AI-assisted mangrove state diagnosis method and system

Abstract

The application discloses a mangrove state diagnosis method and system based on AI assistance, which comprises the following steps: first, performing first feature extraction on remote sensing data of a mangrove to be diagnosed to obtain remote sensing visual features; for each mangrove state of a final classification unit, integrating its own soft prompt, the soft prompt of an upper classification unit and state information to construct mangrove state information; performing second feature extraction on the mangrove state information to obtain state features; calculating matching coefficients of the remote sensing visual features and the state features; and determining a diagnosis conclusion according to the mangrove state corresponding to the state features with the largest matching coefficients. The present application realizes accurate and hierarchical diagnosis of mangrove states by fusing ecological prior knowledge through multi-level soft prompts, combining a double-feature extraction and matching mechanism, and providing technical support for coastal ecological protection and restoration decisions.

Description

TECHNICAL FIELD

[0001] The present application relates to the technical field of artificial intelligence, in particular to an AI-assisted mangrove state diagnosis method and system. BACKGROUND

[0002] As a key coastal ecosystem, mangrove plays an irreplaceable ecological function in carbon fixation and storage, coastal protection, and biodiversity maintenance. Its state diagnosis is the core basis of ecological protection and management. The existing mangrove state diagnosis methods have significant limitations: traditional manual plot investigation is inefficient and has insufficient spatial coverage, making it difficult to quickly obtain regional-scale data; conventional remote sensing classification methods rely only on single-level spectral or texture features, resulting in deviations in the accuracy of details and consistency of ecological levels in the diagnosis results. SUMMARY

[0003] The present application relates to the technical field of artificial intelligence, in particular to an AI-assisted mangrove state diagnosis method and system.

[0004] In a first aspect, an AI-assisted mangrove state diagnosis method is provided, comprising:

[0005] Performing a first feature extraction operation on the remote sensing data of the mangrove to be diagnosed to obtain remote sensing visual features;

[0006] For each mangrove state in the plurality of mangrove states at the last-level classification unit, constructing a mangrove state information from the soft hint of the mangrove state, the soft hint of the mangrove state of the upper-level classification unit of the mangrove state, and the mangrove state;

[0007] Performing a second feature extraction operation on the mangrove state information to obtain state features;

[0008] Determining the matching coefficients of the remote sensing visual features and the plurality of state features;

[0009] Based on the mangrove state corresponding to the state feature with the largest matching coefficient in the plurality of state features, determining the mangrove state corresponding to the remote sensing data of the mangrove to be diagnosed.

[0010] In a possible implementation, determining the mangrove state corresponding to the remote sensing data of the mangrove to be diagnosed based on the mangrove state corresponding to the state feature with the largest matching coefficient in the plurality of state features comprises:

[0011] Obtaining the mangrove state corresponding to the state feature with the largest matching coefficient, and the mangrove state of the upper-level classification unit of the mangrove state;

[0012] The mangrove state corresponding to the state feature with the largest matching coefficient and the mangrove state of the upper classification unit of the mangrove state are jointly used as the mangrove state corresponding to the to-be-diagnosed mangrove remote sensing data.

[0013] In a possible implementation, the mangrove state corresponding to the state feature with the largest matching coefficient is determined as the mangrove state corresponding to the to-be-diagnosed mangrove remote sensing data.

[0014] In a possible implementation, the mangrove state corresponding to the state feature with the largest matching coefficient is determined as the mangrove state corresponding to the to-be-diagnosed mangrove remote sensing data.

[0015] In a possible implementation, the mangrove state diagnosis method is implemented based on a pre-trained mangrove state recognition model.

[0016] In a possible implementation, the mangrove state exists in multiple classification units, and for any two adjacent classification units, a first classification unit and a second classification unit, in the multiple classification units, the first classification unit exists in N mangrove states, the second classification unit exists in M mangrove states, and one or more mangrove states in the M mangrove states exist in the N mangrove states as upper mangrove states, and the N is not more than the M.

[0017] Before constructing the mangrove state information of each mangrove state in the multiple mangrove states in the last-level classification unit from the soft hint of the mangrove state, the soft hint of the mangrove state of the upper classification unit of the mangrove state, and the mangrove state, the method further includes:

[0018] Performing a first feature extraction operation on the sample mangrove remote sensing data to obtain sample remote sensing visual features;

[0019] For each classification unit in the multiple classification units, based on the sample state features and the sample remote sensing visual features corresponding to the multiple mangrove states of the classification unit, a state confidence sequence corresponding to each mangrove state of the classification unit is calculated.

[0020] Based on the state confidence sequence corresponding to each mangrove state of the classification unit, a first error parameter corresponding to the classification unit is calculated.

[0021] For the second classification unit in the multiple classification units, based on the state confidence sequence corresponding to the multiple mangrove states of the second classification unit and the state confidence sequence corresponding to the upper mangrove state of each mangrove state of the second classification unit, a second error parameter of the first classification unit is calculated.

[0022] determine a comprehensive error variable based on the first error variable and the second error variable;

[0023] adjust the parameters of the soft hints corresponding to the plurality of mangrove states of each classification unit in the plurality of classification units based on the comprehensive error variable until a training termination condition is met, to obtain a trained mangrove state recognition model.

[0024] In a possible implementation, the calculation of the state confidence sequence corresponding to the plurality of mangrove states of each classification unit in the plurality of classification units based on the sample state features corresponding to the plurality of mangrove states of the classification unit and the sample remote sensing visual features includes:

[0025] determining the sample state features corresponding to the plurality of mangrove states of the classification unit based on the classification unit level order of the classification unit in the plurality of classification units;

[0026] calculating the state confidence sequence corresponding to the plurality of mangrove states of the classification unit based on the sample state features corresponding to the plurality of mangrove states of the classification unit, the sample remote sensing visual features, and the ecological sensitivity adjustment factor.

[0027] In a possible implementation, the determination of the sample state features corresponding to the plurality of mangrove states of the classification unit based on the classification unit level order of the classification unit in the plurality of classification units includes:

[0028] if the classification unit is the topmost classification unit in the plurality of classification units, for each mangrove state in the plurality of mangrove states of the classification unit, constructing sample mangrove state information from the soft hint of the mangrove state and the mangrove state;

[0029] performing a second feature extraction operation on the sample mangrove state information to obtain sample state features.

[0030] In a possible implementation, the determination of the sample state features corresponding to the plurality of mangrove states of the classification unit based on the classification unit level order of the classification unit in the plurality of classification units further includes:

[0031] if the classification unit is a non-topmost classification unit in the plurality of classification units, for each mangrove state in the plurality of mangrove states of the classification unit, constructing sample mangrove state information from the soft hint of the mangrove state, the soft hint of the mangrove state of the upper classification unit of the mangrove state, and the mangrove state;

[0032] The sample mangrove state information is subjected to a second feature extraction operation to obtain sample state features.

[0033] In a possible implementation, the second error parameter of the first classification unit is calculated based on the state confidence sequence corresponding to each mangrove state of the second classification unit and the state confidence sequence corresponding to the upper mangrove state of each mangrove state of the second classification unit, and the calculation includes:

[0034] A reference classification unit of the first classification unit is established, the number of mangrove state categories of the reference classification unit is the same as the number of mangrove state categories of the second classification unit, and each mangrove state of the reference classification unit is matched with each mangrove state of the second classification unit;

[0035] For each mangrove state of the second classification unit, the state confidence sequence of the upper mangrove state of the mangrove state is determined based on the state confidence sequence of the upper mangrove state of the mangrove state.

[0036] The second error parameter of the first classification unit is calculated based on the state confidence sequence of each mangrove state of the second classification unit and the state confidence sequence of the corresponding mangrove state of the reference classification unit.

[0037] In a second aspect, an embodiment of the present application provides a server system, and the readable storage medium includes a computer program, and the computer program controls a computer device where the readable storage medium is located to execute the method of the first aspect when running.

[0038] Compared with the prior art, the present application provides the following beneficial effects: by using the AI-assisted mangrove state diagnosis method and system disclosed in the present application, remote sensing visual features are obtained by performing first feature extraction on remote sensing data of a mangrove to be diagnosed; for each mangrove state of the last classification unit, the mangrove state information is constructed by integrating its own soft prompt, the soft prompt of the upper classification unit, and the state information; the state features are obtained by performing second feature extraction on the mangrove state information; the matching coefficients of the remote sensing visual features and the state features are calculated; and the diagnosis conclusion is determined according to the mangrove state corresponding to the state feature with the largest matching coefficient. The present application realizes the precise and hierarchical diagnosis of the mangrove state by fusing ecological prior knowledge through multi-level soft prompts, combining a double-feature extraction and matching mechanism, and provides technical support for coastal ecological protection and restoration decision-making. BRIEF DESCRIPTION OF DRAWINGS

[0039] In order to more clearly illustrate the technical solutions of the embodiments of the present application, the following will briefly introduce the drawings needed to be used in the embodiments. It should be understood that the following drawings only show some of the embodiments of the present application, and therefore should not be considered as a limitation to the scope. For those skilled in the art, other related drawings can also be obtained without creative labor.

[0040] Figure 1 The step flowchart of the AI-assisted mangrove state diagnosis method provided by the embodiments of the present application is shown in the figure.

[0041] Figure 2 The structural schematic block diagram of the computer device provided by the embodiments of the present application is shown in the figure. DETAILED DESCRIPTION

[0042] In order to make the purpose, technical solutions and advantages of the embodiments of the present application more clear, the following will combine the drawings in the embodiments of the present application to clearly and completely describe the technical solutions in the embodiments of the present application. Obviously, the described embodiments are some of the embodiments of the present application, not all. The components of the embodiments of the present application described and shown in the drawings can be arranged and designed in various different configurations.

[0043] The specific embodiments of the present application will be described in detail below in combination with the drawings.

[0044] In order to solve the technical problems in the foregoing background art, Figure 1 The flowchart of the AI-assisted mangrove state diagnosis method provided by the embodiments of the present application is shown in the figure, and the following will introduce the AI-assisted mangrove state diagnosis method in detail.

[0045] Step S201, performing a first feature extraction operation on the remote sensing data of the mangrove to be diagnosed to obtain remote sensing visual features;

[0046] Step S202, for each mangrove state in the plurality of mangrove states in the terminal classification unit, constructing a mangrove state information from the soft hint of the mangrove state, the soft hint of the mangrove state of the upper classification unit of the mangrove state, and the mangrove state;

[0047] Step S203, performing a second feature extraction operation on the mangrove state information to obtain state features;

[0048] Step S204, determining the matching coefficients of the remote sensing visual features and the plurality of state features;

[0049] Step S205, determining the mangrove state corresponding to the remote sensing data of the mangrove to be diagnosed based on the mangrove state corresponding to the state feature with the largest matching coefficient in the plurality of state features.

[0050] In the embodiments of the present application, for example, in the implementation process of the AI-assisted mangrove state diagnosis method of the present application, the server as the execution subject sequentially completes each step to realize accurate diagnosis of the state of the mangrove. First, the server receives remote sensing data of the mangrove to be diagnosed. Such data is usually collected and generated by satellite remote sensing platforms, unmanned aerial vehicle remote sensing systems, etc., covering multispectral images, high-resolution spatial images, etc., including spectral reflectance characteristics, spatial distribution patterns, texture structure characteristics, and intertidal zone water environment information of mangrove vegetation, etc. Taking the ecological monitoring scene of a mangrove protection area in the southeast coastal area of China as an example, when the protection area needs to diagnose the state of the mangrove in autumn, after the server receives the multi-temporal remote sensing images of the area transmitted by the remote sensing satellite, the first feature extraction model (such as a deep convolutional neural network architecture) is called to carry out the first feature extraction operation. In this process, the multi-layer convolutional layers of the first feature extraction model will capture the spectral characteristics (such as the difference in near-infrared reflectance of healthy mangroves showing high reflectance and degenerative mangroves due to the decrease in chlorophyll content), spatial texture characteristics (such as the cluster distribution pattern of mangrove communities, the branch shape and density characteristics of the tidal creek system) of the mangrove vegetation in the image layer by layer; then, the pooling layer reduces the dimension and abstracts the extracted features, and the fully connected layer further integrates the feature information, finally generating a remote sensing visual feature vector containing key ecological information such as vegetation coverage, community structure complexity, and water and mangrove boundary contour characteristics. The vector accurately describes the visual form and ecological state of the mangrove to be diagnosed, providing basic data support for the matching of state characteristics in the subsequent steps.

[0051] Then, the server constructs the mangrove state information for each of the plurality of mangrove states in the last-level classification unit. The mangrove states are divided into a multi-level classification unit structure according to the principles of ecological taxonomy, for example, the top-level classification unit is "health state" (including healthy, sub-healthy, and degraded three sub-states), the middle-level classification unit is "stress type" (for sub-healthy and degraded states, subdivided into sub-categories such as alien species invasion, sea level rise influence, and human activity disturbance), and the last-level classification unit is "specific stress performance" (such as "alien species invasion-mutual grass invasion causing mangrove seedling loss" and "sea level rise influence-high tide frequent flooding causing mangrove growth deformity"). Each mangrove state corresponds to a predefined soft prompt, and the soft prompt is a semantic coding vector that integrates ecological expert knowledge and historical monitoring data rules. For example, the soft prompt vector of the "healthy" state includes numerical representations of ecological parameters such as high vegetation coverage threshold, community structure integrity index, and biodiversity richness characteristics. The soft prompt vector of the "alien species invasion" state encodes spectral recognition features of invasive species and typical texture change patterns of mangrove communities after disturbance. The server integrates the soft prompt of each mangrove state in the last-level classification unit, the soft prompts of its upper-level classification units (such as the "alien species invasion" corresponding to the middle-level classification unit and the "sub-healthy" corresponding to the top-level classification unit), and the identification information of the state to construct the mangrove state information. Taking the last-level state "alien species invasion-mutual grass moderate invasion causing community structure singleness" as an example, the server first obtains the soft prompt of the state itself (including the texture fragmentation feature and species composition proportion feature corresponding to "community structure singleness"), then obtains the soft prompts of the middle-level classification unit "alien species invasion" (including the typical spectral features of mutual grass and the growth restriction pattern of mangrove after invasion) and the top-level classification unit "sub-healthy" (including the vegetation coverage decay interval and biomass decline feature coding), and finally integrates these three types of information to form mangrove state information covering multiple ecological semantics, ensuring that the subsequent feature extraction process can fully utilize hierarchical ecological knowledge resources.

[0052] Subsequently, the server performs a second feature extraction operation on the constructed mangrove state information to obtain state features. The second feature extraction model adopts a Transformer-based encoder architecture with a multi-head self-attention mechanism for processing multi-source hierarchical information. For the constructed mangrove state information, the model captures the correlation between different classification unit soft prompts (such as the causal correlation between "sub-health" and "invasion of alien species," where the invasion of alien species may lead to a decrease in the health level of the mangrove to sub-health; the performance correlation between "invasion of alien species" and "single community structure," where the invasion of Spartina alterniflora may occupy the growth space of mangroves, leading to a single community species composition) through the multi-head self-attention layer; and then encodes and abstracts the captured correlation information through a feedforward neural network to finally generate a state feature vector that accurately represents the core ecological features of the mangrove state. Taking the processing of state information of "Spartina alterniflora moderate invasion leading to single community structure" as an example, the second feature extraction model focuses on the destruction mode of "invasion of alien species" to "community structure" (such as the texture feature correlation between the invasion of Spartina alterniflora in the intertidal zone leading to sparse distribution of mangrove seedlings) and the overall restriction degree of mangrove growth in the "sub-health" state (such as the spatial correspondence between the vegetation coverage decrease interval and the spectral reflection abnormal area) during operation, and converts these ecological logics into high-dimensional numerical features to form a state feature vector unique to the state, providing comparable feature dimensions for the subsequent matching process with remote sensing visual features.

[0053] Subsequently, the server needs to determine the matching coefficients of the remote sensing visual features and the plurality of state features. The matching calculation section adopts a cosine similarity algorithm. The server inputs the remote sensing visual feature vector obtained through the first feature extraction and the state feature vector of each final state into the matching calculation module, and quantifies the matching degree between the two by calculating the cosine similarity between the vectors. For the remote sensing visual feature vector V of the mangrove to be diagnosed, and the state feature vector of the i-th state of the final classification unit , the matching coefficient is calculated according to the formula (where "·" represents the inner product operation, and are the lengths of the vectors , ). After traversing all the mangrove states of the final classification unit, the server obtains a set of matching coefficients , each coefficient corresponds to the degree of fit between the remote sensing data features and the state features. For example, if the mangrove to be diagnosed is invaded by Spartina alterniflora, resulting in a single community structure, its remote sensing visual features such as “community texture fragmentation” (the distribution of mangrove clusters is broken, showing scattered distribution characteristics), “local abnormal near-infrared reflectance” (there is a difference in near-infrared reflectance between Spartina alterniflora and mangrove, resulting in abnormal spectral features in the local area of the image) and other features are highly matched with the state feature vector of “invasion of alien species - moderate invasion of Spartina alterniflora leading to single community structure” in the spectral dimension (the spectral code of Spartina alterniflora matches the abnormal spectral area in the image) and the texture dimension (the texture code of community fragmentation matches the distribution characteristics of mangrove in the image). The matching coefficient of the corresponding state feature will be significantly higher than that of other states.

[0054] Finally, the server determines the state of the mangrove to be diagnosed based on the state feature with the largest matching coefficient among the multiple state features. In practical applications of ecological monitoring, if the decision-making needs of ecological management from macro health level to micro stress details need to be met (such as when formulating an ecological restoration plan, it is necessary to clearly understand the overall health level of the mangrove, the main stress type and specific performance, so as to deploy targeted restoration measures), the server will extract the final state corresponding to the state feature with the largest matching coefficient and the upper classification unit state. Still taking the above-mentioned state of “invasion of alien species - moderate invasion of Spartina alterniflora leading to single community structure” as an example, the upper classification unit state is “invasion of alien species” of the intermediate classification unit and “sub-health” of the top classification unit. The server will jointly use the three levels of states (sub-health, invasion of alien species, and moderate invasion of Spartina alterniflora leading to single community structure) as the diagnosis result to generate a multi-level state report, providing comprehensive information from macro to micro for managers. If the application scenario is research in the field of science focusing on the specific stress mechanism (such as focusing on the influence mechanism of Spartina alterniflora invasion on the community structure of mangrove), only the accurate final state information is needed, then the server directly outputs the final state “invasion of alien species - moderate invasion of Spartina alterniflora leading to single community structure”. During the operation of the server, the output form is automatically selected according to the pre-configured diagnosis strategy (set by the specific needs of ecological monitoring tasks, such as the differentiated needs of different scenarios such as ecological management and scientific research), thereby completing the whole process of remote sensing data collection and intelligent diagnosis of the state of mangrove, and providing accurate technical support for the ecological protection, restoration and scientific management of mangrove.

[0055] In the embodiment of the present application, the state of the mangrove to be diagnosed is determined based on the state feature with the largest matching coefficient among the multiple state features. The determination can be implemented through the following examples.

[0056] obtaining a mangrove state corresponding to the state feature with the largest matching coefficient and a mangrove state of an upper classification unit of the mangrove state;

[0057] The mangrove state corresponding to the state feature with the largest matching coefficient and the mangrove state of the upper classification unit of the mangrove state are jointly used as the mangrove state corresponding to the to-be-diagnosed mangrove remote sensing data.

[0058] In the embodiment of the present application, in an exemplary red mangrove ecological monitoring scene, when the server as the execution subject carries out operation, first, according to the previous matching coefficient calculation result, the state feature with the largest matching coefficient is accurately located to the final level mangrove state “invasion of alien species-specific herbaceous plant medium invasion causes community structure to be single”. The determination of this final state is derived from the visual features of the to-be-diagnosed mangrove remote sensing data (such as the texture feature of “red tree cluster distribution fragmentation” presented by the image, and the feature of “local abnormal reflection of specific spectral band”), which is highly consistent with the final state feature vector in the ecological semantic dimension (such as the texture pattern coding of the invasion plant leading to the sparse distribution of mangrove seedlings, and the coding of the spectral difference rule between the invasion plant and the mangrove), and the matching coefficient after quantization by cosine similarity is much higher than that of other final states.

[0059] Then, the server automatically calls the mangrove state hierarchical association database according to the hierarchical classification rule of the mangrove state (the top classification unit is the health state, the intermediate classification unit is the stress type, and the final classification unit is the specific stress performance), and traces the upper classification unit information of the final state: at the intermediate classification unit level, the state belongs to “invasion of alien species” (this state encodes the common ecological features of the invasion stress, such as the typical rule of the invasion species expanding and occupying the red mangrove habitat); at the top classification unit level, the upper state corresponding to “invasion of alien species” is “sub-health” (this state represents that the overall health level of the mangrove forest is in the interval that needs attention but has not reached degradation, covering the macro ecological index features such as vegetation coverage decline and biomass decline).

[0060] Finally, the server integrates the final state “invasion of alien species-specific herbaceous plant medium invasion causes community structure to be single”, the intermediate classification unit state “invasion of alien species”, and the top classification unit state “sub-health” as the mangrove state corresponding to the to-be-diagnosed mangrove remote sensing data. When the integration result is output as an ecological monitoring report, it can provide multi-dimensional decision support for the monitoring and management subject: the top “sub-health” clearly locates the macro positioning of the ecological health level to allocate the resource priority, the intermediate classification unit “invasion of alien species” anchors the stress control direction, and the final state provides micro basis for the design of fine repair measures such as “removing the invasion plant and replanting the mangrove seedling”, realizing the whole chain supply of diagnosis information from ecological level judgment to stress detail analysis.

[0061] In the embodiments of the present application, the state of the mangrove corresponding to the state feature with the largest matching coefficient is determined as the state of the mangrove corresponding to the remote sensing data to be diagnosed, which can be implemented through the following examples.

[0062] The state of the mangrove corresponding to the state feature with the largest matching coefficient is determined as the state of the mangrove corresponding to the remote sensing data to be diagnosed.

[0063] In the embodiments of the present application, for example, when the server completes the state diagnosis as the execution subject, the final state of the mangrove corresponding to the state feature with the largest matching coefficient is accurately identified by relying on the matching coefficient calculation process of the remote sensing visual features and each final state feature in the early stage. For example, when the core visual feature of the remote sensing data to be diagnosed is "the distribution of the mangrove cluster presents a fragmented pattern (such as the originally continuous mangrove community is divided into scattered patches)" and "the reflectivity of the local area is abnormal under a specific spectral band (there is a significant difference from the typical spectral feature of the mangrove)", the server quantifies the matching degree of each final state feature vector and the remote sensing visual feature vector by using the cosine similarity algorithm, and finds that the matching coefficient of the final state "invasion of alien species - moderate invasion of specific herbaceous plants leading to single community structure" is significantly higher than that of other states. The reason is that the state feature vector encodes the "texture pattern rule of the invasion of herbaceous plants occupying the habitat of the mangrove leading to sparse distribution of seedlings" and "the difference in reflectivity between the invasive species and the mangrove in the target spectral band", which is highly consistent with the spatial correlation characteristics of the fragmented distribution of the mangrove and the abnormal spectral area in the data to be diagnosed.

[0064] Subsequently, the server directly determines the final state of the mangrove corresponding to the state feature with the largest matching coefficient, i.e. "invasion of alien species - moderate invasion of specific herbaceous plants leading to single community structure", as the state of the mangrove corresponding to the remote sensing data to be diagnosed, according to the demand of the scientific research scene for "accurately obtaining specific stress performance details". After outputting this diagnosis result, the scientific research team can carry out targeted research based on this, such as selecting the mangrove area corresponding to this state and the control area not affected by the invasion of the herbaceous plants, investigating the density of the mangrove seedlings and the proportion of species composition through field quadrat survey, combining with the analysis of the expansion rate of the invasive plants through remote sensing image time series, and then verifying the "driving mechanism of the invasion of specific herbaceous plants to the fragmentation of the community structure of the mangrove", realizing the accurate connection from the intelligent diagnosis of remote sensing data to the empirical analysis of scientific research problems, and meeting the demand of the scientific research scene for accurate positioning of micro stress states.

[0065] In the embodiment of the present application, the mangrove state diagnosis method is realized based on a pre-trained mangrove state recognition model. The mangrove state has multiple classification units. For any two adjacent classification units in the multiple classification units, a first classification unit and a second classification unit, the first classification unit has N mangrove states, the second classification unit has M mangrove states, one or more mangrove states in the M mangrove states have a superior mangrove state existing in the N mangrove states, and the N is not more than the M.

[0066] Before constructing the mangrove state information of each mangrove state in the multiple mangrove states in the final classification unit from the soft hint of the mangrove state, the soft hint of the mangrove state of the upper classification unit of the mangrove state, and the mangrove state, the method further comprises:

[0067] Performing a first feature extraction operation on the sample mangrove remote sensing data to obtain sample remote sensing visual features;

[0068] For each classification unit in the multiple classification units, based on the sample state features and the sample remote sensing visual features corresponding to the multiple mangrove states of the classification unit, the state confidence sequences corresponding to the multiple mangrove states of the classification unit are calculated.

[0069] Based on the state confidence sequences corresponding to the multiple mangrove states of the classification unit, a first error parameter corresponding to the classification unit is calculated.

[0070] For the second classification unit in the multiple classification units, based on the state confidence sequences corresponding to the multiple mangrove states of the second classification unit, and the state confidence sequences corresponding to the superior mangrove states of each mangrove state of the second classification unit, a second error parameter of the first classification unit is calculated.

[0071] Based on the first error parameter and the second error parameter, a comprehensive error parameter is determined.

[0072] Based on the comprehensive error parameter, the parameters of the soft hints corresponding to the multiple mangrove states of each classification unit in the multiple classification units are adjusted until the training termination condition is met, and a trained mangrove state recognition model is obtained.

[0073] In the embodiment of the present application, in the training process of the mangrove state recognition model, the server as the execution subject first performs a first feature extraction operation on the sample mangrove remote sensing data (covering multispectral images and spatial texture data of mangroves in different ecological gradients and seasonal cycles). This operation calls a convolutional neural network with the same architecture as the inference stage to encode the spectral response of the mangrove vegetation in the sample image (such as the strong reflection of healthy mangroves in the near-infrared band and the spectral redshift characteristics of degraded mangroves), spatial pattern (such as the texture complexity of cluster distribution and the branch density of tidal creek systems), and generates a sample remote sensing visual feature vector containing ecological information such as vegetation coverage and community structure integrity.

[0074] Based on the multi-layer classification unit design of the mangrove state (such as the top classification unit containing "state A", "state B", and "state C" for a total of N=3 states, and the second classification unit containing "sub-state α", "sub-state β", "sub-state γ", "sub-state δ", and "sub-state ε" for a total of M=5 states, and the upper states of "sub-state α" and "sub-state β" are "state B", satisfying the N not exceeding M hierarchical association rule), the server processes each classification unit in turn. For the top classification unit, the server first constructs a sample state feature for each state: integrates the initial soft prompt of the state (a vector encoding the health threshold defined by ecological experts and prior knowledge such as community structure indicators) and the state identifier into sample mangrove state information, and then performs a second feature extraction through a Transformer encoder to obtain a sample state feature vector representing the core ecological characteristics of the state. Subsequently, based on the sample state features and sample remote sensing visual features of the states of the top classification unit, the server calculates the confidence of each state through cosine similarity combined with a softmax function (such as when the sample image corresponds to the true label "state B", the confidence of "state B" should be significantly higher than that of other states), forming the state confidence sequence of this classification unit; and then quantifies the difference between the predicted confidence sequence and the true label through a cross-entropy loss function to obtain the first error parameter corresponding to the top classification unit.

[0075] For the second taxon (e.g., "substate α" and "substate β" of the lower taxon), the server first obtains the sample state features for each state (constructed by integrating its own soft hints, the soft hints of the upper taxon "state B," and the state identifier, then extracting the state features through the second feature extraction). The server then calculates the state confidence sequence for the taxon. Simultaneously, it retrieves the state confidence sequence for its superordinate state ("state B") in the top taxon. Using a designed hierarchical constraint loss function (e.g., KL divergence measures the alignment of the state confidence distribution of the lower taxon with that of the superordinate state, ensuring that the confidence of "state B" is appropriately correlated as the confidence of "substate α" increases), the server calculates the second error parameter corresponding to the first taxon (the top taxon). This process ensures consistency of ecological logic across the hierarchy (e.g., when the superordinate taxon is determined to be in "state B," the confidence of the lower taxon's "substate α" must conform to the typical stress patterns observed under "state B").

[0076] The server then weights and fuses the first and second error parameters according to their importance at the ecological level to form a comprehensive error parameter. Based on this, the server uses a gradient descent algorithm with backpropagation to adjust the soft-hint parameters for each mangrove state in each taxonomic unit. If the first error parameter for the top-level "state B" is too high, indicating that the ecological characteristics encoded in its soft-hint do not match the sample's remote sensing visual features, the server then fine-tunes the values ​​of dimensions such as "community structural integrity" and "biomass threshold" in the soft-hint vector to ensure that subsequent feature extraction is more consistent with the true ecological performance of "state B." If the second error parameter for the second-level taxonomic unit "sub-state α" is too high, indicating that its hierarchical association logic with the upper-level "state B" is not met, the server then adjusts the encoding of "invasive species spectral characteristics" and "mangrove disturbance texture pattern" in the soft-hint of "sub-state α" to strengthen its semantic association with the soft-hint of "state B."

[0077] The server continues to iterate the training process until the comprehensive error parameter converges to a preset threshold (e.g., the loss function value decreases by less than 10⁻⁵ for 50 consecutive epochs) or the maximum number of iterations is reached. At this point, the soft-hint parameters for each taxon have fully learned the ecological characteristics and hierarchical associations of the sample data, ultimately resulting in a trained mangrove state recognition model. During the inference phase, this model accurately invokes soft-hints for each taxon, combining remote sensing visual features with state characteristics to achieve intelligent diagnosis of multi-dimensional mangrove state.

[0078] In an embodiment of the present invention, for each of the multiple classification units, the state confidence sequence corresponding to the multiple mangrove states of the classification unit is calculated based on the sample state characteristics and the sample remote sensing visual characteristics corresponding to the multiple mangrove states of the classification unit, which can be implemented through the following examples.

[0079] determine, based on the classification unit hierarchical order of the plurality of classification units, a plurality of sample state features corresponding to a plurality of mangrove state statuses of the classification unit respectively;

[0080] calculate, based on the plurality of sample state features corresponding to the plurality of mangrove state statuses of the classification unit respectively, the ecological sensitivity adjustment factor, and the sample remote sensing visual feature, a state confidence sequence corresponding to the plurality of mangrove state statuses of the classification unit respectively.

[0081] In the embodiment of the present application, for example, in the training phase of the mangrove state recognition model, when the server performs state confidence sequence calculation for each classification unit (such as the top-level classification unit "healthy / sub-healthy / degraded", the intermediate classification unit "stress type (invasion / pollution / sea level impact)", and the final classification unit "specific stress performance"), first, the sample state features are determined according to the classification unit hierarchical order:

[0082] For the top-level classification unit (the topmost level without an upper-level classification unit), the server constructs a sample state feature for each mangrove state (such as "healthy", "sub-healthy", and "degraded"): integrates the initial soft prompt (encodes the prior knowledge of "healthy mangrove near-infrared high reflection threshold" and "degraded mangrove biomass attenuation interval" defined by ecological experts) of the state and the state identifier into sample mangrove state information, and then performs second feature extraction through the Transformer encoder to generate a sample state feature vector (such as the feature vector of the "healthy" state, which needs to include the spectral coding of high vegetation coverage and the texture coding of complete community structure) representing the core ecological features of the state.

[0083] For a non-top-level classification unit (such as the intermediate classification unit "stress type"), when the server constructs a sample state feature for each mangrove state (such as "invasion" and "pollution"), it needs to integrate three parts of information: the soft prompt of the state itself (encoding "typical spectral features of invasive species" and "abnormal patterns of mangrove chlorophyll fluorescence caused by pollution"), the soft prompt of its upper-level classification unit (such as the top-level "sub-healthy") (encoding "sub-healthy mangrove vegetation coverage attenuation interval" and "biodiversity decline features"), and the state identifier, and after the second feature extraction, a sample state feature vector (such as the feature vector of the "invasion" state, which needs to associate the macroscopic ecological decline law of "sub-healthy" with the microscopic spectral interference features of invasive species) is obtained, which integrates the hierarchical ecological semantics.

[0084] After obtaining the sample state feature of each classification unit, the server combines the sample remote sensing visual feature (the ecological feature vector such as spectrum, texture, spatial pattern, etc. extracted from the sample mangrove remote sensing image by the first feature extraction operation) and the ecological sensitivity adjustment factor (the weight parameter preset according to the ecological vulnerability of the mangrove region, such as the adjustment factor of the area of the intertidal zone edge vulnerable to sea level is higher), and calculates the state confidence sequence: taking the “invasion” state of the intermediate classification unit as an example, the server first calculates the matching degree of the “invasion” sample state feature vector and the sample remote sensing visual feature vector by the cosine similarity algorithm (if there is a red tree cluster fragmentation texture caused by invasive species in the sample image, the near-infrared band abnormal reflection area, and the matching degree is significantly improved); then multiply the matching degree and the ecological sensitivity adjustment factor (such as the historical invasion event frequently occurs in this area, and the adjustment factor is set to 1.2) to strengthen the attention to the ecological sensitive features; finally, the weighted matching degree of all states is converted into a probability distribution by the softmax function to form the state confidence sequence of the classification unit (such as the “invasion” state confidence is 0.75, the “pollution” state is 0.15, and the “sea level influence” state is 0.10, reflecting the matching priority of the sample remote sensing feature and each state).

[0085] Through this process, the server not only guarantees the ecological semantic integrity of the sample state feature of the classification unit at different levels (focusing on the macro health level at the top and refining the stress type at the bottom), but also adapts to the actual vulnerability difference of the mangrove region through the ecological sensitivity adjustment factor, so that the state confidence sequence accurately reflects the state matching logic of the sample remote sensing data in the ecological level and spatial heterogeneity dimension, providing reliable probability distribution basis for subsequent error parameter calculation and model training.

[0086] In the embodiment of the application, the determination of the sample state feature corresponding to each of the plurality of mangrove states of the classification unit based on the classification unit order of the classification unit in the plurality of classification units comprises:

[0087] If the classification unit is the topmost classification unit in the plurality of classification units, for each of the plurality of mangrove states of the classification unit, the soft prompt of the mangrove state and the mangrove state are constructed to obtain sample mangrove state information;

[0088] The second feature extraction operation is performed on the sample mangrove state information to obtain the sample state feature.

[0089] In the training process of the mangrove state recognition model, when the server processes the topmost classification unit (for example, “ecological health level”, which includes “healthy”, “sub-healthy” and “degradation”, and has no upper classification unit), the sample state feature of each state is generated according to the following logic:

[0090] Firstly, for each mangrove state of the top-level classification unit, the server integrates the "soft hints" and "state identifiers" to build sample mangrove state information. Taking the "healthy" state as an example, its soft hints are vectors of pre-encoded ecological priori knowledge, covering ecological rules such as typical spectral features of healthy mangroves (such as the threshold encoding of near-infrared reflectance higher than 0.7), spatial pattern features (such as the texture complexity index of community continuous distribution, the encoding of gray level co-occurrence matrix contrast lower than 0.3), and biomass range (such as the numerical encoding of unit area biomass higher than 500 kg / ha); the server will structure the integration of this soft hint and the identifier of the "healthy" state (such as the numerical encoding "0") to form sample mangrove state information carrying the core ecological definition of the "healthy" state (this information is stored in the form of key-value pairs or vector splicing to ensure the integrity of ecological semantics).

[0091] Subsequently, the server calls the second feature extraction module (based on the Transformer encoder architecture) to perform feature extraction on the sample mangrove state information. Taking the sample mangrove state information of the "healthy" state as an example, the multi-head self-attention layer of Transformer will capture the associated logic of multi-dimensional ecological features within the soft hints (such as the symbiotic relationship between "high near-infrared reflectance" and "continuous distribution of community", healthy mangroves have high near-infrared reflectance due to sufficient chlorophyll content, and the community is continuously distributed when not disturbed, and there is strong synergy between them in spatial and spectral dimensions); the feedforward neural network layer encodes and abstracts these associated information, and finally generates a high-dimensional sample state feature vector (this vector not only contains the numerical features of individual ecological indicators, but also integrates the ecological coupling rules between "spectrum-texture-biomass", such as when the near-infrared reflectance is >0.7 and the texture complexity is <0.3, the activation value of the corresponding dimension in the vector will significantly increase, representing the typical feature combination of the "healthy" state).

[0092] Similarly, the server repeats the above process for the "sub-healthy" and "degraded" states of the top-level classification unit: the soft hints of the "sub-healthy" state encode ecological features such as "10%-20% decrease in vegetation coverage compared to healthy" and "reduced biodiversity index but not reaching the critical value", and after integration and second feature extraction, the corresponding sample state features are generated; the soft hints of the "degraded" state encode features such as "vegetation coverage decreased by more than 30%" and "serious community fragmentation (texture complexity >0.5)", and also generate sample state features.

[0093] Through the process, the server builds a complete link of "prior knowledge coding-structured information-feature abstraction" for each mangrove state of the topmost classification unit, ensuring that the sample state features not only carry the rules defined by ecological experts, but also capture the ecological association logic between features through the Transformer, providing accurate feature basis for subsequent calculation of state confidence sequence combined with sample remote sensing visual features, and driving model learning of recognition rules of the top-level ecological grade.

[0094] In the embodiments of the present application, the determination of the sample state features corresponding to the plurality of mangrove states of the classification unit based on the classification unit hierarchical order of the plurality of classification units also provides the following implementation.

[0095] If the classification unit is a non-topmost classification unit in the plurality of classification units, for each mangrove state of the plurality of mangrove states of the classification unit, the soft hint of the mangrove state, the soft hint of the mangrove state of the upper classification unit of the mangrove state, and the sample mangrove state information constructed by the mangrove state are obtained.

[0096] The second feature extraction operation is performed on the sample mangrove state information to obtain sample state features.

[0097] In the embodiments of the present application, for example, in the training process of the mangrove state recognition model, when the server processes a non-topmost classification unit (for example, the "stress type" intermediate classification unit, whose upper layer is the top-level "sub-health" classification unit, and itself contains "invasion", "pollution", "sea level influence" and other mangrove states), the sample state features of each state are generated according to the following logic:

[0098] Firstly, for each mangrove state of the classification unit (take the "invasion" state as an example), the server integrates three types of information to construct sample mangrove state information: first, the soft prompt of the "invasion" state itself (encoding the typical spectral features of invasive species, such as the numerical coding of 0.2 of the reflectivity of specific herbs in the short-wave infrared band compared to mangroves; and the texture features of mangrove communities after invasion, such as the entropy value of the gray level co-occurrence matrix of 0.8 of the fragmentation of cluster distribution); second, the soft prompt of the upper classification unit (the top "sub-health") of the mangrove state (encoding the macro-ecological characteristics of "sub-health", such as the interval coding of 15%-20% of the decrease of vegetation coverage compared to healthy, and the numerical coding of 10%-15% of the decrease of biodiversity index); third, the identification of the "invasion" state (such as the numerical coding "1"). The server integrates these three parts of information into sample mangrove state information that carries both the "invasion" itself stress characteristics and the "sub-health" macro-ecological background (for example, the association field of "invasion spectral features" and "sub-health coverage decay" in the information will strengthen the ecological logic that "invasion is one of the causes of sub-health").

[0099] Next, the server calls the second feature extraction module (based on the Transformer encoder) to perform feature extraction on the sample mangrove state information. Taking the sample information of the "invasion" state as an example, the multi-head self-attention layer of the Transformer captures the ecological associations between multiple levels of soft prompts: on the one hand, it focuses on the spectral and texture feature associations within the "invasion" itself soft prompt (such as the higher the spatial overlap between the spectral anomaly area of the invasive species and the fragmented texture of the mangrove, the stronger the activation value of the corresponding feature dimension); on the other hand, it associates the macro-features of the upper "sub-health" soft prompt (such as when the "sub-health" coverage decay interval matches the red tree death area caused by "invasion", the self-attention mechanism will strengthen the causal association coding of the two). The feedforward neural network layer abstracts and compresses these cross-level and cross-dimension association information, and finally generates a sample state feature vector that integrates "invasion stress details" and "sub-health macro-background" (in this vector, if the sample information simultaneously satisfies "invasive species spectral features exist" and "sub-health coverage decay interval matches", the feature value of the corresponding dimension will significantly increase, accurately representing the feature coupling mode of the ecological process "invasion drives sub-health").

[0100] Similarly, the server repeats the above process for the "pollution" and "sea level impact" states of non-top-level classification units: the soft prompt of the "pollution" state encodes the spectral blue shift features of mangrove leaves caused by pollutants and the texture anomaly patterns of water eutrophication, and generates sample state features after combining with the upper "sub-health" soft prompt; the soft prompt of the "sea level impact" state encodes the growth deformity texture of mangroves caused by high tide inundation and the spectral features of elevation anomaly areas, and also completes feature extraction after integrating the upper soft prompt.

[0101] Through the process, the server constructs a sample state feature of "self-stress feature + upper ecological background + cross-level correlation logic" for each mangrove state of a non-topmost classification unit, which not only guarantees accurate coding of ecological stress details, but also strengthens the inter-level ecological cause-and-effect correlation through upper classification unit soft prompts (e.g., lower stress is the inducement of upper health level change), thereby providing fusion feature support for subsequent calculation of state confidence sequences combined with sample remote sensing visual features, and driving model learning of the correlation recognition rules of multi-level ecological states.

[0102] In the embodiment of the present application, the second error parameter of the first classification unit is calculated based on the state confidence sequence corresponding to each mangrove state of the second classification unit and the state confidence sequence corresponding to the upper mangrove state of each mangrove state of the second classification unit. The following example can be used to implement the embodiment.

[0103] A reference classification unit of the first classification unit is established, the number of mangrove state categories of the reference classification unit is the same as the number of mangrove state categories of the second classification unit, and each mangrove state of the reference classification unit is matched with each mangrove state of the second classification unit.

[0104] For each mangrove state of the second classification unit, the state confidence sequence of the corresponding mangrove state of the reference classification unit is determined based on the state confidence sequence of the upper mangrove state of the mangrove state.

[0105] The second error parameter of the first classification unit is calculated based on the state confidence sequence of each mangrove state of the second classification unit and the state confidence sequence of the corresponding mangrove state of the reference classification unit.

[0106] In the embodiment of the present application, for example, in the training process of the mangrove state recognition model, when the server trains the hierarchical correlation between the first classification unit (e.g., the top-level "health level" including states A, B, and C) and the second classification unit (e.g., the intermediate "stress type" including states α, β, γ, and δ, where the upper level of α and β is B, the upper level of γ is C, and the upper level of δ is A), the following operations are sequentially performed:

[0107] Firstly, the server establishes a reference classification unit of the first classification unit. Since the second classification unit contains 4 mangrove forest states (M=4), the reference classification unit needs to be consistent with the number of states, so a reference classification unit containing 4 states (denoted as (A'), (B'), (C'), (D')) is constructed, and the one-to-one matching of "reference state-second classification unit state-superior state" is realized through ecological correlation rules (such as (A') corresponds to δ (superior state is A), (B') corresponds to α (superior state is B), (C') corresponds to β (superior state is B), (D') corresponds to γ (superior state is C)), which ensures that each state of the reference classification unit can accept the association logic of the second classification unit state and the superior state of the first classification unit.

[0108] Next, for each mangrove forest state of the second classification unit (take α as an example), the server retrieves the state confidence sequence of its superior mangrove forest state (B, belonging to the first classification unit) (assuming that during the training process, the state confidence sequence of the top layer B is ([A:0.2, B:0.7, C:0.1]), indicating the matching probability distribution of the sample remote sensing visual features and A, B, C). According to the rule that "the superior state confidence needs to pass on the ecological correlation to the corresponding state of the reference classification unit", the core confidence of B (0.7) is assigned to (B') of the reference classification unit, and combined with α as the ecological logic of B typical stressor state, the state confidence sequence of (B') in the reference classification unit is generated (such as ([A':0.05, B':0.85, C':0.05, D':0.05]), simulating the hierarchical rule that "the high confidence of α needs to be strongly associated with the high confidence of B").

[0109] Finally, the server calculates the second error parameter of the first classification unit: retrieves the state confidence sequence of the second classification unit α (assuming it is ([α:0.8, β:0.1, γ:0.05, δ:0.05]), indicating the matching probability of the sample features and α, β, γ, δ), and compares it with the state confidence sequence of the reference classification unit (B') ([A':0.05, B':0.85, C':0.05, D':0.05]). The inconsistency between the two is quantified by KL divergence (measuring the difference between two probability distributions): .

[0110] If the high confidence of a (0.8) matches the high confidence of (B') (0.85) with high degree, the KL divergence value is small, indicating that the hierarchical association conforms to the ecological logic; if the confidence of a is high but the confidence of (B') is low (such as the confidence of (B') is 0.1), the divergence value increases sharply, exposing the logical contradiction that "a is the sub-state of B but is out of touch with the confidence of B". The server takes the KL divergence value as the second error parameter of the first classification unit (top layer) to constrain the ecological association consistency of "stress sub-state (second classification unit)" and "health level (first classification unit)", avoid the logical rupture of the hierarchy, and provide a loss signal for the model to learn the association rule of the multi-level state.

[0111] For the second classification unit of β (upper B), γ (upper C) and δ (upper A), the server repeats the above process, calculates the second error parameter of the corresponding reference classification unit, and integrates to finally form the hierarchical constraint loss of the first classification unit, which drives the model to accurately capture the multi-level ecological association of the mangrove state.

[0112] The embodiment of the present application provides a computer device 100, which comprises a processor and a non-volatile memory storing computer instructions. When the computer instructions are executed by the processor, the computer device 100 executes the aforementioned AI-assisted mangrove state diagnosis method. As shown in the figure, Figure 2 Figure 2 The structural block diagram of the computer device 100 provided by the embodiment of the present application. The computer device 100 comprises a memory 111, a processor 112 and a communication unit 113. In order to realize the transmission or interaction of data, the memory 111, the processor 112 and the communication unit 113 are directly or indirectly electrically connected with each other. For example, the electrical connection between these elements can be realized by one or more communication buses or signal lines.

[0113] The foregoing description is made with reference to specific embodiments. However, the above illustrative discussion is not intended to be exhaustive or to limit the disclosure to the precise forms disclosed. Numerous modifications and variations are possible in light of the above teachings. These embodiments are chosen and described in order to best explain the principles of the disclosure and its practical application to thereby enable others skilled in the art to best utilize the disclosure and various embodiments with various modifications as are suited to the particular use contemplated.​

Claims

1. A mangrove status diagnosis method based on AI assistance, characterized in that: include: Perform the first feature extraction operation on the mangrove remote sensing data to be diagnosed to obtain remote sensing visual features; For each mangrove state in a plurality of mangrove states in a final classification unit, constructing mangrove state information by combining a soft prompt of the mangrove state, a soft prompt of a mangrove state of a higher-level classification unit of the mangrove state, and the mangrove state; performing a second feature extraction operation on the mangrove state information to obtain a state feature; determining a matching coefficient between the remote sensing visual feature and a plurality of the state features; Determining the mangrove state corresponding to the mangrove remote sensing data to be diagnosed based on the mangrove state corresponding to the state feature with the largest matching coefficient among the plurality of state features; The mangrove state diagnosis method is implemented based on a pre-trained mangrove state recognition model; The mangrove state exists in multiple taxa, and for any two neighboring taxa among the multiple taxa: a first taxa and a second taxa, the first taxa exists in N mangrove states, the second taxa exists in M ​​mangrove states, and one or more superordinate mangrove states among the M mangrove states exist in the N mangrove states, where N does not exceed M. Before constructing the mangrove state information by combining, for each of the multiple mangrove states in the final classification unit, the soft prompt of the mangrove state, the soft prompt of the mangrove state of the upper classification unit of the mangrove state, and the mangrove state, the method further includes: Performing a first feature extraction operation on the sample mangrove remote sensing data to obtain the sample remote sensing visual features; Determining sample state characteristics corresponding to multiple mangrove states of the taxonomic unit based on the taxonomic unit hierarchy order of the taxonomic unit in the multiple taxonomic units; Calculating state confidence sequences corresponding to the multiple mangrove states of the classification unit based on the sample state characteristics corresponding to the multiple mangrove states of the classification unit, the sample remote sensing visual characteristics, and the ecological sensitivity adjustment factor; Calculating a first error parameter corresponding to the classification unit based on state confidence sequences corresponding to a plurality of mangrove states of the classification unit; For a second classification unit among the plurality of classification units, calculating a second error parameter of the first classification unit based on state confidence sequences corresponding to a plurality of mangrove states of the second classification unit and a state confidence sequence corresponding to a superordinate mangrove state of each mangrove state of the second classification unit; Determining a comprehensive error parameter based on the first error parameter and the second error parameter; Based on the comprehensive error parameters, the parameters of the soft prompts corresponding to the multiple mangrove states of each classification unit in the multiple classification units are adjusted until the training termination conditions are met to obtain a trained mangrove state recognition model.

2. The method according to claim 1, characterized in that The determining, based on the mangrove state corresponding to the mangrove remote sensing data to be diagnosed, the mangrove state corresponding to the state feature having the largest matching coefficient among the plurality of state features, includes: Obtaining the mangrove state corresponding to the state feature with the largest matching coefficient, and the mangrove state of the upper-level classification unit of the mangrove state; The mangrove state corresponding to the state feature with the largest matching coefficient and the mangrove state of the upper-level classification unit of the mangrove state are taken together as the mangrove state corresponding to the mangrove remote sensing data to be diagnosed.

3. The method according to claim 1, characterized in that The determining, based on the mangrove state corresponding to the mangrove remote sensing data to be diagnosed, the mangrove state corresponding to the state feature having the largest matching coefficient among the plurality of state features, includes: The mangrove state corresponding to the state feature with the largest matching coefficient is determined as the mangrove state corresponding to the mangrove remote sensing data to be diagnosed.

4. The method according to claim 1, wherein The determining, based on the order of the classification unit in the classification unit hierarchy of the multiple classification units, sample state features corresponding to the multiple mangrove states of the classification unit, respectively, includes: If the classification unit is a top-level classification unit among the multiple classification units, for each mangrove state among the multiple mangrove states of the classification unit, constructing the soft prompt of the mangrove state and the mangrove state to obtain sample mangrove state information; A second feature extraction operation is performed on the sample mangrove state information to obtain a sample state feature.

5. The method according to claim 1, wherein The determining, based on the order of the classification unit in the classification unit hierarchy of the multiple classification units, sample state features corresponding to the multiple mangrove states of the classification unit, further includes: If the classification unit is a non-top classification unit among the multiple classification units, for each mangrove state among the multiple mangrove states of the classification unit, a soft prompt of the mangrove state, a soft prompt of the mangrove state of a higher-level classification unit of the mangrove state, and the mangrove state are constructed to obtain sample mangrove state information; A second feature extraction operation is performed on the sample mangrove state information to obtain a sample state feature.

6. The method according to claim 1, characterized in that The calculating, for a second classification unit among the plurality of classification units, a second error parameter of the first classification unit based on state confidence sequences corresponding to a plurality of mangrove states of the second classification unit and a state confidence sequence corresponding to a superordinate mangrove state of each mangrove state of the second classification unit, comprises: Establishing a reference taxon for the first taxon, wherein the number of mangrove state categories of the reference taxon is the same as the number of mangrove state categories of the second taxon, and each mangrove state of the reference taxon matches each mangrove state of the second taxon; For each mangrove state of the second taxonomic unit, determining a state confidence sequence of the mangrove state corresponding to the mangrove state in the reference taxonomic unit based on a state confidence sequence of a superordinate mangrove state of the mangrove state; Based on the state confidence sequence of each mangrove state of the second classification unit and the state confidence sequence of the corresponding mangrove state of the reference classification unit, a second error parameter of the first classification unit is calculated.

7. A server system, characterized in that: The method comprises a server, wherein the server is configured to execute the method according to any one of claims 1 to 6.

Images