Method and device for AI camera shooting and water sample collection of coral reef marine fish under non-steady state condition

Abstract

The application discloses a method and device for AI camera shooting and water sample collection of coral reef marine fish under non-steady-state conditions, and belongs to the technical field of marine ecological monitoring. The device comprises a preset bottom-supported frame, a camera unit, an AI processing module, a water sample collection unit and a reagent adding unit. The device is thrown into the target coral reef sea area, the camera unit continuously collects real-time underwater video images of the reef area, and calls fish recognition models and coral reef state recognition models in parallel, synchronously analyzes the video frames, and outputs fish monitoring information and coral reef whitening quantitative data in real time; necessary synchronous sampling trigger instructions are analyzed and generated to control the suction device to extract in-situ seawater in the corresponding area, and a quantitative filling pump is started synchronously to fill the water sample into the same storage container and save the reagent. The application can realize underwater image monitoring, ecological intelligent recognition, event-triggered in-situ sampling and water sample synchronous solid sample integrated operation of the coral reef sea area, and solves key problems such as poor anti-interference ability of traditional equipment and low spatiotemporal matching degree of data.

Description

Technical Field

[0001] This application relates to the field of marine ecological monitoring technology, and in particular to an AI camera and water sample collection method and device for coral reef marine fish under non-steady-state conditions. Background Technology

[0002] Coral reef ecosystems are highly sensitive to water quality fluctuations and biological disturbances. Long-term, in-situ, high-precision monitoring requires the simultaneous acquisition of underwater visual images and in-situ water sample data precisely matched to biological events. This reveals the coupling relationship between ecological processes such as fish activity and coral bleaching and changes in the aquatic environment. Current monitoring methods generally deploy underwater cameras and automatic water samplers independently, operating at different times: underwater cameras continuously record fish behavior and coral growth in the reef area but lack water sample collection capabilities; conventional time-series water samplers can collect water samples at fixed points but cannot perceive underwater ecological dynamics in real time, only supporting timed sampling modes. This separate deployment of the two types of equipment results in a significant spatiotemporal disconnect between image recordings and water sample data, making it difficult to accurately correlate instantaneous biological disturbances and coral bleaching stress phenomena with corresponding water quality conditions, thus hindering research on coral reef ecological response mechanisms.

[0003] Meanwhile, the shallow sea areas where coral reefs are distributed are typical unstable aquatic environments, characterized by continuous turbulence caused by waves and tides, rugged and soft seabed sediment, and high levels of suspended solids in the water. Traditional bottom-mounted camera equipment is easily affected by water flow, causing it to shake and sink, resulting in image jitter; the inlet of conventional water samplers is easily blocked by coral debris and sediment, leading to water sample contamination or even sampling failure. Existing anti-disturbance and stable base structures are mostly only suitable for single camera equipment or single water sampling devices, lacking a synergistic optimization solution for integrated equipment.

[0004] More importantly, the sampling triggering logic of existing intelligent monitoring systems is relatively simple, and most of them use a single judgment threshold. They cannot distinguish the monitoring priorities of different ecological events such as indicative protected fish, common reef fish, and coral bleaching, making it difficult to achieve differentiated and multi-level intelligent sampling control. Summary of the Invention

[0005] This application aims to address the technical problems of poor spatiotemporal matching of data from existing coral reef ecological monitoring equipment, weak anti-interference capability in non-steady-state sea areas, inability to intelligently trigger sampling based on ecological events, and low sampling accuracy. It provides a method and device for AI-powered imaging and water sampling of marine fish in coral reefs under non-steady-state conditions, enabling integrated and coordinated operations of coral reef marine image monitoring, intelligent ecological identification, in-situ water sampling, and simultaneous water sample consolidation.

[0006] This application provides a method and apparatus for AI-based imaging and water sampling of marine fish in coral reefs under unsteady conditions, which adopts the following technical solution: In a first aspect, this application provides an AI camera and water sample collection device for coral reef marine fish under unsteady conditions, including a device body with a hollow internal structure, a bottom support frame, a reagent addition unit, a water sample collection unit, a camera unit, an AI processing module, and an unsteady condition compensation component.

[0007] The bottom support frame features adaptive compensation for unsteady environments. Serving as the load-bearing base of the entire device, it ensures stable placement of the equipment on complex seabeds, effectively mitigating disturbances from ocean currents and turbulent water flow, guaranteeing stable operation throughout the process. The reagent addition unit and water sample collection unit are fixedly mounted at designated locations on the bottom support frame. The water sample collection unit integrates multiple independent storage containers, each capable of rotating in a preset direction to precisely align with the output port of the reagent addition unit, enabling in-situ reagent addition and fixation after water sample collection. The camera unit and AI processing module are correspondingly embedded within the bottom support frame. The AI ​​processing module incorporates an optimized deep learning recognition model, enabling real-time intelligent identification of ecological events from underwater video images captured by the camera unit. When preset ecological events such as fish activity or coral bleaching are detected, it automatically outputs trigger control signals. Unsteady-state condition compensation components are also mounted on the bottom support frame, working in conjunction with it to achieve adaptive compensation and adjustment in unsteady marine environments.

[0008] As a preferred structure of the present invention, the device also integrates a control module, which is embedded in the base support frame and electrically connected to the reagent addition unit, water sample collection unit, camera unit, and AI processing module, serving as the core control hub of the device. The control module incorporates a two-dimensional event judgment logic, which can receive ecological event trigger signals output by the AI ​​processing module in real time. Based on the two-dimensional judgment logic, it performs ecological event validity verification, thereby precisely controlling the container repositioning action of the water sample collection unit, the quantitative reagent addition action of the reagent addition unit, and the start / stop shooting and continuous recording action of the camera unit. This completely eliminates the traditional timed sampling mode, realizing intelligent water sample collection and synchronous image monitoring triggered by ecological events, significantly improving the effectiveness and relevance of the sampling data.

[0009] To further enhance the equipment's operational stability in unstable reef environments, the bottom support frame integrates two major anti-interference structures: a current-stabilizing and disturbance-resistant structure, and a bottom-stabilizing structure. The current-stabilizing and disturbance-resistant structure utilizes a honeycomb current-stabilizing cover plate, installed at the top of the bottom support frame and corresponding to the camera unit's shooting area. This effectively evens out surface turbulence around the equipment, preventing water flow fluctuations from interfering with camera image clarity and equipment stability. The bottom-stabilizing structure mainly consists of counterweights and support leg components. The counterweights are fixedly arranged at the bottom of the bottom support frame, lowering the overall center of gravity of the equipment and improving seabed stability. The support leg components are located at the four corners of the bottom of the bottom support frame, employing a telescopic and adjustable structure to adapt to the complex terrain of coral reef seabeds with varying elevations, enabling the equipment to self-adaptively level.

[0010] Specifically, the outrigger assembly includes multiple adjustable telescopic outriggers and matching ball-joint adaptive foot pads. Each adjustable telescopic outrigger is equipped with a ball-joint adaptive foot pad at its bottom. The adjustable telescopic outriggers enable independent height adjustment at all four corners. Combined with the angle adaptive adjustment function of the ball-joint adaptive foot pads, it can conform to irregular reef bottom terrain, effectively avoiding problems such as equipment suspension, tilting, and slippage, and significantly improving the stability of equipment placement in unstable and complex seabed environments.

[0011] The water sampling unit comprises a suction device, a fixed outlet pipe, a horizontally rotating sample tray, and a rotation drive assembly. The suction device has an input port connected to a water intake, inside which a radially mounted filter effectively intercepts suspended impurities and floating debris, preventing pipe blockage and sample contamination. The output port connects to the fixed outlet pipe to remove the collected seawater. The horizontally rotating sample tray is positioned below the suction device, with multiple independent storage containers evenly arranged circumferentially around it. These containers rotate synchronously with the tray, automatically switching between sampling, filling, and standby positions. The rotation drive assembly is fixedly mounted on the support frame, with its power output connected to the center of the rotating sample tray. This provides stable power for the tray's indexing rotation and position switching, enabling precise alignment and sampling of empty containers.

[0012] In terms of the specific structural design of the reagent addition unit, the unit includes a pressure-resistant reagent storage chamber and a dispensing conduit. The pressure-resistant reagent storage chamber is fixed inside the bottom support frame, which can adapt to the high-pressure underwater environment and prevent the reagent storage chamber from deforming or leaking under pressure. The dispensing conduit is assembled at the bottom of the pressure-resistant reagent storage chamber to achieve directional discharge of reagents. A quantitative dispensing pump is also installed on the dispensing conduit to accurately control the reagent dosage, ensuring the accuracy and standardization of reagent addition for water sample preservation and ensuring the in-situ fixation effect of water samples.

[0013] The AI ​​processing module of this invention incorporates a deep learning model suitable for intelligent underwater image analysis. This model consists of two core modules: a fish identification model and a coral reef status identification model. The two models operate in parallel, enabling simultaneous fish monitoring and coral reef status assessment. The fish identification model employs at least one of the YOLO series object detection networks and the ResNet50 series classification backbone network, capable of identifying real-time information such as fish species, numbers, and activity status within underwater video footage. The coral reef status identification model uses the U-Net series semantic segmentation network to segment and extract coral reef regions from the image, quantifying core data such as coral bleaching area and bleaching percentage, thus achieving a quantitative assessment of coral bleaching status.

[0014] Secondly, this application provides a method for AI-based imaging and water sampling of fish in coral reef areas under unsteady conditions, implemented using any of the aforementioned devices, comprising the following steps: A method for AI-powered imaging and water sampling of marine fish in coral reefs under unsteady conditions, using the apparatus described above, includes the following steps: S1. Deploy the device to the target coral reef area and use the unsteady condition compensation component of the bottom support frame to complete the adaptive bottom leveling; use the honeycomb flow stabilizing cover plate to homogenize the surface turbulence, and use adjustable legs and counterweights to adapt to the undulating terrain of the reef bottom. S2. The camera unit continuously collects real-time underwater video images of the reef area and transmits the video stream to the AI ​​processing module in real time. S3 and the AI ​​processing module call the fish recognition model in parallel to analyze the real-time video frame by frame and output fish monitoring information in real time; based on the fish monitoring information, execute routine monitoring instructions and make early warning condition judgments: if it is determined that there is abnormal information in the fish monitoring information, then S4 is executed simultaneously. S4. The AI ​​processing module starts the coral reef status recognition model, performs pixel-level semantic segmentation and analysis on the coral reef area in the current frame, obtains real-time bleaching quantification data of the coral reef into ecological early warning information, and transmits back the video frame of the coral reef area in the current frame.

[0015] In a further embodiment, the fish monitoring information includes at least: the number of individuals identified per frame of coral reef water environment indicator organisms, the frequency of their appearance within a set monitoring period, and the continuous effective dwell time of a single individual; The corresponding routine monitoring instructions are as follows: when the AI ​​processing module identifies the presence of coral reef water environment indicator organisms in the monitoring field of view, it immediately drives the water sampling unit to collect in-situ seawater samples; the collected water samples are kept for future use for subsequent environmental DNA / eDNA testing.

[0016] In a further embodiment, the criteria for determining the presence of abnormal information in the fish monitoring information are as follows: Set up a continuous multi-segment judgment period. If at least one monitoring indicator is continuously lower than the corresponding health threshold and the abnormal state is maintained stably for N consecutive monitoring periods, it can be judged as abnormal. The monitoring indicators include the number of individuals identified per frame of coral reef water environment indicator organisms, the frequency of occurrence during the monitoring period, and the duration of continuous effective residence of individuals.

[0017] In summary, this application includes the following beneficial technical effects: The integrated structure combines camera monitoring, AI ecological recognition, water sample collection, and reagent sample fixation functions into a single bottom device. Furthermore, the water sampling module corresponds to the camera's field of view, solving the problem of spatiotemporal data fragmentation in traditional split-type monitoring equipment. This enables the linkage and matching of coral reef fish activity, coral status, and water quality parameters, significantly improving the correlation and accuracy of ecological monitoring data.

[0018] The system incorporates a non-steady-state condition compensation component, which uses a top honeycomb flow stabilizing cover to homogenize water flow turbulence. Combined with a bottom counterweight, adjustable telescopic outriggers, and ball joint adaptive feet, it adapts to the complex bottom texture of coral reefs and the non-steady-state water environment of turbulent waves. This effectively solves the problems of traditional equipment being prone to shaking, overturning, and slipping, ensuring long-term stable operation of the equipment.

[0019] The simultaneous operation of water sampling and reagent dispensing allows for immediate in-situ sample consolidation after water sampling, avoiding data distortion caused by component oxidation and microbial mutation during water sample transportation and storage. At the same time, the multi-station independent sampling structure enables the storage of multiple sets of ecological event water samples without cross-contamination, meeting the needs of continuous and refined monitoring.

[0020] Based on a deep learning model, it achieves dual-dimensional intelligent identification of fish communities and coral bleaching status. Relying on a custom event threshold, it intelligently triggers sampling, abandoning the blindness of traditional timed sampling. It can capture in-situ water samples corresponding to abnormal coral reef ecological events, effectively improving the effective sampling rate and adapting to the needs of dynamic ecological monitoring of coral reefs. Attached Figure Description

[0021] Figure 1 This is a main cross-sectional view of an AI camera and water sample collection device for coral reef marine fish under unsteady conditions, as described in Example 1.

[0022] Figure 2 This is a top-section view of an AI camera and water sample collection device for coral reef marine fish under unsteady conditions, as described in Example 1.

[0023] Figure 3 This is a flowchart of the method for AI photography and water sample collection of marine fish in coral reefs under unsteady conditions in Example 2.

[0024] Figure 4This is a flowchart of the two-dimensional event judgment logic in Example 2.

[0025] Explanation of reference numerals in the attached diagram: 1. Honeycomb flow stabilizing cover plate; 2. AI processing module; 3. Camera unit; 4. Water sample collection unit; 5. Pressure-resistant control chamber; 6. Reagent addition unit; 7. Counterweight block; 8. Seat support frame; 31. Camera module; 32. Lighting component; 41. Water collection module; 42. Suction device; 43. Fixed water outlet pipe; 44. Horizontal rotating sample tray; 45. Central drive shaft; 46. Sample storage container; 51. Control module; 52. Drive motor; 53. Power supply module; 54. Data storage module; 61. Pressure-resistant reagent storage chamber; 62. Quantitative pump; 63. Injection conduit; 81. Adjustable telescopic support leg; 82. Ball joint adaptive foot pad. Detailed Implementation

[0026] The following is in conjunction with the appendix Figure 1-4 This application will be described in further detail.

[0027] Example 1 like Figure 1 As shown in the figure, this embodiment discloses an integrated AI camera and water sample collection device for coral reef marine fish under unsteady conditions. It is suitable for the unsteady hydrodynamic environment of shallow coral reefs from 0 to 30m deep. It can achieve stable seabed mounting, real-time AI monitoring of underwater ecology, in-situ water sample collection triggered by biological events, and automatic water sample preservation. It effectively solves the technical problems of traditional equipment such as data spatiotemporal fragmentation, susceptibility to water flow disturbance, lens contamination, water collection blockage, and water sample contamination. The device consists of a seabed support frame 8, a camera unit 3, an AI processing module 2, a water sample collection unit 4, a reagent addition unit 6, a control module 51, an unsteady condition compensation component, and supporting power supply and sample storage structure. The various structures are precisely assembled and work together. The specific implementation structure and working method are as follows.

[0028] The main body of the device uses a corrosion-resistant aluminum alloy base support frame 8 as the load-bearing base. The overall structure adopts a hollow and lightweight frame structure, which reduces the disturbance of seawater buoyancy and avoids the accumulation and entanglement of coral reef debris, making it suitable for the complex and rugged seabed environment of coral reefs. Four sets of adjustable telescopic legs 81 are evenly distributed at the bottom of the frame. The legs adopt an electric telescopic adjustment structure with a telescopic adjustment stroke set from 0 to 150 mm. The height can be independently adjusted according to the undulation of the seabed rocks and the unevenness of the seabed, so as to achieve the leveling of the whole machine.

[0029] In the above embodiments, in order to adapt to the unsteady water flow conditions formed by waves and tides, a high-density, low-center-of-gravity counterweight block 7 is integrated at the lower part of the device frame. The counterweight block adopts a lead-sealed counterweight structure and is embedded and fixed at the center of the bottom of the frame, so that the overall center of gravity of the device is lower than the support plane of the legs, thereby improving the overall anti-overturning and anti-lateral flow impact performance of the machine and reducing the shaking of the whole machine caused by tides and turbulence.

[0030] The top of the device is detachably fitted with a honeycomb flow stabilizing cover plate 1, which has 8-12mm square through honeycomb holes. After the wide lateral turbulence brought by the surface waves is cut and diverted by the honeycomb structure, the large-scale irregular vortices are broken up and transformed into a uniform vertical and gentle water flow. This greatly reduces the direct impact of the lateral water flow on the underwater camera unit's field of view and the water intake module 4, maintaining stable imaging and a stable water intake flow.

[0031] In the above embodiment, each telescopic outrigger is further equipped with a ball-joint adaptive foot pad 82 at its bottom. The foot pad adopts a spherical hinge structure, which can achieve 360° small-angle adaptive deflection. The bottom surface of the foot pad is made of non-slip and wear-resistant rubber material and is provided with non-slip texture, which increases the contact area with the seabed and can effectively adapt to different coral reef bottoms such as soft mud and sand bottoms and hard reef bottoms, avoiding problems such as sinking, slipping, and tilting after the device sits on the bottom.

[0032] The camera unit 3 is fixedly installed on one side of the upper part of the base support frame 8, and the light-collecting axis and the water-collecting module 41 maintain a preset spatial correspondence. The camera unit 3 includes a high-resolution low-light color camera module 31 and a matching supplementary lighting component 32. The self-stabilizing gimbal, air deflector, and self-cleaning component can be selected according to the deployment environment.

[0033] The camera unit 3 is electrically connected to the AI ​​processing module 2 built into the pressure-resistant control chamber 5 via a waterproof signal cable. The AI ​​processing module 2 adopts embedded high-performance industrial control equipment and has a built-in trained deep learning image recognition model. The model can analyze the video stream transmitted by the camera unit 3 frame by frame in real time and automatically identify the core ecological indicators in the monitoring area, including fish species, fish numbers, fish population density, and fish swimming behavior. At the same time, it can accurately identify state parameters such as coral reef coverage, coral bleaching area, and degree of bleaching.

[0034] In practical work, the identification trigger thresholds can be preset according to monitoring needs, specifically including: the appearance of target protected fish, the number of fish gathering exceeding the preset threshold, the continuous expansion of coral bleaching areas, and abnormal disturbances to the bottom ecology. When the AI ​​processing module 2 identifies that the image status meets any preset conditions, it can instantly generate an electrical trigger signal and transmit it to the control module 51, with a trigger delay of ≤0.2s, achieving seamless connection between ecological events and sampling actions.

[0035] The water sampling unit 4 is fixed to the other side of the base support frame 8. Its core components include a water sampling module 41, a suction device 42, a fixed water outlet conduit 43, a horizontally rotating sample tray 44, a central drive shaft 45, and a sample storage container 46. The water sampling module 41 extends into the field of view of the underwater camera unit 3 via a rigid bracket, ensuring that the horizontal distance between the water sampling axis and the optical axis of the camera unit 3 is maintained at a preset value of 35~40cm, and that the orientation of the water sampling module 41 is consistent with the shooting direction of the underwater camera unit 3. The front end of the water sampling module 41 is equipped with a stainless steel filter with a pore size of 200μm to prevent the intake of coarse particles and coral debris. The fixed water outlet conduit 43 remains stationary, and the horizontally rotating sample tray 44 rotates in increments under the drive of the central drive shaft 45, so that the mouths of the sampling bottles are sequentially aligned with the bottom of the fixed water outlet conduit 43. After sampling of a single bottle, it automatically switches to the next bottle position.

[0036] The suction device 42 uses a low-noise waterproof micro water pump, which is connected to the water sampling module 41 through a fixed water outlet pipe 43. The pump start-up and shutdown and the water flow rate are precisely controlled by the control module 51. The water volume for a single sampling can be preset to 50-200mL, which meets the routine sampling standards for marine water quality testing.

[0037] A horizontally rotating sample tray 44 is installed horizontally inside the device and is driven to rotate by a central drive shaft 45 at the bottom. The tray has 12 evenly spaced mounting positions, each holding an independent, sealed, sterile sample storage container 46. This allows for independent water sample collection and storage for 12 different ecological events without cross-contamination. The central drive shaft 45 is driven by a stepper motor, enabling precise control of the sample tray's position switching and ensuring accurate connection and fixing of empty sample containers to the outlet pipe 43 and reagent filling port. A multi-position selection valve works in conjunction with the sample tray, switching pipelines under the command of the control module 51 to achieve independent single-bottle sampling and multi-event cyclic sampling.

[0038] In the above embodiments, to ensure the airtightness and independence of sampling, the sample storage container 46 is equipped with an automatic sealing cover. The cover is equipped with a miniature waterproof electromagnetic drive component, which is electrically connected to the control module 51. When the workstation rotates to the sampling and dispensing position, the control module 51 outputs an opening signal, and the electromagnetic component pulls the cover upward to open the bottle mouth. After the water sample extraction and reagent dispensing are completed, the control module issues a closing command, the electromagnetic drive component resets, and the cover fully closes and presses the container mouth to achieve a seal and lock, isolating external seawater and air from contact with the internal water sample, avoiding water sample oxidation, microbial growth, and secondary pollution from impurities. This enables the non-cross-isolation storage of water samples from multiple different ecological events.

[0039] The reagent addition unit 6 is arranged side by side with the fixed water outlet conduit, and includes a pressure-resistant reagent storage tank 61, a micro-quantitative pump 62, and a dispensing conduit 63. The dispensing port is also aligned with the bottle opening at the sampling station. The control module 51 can automatically add fixative to the bottle after sampling according to a preset volume ratio, so as to achieve in-situ fixation of the water sample and avoid changes in water quality composition during sample recovery.

[0040] When the water sampling unit 4 starts pumping water and the water sample flows into the sample storage container 46, the control module 51 simultaneously triggers the reagent addition unit 6 to work. The preset dose of preservation reagent is precisely added into the sample storage container 46 through a micro quantitative injection pump. The reagent addition volume is precisely matched with the water sample volume, realizing the integrated and synchronous completion of sampling and reagent fixation. No subsequent manual processing is required, which effectively avoids the distortion of detection data caused by microbial changes and component oxidation after water sample collection, and ensures the authenticity of the in-situ water sample data.

[0041] The control module 51 and power supply module 53 are both integrated inside the sealed and pressure-resistant control chamber 5. The control chamber adopts a waterproof and pressure-resistant sealed design, which is suitable for high-pressure and highly corrosive environments in shallow seas within 30m and can achieve long-term underwater self-contained operation. The control module 51 is the core scheduling unit of the device, which is electrically connected to the camera unit 3, AI processing module 2, suction device 42, reagent addition unit 6, and sample tray drive motor 52. It is responsible for receiving AI recognition trigger signals, coordinating and controlling the entire process of sampling, addition, and station switching, and recording a complete set of data such as sampling time, ecological event type, water sample number, and video images, and completing local storage.

[0042] The power supply module 53 uses a high-capacity waterproof lithium battery pack, which can provide continuous and stable power to the whole machine, supporting long-term unattended on-site monitoring without external cables, making it suitable for deployment in coral reef areas. All electrical connections of the whole machine are sealed with waterproof aviation plugs to prevent seawater infiltration and short circuit failures.

[0043] AI processing module 2 employs an embedded GPU / NPU computing platform and deploys pre-trained deep learning models, including but not limited to fish target detection models based on YOLO series or Faster R-CNN, and semantic segmentation models for coral reef coverage and bleaching degree. Real-time video streams are pre-processed and input into the models. When AI processing module 2 detects one of the following preset events (configurable according to research needs): a specific fish species appears continuously for more than a set number of frames / time; the instantaneous change in the number of individual fish exceeds a threshold; the proportion of bleached coral area exceeds a preset percentage and persists for a certain period, it immediately outputs a trigger command to control module 51.

[0044] After receiving the trigger command, the control module 51 (low-power MCU) records the current timestamp and video frame number, and executes the following steps sequentially: First, the control center drive shaft 45 rotates the horizontally rotating sample tray 44 to an empty bottle position. Then, the suction device 42 is activated for a short-term pre-flushing (to drain residual water from the pipeline). After that, seawater is extracted from the in-situ into the target sampling bottle at a set flow rate and volume. After sampling is completed, the reagent adding device is activated simultaneously to inject the storage reagent in proportion. After all actions are completed, the event information corresponding to the sample in that bottle is recorded, awaiting the next trigger.

[0045] In actual use, the device is hoisted by the research vessel and lowered to the seabed of a coral reef at a depth of 0-30m with the assistance of divers. Once seated, it stabilizes under its own weight and with the help of anti-sinking structures. After the system is powered on, the top honeycomb flow stabilizer 1 continuously homogenizes the water flow, while the underwater camera unit 3 continuously collects video data, which is then analyzed by the AI ​​processing module 2. When a preset biological event is detected, water sampling and reagent dispensing are triggered simultaneously, and data and samples are saved concurrently. After the mission, the device is recovered by the research vessel and divers, and the water samples are analyzed in the laboratory, with data transmitted back.

[0046] Example 2 Based on the device in Example 1, this example also proposes an AI-based method for capturing images and collecting water samples of fish in coral reef areas under unsteady conditions, combined with... Figure 3 The specific steps are as follows: S1. The entire machine is lowered to the target coral reef location using hoisting equipment. The machine is then leveled by self-adaptive bottoming using the counterweight block at the bottom of the base support frame, adjustable telescopic legs, and ball joint self-adaptive foot pads. The honeycomb flow stabilizing cover plate on the top of the equipment continuously cuts the surface wave turbulence, weakens the continuous disturbance of the transverse water flow to the camera and water sampling areas, and stabilizes the working flow field of the equipment.

[0047] S2. After the device is powered on, the camera unit continuously captures images of the reef area and transmits the complete video stream in real time to the AI ​​processing module in the pressure tank for synchronous processing.

[0048] S3 and the AI ​​processing module call the fish recognition model in parallel to analyze the real-time video frame by frame and output fish monitoring information in real time; based on the fish monitoring information, execute routine monitoring instructions and make early warning condition judgments: if it is determined that there is abnormal information in the fish monitoring information, then S4 is executed simultaneously. S4. The AI ​​processing module starts the coral reef status recognition model, performs pixel-level semantic segmentation and analysis on the coral reef area in the current frame, obtains real-time bleaching quantification data of the coral reef into ecological early warning information, and transmits back the video frame of the coral reef area in the current frame.

[0049] Therefore, this embodiment does not continuously transmit real-time video streams. Instead, it only retains and transmits single frames of video when an ecological anomaly is detected. This eliminates the need for continuous storage and transmission of massive amounts of video data, significantly reducing device memory usage and power consumption. Furthermore, it avoids interference caused by redundant data in real-time video transmission, effectively improving the accuracy and efficiency of ecological anomaly detection.

[0050] The fish monitoring information includes: monitoring frame data and monitoring time series data of coral reef aquatic environment indicator organisms; it should be noted that the coral reef aquatic environment indicator organisms mentioned in this embodiment are the typical coral reef monitoring fish species pointed to in item 7.3.1.4 of the industry standard HY / T 0460.5-2024, such as various species of butterflyfish, parrotfish, spinyfish, basketfish, and large serpentine.

[0051] The monitoring frame data refers to the number of individuals identified per frame, collected in real time frame by frame by the AI ​​camera unit. Single-frame target individual coordinate position The monitoring time series data is collected during a preset fixed monitoring period. Within, dynamic feature data obtained through time-series aggregation and differential operations, including the population comprehensive decay rate. and the overall migration escape rate of the group .

[0052] Furthermore, the population overall attenuation rate described in this embodiment The calculation process is as follows: ,in, This represents the total number of indicator fish species effectively detected within the current monitoring field of view. , , and These are indicator organisms of the coral reef aquatic environment. Frequency offset, number of individuals offset, and dwell time offset; the specific calculation formulas are as follows: ;in, , and These are indicator organisms of the coral reef aquatic environment. Pre-set frequency thresholds, quantity thresholds, and dwell time thresholds; , and Indicator organisms for coral reef aquatic environments During the preset fixed monitoring period The corresponding actual value within.

[0053] Correspondingly, the overall migration escape rate of the population The calculation formula is: ; It is dimensionless and has a range of values. The higher the value, the stronger the overall trend of indicator organisms in the reef area fleeing the monitoring station. Among them, The total number of fish individuals whose movement direction points outward from the field of view and are determined to be migrating outward; the direction of individual movement is determined by the pixel coordinates of adjacent frames. The displacement vector is obtained by solving the inter-frame difference method. This refers to the total number of individuals that migrated back to the monitoring area from outside the field of view; The total outward displacement flux of all escaping individuals. The theoretical maximum outward displacement flux is pre-calibrated and used to eliminate calculation errors caused by differences in fish size and swimming speed.

[0054] Among them, displacement flux The calculation formula is: ,in, For the first Each migrating individual is determined by the coordinate position of the target individual in adjacent frames. The displacement vector obtained by difference operation; The magnitude represents the displacement vector of a single emigrating individual in the direction of outward escape.

[0055] The corresponding routine monitoring instructions are as follows: when the AI ​​processing module identifies the presence of coral reef water environment indicator organisms in the monitoring field of view, it immediately drives the water sampling unit to collect in-situ seawater samples; the collected water samples are kept for future use for subsequent environmental DNA / eDNA testing.

[0056] Traditional coral reef monitoring often employs intermittent manual sampling on a quarterly or annual basis, which cannot achieve synchronous sampling at the moment of indicator organism appearance. This makes it difficult to obtain in-situ water eDNA information that closely matches the activity of indicator fish, resulting in poor timeliness and weak correlation of subsequent population tracing and aquatic environment assessment data. This solution uses AI to identify indicator organisms in real time, enabling routine and precise water sampling upon the sighting of fish. It accurately preserves in-situ water genetic information during the habitat and activity periods of indicator fish, greatly improving the authenticity and reliability of subsequent eDNA community analysis and ecological tracing results.

[0057] Based on the overall population decay rate and the overall migration escape rate of the group The specific process for determining whether there are abnormalities in fish monitoring information is as follows: Pre-configured population decay threshold Population migration escape rate threshold If continuous Preset fixed monitoring period Inside, satisfy and / or And the total number of relocated individuals If so, it is determined that the indicator species population in the reef area has experienced abnormal group disturbance; After identifying a collective abnormal disturbance, the AI ​​processing module synchronously calls the coral reef status recognition model to obtain the quantification data of coral reef bleaching, and generates an early warning sampling instruction to drive the water sampling unit to complete in-situ seawater sampling, and simultaneously collect and transmit coral reef image data of the monitoring points.

[0058] Existing visual monitoring methods mostly rely on single-frame, instantaneous data to determine ecological anomalies. These methods are highly susceptible to interference from accidental factors such as suspended matter in the water, light and shadow fluctuations, and the instantaneous movement of fish, making them prone to momentary misjudgments and false alarms. This embodiment differs from the approach of judging individual fish indicators. It adopts a macro-level approach, relying on two population characteristic indicators—the overall population decline rate and the overall population migration and escape rate—to determine collective ecological anomalies. Furthermore, it introduces a multi-period continuous steady-state determination mechanism. Ecological disturbance anomalies are only determined when the overall population size, frequency of occurrence, and duration of stay show a continuous decline within multiple consecutive monitoring windows, or when directional escape behavior occurs. Upgrading from individual point identification to population trend analysis effectively filters out false triggers caused by local, instantaneous environmental fluctuations, significantly improving the anti-interference capability and accuracy of coral reef ecological early warnings.

[0059] It should be noted that the real-time coral bleaching quantification value described in this embodiment... The determination method involves calling the U-Net semantic segmentation network built into the AI ​​processing module to perform pixel-level recognition on the real-time images transmitted by the underwater camera unit, and calculating the ratio of bleached coral pixels to the total number of live coral pixels to obtain a real-time bleaching quantification value. Unlike the existing manual transect measurement method, traditional monitoring methods require underwater deployment of tape measure transects, manual visual inspection of bleached corals, measurement of bleached coral lengths, and conversion of bleaching rates. This method only allows for intermittent manual surveys and cannot perform real-time, in-situ measurement of the degree of bleaching around the clock.

[0060] It is understood that some features in the above embodiments can be replaced or omitted, as long as they do not depart from the core concept of integrated triggered water sampling, in-situ reagent fixation, and non-steady-state flow compensation of the present invention. For example, the number of sampling bottles can be adjusted to 6 or 9 bottles according to monitoring needs; the reagent adding device can be expanded to have multiple reagent channels to adapt to the storage requirements of different detection indicators. The device can also be expanded to include sensors for parameters such as temperature, salinity, pH, and dissolved oxygen to further improve the synchronization data. All modifications based on the essence of the present invention fall within its protection scope.

[0061] The above are all preferred embodiments of this application, and are not intended to limit the scope of protection of this application. Therefore, all equivalent changes made in accordance with the structure, shape and principle of this application should be covered within the scope of protection of this application.

Claims

1. An AI-powered camera and water sampling device for coral reef marine fish under unsteady-state conditions, comprising a device body having a hollow internal structure; characterized in that, The device further includes: The bottom support frame has an adaptive compensation function for unsteady environment, which will stabilize the device on the seabed and counteract ocean current disturbances. The reagent addition unit and the water sample collection unit are respectively set at designated positions on the base support frame; the water sample collection unit is equipped with several independent storage containers, and each storage container rotates in a predetermined direction to align with the output port of the reagent addition unit in turn; The camera unit and the AI ​​processing module are installed in the support frame. The AI ​​processing module has a built-in deep learning recognition model for identifying ecological events in video images and outputting a trigger signal when a preset ecological event is identified.

2. The AI ​​camera and water sample collection device for coral reef marine fish under unsteady conditions according to claim 1, characterized in that, Also includes: The control module is integrated and installed within the base support frame, and is electrically connected to the reagent addition unit, water sample collection unit, camera unit and AI processing module respectively; The control module is used to receive the trigger signal output by the AI ​​processing module and has a built-in two-dimensional event judgment logic. Using the two-dimensional event judgment logic, and based on the trigger signal, it controls the container repositioning action of the water sample collection unit, the reagent addition action of the reagent addition unit, and the start and stop shooting action of the camera unit to realize intelligent water sample collection and image monitoring operations triggered by ecological events.

3. The AI ​​camera and water sampling device for coral reef marine fish under unsteady conditions according to claim 1, characterized in that, The base support frame is equipped with a flow stabilization and disturbance rejection structure and a base stabilization structure. The current stabilization and anti-interference structure is a honeycomb current stabilization cover plate, which is disposed on the top of the base support frame and adjacent to the camera unit; The seat stabilization structure includes at least a counterweight and a leg assembly; the counterweight is arranged at the bottom of the seat support frame, and the leg assembly is located at the four corners of the bottom of the seat support frame, and the leg assembly is a telescopic and length-adjustable structure.

4. The AI ​​camera and water sampling device for coral reef marine fish under unsteady conditions according to claim 3, characterized in that, The outrigger assembly includes: adjustable telescopic outriggers and ball joint adaptive foot pads; multiple adjustable telescopic outriggers are provided, and each adjustable telescopic outrigger is equipped with a ball joint adaptive foot pad.

5. The AI ​​camera and water sampling device for coral reef marine fish under unsteady conditions according to claim 1, characterized in that, The water sample collection unit includes: The suction device has a water inlet installed at its input end via a pipe; a filter screen is installed radially inside the water inlet. A fixed water outlet pipe is installed at the output end of the suction device; A horizontally rotating sample tray is located below the suction device; the plurality of independent storage containers are arranged circumferentially along the horizontally rotating sample tray and rotate horizontally to switch work positions synchronously with the horizontally rotating sample tray. A rotary drive assembly is fixed to the base support frame; the output end of the rotary drive assembly is connected to the center position of the horizontally rotating sample disk.

6. The AI ​​camera and water sampling device for coral reef marine fish under unsteady conditions according to claim 1, characterized in that, The reagent addition unit includes: The pressure-resistant reagent storage chamber is fixed within the base support frame; A dispensing conduit is installed at the bottom of the pressure-resistant reagent storage chamber; a metering pump is configured on the dispensing conduit.

7. The AI ​​camera and water sampling device for coral reef marine fish under unsteady conditions according to claim 1, characterized in that, The AI ​​processing module has a built-in deep learning model for intelligent underwater image analysis, which is divided into a fish recognition model and a coral reef status recognition model. The fish recognition model employs at least one of the YOLO series target detection networks and the ResNet50 series classification backbone networks to identify real-time information about fish in the image. The coral reef status identification model uses the U-Net series semantic segmentation network to segment and extract coral reef areas and determine the quantification data of coral bleaching.

8. A method for AI-based imaging and water sampling of marine fish in coral reefs under unsteady conditions, using the apparatus described in any one of claims 1 to 7, characterized in that, Includes the following steps: S1. Deploy the device to the target coral reef area and use the unsteady condition compensation component of the bottom support frame to complete the adaptive bottom leveling; use the honeycomb flow stabilizing cover plate to homogenize the surface turbulence, and use adjustable legs and counterweights to adapt to the undulating terrain of the reef bottom. S2. The camera unit continuously collects real-time underwater video images of the reef area and transmits the video stream to the AI ​​processing module in real time. S3 and the AI ​​processing module call the fish recognition model in parallel to analyze the real-time video frame by frame and output fish monitoring information in real time; based on the fish monitoring information, execute routine monitoring instructions and make early warning condition judgments: if it is determined that there is abnormal information in the fish monitoring information, then S4 is executed simultaneously. S4. The AI ​​processing module starts the coral reef status recognition model, performs pixel-level semantic segmentation and analysis on the coral reef area in the current frame, obtains real-time bleaching quantification data of the coral reef into ecological early warning information, and transmits back the video frame of the coral reef area in the current frame.

9. The method for AI-based imaging and water sampling of marine fish in coral reefs under unsteady conditions according to claim 8, characterized in that, The fish monitoring information includes: monitoring frame data and monitoring time-series data of coral reef aquatic environment indicator organisms; wherein, the monitoring frame data is the number of individuals identified in a single frame, collected in real time frame by frame by the AI ​​camera unit. Single-frame target individual coordinate position The monitoring time series data is collected during a preset fixed monitoring period. Within, dynamic feature data obtained through time-series aggregation and differential operations, including the population comprehensive decay rate. and the overall migration escape rate of the group ; The corresponding routine monitoring instructions are as follows: when the AI ​​processing module identifies the presence of coral reef water environment indicator organisms in the monitoring field of view, it immediately drives the water sampling unit to collect in-situ seawater samples; the collected water samples are kept for future use for subsequent environmental DNA / eDNA testing.

10. The method for AI-based imaging and water sampling of marine fish in coral reefs under unsteady conditions according to claim 8, characterized in that, Based on the overall population decay rate and the overall migration escape rate of the group The specific process for determining whether there are abnormalities in fish monitoring information is as follows: Pre-configured population decay threshold Group migration escape rate threshold ; If continuous Preset fixed monitoring period Inside, satisfy and / or And the total number of relocated individuals If so, it is determined that the indicator species population in the reef area has experienced abnormal group disturbance; After identifying a collective abnormal disturbance, the AI ​​processing module synchronously calls the coral reef status recognition model to obtain the quantification data of coral reef bleaching, and generates an early warning sampling instruction to drive the water sampling unit to complete in-situ seawater sampling, and simultaneously collect and transmit coral reef image data of the monitoring points.

Images