Intelligent monitoring and early warning method for salinity change in mariculture environment

Abstract

The invention relates to the technical field of mariculture intelligent monitoring, in particular to an intelligent monitoring and early warning method for salinity change in a mariculture environment, which comprises the following steps of: acquiring real-time salinity, temperature and underwater video data through sensor arrays arranged on a surface layer, a middle layer, a bottom layer, a water inlet and a water outlet of a culture water area, and carrying out space-time alignment and fusion processing; generating high-dimensional salinity characteristic data; inputting the data into a pre-trained salinity dynamic prediction model, and outputting a salinity prediction value sequence in a future set time period; comparing the prediction result with the dynamic salinity safety threshold range, and generating a graded early warning signal of a corresponding grade; and generating a control command according to the early warning level and issuing the control command to the breeding environment regulation and control equipment to realize accurate correction of salinity. According to the invention, integration of multi-source information fusion, salinity trend prediction and response linkage is realized, and the method has the advantages of high monitoring precision, accurate early warning judgment, strong automatic control capability and the like.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of intelligent monitoring of mariculture, and particularly relates to an intelligent monitoring and early warning method for salinity changes in a mariculture environment. BACKGROUND

[0002] With the rapid development of mariculture, the influence of water environmental parameters on the health, growth rate and survival rate of cultured organisms is increasingly significant, especially salinity as a key water quality indicator, its fluctuation is directly related to the osmotic pressure regulation ability, immune level and feeding behavior of cultured organisms. In coastal or estuarine areas, affected by factors such as rainfall, evaporation, tides and changes in freshwater supply, the salinity of the cultured water body often fluctuates frequently. Some high economic value species (such as Penaeus monodon and Japanese seabream) are particularly sensitive to salinity changes, and if early identification and effective intervention of salinity changes cannot be achieved, it is easy to lead to stress response, outbreak of diseases and even large-scale death.

[0003] Existing salinity monitoring is mainly based on single-point sensor collection, which lacks comprehensive perception ability of spatial heterogeneity in cultured waters, and cannot capture the salinity evolution characteristics at multiple water layers and key exchange channels (such as water inlet and drainage outlet). Although some systems introduce data-driven models, most of them are based on static threshold judgment, and cannot set dynamic adaptive thresholds combined with the species, growth stage and water temperature conditions of cultured organisms, resulting in low warning accuracy and high false alarm rate. At the same time, the existing schemes generally lack linkage mechanism with regulation equipment, and need to rely on manual judgment and manual intervention, with serious response delay. The lack of systematic, closed-loop and personalized intelligent monitoring and early warning technology for salinity changes has become a key technical bottleneck restricting the improvement of fine management and intelligent operation level of modern mariculture. SUMMARY

[0004] The present application provides an intelligent monitoring and early warning method for salinity changes in a mariculture environment, which has a salinity change monitoring and early warning mechanism with dynamic perception, prediction analysis and hierarchical response capabilities, and can improve the intelligent level of aquaculture and ensure the safety of mariculture.

[0005] An intelligent monitoring and early warning method for salinity changes in a mariculture environment, comprising the following steps: S1: synchronously collecting real-time salinity data, temperature data and underwater video data through a sensor array arranged at different spatial sites in the cultured water area; performing spatio-temporal alignment and data fusion processing on the real-time salinity data, temperature data and underwater video data to generate high-dimensional salinity feature data containing spatial distribution and time sequence information; S2: inputting the high-dimensional salinity feature data into a pre-trained salinity dynamic prediction model, and the salinity dynamic prediction model outputs a salinity prediction value sequence in a future set time period; S3: Real-time comparison is made between the salinity prediction value sequence and a preset dynamic salinity safety threshold range, and if any prediction value in the salinity prediction value sequence exceeds the dynamic salinity safety threshold range, a corresponding hierarchical early warning signal is generated; S4: According to the hierarchical early warning signal, a corresponding aquaculture environment regulation device is automatically triggered and started to correct the salinity of the aquaculture environment.

[0006] Optionally, the S1 comprises: S11: Through the sensor array arranged at different spatial sites of the aquaculture water area, real-time salinity data, real-time temperature data and underwater video data from each spatial site are synchronously collected under the control of the same time stamp; S12: The real-time salinity data, real-time temperature data and underwater video data collected in S11 are paired and aligned according to their collection time stamps and sensor spatial site coordinates, and multi-source synchronous data after alignment is generated; S13: The multi-source synchronous data after alignment obtained in S12 is subjected to fusion processing, and the fusion processing comprises extracting the behavior characteristics of the cultured organisms in the underwater video data, and integrating the behavior characteristics with the real-time salinity data and real-time temperature data of the corresponding space-time point, and finally generating high-dimensional salinity feature data containing spatial distribution and time sequence information.

[0007] Optionally, the arrangement spatial sites of the sensor array comprise the surface layer, the middle layer, the bottom layer, the water inlet and the water outlet of the aquaculture water area.

[0008] Optionally, the S2 comprises: S21: The high-dimensional salinity feature data is subjected to standardization and dimension transformation processing according to the input format required by the pre-trained salinity dynamic prediction model, and model input data is generated; S22: The model input data is input into the pre-trained salinity dynamic prediction model, and the pre-trained salinity dynamic prediction model is used for calculation and analysis, and an initial prediction result is output; S23: The initial prediction result is subjected to inverse standardization and data smoothing processing, and a salinity prediction value sequence with explicit time stamps within a future set time period is generated.

[0009] Optionally, the pre-trained salinity dynamic prediction model comprises a data input layer, a feature extraction module, a time series prediction module and a model output layer; the data input layer is used to receive the high-dimensional salinity feature data, the feature extraction module is used to extract space-time features from the high-dimensional salinity feature data, the time series prediction module is used to calculate future salinity changes according to the space-time features, and the model output layer is used to output the salinity prediction value sequence.

[0010] Optionally, the S3 comprises: S31: According to the variety, growth stage and current water temperature of the target cultured organism, querying the pre-stored biological stress salinity database to obtain and set the current required dynamic salinity safety threshold range; S32: Comparing the salinity prediction value sequence with the obtained dynamic salinity safety threshold range point by point in real time, analyzing and recording the degree and duration of each prediction value in the salinity prediction value sequence exceeding the dynamic salinity safety threshold range; S33: According to the obtained degree and duration of exceeding the dynamic salinity safety threshold range, logical judgment is carried out according to the pre-stored grading warning rule to generate the corresponding grading warning signal.

[0011] Optionally, the pre-stored biological stress salinity database in the S31 stores the sub-lethal stress salinity critical value of the cultured organism of different varieties, different growth stages under different water temperatures, and the upper limit and lower limit of the dynamic salinity safety threshold range are set based on the sub-lethal stress salinity critical value.

[0012] Optionally, the S4 comprises: S41: Receiving the grading warning signal generated in step S3, and analyzing the specific control instruction corresponding to the grading warning signal according to the pre-stored warning response strategy library; S42: According to the specific control instruction, a device control command corresponding to the level of the grading warning signal is generated, which can be recognized and executed by various types of cultured environment regulation and control equipment; S43: The device control command is sent to the corresponding cultured environment regulation and control equipment, triggering and starting the cultured environment regulation and control equipment to run to correct the salinity of the cultured environment.

[0013] Optionally, the cultured environment regulation and control equipment comprises one or more combinations of fresh water supply system, drainage system, oxygenator and automatic feeder, and different levels of grading warning signals trigger different combinations of cultured environment regulation and control equipment to run cooperatively.

[0014] The beneficial effects of the present application are: The present application, by arranging sensor arrays at different spatial points (including surface layer, middle layer, bottom layer, water inlet and drainage outlet) in the cultured water area, combining with the unified timestamp control mechanism, constructs high-dimensional salinity feature data with spatial distribution and time sequence characteristics. This data integrates real-time salinity, temperature and behavior characteristic information of cultured organisms. Compared with the traditional monitoring method which only relies on a single physical parameter, it can more comprehensively reflect the salinity dynamic evolution process of the cultured water body, improve the integrity and timeliness of the monitoring data, and provide a reliable foundation for subsequent model prediction.

[0015] The salinity dynamic prediction model constructed by the application adopts a feature extraction structure in series connection of a long short-term memory network and a time sequence convolution network, and performs multi-time step prediction output through a fully connected neural network, so that the accurate prediction of the future salinity change trend is realized. Meanwhile, the salinity safety threshold range is dynamically set according to the species, growth stage and current water temperature of the target cultured organism, and a multi-level early warning mechanism is triggered according to the deviation degree and duration of the predicted value, so that compared with the static threshold determination mode, the sensitivity of early warning and the matching degree of biological adaptability are significantly improved.

[0016] The application automatically links the aquaculture environment regulation equipment based on the hierarchical early warning signal, and a combined response strategy including a freshwater supply system, a drainage system, an oxygenator and an automatic feeder, can dynamically generate equipment control commands according to the difference of early warning levels and remotely issue and execute the commands, so that the automatic correction control of salinity anomaly is realized. The mechanism supports differential response, multi-device cooperation and closed-loop feedback, effectively reduces the manual intervention delay, improves the regulation efficiency and stability of the aquaculture system in response to salinity fluctuation, and significantly enhances the aquaculture safety guarantee capability under the condition of sudden salinity anomaly. BRIEF DESCRIPTION OF DRAWINGS

[0017] In order to more clearly illustrate the technical solutions in the application or the prior art, the following will briefly introduce the drawings needed to be used in the embodiments or prior art description. Obviously, the drawings in the following description are only a part of the application, and other drawings can also be obtained by those skilled in the art without creative labor.

[0018] Fig. 1 The method flowchart of the embodiment of the application is shown in the figure. Fig. 2 The S3 flowchart of the embodiment of the application is shown in the figure. DETAILED DESCRIPTION

[0019] The application will be described in detail below in combination with the drawings and specific embodiments. It should be noted that in order to make the embodiments more detailed, the following embodiments are the best, preferred embodiments, and other alternative ways can also be used by those skilled in the art to implement; and the drawings are only used to describe the embodiments more specifically, and are not intended to specifically limit the application.

[0020] It should be noted that in the specification, "one embodiment", "embodiment", "exemplary embodiment", "some embodiments" and the like indicate that the described embodiments can include specific features, structures or characteristics, but not necessarily every embodiment includes the specific features, structures or characteristics. In addition, when a specific feature, structure or characteristic is described in combination with an embodiment, it should be within the knowledge of those skilled in the art to realize this feature, structure or characteristic in combination with other embodiments (whether or not explicitly described).

[0021] Generally, terminology can be understood at least in part from the context of usage. For example, the term "one or more" as used herein, depending at least in part upon context, can be used to describe any feature, structure, or characteristic in a singular sense or can be used to describe combinations of features, structures or characteristics in a plural sense. Additionally, the term, "based on," can be understood as not necessarily

[0022] As shown in Figs. 1-2 A method for intelligent monitoring and early warning of salinity changes in a mariculture environment, comprising the following steps: S1: Through the sensor array arranged at different spatial points in the cultivation water area, real-time salinity data, temperature data and underwater video data are synchronously collected; real-time salinity data, temperature data and underwater video data are processed for spatio-temporal alignment and data fusion to generate high-dimensional salinity feature data containing spatial distribution and time sequence information, specifically: S11: Through the sensor array arranged at different spatial points in the cultivation water area, real-time salinity data, real-time temperature data and underwater video data from each spatial point are synchronously collected under the control of the same timestamp. Specifically, the sensor array is composed of multiple integrated collection units, each of which is fixedly deployed at a specific spatial point in the cultivation water area. The spatial points of the sensor array include the surface layer, the middle layer, the bottom layer, the water inlet and the water outlet of the cultivation water area. Among them, the surface layer position is not more than 0.3 meters from the water surface, the middle layer position is located at the geometric midpoint between the surface layer and the bottom layer, the bottom layer position is not more than 0.2 meters from the pool bottom, the water inlet position is set at the front end of the water body inlet pipe 0.5 meters, and the water outlet position is set in the water outlet channel 1.0 meters.

[0023] At each of the above-mentioned spatial points, at least one real-time salinity sensor, one real-time temperature sensor and one underwater video acquisition module are configured. The collection of real-time salinity data, real-time temperature data and underwater video data is synchronously controlled by a unified clock control system to ensure that all collection units perform data collection operations at the same timestamp. The timestamp synchronization adopts a distributed time control device, with an error controlled within 10 milliseconds to ensure the time consistency between data.

[0024] S12: The real-time salinity data, real-time temperature data and underwater video data collected in S11 are paired and aligned according to their time stamps and sensor spatial site coordinates to generate aligned multi-source synchronous data. Specifically, all collected data are preliminarily classified according to time stamps, and data with a collection time error exceeding ±10 milliseconds are excluded. Subsequently, according to the spatial site coordinates recorded when the sensors are laid out, the real-time salinity data, real-time temperature data and underwater video data from the same spatial site under the same time stamp are combined into complete data entries to construct a standardized data structure. Each data entry contains the following fields: collection time stamp, spatial site number, real-time salinity value, real-time temperature value, underwater video frame index number and corresponding original video data reference address.

[0025] The pairing and alignment process is performed by a central coordination computing node, and all data are collected to the control platform for unified processing through wired mode, ensuring accurate pairing and generating aligned multi-source synchronous data with timing consistency and spatial correspondence, providing high-precision input for subsequent fusion processing.

[0026] S13: The aligned multi-source synchronous data obtained in S12 are subjected to fusion processing, which includes extracting the behavior characteristics of the cultured organisms in the underwater video data and integrating the behavior characteristics with the real-time salinity data and real-time temperature data corresponding to the space-time point, to finally generate high-dimensional salinity feature data containing spatial distribution and time sequence information. Specifically, the underwater video data is first input into the behavior characteristic extraction module, which includes an image preprocessing unit, an object detection unit and a behavior coding unit. The image preprocessing unit performs image enhancement, denoising and smoothing processing on the video frames; the object detection unit uses a target detection model based on a deep convolutional neural network to identify the position information of the cultured organism individuals in the video; and the behavior coding unit calculates the behavior characteristics of the organisms according to their motion trajectories in consecutive frames, including motion speed, direction change frequency and aggregation distribution density.

[0027] Subsequently, the extracted behavior characteristics of the cultured organisms are integrated with the corresponding real-time salinity data and real-time temperature data under the same time stamp and the same spatial site to form structured fusion samples. All fusion samples are uniformly coded into multi-dimensional feature vectors, and the feature dimensions include spatial coordinates, time stamps, real-time salinity values, real-time temperature values and multiple behavior characteristic index values. All feature vectors are arranged in chronological order and spatial distribution to construct the final high-dimensional salinity feature data, which is used as input to the subsequent salinity dynamic prediction model.

[0028] S2: The high-dimensional salinity feature data is input into the pre-trained salinity dynamic prediction model, which outputs a salinity prediction value sequence for a future set time period, specifically: S21: The high-dimensional salinity feature data is standardized and dimensionally transformed according to the input format required by the pre-trained salinity dynamic prediction model to generate model input data. Specifically, first, each feature variable constituting the high-dimensional salinity feature data is standardized. The standardization method used is mean normalization or Z-score standard deviation normalization to ensure that the numerical range of all variables is on a unified scale, improving the stability and convergence speed of the model input.

[0029] Subsequently, the standardized high-dimensional salinity feature data is dimensionally transformed according to the structural requirements of the pre-trained salinity dynamic prediction model. The transformation includes reconstructing the original two-dimensional data (time x feature) into a three-dimensional tensor format (sample number x time step x feature dimension), where the time step is the time window length set during model training, and the feature dimension includes spatial coordinates, time stamp, real-time salinity value, real-time temperature value, and behavior feature encoding result of the cultured organism. The model input data generated after dimension transformation is used as the input for subsequent model calculation.

[0030] S22: The model input data is input into the pre-trained salinity dynamic prediction model, and the pre-trained salinity dynamic prediction model is calculated and analyzed to output the initial prediction result. The pre-trained salinity dynamic prediction model includes a data input layer, a feature extraction module, a time series prediction module, and a model output layer.

[0031] Specifically, the data input layer first receives the above-mentioned model input data and inputs it into the feature extraction module. The feature extraction module is composed of a long short-term memory network layer and a time series convolution network layer in series. The long short-term memory network layer is used to capture the long-term temporal dependence in the high-dimensional salinity feature data, and its output is input into the time series convolution network layer as an intermediate result to further extract multi-scale local spatio-temporal features and construct a complete spatio-temporal feature representation.

[0032] The spatio-temporal features are passed to the time series prediction module, which is composed of a fully connected neural network. The fully connected neural network inputs the spatio-temporal features and performs forward propagation operations to output the initial prediction results corresponding to multiple future time steps. The initial prediction results are standardized prediction value sequences and do not contain real salinity values in a physical sense.

[0033] The pre-trained salinity dynamic prediction model is trained using a historical data set. The historical data set includes historical high-dimensional salinity feature data and corresponding actual salinity value sequences for a future period of time. During training, the least mean square error is used as the loss function, and the Adam optimizer is used to update the parameters until the prediction error of the model on the validation set converges below a set threshold, ensuring the stability and generalization ability of the model.

[0034] S23: De-standardization and data smoothing processing are performed on the initial prediction result to generate a salinity prediction value sequence with a clear timestamp in a future set time period. Specifically, first, according to the standardization method adopted in S21, the initial prediction result is de-standardized to restore the physical order of magnitude of the true salinity value.

[0035] Subsequently, the salinity value sequence after de-standardization is subjected to data smoothing processing, and the smoothing method adopted is the exponential weighted moving average algorithm, so as to reduce the short-period fluctuations in the model prediction process and improve the continuity and interpretability of the prediction curve. After smoothing processing, combined with the starting timestamp and prediction step of the input model data, the timestamp corresponding to each prediction value is labeled one by one, and finally a salinity prediction value sequence with a clear time sequence identifier is formed, which serves as the input basis for subsequent graded warning.

[0036] S3: The salinity prediction value sequence is compared with the preset dynamic salinity safety threshold range in real time, and if any prediction value in the salinity prediction value sequence exceeds the dynamic salinity safety threshold range, a corresponding graded warning signal is generated, specifically: S31: According to the variety, growth stage and current water temperature of the target cultured organism, the pre-set biological stress salinity database is queried to obtain and set the current required dynamic salinity safety threshold range. Specifically, first, the variety information of the current cultured organism is identified through the image recognition module in the underwater video data, the growth stage information of the variety at present is obtained through system setting or feeding log, and the current water temperature is determined by using the real-time temperature data collected in S1.

[0037] The above three parameters are input into the biological stress salinity database as joint search conditions to obtain the sub-lethal stress salinity critical value under the corresponding conditions. The biological stress salinity database is a pre-set structured data table, which contains sub-lethal stress salinity critical value data of multiple cultured varieties (such as marbled shrimp, Japanese seabream, abalone, etc.), different growth stages (such as seedling stage, rapid growth stage, mature stage) of each variety, and corresponding water temperature intervals (such as 16-20℃, 21-25℃, 26-30℃).

[0038] According to the query result, the current applicable dynamic salinity safety threshold range is set, the upper limit and the lower limit of which are respectively the sub-lethal stress salinity critical value floating up and down by a certain safety margin (such as ±1.0 PSU, PSU is the practical salinity unit), so as to construct a salinity safety judgment boundary with dynamic adaptability.

[0039] S32: The salinity prediction value sequence is compared with the obtained dynamic salinity safety threshold range point by point in real time, and the degree and duration of each prediction value in the salinity prediction value sequence exceeding the dynamic salinity safety threshold range are analyzed and recorded. Specifically, let the salinity prediction value sequence be , the dynamic salinity safety threshold range is The following judgment is performed for each prediction value: If , it is determined to be "low salinity overrun", and the overrun degree is calculated as . If , it is determined to be "high salinity overrun", and the overrun degree is calculated as . If ， , it is determined to be "normal".

[0040] The time step number of the continuous occurrence of the overrun state is the overrun duration of the prediction value section. The system records the start time stamp, end time stamp, maximum overrun degree value and corresponding overrun type of each overrun state section, generates a prediction salinity anomaly marker sequence, which is used as the basis for subsequent warning level judgment.

[0041] S33: According to the degree and duration of exceeding the dynamic salinity safety threshold range obtained, logical judgment is performed according to the pre-set graded warning rules to generate corresponding graded warning signals. The graded warning rules are pre-set in the system database and stored in a rule table structure. According to the combined index of "overrun degree x duration", the warning levels are classified into three levels: First-level warning: any prediction value overrun degree exceeding 1.5 PSU occurs within 3 consecutive time steps; Second-level warning: the prediction value is continuously in the overrun state for 5 or more consecutive time steps, but the overrun degree is less than 1.5 PSU; Third-level warning: single-point transient overrun (overrun degree ≥ 0.5 PSU) occurs, but the duration does not exceed 2 time steps.

[0042] The logical judgment module reads the anomaly marker sequence generated in S32, matches the above rule table section by section, and outputs the corresponding graded warning signal if it meets a certain rule condition. Each graded warning signal includes: warning level, warning trigger time, overrun type, duration and maximum overrun value.

[0043] S4: According to the graded warning signal, automatically trigger and start the corresponding aquaculture environment control device to correct the salinity of the aquaculture environment, specifically: S41: Receive the graded warning signal generated by S3, and parse the specific control instructions corresponding to the graded warning signal according to the pre-set warning response strategy library. Specifically, continuously monitor the graded warning signal output by S3, and immediately extract the warning level, warning type, overrun direction (high salt or low salt) and duration fields in it when it is generated.

[0044] The hierarchical early warning signal is taken as an input to query the early warning response strategy library. The early warning response strategy library is a structured rule set, and different levels of hierarchical early warning signals correspond to control response rules. Each rule specifies the control logic to be activated under different combinations of salinity overrun types (high salt, low salt) and different early warning levels (level one, level two, and level three). The parsing process is completed by the response parsing module, and the output is a standardized control instruction set that clearly specifies the types of subsequent breeding environment control devices to be started, the running time, the start-stop conditions, and the execution priority.

[0045] S42: According to the specific control instruction, a device control command corresponding to the level of the hierarchical early warning signal is generated, which can be recognized and executed by various types of breeding environment control devices. The control command generation module encodes and packages the specific control instruction obtained in S41 according to the device communication protocol format to form a standardized device control command.

[0046] The device control command contains fields such as device identifier, action type (start / stop / adjust), execution time, adjustment parameters (such as drainage flow rate, freshwater supply amount), and execution trigger conditions, which can be directly recognized and executed by the target breeding environment control device. The content of the control command is differentiated according to the level of the hierarchical early warning signal, for example: For a level one early warning signal, the generated device control command will simultaneously activate the freshwater supply system and the drainage system to work together to quickly reduce the salinity; For a level two early warning signal, the control command will control the freshwater supply system separately to make slow corrections; For a level three early warning signal, an oxygenator is activated to run to improve water disturbance and assist in salinity diffusion.

[0047] In the case of low salinity overrun, the control command can also include limiting the drainage system and increasing the proportion of concentrated water reflux.

[0048] S43: The device control command is sent to the corresponding breeding environment control device to trigger and start the breeding environment control device to run to correct the salinity of the breeding environment. The device control platform sends the device control command to various types of breeding environment control devices through serial communication, Modbus protocol, or TCP / IP network interface.

[0049] The breeding environment control device includes one or more combinations of a freshwater supply system, a drainage system, an oxygenator, and an automatic feeder. Different levels of hierarchical early warning signals trigger different combinations of breeding environment control devices to run cooperatively. For example: In the case of rapid salinity rise and triggering a level one early warning signal, the drainage system and the freshwater supply system are started simultaneously, and the oxygenator running frequency is increased, forming a three-device cooperative running mechanism; If the corresponding three-level early warning signal is short-term and mild over-limit, the system only controls the automatic feeder to suspend feeding to reduce the biological metabolic stress burden; If it is a low-salt over-limit situation, the control command will trigger the concentrated salt water injection module (as part of the drainage system) to operate to improve the salinity level of the water body.

[0050] The equipment operating state is monitored in real time at the control center, the system dynamically evaluates the correction effect according to the execution feedback signal, and automatically issues a shutdown command when the conditions are met, ensuring that the salinity returns to the dynamic salinity safety threshold range.

[0051] The present application encompasses any alternative, modification, equivalent method and scheme made on the essence and scope of the present application. In order for the public to have a thorough understanding of the present application, specific details are described in the following preferred embodiments of the present application, and the present application can also be fully understood without these descriptions for those skilled in the art. In addition, in order to avoid unnecessary confusion to the essence of the present application, well-known methods, processes, procedures, elements and circuits, etc. are not described in detail.

[0052] The above is only the preferred embodiment of the present application, and it should be pointed out that for ordinary skilled in the art, without departing from the principles of the present application, a number of improvements and refinements can be made, which should be considered as the protection scope of the present application.

Claims

1. An intelligent monitoring and early warning method for salinity changes in a mariculture environment, characterized in that, The method comprises the following steps: S1: synchronously collecting real-time salinity data, temperature data and underwater video data through a sensor array arranged at different spatial sites in the aquaculture water area; spatial alignment and data fusion processing of the real-time salinity data, temperature data and underwater video data to generate high-dimensional salinity feature data containing spatial distribution and time sequence information; S2: inputting the high-dimensional salinity feature data into a pre-trained salinity dynamic prediction model, and the salinity dynamic prediction model outputs a salinity prediction value sequence in a future set time period; S3: real-time comparison of the salinity prediction value sequence and a preset dynamic salinity safety threshold range, if any prediction value in the salinity prediction value sequence exceeds the dynamic salinity safety threshold range, a corresponding graded early warning signal is generated; S4: according to the graded early warning signal, automatically triggering and starting the aquaculture environment regulation equipment corresponding to the warning level to correct the salinity of the aquaculture environment.

2. The intelligent monitoring and early warning method for salinity change in a mariculture environment according to claim 1, characterized in that, The S1 comprises: S11: synchronously collecting real-time salinity data, real-time temperature data and underwater video data from each spatial site under the control of the same timestamp through a sensor array arranged at different spatial sites in the aquaculture water area; S12: pairing and aligning the real-time salinity data, real-time temperature data and underwater video data collected in S11 according to their collection timestamps and sensor spatial site coordinates to generate aligned multi-source synchronous data; S13: fusion processing of the aligned multi-source synchronous data obtained in S12, the fusion processing comprising extracting the behavior characteristics of the cultured organisms in the underwater video data, and integrating the behavior characteristics with the real-time salinity data, real-time temperature data of the corresponding space-time point, and finally generating high-dimensional salinity feature data containing spatial distribution and time sequence information.

3. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 2, characterized in that, The arrangement spatial sites of the sensor array include the surface layer, middle layer, bottom layer, water inlet and water outlet of the aquaculture water area.

4. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 3, characterized in that, The S2 comprises: S21: standardizing and dimension transforming the high-dimensional salinity feature data according to the input format required by the pre-trained salinity dynamic prediction model to generate model input data; S22: inputting the model input data into the pre-trained salinity dynamic prediction model, and outputting an initial prediction result by calculation and analysis of the pre-trained salinity dynamic prediction model; S23: reverse standardization and data smoothing processing of the initial prediction result to generate a salinity prediction value sequence with explicit timestamps in a future set time period.

5. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 4, characterized in that, The pre-trained salinity dynamic prediction model comprises a data input layer, a feature extraction module, a time series prediction module and a model output layer; the data input layer is used to receive the high-dimensional salinity feature data, the feature extraction module is used to extract space-time features from the high-dimensional salinity feature data, the time series prediction module is used to calculate future salinity changes according to the space-time features, and the model output layer is used to output the salinity prediction value sequence.

6. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 5, characterized in that, The S3 comprises: S31: according to the variety, growth stage and current water temperature of the target cultured organisms, querying a pre-set biological stress salinity database to obtain and set the current required dynamic salinity safety threshold range; S32: Real-time point-by-point comparison of the salinity prediction value sequence with the obtained dynamic salinity safety threshold range, analysis and recording of the degree and duration of each prediction value in the salinity prediction value sequence exceeding the dynamic salinity safety threshold range; S33: According to the obtained degree and duration of exceeding the dynamic salinity safety threshold range, logical judgment is made according to the pre-set graded warning rules to generate corresponding graded warning signals.

7. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 6, characterized in that, The biological stress salinity database pre-set in S31 stores the sub-lethal stress salinity critical values of different varieties, different growth stages of cultured organisms under different water temperatures, and the upper and lower limits of the dynamic salinity safety threshold range are set based on the sub-lethal stress salinity critical values.

8. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 7, characterized in that, The S4 includes: S41: Receive the graded warning signal generated in step S3, and parse the specific control instructions corresponding to the graded warning signal according to the pre-set warning response strategy library; S42: According to the specific control instructions, generate device control commands corresponding to the level of the graded warning signal, which can be recognized and executed by various types of aquaculture environment control equipment; S43: Send the device control command to the corresponding aquaculture environment control equipment, trigger and start the operation of the aquaculture environment control equipment to correct the salinity of the aquaculture environment.

9. The intelligent monitoring and early warning method for salinity variation in a mariculture environment according to claim 8, characterized in that, The aquaculture environment control equipment includes one or more combinations of fresh water supply system, drainage system, oxygenator and automatic feeder, and different levels of graded warning signals trigger different combinations of aquaculture environment control equipment to operate cooperatively.

Images