A ship side intelligent identification follow-up type sound-light repelling anti-piracy method and system

Abstract

This invention relates to the field of ship defense technology and discloses a ship hull-side intelligent identification and follow-up acoustic-optical denial anti-piracy method and system. The method includes: scanning the sea surface using a wide-angle polarization visual sensor to extract target motion parameters and calculate a comprehensive threat index, identifying high-threat targets and initiating locking; calculating phase delay and strong light spatial distribution based on target location using phased array electronic scanning technology to establish a mechanically non-mechanically followed acoustic-optical beam locking system; extracting target torso posture and facial expression features during the attack process, calculating a biofeedback effectiveness index, and introducing a phase chaotic perturbation and optical flux frequency domain chaotic modulation model to randomly jump-modulate the acoustic wave phase and strong light frequency; and monitoring the comprehensive threat index in real time until the target moves away and the denial stops. This invention utilizes electronic follow-up to solve the mechanical lag problem and breaks through the target's psychological and equipment shielding through biofeedback and chaotic modulation, improving the reliability and deterrent effect of non-lethal ship defense.

Description

Technical Field

[0001] This invention relates to the field of ship defense technology, specifically to a ship side intelligent identification and follow-up acoustic and light denial anti-piracy method and system. Background Technology

[0002] With the increasing frequency of global ocean trade, high-value vessels such as liquefied natural gas (LNG) carriers and very large crude carriers (VLCCs) face severe challenges to navigation safety in high-risk waters such as the Gulf of Aden and the Gulf of Guinea. In ship defense systems, non-lethal acoustic and visual denial systems have become the mainstream configuration for anti-piracy on merchant ships due to their effectiveness in deterring targets and compliance with international maritime laws. These systems typically utilize high-intensity acoustic devices and high-glare optical devices to induce unbearable physiological discomfort in targets, forcing pirates to abandon their boarding intentions and thus ensuring the safety of the ship and crew without the use of lethal force.

[0003] In existing technologies, shipboard acoustic and optical denial systems mostly employ a radar / AIS guidance combined with a mechanical gimbal-driven architecture. This means that after radar detects a target, a servo motor drives the mechanical gimbal to rotate, causing the acoustic and optical emission units to align with the target. In terms of attack mode, existing equipment typically uses a continuous, constant high-decibel acoustic alarm combined with a strong light for suppression. This technical solution has a certain defensive effect against the conventional approach of low-speed, single targets and played an important role in early maritime defense.

[0004] However, in the face of increasingly professional and well-equipped modern pirate attacks, existing technologies have some shortcomings, mainly in the following two aspects: First, modern pirates often use wolf pack tactics, driving multiple speedboats to approach in high-speed S-shaped maneuvers. Traditional mechanical gimbals are limited by motor inertia and mechanical transmission gaps, resulting in physical lag. Their response speed often cannot keep up with the changes in the angular velocity of the speedboat, causing the acoustic and optical beams to be unable to continuously lock onto the target, thus reducing the denial effect. Secondly, existing denial systems typically use sound waves of constant frequency. These single sound waves are easily perceived by the human brain as background noise and automatically filtered out, leading to psychological adaptation and shielding. Furthermore, with the widespread adoption of active noise-canceling (ANC) headphones, and given the highly predictable nature of continuous, constant sound wave waveforms, the built-in noise-canceling algorithms in these headphones can easily predict the next waveform and generate an inverse wave to cancel it out. Pirates can weaken sound wave attacks by wearing such equipment, thus reducing the effectiveness of sound wave defense. Simultaneously, the strong light from existing denial systems is usually of fixed frequency or a constant light source. The human pupil can quickly adapt to this light intensity through contraction, making it difficult to produce an effective flash blindness effect or induce dizziness, nausea, or other physiological discomfort. More importantly, existing systems lack the ability to perceive the biological feedback of targets and cannot adjust attack strategies in real time according to the pirate's physiological state, resulting in a single attack pattern and vulnerability to countermeasures. Therefore, there is an urgent need for a shipboard intelligent identification and follow-up sound and light denial anti-piracy method and system to solve the above problems. Summary of the Invention

[0005] To address the problems in related technologies, this invention provides a ship side intelligent identification and follow-up acoustic and light denial anti-piracy method to overcome the aforementioned technical problems in existing related technologies.

[0006] To solve the aforementioned technical problem, the present invention is achieved through the following technical solution: In a first aspect, embodiments of the present invention provide a ship's side intelligent identification and follow-up acoustic-optical denial anti-piracy method, specifically including: scanning the sea surface using a wide-angle polarization visual sensor, filtering out sea surface glare clutter using a polarization difference algorithm and extracting the target's distance, relative speed, and angle of attack; constructing a dynamic threat assessment model to calculate a comprehensive threat index; if the index exceeds a preset threat threshold, it is determined to be a high-threat target and locking is initiated; based on the azimuth angle of the high-threat target, calculating the phase delay and initial weight vector required for each element of the acoustic phased array to form a highly directional sound beam, and simultaneously calculating the intensity matrix... The system activates the region and generates a spatial distribution function of light intensity. It then establishes a non-mechanically driven acoustic and optical energy beam lock through electronic scanning. Hierarchical visual perception is initiated to extract the torso posture and facial expression features of high-threat targets, and a biofeedback effectiveness index is calculated. If the index indicates a decline in strike effectiveness, a phase chaos perturbation model and a light flux frequency domain chaos modulation model are introduced to randomly modulate the acoustic wave phase and the high-intensity flashing frequency. The system monitors the comprehensive threat index in real time. When the target moves away and the comprehensive threat index remains below the safety threshold for a preset time window, the denial process is stopped and the defense data is archived.

[0007] As a preferred embodiment of the ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method of the present invention, the expression of the dynamic threat assessment model is as follows: ; In the formula, for The overall threat index at any given moment. For the target distance, For relative velocity, This is a preset normalization constant for the collision hazard time. To prevent tiny constants with a denominator of zero, From the angle of attack for high-threat targets. To determine the confidence level for target classification based on visual and voiceprint recognition. , , These are the weighting coefficients for each item. This is the distance attenuation factor.

[0008] As a preferred embodiment of the intelligent identification and follow-up acoustic and optical deterrence anti-piracy method for ship hull sides described in this invention, the initial weight vector is calculated using the following formula: ; In the formula, For the first The initial weight vector of each transducer To normalize the maximum emission amplitude, Represents the imaginary unit. No. The phase delay required for each array element relative to the reference array element The pointing angle of the acoustic beam.

[0009] As a preferred embodiment of the intelligent identification and follow-up acoustic and optical deterrence anti-piracy method for ship hull sides described in this invention, the expression for the spatial distribution function of the light intensity is: ; In the formula, Represents the spatial distribution function of light intensity. The normalized reference luminous intensity required to reach the target location, Atmospheric transmittance, The maximum physical luminous intensity allowed by the hardware. The physical coordinates on the high-intensity light matrix panel. Azimuth of high-threat targets The center coordinates mapped onto the high-intensity light panel. This is the beam divergence angle coefficient.

[0010] As a preferred embodiment of the ship hull-side intelligent identification and follow-up acoustic and optical deterrence anti-piracy method of the present invention, the calculation formula for the biofeedback effectiveness index is as follows: ; In the formula, The biofeedback effectiveness index and the posture defense index are both included. Facial pain index, Let be the weighting coefficients for each item. Let be the distance confidence function. This represents the theoretical maximum limit of facial pain response.

[0011] As a preferred embodiment of the ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method of the present invention, the expression of the phase chaotic perturbation model is as follows: ; In the formula, The adjusted phased array weight vector. This is the adaptive phase perturbation term.

[0012] As a preferred embodiment of the ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method of the present invention, the adaptive phase perturbation term The calculation formula is: ; In the formula, For acoustic chaos factors, For maximum modulation depth, This represents the total system time delay.

[0013] As a preferred embodiment of the ship hull-side intelligent identification and follow-up acoustic-optical denial anti-piracy method of the present invention, the expression of the optical flux frequency domain chaotic modulation model is as follows: ; In the formula, This is the frequency of strong light flicker. Based on the interference frequency, For frequency agile bandwidth, It is the optical frequency chaos factor.

[0014] As a preferred embodiment of the ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method described in this invention, the method utilizes a polarization difference algorithm to filter out sea surface glare clutter, specifically based on Stokes vector calculation of the linear polarization degree spectrum. and polarization angle spectrum And calculate the enhanced image after flare removal. The enhanced image The calculation formula is: ; In the formula, The original total light intensity, To prevent the minimum value where the denominator is zero, As a glare inhibitor, The maximum linear polarization degree calculated for the current frame. This is a non-linear adjustment coefficient; A direction-selective mask is used to weighted suppress the horizontal polarization characteristics of light reflected from the sea surface.

[0015] Secondly, embodiments of the present invention provide a system for a ship's side intelligent identification and follow-up acoustic-optical denial anti-piracy method, comprising: a panoramic warning and assessment module, used to scan the sea surface using a wide-angle polarization visual sensor, filter out glare clutter using a polarization difference algorithm and extract target motion parameters, and calculate a comprehensive threat index to determine whether to initiate locking; an acoustic-optical follow-up locking module, used to calculate the phase delay of each element of the acoustic phased array and the spatial distribution of light intensity of the strong light matrix based on the target's azimuth, and establish acoustic-optical energy beam locking without mechanical hysteresis through electronic scanning; a biofeedback monitoring module, used to extract the torso defensive posture characteristics and facial pain expression characteristics of the target during the attack process, and calculate the biofeedback effectiveness index in real time; and an adaptive chaos control module, used to dynamically adjust the attack strategy according to the biofeedback effectiveness index, introduce a phase chaos perturbation model and a light flux frequency domain chaos modulation model when the attack effect decreases, randomly modulate the acoustic wave phase and the strong light frequency, and determine whether to stop the denial based on the comprehensive threat index.

[0016] The present invention has the following beneficial effects: 1. This invention adopts an architecture of coaxial deployment of acoustic phased array and matrix-type high-intensity light unit, and uses phased array electronic scanning technology to replace traditional mechanical rotation; by adjusting the emission phase of the array unit and the electronically gated high-intensity light module, the system can change the direction of the acousto-optic beam in a very short time, eliminating the lag of physical motion, thereby enabling continuous and stable locking of high-speed changing direction incoming targets, which helps to improve the hit rate and coverage of dynamic denial.

[0017] 2. This invention introduces a multimodal biofeedback closed-loop mechanism. By using visual sensors to extract macroscopic torso defensive actions such as covering ears and curling up, as well as microscopic facial pain features such as closing eyes and frowning, the biofeedback efficacy index is calculated in real time. Based on this, the phase perturbation of sound waves and the flashing frequency of strong light are dynamically adjusted. This mechanism ensures that the system can automatically switch attack strategies according to the pirate's actual physiological reactions, i.e., adapting to the enemy's defense, thereby helping to ensure that the denial action can always produce good results.

[0018] 3. This invention constructs a phase chaotic perturbation model. When the system detects a decrease in the effectiveness of the attack, it no longer emits constant sound waves. Instead, it introduces random phase noise with nonlinear chaotic characteristics into the array unit. This process makes the waveform of the synthesized sound wave highly unpredictable in time and space, like "garbled code". This causes the prediction algorithm of the active noise-canceling headphones to be unable to generate an effective anti-phase cancellation wave. At the same time, it is easier to break through the psychological adaptation defense of the human auditory system, which helps to aggravate the physiological discomfort of the target pirates and achieve the purpose of ship defense and repelling pirates.

[0019] 4. This invention introduces biofeedback into the high-intensity light control loop and constructs a chaotic modulation model of light flux in the frequency domain. Thus, during the denial attack process, by controlling the flashing frequency of the high-intensity light to jump irregularly within the frequency band sensitive to the human visual nerve, it utilizes the flash blindness effect of the human eye, disrupts the adaptive adjustment mechanism of the visual nerve to the light environment, and induces a strong sense of dizziness and nausea in the vision of the target personnel. This complements the sound wave attack, enhances the deterrent power of non-lethal denial, and achieves ship defense.

[0020] Of course, any product implementing this invention does not necessarily need to achieve all of the advantages described above at the same time. Attached Figure Description

[0021] To more clearly illustrate the technical solutions of the embodiments of the invention, the accompanying drawings used in the description of the embodiments will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the invention. For those skilled in the art, the drawings can be obtained from these drawings without creative effort.

[0022] Figure 1The present invention provides a flowchart of a ship side intelligent identification and follow-up sound and light denial anti-piracy method.

[0023] Figure 2 The system module architecture diagram provided for this invention.

[0024] Figure 3 A comparison diagram of chaotic modulation waveforms provided by the present invention. Detailed Implementation

[0025] 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. Example

[0026] In modern ocean shipping, traditional mechanical follow-up sound and light denial equipment suffers from problems such as slow response and limited attack modes in the face of the high-speed maneuverability of pirate speedboats' "wolf pack tactics" and the upgrading of countermeasures such as noise-canceling earplugs.

[0027] To solve the above technical problems, such as Figure 1 As shown, Embodiment 1 of the present invention provides a ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method. Specifically, Embodiment 1 uses an LNG (liquefied natural gas) carrier navigating in the high-risk waters of the Gulf of Aden as an example: the ship's hull is equipped with a linearly distributed intelligent defense zone. This intelligent defense zone includes several acoustic phased array units, a matrix-type strong light denial unit composed of a high-brightness LED array, a polarization vision sensor group, and an edge computing control terminal. Furthermore, the several acoustic phased array units are composed of piezoelectric ceramic transducer arrays with spacing... .

[0028] In the specific implementation of Example 1: First, the sea surface is scanned using a wide-angle polarization vision sensor. The polarization difference algorithm is used to filter out sea surface glare clutter and extract the target's distance, relative speed, and angle of attack. A dynamic threat assessment model is constructed to calculate the comprehensive threat index. If the index exceeds a preset threat threshold, it is determined to be a high-threat target and locking is initiated. This method uses the polarization difference algorithm to pre-eliminate the "white shadow" in the image caused by strong sea surface reflection at the optical physical level. At the same time, it combines multi-dimensional motion parameters and classification confidence to construct a dynamic threat assessment model, ensuring that the system can accurately identify high-speed maneuvering pirate speedboats and effectively filter out non-threat interference such as fishing boats and buoys, achieving objectivity and timeliness of defense activation. Secondly, when the system locks onto a target, based on the azimuth of the high-threat target, it calculates the required phase delay and initial weight vector of each element of the acoustic phased array to form a highly directional acoustic beam. At the same time, it calculates the activation region of the strong light matrix and generates a light intensity spatial distribution function. It then establishes a mechanically servo-free acoustic-optical energy beam lock through electronic scanning. This method uses phased array electronic scanning technology to replace the traditional mechanical servo architecture. Through microsecond-level phase offset control and electronic mapping of the strong light matrix, the acoustic-optical energy beam can lock onto the target instantaneously without mechanical inertia or physical wear. This eliminates the response lag problem of traditional gimbals when facing pirate "wolf pack tactics" and S-shaped high-speed maneuvers, and improves the stability of dynamic tracking and the energy concentration of the strike. Then, during the attack, tiered visual perception is activated to extract the torso posture and facial expression features of high-threat targets, and a biofeedback effectiveness index is calculated. If the index indicates a decline in attack effectiveness, a phase chaotic perturbation model and a light flux frequency domain chaotic modulation model are introduced to randomly jump-modulate the acoustic wave phase and the frequency of intense light flashing. This method introduces a biofeedback closed-loop control mechanism. When the target is detected to have developed psychological adaptation or to use active noise reduction equipment to counteract the decline in attack effectiveness, a chaotic perturbation model is introduced to break the predictive rules of the acoustic and optical waveforms. "Acoustic scrambling" and disordered flashing forcefully disrupt the body's adaptive adjustment mechanism and electronic noise reduction algorithm, ensuring that the physiological deterrent force of the denial means remains high, thereby ensuring that the denial action always produces good results. Finally, when the target is forced to turn away or stop moving, and the threat index remains below the safety threshold, the system automatically stops the attack, archives the defense data, and returns to the alert state.

[0029] Furthermore, to better illustrate the technical solution of Embodiment 1 of the present invention, a detailed description of the intelligent identification and follow-up acoustic and visual deterrence method for anti-piracy on the ship's side is provided, specifically including the following: S1. Panoramic Vigilance and Dynamic Threat Assessment, which includes the following sub-steps: S11. The system scans the sea surface in real time using a wide-angle polarization vision sensor on the side of the ship. When an abnormal buoy is detected, a polarization difference algorithm is used to filter out sea surface glare clutter and extract the target's distance. Relative approach speed and the angle of attack At the same time, an alarm was sounded to alert staff.

[0030] Specifically, in this embodiment 1, in order to accurately extract the target under strong interference from sunlight reflection on the sea surface, the specific steps for filtering out sea surface glare clutter using the polarization difference algorithm are as follows: S111: Perform multi-channel polarization image acquisition. Using a wide-angle polarization vision sensor, simultaneously acquire raw light intensity images in four different polarization directions, respectively... , , and .

[0031] in, The coordinates are pixel coordinates, and the subscripts indicate the angle of the polarizer's transmission axis. 0° is usually parallel to the sea level.

[0032] S112: Perform Stokes vector calculation. For each pixel, calculate the Stokes vector describing the full polarization state of the beam through four-way intensity calculation. : ; in, This represents the total light intensity image, equivalent to the image captured by a regular camera, and includes strong glare. Characterizes the difference between horizontal and vertical polarization components; Characterizes the difference between the 45° and 135° polarization components.

[0033] S113: Construct polarization feature maps. Calculate the linear polarization degree map (DoLP) and polarization angle map (AoLP) of the scene at this moment as intermediate variables for denoising: ; ; in, Represents the linear polarization degree spectrum; This represents a polarization angle spectrum.

[0034] S114: Constructing an adaptive glare suppression filter. Utilizing the high intensity and high polarization of sea surface glare, a weighted polarization difference filter is constructed. By introducing a DoLP negative feedback mechanism, the enhanced image after glare removal is calculated. : ; In the formula, This represents the original total light intensity, including the target and clutter. To prevent the minimum value where the denominator is zero; As a glare inhibitor, For the maximum linear polarization degree calculated in the current frame, for the sea surface flare point, Approaching The glare suppression factor approaches 0, thus suppressing highlights; for pirate ships / personnel (diffuse reflection). The glare suppression factor is low, approaching 1, thus preserving the target brightness; This is a nonlinear adjustment coefficient used to accelerate the attenuation of highly polarized clutter; To select the direction of the mask, since the light reflected from the sea surface is mainly horizontally polarized ( ), set mask logic: when (like When the value is 0, an additional penalty weight is applied; otherwise, it is set to 1 to filter out residual horizontal light.

[0035] For example: In this embodiment 1, the orientation selection mask The specific construction method is as follows: The system calculates a dynamic angle tolerance threshold including motion blur compensation based on the read hull roll angular velocity to adapt to measurement errors caused by severe rolling under high sea states; coordinate system rotation compensation is performed on the original polarization angle map using real-time roll angle data to calculate the absolute polarization deviation of each pixel relative to the physical sea level, and a limit attitude gating function based on the field of view boundary is constructed simultaneously to prevent false detection under extreme attitudes; a nonlinear weighted mask is generated based on the absolute polarization deviation and the dynamic threshold to suppress the sea surface glare clutter region exhibiting horizontal polarization characteristics (weights approaching 0), while retaining the target diffuse reflection region exhibiting non-horizontal polarization characteristics (weights approaching 1), thereby generating a direction selection mask with dynamic line-of-sight stability. Furthermore, sensitivity analysis shows that as long as the extinction ratio of the polarization sensor is greater than 100:1 and the angle calculation accuracy is better than 0.5°, this method can effectively separate the sea surface background from the target, meeting the needs of engineering applications.

[0036] In this embodiment 1, the nonlinear adjustment coefficient The determination method is as follows: Based on the principle of maximum image entropy, the one-dimensional information entropy of the image after de-flare processing is defined as the objective function that measures the degree of image detail preservation and gray-level distribution uniformity. Subsequently, within the preset physical effective range, the golden section search algorithm is used to iteratively select candidate candidates. The system calculates the corresponding image entropy value and continuously narrows the search range by comparing the entropy values ​​of the points within the interval to approach the extreme point. Finally, it selects the value that maximizes the objective function as the optimal adjustment coefficient, thereby automatically obtaining the objective parameter that achieves the best balance between glare suppression and target texture preservation.

[0037] S115: Perform target binarization extraction. This is applied to the image after the polarization cleaning described above. Adaptive thresholding (such as the Otsu algorithm) is used to extract the highlighted connected components. These extracted connected components represent the true target obscured by the glare.

[0038] S116: Based on the connected components of the real target extracted in step S115, calculate the geometric motion parameters of the real target, specifically including: S116-1 Calculate the centroid coordinates of the target connected component. ,in The horizontal pixel coordinates of the image. These are the pixel coordinates in the vertical direction.

[0039] S116-2. Based on the pinhole camera imaging model and combined with camera intrinsic parameters, calculate the horizontal azimuth angle of the target relative to the camera's optical axis (i.e., the direction of the ship's side normal by default). The calculation formula is: ; In the formula, From the angle of attack; The horizontal pixel coordinates of the target centroid; The horizontal coordinates of the camera's principal point (usually half the image width); This is the equivalent focal length of the camera in the horizontal direction; It is the arctangent function.

[0040] If the system consists of multiple cameras stitched together, the physical yaw angle offset of the camera installation must also be added. ,Right now .

[0041] S116-3. Using sea level as the reference plane, and employing the monocular ranging principle, combined with the camera installation height... and the angle of depression of the camera's optical axis relative to the horizontal sea level Solving for the physical distance in a straight line : ; in, Let be the angle subtended by the target in the vertical direction of the image. Vertical focal length It is the vertical principal point.

[0042] S116-4, For continuous time and The distance is differentially analyzed and filtered to calculate the relative velocity. : ; in, This indicates Kalman filtering for smoothing. This represents the sampling time interval.

[0043] S12. The system constructs a dynamic threat assessment model and calculates the comprehensive threat index of the target. This is used to determine whether to initiate a denial-of-attack strategy. The specific formula for the dynamic threat assessment model is expressed as: ; In the formula, for The overall threat index at any given time, with a range of values. The closer the value is to 1, the greater the threat. Distance to target; It is the relative velocity; This is a preset normalization constant for the collision hazard time; To prevent tiny constants with a denominator of zero; The angle from which the target is attacked; The confidence score for target classification based on visual and voiceprint recognition (e.g., 1 for "manned speedboat" and 0 for "fishing net buoy"). These are dimensionless weighting coefficients; This is the distance attenuation factor.

[0044] For example: In this embodiment 1, the multimodal fusion confidence level Based on the DS evidence theory, the method is as follows: The first step is to construct the basic probability assignment function (BPA), including the visual BPA ( ) and auditory BPA ( ): Visual BPA ( Let the output probability of the visual classifier be... Define the degree of confidence that vision has in the existence of pirates as... The level of trust in "uncertainty" is ; Specifically, to achieve efficient and high-precision target attribute determination on edge computing devices, the visual classifier employs a lightweight convolutional neural network (CNN) architecture. The specific steps are as follows: A1. Based on the target connected components extracted by the polarization difference algorithm in step S11, extract the corresponding image sub-blocks. Scale these sub-blocks to a standard input size (e.g., ...). (pixels), and then normalized.

[0045] A2. An improved MobileNetV3-Small network is used as the feature extraction backbone. This network includes depthwise separable convolutional layers and SE channel attention modules, enabling rapid extraction of target texture and contour features on computationally limited shipborne terminals.

[0046] Classification Header: The network end connects to a fully connected layer, with 2 output nodes, corresponding to "non-threat targets (fishing boats / buoys)" and "threat targets (pirate speedboats)" respectively.

[0047] A3. The output vector of the fully connected layer... Input the Softmax function to calculate the posterior probability of belonging to the "threat target" category. : ;this The value directly reflects the visual level of confidence in the existence of pirates. .

[0048] Hearing BPA ( Let the matching coefficient between the voiceprint collected by the hydrophone and the pirate speedboat database be . (Normalized to [0,1]). The level of trust in hearing is defined as... The uncertainty is ; Specifically, in order to quantify the similarity between the hydrophone-collected signal and the acoustic signature of the pirate speedboat, this embodiment uses a cosine similarity matching algorithm based on MFCC features to obtain auditory trust levels. The specific steps are as follows: B1. Raw underwater acoustic signals collected by the hydrophone Perform bandpass filtering (e.g.) Low-frequency wave noise and high-frequency marine biological noise are filtered out. Then, windowing and frame splitting processing is performed (Hamming window, frame length 25ms, frame shift 10ms).

[0049] B2. Perform a Fast Fourier Transform (FFT) on each frame of the signal to calculate the power spectrum, and extract energy features using a Mel filter bank. Take the logarithm of the filter bank output and perform a Discrete Cosine Transform (DCT) to extract the first 13 dimensions of coefficients as Mel frequency cepstral coefficients (MFCC).

[0050] Calculate the mean vector of MFCC coefficients for all frames within the current time window. , which serves as the voiceprint feature vector of the current target.

[0051] B3. The system has a pre-installed "Pirate Speedboat Typical Voiceprint Database", which stores... Standard voiceprint feature vectors of a known pirate ship type .

[0052] B4. Calculate the current feature vector. Nearest neighbor reference vector in the database Cosine similarity: ;in, The range of values ​​is .

[0053] B5. To map physical similarity to probability confidence levels in the [0,1] interval and filter out noise interference from low-similarity levels, a modified Sigmoid activation function is used as the normalization mapping: ; In the formula: The baseline threshold for voiceprint matching can be determined based on the cosine similarity value corresponding to the point of equal error rate between the statistical false recognition rate and the false rejection rate of the system for pirate voiceprints under a preset signal-to-noise ratio environment. For example, a value of 0.6 means that significant trust is only generated when the similarity is higher than 0.6. The steepness coefficient can be determined by fitting the Sigmoid curve of the experimental test data to maximize the probability discrimination near this threshold. For example, a value of 10.0 can be used to control the sensitivity of the decision.

[0054] The second step is to use the orthogonal sum rule to... and Perform fusion and calculate the trust level after fusion. : ; In the formula, The conflict coefficient is calculated using the following formula: ; In the simplified model of this embodiment, if the sensor only outputs "is target" or "uncertain", then the conflict coefficient is... This can be ignored.

[0055] In this embodiment 1, the distance attenuation factor The method for determining this is as follows: Let the maximum effective suppression distance of the acoustic denial system be... At this distance, the sound pressure level decays to the critical deterrence threshold. In the mathematical model, the distance term at this point... The contribution should decay to the initial value. (Common cutoff standards in engineering). The calculation formula is: .

[0056] In this embodiment 1, the weighting coefficient The offline self-learning method based on particle swarm optimization (PSO) was determined, and the steps are as follows: Step A: Construct the sample set. Collect a historical dataset containing positive samples (real pirate attack video and data) and negative samples (normal fishing boat approach and buoy drift data). ,in Extracted by the sensor Equal state vector, The label is real (1 for threat, 0 for non-threat).

[0057] Step B: Define the fitness function In the dataset The F1-Score (harmonic mean) is used to balance precision and recall. ; Among them, the system prediction results If and only if the calculated .

[0058] Step C: PSO iterative optimization.

[0059] Initialization: in three-dimensional space Random initialization One particle.

[0060] Iteration: Each particle updates its velocity and position based on its own historical best position and the group's historical best position.

[0061] Constraints: Limitations and .

[0062] Termination: When the fitness function no longer significantly improves or reaches the maximum number of iterations, output the globally optimal particle position. .

[0063] In this embodiment 1, the preset collision danger time normalization constant This is equal to the sum of three factors: the inherent response delay of the system hardware from perception to excitation, the medical statistical threshold time for the human vestibular and retinal systems to produce a physiological rejection effect on sound and light stimuli, and the hydrodynamic limit time required for the target speedboat to complete an emergency evasive maneuver at a typical attack speed. This method, by referencing hardware specification indicators, medical assessment standards for non-lethal weapon effects, and the International Maritime Organization's ship maneuvering hydrodynamics, [integrates these factors]. This is solidified into an objective critical time window that can effectively block collision risks in a physically sound manner, ensuring the physical interpretability and robustness of the threat assessment model.

[0064] S13. Access Determination: Setting Threat Threat Thresholds .like If the system determines that the target is a high-threat target (such as a pirate ship), it immediately activates the phased array denial unit; if Then, remain silent and monitor.

[0065] For example, in this embodiment 1, in order to minimize the false alarm rate while ensuring a high detection rate, this embodiment uses an adaptive method based on ROC curve analysis and Youden's exponent maximization to determine the threat threshold. The specific steps are as follows: Step 1: Construct a validation dataset. Collect and organize data on typical targets encountered by ships during their historical voyages to construct a labeled validation dataset. .in, For the first The comprehensive threat index input value calculated for each sample at the moment of encounter, based on the aforementioned The parameters are calculated accordingly; This is the true category label for the sample. This indicates that the threat is confirmed as a pirate threat (positive sample). This indicates that the vessel is a regular merchant ship or fishing vessel (negative sample).

[0066] Step 2: Traverse the candidate thresholds. Within the interval... Internally set fine step size (e.g.) ), generating a series of candidate thresholds For each Statistical analysis of its role in the dataset The classification performance on the screen: True case rate: That is, the percentage of all real pirates who were correctly identified.

[0067] False positive rate: This refers to the proportion of all ordinary ships that are mistakenly identified as pirates.

[0068] Step 3: For each Calculate its corresponding Youden index : ; when At its maximum, the system achieves a mathematically optimal balance between recall and precision.

[0069] Step 4: Select the candidate threshold that maximizes the Youden index as the final, fixed system threat threshold. .

[0070] Specifically, for example, in this embodiment 1, weights are set. Distance decay factor Threat threshold Assuming in At that moment, the sensor detected a speedboat at a distance. ,by high-speed direct impact on the ship ( ), and classification confidence Substitute the values ​​into the formula to calculate the final threat index: .because The system identifies the threat as extremely high and immediately initiates defense. This step, through multi-dimensional physical quantity coupling calculations, can effectively distinguish between normal passing fishing boats and pirate attacks, avoiding false alarm interference.

[0071] S2. Mechanical-free servo locking based on phased array, specifically including the following sub-steps: S21. Obtain ambient temperature in real time through the shipboard weather station. (Celsius), relative humidity and wind speed vector The velocity of sound is corrected using a formula that takes temperature and humidity into account: ; Introducing wind speed compensation to correct target orientation: calculating crosswind Caused drift angle The actual pointing angle of the emitted acoustic beam. It should be noted that, Target azimuth, target angle of approach It is the horizontal angle between the target and the optical axis of the vision sensor. The optical axis of the vision sensor and the normal of the acoustic phased array are parallel (coaxial) during installation, therefore numerically... .

[0072] The system is based on the pointing angle of the acoustic beam. Calculate the phase delay required for each element of the acoustic phased array to form a highly directional sound beam: ; In the formula, For the first Each array element is relative to the reference array element (usually 1). The required phase delay; For carrier frequency; To correct for the speed of sound; The distance between array elements.

[0073] S22. Control using beamforming weight formula The transmitted signals of each array element. In the initial stage, the initial weight vector... The calculation is as follows: ; In the formula, For the first The initial weight vector of each transducer; This represents the normalized maximum emission amplitude. This represents the imaginary unit. The formula, by precisely controlling the phase of each unit, utilizes the principle of wave interference in space. The energy peaks in one direction are synthesized, while they cancel each other out in other directions.

[0074] Specifically, for example: in this embodiment 1, assuming the carrier frequency Array element spacing (Approximately half wavelength, to avoid grating lobes), the pointing angle of the acoustic beam. °, (Represents the system at full power output), corrected sound velocity The system calculates the first... The phase compensation amount for each array element is: Initial weight vector The complex number results indicate that, in order to ensure that the acoustic wave of the first array element is in phase with other array elements at a 26.77° angle, the system needs to maintain the drive signal of this array element at 100% of its maximum amplitude and introduce approximately... The phase lag in radians, in actual hardware (FPGA or DSP), means that the output driving sine wave needs to be delayed by approximately 20 microseconds relative to the reference clock (i.e., This process precisely achieves electronic deflection of the beam. By precisely controlling the phase delay parameter of each transmitting unit in the array, and utilizing the principle of coherent superposition of waves, the multiple sound waves propagate through space, causing only the acoustic beam's pointing angle to change. The beam forms an energy-enhanced in-phase interference peak (i.e., the main lobe of the beam) on the upper part of the beam, while canceling each other out in other directions. This replaces the physical rotation of the traditional mechanical gimbal with electronic scanning, eliminating motion lag caused by mechanical inertia and achieving microsecond-level precise locking and continuous energy suppression of high-speed S-shaped maneuvering targets.

[0075] S23. Constructing electron servo mapping of a strong light matrix.

[0076] The system reads the real-time visibility measured by the shipborne visibility meter. Using Koschmieder's law, the atmospheric extinction coefficient in a marine environment is estimated to be approximately... Combined with target distance Calculate atmospheric transmittance : .

[0077] The system is equipped with a matrix-style high-intensity light blocking unit composed of a high-brightness LED array, which is deployed coaxially with the acoustic phased array. To achieve spatial synchronization between the high-intensity light and the acoustic beam, the system adjusts the beam direction according to the pointing angle of the acoustic beam. Calculate the activation region index of the intensity matrix And generate the spatial distribution function of light intensity. : ; In the formula, To achieve the normalized reference luminescence intensity required to reach the target location, ensure that... Sufficient lux illuminance is generated at the location; Atmospheric transmittance, and ; The maximum physical luminous intensity allowed by the hardware; The function is used for extreme case truncation to prevent interference in extremely dense fog ( The compensation value calculated under these circumstances exceeds the equipment burnout limit; These are the physical coordinates on the high-intensity light matrix panel; Target azimuth The center coordinates mapped onto the high-intensity light panel; This is the beam divergence angle coefficient. This step uses electronic gates to select specific LED module areas, replacing mechanical rotation, to achieve beam coverage of the target area in a very short time.

[0078] Specifically, for example: in this embodiment 1, it is assumed that the system is in a complex weather environment with dense sea fog, and the target distance is... ,visibility Calculate transmittance Assuming a reference luminous intensity (i.e., the source intensity that can cause glare under ideal clear weather conditions), the maximum physical light intensity allowed by the hardware. Calculate atmospheric transmittance. This indicates that under the current sea fog, without compensation, only about 14.1% of the light energy can penetrate the fog and reach the target; calculations ,because The system determined that the intensity was within the hardware safety range, and therefore set the center drive light intensity to [value missing]. This result demonstrates that, compared to clear weather, the system automatically increased the luminous intensity by approximately seven times, thus forcibly penetrating sea fog. (Assuming the target's location...) Projection mapping coefficients of the high-intensity light matrix The system calculates the center physical coordinates of the beam on the high-intensity light panel. The system locks onto the physical coordinates of the high-intensity light matrix. The LED module at that location serves as the activation center (corresponding to the first...) (e.g., column 16), control it with The intensity of light emission. Simultaneously, control its adjacent positions (such as...) The module at the target location emits auxiliary light at approximately 60% intensity to form a focused light cone. This step establishes a real-time electronic mapping relationship between the target's orientation and the physical coordinates of the LED array, and uses electronically selected light-emitting units in specific areas to replace the physical rotation of the traditional mechanical gimbal, achieving lag-free follow-up coverage of high-speed moving targets by a strong light beam in an extremely short time.

[0079] S3. After manual authorization, the system executes multimodal biofeedback closed-loop and adaptive mutation attack, specifically including the following sub-steps: S31. Hierarchical visual locking and feature extraction: After the system emits the initial sound wave, it initiates a hierarchical visual perception logic from macroscopic to microscopic: S311, Primary Feature (Macro): Utilizing a telephoto camera combined with geometric feature analysis of key points in the human skeleton, the torso posture of the target individual is identified. Emphasis is placed on monitoring large-amplitude defensive movements such as "hands covering ears," "body curling up," and "head burying," and calculating the postural defense index. The specific steps are as follows: First, key point extraction and normalization benchmark establishment are performed. The system first acquires the first key point from the visual sensor. Frame images, using skeleton detection algorithms to extract the target person's... Coordinates of key points To eliminate the influence of target distance (image scale) on distance calculation, a normalized reference length must be established. The Euclidean distance from the neck to the center of the pelvis is selected as the benchmark (this trunk length is relatively rigid in a short time and is suitable as a scale): ; In the formula, Denotes the Euclidean norm; To prevent tiny quantities with a denominator of zero; for At any given moment, the skeleton detection algorithm extracts the two-dimensional coordinate vectors of key points on the neck of the target person; for At any given moment, the skeleton detection algorithm extracts the two-dimensional coordinate vector of the key point at the center of the target person's pelvis. All subsequent distance calculations must be divided by this value. To achieve dimensionless designation.

[0080] Then, we define geometric feature sub-indices for two core defensive behaviors: the ear protection index. With the curling index .

[0081] Among them, the ear protection index This is used to quantify how close the wrist is to the ear. The normalized distances from the left wrist to the left ear and from the right wrist to the right ear are calculated: ; In the formula, for At any given moment, the skeleton detection algorithm extracts the two-dimensional coordinate vector of the key point on the left wrist of the target person; for At any given moment, the skeleton detection algorithm extracts the two-dimensional coordinate vector of the key point of the left ear of the target person; for At any given moment, the skeleton detection algorithm extracts the two-dimensional coordinate vector of the key point on the right wrist of the target person; for At any given moment, the skeleton detection algorithm extracts the two-dimensional coordinate vector of the key point of the right ear of the target person; Mapping the distance using Gaussian radial basis functions The probability value (the closer the distance, the larger the value): In the formula, This is the sensitivity coefficient; when the wrist touches the ear, When the hand is more than half the length of the torso away from the head, It rapidly decays to 0.

[0082] Among them, the curling index Used to quantify a sudden drop in the target's center of gravity or a curvature of the spine. This is determined by comparing the current vertical projection height with the target's historical average height (under normal driving conditions). ; In the formula, The vertical projection length of the vector from the neck to the pelvis at the current moment; This is the sliding average of normal upright sitting height detected over a past period (e.g., the past 3 seconds). When the pirate suddenly bends over, the molecule decreases. Increase.

[0083] Finally, considering that "ear protection" and "curling up" may occur simultaneously or independently (e.g., only covering the ears without lowering the head, or only lowering the head to avoid the movement without covering the ears), max pooling is used to calculate the final result. This ensures that any defensive action triggers a high feedback value. Its expression is: .

[0084] Specifically, if (Normal driving) .like (Covering ears violently) (Slightly lowering the head), then .

[0085] For example, sensitivity coefficient The determination method is as follows: First, based on standard human anatomical statistics, the average distance from the tragus to the ipsilateral acromion is selected as the physical effective boundary of the ear-protecting action, and its ratio to the trunk length is defined as the normalized anatomical threshold. Secondly, based on Gaussian distribution The cutoff principle sets a cutoff value where the probability of ear protection decreases to near zero when the wrist position exceeds the physical boundary (set to 0). Finally, solve... .

[0086] S312. Secondary Features (Microscopic): When the distance or zoom level is sufficient, activate the facial region of interest (ROI) to extract features such as eyebrow spacing contraction and eyelid closure, and calculate the facial pain index. .

[0087] Specifically, facial pain index Based on the Facial Action Coding System (FACS), the calculation focuses on quantifying two geometric features strongly correlated with physiological pain: "eyelid closure (AU6 / AU43)" and "brow furrowing (AU4)". The specific process is as follows: First, the sharpness of the extracted regions of interest (ROI) is assessed, and the distance confidence function is calculated. The Laplacian operator is used to extract the high-frequency components of the image, and their variance is calculated as the sharpness score. : ; A distance confidence function is constructed based on this score. : ; In the formula, Represents the statistical variance operator; The second derivative of the grayscale image of the face; The preset sharpness threshold; This is the transition slope coefficient. This formula characterizes the transition slope coefficient when the image is blurred (…). When (low), In step S32, the facial item weights are automatically reset to zero; when the image is clear, Approaching 1, enable facial feedback.

[0088] Then, in Provided the image is available, 68 key facial points were extracted. To eliminate errors caused by different face sizes, the distance between the outer corners of both eyes was selected. As a normalization benchmark: This section uses the standard dlib68 point index, with 36 representing the left corner of the eye and 45 representing the right corner of the eye.

[0089] Extract two core pain features: Eye aspect ratio Used to determine if the eyes are tightly closed due to pain. Calculate the average opening and closing degree of both eyes: ; When eyes are open normally When tightly closed .

[0090] Normalized eyebrow spacing Used to determine if eyebrows are wrinkled. Calculate the distance between the inner points of the left and right eyebrows (21 o'clock and 22 o'clock): ; When the eyebrows are calm Larger, when frowning in pain Significantly reduced.

[0091] because and All of these are physical quantities and need to be mapped to... The probability of suffering. Using a reverse linear rectified function model, a statistical baseline value under normal conditions is set: Eye pain : ; It indicates that the more closed the eyes are, the larger the difference, and the closer the component is to 1; The aspect ratio of the eye when it is normally open.

[0092] Pain level in the eyebrows : ; It states that the closer the eyebrows are to each other and the smaller the distance between them, the closer their weight is to 1; The distance between the eyebrows when the eyebrows are at rest.

[0093] The method for determining the frown sensitivity range is as follows: A standard facial expression dataset is selected as the sample source, and the normalized average eyebrow spacing is calculated for a large number of samples under the "calm" label. The normalized eyebrow spacing statistical mean when the expression reaches its peak under the "pain" label The difference between the two is defined as the effective physiological dynamic range of corrugator supercilii contraction, i.e. .

[0094] Finally, calculate the overall facial pain index. : ; In the formula, The weighting coefficient for eye features is determined as follows: First, an internationally recognized standard pain facial expression database (such as UNBC-McMaster) is selected as the sample source; second, the Pearson correlation coefficients between the intensity of eye closure and eyebrow contraction and the actual pain label are calculated, denoted as . and Finally, Defined as the proportion of the eye correlation coefficient to the sum of the absolute values ​​of the two, i.e. This allows us to quantify the relative importance of ocular features in pain expression based on objective statistical data.

[0095] S32, Attitude-Based Defense Index and facial pain index The current strike effectiveness is calculated using a weighted average: ; In the formula, The biofeedback efficacy index; This refers to the posture defense index; Facial pain index; These are the weighting coefficients; For distance confidence function; This is the theoretical maximum facial pain response limit, due to the above. The range of values ​​is Therefore, in this embodiment The value is 1; For example, in this embodiment 1, the weighting coefficient The method for determining it is as follows: First, select publicly available computer vision datasets with industry credibility, corresponding to the primary features (pose) and secondary features (expression) of this system: Pose baseline: The MS COCO Keypoint Challenge dataset (a computer vision dataset released by Microsoft) was selected. This dataset covers human skeleton detection scenarios with complex backgrounds, occlusion, and multiple poses, and has statistical similarity to the pirate deck activity scene.

[0096] Facial expression benchmark: The EmotioNet facial expression dataset was selected, which focuses on facial unit (AU) recognition in non-laboratory environments, under natural lighting and partial occlusion.

[0097] Then, the average accuracy of the skeleton detection algorithm mentioned above is used as the objective reliability score for this feature channel: let the baseline score of pose detection in complex scenes be... Let the benchmark score for face unit detection in a field scene be set. .

[0098] Finally, the weighting coefficient is defined as the proportion of the reliability score of that feature channel in the total system information. Furthermore, to ensure the system's logical closed loop, a weighting coefficient is set... The calculation formula is as follows: .

[0099] Specifically, for example: based on industry benchmark data, take ,Pick Substitute into the formula to calculate: , .

[0100] S33. Constructing a Phase Chaotic Perturbation Model: For active noise-canceling headphones that pirates might wear, the system no longer simply changes the volume, but instead dynamically disrupts the phase consistency of the array, breaking the coherent superposition of sound waves at the target location. The adjusted phased array weight vector... The calculation formula is as follows: ; ; In the formula, For adaptive phase perturbation terms; As an acoustic chaos factor, it obeys A random variable with a given distribution; This represents the maximum modulation depth. In the formula, This ensures that the sound wave energy still points in a direction on a macroscopic scale. ; It is random phase noise added to each array element; Invalid feedback item; This represents the total system time delay. For example: For example: In this embodiment 1, the acoustic chaos factor A Logistic chaotic mapping is used to generate phase transitions that achieve full array synchronization. The specific method is as follows: First, the system maintains a globally unified chaotic state variable. For the i-th... Each time step (corresponding time) The state value at the next time step is calculated using the following iterative formula: ;in, Let be the bifurcation parameter, and in nonlinear dynamics, if and only if When the logistic mapping reaches its maximum Lyapunov exponent, that is... The system is in a fully mapped state and can traverse... All values ​​within the interval, exhibiting purely chaotic characteristics. As the initial value, take Any constant within the interval except for fixed points (e.g.) Furthermore, all array elements share the same initial value. Finally, the iteratively generated... Mapping to the formula requirements Interval, calculate the acoustic chaos factor: This step ensures that the spatial directivity of the beam (i.e., the direction of the main lobe) remains unchanged by applying a time-synchronized but numerically random phase offset to all array elements, while making the phase of the synthesized sound wave highly unpredictable on the time axis, thereby effectively disrupting the phase prediction mechanism of the active noise reduction algorithm.

[0101] In this embodiment 1, the maximum modulation depth Based on the principle of wave interference, the phase circumferential complete decoherence boundary method is used to determine the maximum phase perturbation amplitude, which must cover... The antiphase interference region was used to achieve complete collapse of beamforming, and the physical limit value was derived. .

[0102] In this embodiment 1, the total system time delay The delay is the dynamic sum of the system's inherent latency, physical propagation latency, and physiological response latency. The inherent latency is obtained through the difference between the hardware timestamps of visual acquisition and command issuance; the physical propagation latency is calculated based on the real-time ratio of the target distance to the speed of sound in air; and the physiological response latency is dynamically weighted and synthesized based on the type of limb or facial defensive action identified by the system, using a pre-set medical statistical response time constant. Through the real-time superposition of these three factors, the system can automatically correct for timing deviations caused by changes in target distance and differences in response type, achieving precise targeting of denial-of-attack and biofeedback in the time dimension.

[0103] Reference Figure 3 The above-mentioned chaotic modulation waveform comparison diagram shows that... The calculation formula characterizes: when When the attack is effective, ( The chaotic phase perturbation term approaches 0. "disappear", At this point, the system maintains the initial weight vector. Each array of sound waves is continuous and constant; when If it is lower, then ( As the chaotic phase perturbation term increases, Intervene, calculate and execute the revised weight vector At this point, the system adds random phase noise to each array element. This random disturbance is as chaotic as "scrambled code," unlike existing constant sound waves, which are easily "automatically filtered" and psychologically blocked by the human brain. The phase and amplitude of the randomly disturbed sound waves will randomly jump within microseconds, which not only easily causes nausea and irritability in humans, but also makes the active noise cancellation algorithm built into the active noise cancellation headphones unable to predict the waveform of the next moment, thus failing. Therefore, even if the pirates wear active noise cancellation headphones, they cannot effectively block the sound waves, thereby breaking their psychological defenses and achieving a better rejection effect.

[0104] Specifically, for example: in this embodiment 1, when the system detects that the biofeedback efficacy index has dropped to At that time, the chaotic perturbation mode is automatically activated. Assuming the maximum modulation depth... The random acoustic chaos factor generated by the Logistic mapping at the current time. The system first calculates the additional phase disturbance. ; and then superimposed it onto the original directional phase of the first array element ( Generate the corrected weight vector on the ) Compared to the initial weights without introducing chaos (Corresponding to a phase of 0.376 rad), the corrected weight vector A significant phasor rotation (phase shift of approximately 115°) occurred in the complex plane. This step, by introducing chaotic phase noise negatively correlated with biological feedback while maintaining the macroscopic directivity of the beam, causes the waveform of the synthesized sound wave to undergo a drastic nonlinear jump on the microscopic time scale, i.e., from a constant wave to "acoustic gibberish." This highly unpredictable random perturbation not only effectively disrupts the anti-phase cancellation mechanism of active noise-canceling headphones based on historical waveform predictions, but also breaks through the psychological adaptation defenses of the human auditory system, thereby enhancing the physiological deterrent effect of denial strikes when the target develops tolerance.

[0105] S34. Construct a frequency-domain chaotic modulation model for optical flux. The system will incorporate the biological feedback efficiency index. A high-intensity beam control circuit is introduced. When a decrease in impact effectiveness is detected, the constant high-intensity beam is no longer emitted; instead, it switches to a strobe-induced dazzle mode. High-intensity beam strobe frequency. The calculation formula is as follows: ; In the formula, Based on the interference frequency; For frequency agile bandwidth; This is the optical frequency chaos factor, which differs from the aforementioned acoustic chaos factor. The same method is used to generate it. Specifically, a fixed initial value is taken (e.g., ...). ), generate time random variables that follow a distribution in [-1,1], in order to realize the frequency of strong light in Disordered transitions within the bandwidth.

[0106] For example, frequency agile bandwidth The method for determining it is as follows: First, we use D.H. Kelly's standard model of visual psychophysics to describe the sensitivity of the human eye to flashing light stimuli of different frequencies. The standard model of visual psychophysics is a bandpass filter determined by the time constant of the retinal photochemical response, and its analytical expression is: ; In the formula, The blinking frequency; The amplitude normalization constant is taken as in this calculation. (Because we only focus on relative bandwidth, not absolute gain); and This is a physiological statistical objective constant under photopic vision conditions. Based on standard photopic vision (illuminance)... ) statistical data, take ;Pick This value corresponds to the average delay time constant of human optic nerve processing.

[0107] Then, the peak sensitivity frequency is calculated. To maximize the glare effect, This is typically set to the peak frequency of the curve. Seeking information about Taking the first derivative of and setting it to 0, solve for the extreme points: ; Simplifying, we get: ; Right now: ; Substitute objective constants The physical peak frequency is calculated as follows: The calculation result is consistent with the human brain. wave and The transition frequency bands of the waves are highly compatible, and the system settings are correct. ).

[0108] To ensure that the frequency of chaotic transitions always falls within the highly sensitive region where the human eye can produce a strong physiological response, the definition is... for The half-power bandwidth of the curve, i.e., finding the two cutoff frequencies. and ,satisfy: : Will Substitute into the equation: ; Substitution The equation is transformed into a transcendental equation: ; Solve for the two real roots of this equation using Newton's method of iteration: The low-frequency cutoff point is obtained: ; The high-frequency cutoff point is obtained: ; Finally, calculate the frequency agility bandwidth. : ; The results indicate that the frequency agility bandwidth in Example 1 is... Determined as At this time, the frequency of the strong light flicker In Right now Chaotic transitions occur within a specific range. This range fully covers the Theta, Alpha, and Beta wave frequency bands, which are most sensitive to human vision, ensuring the maximum physiological rejection effect is generated in different pirate individuals.

[0109] The above-mentioned high-intensity light flicker frequency The calculation formula characterizes: when (When the target exhibits painful reactions such as covering their eyes or lowering their head, indicating an effective strike) Approaching 0, at this point The system maintains the basic interference frequency. ;when When the target does not respond and the attack is ineffective, Increase the frequency; at this time, the system executes a strong light strobe. The stroboscopic frequency undergoes large, chaotic jumps within the sensitive physiological frequency band. This unpredictable nonlinear stroboscopic effect utilizes the stroboscopic blindness effect to disrupt the optic nerve's adaptation to the environment, forcing the target to experience disorientation and nausea, thus forming a complementary denial defense against sound wave attacks.

[0110] Specifically, for example, in this embodiment 1, the system sets the basic interference frequency. Frequency agility bandwidth When biofeedback monitoring shows that the biofeedback effectiveness index is as low as At that time, the system activates the strong light chaotic mode. Assume the random optical frequency chaotic factor generated at the current moment... The system calculates the frequency offset as This instantly adjusts the frequency of the strong light emission to This step involves introducing nonlinear random frequency jumps controlled by biofeedback within the frequency band most sensitive to the human eye's visual nerves. This forces the flashing frequency of the intense light to jump irregularly from one sensitive point (e.g., 13Hz) to another (e.g., 8.72Hz) within a very short time. This unpredictable "light frequency scrambling" can effectively disrupt the adaptive adjustment mechanism of the human eye's pupil and retina. By utilizing the flash blindness effect, it induces strong disorientation and physiological vertigo in the target person's brain, thereby constructing a second physiological defense line on top of sound wave rejection.

[0111] S4. Calculate the comprehensive threat index in real time based on the dynamic threat assessment model. ,when Drop below the safety threshold, i.e. If the duration exceeds 30 seconds, the system will automatically stop the denial, generate a defense report, and upload it to the cloud.

[0112] For example, in this embodiment 1, the system sets the safety observation window to 30 seconds to prevent pirates from feigning retreat and then launching a surprise attack. Specifically, for example, after 15 seconds of high-intensity chaotic audio-visual suppression, the visual sensor detects that the target speedboat makes a sharp turn (the heading angle changes by more than 90 degrees) and speeds away, while the pirate crew members are in a prone position. At this time, the system calculates the relative speed. The value becomes negative (away from the target), resulting in a negative threat index. The value rapidly decreased to 0.2 (below the threshold of 0.6). The system maintained lock but did not emit energy. After tracking for 30 seconds, it confirmed that the target had sailed beyond a safe distance of 3 kilometers. The threat was then officially declared neutral, the power module was shut down, and a defense report containing the encounter time, target characteristics, and engagement video was automatically generated and uploaded to the ship-to-shore server.

[0113] Example 2 As a second embodiment of the present invention, such as Figure 2 As shown in Example 1, this example also discloses a system for a ship side intelligent identification and follow-up sound and light denial anti-piracy method, specifically including: a panoramic warning assessment module, a sound and light follow-up locking module, a biofeedback monitoring module, and an adaptive chaos control module.

[0114] The panoramic surveillance and assessment module is used to continuously scan the sea surface using a wide-angle polarized vision sensor, filter out sea surface glare clutter using a polarization difference algorithm, and extract the distance to the target when an abnormal target is detected. Relative approach speed and the angle of attack Then, a comprehensive threat index is calculated based on a dynamic threat assessment model. And based on threat thresholds Determine whether to initiate target locking; The acoustic-optical tracking module is used to, when a target is determined to be a high threat, lock the target based on its real-time location. Calculate the phase delay required for each element of the acoustic phased array. And the spatial distribution of light intensity in the activation region of the high-intensity matrix, using electronic scanning technology to replace mechanical rotation, to establish a sound and light energy beam locking without mechanical lag; The biofeedback monitoring module is used to extract, in a tiered manner, the target's torso defensive posture features (such as covering ears, curling up) and microscopic facial pain expression features (such as closing eyes, frowning) during the denial of attack, and to calculate in real time a biofeedback effectiveness index reflecting the attack effect based on a weighted algorithm. ; The adaptive chaos control module is used to determine the biofeedback efficiency index. The attack strategy is dynamically adjusted. When a decrease in the effectiveness of the attack is detected, a phase chaotic perturbation model and a light flux frequency domain chaotic modulation model are automatically introduced to perform random nonlinear modulation on the acoustic wave phase and the strong light scintillation frequency, and the comprehensive threat index is continuously monitored. The denial will cease once the target has withdrawn and the threat level is below the safety threshold.

[0115] In the specific implementation of Implementation 2 above, firstly, the system solves the problem of target identification under strong light interference through the panoramic warning assessment module. This module utilizes the polarization characteristics of light to accurately filter out clutter and highlight the target outline in the dazzling sea surface light, while simultaneously calculating the comprehensive threat index by combining multi-dimensional motion parameters. This method effectively prevents missed detections due to sea surface reflections or misidentification of ordinary fishing boats as pirates, ensuring the accuracy and timeliness of the defense system's activation. Secondly, through the acoustic-optical tracking module, the system overcomes the physical inertia limitations of traditional mechanical gimbals. Against the high-speed S-shaped maneuvers of pirate speedboats, this module utilizes phased array electronic scanning technology to achieve microsecond-level beam deflection. This method eliminates the hysteresis effect of mechanical motion, ensuring that high-intensity acoustic and optical energy consistently and accurately covers rapidly moving targets, significantly improving the hit rate of dynamic denial. Then, through the biofeedback monitoring module, the system can perceive the pirate's physiological pain state in real time, determining the effectiveness of the current attack. This method quantifies the actual deterrent effect of non-lethal weapons by capturing subtle features of limbs and faces, and adjusts the strike strategy accordingly in real time. Finally, the adaptive chaos control module executes mutated strikes based on the active noise cancellation devices and psychological adaptability that modern pirates may wear. Specifically, when biofeedback indicates that pirates have developed tolerance, the module generates unpredictable acoustic and optical chaotic signals. This method uses randomly changing phase and frequency to disrupt the prediction mechanism of the active noise reduction algorithm and break the adaptability of human senses, thereby significantly enhancing the denial deterrence and forcing pirates to abandon their attack, thus achieving ship defense.

[0116] Furthermore, a skeleton detection algorithm is provided, which uses a regression-based single-stage deep learning model (YOLO-Pose) to extract the skeleton of the target person. Coordinates of key points Specifically, it includes: Network architecture design: The YOLO-Pose model uses CSPDarknet as the backbone feature extraction network and leverages a path aggregation network (PANet) for multi-scale feature fusion. In the detection head section, a design was implemented to meet the specific requirements of this embodiment, including... The decoupled head for each output channel. (Corresponding to the 17 major joints defined in the COCO dataset, covering the nose, eye, ear, shoulder, elbow, wrist, hip, knee, and ankle); Each anchor box directly outputs a high-dimensional vector. .

[0117] Output vector definition: For each grid cell on the feature map, the network outputs the target vector. The definition is as follows: ; In the formula, The coordinates of the center and the width and height of the bounding box; The target has a confidence level; For category probabilities (mainly "personnel" in this case); For the first Normalized coordinate offset of each key point; For the first The visibility score of each key point. In this embodiment During the steps, if If the area is obscured (e.g., a pirate hiding behind the ship's side), it is considered to be obscured and will not be included in the defensive posture calculation.

[0118] Loss function construction: To ensure the accuracy of keypoint localization, the model training uses object keypoint similarity as the core component of the loss function. Total Loss Function Represented as: ; Among them, keypoint regression loss Specifically defined as: ; In the formula, This represents the number of positive samples (i.e., targets labeled "pirates" in the image); Key points for prediction Euclidean distance between the actual labeled point and the actual labeled point; This is a scaling factor for the target scale, used to balance the weights of large targets (nearby pirates) and small targets (distant pirates), ensuring that the algorithm has scale invariance; For the first The normalization constants of key points reflect the ease of labeling the anatomical region, for example, the "eye". The value is usually smaller than "hip" because eye position is more precise and allows for less error; It is a tiny constant; This serves as a visibility marker for key points.

[0119] Inference and Decoding: On the edge computing terminal of Example 1, the system performs non-maximum suppression (NMS) on the model output. When the keypoint coordinates of the output... After inverse coordinate transformation and mapping back to the original image dimensions, the desired dimensions obtained in the aforementioned steps are obtained. Then enter to In the defensive posture calculation module.

[0120] Furthermore, to verify the robustness of the skeleton detection algorithm under harsh sea conditions and to ensure the long-term stable operation of the system on the shipborne embedded terminal, this embodiment adopts the following training strategy and hardware deployment architecture: Base samples: 80,000 human portrait samples from the COCOKeypoint dataset were selected as the general feature base.

[0121] Specific Enhancement Samples: 45,000 real sea state video frames were collected and labeled, specifically covering scenarios of "low illumination (<5 Lux)," "strong water surface reflection," and "ship hull obstruction." Mosaic-9 data augmentation (9-image stitching to simulate complex backgrounds) and Gamma correction randomization were introduced during the training preprocessing stage. This forces the model to learn structured skeleton features rather than relying on local textures, thereby ensuring that the detection rate of human poses is not less than 85% in nighttime or hazy environments.

[0122] Edge hardware acceleration implementation: The algorithm is deployed on an edge computing terminal based on NVIDIA Jetson AGX Orin. The TensorRT engine is used to perform operator fusion and INT8 mixed-precision quantization on the trained YOLO-Pose model, ensuring accurate regression error at key points. Under the premise of pixel, the inference time of a single frame is compressed to 22ms, which meets the timeliness requirements of real-time dynamic locking.

[0123] Hardware-level random number seed management mechanism: Regarding the chaotic perturbation generation involved in steps S33 and S34 of Example 1, to prevent waveform repetition (i.e., "false chaos") caused by the short period of the software pseudo-random number generator (PRNG) or multi-threaded concurrent competition, the system adopts a two-layer seed management strategy consisting of a hardware entropy source and thread-independent operation. Entropy source acquisition: Thermal noise is collected using the onboard hardware true random number generator (TRNG) as the initial high entropy seed, abandoning the traditional approach of relying solely on the system clock as the seed, and physically eliminating the possibility of cracking chaotic sequences by exhaustively searching for timestamps.

[0124] Thread isolation: At the operating system level, each independent thread responsible for sound wave control and strong light modulation is allocated a random state register with thread-local storage (TLS). At system startup, the main control program injects independent, unrelated seeds into the PRNG state machine of each thread via TRNG, ensuring that the chaotic sequences of the sound and light dual channels are mathematically strictly orthogonal, do not interfere with each other, and are unpredictable.

[0125] In the description of this specification, references to terms such as "an embodiment," "example," "specific example," etc., indicate that a specific feature, structure, material, or characteristic described in connection with that embodiment or example is included in at least one embodiment or example of the invention. In this specification, the illustrative expressions of the above terms do not necessarily refer to the same embodiment or example. Furthermore, the specific features, structures, materials, or characteristics described may be combined in any suitable manner in one or more embodiments or examples.

[0126] The preferred embodiments of the invention disclosed above are merely illustrative of the invention. These preferred embodiments do not exhaustively describe all details, nor do they limit the invention to the specific implementations described. Clearly, many modifications and variations can be made based on the content of this specification. This specification selects and specifically describes these embodiments to better explain the principles and practical applications of the invention, thereby enabling those skilled in the art to better understand and utilize the invention.

Claims

1. A ship side intelligent identification servo-type acousto-optic anti-piracy method, characterized in that, include: The sea surface is scanned by a wide-angle polarization vision sensor. The polarization difference algorithm is used to filter out sea surface glare clutter and extract the target's distance, relative speed and angle of attack. A dynamic threat assessment model is constructed to calculate the comprehensive threat index. If the index exceeds the preset threat threshold, it is determined to be a high-threat target and locking is initiated. Based on the azimuth angle of high-threat targets, the phase delay and initial weight vector required for each element of the acoustic phased array are calculated to form a highly directional acoustic beam. At the same time, the activation region of the strong light matrix is ​​calculated and the light intensity spatial distribution function is generated. Mechanically responsive acousto-optic energy beam locking is established through electronic scanning. Initiate hierarchical visual perception, extract torso posture features and facial expression features of high-threat targets, and calculate the biofeedback effectiveness index; If the index indicates a decrease in the effectiveness of the attack, a phase chaotic perturbation model and a light flux frequency domain chaotic modulation model are introduced to randomly jump-modulate the acoustic wave phase and the strong light flicker frequency. Real-time monitoring of the comprehensive threat index; when the target moves away and the comprehensive threat index remains below the security threshold for a preset time window, the denial process stops and the defense data is archived.

2. The ship side intelligent identification follow-up type acousto-optic anti-piracy method according to claim 1, characterized in that, The expression for the dynamic threat assessment model is: ； In the formula, for The overall threat index at any given moment. For the target distance, For relative velocity, This is a preset normalization constant for the collision hazard time. To prevent tiny constants with a denominator of zero, From the angle of attack for high-threat targets. To determine the confidence level for target classification based on visual and voiceprint recognition. , , These are the weighting coefficients for each item. This is the distance attenuation factor.

3. The intelligent identification and follow-up acoustic and optical deterrence method for anti-piracy on the ship's side as described in claim 1, characterized in that, The formula for calculating the initial weight vector is as follows: ； In the formula, For the first The initial weight vector of each transducer To normalize the maximum emission amplitude, Represents the imaginary unit. No. The phase delay required for each array element relative to the reference array element The pointing angle of the acoustic beam.

4. The ship hull-side intelligent identification and follow-up acoustic and optical deterrence anti-piracy method according to claim 3, characterized in that, The expression for the spatial distribution function of light intensity is: ; In the formula, Represents the spatial distribution function of light intensity. The normalized reference luminous intensity required to reach the target location, Atmospheric transmittance, The maximum physical luminous intensity allowed by the hardware. The physical coordinates on the high-intensity light matrix panel. Azimuth of high-threat targets The center coordinates mapped onto the high-intensity light panel. This is the beam divergence angle coefficient.

5. The intelligent identification and follow-up acoustic and optical deterrence method for anti-piracy on the ship's side according to claim 4, characterized in that, The formula for calculating the biofeedback efficacy index is as follows: ； In the formula, The biofeedback efficacy index The posture defense index, Facial pain index, Let be the weighting coefficients for each item. Let be the distance confidence function. This represents the theoretical maximum limit of facial pain response.

6. The intelligent identification and follow-up acoustic and optical deterrence method for anti-piracy on the ship's side according to claim 5, characterized in that, The expression for the phase chaotic perturbation model is: ； In the formula, The adjusted phased array weight vector. This is the adaptive phase perturbation term.

7. The intelligent identification and follow-up acoustic and optical deterrence method for anti-piracy on the ship's side according to claim 6, characterized in that, The adaptive phase perturbation term The calculation formula is: ； In the formula, For acoustic chaos factors, For maximum modulation depth, This represents the total system time delay.

8. The intelligent identification and follow-up acoustic and optical deterrence method for anti-piracy on the ship's side according to claim 5, characterized in that, The expression for the frequency domain chaotic modulation model of optical flux is: ； In the formula, This is the frequency of strong light flicker. Based on the interference frequency, For frequency agile bandwidth, It is the optical frequency chaos factor.

9. The intelligent identification and follow-up acoustic and optical deterrence method for anti-piracy on the ship's side according to claim 1, characterized in that, The polarization difference algorithm is used to filter out sea surface glare clutter, specifically based on the calculation of the linear polarization degree spectrum using Stokes vectors. and polarization angle spectrum And calculate the enhanced image after flare removal. The enhanced image The calculation formula is: ； In the formula, The original total light intensity, To prevent the minimum value where the denominator is zero, As a glare inhibitor, The maximum linear polarization degree calculated for the current frame. This is a non-linear adjustment coefficient; A direction-selective mask is used to weighted suppress the horizontal polarization characteristics of light reflected from the sea surface.

10. A system for a ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method, employing the ship hull-side intelligent identification and follow-up acoustic and optical denial anti-piracy method as described in any one of claims 1 to 9, characterized in that... include: The panoramic warning and assessment module is used to scan the sea surface with a wide-angle polarized vision sensor, use polarization difference algorithm to filter out glare clutter and extract target motion parameters, and calculate the comprehensive threat index to determine whether to activate the lock. The acoustic-optical follow-up locking module is used to calculate the phase delay of each element of the acoustic phased array and the spatial distribution of the light intensity of the strong light matrix based on the target orientation, and to establish a mechanically hysteresis-free acoustic-optical energy beam locking through electronic scanning. The biofeedback monitoring module is used to extract the torso defensive posture features and facial pain expression features of the target during the strike, and to calculate the biofeedback effectiveness index in real time. The adaptive chaos control module is used to dynamically adjust the attack strategy based on the biofeedback effectiveness index. When the attack effect decreases, it introduces a phase chaos perturbation model and a light flux frequency domain chaos modulation model to randomly modulate the acoustic wave phase and the strong light frequency, and determines whether to stop the denial based on the comprehensive threat index.

Images