A mangrove forest weak label quality evaluation and time sequence expansion method and system

Abstract

The present application belongs to the technical field of mangrove weak tag monitoring data optimization, and particularly relates to a mangrove weak tag quality evaluation and time sequence expansion method and system. The method comprises the following steps: analyzing mangrove spatial benchmark multi-source comprehensive data based on mangrove monitoring heterogeneous data; performing mangrove multi-dimensional feature fusion analysis on the mangrove spatial benchmark multi-source comprehensive data to generate mangrove multi-dimensional feature fusion data; analyzing mangrove global grid credible weak tag data based on the mangrove multi-dimensional feature fusion data; performing mangrove layer sample correction and optimization processing based on the mangrove global grid credible weak tag data and mangrove benchmark mapping mask data to generate mangrove layer sample correction and optimization data; and performing pseudo tag design for mangrove layer time sequence expansion on the mangrove layer sample correction and optimization data. The present application analyzes mangrove monitoring data to realize intelligent weak tag quality evaluation and time sequence expansion.

Description

Technical Field

[0001] This invention belongs to the field of mangrove weak label monitoring data optimization technology, and in particular relates to a method and system for mangrove weak label quality assessment and time-series expansion. Background Technology

[0002] Mangroves are distributed in coastal wetland areas, and their spatial distribution data is an important data source for coastal wetland monitoring. Remote sensing intelligent interpretation technology has become the mainstream method for large-scale mangrove time-series monitoring, which highly relies on high-quality sample labels to support monitoring accuracy. Affected by factors such as the cloudy and rainy climate of coastal areas, remote sensing seasonal differences, complex coastal topography, and fragmented mangrove patches and mixed ground features, mangrove monitoring technology is unable to obtain high-precision, strongly labeled samples with full coverage and throughout the year. The industry often uses weak labels to expand samples to meet time-series monitoring needs. However, existing mangrove weak label quality assessment methods have obvious defects. They mostly rely solely on remote sensing spectral features to construct labels without combining multi-dimensional constraints such as topography, coastal location, and patch topology. They also lack a credibility grading and verification mechanism, resulting in prominent problems such as mislabeling, omissions, and uneven credibility. At the same time, existing schemes lack targeted patch anomaly correction and time-series expansion capabilities, cannot optimize anomaly samples in the northern edge ecotone of mangroves, cannot support long-term continuous monitoring, and are prone to problems such as dynamic identification bias and data gaps in mangroves. Summary of the Invention

[0003] Based on this, the present invention provides a method and system for weak labeling quality assessment and time-series expansion of mangroves to solve at least one of the above-mentioned technical problems.

[0004] To achieve the above objectives, a method for weakly labeled mangrove quality assessment and time-series expansion includes the following steps: Step S1: Obtain heterogeneous mangrove monitoring data for the entire study area. The heterogeneous mangrove monitoring data includes mangrove baseline mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Based on the heterogeneous mangrove monitoring data, perform spatial comprehensive baseline processing of mangrove multi-source data to generate mangrove spatial baseline multi-source comprehensive data. Step S2: Perform multi-dimensional feature fusion analysis on the multi-source integrated spatial benchmark data of mangroves to generate multi-dimensional feature fusion data of mangroves; Step S3: Perform a comprehensive raster-based analysis of mangrove forests based on the multi-dimensional feature fusion data to generate comprehensive raster-based data of mangrove forests with reliable weak labels. Step S4: Based on the mangrove global raster reliable weak label data and mangrove baseline mapping mask data, perform mangrove layer sample correction and optimization processing to generate mangrove layer sample correction and optimization data; perform pseudo-label design processing for mangrove layer time-series extension on the mangrove layer sample correction and optimization data to generate mangrove layer pseudo-label time-series extension data.

[0005] Furthermore, step S1 includes the following steps: Step S11: Obtain heterogeneous monitoring data of mangroves across the entire study area; Step S12: Using multi-year continuous time-series remote sensing data as a unified geographic benchmark, perform coordinate projection unified transformation and resolution resampling registration on all raster data in the heterogeneous mangrove monitoring data to generate raster spatial benchmark aligned data. Step S13: Based on the unified projection and resolution of the raster spatial reference alignment data, perform rasterization conversion on the coastal zone vector auxiliary data within the heterogeneous mangrove monitoring data to generate integrated vector raster registration data; Step S14: Perform categorized preprocessing operations on the vector raster integrated registration data, including normalizing the seasonal features of time-series remote sensing data, extracting topographic data factors, and extracting coastal location distance raster data to generate categorized mangrove heterogeneous feature data. Then, perform categorized data standardization processing on the categorized mangrove heterogeneous feature data to generate categorized mangrove standard feature data. Step S15: Perform unified pixel spatial layer benchmark encapsulation processing on the standard feature data of classified mangroves to generate multi-source integrated spatial benchmark data of mangroves.

[0006] Furthermore, step S2 includes the following steps: Step S21: Perform mangrove hierarchical separation processing based on multi-source integrated mangrove spatial benchmark data to generate mangrove hierarchical separation data, wherein the mangrove hierarchical separation data includes remote sensing image hierarchical data, topographic and coastal hierarchical data, and mangrove mapping mask patch hierarchical data. Step S22: Perform pixel-scale spectral temporal feature analysis based on remote sensing image layered data to generate mangrove spectral temporal feature data; Step S23: Perform mangrove topographic and coastal constraint feature analysis based on topographic and coastal layered data to generate mangrove topographic and coastal constraint feature data; Step S24: Perform patch neighborhood topological connectivity feature analysis based on mangrove mapping mask patch layered data to generate mangrove patch neighborhood topological connectivity feature data; Step S25: Perform multi-dimensional feature matching and fusion processing on mangrove spectral temporal feature data, mangrove topographic and coastal constraint feature data, and mangrove patch neighborhood topological connectivity feature data to generate multi-dimensional feature fusion data of mangroves.

[0007] Furthermore, step S22 includes the following steps: Pixel-scale spectral feature analysis is performed on the layered data of remote sensing images to generate layered pixel-scale spectral feature data of remote sensing images. Then, temporal information interpretation and extraction processing is performed on the layered pixel-scale spectral feature data of remote sensing images to generate mangrove spectral temporal feature data.

[0008] Furthermore, step S23 includes the following steps: Based on the topographic and coastal layered data, topographic and coastal location analysis is performed to generate topographic and coastal location analysis data; topographic and coastal location feature analysis is performed on the topographic and coastal location analysis data to generate topographic and coastal location feature data; and mangrove topographic and coastal constraint feature analysis is performed on the topographic and coastal location feature data to generate mangrove topographic and coastal constraint feature data.

[0009] Furthermore, step S24 includes the following steps: Based on the layered data of mangrove mapping mask patches, spatial topological analysis of mangrove patches is performed to generate spatial topological data of mangrove patches. Then, the topological connectivity feature analysis of the patch neighborhoods of the spatial topological data of mangrove patches is performed to generate topological connectivity feature data of the patch neighborhoods of mangrove patches.

[0010] Furthermore, step S3 includes the following steps: Step S31: Perform mangrove multidimensional feature credibility mapping processing on the mangrove multidimensional feature fusion data to generate mangrove multidimensional feature credibility mapping data; Step S32: Perform neighborhood cross-pixel association derivative feature analysis on the mangrove multidimensional feature fusion data to generate neighborhood cross-pixel association derivative feature data. Then, optimize the multidimensional feature cross-pixel association credibility derivative feature mapping of the mangrove multidimensional feature credibility mapping data using the neighborhood cross-pixel association derivative feature data to generate optimized mangrove multidimensional feature credibility mapping data. Step S33: Perform multi-factor fitness assignment weight analysis on mangroves based on the multi-dimensional feature fusion data of mangroves, and generate multi-factor fitness assignment weight data of mangroves. Step S34: Perform pixel-by-pixel mangrove multidimensional fit labeling on the optimized mangrove multidimensional feature credibility mapping data using the mangrove multi-factor fit weighting data to generate mangrove multidimensional fit labeling data. Step S35: Perform a mangrove whole-domain raster-based reliability assessment on the mangrove multidimensional adaptability identification data through a preset reliability judgment threshold to generate mangrove whole-domain raster-based reliability assessment data. Step S36: Design a mangrove whole-area raster trustworthy weak label for the mangrove whole-area raster trustworthy assessment data, and generate mangrove whole-area raster trustworthy weak label data.

[0011] Furthermore, step S4 includes the following steps: Step S41: Perform validity segmentation of mangrove whole-domain sample layers using mangrove whole-domain raster reliable weak label data and mangrove benchmark mapping mask data to generate mangrove whole-domain sample layer validity classification data; Step S42: Extract anomalous local patches on the northern edge of the mangrove whole-area sample layer validity classification data to generate anomalous local patch data on the northern edge; Step S43: Extract the effective layer of the northern edge patch of the mangrove from the effective classification data of the mangrove whole area sample layer, generate the effective layer data of the northern edge patch of the mangrove, and perform joint correction feature analysis of the northern edge patch of the sample layer through the effective layer data of the northern edge patch of the mangrove, to generate the joint correction feature data of the northern edge patch of the sample layer. Step S44: Adaptive correction of anomalous local patches on the northern edge is performed on the northern edge patch data using the joint correction feature data of the northern edge patches in the sample layer, generating corrected local patch data on the northern edge. Step S45: By correcting local patch data on the northern edge, perform sample correction and optimization processing on the mangrove layer validity classification data of the whole mangrove sample layer to generate mangrove layer sample correction and optimization data; Step S46: Perform layer time series analysis processing based on the mangrove layer sample correction and optimization data to generate mangrove layer sample time series analysis data; Step S47: Perform mangrove state trend feature analysis on the time series parsing data of the mangrove layer samples to generate mangrove state trend feature data; Step S48: Map the mangrove state trend feature data to the mangrove layer sample correction and optimization data, perform pseudo-label design processing for the time-series extension of the mangrove layer, and generate mangrove layer pseudo-label time-series extension data.

[0012] Furthermore, step S41 includes the following steps: The raster-based weak label data and benchmark mapping mask data of mangrove forests are subjected to pixel-level overlay binding processing to generate mangrove binary identification overlay data. Based on the mangrove binary identification overlay data, a two-dimensional identification matrix for mangrove samples is designed, wherein the two-dimensional identification matrix for mangrove samples consists of the first identification dimension of the benchmark mask binary annotation result and the second identification dimension of the weak label confidence annotation result. The validity classification data of mangrove forest sample layers is generated by using the mangrove sample binary identification matrix to classify the validity of mangrove forest sample layers.

[0013] This specification provides a mangrove weak labeling quality assessment and time-series extension system for performing the mangrove weak labeling quality assessment and time-series extension method as described above. The mangrove weak labeling quality assessment and time-series extension system includes: The multi-source data monitoring and spatial integrated benchmark module is used to acquire heterogeneous mangrove monitoring data for the entire study area. The heterogeneous mangrove monitoring data includes mangrove benchmark mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Based on the heterogeneous mangrove monitoring data, spatial integrated benchmark processing of mangrove multi-source data is performed to generate mangrove spatial benchmark multi-source integrated data. The mangrove multidimensional feature fusion module is used to perform multidimensional feature fusion analysis on multi-source comprehensive data of mangrove spatial benchmarks and generate multidimensional feature fusion data of mangroves. The mangrove weak label quality assessment module is used to perform reliable weak label analysis of the entire mangrove raster based on multi-dimensional feature fusion data of mangroves, and generate reliable weak label data of the entire mangrove raster. The mangrove time-series extension module is used to perform mangrove layer sample correction and optimization processing based on mangrove global raster reliable weak label data and mangrove baseline mapping mask data, generating mangrove layer sample correction and optimization data; and to perform pseudo-label design processing on mangrove layer time-series extension of mangrove layer sample correction and optimization data, generating mangrove layer pseudo-label time-series extension data.

[0014] The beneficial effects of this application are as follows: This invention integrates heterogeneous data from multiple sources, including reference masks, topography, coastline vectors, and multi-year time-series remote sensing data. Using time-series remote sensing data as a unified geographic reference, it completes projection transformation, resolution resampling, and integrated vector raster registration, solving the technical problems of inconsistent coordinate systems, scale inconsistencies, and the inability to fuse and combine vector raster data from multiple sources. Simultaneously, through seasonal phase normalization, extraction of topographic and coastal location factors and data standardization, and layer reference encapsulation processing, it eliminates interference caused by differences in remote sensing seasonal phases and data dimensions, achieving standardized and integrated integration of multi-source heterogeneous data. This provides a unified, standardized, and high-precision spatial data foundation for subsequent multi-dimensional feature fusion and precise analysis of weak labels, ensuring the accuracy and consistency of subsequent data analysis. The preprocessed multi-source integrated data is hierarchically separated into three independent hierarchical data categories: remote sensing, topography and coastline, and patch masks. Targeted mining of pixel spectral temporal features, topographic and coastal constraint features, and patch neighborhood topological connectivity features overcomes the limitations of traditional single-reliance on spectral feature analysis. By integrating multi-dimensional feature matching, this study fully considers the spectral temporal variation patterns of mangroves, coastal topographic growth constraints, and patch spatial topological distribution characteristics. This enriches the feature dimensions of mangroves, effectively compensating for the shortcomings of single feature information being one-sided and having poor anti-interference capabilities, and significantly improving the comprehensiveness and accuracy of subsequent weak label credibility judgment. Based on multi-dimensional feature fusion data, weak label credibility analysis is conducted. Through multi-dimensional feature credibility mapping, neighborhood cross-pixel correlation feature optimization, and multi-factor adaptive weight allocation, technical problems such as single and fixed scoring methods for weak labels are solved. This effectively utilizes pixel neighborhood spatial correlation information to correct single-point feature biases and improves the spatial rationality of credibility evaluation. Simultaneously, through pixel-by-pixel multi-dimensional fit identification and hierarchical credibility threshold determination, refined hierarchical quality control of weak labels across the entire raster is achieved. This accurately distinguishes label samples of different credibility levels, avoiding the problems of traditional weak labels lacking hierarchy and mixed credibility, reducing the probability of mislabeling and omission, and providing a highly reliable and quantifiable weak label data foundation for subsequent sample correction and temporal expansion. By constructing a two-dimensional discrimination matrix, accurate classification of sample validity is achieved. Anomalous patches on the northern edge of mangroves are extracted and adaptively corrected, effectively solving the problems of handling fragmented anomalous samples in the coastal ecotone and large edge recognition bias, significantly improving the accuracy and stability of the overall sample label. Simultaneously, based on optimized sample data, time-series analysis and state trend feature mining are conducted. Pseudo-label time-series expansion is completed through trend feature mapping, compensating for the shortcomings of traditional mangrove sample time-series coverage being incomplete and having data gaps. This effectively enriches the long-term time-series sample dataset, providing continuous and reliable label data support for long-term dynamic monitoring and ecological evolution analysis of mangroves.

[0015] Therefore, the mangrove weak label quality assessment and temporal extension method of this invention integrates multi-dimensional features of spectral time series, topography, coastline, and patch topology to conduct weak label quality assessment and constructs a credibility grading verification mechanism, effectively improving the problems of mislabeling, omissions, and uneven label credibility caused by single feature modeling. Simultaneously, it achieves adaptive correction for anomalous patches in the northern mangrove ecotone and completes temporal extension of sample pseudo-labels, overcoming the shortcomings of existing technologies, effectively eliminating data gaps and identification biases in long-term time series monitoring, and significantly improving the overall accuracy and data continuity of mangrove time series monitoring. Attached Figure Description

[0016] Figure 1 This is a flowchart illustrating the steps of a method for assessing and extending the quality of weakly labeled mangroves according to the present invention. Figure 2 for Figure 1 A detailed flowchart illustrating the implementation steps of step S4. The realization of the objective, functional features and advantages of the present invention will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0017] The technical method of the present invention will now be clearly and completely described with reference to the accompanying drawings. Obviously, the described embodiments are only some, not all, of the embodiments of the present invention. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without inventive effort are within the scope of protection of the present invention.

[0018] Furthermore, the accompanying drawings are merely illustrative of the invention and are not necessarily drawn to scale. The same reference numerals in the drawings denote the same or similar parts, and therefore repeated descriptions of them will be omitted. Some block diagrams shown in the drawings are functional entities and do not necessarily correspond to physically or logically independent entities. Functional entities may be implemented in software, or in one or more hardware modules or integrated circuits, or in different network and / or processor methods and / or microcontroller methods. The term "and / or" as used herein includes any and all combinations of one or more of the associated items listed.

[0019] To achieve the above objectives, please refer to Figures 1 to 2 This invention provides a method and system for quality assessment and time-series expansion of weakly labeled mangroves. In the embodiments of this invention, please refer to... Figure 1 The diagram shown is a flowchart illustrating the steps of a mangrove weak label quality assessment and time-series expansion method according to the present invention. The mangrove weak label quality assessment and time-series expansion method includes the following steps: Step S1: Obtain heterogeneous mangrove monitoring data for the entire study area. The heterogeneous mangrove monitoring data includes mangrove baseline mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Based on the heterogeneous mangrove monitoring data, perform spatial comprehensive baseline processing of mangrove multi-source data to generate mangrove spatial baseline multi-source comprehensive data. In this embodiment of the invention, a comprehensive collection and aggregation of heterogeneous mangrove monitoring data covering the entire study area is performed. The aggregated heterogeneous data strictly covers four basic data types: mangrove baseline mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Each type of data fully covers the entire spatial range of the pre-defined study area. Among them, the mangrove baseline mapping mask data carries the known spatial distribution range and patch boundary benchmark information of mangroves in the study area; the topographic raster data carries the overall topographic base information of the study area, such as surface elevation and slope morphology; the coastal zone vector auxiliary data carries the coastal zone spatial auxiliary information, such as shoreline trend, coastal zoning, and tidal flat range; and the multi-year continuous time-series remote sensing data carries the long-term surface spectral and vegetation dynamic observation information of the study area. The multiple types of data complement each other, forming a multi-format, multi-dimensional, and multi-temporal heterogeneous mangrove monitoring data system. After completing the collection of heterogeneous data across the entire region, multi-source data spatial comprehensive benchmark processing was carried out to unify the spatial framework and expression standards of all heterogeneous data. Using the spatial coordinate system and pixel scale of multi-year continuous time-series remote sensing data as a unified geographic benchmark, global spatial coordinate projection transformation and resolution resampling registration were performed on all raster-type heterogeneous data to ensure uniformity in the spatial location, pixel scale, and projection framework of all raster data, eliminating spatial misalignment and scale inconsistencies among multi-source raster data. Global rasterization transformation was performed on coastal zone vector auxiliary data, and grid mapping of vector elements was completed according to the unified raster spatial standard, achieving integrated spatial registration of vector and raster data. Classified professional preprocessing operations were performed on the registered integrated data. Seasonal feature unification and normalization processing was carried out on multi-year time-series remote sensing data to weaken feature disturbances caused by differences in illumination and phenology across different seasons. Topographic correlation factors were extracted from topographic raster data to mine topographic feature information suitable for mangrove growth and distribution. Coastal location distance features were extracted from coastal zone auxiliary raster data to construct a global coastal location feature layer. All extracted feature data from different layers undergo unified standardization processing, unifying the numerical expression range and dimensional system of various feature data to eliminate interference from numerical differences across different feature dimensions. Finally, using a single pixel as the basic spatial unit, a global spatial layer benchmark encapsulation and integration is performed on all standardized layered feature layers. This allows multiple types of standard feature information to be integrated into the same spatial pixel, constructing a unified spatial benchmark, well-structured layers, complete feature dimensions, and continuous spatial coverage of mangrove spatial benchmark multi-source comprehensive data. This provides standardized and reliable global basic data support for subsequent multi-dimensional feature fusion analysis and weak label credibility analysis.

[0020] Step S2: Perform multi-dimensional feature fusion analysis on the multi-source integrated spatial benchmark data of mangroves to generate multi-dimensional feature fusion data of mangroves; In this embodiment of the invention, a multi-dimensional feature fusion analysis is conducted on multi-source comprehensive data of mangrove spatial benchmarks. The integrated comprehensive data is decomposed into three independent and spatially matched hierarchical data systems through a hierarchical separation approach: remote sensing image hierarchical data, topographic and coastal hierarchical data, and mangrove mapping mask patch hierarchical data. This achieves decoupling and hierarchical layering of spectral temporal dimensions, environmental constraint dimensions, and spatial topological dimensions. Based on the remote sensing image hierarchical data, spectral feature analysis and temporal information extraction at the pixel scale are performed across the entire domain. The spectral evolution patterns and vegetation phenological change characteristics over long time series of single pixels are explored, constructing a spectral temporal feature system capable of characterizing the temporal growth state of mangroves. Based on the topographic and coastal hierarchical data, environmental analysis of the entire domain's topography and coastal location is conducted, exploring the constraint mechanisms of environmental factors such as elevation, slope, and shoreline distance on the spatial distribution of mangroves, and constructing a comprehensive topographic and coastal constraint feature system. Based on layered mangrove mapping mask patch data, spatial topology analysis and neighborhood connectivity analysis of mangrove patches were conducted to explore the spatial morphological structure, neighborhood association status, and patch cluster distribution characteristics of mangrove patches, constructing a patch neighborhood topology connectivity feature system. Building upon the construction of three independent feature dimensions, multi-dimensional feature matching and fusion processing was carried out. Using unified pixel spatial location as the matching carrier, multi-level fusion and overlay of spectral temporal features, topographic and coastal constraint features, and patch topological features were achieved. This eliminated the problem of one-sided information from a single feature dimension, forming multi-dimensional mangrove feature fusion data that covers the entire region, has complete dimensions, and strong spatial correlations, providing comprehensive feature support for subsequent quantitative assessment of weakly labeled reliability.

[0021] Step S3: Perform a comprehensive raster-based analysis of mangrove forests based on the multi-dimensional feature fusion data to generate comprehensive raster-based data of mangrove forests with reliable weak labels. In this embodiment of the invention, a systematic analysis and processing of weak, credible labels for the entire raster network is carried out based on multi-dimensional feature fusion data of mangroves. Relying on the complete spatial feature information contained in the fusion data—including spectral time-series dimension, topographic and coastal constraint dimension, and patch topological connectivity dimension—the quantification and structured generation of weak label credibility metrics for raster units across the entire region are completed. By performing a unified credibility feature mapping transformation on the multi-dimensional fusion features across the entire region, multi-source spatial features with different attributes and representation forms are uniformly transformed into quantitative credibility information that reflects the rationality of mangrove pixel attribution, completing the overall transformation from feature space to credibility evaluation space. Combining the spatial correlation characteristics of raster neighborhoods, cross-pixel feature derivation and credibility optimization control are carried out across the entire region. Utilizing the continuity of land cover distribution and the correlation of patch structure within a local spatial range, the evaluation bias caused by fluctuations in single-point pixel features is improved, ensuring that the spatial distribution of credibility across the entire region conforms to the actual patch distribution pattern and edge transition characteristics of mangroves. By employing a differentiated weight allocation mechanism based on multi-dimensional feature spatial adaptability, the effective contribution capabilities of various feature dimensions at different spatial locations are distinguished, enabling adaptive integration and expression of the credibility of multi-feature fusion across the entire domain, pixel by pixel. This results in multi-dimensional adaptation feature results that comprehensively reflect the adaptation level of mangrove distribution in a single pixel. Based on a unified, pre-defined hierarchical credibility judgment standard, a hierarchical credibility assessment is conducted on the adaptation results of the entire domain raster, achieving a systematic distinction between weak labels of different quality levels. Based on the hierarchical assessment results, a standardized and structured construction of weak labels for the entire domain raster is completed, forming mangrove entire domain raster credible weak label data with hierarchical credibility attributes, full spatial coverage, and distinguishable quality differences. This provides a quality-controllable and hierarchically clear weak label data foundation for subsequent sample correction and optimization and temporal pseudo-label expansion.

[0022] Step S4: Based on the mangrove global raster reliable weak label data and mangrove baseline mapping mask data, perform mangrove layer sample correction and optimization processing to generate mangrove layer sample correction and optimization data; perform pseudo-label design processing for mangrove layer time-series extension on the mangrove layer sample correction and optimization data to generate mangrove layer pseudo-label time-series extension data.

[0023] In this embodiment of the invention, the entire process of mangrove layer sample correction and optimization and temporal pseudo-label expansion is completed through the collaborative processing of global raster reliable weak label data and mangrove benchmark mapping mask data. Combining the complementary advantages of these two types of high-precision spatial data, the technical problems of patch anomalies, edge misclassification, inconsistent sample quality, and discontinuous temporal sample coverage in traditional mangrove weak label samples are solved. Pixel-level spatial matching and layer overlay association are achieved between global raster reliable weak label data and mangrove benchmark mapping mask data. Based on the standard spatial distribution boundary of the benchmark mask and the multi-level reliability attributes of the weak labels, a two-dimensional sample validity identification system is established to complete the refined validity division of the entire sample layer in the study area, accurately distinguishing the spatial distribution areas of valid standard samples, low-confidence samples requiring optimization, and spatially mismatched abnormal samples within the entire area. To address the fragmented nature of mangrove ecosystem ecotone patches, the mixed distribution of land features, and the susceptibility to labeling errors, this study specifically extracts and categorizes anomalous local patches across the entire region. Based on the spatial morphology, neighborhood topology, and reliability distribution patterns of high-quality, valid patches within the region, a joint patch correction feature system adapted to the unique ecological scenario of the northern edge is constructed. Adaptive correction and optimization processing is then applied to anomalous patches, correcting labeling defects and spatial structural anomalies in localized areas. This achieves overall correction and optimization of the entire mangrove sample layer, resulting in mangrove layer sample correction and optimization data with spatial distribution consistent with the actual ecological pattern, standardized labeling, and clear reliability levels. Building upon the optimized sample data, long-term spatiotemporal analysis of the layer is conducted. The spatial attributes of single-period optimized samples are deeply coupled and matched with multi-year continuous time-series remote sensing observations to analyze the evolution patterns of mangrove patch expansion, contraction, and stable existence at different time points, extracting long-term mangrove state trend characteristics corresponding to the entire region's raster units. By inferring the regularity of each missing time-series observation node through temporal trend characteristics, the sample completion and pseudo-label structured design of the time-series dimension are completed, realizing the extension and expansion of static optimized samples to a long-term continuous sample system. Finally, the pseudo-label time-series extended data of mangrove layers with complete time-series coverage, evolution logic that conforms to the ecological evolution law of mangroves, and stable and reliable spatial annotation are generated, providing continuous and high-quality sample data support for long-term dynamic monitoring and ecological evolution analysis of mangroves.

[0024] Furthermore, step S1 includes the following steps: Step S11: Obtain heterogeneous monitoring data of mangroves across the entire study area; In this embodiment of the invention, relying on a pre-defined mangrove monitoring coverage area, the heterogeneous mangrove monitoring data for complete coverage of the study area is collected. The types of data collected strictly match the requirements of mangrove weak label quality assessment and time-series expansion for comprehensive analysis. Specifically, they include mangrove baseline mapping mask data, comprehensive topographic raster data, coastal zone vector auxiliary data, and long-term time-series remote sensing data with continuous coverage year by year. The spatial coverage of all collected data is completely aligned with the pre-defined study area boundary, achieving full data coverage without any blank areas in the study area. Various heterogeneous data retain their original spatial and attribute information, providing a complete original data foundation for subsequent multi-source data registration, feature extraction, and fusion analysis, ensuring the data integrity of subsequent mangrove spatial baseline construction work.

[0025] Step S12: Using multi-year continuous time-series remote sensing data as a unified geographic benchmark, perform coordinate projection unified transformation and resolution resampling registration on all raster data in the heterogeneous mangrove monitoring data to generate raster spatial benchmark aligned data. In this embodiment of the invention, multi-year continuous time-series remote sensing data of the study area is set as a unified geographic reference framework for multi-source raster data across the entire region. Using the spatial projection coordinate system and spatial resolution of this reference data as the standard, unified coordinate projection transformation and spatial resolution resampling registration are performed on all heterogeneous raster data within the study area. The spatial projection parameters of various types of raster data are uniformly adjusted to ensure that all raster data are adapted to a geographic coordinate system consistent with the time-series remote sensing data. Simultaneously, a resampling algorithm is used to achieve scale unification of raster data at different resolutions, eliminating coordinate offsets, scale mismatches, and spatial misalignments among multi-source raster data. After registration, all raster data possess a unified spatial reference system, forming raster spatial reference aligned data with mutually matching spatial locations and a unified scale system, laying a unified spatial foundation for subsequent vector data rasterization and fusion processing.

[0026] Step S13: Based on the unified projection and resolution of the raster spatial reference alignment data, perform rasterization conversion on the coastal zone vector auxiliary data within the heterogeneous mangrove monitoring data to generate integrated vector raster registration data; In this embodiment of the invention, based on the unified projection parameters and spatial resolution determined by the raster spatial reference alignment data, the collected coastal zone vector auxiliary data undergoes a full-domain rasterization conversion process. Following a unified pixel scale and spatial boundary, the vector format coastal zone boundary, shoreline location, and coastal geomorphological auxiliary information are mapped element by element to the raster grid cells, achieving format conversion and spatial registration from vector data to raster data. This ensures that the converted vector-derived raster data fully matches the spatial location, pixel scale, and projection system of the remaining raster monitoring data. This processing method achieves integrated spatial fusion of vector and raster data, forming integrated vector-raster registered data, breaking down the spatial fusion barriers between different formats and multiple sources.

[0027] Step S14: Perform categorized preprocessing operations on the vector raster integrated registration data, including normalizing the seasonal features of time-series remote sensing data, extracting topographic data factors, and extracting coastal location distance raster data to generate categorized mangrove heterogeneous feature data. Then, perform categorized data standardization processing on the categorized mangrove heterogeneous feature data to generate categorized mangrove standard feature data. In this embodiment of the invention, categorized preprocessing operations are performed on vector raster integrated registration data, and seasonal feature normalization processing is performed on time-series remote sensing data to eliminate spectral feature deviations caused by differences in illumination angle, vegetation phenology, and meteorological environment in remote sensing images from different seasons, thus unifying the seasonal feature expression standards of time-series remote sensing data. Multiple topographic factor extraction operations are performed on topographic raster data to mine topographic feature information related to mangrove growth and distribution, such as elevation, slope, aspect, and tidal level difference in the study area. Coastal zone raster data undergoes coastal location distance raster extraction to generate distance feature layers from pixels to the shoreline, nearshore waters, and tidal flat boundaries, completing the hierarchical extraction of multiple heterogeneous features to obtain classified mangrove heterogeneous feature data. All hierarchically extracted heterogeneous feature data undergo unified standardization processing to unify the numerical range and dimensional system of various feature data, avoiding analytical interference caused by differences in the numerical magnitude of different feature data, and generating classified mangrove standard feature data.

[0028] Step S15: Perform unified pixel spatial layer benchmark encapsulation processing on the standard feature data of classified mangroves to generate multi-source integrated spatial benchmark data of mangroves.

[0029] In this embodiment of the invention, a single pixel is used as the smallest spatial unit. A global spatial layer benchmark encapsulation process is performed on the standard feature data of classified mangroves. Multiple standard feature layers, including normalized remote sensing seasonal features, topographic factor features, and coastal location distance features, are layered and bound according to a unified pixel spatial location. Single pixels at the same geographical location aggregate to carry multi-dimensional standard feature information, completing the spatial integration and encapsulation of multi-source layered feature data. This constructs a unified multi-source comprehensive spatial benchmark data for mangroves with a unified layer structure, consistent spatial benchmark, and complete feature dimensions. This provides standardized and integrated global spatial feature data support for subsequent core processes such as multi-dimensional feature fusion of mangroves and weak label credibility analysis.

[0030] Furthermore, step S2 includes the following steps: Step S21: Perform mangrove hierarchical separation processing based on multi-source integrated mangrove spatial benchmark data to generate mangrove hierarchical separation data, wherein the mangrove hierarchical separation data includes remote sensing image hierarchical data, topographic and coastal hierarchical data, and mangrove mapping mask patch hierarchical data. In this embodiment of the invention, mangrove hierarchical separation processing is performed based on multi-source integrated mangrove spatial benchmark data. Relying on the attribute dimensions and spatial semantic differences of various feature data, the integrated multi-source layers are decomposed and split into layers, achieving decoupling and layering of different types of geographic feature data, resulting in structurally independent and dimensionally clear mangrove hierarchical separation data. The layering process strictly classifies data according to semantic attributes. Image data carrying temporal spectral information is uniformly classified as remote sensing image layered data, which fully preserves the temporal spectral variation information and spatial distribution structure of pixels across the entire region. Geographical environmental data carrying topographic relief and coastal location constraints are uniformly classified as topographic coastal layered data, which covers the spatial distribution information of topographic factors and coastal location distances across the entire study area. Mask data carrying the boundaries and distribution range of mangrove spatial patches are uniformly classified as mangrove mapping mask patch layered data, which fully records the spatial outline and overall distribution pattern of mangrove patches. The three-layered data maintains the same spatial benchmark and pixel grid structure as the original comprehensive data. Each layer of data is independent of each other and their spatial positions are completely corresponding, providing structured data support for subsequent multi-dimensional feature mining and specialized feature analysis, and ensuring targeted extraction and operation of features of different dimensions.

[0031] Step S22: Perform pixel-scale spectral temporal feature analysis based on remote sensing image layered data to generate mangrove spectral temporal feature data; In this embodiment of the invention, pixel-scale spectral temporal feature analysis is performed based on layered remote sensing image data. Using a single pixel as the basic analysis scale, the spectral response information of each pixel within the layered data is analyzed across the entire domain. The differences in spectral values, the magnitude of spectral changes, and the temporal evolution patterns of pixels at different time points are extracted to construct layered pixel-scale spectral feature data of the remote sensing image. Relying on the continuous time dimension attribute of long-term remote sensing data, temporal information interpretation and extraction are performed on the static spectral features of pixels across the entire domain. This uncovers the phenological evolution characteristics, seasonal change patterns, and vegetation spectral stability characteristics of single pixels over long time series. The spectral temporal characteristics of non-mangrove vegetation, bare beaches, water bodies, and other disturbing features are removed, while retaining spectral temporal information adapted to mangrove growth patterns. Through pixel-by-pixel and time-by-time stage feature analysis, the static spectral features and temporal dynamic features of all pixels across the entire domain are integrated, completing the structured integration of the entire domain's spectral temporal information. This generates dimensionally complete and spatially comprehensive mangrove spectral temporal feature data, providing core spectral temporal dimension feature support for subsequent multi-dimensional feature fusion.

[0032] Step S23: Perform mangrove topographic and coastal constraint feature analysis based on topographic and coastal layered data to generate mangrove topographic and coastal constraint feature data; In this embodiment of the invention, mangrove topographic and coastal constraint characteristics are analyzed based on topographic and coastal layered data. Spatial semantic decomposition is performed on the topographic elevation, slope, and coastal distance factors included in the layered data to clarify the correlation between various topographic and coastal factors and the spatial distribution of mangroves, thus completing the construction of comprehensive topographic and coastal location analysis data. Based on the location analysis results, topographic and coastal location characteristic analysis is conducted, sorting out the topographic undulation, coastal distance distribution patterns, and tidal flat location attributes of different location units. Coastal topographic location intervals suitable for mangrove growth are delineated, and topographic and coastal location constraint characteristics across the entire region are extracted to form standardized topographic and coastal location characteristic data. Combining the environmental attributes of mangrove coastal growth, this study conducts a comprehensive constraint feature extrapolation based on topographic and coastal location data. It distinguishes the environmental differences between gently sloping mudflats and steep shorelines based on topographic factors, and differentiates the spatial range of suitable nearshore growth areas from unsuitable offshore growth areas based on coastal distance factors. This fully explores the constraining effect of topography and coastal environment on the spatial distribution of mangroves, integrates all constraint feature information, and generates comprehensive mangrove topographic and coastal constraint feature data covering the entire region, thus overcoming the technical shortcomings of single spectral features failing to reflect environmental constraints.

[0033] Step S24: Perform patch neighborhood topological connectivity feature analysis based on mangrove mapping mask patch layered data to generate mangrove patch neighborhood topological connectivity feature data; In this embodiment of the invention, patch neighborhood topological connectivity feature analysis is performed based on layered data of mangrove mapping mask patches. Using the outlines of mangrove patches in the layered data as the analysis carrier, the spatial morphology, patch boundary range, and spatial distribution pattern of all mangrove patches within the entire area are identified. The spatial topological relationships of each independent patch are statistically analyzed, and the spatial arrangement, gap distribution, and connectivity range of adjacent patches are sorted out to construct mangrove patch spatial topological data covering the entire study area. Based on the patch spatial topological data, refined patch neighborhood topological connectivity feature analysis is carried out. Taking a single mangrove patch as the core unit, the analysis is extended to the connectivity characteristics of land features, patch adjacency relationships, and patch cluster distribution characteristics within the surrounding neighborhood of the patch. The spatial connectivity attributes of the patch neighborhood, the topological dependency characteristics of adjacent patches, and the spatial association characteristics between scattered small patches and the main patch are extracted. The neighborhood topological information of all patches in the entire area is unified and integrated to form complete mangrove patch neighborhood topological connectivity feature data. This adds spatial topological dimension feature support for subsequent multi-dimensional feature fusion and improves the completeness of the expression of mangrove spatial distribution features.

[0034] Step S25: Perform multi-dimensional feature matching and fusion processing on mangrove spectral temporal feature data, mangrove topographic and coastal constraint feature data, and mangrove patch neighborhood topological connectivity feature data to generate multi-dimensional feature fusion data of mangroves.

[0035] In this embodiment of the invention, multi-dimensional feature matching and fusion processing is performed on the mangrove spectral temporal feature data, mangrove topographic and coastal constraint feature data, and mangrove patch neighborhood topological connectivity feature data generated in the previous steps. Using a unified pixel spatial grid as the matching benchmark, precise spatial correspondence of the three types of feature data with different dimensions is achieved. Using a single pixel as the fusion unit, the spectral temporal dynamic features, topographic and coastal environmental constraint features, and patch neighborhood topological connectivity features corresponding to the pixel are matched in multiple dimensions to establish a structural system for the collaborative expression of multiple features in a single pixel, breaking down the feature barriers of the spectral temporal dimension, environmental constraint dimension, and spatial topological dimension. The feature information after multi-dimensional matching is then fused and integrated to retain the effective information of each dimension, achieving complementary gains of different types of features and avoiding the problems of one-sided information and weak anti-interference ability of single-dimensional feature data. Through multi-feature matching and fusion operations on a pixel-by-pixel basis across the entire study area, multi-dimensional features of all regions are integrated to generate multi-dimensional feature fusion data of mangroves that covers the entire area, has multiple dimensions, and is complete in information. This provides a comprehensive and reliable multi-dimensional feature data foundation for subsequent core processes such as mangrove weak label credibility assessment and weak label quality optimization.

[0036] Furthermore, step S22 includes the following steps: Pixel-scale spectral feature analysis is performed on the layered data of remote sensing images to generate layered pixel-scale spectral feature data of remote sensing images. Then, temporal information interpretation and extraction processing is performed on the layered pixel-scale spectral feature data of remote sensing images to generate mangrove spectral temporal feature data.

[0037] In this embodiment of the invention, based on the hierarchical separation of multi-source mangrove data and the acquisition of standardized remote sensing image layered data, pixel-scale spectral feature analysis is conducted across the entire area. Using the basic pixels of the remote sensing image layered data as the basic analysis unit, the spectral response information of all pixels within the entire study area is analyzed and its features are mined layer by layer. This fully explores the spectral reflectance patterns corresponding to different land cover types, accumulates basic spectral characterization information of all pixels in the entire area, and constructs a structurally complete and spatially comprehensive layered pixel-scale spectral feature data of remote sensing images. This process relies on the inherent spectral differences of different land cover types such as mangrove vegetation, coastal bare beaches, nearshore waters, and terrestrial vegetation to systematically sort out the spectral attributes of all pixels in the entire area, distinguish the spectral characterization patterns of different land cover types, and construct a basic feature dataset that reflects the spectral distribution pattern of land cover in the study area, providing stable static spectral feature support for subsequent temporal information mining. After constructing the static spectral feature data, we conducted full-domain temporal information interpretation and extraction processing on the layered pixel-scale spectral feature data of remote sensing images. Based on the temporal dimension attributes of multi-year continuous temporal remote sensing data, we continuously analyzed the spectral feature variation patterns of pixels at the same spatial location at different temporal nodes, sorted out the spectral evolution trend, seasonal fluctuation patterns, and long-term stable spectral characterization status within the vegetation growth cycle, extracted temporal information that can reflect the continuous growth characteristics of mangrove vegetation, and screened out invalid spectral change information caused by short-term meteorological interference and instantaneous surface environmental fluctuations. We integrated the static spectral attributes and dynamic temporal evolution features of pixels across the entire domain, completed the structured integration of multi-temporal dimension features, and generated mangrove spectral temporal feature data, providing refined spectral temporal dimension support for subsequent multi-dimensional feature fusion and weak label credibility assessment.

[0038] Furthermore, step S23 includes the following steps: Based on the topographic and coastal layered data, topographic and coastal location analysis is performed to generate topographic and coastal location analysis data; topographic and coastal location feature analysis is performed on the topographic and coastal location analysis data to generate topographic and coastal location feature data; and mangrove topographic and coastal constraint feature analysis is performed on the topographic and coastal location feature data to generate mangrove topographic and coastal constraint feature data.

[0039] In this embodiment of the invention, topographic and coastal layered data obtained after hierarchical separation processing are used to conduct comprehensive topographic and coastal location analysis. Using a unified pixel grid as the spatial carrier, the topographic and coastal location elements contained in the layered data are spatially decomposed and semantically deconstructed, completing a systematic analysis of the overall surface topography and coastal spatial pattern. This process spatially divides the study area into topographic relief units, tidal flat spatial ranges, and shoreline distribution patterns. It then performs layer-by-layer matching and association of the topographic base attributes and coastal spatial location attributes corresponding to each pixel, realizing the correspondence between topographic elements, coastal elements, and surface spatial units. This constructs topographic and coastal location analysis data that can fully reflect the topographic base and coastal spatial pattern of the study area, providing a standardized spatial analysis carrier for subsequent location feature mining. Based on the completed topographic and coastal location analysis data, a comprehensive topographic and coastal location feature analysis was conducted. This analysis explored the distribution patterns of topography and geomorphology in different spatial units, as well as the distribution patterns of coastal location. The distribution characteristics of spatial elements such as elevation undulation, slope morphology, and shoreline distance were analyzed. The spatial attributes of coastal wetlands in different location units were summarized, and the structured extraction and integration of the comprehensive coastal topographic and location features were completed, forming a complete and spatially continuous topographic and coastal location feature data set. Based on the topographic and coastal location feature data, a topographic and coastal constraint feature analysis was conducted focusing on mangrove distribution patterns. Combining the unique coastal growth adaptation characteristics of mangroves with the overall topographic and coastal location conditions, the environmental constraint levels of different spatial regions were classified. The constraining effects of topography and coastal spatial patterns on mangrove vegetation growth and patch distribution were analyzed. The topographic restriction features and coastal location constraint features across the entire region were integrated, and multiple types of environmental constraint information were fused and integrated to generate mangrove topographic and coastal constraint feature data that objectively reflects the environmental constraints on mangrove coastal growth. This provides core environmental feature support for subsequent multi-dimensional feature fusion and weak label credibility judgment.

[0040] Furthermore, step S24 includes the following steps: Based on the layered data of mangrove mapping mask patches, spatial topological analysis of mangrove patches is performed to generate spatial topological data of mangrove patches. Then, the topological connectivity feature analysis of the patch neighborhoods of the spatial topological data of mangrove patches is performed to generate topological connectivity feature data of the patch neighborhoods of mangrove patches.

[0041] In this embodiment of the invention, a comprehensive spatial topological analysis of mangrove patches is conducted based on the layered data of mangrove mapping mask patches obtained through pre-processing. Using the spatial outlines and distribution range of mangrove patches carried by the layered data as the basis for analysis, the spatial morphology, boundary trends, patch size, and spatial arrangement of all independent mangrove patches within the entire study area are systematically analyzed and organized. Through the spatial unit division and topological relationship analysis of the patches across the entire area, the spatial boundary composition of each independent patch, the complete internal structure of the patch, and the spatial interval between different patches are identified. The overall spatial distribution pattern and individual spatial structural attributes of the mangrove patches across the entire area are recorded, completing the structured integration of the topological information of the patches across the entire area. This constructs comprehensive and spatially continuous mangrove patch spatial topological data, providing a standardized spatial topological data base for subsequent neighborhood feature mining. Based on established mangrove patch spatial topology data, a refined analysis of patch neighborhood topological connectivity features is conducted. Taking each mangrove patch as the core spatial unit, the analysis extends outwards to its corresponding neighborhood spatial scope, systematically clarifying the spatial connectivity relationships between the core patch and surrounding adjacent mangrove patches, coastal mudflats, water bodies, and other land cover units. The spatial connectivity structure of the patch neighborhood, patch cluster distribution characteristics, and spatial dependencies between scattered patches and the main patch are quantitatively extracted. The topological association patterns and spatial structural features of the entire patch neighborhood are analyzed, and the neighborhood topological connectivity information of all patch units is integrated. This completes the comprehensive extraction and summarization of patch spatial structural dimension features, generating mangrove patch neighborhood topological connectivity feature data with rich structural dimensions and complete spatial correlations, effectively supplementing the structural feature dimensions of mangrove spatial distribution.

[0042] Furthermore, step S3 includes the following steps: Step S31: Perform mangrove multidimensional feature credibility mapping processing on the mangrove multidimensional feature fusion data to generate mangrove multidimensional feature credibility mapping data; In this embodiment of the invention, based on the previously constructed multi-dimensional feature fusion data of mangroves, a multi-dimensional feature credibility mapping process is carried out across the entire mangrove area. This fusion data uniformly integrates three core feature dimensions: spectral temporal features, topographic and coastal constraint features, and patch neighborhood topological connectivity features. All feature dimensions are uniformly matched with the same spatial grid scale and geographical benchmark, achieving seamless spatial coverage of the entire study area. This step uses a publicly interpretable multi-feature linear credibility mapping model to complete the spatial transformation from feature dimensions to credibility dimensions. The core calculation formula of the model is: Where R represents the basic confidence value of a single pixel, and F1, F2, and F3 correspond to the values ​​of spectral time-series features, topographic and coastal constraint features, and patch neighborhood topological connectivity features, respectively. These are the basic weight coefficients corresponding to each feature dimension. The model training data uses a multi-year continuous mangrove spatial distribution sample feature set in the study area. The sample feature set covers all types of feature samples from concentrated mangrove distribution areas, fragmented patch areas, and coastal transition zones. The weight coefficient ratio is determined based on the feature contribution patterns of various types of samples, forming a fixed weight system adapted to the distribution characteristics of coastal mangroves. The model uses a single raster pixel as an independent computational unit to complete a pixel-by-pixel traversal operation across the entire study area, unifying the multidimensional feature values ​​with different dimensions and physical attributes into reliable quantitative values ​​with continuous interval distribution, eliminating computational interference caused by the heterogeneity of multi-source features. After the full-domain operation is completed, the reliable quantitative results of all raster pixels are integrated with the spatial location correspondence to construct a reliable mapping data of multidimensional mangrove features with complete spatial structure and unified numerical dimensions, providing a standardized quantitative data foundation for subsequent neighborhood feature optimization and hierarchical evaluation.

[0043] Step S32: Perform neighborhood cross-pixel association derivative feature analysis on the mangrove multidimensional feature fusion data to generate neighborhood cross-pixel association derivative feature data. Then, optimize the multidimensional feature cross-pixel association credibility derivative feature mapping of the mangrove multidimensional feature credibility mapping data using the neighborhood cross-pixel association derivative feature data to generate optimized mangrove multidimensional feature credibility mapping data. In this embodiment of the invention, based on established multi-dimensional feature fusion data of mangroves, a comprehensive neighborhood cross-pixel correlation derivative feature analysis is conducted. Using a single pixel as the central unit, local spatial analysis units are constructed based on the raster set within the spatial neighborhood, fully exploring the spatial continuity and patch integrity characteristics of mangrove surface distribution. Through linkage comparison and correlation analysis of the multi-dimensional fusion features of the central pixel and surrounding neighboring pixels, the homogeneous distribution state of features, the magnitude of spatial gradient changes, and the connectivity patterns of patch structures within the local spatial range are statistically analyzed. Derivative feature information that can characterize local spatial correlation, regional feature consistency, and spatial heterogeneity is extracted, overcoming the spatial information fragmentation problem caused by the independent analysis mode of a single pixel. After completing the correlation feature mining and aggregation for each local spatial unit across the entire domain, a comprehensive, spatially continuous neighborhood cross-pixel correlation derivative feature data that can characterize the spatial correlation patterns of the entire raster domain is formed. Using neighborhood-related derived features as spatial constraints, the multi-dimensional mangrove feature credibility mapping data generated in the previous step is optimized using global derived feature mapping. The continuity constraint relationship of neighborhood spatial features is used to correct credibility deviations caused by abrupt changes in local features and interference from minor ground features in single pixels. The optimization process smooths out anomalous values ​​at single points based on the overall distribution of local spatial features, maintaining the interval uniformity of credibility values ​​in contiguous mangrove areas, while preserving reasonable gradient changes at patch edges and tidal flat transition zones to avoid homogenization of global values ​​and smoothing out real spatial differences. After iterative optimization and correction across the entire domain, pixel by pixel and neighborhood window by neighborhood window, the original discrete credibility distribution is improved overall, generating optimized multi-dimensional mangrove feature credibility mapping data with stronger spatial continuity and regional distribution patterns that better reflect the actual spatial pattern of mangroves. This provides a more spatially reasonable and stable foundation of credibility data for subsequent multi-factor weighting and hierarchical credibility assessment.

[0044] Step S33: Perform multi-factor fitness assignment weight analysis on mangroves based on the multi-dimensional feature fusion data of mangroves, and generate multi-factor fitness assignment weight data of mangroves. In this embodiment of the invention, a global multi-factor adaptation weight allocation analysis is conducted on multi-dimensional feature fusion data of mangroves. The static allocation mode with uniform weights across the entire domain is abandoned, and a spatially adaptive dynamic weight allocation algorithm is adopted to achieve differentiated spatial configuration of weights for each feature dimension. The algorithm uses three core factors—spectral temporal stability corresponding to a single pixel, topographic and coastal environmental adaptability, and patch topological integrity—as the basis for weight allocation, and constructs a dynamic weight allocation operation logic: In the formula ω i For the dynamic weights corresponding to the feature dimensions, F1, F2, and F3 represent the adaptation scores for corresponding dimensional features in a single pixel, corresponding to spectral temporal feature values, topographic and coastal constraint feature values, and patch neighborhood topological connectivity feature values, respectively. The algorithm's training data is derived from a multi-dimensional feature statistics dataset of the entire study area's raster network. By statistically analyzing the effective expressive power of each feature dimension in different spatial regions, the calculation rules for the adaptation scores of each feature dimension are determined. In areas with concentrated and contiguous mangrove distribution, the adaptation scores for spectral temporal features and patch topological features are higher, and corresponding dimensions are assigned greater weights. In coastal-tidal flat transition zones and scattered patch distribution areas, the adaptation scores for topographic and coastal constraint features are higher, and corresponding dimension weights are increased accordingly. The algorithm performs iterative calculation of pixel-by-pixel weights across the entire study area, achieving adaptive matching of feature weights at different spatial locations, aligning with the environmental and spatial structural characteristics of mangrove growth in local areas. After the overall weight calculation is completed, the multi-dimensional dynamic weight allocation information of all raster units is summarized to generate multi-factor adaptation degree weight allocation data for mangroves with spatially differentiated distribution characteristics, providing dynamic weight support for subsequent accurate fusion calculations of multi-dimensional adaptation degrees.

[0045] Step S34: Perform pixel-by-pixel mangrove multidimensional fit labeling on the optimized mangrove multidimensional feature credibility mapping data using the mangrove multi-factor fit weighting data to generate mangrove multidimensional fit labeling data. In this embodiment of the invention, the mangrove multi-factor fit weight data is used to perform pixel-by-pixel mangrove multi-dimensional fit identification processing on the optimized mangrove multi-dimensional feature credibility mapping data, and a dynamic weighted fusion model is used to complete the integration calculation of multi-dimensional credibility information. Where S is the multidimensional fit of a single pixel, and R1, R2, and R3 are the optimized confidence scores for each dimension of the feature, respectively. The dynamic weight coefficients for each pixel are derived from the dynamic weight data of that pixel. The weight ratios for each pixel are not uniform, adapting to the local spatial feature distribution. The model uses a single raster pixel as the smallest computational unit, traversing all spatial locations across the entire study area to complete the weighted fusion of multi-dimensional reliability information for each pixel. This comprehensively quantifies the matching degree between spectral temporal features, topographic and coastal constraints, patch topological connectivity features, and the actual growth and distribution of mangroves at a single pixel location. This computational mechanism effectively amplifies the expressive role of advantageous features in local areas, weakens the interference of anomalous feature dimensions on the overall computational results, and avoids the shortcomings of static weighting models in adapting to regional spatial differences. After the weighted fusion computation of all raster units across the entire area is completed, the multi-dimensional adaptation quantification results and spatial location correspondences of all pixels are uniformly summarized, forming seamless, continuously varying numerical multi-dimensional adaptation identification data for mangroves, thus completing a refined quantitative representation of the spatial distribution adaptation level of mangroves across the entire area.

[0046] Step S35: Perform a mangrove whole-domain raster-based reliability assessment on the mangrove multidimensional adaptability identification data through a preset reliability judgment threshold to generate mangrove whole-domain raster-based reliability assessment data. In this embodiment of the invention, a multi-level credibility threshold judgment system adapted to the ecological distribution characteristics of coastal mangroves is established to conduct raster-level credibility assessment processing on the multi-dimensional adaptability label data of mangroves across the entire region. The threshold system is constructed based on years of statistical data on the field distribution of mangroves in the study area and the distribution characteristics of the overall adaptability values. It divides the data into multiple levels of continuous threshold intervals, each corresponding to an independent weak label credibility level, forming a standardized, hierarchical credibility grading judgment standard. The grading judgment system includes three levels: high credibility interval, medium credibility interval, and low credibility interval. The threshold nodes at each level are determined comprehensively based on the extreme value distribution, mean distribution, and actual mangrove patch distribution range of the overall adaptability values, adapting to the scattered and complex distribution characteristics of mangrove patches and shorelines in coastal areas. The grading assessment operation uses a single raster pixel as the basic judgment unit, uniformly applying the same threshold judgment rules across the entire region. It performs interval matching and classification on the multi-dimensional adaptability values ​​of all pixels one by one, distinguishing the label credibility levels of different spatial units. The calculation process preserves the adaptation gradient characteristics of patch edges and tidal flat transition zones, without uniformly smoothing out minor local numerical differences, thus fully distinguishing the credibility level differences between concentrated distribution areas, scattered patch areas, and transition zones. After the global grading calculation is completed, the credibility level identification results and spatial distribution information of all raster pixels are integrated to generate a comprehensive raster-based credibility assessment data for mangrove forests with multi-level credibility attributes and a complete and continuous spatial distribution, providing standardized grading data support for differentiated weak-label structured design.

[0047] Step S36: Design a mangrove whole-area raster trustworthy weak label for the mangrove whole-area raster trustworthy assessment data, and generate mangrove whole-area raster trustworthy weak label data.

[0048] In this embodiment of the invention, a comprehensive mangrove raster-based reliability assessment data is used to design a comprehensive mangrove raster-based weak label system. Based on the multi-level reliability classification, differentiated weak label attribute assignments and structured construction are completed for all raster units. The weak label assignment rules correspond to the three-level interval system of the reliability assessment. High-level reliable weak label attributes are configured for raster units in high-reliability intervals, medium-level reliable weak label attributes are configured for raster units in medium-reliability intervals, and low-level reliable weak label attributes are configured for raster units in low-reliability intervals, thus constructing a hierarchical comprehensive weak label dataset. The label assignment logic relies entirely on the full-process calculation results of pre-multidimensional feature mapping, neighborhood optimization, dynamic weight fusion, and hierarchical threshold determination. The label level has a strict correspondence with the pixel multidimensional feature fit and spatial distribution rationality, ensuring that the label attributes can objectively reflect the reliability of the mangrove identification result of a single pixel. All raster units in the comprehensive area follow a unified label assignment rule, achieving full coverage weak label annotation of the study area and completely preserving the label quality differences between concentrated mangrove distribution areas, shoreline transition areas, and scattered patch distribution areas. After all raster cells have been labeled and associated with attributes, the label level information, credible quantification information and spatial location information of all spatial pixels are integrated to construct a hierarchical, spatially complete and controllable mangrove whole-area raster credible weak label data. This provides high-quality, hierarchical and quantifiable weak label data support for subsequent mangrove sample correction and optimization, abnormal patch correction and long-term pseudo-label time-series expansion processes.

[0049] Furthermore, as an embodiment of the present invention, reference is made to... Figure 2 As shown, Figure 1 A detailed flowchart illustrating the implementation steps of step S4 is provided in this embodiment. Step S4 includes: Step S41: Perform validity segmentation of mangrove whole-domain sample layers using mangrove whole-domain raster reliable weak label data and mangrove benchmark mapping mask data to generate mangrove whole-domain sample layer validity classification data; In this embodiment of the invention, the validity of mangrove whole-area sample layers is classified using mangrove whole-area raster data with reliable weak labels and mangrove baseline mapping mask data. The two types of data are uniformly matched with a consistent spatial projection system, pixel grid size, and the entire boundary of the study area. A two-dimensional identification matrix coupling model is constructed to achieve refined classification of sample validity. The model sets two independent identification dimensions: the first dimension is the binary annotation dimension of the mangrove baseline mapping mask, which, based on the whole-area spatial partitioning attribute of the baseline mask, divides the baseline mangrove distribution area and the non-mangrove distribution area, constructing a whole-area binary spatial baseline matrix; the second dimension is the weak label reliability level annotation dimension, which inherits the high, medium, and low reliability weak label level system, constructing a whole-area raster reliability level matrix. The core operation structure of the model is a dual-matrix pixel-level binding and superposition operation, with the operation formula C(M,R), where C represents the sample validity classification attribute, M represents the baseline mask binary attribute parameter, and R represents the weak label reliability level parameter. The model training and identification are based on a benchmark sample set of mangrove spatial distribution over many years in the study area and a global weakly labeled hierarchical feature sample set. Through the combination and matching relationship of these two types of parameters, four categories of global sample states are formed: high-confidence valid samples within the benchmark area, low-confidence samples requiring optimization within the benchmark area, incorrectly labeled samples outside the benchmark area, and unlabeled blank samples across the entire region. After completing pixel-by-pixel matrix matching and classification of all raster units across the entire region, the sample attribute categories, spatial distribution locations, and confidence correlation information of the global raster are integrated to generate a clear classification system and spatially complete mangrove global sample layer validity classification data.

[0050] Step S42: Extract anomalous local patches on the northern edge of the mangrove whole-area sample layer validity classification data to generate anomalous local patch data on the northern edge; In this embodiment of the invention, abnormal local patches on the northern edge of the mangrove whole-area sample layer validity classification data are extracted. Combined with the zonal distribution pattern of mangrove coastal vegetation, a fixed spatial range of the northern edge ecotone of the mangrove forest in the study area is delineated. This area belongs to the northern transition zone of mangrove vegetation distribution and has long exhibited spatial characteristics of interwoven vegetation growth, fragmented and discrete patches, and mixed land cover types. It is a concentrated distribution area of ​​abnormal patches in the whole-area sample. Using the validity classification results of the whole-area sample layer as the core judgment criterion, a comprehensive screening is conducted across the entire spatial area of ​​the northern edge ecotone, focusing on identifying local patch units where the spatial range of the baseline mask does not match the weak label annotation attributes, the confidence level is in the low range, the spatial connectivity is small, and the morphology is scattered and fragmented. A whole-area connected domain topological statistical algorithm is used to complete the structured extraction of abnormal patches. By statistically analyzing the spatial connectivity scale, morphological structural characteristics, and sample validity attributes of each independent connected patch in the northern edge area, normal patches with continuous structure and compliant attributes are distinguished from defective patches with discrete morphology and abnormal attributes. The spatial contours, coverage areas, and attribute defect types of all abnormal patches are batch-collected. The entire extraction process only completes the localization, aggregation, and feature labeling of abnormal patches, without altering the attribute information of the original global sample classification data, thus fully preserving the original spatial structure and label defect features of the abnormal patches. After the global northern edge region screening and extraction is completed, the spatial vector information and attribute feature information of all abnormal local patches are summarized to generate northern edge abnormal local patch data with complete boundaries, clear features, and accurate range, providing targeted data support for subsequent targeted adaptive correction processing.

[0051] Step S43: Extract the effective layer of the northern edge patch of the mangrove from the effective classification data of the mangrove whole area sample layer, generate the effective layer data of the northern edge patch of the mangrove, and perform joint correction feature analysis of the northern edge patch of the sample layer through the effective layer data of the northern edge patch of the mangrove, to generate the joint correction feature data of the northern edge patch of the sample layer. In this embodiment of the invention, the effective layer of mangrove patches on the northern edge is extracted from the validity classification data of the mangrove whole-area sample layer. For a pre-defined spatial range of the northern edge ecotone, attribute screening and spatial aggregation are performed on all mangrove patch units within the region. Natural mangrove patch units with compliant sample validity attributes, a medium-to-high confidence level, complete spatial connectivity, and regular and continuous morphology are retained. Fragmented and invalid patches caused by transient environmental disturbances, misidentification of land features, and label disorder are removed. All compliant and valid patch units are integrated to form a complete spatial layer, thus constructing the effective layer data of mangrove patches on the northern edge. This layer fully reflects the true and stable spatial distribution pattern and sample labeling rules of mangroves in the northern edge region and can serve as a standard feature reference system for correcting abnormal patches. Based on established effective layer data of patches along the northern edge of the forest, this study conducts multi-dimensional joint correction feature analysis of samples. A joint feature system is constructed, encompassing patch morphology, neighborhood topology, and confidence gradient dimensions. A multi-dimensional feature fusion analysis model is built, using all regular patches within the effective layer as training samples. The model extracts patch contour morphology parameters, spatial connectivity topology, and edge confidence gradient distribution features in batches, systematically collecting standard spatial distribution paradigms and attribute matching patterns of mangrove patches along the northern edge. Through multi-dimensional feature fusion and integration, a standardized feature set adapted to the characteristics of the northern edge ecotone is formed, generating comprehensive and realistically relevant joint correction feature data for sample layers of northern edge patches.

[0052] Step S44: Adaptive correction of anomalous local patches on the northern edge is performed on the northern edge patch data using the joint correction feature data of the northern edge patches in the sample layer, generating corrected local patch data on the northern edge. In this embodiment of the invention, adaptive correction processing of anomalous local patches on the northern edge is performed using joint correction feature data of patches on the northern edge of the sample layer. A multi-dimensional feature similarity matching correction model is used to complete the attribute optimization and structural correction of the anomalous patches. The core calculation formula of the model is as follows: Where Sim is the overall feature matching degree of the patch. Adjust the weight coefficients of features for each dimension. The matching score is calculated by combining the local features of the anomalous patch with the standard joint correction features, where n represents the total number of feature dimensions. The model training dataset is entirely derived from the standard feature sample set of the effective layer of the northern edge patch. The weight coefficients are set according to the proportion of influence of each dimension feature on the distribution status of the northern edge mangroves, ensuring that the correction rules conform to the ecological distribution characteristics of the coastal northern edge mangroves. The model performs patch-by-patch, multi-dimensional feature matching operations on each anomalous local patch on the northern edge, quantifying the attribute deviation type, morphological defect degree, and credibility disorder of the anomalous patch. Based on the matching score, the sample validity attribute and credibility level of the pixels inside the patch are adjusted accordingly. The correction process follows the natural gradation law of the northern edge ecotone, retaining reasonable feature transition differences at the patch edges, and only correcting anomalous features and incorrectly labeled attributes that deviate from the standard distribution paradigm, without smoothing out the normal spatial feature fluctuations of the ecological transition zone. After all anomalous patches on the northern edge have completed adaptive correction iterations, the corrected patch spatial contours, structural morphology, and sample attribute information are uniformly integrated to generate northern edge corrected local patch data with regular spatial structure and standardized attribute labeling.

[0053] Step S45: By correcting local patch data on the northern edge, perform sample correction and optimization processing on the mangrove layer validity classification data of the whole mangrove sample layer to generate mangrove layer sample correction and optimization data; In this embodiment of the invention, the validity classification data of the mangrove forest whole-domain sample layer is optimized by correcting local patch data on the northern edge. The spatial structure and standardized sample attributes of the locally corrected northern edge patches are used to accurately replace the corresponding abnormal patch areas in the whole-domain sample classification data. This achieves local update and optimization correction of defective samples on the northern edge, while the original classification attributes of other sample areas with compliant attributes and normal structure remain unchanged, ensuring the overall stability and continuity of the whole-domain sample dataset. During the correction and replacement process, attribute smoothing is performed on pixels in the transition areas of patch boundaries to weaken the spatial attribute abruptness caused by local correction, ensuring a natural transition of sample attributes between the interior and edge areas of the patches, and conforming to the real spatial distribution of mangroves. This correction and optimization specifically addresses the technical problems commonly found in the northern edge ecotone, such as fragmented patch morphology, disordered sample labeling, abnormal confidence levels, and mislabeling or omission of edge labels. It unifies the labeling specifications and quality standards of mangrove samples throughout the whole domain, taking into account both the integrity of the whole-domain sample system and the adaptability to the special ecological areas of the northern edge. After all abnormal areas in the entire domain have been corrected and integrated, a mangrove layer sample correction and optimization data with no obvious sample defects, reasonable spatial distribution, and unified attribute system has been formed, providing a high-quality and highly regular static sample data base for subsequent time series analysis, trend feature mining, and pseudo-label time series expansion.

[0054] Step S46: Perform layer time series analysis processing based on the mangrove layer sample correction and optimization data to generate mangrove layer sample time series analysis data; In this embodiment of the invention, layer time-series analysis is performed on the mangrove layer sample correction and optimization data to build a spatiotemporal coupling framework between static samples and long-term time-series observation data. Using a complete natural year as the basic time-series analysis unit, the corresponding relationship between each time-series node and the spatial grid is established. The spatial attributes, validity classification, and reliability level information of single-period high-quality corrected samples are mapped and matched grid-by-grid and time-by-time to the multi-year continuous time-series remote sensing spatial system, realizing the expansion and adaptation of static optimized samples to long-term spatiotemporal datasets. The time-series analysis process systematically sorts out the spatial distribution pattern, patch coverage, and basic changes in spatial expansion and contraction of effective mangrove sample areas under each year's time-series nodes, collects the spatial existence status of samples at different time-series stages, and comprehensively sorts out the spatiotemporal evolution of mangrove samples over multiple years. The analysis scope covers the entire time-series observation cycle of the study area, achieving full-domain spatiotemporal matching without gaps in the time-series dimension or misalignments in the spatial dimension, constructing a structured time-series sample sequence. By deeply analyzing and reorganizing the spatiotemporal dimensions, the time limitations of single-period samples are broken, and static planar sample data is transformed into continuous dynamic spatiotemporal sample sequence data. This generates time-series analysis data of mangrove layer samples with complete temporal dimensions, accurate spatial attributes, and clear evolutionary context, laying a spatiotemporal data foundation for subsequent long-term trend feature extraction and analysis.

[0055] Step S47: Perform mangrove state trend feature analysis on the time series parsing data of the mangrove layer samples to generate mangrove state trend feature data; In this embodiment of the invention, the temporal analysis data of mangrove layer samples are used to analyze the trend characteristics of mangrove status. A temporal gradient evolution analysis model is employed to extract multidimensional trend features across the entire domain. The core input data sources are the annual continuous temporal sample spatial distribution sequence, sample validity change sequence, and patch coverage evolution sequence. Based on temporal gradient difference calculation and state clustering algorithms, the long-term, continuous evolutionary patterns of mangrove vegetation are explored. The model is developed through two parts: temporal difference calculation and state classification. The first part obtains the annual evolution gradient parameters by calculating the difference between the spatial coverage of adjacent temporal node samples, the number of effective pixels, and the mean confidence level. The second part, through the clustering statistics of multi-year gradient parameters, classifies four types of temporal evolution states: stable existence, annual expansion, annual contraction, and phased fluctuation. The model training data relies on a multi-year temporal sample analysis dataset across the entire domain. Through the statistical analysis of gradient patterns in massive temporal samples, a state judgment standard suitable for the regional mangrove evolution characteristics is formed, filtering out short-term invalid change information caused by single-year meteorological fluctuations and instantaneous surface disturbances, retaining only long-term, continuous, and regular evolutionary trend characteristics. After completing the temporal trend analysis of all spatial patches and grid units in the entire region, the long-term evolution status, trend gradient parameters and temporal change patterns of each spatial location are summarized to generate mangrove state trend feature data with rich temporal dimensions, clear evolution characteristics and clear state division, providing core trend feature support for subsequent temporal pseudo-label inference and expansion.

[0056] Step S48: Map the mangrove state trend feature data to the mangrove layer sample correction and optimization data, perform pseudo-label design processing for the time-series extension of the mangrove layer, and generate mangrove layer pseudo-label time-series extension data.

[0057] In this embodiment of the invention, the trend feature data of the mangrove status across the entire region is mapped to the corrected and optimized sample data of the mangrove layer. Pseudo-label design processing for the temporal expansion of the mangrove layer is carried out, and a trend-driven temporal pseudo-label inference model is constructed. The model uses high-quality single-period corrected and optimized samples as the baseline label base and the temporal trend features of the entire region as the inference constraint condition to achieve sample completion and label expansion for long-term unlabeled nodes. The core inference logic of the model is the coupled mapping of spatial baseline labels and temporal trend features. For temporal nodes with existing valid samples, the original optimized label attributes are retained. For missing temporal nodes without valid sample coverage, pseudo-labels are assigned based on the long-term trend status of the corresponding spatial location. Stable areas continue the baseline label attributes, expanding areas increase the effective mangrove label range year by year, and shrinking areas decrease the effective mangrove label range year by year. The pseudo-label generation process strictly follows the natural growth and evolution patterns of mangrove vegetation and the characteristics of coastal ecological evolution. The label distribution status of all time-series nodes matches the long-term time-series trend, avoiding problems such as abrupt changes in time-series labels and disordered evolutionary logic, and filling the gaps in sample gaps and insufficient time-series coverage in long-term time-series monitoring. Through systematic pseudo-label design and supplementation expansion on a time-series and spatial unit-by-spatial basis, the system extends from short-term high-quality samples to a multi-year continuous time-series sample system, generating mangrove layer pseudo-label time-series extended data with complete time-series coverage, compliant evolutionary logic, and accurate spatial annotation, effectively supporting long-term dynamic monitoring and ecological evolution analysis of mangroves across the entire region.

[0058] Furthermore, step S41 includes the following steps: The raster-based weak label data and benchmark mapping mask data of mangrove forests are subjected to pixel-level overlay binding processing to generate mangrove binary identification overlay data. Based on the mangrove binary identification overlay data, a two-dimensional identification matrix for mangrove samples is designed, wherein the two-dimensional identification matrix for mangrove samples consists of the first identification dimension of the benchmark mask binary annotation result and the second identification dimension of the weak label confidence annotation result. The validity classification data of mangrove forest sample layers is generated by using the mangrove sample binary identification matrix to classify the validity of mangrove forest sample layers.

[0059] In this embodiment of the invention, pixel-level overlay binding processing is performed on the formed mangrove whole-area raster reliable weak label data and mangrove baseline mapping mask data. The two types of layer data are uniformly matched with a consistent spatial projection framework, pixel grid size, and study area boundary range, ensuring that the spatial position of each pixel corresponds throughout the whole area, and realizing accurate spatial correlation matching of attribute information of the two types of heterogeneous layers. The overlay binding operation uses a single pixel as the basic processing unit, and synchronously collects and binds the baseline mask spatial attributes and weak label reliability attributes corresponding to each pixel, so that a single pixel simultaneously carries the baseline spatial attribution information and label reliability status information. After all pixels in the whole area have completed layer-by-layer overlay matching, a mangrove binary identification overlay data with strict spatial correspondence and complete dual-attribute correlation is constructed. Based on this binary identification overlay data, a structured two-dimensional identification matrix for mangrove samples was constructed. The matrix sets two independent yet collaboratively constrained identification dimensions. The first dimension is the binary annotation result of the baseline mask, which uses the baseline mapping mask to complete the binary region division of the entire space, distinguishing the mangrove baseline distribution area to which each pixel belongs from the non-mangrove baseline distribution area, thus achieving basic discrimination at the spatial belonging level. The second dimension is the weak label confidence annotation result, which inherits the hierarchical attributes of the global raster confidence weak labels, achieving refined discrimination of pixel label quality levels. The two-dimensional identification matrix forms a multi-class sample attribute identification paradigm through the cross-combination of the two dimensions, establishing a standardized discrimination system for the entire sample. Based on the constructed two-dimensional identification matrix, the validity classification of the entire sample layer was carried out. The two-dimensional attribute matching and classification were completed for all pixels in the entire study area. Based on the results of different dimension combinations, various sample types were distinguished, such as compliant and valid samples in the benchmark area, low-confidence samples to be optimized in the benchmark area, and abnormal mislabeled samples in the non-benchmark area. The systematic hierarchical classification of the entire sample was completed, and finally, the validity classification data of the mangrove sample layer with unified discrimination logic, complete spatial coverage, and clear type classification was generated, which provides standardized sample classification support for subsequent operations such as the extraction of abnormal patches on the northern edge and sample correction and optimization.

[0060] This specification provides a mangrove weak labeling quality assessment and time-series extension system for performing the mangrove weak labeling quality assessment and time-series extension method as described above. The mangrove weak labeling quality assessment and time-series extension system includes: The multi-source data monitoring and spatial integrated benchmark module is used to acquire heterogeneous mangrove monitoring data for the entire study area. The heterogeneous mangrove monitoring data includes mangrove benchmark mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Based on the heterogeneous mangrove monitoring data, spatial integrated benchmark processing of mangrove multi-source data is performed to generate mangrove spatial benchmark multi-source integrated data. The mangrove multidimensional feature fusion module is used to perform multidimensional feature fusion analysis on multi-source comprehensive data of mangrove spatial benchmarks and generate multidimensional feature fusion data of mangroves. The mangrove weak label quality assessment module is used to perform reliable weak label analysis of the entire mangrove raster based on multi-dimensional feature fusion data of mangroves, and generate reliable weak label data of the entire mangrove raster. The mangrove time-series extension module is used to perform mangrove layer sample correction and optimization processing based on mangrove global raster reliable weak label data and mangrove baseline mapping mask data, generating mangrove layer sample correction and optimization data; and to perform pseudo-label design processing on mangrove layer time-series extension of mangrove layer sample correction and optimization data, generating mangrove layer pseudo-label time-series extension data.

[0061] Therefore, the embodiments should be considered as exemplary and non-limiting in all respects, and the scope of the invention is defined by the appended claims rather than the foregoing description. Thus, all variations falling within the meaning and scope of the equivalents of the application are intended to be included within the invention.

[0062] The above description is merely a specific embodiment of the present invention, enabling those skilled in the art to understand or implement the invention. Various modifications to these embodiments will be readily apparent to those skilled in the art, and the general principles defined herein may be implemented in other embodiments without departing from the spirit or scope of the invention. Therefore, the present invention is not to be limited to the embodiments shown herein, but is to be accorded the widest scope consistent with the principles and novel features of the invention herein.

Claims

1. A method for quality assessment and time-series expansion of weakly labeled mangrove forests, characterized in that, Includes the following steps: Step S1: Obtain heterogeneous mangrove monitoring data for the entire study area. The heterogeneous mangrove monitoring data includes mangrove baseline mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Spatial integrated benchmark processing of mangrove multi-source data is performed based on heterogeneous mangrove monitoring data to generate mangrove spatial benchmark multi-source integrated data. Step S2: Perform multi-dimensional feature fusion analysis on the multi-source integrated spatial benchmark data of mangroves to generate multi-dimensional feature fusion data of mangroves; Step S3: Perform a comprehensive raster-based analysis of mangrove forests based on the multi-dimensional feature fusion data to generate comprehensive raster-based data of mangrove forests with reliable weak labels. Step S4: Based on the mangrove whole-area raster reliable weak label data and mangrove benchmark mapping mask data, perform mangrove layer sample correction and optimization processing to generate mangrove layer sample correction and optimization data; The mangrove layer sample correction and optimization data is processed by designing pseudo-labels for time-series extension of the mangrove layer, generating time-series extended pseudo-label data for the mangrove layer.

2. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 1, characterized in that, Step S1 includes the following steps: Step S11: Obtain heterogeneous monitoring data of mangroves across the entire study area; Step S12: Using multi-year continuous time-series remote sensing data as a unified geographic benchmark, perform coordinate projection unified transformation and resolution resampling registration on all raster data in the heterogeneous mangrove monitoring data to generate raster spatial benchmark aligned data. Step S13: Based on the unified projection and resolution of the raster spatial reference alignment data, perform rasterization conversion on the coastal zone vector auxiliary data within the heterogeneous mangrove monitoring data to generate integrated vector raster registration data; Step S14: Perform categorized preprocessing operations on the vector raster integrated registration data, including normalizing the seasonal features of time-series remote sensing data, extracting topographic data factors, and extracting coastal location distance raster data to generate categorized mangrove heterogeneous feature data. Then, perform categorized data standardization processing on the categorized mangrove heterogeneous feature data to generate categorized mangrove standard feature data. Step S15: Perform unified pixel spatial layer benchmark encapsulation processing on the standard feature data of classified mangroves to generate multi-source integrated spatial benchmark data of mangroves.

3. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 1, characterized in that, Step S2 includes the following steps: Step S21: Perform mangrove hierarchical separation processing based on multi-source integrated mangrove spatial benchmark data to generate mangrove hierarchical separation data, wherein the mangrove hierarchical separation data includes remote sensing image hierarchical data, topographic and coastal hierarchical data, and mangrove mapping mask patch hierarchical data. Step S22: Perform pixel-scale spectral temporal feature analysis based on remote sensing image layered data to generate mangrove spectral temporal feature data; Step S23: Perform mangrove topographic and coastal constraint feature analysis based on topographic and coastal layered data to generate mangrove topographic and coastal constraint feature data; Step S24: Perform patch neighborhood topological connectivity feature analysis based on mangrove mapping mask patch layered data to generate mangrove patch neighborhood topological connectivity feature data; Step S25: Perform multi-dimensional feature matching and fusion processing on mangrove spectral temporal feature data, mangrove topographic and coastal constraint feature data, and mangrove patch neighborhood topological connectivity feature data to generate multi-dimensional feature fusion data of mangroves.

4. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 3, characterized in that, Step S22 includes the following steps: Pixel-scale spectral feature analysis is performed on the layered data of remote sensing images to generate layered pixel-scale spectral feature data of remote sensing images. Then, temporal information interpretation and extraction processing is performed on the layered pixel-scale spectral feature data of remote sensing images to generate mangrove spectral temporal feature data.

5. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 3, characterized in that, Step S23 includes the following steps: Based on the topographic and coastal layered data, topographic and coastal location analysis is performed to generate topographic and coastal location analysis data; topographic and coastal location feature analysis is performed on the topographic and coastal location analysis data to generate topographic and coastal location feature data; and mangrove topographic and coastal constraint feature analysis is performed on the topographic and coastal location feature data to generate mangrove topographic and coastal constraint feature data.

6. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 3, characterized in that, Step S24 includes the following steps: Based on the layered data of mangrove mapping mask patches, spatial topological analysis of mangrove patches is performed to generate spatial topological data of mangrove patches. Then, the topological connectivity feature analysis of the patch neighborhoods of the spatial topological data of mangrove patches is performed to generate topological connectivity feature data of the patch neighborhoods of mangrove patches.

7. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 1, characterized in that, Step S3 includes the following steps: Step S31: Perform mangrove multidimensional feature credibility mapping processing on the mangrove multidimensional feature fusion data to generate mangrove multidimensional feature credibility mapping data; Step S32: Perform neighborhood cross-pixel association derivative feature analysis on the mangrove multidimensional feature fusion data to generate neighborhood cross-pixel association derivative feature data. Then, optimize the multidimensional feature cross-pixel association credibility derivative feature mapping of the mangrove multidimensional feature credibility mapping data using the neighborhood cross-pixel association derivative feature data to generate optimized mangrove multidimensional feature credibility mapping data. Step S33: Perform multi-factor fitness assignment weight analysis on mangroves based on the multi-dimensional feature fusion data of mangroves, and generate multi-factor fitness assignment weight data of mangroves. Step S34: Perform pixel-by-pixel mangrove multidimensional fit labeling on the optimized mangrove multidimensional feature credibility mapping data using the mangrove multi-factor fit weighting data to generate mangrove multidimensional fit labeling data. Step S35: Perform a mangrove whole-domain raster-based reliability assessment on the mangrove multidimensional adaptability identification data through a preset reliability judgment threshold to generate mangrove whole-domain raster-based reliability assessment data. Step S36: Design a mangrove whole-area raster trustworthy weak label for the mangrove whole-area raster trustworthy assessment data, and generate mangrove whole-area raster trustworthy weak label data.

8. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 1, characterized in that, Step S4 includes the following steps: Step S41: Perform validity segmentation of mangrove whole-domain sample layers using mangrove whole-domain raster reliable weak label data and mangrove benchmark mapping mask data to generate mangrove whole-domain sample layer validity classification data; Step S42: Extract anomalous local patches on the northern edge of the mangrove whole-area sample layer validity classification data to generate anomalous local patch data on the northern edge; Step S43: Extract the effective layer of the northern edge patch of the mangrove from the effective classification data of the mangrove whole area sample layer, generate the effective layer data of the northern edge patch of the mangrove, and perform joint correction feature analysis of the northern edge patch of the sample layer through the effective layer data of the northern edge patch of the mangrove, to generate the joint correction feature data of the northern edge patch of the sample layer. Step S44: Adaptive correction of anomalous local patches on the northern edge is performed on the northern edge patch data using the joint correction feature data of the northern edge patches in the sample layer, generating corrected local patch data on the northern edge. Step S45: By correcting local patch data on the northern edge, perform sample correction and optimization processing on the mangrove layer validity classification data of the whole mangrove sample layer to generate mangrove layer sample correction and optimization data; Step S46: Perform layer time series analysis processing based on the mangrove layer sample correction and optimization data to generate mangrove layer sample time series analysis data; Step S47: Perform mangrove state trend feature analysis on the time series parsing data of the mangrove layer samples to generate mangrove state trend feature data; Step S48: Map the mangrove state trend feature data to the mangrove layer sample correction and optimization data, perform pseudo-label design processing for the time-series extension of the mangrove layer, and generate mangrove layer pseudo-label time-series extension data.

9. The method for quality assessment and time-series expansion of weakly labeled mangroves according to claim 8, characterized in that, Step S41 includes the following steps: The raster-based weak label data and benchmark mapping mask data of mangrove forests are subjected to pixel-level overlay binding processing to generate mangrove binary identification overlay data. Based on the mangrove binary identification overlay data, a two-dimensional identification matrix for mangrove samples is designed, wherein the two-dimensional identification matrix for mangrove samples consists of the first identification dimension of the benchmark mask binary annotation result and the second identification dimension of the weak label confidence annotation result. The validity classification data of mangrove forest sample layers is generated by using the mangrove sample binary identification matrix to classify the validity of mangrove forest sample layers.

10. A mangrove weak-label quality assessment and time-series expansion system, characterized in that, For executing the mangrove weak label quality assessment and time-series extension method as described in claim 1, the mangrove weak label quality assessment and time-series extension system includes: The multi-source data monitoring and spatial integrated benchmark module is used to acquire heterogeneous mangrove monitoring data for the entire study area. The heterogeneous mangrove monitoring data includes mangrove benchmark mapping mask data, topographic raster data, coastal zone vector auxiliary data, and multi-year continuous time-series remote sensing data. Based on the heterogeneous mangrove monitoring data, spatial integrated benchmark processing of mangrove multi-source data is performed to generate mangrove spatial benchmark multi-source integrated data. The mangrove multidimensional feature fusion module is used to perform multidimensional feature fusion analysis on multi-source comprehensive data of mangrove spatial benchmarks and generate multidimensional feature fusion data of mangroves. The mangrove weak label quality assessment module is used to perform reliable weak label analysis of the entire mangrove raster based on multi-dimensional feature fusion data of mangroves, and generate reliable weak label data of the entire mangrove raster. The mangrove time-series extension module is used to perform mangrove layer sample correction and optimization processing based on mangrove global raster reliable weak label data and mangrove baseline mapping mask data, generating mangrove layer sample correction and optimization data; and to perform pseudo-label design processing on mangrove layer time-series extension of mangrove layer sample correction and optimization data, generating mangrove layer pseudo-label time-series extension data.

Images