Marine ranching site selection method and system

Abstract

The application discloses a marine ranching site selection method and system, which is applied to the field of data processing and comprises the following steps: acquiring seabed environment parameters of each target sea area; obtaining at least temperature field distribution results and biological field distribution results based on time sequence change information of the seabed environment parameters; determining full life cycle information of a target breeding object and corresponding growth demand parameter sets; performing space-time matching analysis on the temperature field distribution results, the biological field distribution results and the growth demand parameter sets, and calculating corresponding comprehensive ranching indexes of each target sea area; and obtaining site selection decision results of each target sea area based on at least shipping channel information, ecological protection information and the corresponding comprehensive ranching indexes of each target sea area. The marine ranching site selection method provided by the application can realize scientific and effective marine ranching site selection by comprehensively considering seabed time sequence sensing data, biological image analysis and full life cycle demand of a breeding object and combining a multi-target optimization algorithm.

Description

Technical Field

[0001] This invention relates to the field of data processing technology, and in particular to a method and system for site selection in marine ranching. Background Technology

[0002] Developing marine ranches with ecological restoration and sustainable utilization as the core has become an important strategic direction. Scientific site selection is the key prerequisite for the success or failure of marine ranch construction. Traditional site selection methods rely on historical experience, static hydrological maps or single environmental factors (such as water temperature and salinity) for manual judgment, which is highly subjective and has poor site selection effect.

[0003] Existing site selection methods use simple weighted scoring to process raw marine data and then obtain site selection decisions. However, some data within the marine area are affected by the environment and have low accuracy, making it difficult for existing site selection methods to support refined and intelligent aquaculture planning. Summary of the Invention

[0004] This invention provides a method and system for selecting sites for marine ranching. By comprehensively considering seabed time-series sensor data, biological image analysis, and the needs of the entire life cycle of farmed organisms, and combining multi-objective optimization algorithms, it can achieve scientific and effective site selection for marine ranching.

[0005] To address the aforementioned technical problems, embodiments of the present invention provide a method for selecting sites for marine ranching, comprising:

[0006] Acquire seabed environmental parameters collected by seabed sensors deployed in various target sea areas;

[0007] Based on the temporal variation information of the seabed environmental parameters, at least the temperature field distribution results and the biofield distribution results are obtained. The temperature field distribution results are quantified based on a temporal sliding window to assess the ecological impact of temperature field anomalies on fish stress. The biofield distribution results are obtained at least by analyzing a target image, where the target image is separated from the noise region by pixel-level labeling.

[0008] Determine the full life cycle information of the proposed aquaculture species, and determine the corresponding set of growth requirement parameters based on the full life cycle information;

[0009] The temperature field distribution results are spatiotemporally matched with the biofield distribution results and the growth requirement parameter set to calculate the comprehensive grazing index corresponding to each of the target sea areas.

[0010] Based at least on the shipping channel information, ecological protection information and the corresponding comprehensive grazing index of each of the target sea areas, a multi-objective optimization algorithm is used to obtain the site selection decision result for each of the target sea areas.

[0011] As one preferred embodiment, the process of obtaining at least the temperature field distribution result based on the time-series variation information of the seabed environmental parameters includes:

[0012] The span of the sliding window is determined based on the seasonal temperature variation characteristics of the target sea area.

[0013] The temperature sequence within the span is sequentially subjected to multi-scale decomposition and nonlinear trend modeling to obtain a temperature prediction model; wherein, the multi-scale decomposition includes extracting the high-frequency perturbation component and the low-frequency trend component of the temperature sequence.

[0014] The temperature anomaly events output by the temperature prediction model are coupled with the heat stress threshold of the selected aquaculture objects to obtain a stress risk heat map.

[0015] The temperature field distribution results are obtained based at least on the stress risk heat map.

[0016] As one preferred embodiment, the process of obtaining at least the biofield distribution results based on the temporal variation information of the seabed environmental parameters includes:

[0017] The pixel blocks in the underwater optical image acquired by the seabed sensor are randomly masked to generate several missing sub-images. The number of missing sub-images is determined based on the local pixel dependency strength of the underwater optical image.

[0018] Each of the missing sub-images is input into a pre-trained image reconstruction model for image reconstruction, and the reconstruction error of each pixel block is calculated based on the comparison results between the image reconstruction results and the underwater optical images.

[0019] Based on the reconstruction error, at least the noisy regions of the underwater optical image are marked;

[0020] The marked noise region is separated from the underwater optical image to obtain the biofield distribution result with the target region.

[0021] As one preferred embodiment, the random masking process includes K iterations, where K is a positive integer;

[0022] In the m-th process of the K processes, the pixel block to be masked is designed to be determined based on the spatial dependency between each pixel in the underwater optical image and its neighboring pixels. The spatial dependency is obtained at least through a local autocorrelation function, a structural tensor, or a neighborhood relationship modeling method based on a graph neural network, where 1≤m≤N.

[0023] As one preferred option, the species to be cultured is a specific fish species;

[0024] The process of determining the full life cycle information of the proposed aquaculture species and determining the corresponding set of growth requirement parameters based on the full life cycle information includes:

[0025] Based on the physiological and metabolic characteristics of the specific fish at different developmental stages, various life stages are divided, including at least the hatching period, juvenile period, young fish period, adult fish period, and reproductive period.

[0026] Quantify the dynamic response curve of each life stage to growth requirement information, wherein the growth requirement information includes at least water temperature gradient, dissolved oxygen concentration, photoperiod, feed particle size and nutrient composition and water flow velocity;

[0027] Based on the fuzzy comprehensive evaluation model, the dynamic response curves corresponding to each life stage are transformed into membership functions, and weighted fusion is performed based on the target weights selected by the aquaculture users to obtain the set of growth requirement parameters.

[0028] As one preferred embodiment, the step of performing spatiotemporal matching analysis on the temperature field distribution results, the biofield distribution results, and the growth requirement parameter set to calculate the comprehensive grazing index corresponding to each of the target sea areas includes:

[0029] The temperature field distribution results, the biofield distribution results, and the growth requirement parameter set are mapped to a coordinate system with latitude, longitude, water depth, and time as the dimensions to obtain a three-dimensional spatiotemporal model.

[0030] Based on the aforementioned three-dimensional spatiotemporal model, the spatiotemporal evolution results of the target sea area are obtained;

[0031] The change information of the spatiotemporal evolution results is quantified to obtain the comprehensive grazing index corresponding to each target sea area.

[0032] As one preferred embodiment, obtaining the spatiotemporal evolution results of the target sea area based on the three-dimensional spatiotemporal model includes:

[0033] Determine the geographical environmental parameters of the target sea area;

[0034] Based on the aforementioned geographical environmental parameters, the frequency of occurrence of various extreme environmental events corresponding to the target sea area is determined;

[0035] Based on the frequency of occurrence, the created environmental simulation events are applied to the three-dimensional spatiotemporal model;

[0036] The spatiotemporal evolution results of the target sea area are obtained based at least on the response results of the three-dimensional spatiotemporal model.

[0037] As one preferred embodiment, the number of environmental simulation events applied is at least two, and the application time of each environmental simulation event satisfies a preset condition;

[0038] During the process of applying each of the environmental simulation events to the three-dimensional spatiotemporal model, the attenuation index of each environmental simulation event is determined according to the degree of correlation between the various environmental simulation events.

[0039] During the application time, the initial response result of the three-dimensional spatiotemporal model is corrected based at least on the decay index to obtain the response result.

[0040] As one preferred embodiment, the biofield distribution results include at least the planktonic benthic organism regeneration rate and nutrient cycling flux;

[0041] During the spatiotemporal matching analysis, the matching degree between the expected aquaculture density and the ecological carrying capacity threshold function is compared in real time, and the aquaculture load is determined based on the matching degree result; the ecological carrying capacity threshold function is determined by the planktonic benthic organism regeneration rate and the nutrient cycling flux.

[0042] If the aquaculture load is detected to exceed the threshold, the current comprehensive aquaculture index will be subjected to non-linear decay processing.

[0043] Another embodiment of the present invention provides a marine ranching site selection system, comprising:

[0044] The acquisition module is used to acquire seabed environmental parameters collected by seabed sensors deployed in various target sea areas;

[0045] The distribution results module is used to obtain at least temperature field distribution results and biofield distribution results based on the temporal variation information of the seabed environmental parameters. The temperature field distribution results are used to quantify the ecological impact of temperature field anomalies on fish stress based on a temporal sliding window. The biofield distribution results are obtained at least by analyzing a target image, where the target image is separated from the noise region and the target region through pixel-level marking.

[0046] The life cycle module is used to determine the full life cycle information of the object to be farmed, and to determine the corresponding set of growth requirement parameters based on the full life cycle information;

[0047] The comprehensive grazing index module is used to perform spatiotemporal matching analysis on the temperature field distribution results, the biofield distribution results, and the growth requirement parameter set to calculate the comprehensive grazing index corresponding to each of the target sea areas.

[0048] The site selection decision module is used to obtain the site selection decision result for each of the target sea areas by employing a multi-objective optimization algorithm, based at least on the shipping channel information, ecological protection information and the corresponding comprehensive grazing index of each target sea area.

[0049] Compared with the prior art, the beneficial effects of the embodiments of the present invention are at least one of the following:

[0050] (1) By deploying a seabed sensor network to continuously acquire high spatiotemporal resolution environmental parameters, and combining the time-series sliding window mechanism to dynamically model the temperature field, it is possible to accurately identify short-term temperature anomalies and their potential impact on heat stress in specific fish species, effectively avoid aquaculture losses caused by sudden heat waves or cold surges, and thus significantly improve aquaculture survival rate and production stability.

[0051] (2) Overcoming the shortcomings of traditional remote sensing or manual sampling which are easily affected by turbid water, light attenuation and human error, this invention introduces underwater optical image processing technology based on pixel-level marking and image reconstruction error analysis. Through random masking and multi-round reconstruction comparison, it automatically separates the noise area from the real biological target area, which significantly improves the accuracy and robustness of biological distribution identification.

[0052] (3) The comprehensive grazing index is integrated with spatial constraints such as shipping channels and ecological protection red lines. A multi-objective optimization algorithm is used to generate site selection decision results that take into account safety, compliance and economy. The whole scheme forms a closed-loop intelligent system of "perception-modeling-matching-optimization-decision". It is not only applicable to single fish species ranch planning, but can also be extended to multi-trophic level comprehensive aquaculture scenarios, providing strong technical support for the sustainable development of scientific marine ranches. Attached Figure Description

[0053] Figure 1 This is a flowchart illustrating the marine ranching site selection method in one embodiment of the present invention;

[0054] Figure 2 This is a schematic diagram of the marine ranching site selection system in one embodiment of the present invention;

[0055] Figure label:

[0056] The modules are: 11. Acquisition module; 12. Distribution results module; 13. Life cycle module; 14. Comprehensive grazing index module; and 15. Site selection decision module. Detailed Implementation

[0057] The technical solutions of the embodiments of the present invention will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. The purpose of providing these embodiments is to make the disclosure of the present invention more thorough and comprehensive. Based on the embodiments of the present invention, all other embodiments obtained by those of ordinary skill in the art without creative effort are within the scope of protection of the present invention.

[0058] In the description of this application, the terms "first," "second," "third," etc., are used for descriptive purposes only and should not be construed as indicating or implying relative importance or implicitly specifying the number of indicated technical features. Therefore, a feature defined with "first," "second," "third," etc., may explicitly or implicitly include one or more of that feature. In the description of this application, unless otherwise stated, "a plurality of" means two or more.

[0059] In the description of this application, it should be noted that, unless otherwise expressly specified and limited, the terms "installation," "connection," and "linking" should be interpreted broadly. For example, they can refer to fixed connections, detachable connections, or integral connections; they can refer to mechanical connections or electrical connections; they can refer to direct connections or indirect connections through an intermediate medium; and they can refer to the internal communication between two components. The terms "vertical," "horizontal," "left," "right," "upper," "lower," and similar expressions used herein are for illustrative purposes only and do not indicate or imply that the device or component referred to must have a specific orientation, or be constructed and operated in a specific orientation, and therefore should not be construed as limiting the invention. The term "and / or" as used herein includes any and all combinations of one or more of the associated listed items. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0060] In the description of this application, it should be noted that, unless otherwise defined, all technical and scientific terms used in this invention have the same meaning as commonly understood by one of ordinary skill in the art. The terminology used in this specification is for the purpose of describing specific embodiments only and is not intended to limit the invention. Those skilled in the art can understand the specific meaning of the above terms in this application based on the specific circumstances.

[0061] One embodiment of the present invention provides a method for site selection in marine ranching. For details, please refer to [link / reference]. Figure 1 , Figure 1 The diagram shown is a flowchart of a marine ranching site selection method according to one embodiment of the present invention, which includes steps S1 to S5:

[0062] S1. Obtain seabed environmental parameters collected by seabed sensors deployed in various target sea areas;

[0063] S2. Based on the temporal variation information of the seabed environmental parameters, at least the temperature field distribution result and the biofield distribution result are obtained. The temperature field distribution result is based on a temporal sliding window to quantify the ecological impact of temperature field anomalies on fish stress. The biofield distribution result is obtained at least by analyzing the target image, where the target image is separated from the noise region and the target region through pixel-level marking.

[0064] S3. Determine the full life cycle information of the proposed aquaculture species, and determine the corresponding set of growth requirement parameters based on the full life cycle information;

[0065] S4. Perform a spatiotemporal matching analysis on the temperature field distribution results, the biofield distribution results, and the growth requirement parameter set to calculate the comprehensive grazing index corresponding to each of the target sea areas.

[0066] S5. Based at least on the shipping channel information, ecological protection information and the corresponding comprehensive grazing index of each of the target sea areas, a multi-objective optimization algorithm is used to obtain the site selection decision result for each of the target sea areas.

[0067] In this embodiment of the invention, the raw data are obtained through a seabed sensor network. Details of the seabed sensor network, such as model, distribution location, and quantity, need to take into account both the geographical characteristics of the current candidate sea area and the preliminary planning of aquaculture. Preferably, within the candidate target sea area, a corresponding number of seabed sensor nodes can be deployed based on seabed topographic features, water depth gradient, and historical ecological data (which can be obtained from relevant databases) to form a sensor network that covers the entire area and has spatial representativeness. Each sensor node can independently test multiple types of data, and multiple single-function sensors can also form the above-mentioned sensor node. In this embodiment of the invention, no specific limitation is made. The collected seabed environmental parameters include, but are not limited to, multi-dimensional environmental parameters such as bottom water temperature, images, salinity, dissolved oxygen concentration, pH value, redox potential, ocean current speed and direction, turbidity, chlorophyll a fluorescence intensity, nitrate / phosphate concentration, and seabed type (such as sandy, muddy, or rocky reef). In addition, since the image data needs to be processed in the subsequent embodiments of this invention, in order to ensure the accuracy of the image data, some sensor nodes can also be equipped with underwater optical imaging units or acoustic Doppler profilers (ADCP), which will not be described in detail here.

[0068] In the above embodiments, all sensor nodes are connected to shore-based or buoy data relay stations via underwater cable or wireless communication methods (such as underwater acoustic communication or LoRa long-range low-power communication) to achieve continuous, high-frequency (e.g., once every 10 minutes to 1 hour) acquisition and real-time transmission of environmental parameters. Preferably, to ensure data reliability, self-calibration and outlier detection mechanisms can be designed, such as through moving average filtering, mutation point identification, or cross-validation with neighboring node data, to eliminate invalid data caused by biological attachment, equipment drift, or transient interference. The above steps acquire high-dimensional, high-frequency, and high-reliability seabed environmental data, and provide solid data support for subsequent construction of dynamic temperature fields, accurate inversion of biofields, and achieving full life-cycle demand matching.

[0069] Furthermore, in the above embodiments, at least the biofield distribution results and temperature field distribution results need to be obtained, which will be explained one by one below.

[0070] To address the temperature field distribution results, preferably, the continuous temperature time series collected by each seabed sensor node is preprocessed to remove outliers and perform time alignment, forming a temperature observation dataset covering the entire target sea area with a unified time reference. Based on this, historical water temperature data or preliminary observation results of the target sea area are analyzed to identify its seasonal temperature change characteristics, such as the month of thermocline appearance, the duration of the lowest winter temperature, and the periods of frequent summer heat waves. According to these characteristics, the span of the sliding window is dynamically set—a shorter window (e.g., 7-15 days) is used during transitional seasons with drastic temperature changes (such as spring and autumn) to capture rapid changes, while a longer window (e.g., 30-60 days) is used during periods of stable temperature changes (such as midsummer or midwinter) to enhance trend stability, thereby achieving adaptive adjustment of the window length.

[0071] Then, for the temperature sequence within each sliding window, a multi-scale signal decomposition method (such as wavelet transform, empirical mode decomposition (EMD), or variational mode decomposition (VMD) in existing technologies) is used to decompose it into multiple intrinsic modal components, clearly separating the high-frequency disturbance components reflecting short-term disturbances (such as instantaneous warming / cooling caused by tides and storms) from the low-frequency trend components characterizing long-term climate trends (such as seasonal warming and warming background). Subsequently, these components are input into nonlinear time series prediction models (such as long short-term memory networks (LSTM), gated recurrent units (GRU), or Transformers in existing technologies) for joint modeling to construct the corresponding temperature prediction model. The performance of this model is affected by the various algorithms listed in parentheses above, resulting in varying functional effects, but it is necessary to ensure that the water temperature evolution path is predicted over a certain period of time in the future.

[0072] Considering that the core of this scheme is to achieve scientific aquaculture, a comprehensive analysis of each aquaculture species is required. Taking fish as an example, it is necessary to obtain a database of heat stress physiological thresholds for each fish species (this database contains the tolerance limits of different life stages to critical high temperatures, duration of high temperatures, and rate of warming, and can be accessed through existing websites such as https: / / www.ebiotrade.com / ). By coupling the predicted water temperature evolution path over a certain period of time with this threshold, the probability and intensity of heat stress during the prediction period can be obtained, thereby generating a stress risk heat map with spatial resolution. Preferably, this heat map represents the risk level of each region in the form of a color gradient or a numerical matrix, with high-risk areas (such as areas that frequently exceed the threshold) being prominently marked, thus obtaining the temperature field distribution results.

[0073] Due to the complexity of the marine environment, image data is affected by factors such as suspended particles, light scattering, biological attachment, equipment vibration, and low illumination. This results in a large number of noisy areas (such as blurred patches, halos, artifacts, or non-biological interference) mixed among real biological targets (such as algae, shellfish, sea cucumbers, and fish communities), severely affecting the accuracy of subsequent biomass estimation and distribution identification. Therefore, as mentioned above, the embodiments of the present invention require a series of processing steps on the image data to ensure the final location accuracy. Furthermore, in the embodiments of the present invention, to effectively separate noise, the original underwater optical image data is first divided into blocks, into several pixel blocks (the division criteria are based on the convenience of subsequent processing, and can also be a fixed area division, which is not limited in the embodiments of the present invention). Subsequently, based on the pixel dependency intensity of local regions of the image, which is the spatial correlation between neighboring pixels, preferably obtained through local autocorrelation function, structural tensor, or graph neural network modeling, the random mask processing can be determined. The intensity and number of missing sub-images are considered. For example, in areas with complex textures and dense biological populations (high intensity dependence), fewer but more refined mask sub-images are generated; in smooth or noise-dominated areas (low intensity dependence), more random missing samples are generated to enhance robustness. Then, each missing sub-image (i.e., an image in which some pixels are randomly set to zero or occluded) is input into a pre-trained image reconstruction model, which can be a diffusion model, etc. Through the processing of this model, the missing parts in the image can be filled in and reconstructed. Of course, the underwater sample image dataset on which the training is based can be typical data or historical data, which is not limited in the embodiments of this invention.

[0074] After the above processing, the reconstructed image is compared pixel-by-pixel with the original complete image, and the reconstruction error of each pixel block (such as mean squared error (MSE), structural similarity difference (SSIM), or perceptual loss) is calculated. Real biological targets (i.e., target regions) are usually accurately reconstructed after masking due to their stable spatial structure and semantic consistency, with relatively small errors; while noisy regions, lacking semantic regularity, have significantly higher reconstruction errors. Therefore, high-error regions can be marked with noise to form noise masks or noise labels. Then, logical operations are used to remove or zero out this mask from the underwater optical image, thereby separating high-confidence target regions and obtaining the biofield distribution results. This embodiment, through special image design, can effectively identify target areas in images. Compared with existing denoising techniques, this embodiment retains more target area information in the scene, avoiding repeated denoising or distortion caused by denoising. This eliminates the influence of irrelevant areas in underwater optical images. Furthermore, since the image data collected by the aforementioned sensor network in the target sea area is enormous, separating and removing irrelevant factors from the image, and only needing to process the target area, also helps to improve the processing efficiency of the entire image dataset.

[0075] Furthermore, in the above embodiments, the random masking process is not a single process, but needs to be repeated K times. This is to reduce the errors caused by different masking strategies. Specifically, it needs to be processed K times (K is a positive integer), and each processing executes a different masking strategy for the original image or its sub-regions to fully obtain the structural information and noise features inside the image. In this way, details and patterns that are difficult to discover using a single masking strategy can be identified more accurately.

[0076] In the m-th (1≤m≤K) masking process, preferably, the spatial dependency between each pixel and its neighboring pixels needs to be considered. This dependency reflects the degree of correlation between pixels within a local region of the image, which helps determine which pixel blocks may contain important information (such as the outline or texture of an organism) and which may be noise. This process can be achieved using a local autocorrelation function; high autocorrelation values ​​are typically used to characterize a region with strong intrinsic structure (e.g., part of an organism), while low autocorrelation values ​​indicate noise or other unstructured elements. Alternatively, structural tensors can be used to identify features such as lines and edges in the image, thereby guiding mask design and protecting these critical structures from being affected. Graph neural network technology can also be used to treat the image as a graph composed of nodes (pixels) and edges (relationships between pixels). By learning from this graph, complex local structural features can be captured, and the masking operation can be determined accordingly. Therefore, in the above embodiment, in a dense biological region, the graph neural network may suggest preserving more complete pixel blocks for subsequent reconstruction; while in noise-dominated regions, a larger mask range will be recommended. Each masking process described above can more accurately locate key information and potential noise in the image, thereby optimizing the subsequent image reconstruction process.

[0077] Considering the huge demand for fish farming in marine ranches, in this embodiment of the invention, the target species to be farmed is a specific fish species. This embodiment of the invention achieves quantitative processing of growth requirements by dividing different life cycle stages. Of course, different life cycle stages can also be divided for other non-fish farming species, which will not be described one by one.

[0078] The inventors discovered through research that different stages of fish have different environmental requirements. Based on the significant transition points of their physiological and metabolic characteristics, this embodiment divides the fish into the hatching period, juvenile period, young fish period, adult fish period, and breeding period. The relevant information is shown in Table 1 below (State and Information of Fish Life Stages).

[0079] Table 1. Status and Information of Fish Life Stages

[0080]

[0081] For each life stage, this embodiment quantifies its dynamic response to key environmental factors. Preferably, dynamic response curves for multidimensional growth requirements can be constructed by integrating laboratory controlled environment experimental data (such as survival rates under different temperature gradients and feeding rates at specific dissolved oxygen concentrations), in-situ field observation records (such as the distribution density of natural populations in different flow velocity areas), and authoritative aquatic databases (such as FAO FishBase, a standard of the Chinese Academy of Fishery Sciences). These curves precisely describe the tolerance range, optimal range, and critical threshold of various life stages to parameters such as water temperature gradient (e.g., the optimal temperature for juveniles is 18-22℃, and stress occurs when the temperature exceeds 25℃), dissolved oxygen concentration (e.g., >5 mg / L for adults and >6 mg / L for breeding), photoperiod (e.g., simulating natural photoperiods to induce gonad development during breeding), feed particle size and nutrient composition (e.g., 50-100 μm rotifers for juveniles and 200-500 μm formulated feed with a protein content >45% for young fish), and water flow velocity (e.g., <5 cm / s during hatching to avoid eggs sinking to the bottom and causing oxygen deficiency, and 10-30 cm / s for adults to promote feeding).

[0082] To obtain the set of growth requirement parameters, the aforementioned nonlinear and fuzzy biological responses are transformed into calculable quantitative results. Specifically, this invention introduces a fuzzy comprehensive evaluation model, which maps each dynamic response curve to a membership function (such as a triangular, Gaussian, or S-shaped function) to characterize the membership degree (between 0 and 1) of a certain environmental value to a specific life stage. For example, when the water temperature is 20℃, the membership degree for the juvenile stage may be 0.95 (highly suitable), while for the reproductive stage it may be only 0.3 (unsuitable). Subsequently, adjustments are made according to the different needs of marine ranching users, such as pursuing high growth rate, high disease resistance, or high reproductive success rate. For example, if the goal is rapid breeding and market entry, the weight of the juvenile and adult stages is higher; if the goal is germplasm resource protection, the weight of the reproductive stage is prominent. Finally, by weighted fusion of the membership functions of each stage, a set of growth requirement parameters covering the entire life cycle and adaptable to different time periods is generated.

[0083] Furthermore, in the above embodiments, comprehensive analysis is required. Considering that the data from the sensor network may not cover all dimensions, this embodiment introduces a three-dimensional spatiotemporal model to facilitate accurate subsequent analysis.

[0084] First, the temperature field distribution results, biofield distribution results, and growth requirement parameter set mentioned above are standardized and spatiotemporally aligned. Then, these data are uniformly mapped to a four-dimensional coordinate system with longitude (x), latitude (y), and water depth (z) as spatial dimensions and time (t) as the fourth dimension, constructing a high-resolution three-dimensional spatiotemporal model. This three-dimensional spatiotemporal model is actually a dynamic cube combining 3D and time dimensions.

[0085] In this model, each voxel not only contains spatial location information but also carries the multidimensional environmental state and adaptation requirements of that location at a specific point in time, forming a dynamically evolving digital marine environment twin. Of course, in addition to the aforementioned three-dimensional spatiotemporal model, other display models of different scales can be constructed according to actual needs, which will not be elaborated further.

[0086] Based on the aforementioned three-dimensional spatiotemporal model, the spatiotemporal evolution of the target sea area over a future period can be simulated and predicted. For example, it can simulate how spring upwelling alters bottom temperature and nutrient distribution, thereby affecting the timing and location of phytoplankton blooms; or predict how summer heat waves cause a sudden drop in dissolved oxygen in a certain water layer, thus conflicting with the needs of fish during their breeding season. Preferably, in this embodiment, extreme environmental events are introduced to achieve a more realistic simulation, which will be described in detail below. Finally, the change information in the above spatiotemporal evolution results is quantified to obtain an accurate comprehensive susceptibility index. Preferably, the comprehensive susceptibility index is obtained by fusing the above multiple indicators. The fusion process can employ methods such as weighted summation, TOPSIS multi-criteria decision-making, or neural network regression. The value of this comprehensive susceptibility index is used to characterize the overall susceptibility of the sea area to the target fish species during that period.

[0087] In addition, to achieve comprehensive data display, time integration or stability analysis can be performed on the index for each period of the year or the breeding cycle, which can further generate an annual comprehensive grazing index or a seasonal recommended level map, which will not be elaborated further in this embodiment.

[0088] In addition, the comprehensive aquaculture index also needs to be adjusted according to the actual situation. Taking the planktonic benthic organism regeneration rate and nutrient cycling flux in the biofield distribution results as examples, the two together reflect the self-purification capacity, food supply potential and material cycling efficiency of the target sea area ecosystem. They are the core indicators for measuring whether it can support aquaculture activities. Based on this, the ecological carrying capacity threshold function can be obtained. For example, the ecological carrying capacity threshold is equal to the first coefficient multiplied by the planktonic benthic organism regeneration rate, plus the second coefficient multiplied by the nutrient cycling flux. The first and second coefficients are weighted coefficients calibrated according to the trophic level, excretion characteristics and ecosystem sensitivity of the aquaculture objects.

[0089] During spatiotemporal matching analysis, the matching degree between the expected breeding density and the ecological carrying capacity threshold function is compared in real time, and the breeding load is determined based on the matching degree result. This load is the actual load. If the overload condition is found, the current comprehensive grazing index needs to be corrected in order to obtain an accurate comprehensive grazing index.

[0090] The above embodiments mention extreme environmental events. Preferably, the first step is to systematically identify and quantify the geographic environmental parameters of the target sea area. These geographic environmental parameters include, but are not limited to: seabed topography slope and landform type (such as continental shelf, trench, reef), water depth distribution, shoreline tortuosity, distance from river mouth or land-based input, location of ocean current main axis, historical typhoon path density, red tide high-incidence area indicators, and active areas of ocean fronts or upwelling. Geographic environmental parameters can be obtained from relevant public websites and spatially rasterized to form structured geographic feature vectors. Then, based on the above geographic environmental parameters, statistical modeling or machine learning methods (such as Poisson regression, random forest, or LSTM time series models) are used to determine the historical frequency and spatial probability distribution of various extreme environmental events in this sea area. For example, typical extreme events include: marine heat waves (surface temperatures exceeding the 90th percentile seasonal threshold for more than 5 consecutive days), severe storm events (accompanied by large waves, strong currents, and bottom disturbances), hypoxic events (dissolved oxygen <2 mg / L for several days), harmful algal blooms (such as Karenia mikimotoi and Heterosigma rubiginosa blooms), and sudden salinity changes (triggered by heavy rainfall or river floods). For instance, bays near estuaries and with enclosed topography may have a higher frequency of hypoxic events and red tides; while the sea areas located in the typhoon corridor of the western Pacific have a significantly higher frequency of storm disturbances. This frequency of occurrence is not only expressed as an annual average but can also be refined to seasonal probabilities (e.g., 80% probability of summer heat waves and 60% probability of winter storms). Then, according to the determined frequency of occurrence, the created environmental simulation events are dynamically applied to a three-dimensional spatiotemporal model. For example, when simulating the evolution over the next 90 days, a heat wave event is injected on day 20 with probability, and a hypoxic disturbance is superimposed on day 50 to form a composite stress scenario.

[0091] Through the above process, the response results of the three-dimensional spatiotemporal model under the applied environmental simulation events can be monitored and recorded in real time, thereby obtaining the spatiotemporal evolution results of the target sea area. In other words, this process generates a spatiotemporal evolution result that is closer to the complexity of the real ocean by integrating the disturbance response information into the natural evolution trend under the undisturbed state.

[0092] To further reflect the complex real-world marine environment, in the above embodiments, at least two environmental simulation events are applied. The application time of each event strictly meets preset conditions. For example, the heat wave event is set to be applied on the 15th to 25th day of the natural rise in water temperature during summer, and the storm event is set to be triggered on the 30th to 35th day of the typhoon season. The event time windows are allowed to partially overlap (such as a storm occurring during a heat wave) to simulate typical scenarios of multiple disasters occurring simultaneously in the real ocean.

[0093] Of course, there will be mutual influences between various environmental simulation events, leading to intensity attenuation. To quantify this mutual influence, preferably, the mutual reinforcement, weakening, or independent relationship between any two environmental simulation events in terms of intensity, duration, and spatial range can be evaluated first, and then the attenuation index can be determined. For example, the strong mixing effect caused by a storm may alleviate surface heat waves, and heat waves may also reduce dissolved oxygen saturation in water bodies. Within the application time window of the environmental simulation event, the response result is dynamically corrected based on the above attenuation index. Specifically, the initial response value can be multiplied (or divided) by the corresponding attenuation index, or it can be input into a coupled response function for nonlinear adjustment, thereby obtaining a corrected response result that better reflects the reality of multi-disturbance coupling. Through the above process, this embodiment breaks through the limitations of traditional single-event simulation, realizes multi-hazard, dynamic coupling, and effect correction formal assessment of environmental risks in marine ranch site selection, significantly improves the reliability and foresight of the comprehensive ranching index under extreme climate backgrounds, and provides a scientific basis for constructing a highly resilient and interference-resistant marine ranch layout.

[0094] Furthermore, in the above embodiments, it is necessary to obtain accurate biofield distribution results from image data. In order to achieve accurate biometric identification in the image, it can be distinguished by "vortex structure". The principle is that biological objects, such as plankton, move more frequently, so the vortex frequency they generate is higher; while the intensity of underwater impurity vortices is stronger and the frequency is lower. Therefore, biological vortices and non-biological vortices can be identified by reconstructing the frequency domain information in the image. Specifically, it includes the following steps S11~S13:

[0095] S11. Decompose each sub-region in the image to be identified into concentric rings;

[0096] S12. Calculate the gradient direction distribution histogram of all pixels in each ring, and combine them based on the corresponding concentric ring order to obtain the ring feature vector.

[0097] S13. Based on the analysis results of the annular feature vector, the corresponding sub-region is determined as the eddy region, and the biological eddy region is determined based on the frequency domain information of the eddy region.

[0098] In step S11 above, the raw image acquired by the underwater optical sensor is divided into several local sub-regions (e.g., divided using a sliding window or grid method, with each sub-region having a size of 64×64 or 128×128 pixels) to focus on analyzing local flow field or biological motion characteristics. For each sub-region, a set of concentric rings is constructed with its geometric center as the origin, i.e., multiple annular regions with the same center but different radii. For example, the sub-region is divided into 3 to 5 rings from the center outwards, with each ring covering a certain radial distance (e.g., ring 1: 0–10 pixels, ring 2: 10–20 pixels, and so on). Since vortices typically exhibit rotating flow around a central point, with their velocity and structure changing radially in a gradient manner, this annular division method can match the physical morphology of vortices.

[0099] In step S12 above, within each concentric ring, the image gradient direction of all pixels is calculated, that is, the direction of local brightness change is obtained (usually quantized into several bins within the range of 0°~180° or 0°~360°, such as 8 or 16 directions). Subsequently, a histogram of the gradient direction distribution within the ring is plotted, which reflects the dominant directional pattern of texture or edge within the ring.

[0100] Because vortex structures exhibit near-circular rotational symmetry under ideal conditions, their gradient directions show continuous and smooth directional changes (e.g., increasing clockwise or counterclockwise) along the rings, rather than a random or linear distribution. Therefore, by sequentially splicing the rings in a concentric order from the inside out, a high-dimensional ring feature vector is formed. This vector not only preserves radial structural information but also possesses invariance to overall image rotation. Even if the vortex region rotates within the image, its ring gradient distribution pattern remains consistent, thus significantly improving recognition robustness.

[0101] In step S13 above, the ring feature vector corresponding to each sub-region is input into a pre-trained classification model (such as Support Vector Machine (SVM), Random Forest, or Lightweight Neural Network). This model has been trained on a large amount of labeled data (including real eddy and non-eddy samples) and can determine whether the sub-region belongs to an eddy region. The judgment criteria include, but are not limited to, whether the ring feature vector exhibits significant directional continuity, the phase consistency of the gradient directions between each ring, and whether the overall entropy value is lower than the random noise threshold.

[0102] Furthermore, to distinguish between abiotic eddy regions (such as non-biological eddies caused by topography or tidal currents) and biotic eddy regions, this embodiment introduces frequency domain information analysis. Specifically, the time-series image frames or multispectral / polarization response features of the identified eddy regions are extracted, and short-time Fourier transform or wavelet time-frequency analysis is performed to observe whether they have biological rhythmic frequency components (such as the tail-wagging frequency of fish schools at 0.5~3 Hz, and the periodicity of diurnal vertical migration of plankton). If there are significant frequency peaks in the frequency domain that match known biological activity characteristics, it is determined to be a biotic eddy region; otherwise, it is classified as an abiotic hydrodynamic eddy.

[0103] The above process, through frequency domain information analysis, reasonably distinguishes between biological eddies and non-biological eddies, achieving accurate identification of image data. In addition, the identified "biological eddy region" can serve as an important component of the above biofield distribution results, used to indicate areas of high biological activity, potential fish gathering areas, or ecological hotspots, providing key ecological location information for marine ranch site selection.

[0104] Another embodiment of the present invention provides a marine ranching site selection system; for details, please refer to [link to relevant documentation]. Figure 2 , Figure 2 The diagram shown is a structural block diagram of a marine ranching site selection system according to one embodiment of the present invention, which includes:

[0105] The acquisition module 11 is used to acquire seabed environmental parameters collected by seabed sensors deployed in various target sea areas.

[0106] The distribution result module 12 is used to obtain at least temperature field distribution results and biofield distribution results based on the temporal variation information of the seabed environmental parameters. The temperature field distribution results are used to quantify the ecological impact of temperature field anomalies on fish stress based on a temporal sliding window. The biofield distribution results are obtained at least by analyzing a target image, where the target image is separated from the noise region and the target region by pixel-level marking.

[0107] The life cycle module 13 is used to determine the full life cycle information of the object to be farmed, and to determine the corresponding set of growth requirement parameters based on the full life cycle information;

[0108] The comprehensive grazing index module 14 is used to perform spatiotemporal matching analysis on the temperature field distribution results, the biofield distribution results, and the growth requirement parameter set to calculate the comprehensive grazing index corresponding to each of the target sea areas.

[0109] The site selection decision module 15 is used to obtain the site selection decision result for each of the target sea areas by using a multi-objective optimization algorithm, based at least on the shipping channel information, ecological protection information and the corresponding comprehensive grazing index of each target sea area.

[0110] The marine ranching site selection method and system provided in this invention first acquires seabed environmental parameters collected by seabed sensors deployed in various target sea areas; then, based on the temporal variation information of the seabed environmental parameters, at least temperature field distribution results and biofield distribution results are obtained. The temperature field distribution results quantify the ecological impact of temperature field anomalies on fish stress based on a temporal sliding window; the biofield distribution results are obtained at least by analyzing target images, where pixel-level marking separates noise and target regions. This invention continuously acquires high spatiotemporal resolution environmental parameters by deploying a seabed sensor network. This invention, combined with a time-series sliding window mechanism, dynamically models the temperature field, enabling accurate identification of short-term temperature anomalies and their potential impact on heat stress in specific fish species. This effectively avoids aquaculture losses caused by sudden heat waves or cold surges, thereby significantly improving survival rates and production stability. Furthermore, this invention overcomes the limitations of traditional remote sensing or manual sampling, which are susceptible to turbid water, light attenuation, and human error. It introduces underwater optical image processing technology based on pixel-level labeling and image reconstruction error analysis. Through random masking and multi-round reconstruction comparison, it automatically separates noise regions from real biological target regions, significantly improving the accuracy and robustness of biological distribution identification. Next, the entire life cycle information of the proposed aquaculture species is determined, and the corresponding set of growth requirement parameters is determined based on this information. Then, the temperature field distribution results, the biological field distribution results, and the growth requirement parameter set are spatiotemporally matched and analyzed to calculate the comprehensive aquaculture index for each target sea area. Finally, based at least on the shipping channel information, ecological protection information, and the corresponding comprehensive aquaculture index for each target sea area, a multi-objective optimization algorithm is used to obtain the site selection decision result for each target sea area. The entire process integrates the comprehensive grazing index with spatial constraints such as shipping channels and ecological protection red lines. It uses a multi-objective optimization algorithm to generate site selection decision results that take into account safety, compliance and economy. The whole solution forms a closed-loop intelligent system of "perception-modeling-matching-optimization-decision". It is not only applicable to single fish species ranch planning, but can also be extended to multi-trophic level integrated aquaculture scenarios, providing strong technical support for the sustainable development of scientific marine ranches.

[0111] The embodiments described above are merely illustrative of several implementations of the present invention, and while the descriptions are specific and detailed, they should not be construed as limiting the scope of the present invention. It should be noted that those skilled in the art can make various modifications and improvements without departing from the concept of the present invention, and these modifications and improvements all fall within the scope of protection of the present invention. Therefore, the scope of protection of this patent should be determined by the appended claims.

Claims

1. A method for site selection in marine ranching, characterized in that, include: Acquire seabed environmental parameters collected by seabed sensors deployed in various target sea areas; Based on the temporal variation information of the seabed environmental parameters, at least the temperature field distribution results and the biofield distribution results are obtained. The temperature field distribution results are quantified based on a temporal sliding window to assess the ecological impact of temperature field anomalies on fish stress. The biofield distribution results are obtained at least by analyzing a target image, where the target image is separated from the noise region by pixel-level labeling. Determine the full life cycle information of the proposed aquaculture species, and determine the corresponding set of growth requirement parameters based on the full life cycle information; The temperature field distribution results, biofield distribution results, and growth requirement parameter set are subjected to spatiotemporal matching analysis to calculate the comprehensive aquaculture index corresponding to each of the target sea areas. This includes: mapping the temperature field distribution results, biofield distribution results, and growth requirement parameter set to a coordinate system with latitude, longitude, water depth, and time as dimensions to obtain a three-dimensional spatiotemporal model; obtaining the spatiotemporal evolution results of the target sea areas based on the three-dimensional spatiotemporal model; and quantifying the change information of the spatiotemporal evolution results to obtain the comprehensive aquaculture index corresponding to each of the target sea areas. The quantification process is configured to fuse multiple indicators to obtain the change information of the spatiotemporal evolution results. The comprehensive aquaculture index is used to characterize the comprehensive suitability of the target sea areas for target fish species. Based at least on the shipping channel information, ecological protection information and the corresponding comprehensive grazing index of each of the target sea areas, a multi-objective optimization algorithm is used to obtain the site selection decision result for each of the target sea areas.

2. The marine ranching site selection method as described in claim 1, characterized in that, Based on the time-series variation information of the seabed environmental parameters, at least the temperature field distribution results are obtained, including: The span of the sliding window is determined based on the seasonal temperature variation characteristics of the target sea area. The temperature sequence within the span is sequentially subjected to multi-scale decomposition and nonlinear trend modeling to obtain a temperature prediction model; wherein, the multi-scale decomposition includes extracting the high-frequency perturbation component and the low-frequency trend component of the temperature sequence. The temperature anomaly events output by the temperature prediction model are coupled with the heat stress threshold of the selected aquaculture objects to obtain a stress risk heat map. The temperature field distribution results are obtained based at least on the stress risk heat map.

3. The marine ranching site selection method as described in claim 2, characterized in that, The time-series variation information based on the seabed environmental parameters yields at least the biofield distribution results, including: The pixel blocks in the underwater optical image acquired by the seabed sensor are randomly masked to generate several missing sub-images. The number of missing sub-images is determined based on the local pixel dependency strength of the underwater optical image. Each of the missing sub-images is input into a pre-trained image reconstruction model for image reconstruction, and the reconstruction error of each pixel block is calculated based on the comparison results between the image reconstruction results and the underwater optical images. Based on the reconstruction error, at least the noisy regions of the underwater optical image are marked; The marked noise region is separated from the underwater optical image to obtain the biofield distribution result with the target region.

4. The marine ranching site selection method as described in claim 3, characterized in that, The random masking process includes K iterations, where K is a positive integer; In the m-th process of the K processes, the pixel block to be masked is designed to be determined based on the spatial dependency between each pixel in the underwater optical image and its neighboring pixels. The spatial dependency is obtained at least through a local autocorrelation function, a structural tensor, or a neighborhood relationship modeling method based on a graph neural network, where 1≤m≤K.

5. The marine ranching site selection method as described in claim 1, characterized in that, The species to be cultured are specific fish. The process of determining the full life cycle information of the proposed aquaculture species and determining the corresponding set of growth requirement parameters based on the full life cycle information includes: Based on the physiological and metabolic characteristics of the specific fish at different developmental stages, various life stages are divided, including at least the hatching period, juvenile period, young fish period, adult fish period, and reproductive period. Quantify the dynamic response curve of each life stage to growth requirement information, wherein the growth requirement information includes at least water temperature gradient, dissolved oxygen concentration, photoperiod, feed particle size and nutrient composition and water flow velocity; Based on the fuzzy comprehensive evaluation model, the dynamic response curves corresponding to each life stage are transformed into membership functions, and weighted fusion is performed based on the target weights selected by the aquaculture users to obtain the set of growth requirement parameters.

6. The marine ranching site selection method as described in claim 1, characterized in that, The spatiotemporal evolution results of the target sea area obtained based on the three-dimensional spatiotemporal model include: Determine the geographical environmental parameters of the target sea area; Based on the aforementioned geographical environmental parameters, the frequency of occurrence of various extreme environmental events corresponding to the target sea area is determined; Based on the frequency of occurrence, the created environmental simulation events are applied to the three-dimensional spatiotemporal model; The spatiotemporal evolution results of the target sea area are obtained based at least on the response results of the three-dimensional spatiotemporal model.

7. The marine ranching site selection method as described in claim 6, characterized in that, The number of environmental simulation events applied is at least two, and the application time of each environmental simulation event meets a preset requirement; During the process of applying each of the environmental simulation events to the three-dimensional spatiotemporal model, the attenuation index of each environmental simulation event is determined according to the degree of correlation between the various environmental simulation events. During the application time, the initial response result of the three-dimensional spatiotemporal model is corrected based at least on the decay index to obtain the response result.

8. The marine ranching site selection method as described in claim 1, characterized in that, The biofield distribution results include at least the regeneration rate of planktonic benthic organisms and the nutrient cycling flux; During the spatiotemporal matching analysis, the matching degree between the expected aquaculture density and the ecological carrying capacity threshold function is compared in real time, and the aquaculture load is determined based on the matching degree result; the ecological carrying capacity threshold function is determined by the planktonic benthic organism regeneration rate and the nutrient cycling flux. If the aquaculture load is detected to exceed the threshold, the current comprehensive aquaculture index will be subjected to non-linear decay processing.

9. A marine ranching site selection system, characterized in that, include: The acquisition module is used to acquire seabed environmental parameters collected by seabed sensors deployed in various target sea areas; The distribution results module is used to obtain at least temperature field distribution results and biofield distribution results based on the temporal variation information of the seabed environmental parameters. The temperature field distribution results are used to quantify the ecological impact of temperature field anomalies on fish stress based on a temporal sliding window. The biofield distribution results are obtained at least by analyzing a target image, where the target image is separated from the noise region and the target region through pixel-level marking. The life cycle module is used to determine the full life cycle information of the object to be farmed, and to determine the corresponding set of growth requirement parameters based on the full life cycle information; The comprehensive aquaculture index module is used to perform spatiotemporal matching analysis on the temperature field distribution results, the biofield distribution results, and the growth requirement parameter set to calculate the comprehensive aquaculture index corresponding to each of the target sea areas. This includes: mapping the temperature field distribution results, the biofield distribution results, and the growth requirement parameter set to a coordinate system with latitude, longitude, water depth, and time as dimensions to obtain a three-dimensional spatiotemporal model; obtaining the spatiotemporal evolution results of the target sea areas based on the three-dimensional spatiotemporal model; and quantifying the change information of the spatiotemporal evolution results to obtain the comprehensive aquaculture index corresponding to each of the target sea areas. The quantification process is configured to fuse multiple indicators to obtain the change information of the spatiotemporal evolution results. The comprehensive aquaculture index is used to characterize the overall suitability of the target sea area for the target fish species. The site selection decision module is used to obtain the site selection decision result for each of the target sea areas by employing a multi-objective optimization algorithm, based at least on the shipping channel information, ecological protection information and the corresponding comprehensive grazing index of each target sea area.

Images