Target identification-based self-adaptive hidden control method and system for ocean monitoring platform

Abstract

The invention discloses a self-adaptive concealment control method and system for a marine monitoring platform based on target recognition, and belongs to the technical field of marine monitoring. The method comprises the steps that sea surface images are collected in real time through a wide-angle camera; operating the lightweight target detection model to perform threat target identification, and outputting a target bounding box coordinate, a target type label and a threat confidence coefficient; in response to the threat triggering condition being satisfied, calculating a threat level according to the threat confidence and the estimated target distance, and determining a buoyancy adjustment amount, a target depth and a silence duration based on the threat level; the gear transmission mechanism is driven to drive the parallel injector group to suck seawater to realize rapid diving; according to the marine monitoring platform, the problem that an existing marine monitoring platform lacks autonomous threat perception and active hiding and avoiding capabilities is solved, the detection accuracy is not lower than 90%, the peak power consumption is not higher than 10 W, and the silent state is maintained for 72 hours or longer.

Description

Technical Field

[0001] This invention relates to the field of marine monitoring technology, specifically to an adaptive concealment control method and system for marine monitoring platforms based on target recognition. Background Technology

[0002] With the increasing strategic importance of the ocean globally and the rapid development of the marine economy, marine monitoring technology is playing an increasingly important role in military defense, ecological protection, and resource exploration. Traditional marine monitoring mainly relies on fixed buoy systems. While these systems can provide stable data acquisition services under normal conditions, they face severe survivability challenges when operating in complex marine environments, especially sensitive areas. As a crucial node in the maritime situational awareness network, the stealth and survivability of marine monitoring platforms directly impacts the effectiveness and reliability of the overall monitoring system.

[0003] Existing marine buoy systems suffer from the following technical deficiencies: First, their stealth capabilities are severely inadequate. Traditional buoys employ a passive floating design, remaining continuously exposed on the sea surface, making them highly susceptible to detection by satellite reconnaissance, shipborne radar, and visual observation. According to actual sea area test data, the capture rate of traditional buoys in challenging environments exceeds 90%, with an average time of less than 10 minutes from detection to dismantling, making it difficult to perform long-term monitoring missions in sensitive waters. The exposure problem of traditional buoys stems primarily from their fixed surface-floating state, lacking the ability to actively adjust buoyancy to achieve submersion and stealth. Second, they lack autonomous threat perception capabilities. Existing buoy systems typically lack environmental awareness, failing to identify approaching potential threats and passively awaiting human intervention, resulting in response times of several hours, which cannot meet rapid response requirements. Even systems equipped with simple sensors are limited to environmental parameter collection, lacking target detection and threat identification capabilities. Third, they lack proactive avoidance capabilities. Even if some systems can receive remote control commands, the actual avoidance effect is limited due to the susceptibility of communication links to interference and latency. In remote control mode, the end-to-end latency from threat detection to command issuance to platform response typically exceeds 30 minutes, making it difficult to deal with rapidly approaching threat targets.

[0004] To address these issues, international research has been conducted. The U.S. Defense Advanced Research Projects Agency (DARPA) proposed a buoyancy-based payload project that developed a covert monitoring node with diving capabilities. However, this solution uses a traditional motor-driven buoyancy adjustment system with a power consumption exceeding 50W. Under limited battery capacity constraints, its endurance is severely limited, and the noise and electromagnetic signals generated by the motor easily reveal the platform's location. While the deep-sea buoy system developed by the Japan Agency for Marine-Earth Science and Technology (JAMSTEC) possesses strong diving capabilities, its design objective is deep-sea scientific exploration rather than covert monitoring, lacking threat perception and autonomous decision-making capabilities. Some European research institutions have explored buoyancy adjustment schemes based on ballast tanks, but these ballast tanks are large and slow to respond, making them unsuitable for rapid covert operations.

[0005] In the prior art, Chinese invention patent CN114596335A discloses an unmanned surface vessel (USV) target detection and tracking method and system. This scheme replaces the YOLOv4 backbone feature extraction network CSPDarknet53 with the MobileNetv3 structure to achieve model lightweighting, and integrates the KCF correlation filter target tracking algorithm to achieve scale-adaptive tracking of sea surface targets. This technical solution reduces the number of backbone network parameters through depthwise separable convolutions, and adds a CBAM attention mechanism in MobileNetv3 and between PANet and YOLO_HEAD to improve detection accuracy. However, this technical solution has the following limitations: its core objective is to track and follow targets rather than to avoid threats and achieve self-concealment, and its system design philosophy is fundamentally different from the needs of covert monitoring; its system architecture is designed for unmanned surface vessels to actively track scenarios, and does not involve buoyancy adjustment and diving concealment mechanisms, so it cannot achieve active concealment of the platform; its model parameter count and computational complexity are still high. Even after lightweight improvements, the YOLOv4 framework still has more than 20MB of model parameters, making it difficult to deploy on ultra-low power embedded platforms to achieve long-term autonomous operation; its KCF tracking algorithm requires continuous operation and consumes computing resources, making it unsuitable for covert monitoring scenarios that require long-term silent standby.

[0006] Therefore, there is an urgent need for a control method for marine monitoring platforms that integrates threat perception, autonomous decision-making, and proactive concealment capabilities to solve the technical problems of insufficient concealment and lack of autonomous threat perception and proactive avoidance capabilities in existing technologies. Summary of the Invention

[0007] In view of the above-mentioned technical problems in the existing technology, the present invention provides an adaptive covert control method and system for marine monitoring platforms based on target recognition, which aims to realize autonomous threat perception and adaptive covert control of marine monitoring platforms.

[0008] The present invention adopts the following technical solution: An adaptive covert control method for a marine monitoring platform based on target recognition includes: a sea surface image acquisition step, in which sea surface images are acquired in real time using a wide-angle camera and transmitted to an embedded processor; a threat target identification step, in which the embedded processor runs a lightweight target detection model to identify threat targets in the sea surface images and outputs the target bounding box coordinates, target type label, and threat confidence; a covert decision generation step, in response to a threat confidence greater than a preset confidence threshold and a target type label belonging to a preset threat type set, a threat level is calculated based on the threat confidence and estimated target distance, and a buoyancy adjustment amount and target depth are determined based on the threat level, generating a covert trigger signal; a buoyancy adjustment execution step, in response to the covert trigger signal, a gear transmission mechanism drives a parallel injector assembly to draw in seawater, causing the marine monitoring platform to dive to the target depth; and a silent state maintenance step, in which an active signal source is turned off, a silent duration is determined based on the threat level, and in response to the expiration of the silent duration and the absence of a sea surface threat, the platform is controlled to surface and recover.

[0009] Compared with existing technologies, this invention has the following advantages: It achieves real-time identification of threat targets through a lightweight target detection model, with a detection accuracy of no less than 90% and a single-frame detection time of no more than 200ms; it achieves graded response based on threat level through threat level calculation and adaptive buoyancy adjustment, avoiding overreaction or underreaction; it achieves low-power buoyancy adjustment through a gear-driven injector mechanical structure, with peak power consumption not exceeding 10W and a diving time of no more than 3 seconds; it achieves long-term concealment through silent state maintenance, with standby power consumption not exceeding 3W, and can maintain a silent state for more than 72 hours; the overall concealment survival rate reaches over 90%, significantly better than traditional buoy systems. Attached Figure Description

[0010] Figure 1 A flowchart illustrating the adaptive concealment control method for a marine monitoring platform based on target recognition provided in an embodiment of the present invention; Figure 2 A structural block diagram of an adaptive covert control system for a marine monitoring platform based on target recognition, provided in an embodiment of the present invention; Detailed Implementation

[0011] Please refer to the attached document. Figures 1-2 The technical solution of the present invention will now be described in detail with reference to the accompanying drawings and embodiments.

[0012] like Figure 1 As shown, the adaptive concealment control method for marine monitoring platforms based on target recognition provided in this embodiment of the invention includes the following steps: Step S1: Sea Surface Image Acquisition Step. The sea surface image acquisition module 1 acquires sea surface images in real time using a wide-angle camera, providing raw data for subsequent threat target identification. The wide-angle camera is installed on the top hatch of the marine monitoring platform, approximately 20 centimeters above the sea surface, ensuring a wide field of view. In this embodiment, the wide-angle camera uses an OpenMV H7 series embedded vision module, equipped with a wide-angle lens with a field of view of 140 degrees, capable of covering a large area of ​​sea around the platform.

[0013] The wide-angle camera's frame rate is set to 30 frames per second, and the image resolution is set to 1280×720 pixels. Choosing a frame rate of 30 frames per second ensures real-time performance while controlling the data processing load; for sea surface target detection tasks, 30 frames per second is sufficient to capture target motion changes. The 1280×720 pixel resolution strikes a balance between clarity and processing efficiency, providing sufficient detail for target recognition without placing an excessive computational burden on the embedded processor. The acquired sea surface images are stored in RGB format, with each frame containing approximately 2.76MB of data, and are transmitted in real-time to the embedded processor of the threat target recognition module 2 via a USB interface.

[0014] The camera module features a waterproof, sealed design with an IP68 protection rating, enabling long-term stable operation in marine environments. A hydrophobic coated glass front for the lens effectively prevents seawater splashes and salt spray from affecting image quality. The camera supports automatic exposure and automatic white balance, adapting to various lighting conditions from strong sunlight to overcast skies and light fog. In nighttime or extremely low light conditions, the camera automatically switches to a high-sensitivity mode, maintaining basic imaging capabilities by increasing exposure time and gain, although image noise will significantly increase in this situation.

[0015] Step S2: Threat Target Identification. Threat target identification module 2 runs a lightweight target detection model to identify threat targets in the sea surface image, achieving intelligent conversion from raw image to structured threat information. The core of this step is to achieve efficient and accurate target detection on a resource-constrained embedded platform, which requires the target detection model to strike a balance between computational complexity and detection accuracy.

[0016] In this embodiment, the lightweight object detection model uses the YOLOv8n architecture as its basic framework. YOLOv8 is a well-known single-stage object detection algorithm, designed to complete object localization and classification through a single forward propagation. YOLOv8n is the smallest model in the YOLOv8 series, with an original model parameter size of approximately 3.2MB. To further reduce model complexity, this embodiment employs depthwise separable convolution technology for model compression. Depthwise separable convolution is a well-known convolution decomposition technique. Its basic idea is to decompose standard convolution into two steps: channel-wise convolution and pointwise convolution, thereby reducing the number of parameters and computational load to achieve model lightweighting. After depthwise separable convolution compression, the object detection model parameter size in this embodiment is compressed to no more than 10MB, enabling deployment and operation on embedded platforms with limited memory.

[0017] The model's inference on the embedded processor employs fixed-point quantization optimization. After converting floating-point operations to 8-bit integer operations, the single-frame detection time is reduced from the original 400 milliseconds to no more than 200 milliseconds, meeting real-time detection requirements. The embedded processor used is a Raspberry Pi 4B development board, equipped with a Broadcom BCM2711 quad-core ARM Cortex-A72 processor with a main frequency of 1.5GHz, and 4GB of LPDDR4 memory, enabling smooth operation of the lightweight target detection model. The detection accuracy reaches over 90% on the standard maritime target test set, with a false positive rate controlled below 5%.

[0018] The model's training dataset contains over 3000 labeled images, covering various sea conditions and target types to ensure good generalization ability. Training data sources include publicly available maritime target detection datasets, a self-built nearshore real-world image dataset, and synthetic data generated through data augmentation. Data augmentation techniques include random cropping, horizontal flipping, brightness adjustment, contrast adjustment, Gaussian blur, and motion blur to simulate image features under different weather, lighting, and sea condition conditions. The training process employs a transfer learning strategy, using YOLOv8n weights pre-trained on the COCO dataset as initial parameters, fine-tuned on the maritime target dataset, and training for 100 epochs with a batch size of 16. The learning rate is gradually decayed from 0.01 to 0.0001 using cosine annealing.

[0019] The output of the object detection model contains three types of information: the target bounding box coordinates, which represent the position and size of the target in the image in the form of a quadruple, including the x-coordinate of the top-left corner of the bounding box, the y-coordinate of the top-left corner, the width w of the bounding box, and the height h of the bounding box, with coordinate values ​​in pixels; the target type label, which distinguishes different target categories in the form of integer codes, including military vessels (coded as 1), patrol boats (coded as 2), law enforcement vessels (coded as 3), civilian fishing vessels (coded as 4), commercial cargo ships (coded as 5), etc., defining a total of 12 target categories covering common maritime targets; and the threat confidence, which represents the model's degree of confidence in the detection results in the form of a floating-point number, with a value ranging from 0 to 1. The higher the confidence, the more confident the model is in the detection results.

[0020] The model employs a multi-scale feature fusion strategy during detection. The YOLOv8n backbone network outputs three feature maps at different resolutions, corresponding to 1 / 8, 1 / 16, and 1 / 32 scales of the original image, respectively. For small, distant targets, high-resolution low-level features provide detailed spatial localization information, while the 1 / 8 scale feature map retains more edge and texture details. For large, near targets, low-resolution high-level features provide rich semantic information, while the 1 / 32 scale feature map contains more abstract target category features. The feature pyramid network fuses features from different levels through top-down and bottom-up path aggregation, enabling the detector to utilize both detailed and semantic information simultaneously. This strategy effectively solves the challenge of detecting targets with large scale variations in ocean scenes, maintaining stable detection performance regardless of whether the target is far or near.

[0021] In the post-processing stage, the model employs non-maximum suppression (NMS) to remove redundant detection boxes. This algorithm is a well-known technique in object detection; its basic logic is that when multiple detection boxes correspond to the same target, the detection box with the highest confidence is retained, while other detection boxes with an overlap exceeding a preset threshold are suppressed. The overlap threshold for NMS is set to 0.5, striking a balance between avoiding missed detections and suppressing duplicate detections. After NMS, the detection results output per frame typically do not exceed 10 targets, reducing the processing burden on subsequent decision-making modules.

[0022] Step S3: Covert Decision Generation Step. The covert decision generation module 3 receives the target bounding box coordinates, target type label, and threat confidence level output by the threat target identification module 2, and generates a covert control decision according to preset decision rules. The decision generation process includes four stages: threat determination, threat level calculation, buoyancy adjustment determination, and silence time determination. Each stage is executed sequentially to form a complete decision chain.

[0023] The threat assessment process first checks whether the threat confidence level exceeds a preset confidence threshold. In this embodiment, the baseline value of the preset confidence threshold is set to 0.7. This threshold ensures detection reliability while controlling the false alarm rate. Setting the threshold too low will lead to a large number of false alarms, causing the platform to frequently perform unnecessary dives, consuming battery power and increasing mechanical wear; setting the threshold too high will lead to missed alarms, preventing the platform from taking cover in time when a threat approaches. After extensive simulation tests and verification in actual sea areas, the threshold of 0.7 achieves a good balance between the false alarm rate and the missed alarm rate, with a false alarm rate of approximately 5% and a missed alarm rate of approximately 3%.

[0024] Preferably, the preset confidence threshold is dynamically adjusted based on the current sea state light intensity. Light intensity is indirectly reflected by the camera's automatic exposure parameters; a longer exposure time indicates weaker ambient light. When the sea state light intensity is lower than the preset light reference value, image quality deteriorates, leading to a general decrease in model confidence. In this case, the threshold needs to be lowered to maintain detection sensitivity. Specifically, the dynamic confidence threshold is calculated according to the following formula: , in: The confidence threshold is dynamically adjusted, with its value limited to between 0.5 and 0.85 to prevent the threshold from being too low or too high, which could affect system performance. The unit is a dimensionless numerical value; The baseline confidence threshold represents the confidence level determination threshold under standard lighting conditions, with a value of 0.7. The unit is a dimensionless numerical value; The light sensitivity coefficient represents the threshold adjustment amount corresponding to a unit change in light intensity. Its value ranges from 0.001 to 0.005; in this embodiment, it is set to 0.002. The unit is the change in threshold per lux; The preset light reference value corresponds to the standard light intensity at noon on a clear day, and is approximately 1000 lux in reciprocal form of the exposure time. The unit is lux; The current sea state light intensity is calculated from the camera's exposure parameters and expressed as the reciprocal of the exposure time. The unit is lux. The calculation logic of the above formula is: when the current light intensity... Below reference light intensity hour, For positive values, the confidence threshold is... The corresponding reduction improves detection sensitivity under low light conditions; when the current light intensity is higher than or equal to the reference light intensity, If the value is zero or negative, the confidence threshold remains unchanged or slightly increases. This dynamic adjustment mechanism ensures stable detection performance under different lighting conditions, avoiding significant fluctuations in detection sensitivity due to changes in lighting.

[0025] The threat assessment process simultaneously checks whether the target type label belongs to a preset threat type set. In this embodiment, the preset threat type set includes military vessels, patrol boats, and law enforcement vessels, corresponding to type codes 1, 2, and 3, but excludes civilian fishing vessels and commercial cargo ships. This setting aims to avoid overreacting to conventional civilian vessels, conserve platform energy, and reduce unnecessary stealth maneuvers. The subsequent threat level calculation process only proceeds when both the threat confidence and target type meet the trigger conditions. If multiple targets are detected in the same frame, the process proceeds to the next stage as soon as any one of them meets the trigger conditions, with the target having the highest threat level used as the decision-making basis.

[0026] The threat level calculation process determines the threat level based on a combination of threat confidence and estimated target distance. First, the estimated target distance is calculated based on the target's bounding box coordinates. Target distance estimation employs the principle of monocular vision ranging, using the target's image size in the image to infer its distance from the platform. The specific calculation formula is as follows: , in: To estimate target distance, the horizontal distance between the threatening target and the marine monitoring platform is expressed in meters; The actual height of the preset target represents the vertical distance from the waterline to the top of the mast of the threatening vessel. Based on statistics of typical threatening targets, the typical height is taken as 10 meters for military vessels, 5 meters for patrol boats, and 6 meters for law enforcement vessels. The unit is meters; The focal length is the equivalent focal length of the camera's optical system. In this embodiment, the wide-angle camera has an equivalent focal length of 3.5 mm. The unit is millimeters; The height h is the pixel height of the target bounding box, extracted from the bounding box coordinate quadruple output by the detection model. It represents the number of vertical pixels the target occupies in the image. The unit is pixels; The physical size corresponding to a single pixel is determined by both the image sensor size and the image resolution. In this embodiment, the sensor size is 1 / 2.5 inches, meaning the sensor diagonal length is approximately 10.2 millimeters, and the resolution is 1280×720 pixels. The calculated... It is approximately 4.8 micrometers, or 0.0048 millimeters. The unit is millimeters per pixel. The calculation logic of the above formula is based on the geometric relationship of perspective projection: the actual height of the target. With focal length The product of and equals the target distance At the target imaging height The product of these can be used to infer the target distance. As the target approaches, its pixel size in the image... The distance is increased, and the distance between the target and the platform can be calculated accordingly. The estimation error of this method is proportional to the deviation from the actual height of the target. When the deviation between the actual height of the target and the preset value is within 20%, the distance estimation error can be controlled within 15%.

[0027] After obtaining the estimated target distance, a threat level score is calculated. The threat level score considers both threat confidence and target distance; the higher the confidence level and the closer the distance, the higher the threat level. The specific calculation formula is as follows: , in: The threat level score represents a comprehensive assessment of the threat posed by a target to the marine monitoring platform. The score ranges from 0 to 1, where 0 indicates no threat and 1 indicates the highest threat. The unit is a dimensionless numerical value; Threat confidence is the target detection confidence value output by the threat target identification module. It represents the degree of confidence the target detection model has in the detection results, and its value ranges from 0 to 1. The unit is a dimensionless numerical value; To estimate the target distance, the distance is calculated using the aforementioned target distance estimation formula, with the unit being meters; The maximum effective detection distance represents the farthest distance at which the target detection model can reliably detect threatening targets. The value ranges from 800 to 1200 meters; in this embodiment, it is set to 1000 meters. Targets beyond this distance are considered not to pose a threat at this time. The unit is meters; The distance attenuation index controls the weight of distance factors on threat level, with a value ranging from 1.2 to 1.8. In this embodiment, a value of 1.5 is used. This parameter controls the degree of influence of distance on threat level; the larger the value, the more significant the increase in the level of near-range threats. The unit is a dimensionless numerical value. The calculation logic of the above formula is as follows: first, calculate the distance factor. The distance factor is 0 when the target distance equals the maximum effective detection distance, and 1 when the target distance is 0; then the distance factor is adjusted. The exponentiation operation makes the influence of distance non-linear, giving higher weight to targets at closer range; finally, it is multiplied by the threat confidence score to obtain the comprehensive threat level score. When the target distance is equal to the maximum effective detection distance, the threat level score is 0 regardless of the confidence level; when the target distance is 0, the threat level score is equal to the threat confidence score itself.

[0028] The buoyancy adjustment calculation stage determines the required buoyancy adjustment based on the threat level score, which is the volume of seawater the injector assembly needs to draw in. The buoyancy adjustment determines the platform's diving depth; the higher the threat level, the deeper the dive is required to enhance stealth. The specific calculation formula is as follows: , in: The total buoyancy adjustment amount represents the total volume of seawater that the syringe assembly needs to draw in, in milliliters. The basic buoyancy adjustment value represents the minimum volume of seawater the syringe assembly needs to draw in under the lowest threat level conditions, corresponding to the minimum diving depth required. Its value ranges from 300 to 400 ml; in this embodiment, it is 350 ml. The unit is milliliters; This is an adaptive adjustment coefficient, representing the maximum additional water intake when the threat level changes from 0 to 1. It is used to dynamically adjust the diving depth according to the threat level, and its value ranges from 500 to 700 ml. In this embodiment, a value of 600 ml is used. The unit is milliliters; The threat level score is obtained from the aforementioned threat level calculation formula, and its value ranges from 0 to 1. The unit is a dimensionless numerical value. The calculation logic of the above formula is as follows: the total buoyancy adjustment consists of two parts, the basic buoyancy adjustment... Ensure the platform descends to at least the minimum safe depth whenever any covert maneuver is triggered, with adaptive adjustment. The buoyancy adjustment is dynamically increased based on the threat level to achieve deeper dives. According to this formula, when the threat level score is 0, the buoyancy adjustment is 350 ml, corresponding to a diving depth of approximately 1 meter; when the threat level score is 1, the buoyancy adjustment is 350 + 600 = 950 ml, corresponding to a diving depth of approximately 3 meters. This formula ensures that shallow dives are performed at low threat levels to save energy and shorten recovery time, while deeper dives are performed at high threat levels to enhance concealment.

[0029] The target depth is calculated based on the buoyancy adjustment and the platform's buoyancy characteristics. The platform employs a sealed cabin design and initially operates in zero buoyancy equilibrium, meaning gravity equals buoyancy, and the platform floats on the sea surface. When seawater is drawn in by the syringe, the platform's total weight increases, gravity exceeds buoyancy, and the platform begins to submerge. The submersion depth is approximately proportional to the amount of water drawn in, with the proportionality coefficient determined by the platform's buoyancy characteristics. In this embodiment, the platform's cross-sectional area is approximately 0.1 square meters. Drawing in 350 ml of seawater generates approximately 3.5 N of negative buoyancy. According to Newton's second law and fluid resistance balance calculations, the final stable submersion depth is approximately 1 meter; drawing in 950 ml of seawater generates approximately 9.5 N of negative buoyancy, resulting in a final stable submersion depth of approximately 3 meters. The actual submersion depth is also affected by seawater density, platform attitude, and water flow. Real-time monitoring by depth sensors and closed-loop control ensure that the target depth is reached.

[0030] The silence time determination process calculates the silence duration based on the threat level score, which is the length of time the platform remains silent underwater. The silence duration needs to be long enough to ensure the threat target leaves the detection range, but not too long to avoid delaying normal monitoring tasks. The specific calculation formula is as follows: , in: The silence duration refers to the total time that the marine monitoring platform remains silent underwater, expressed in hours. The preset minimum silence time represents the shortest duration for which the platform needs to remain silent in low-threat scenarios. The value ranges from 12 to 24 hours; this embodiment uses 24 hours. The unit is hours; The preset maximum silence time represents the longest period the platform needs to remain silent in high-threat scenarios. The value ranges from 72 to 96 hours; this embodiment uses 72 hours. The unit is hours; The threat level score is obtained from the aforementioned threat level calculation formula, and its value ranges from 0 to 1. The unit is a dimensionless numerical value. The calculation logic of the above formula is: the duration of silence is at its minimum value. and maximum value Linear interpolation between the two, threat level score As an interpolation weight, according to the formula, when the threat level score is 0, the silence duration is 24 hours; when the threat level score is 1, the silence duration is 24 + (72 - 24) × 1 = 72 hours. This formula ensures that the silence time is extended at high threat levels to wait for the threat to be completely eliminated, and shortened at low threat levels to restore monitoring capabilities as quickly as possible. The silence time setting also takes into account the patrol cycle and search capabilities of typical threat targets, and the maximum silence time of 72 hours can cover the duration of most search operations.

[0031] After completing the above calculations, the concealment decision generation module 3 outputs the concealment trigger signal, buoyancy adjustment amount, target depth, and silence duration to the subsequent execution modules. All decision parameters are packaged into data frames and sent to the main controller via the onboard serial port, where the main controller coordinates the timing of actions of each execution module.

[0032] Step S4: Buoyancy adjustment execution step. In response to the concealment trigger signal, buoyancy adjustment module 4 executes a buoyancy adjustment action, enabling the marine monitoring platform to rapidly descend to the target depth. This embodiment employs a gear-driven injector with a non-powered mechanical structure, significantly reducing power consumption and noise compared to traditional motor-driven solutions. It also avoids electromagnetic radiation generated during motor operation, enhancing the platform's concealment performance.

[0033] The gear transmission mechanism is designed using a servo motor drive scheme. A servo motor is an angle servo motor capable of precisely rotating to a specified angle and maintaining its position according to a control signal. This embodiment uses a waterproof digital servo motor with a torque of no less than 25 kg·cm and an operating voltage range of 6V to 12V; this embodiment uses a 12V power supply to achieve maximum torque output. The servo motor output shaft is connected to the drive gear, and the drive gear meshes with the driven gear to form a gear transmission mechanism. The transmission ratio is set to 1:2 to 1:3; this embodiment uses a 1:2 transmission ratio, meaning that for every 1 revolution of the drive gear, the driven gear rotates 0.5 revolutions. Through gear transmission, the servo motor output torque is amplified to 50 kg·cm to 75 kg·cm; in this embodiment, it is 50 kg·cm, sufficient to drive the syringe piston to overcome seawater resistance and sealing friction. A rack is fixed on the driven gear shaft, converting the rotational motion into linear motion. The rack is mechanically connected to the syringe piston rod, driving the piston to move back and forth.

[0034] The parallel injector assembly consists of three medical-grade injectors connected in parallel. Each injector has a capacity of 350 ml, an inner diameter of approximately 45 mm, and a piston stroke of approximately 220 mm. The three injectors are connected in parallel via a Y-shaped tee pipe, sharing a single seawater inlet and outlet. The seawater inlet and outlet are located at the bottom of the platform and are equipped with solenoid valves to control the direction of seawater flow. Filters prevent sediment and debris from entering the injectors and clogging the pipes. The piston rods of the three injectors move synchronously through a mechanical linkage structure, driven by the same rack, ensuring consistency and synchronicity in the suction and discharge processes. The advantage of the parallel design is increased overall buoyancy adjustment. The total capacity of the three 350 ml injectors reaches 1050 ml, corresponding to a total buoyancy adjustment capacity of approximately 10.5 N, which can meet the platform's diving requirements to a depth of 3 meters.

[0035] The descent process is as follows: The main controller receives the buoyancy adjustment parameters output by the concealment decision generation module 3, converts them into the required rotation angle of the servo motor, and sends them to the servo motor drive board via a PWM signal. Upon receiving the position command, the servo motor begins to rotate, driving the drive gear. The drive gear meshes with the driven gear, transmitting the rotational motion to the driven gear. The rack on the driven gear shaft moves accordingly, pushing the pistons of the three syringes backward synchronously. The backward movement of the pistons creates a negative pressure inside the syringes, allowing seawater to enter through the inlet pipes under atmospheric pressure. As seawater enters, the total weight of the platform increases. When gravity exceeds buoyancy, the platform begins to descend. The descent speed is determined by both the magnitude of the negative buoyancy and fluid resistance. Initially, the descent speed is faster; as the speed increases, fluid resistance increases, eventually reaching speed equilibrium.

[0036] Actual test data shows that the servo motor's no-load speed under 12V power is approximately 0.15 seconds / 60 degrees, and its loaded speed is approximately 0.3 seconds / 60 degrees. Each centimeter movement of the rack corresponds to the piston drawing in approximately 15 milliliters of seawater. To complete the 350 milliliters of water intake, the piston needs to move approximately 23 centimeters, corresponding to a driven gear rotation of approximately 120 degrees and a driving gear rotation of approximately 240 degrees. The servo motor's action time is approximately 1.2 seconds. With the three syringes operating synchronously, the entire water intake process can be completed within 1.2 seconds. The total time from the start of water intake to the platform's descent to the target depth does not exceed 3 seconds, meeting the timeliness requirements for rapid concealment. With a target approaching at a speed of 30 knots, the target only moves approximately 46 meters within 3 seconds, by which time the platform has already completed its descent and entered a concealed state.

[0037] The power consumption characteristics of this mechanical structure offer significant advantages. The servo motor consumes power only during diving and surfacing maneuvers, consuming approximately 8W during each maneuver, lasting about 1.2 seconds, with a single maneuver consuming approximately 2.7mWh. Including the power consumption of the solenoid valve and control circuitry, the total power consumption for a single complete diving maneuver is approximately 5mWh. A complete diving-surfacing cycle consumes approximately 10mWh. Compared to the prior art document CN114596335A, which does not address buoyancy adjustment, and the power consumption of over 50W in traditional motor-driven solutions, this embodiment reduces power consumption by more than 80%. Supported by a 22.2V / 10Ah lithium battery, the platform can perform over 2200 diving-surfacing cycles, meeting long-term deployment requirements. Furthermore, the mechanical structure emits no electromagnetic radiation during operation and does not generate detectable electronic signals, further enhancing stealth capabilities. The sound generated by the servo motor during operation has a limited underwater propagation distance; tests show that it attenuates to below ambient noise levels at a distance of 10 meters, thus not revealing the platform's location.

[0038] The buoyancy adjustment execution module 4 also includes a depth sensor to monitor the actual diving depth and achieve closed-loop depth control. This embodiment uses an MS5837-30BA series pressure sensor with a measurement range of 0 to 30 bar, corresponding to a water depth of 0 to 300 meters. The measurement accuracy is better than ±0.2%, with a measurement error of less than ±6 millimeter within a 3-meter water depth range, and a response time of less than 50 milliseconds. The depth sensor communicates with the main controller via an I2C interface, reporting the current depth value in real time. When the actual depth reaches the target depth, the main controller stops the water intake action, the servo motor maintains its current position, and the platform hovers at the target depth. If there is a deviation between the actual depth and the target depth, the main controller compensates by fine-tuning the servo motor angle to ensure the platform remains stably hovered at the target depth.

[0039] Step S5: Silence Maintenance Step. After the marine monitoring platform submerges to the target depth, the silence maintenance module 5 takes over platform control and is responsible for maintaining underwater silence until the threat is eliminated. The achievement of silence includes four stages: active signal source shutdown, depth maintenance monitoring, threat elimination detection, and surfacing trigger control. These stages work together to ensure the platform remains safely concealed during silence and resumes normal functioning at the appropriate time.

[0040] The active signal source shutdown process cuts off all electromagnetic radiation sources that could potentially reveal the platform's location. In this embodiment, the active signal sources include a LoRa wireless communication module, a BeiDou / GPS dual-mode satellite positioning module, and a UHF data transmission module. The LoRa module operates in the 433MHz or 868MHz frequency band and is used for data communication with shore base stations or other monitoring nodes. Its transmission power is approximately 100mW, and its communication distance can reach tens of kilometers. However, it is also a signal source that can be located by radio direction finding equipment. Although the BeiDou / GPS module primarily receives satellite signals, its local oscillator circuit and digital processing circuit also generate weak electromagnetic leakage. The UHF module is used for short-range, high-speed data transmission, with a transmission power as high as 1W. After shutting down these modules, the platform enters an electromagnetic silence state, not actively emitting any radio frequency signals, significantly reducing the probability of being detected by radio direction finding equipment.

[0041] In silent mode, the platform retains only essential sensors and control circuitry in low-power standby mode. Essential sensors include a depth sensor for depth monitoring, a temperature sensor for battery protection, and an accelerometer for attitude monitoring. The control circuitry includes the main controller, servo drive board, and power management module. The main controller enters low-power sleep mode, disabling unnecessary peripheral clocks and retaining only timers and interrupt wake-up functions, reducing standby power consumption to the milliwatt level. The entire platform's standby power consumption is controlled below 3W, primarily due to depth sensor sampling, main controller sleep maintenance, and power management module losses. This power consumption level allows the platform to maintain a silent state for approximately 74 hours with a 22.2V / 10Ah battery capacity, exceeding the preset maximum silent time requirement of 72 hours.

[0042] The depth monitoring system continuously monitors the platform's depth position using depth sensors to ensure stable hovering during periods of silence. Seawater density varies with depth, temperature, and salinity, potentially causing slight fluctuations in platform buoyancy. Temperature changes affect seawater density; higher temperatures decrease density and buoyancy, while lower temperatures increase density and buoyancy. Salinity changes also affect density; freshwater mixing zones near estuaries have lower salinity and density. Furthermore, tides and currents also exert thrust on the platform. The silence maintenance module 5 reads depth sensor data every 10 seconds and calculates the deviation between the current depth and the target depth. When the depth deviation exceeds a preset threshold of 0.2 meters, the main controller wakes from sleep mode and sends a fine-tuning command to the buoyancy adjustment execution module 4. This fine-tuning is achieved by slightly adjusting the servo angle, drawing in or expelling a small amount of seawater to correct the depth position. Each fine-tuning operation does not exceed 20 milliliters, corresponding to a depth change of approximately 5 centimeters, avoiding excessive energy consumption or significant noise from large movements.

[0043] The threat elimination detection phase assesses the sea surface threat status before the silence period expires. Since the platform is underwater and cannot directly observe the sea surface via cameras, an indirect detection method is employed. This embodiment utilizes an accelerometer to detect sea surface vibrations. When a vessel passes by, the propeller and hull generate low-frequency vibrations in the water, which propagate to the platform and cause changes in the accelerometer readings. The accelerometer collects triaxial acceleration data at a sampling rate of 100Hz, extracts vibration signals in the 1Hz to 20Hz frequency band using a bandpass filter, and calculates the vibration energy spectral density. When the vibration energy exceeds a preset threshold, it is determined that a vessel is present on the sea surface, and the silence state continues; when the vibration energy is below the threshold and the duration exceeds 30 minutes, it is determined that there is no threat on the sea surface. This detection method can perceive the sea surface conditions without exposing the platform, providing a basis for surfacing decisions.

[0044] After the threat is confirmed to be eliminated, the surfacing trigger control system controls the platform to surface and resume normal operation. Surfacing trigger requires two conditions to be met simultaneously: the silence duration has reached the preset value calculated by the concealment decision generation module 3; and the threat elimination detection confirms that there is no threat on the sea surface. After both conditions are met, the silence state maintenance module 5 sends a surfacing command to the main controller, and the main controller sends a surfacing trigger signal to the buoyancy adjustment execution module 4.

[0045] The ascent process is the reverse of the descent process: the main controller sends a reverse position command to the servo motor, which rotates in the opposite direction, driving the gear set to move the pistons of the three syringes forward synchronously, expelling the seawater from the syringes through the outlet pipe. During the dewatering process, the platform's total weight decreases, and when buoyancy exceeds gravity, the platform begins to rise. To avoid excessively rapid ascent that could cause platform instability or create significant splashes on the surface, the dewatering speed is controlled at half the intake speed, resulting in an ascent speed of approximately 0.1 meters per second. Ascending from a depth of 3 meters to the surface takes approximately 30 seconds, during which the main controller continuously monitors depth changes to ensure a smooth and controllable ascent.

[0046] After the platform surfaces, the depth sensor detects a depth value close to zero, and the main controller determines that the platform has surfaced. At this point, the silent state maintenance module 5 sends recovery commands to each functional module, sequentially activating the power module, camera module, communication module, and positioning module. After each module completes its power-on self-test, the platform resumes normal monitoring and communication functions. The sea surface image acquisition module 1 restarts image acquisition, the threat target identification module 2 resumes threat detection, and the entire system returns to step S1, forming a complete closed-loop control process. The platform uploads the status logs from the silent period to the shore base station via the LoRa module for subsequent analysis and system optimization.

[0047] This invention, through the coordinated operation of the five steps described above, achieves autonomous threat perception and adaptive covert control for marine monitoring platforms. Compared with existing technologies, the technical advantages of this invention are reflected in the following aspects: In terms of threat perception, the lightweight target detection model runs in real time on an embedded platform, with a detection accuracy of no less than 90% and a single-frame detection time of no more than 200ms. It can promptly trigger covert actions after a threat target enters the effective range. Compared with the tracking-guided design in the prior art document CN114596335A, this invention achieves threat perception guided by avoidance. In terms of decision generation, the threat level is calculated by comprehensively considering threat confidence and target distance, and the buoyancy adjustment amount and silence time are adaptively determined to avoid overreaction or underreaction. The decision-making process is completely autonomous and requires no manual intervention. In terms of execution efficiency, the peak power consumption of the gear-driven injector's mechanical structure does not exceed 10W, and the diving time does not exceed 3 seconds, meeting the timeliness requirements for rapid covertness. In terms of covertness effect, the standby power consumption after turning off the active signal source does not exceed 3W, and it can maintain a silent state for more than 72 hours, with a covert survival rate of over 90%, which is significantly better than the survival rate of less than 10% of traditional buoy systems.

[0048] like Figure 2 As shown in the figure, the adaptive covert control system for a marine monitoring platform based on target recognition provided in this embodiment of the invention includes a sea surface image acquisition module 1, a threat target recognition module 2, a covert decision generation module 3, a buoyancy adjustment execution module 4, and a silent state maintenance module 5. These modules are interconnected via a data bus and control signals to form a complete closed-loop control system. The system adopts a modular design, with each functional module independently packaged and communicating through standardized interfaces, facilitating maintenance, upgrades, and troubleshooting.

[0049] The sea surface image acquisition module 1 is used to acquire sea surface images in real time via a wide-angle camera, providing raw data input for threat target identification. As described in the method embodiment, the wide-angle camera has a field of view of not less than 120 degrees, an acquisition frame rate of not less than 30 frames per second, and an image resolution of not less than 640×480 pixels. The hardware components of the sea surface image acquisition module 1 include an optical lens, an image sensor, an image signal processor, and a data interface circuit. The optical lens adopts an aspherical design to effectively correct wide-angle distortion, and the edge imaging quality loss is controlled within 10%. The image sensor adopts a back-illuminated CMOS chip, which has high sensitivity and low noise characteristics and supports wide dynamic range imaging. The image signal processor completes automatic exposure, automatic white balance, and image compression functions, and outputs JPEG or H.264 encoded image data. The data interface circuit is connected to the embedded processor of the threat target identification module 2 via a USB or MIPI interface.

[0050] Threat target identification module 2 is used to run a lightweight target detection model to identify threat targets in sea surface images, realizing intelligent conversion from raw images to threat information. As described in the method embodiment, the lightweight target detection model uses a depthwise separable convolutional structure for model compression, with model parameters not exceeding 10MB and single-frame detection time not exceeding 200ms. The hardware core of threat target identification module 2 is an embedded processor. In this embodiment, a Raspberry Pi 4B development board is used, whose 1.5GHz quad-core ARM processor and 4GB of memory can meet the real-time inference requirements. The software core is an optimized YOLOv8n detection model, deployed on a TensorFlow Lite or NCNN inference framework. Threat target identification module 2 outputs three types of detection results: target bounding box coordinates, target type label, and threat confidence, which are sent to the covert decision generation module 3 via UART serial port.

[0051] The stealth decision generation module 3 is used to generate stealth control decisions based on threat detection results, realizing intelligent conversion from threat information to control parameters. As described in the method embodiment, in response to a threat confidence level greater than a preset confidence threshold and a target type label belonging to a preset threat type set, the stealth decision generation module 3 calculates the threat level based on the threat confidence level and estimated target distance, determines the buoyancy adjustment amount and target depth based on the threat level, and generates a stealth trigger signal. The function of the stealth decision generation module 3 can be performed by the embedded processor of the threat target identification module 2, or it can be implemented using a separate low-power microcontroller. In this embodiment, an STM32F103 microcontroller is used as the hardware carrier of the stealth decision generation module 3, whose 72MHz main frequency and 64KB Flash storage space are sufficient to run the decision algorithm. The decision algorithm includes the threat level calculation, buoyancy adjustment amount calculation, and silence time calculation detailed in the method embodiment, all of which are implemented using fixed-point arithmetic to improve computational efficiency.

[0052] The buoyancy adjustment execution module 4 is used to execute buoyancy adjustment actions in response to concealed trigger signals, realizing the electromechanical conversion from control parameters to physical motion. As described in the method embodiment, the buoyancy adjustment execution module 4 drives the gear transmission mechanism to drive the parallel injector assembly to draw in or expel seawater, causing the marine monitoring platform to submerge or surface. The hardware components of the buoyancy adjustment execution module 4 include a waterproof servo motor, a gear transmission mechanism, a parallel injector assembly, a seawater pipeline, a solenoid valve, and a servo motor drive board. The waterproof servo motor is a digital servo motor with a torque of not less than 25 kg·cm. The gear transmission mechanism has a transmission ratio of 1:2 to 1:3. The parallel injector assembly includes three 350 mL capacity medical-grade injectors. The seawater pipeline is made of corrosion-resistant silicone material. The solenoid valve controls the direction of seawater flow. The servo motor drive board receives PWM signals from the main controller to drive the servo motor to rotate.

[0053] The silence state maintenance module 5 is used to maintain an underwater silence state after the platform submerges, realizing closed-loop control from covert execution to state recovery. As described in the method embodiment, after the marine monitoring platform submerges, the silence state maintenance module 5 shuts down the active signal source, determines the silence duration based on the threat level, and controls the buoyancy adjustment execution module 4 to discharge seawater and cause the marine monitoring platform to surface when the silence duration expires and no sea surface threat is detected. The function of the silence state maintenance module 5 is undertaken by the system main controller. During the silence period, the main controller enters a low-power sleep mode, retaining only the timer wake-up function. The depth sensor uses an MS5837 series pressure sensor, with a measurement accuracy better than ±2 meters. The accelerometer uses an ADXL345 triaxial digital accelerometer to detect sea surface vibrations and determine the threat elimination status.

[0054] The overall hardware architecture of this system comprises five parts: a main controller, sensor components, actuator components, communication components, and power supply components. The main controller uses an STM32F103 series microcontroller with a 72MHz operating frequency, 64KB of Flash memory, and 20KB of RAM. It is responsible for overall system scheduling, decision algorithm execution, and inter-module communication coordination. The sensor components include an OpenMV wide-angle camera for image acquisition, an MS5837 depth sensor for depth monitoring, an ADXL345 accelerometer for vibration detection, and a DS18B20 temperature sensor for battery protection. The actuator components include an MG996R waterproof servo motor, a 1:2 gear ratio gear set, three 350mL medical syringes, silicone seawater tubing, and a 12V solenoid valve. The communication components include an SX1278 LoRa wireless module, a Beidou / GPS dual-mode positioning module, and a USB debugging interface. The power supply component uses a 22.2V / 10Ah six-cell lithium battery pack, along with a DC-DC power management module to provide 5V, 12V, and 3.3V power to each component.

[0055] This system, through the coordinated operation of five functional modules, enables the marine monitoring platform to achieve autonomous threat perception and adaptive stealth control. The overall system performance indicators include: detection accuracy of no less than 90%, single-frame detection time of no more than 200ms, dive time of no more than 3 seconds, adjustable dive depth of 1 to 3 meters, peak power consumption of no more than 10W, standby power consumption of no more than 3W, adjustable silent duration of 24 to 72 hours, and stealth survival rate of no less than 90%.

[0056] The embodiments of the present invention are not limited to the specific embodiments described above. Those skilled in the art can make various equivalent changes or substitutions based on the technical solutions of the present invention, and all such changes or substitutions should be included within the protection scope of the present invention.

Claims

1. An adaptive concealment control method for marine monitoring platforms based on target recognition, characterized in that, include: The sea surface image acquisition step involves acquiring sea surface images in real time using a wide-angle camera. The wide-angle camera has a field of view of not less than 120 degrees, an acquisition frame rate of not less than 30 frames per second, and an image resolution of not less than 640×480 pixels. The acquired sea surface images are then transmitted to an embedded processor. In the threat target identification step, the embedded processor runs a lightweight target detection model to identify threat targets in the sea surface image. The lightweight target detection model uses a depthwise separable convolutional structure for model compression and outputs the target bounding box coordinates, target type label, and threat confidence. The target type label includes military vessel type and civilian vessel type. In the covert decision generation step, in response to the threat confidence being greater than a preset confidence threshold and the target type label belonging to a preset threat type set, the covert decision generation module calculates the threat level based on the threat confidence and the target distance estimated based on the target bounding box coordinates, determines the buoyancy adjustment amount and target depth based on the threat level, and generates a covert trigger signal; In response to the concealed trigger signal, the buoyancy adjustment execution step involves the buoyancy adjustment execution module driving the gear transmission mechanism to drive the parallel injector group to draw in seawater. The amount of water drawn in the parallel injector group corresponds to the buoyancy adjustment amount, enabling the marine monitoring platform to dive to the target depth within a preset time. In the silent state maintenance step, after the marine monitoring platform descends to the target depth, the silent state maintenance module shuts off the active signal source, determines the silent duration based on the threat level, monitors the current depth through the depth sensor, and in response to the expiration of the silent duration and the absence of a sea surface threat, controls the buoyancy adjustment execution module to discharge seawater to allow the marine monitoring platform to rise to the sea surface.

2. The method according to claim 1, characterized in that, The lightweight target detection model is a YOLOv8n model that has undergone depthwise separable convolution compression. The model parameter size does not exceed 10MB, the single-frame detection time does not exceed 200ms, and the detection accuracy is not less than 90%.

3. The method according to claim 1, characterized in that, The preset confidence threshold is dynamically adjusted according to the current sea state and light intensity. When the light intensity is lower than the preset light reference value, the preset confidence threshold is reduced to improve detection sensitivity.

4. The method according to claim 1, characterized in that, The preset threat type set includes military vessels, patrol boats, and law enforcement vessels, but excludes civilian fishing boats and commercial cargo ships.

5. The method according to claim 1, characterized in that, The calculation of the buoyancy adjustment includes: calculating the estimated target distance based on the target pixel height in the target bounding box coordinates, the preset actual target height, and the camera focal length; calculating the threat level score based on the threat confidence and the estimated target distance; and adding an adaptive adjustment amount to the basic buoyancy adjustment amount based on the threat level score.

6. The method according to claim 1, characterized in that, The gear transmission mechanism has a transmission ratio of 1:2 to 1:3 and is driven by a servo motor with a torque of not less than 25 kg·cm.

7. The method according to claim 1, characterized in that, The parallel syringe group includes no less than three syringes connected in parallel, each syringe having a capacity of no less than 350 mL and a total buoyancy adjustment capacity of no less than 10 N.

8. The method according to claim 1, characterized in that, The active signal source includes a wireless communication module, a satellite communication module, and a positioning module. After the active signal source is turned off, the platform is in an electromagnetic silence state, and the standby power consumption does not exceed 3W.

9. The method according to claim 1, characterized in that, The duration of silence is determined based on the threat level. The higher the threat level, the longer the duration of silence. The range of the duration of silence is from the preset minimum silence time to the preset maximum silence time.

10. An adaptive covert control system for a marine monitoring platform based on target recognition, used to implement the method described in any one of claims 1-9, characterized in that, include: The sea surface image acquisition module is used to acquire sea surface images in real time through a wide-angle camera. The field of view of the wide-angle camera is not less than 120 degrees, the acquisition frame rate is not less than 30 frames per second, and the image resolution is not less than 640×480 pixels. The threat target identification module is used to run a lightweight target detection model to identify threat targets in the sea surface image. The lightweight target detection model uses a depthwise separable convolutional structure for model compression and outputs the target bounding box coordinates, target type label, and threat confidence. The covert decision generation module is used to calculate the threat level based on the threat confidence level and the estimated target distance when the threat confidence level is greater than a preset confidence threshold and the target type label belongs to a preset threat type set, determine the buoyancy adjustment amount and target depth based on the threat level, and generate a covert trigger signal. The buoyancy adjustment execution module is used to respond to the concealed trigger signal, drive the gear transmission mechanism to drive the parallel injector group to draw in seawater, so that the marine monitoring platform can dive to the target depth; The silence state maintenance module is used to shut down the active signal source after the marine monitoring platform dives, determine the silence duration based on the threat level, and control the buoyancy adjustment execution module to discharge seawater to make the marine monitoring platform float when the silence duration expires and no sea surface threat is detected.

Images