the original evidence map EM(x,y) to obtain the normalized evidence map EM. norm (x,y); Gaussian smoothing filtering is applied to the enhanced evidence map to obtain the EM. smooth The optimal segmentation threshold T is adaptively calculated using the Otsu method. opt Generate the target binarization mask B(x,y): A geometric reconstruction algorithm based on edge linear regression inference is used to solve the problem of screen corner occlusion or incomplete outline; To address the jitter problem in event camera detection, a finite state machine (FSM) tracking model is constructed. This FSM introduces the intersection-over-union (IoU) ratio and centroid drift vector as observation metrics for spatiotemporal consistency. The region of interest (ROI) detection algorithm enters a "locked state" only when multiple consecutive frames satisfy the stability criterion. A temporal moving average filter is then applied to smooth the ROI parameters, ultimately yielding the point set P containing the coordinates of the four vertices of the sub-pixel-level ROI. src .

5. The underwater backscattering communication method based on visible light according to claim 1, characterized in that, S5 includes: Construct a polygonal binary mask M(x,y) based on the vertex coordinates of the region of interest, and traverse the original event stream S. raw Perform spatial logic and operations, retaining only the valid event set S inside the mask. ROI ; To address the modulation characteristics of on / off keying modulation, polarity-selective filtering is implemented; a time-domain circular buffer queue is constructed to reassemble the cleaned event stream in an orderly manner according to timestamps, and the effective event density is monitored in real time.

6. The underwater backscattering communication method based on visible light according to claim 1, characterized in that, S6 includes: Construct a time-domain event density histogram H(t), and select the low quantile P. 20 As a basis estimate of background noise N floor Based on this, an adaptive segmentation threshold is set. This achieves dynamic boundary delimitation of the signal interval, where α is the dynamic scaling factor set by the spatiotemporal density clustering algorithm; the density peak clustering algorithm is executed to identify those that satisfy... High-density active regions; introduce a temporal neighborhood merging mechanism for intervals Δt gap Adjacent intervals smaller than the set tolerance are integrated for connectivity; based on the merged time interval [t] start ,t end Perform indexed slicing on the original event stream, t start t end Let C be the start and end points of the merged time interval, and extract the set of event clusters C that independently correspond to a single frame image. k This completes the conversion from discrete event streams to discrete digital frames.

7. The underwater backscattering communication method based on visible light according to claim 4, characterized in that, S7 includes: The point set P of the sub-pixel level region of interest coordinates based on S4 solution. src Construct the reference vertex P in the target orthogonal projection space. dst The two-dimensional homography matrix H is solved using the direct linear transformation algorithm, and the projective mapping relationship from the source plane to the target plane is established. Using this projective mapping relationship, for each event cluster set C k All discrete event coordinates (x, y) within the system are subjected to homography transformation, and the original discrete event coordinates (x, y) are remapped to the corrected coordinate system (x', y'). The transformed sparse event point cloud is resampled and rasterized using bilinear interpolation or nearest neighbor interpolation strategies to generate a standardized event distribution map with a uniform scale and orthogonal grid structure.

8. The underwater backscattering communication method based on visible light according to claim 7, characterized in that, S8 includes: Based on the standardized event distribution map generated by S7, a topology consistent with the LCD screen layout at the transmitter is constructed. Spatial demodulation mesh, traversing mesh cells (r, c), where r and c represent the r-th row and c-th column of the mesh array respectively, and calculating the effective event density D within each mesh cell. r,c And set a decision threshold. Area cell This represents the total number of pixels in each grid cell, and β is the duty cycle coefficient. Binarization and quantization operations are performed when... When the grid is set to a value of logic '1', it is determined to be logic '0' otherwise, thus generating a transient binary data matrix M. k The sliding window algorithm is used to scan the continuous binary sequence, and the frame start boundary is established by matching the autocorrelation peak characteristics of the preset Barker code sequence; then the payload data is reconstructed, and finally the recovered digital information stream is output.

Images