Coastal area coverage surveying and mapping optimization method and device based on artificial intelligence

Abstract

The invention provides a coastal area coverage surveying and mapping optimization method and device based on artificial intelligence, and relates to the technical field of remote sensing surveying and mapping. The method comprises the following steps: acquiring multi-source remote sensing data of a coastal area and preprocessing the multi-source remote sensing data to obtain standardized processing data; inputting the standardized processing data into a preset artificial intelligence model, and fusing the spectral features and the spatial features through a feature extraction module based on a shift window attention mechanism to obtain hierarchical fusion features; and the hierarchical fusion features are processed through a classification detection module, a coastal region coverage classification result and a target detection result are output, and surveying and mapping optimization is realized. Through hierarchical feature extraction and multi-source data fusion of the artificial intelligence model, the problems of spectral and spatial feature separation, poor multi-source data adaptability and weak anti-interference capability in a traditional surveying and mapping method are solved, the coastal region coverage classification precision and the ship and other target detection efficiency are remarkably improved, and the method is suitable for popularization and application. And technical support is provided for coastal ecological protection and sustainable management.

Description

Technical Field

[0001] This invention relates to the field of remote sensing mapping technology, specifically to an artificial intelligence-based method and apparatus for optimizing coastal area coverage mapping. Background Technology

[0002] Coastal areas, as a key ecological zone connecting land and sea, bear important ecological functions such as biodiversity conservation, carbon storage, and coastal protection. However, these areas face multiple threats, including human activity disturbance, sea-level rise, and dynamic tidal changes. Accurate coverage mapping and dynamic monitoring have become core prerequisites for coastal ecological protection and sustainable management.

[0003] Existing methods for coastal cover mapping mainly rely on traditional machine learning algorithms (such as KNN and SVM) and conventional convolutional neural networks (CNN), but they have significant technical limitations: First, traditional machine learning algorithms can only rely on single spectral or spatial features for classification, and cannot achieve deep integration of the two, resulting in low classification accuracy in complex coastal environments (OA is generally below 80%). Second, conventional CNN models have shortcomings in long-distance information interaction, making it difficult to handle the heterogeneous features of multi-source remote sensing data, and have poor adaptability. Third, tidal changes cause dynamic changes in the spectral features and visibility of coastal vegetation (such as mangroves), and existing methods lack targeted anti-interference mechanisms, resulting in unstable intertidal mapping accuracy. Fourth, the detection of moving targets such as ships and coastal cover classification tasks are independent of each other, and no collaborative optimization mechanism has been formed, resulting in low overall mapping efficiency.

[0004] Therefore, how to construct an intelligent mapping method that can deeply integrate the spectral and spatial characteristics of multi-source remote sensing data, adapt to complex environmental interference such as tides, and take into account both coverage classification and target detection has become an urgent technical problem to be solved in the field of coverage mapping in coastal areas. Summary of the Invention

[0005] This invention provides an artificial intelligence-based method and apparatus for optimizing coastal area coverage mapping, which solves the problems of separation of spectral and spatial features, poor adaptability of multi-source data, and weak anti-interference ability in traditional mapping methods.

[0006] To achieve the above objectives, the present invention is implemented through the following technical solution:

[0007] In a first aspect, the present invention provides an artificial intelligence-based method for optimizing coastal area coverage mapping, comprising the following steps:

[0008] Acquire multi-source remote sensing data of coastal areas, and preprocess the multi-source remote sensing data to obtain standardized processed data;

[0009] The standardized data is input into a preset artificial intelligence model, and the feature extraction module of the preset artificial intelligence model fuses spectral features and spatial features to obtain hierarchical fused features. The feature extraction module is constructed based on the shift window attention mechanism.

[0010] The classification and detection module of the preset artificial intelligence model processes the hierarchical fusion features and outputs the coastal area coverage classification results and target detection results, thereby realizing the optimization of coastal area coverage mapping.

[0011] As a further improvement to the technical solution of the present invention, the multi-source remote sensing data includes at least two of the following: hyperspectral data, optical airborne data, and multi-temporal satellite data. The spectral coverage range of the hyperspectral data is 400-2500nm, and the image resolution of the optical airborne data is not less than 512×512 pixels.

[0012] As a further improvement to the technical solution of the present invention, the preprocessing of the multi-source remote sensing data includes:

[0013] Denoising was performed on each type of remote sensing data to remove low signal-to-noise ratio bands and invalid interference bands.

[0014] Standardization is achieved through detector calibration to eliminate systematic errors introduced by data acquisition equipment.

[0015] Dimension reduction techniques are used to reduce redundancy in high-dimensional data while preserving key feature bands.

[0016] The processed data from different sources are formatted and spatiotemporally registered to obtain the standardized processed data.

[0017] As a further improvement to the technical solution of the present invention, the feature extraction module includes a block embedding unit, a Swing Transformer concatenation module and a block merging unit connected in sequence. The Swing Transformer concatenation module contains two consecutive sub-modules, which respectively adopt a window-based multi-head self-attention mechanism (W-MSA) and a shift-window-based multi-head self-attention mechanism (SW-MSA).

[0018] As a further improvement to the technical solution of this invention, the process of fusing spectral features and spatial features includes:

[0019] The standardized data is divided into non-overlapping blocks and converted into a vector of preset dimensions by using block embedding units.

[0020] Local and global features are extracted through the Swin Transformer concatenation module, computational complexity is reduced through the multi-head self-attention mechanism (W-MSA), and cross-window information interaction is achieved through the multi-head self-attention mechanism (SW-MSA).

[0021] By downsampling the feature map through block merging units, the feature hierarchy is enhanced, and deep fusion of spectral and spatial features is achieved.

[0022] As a further improvement to the technical solution of the present invention, the method further includes:

[0023] To address the impact of tidal dynamics on coastal areas, the mangrove inundation index was introduced as a tidal characterization parameter, and a tidal adaptation model was constructed by combining multi-temporal remote sensing data.

[0024] The tidal adaptation model is embedded into the classification and detection module to correct spectral feature distortion caused by tidal changes and improve the classification accuracy of intertidal zone cover.

[0025] As a further improvement to the technical solution of the present invention, the coverage classification results output by the classification and detection module include at least six types of coastal coverage, such as mangroves, salt marshes, seagrass, rivers, land, and ocean. The target detection results include the location, quantity, and category information of ship targets, and the ship target detection accuracy is not less than 90.8%.

[0026] Secondly, the present invention provides an artificial intelligence-based coastal area coverage mapping optimization device, comprising:

[0027] The data preprocessing unit is used to acquire multi-source remote sensing data of coastal areas and perform noise reduction, standardization, dimensionality reduction and spatiotemporal registration processing on the multi-source remote sensing data to obtain standardized data.

[0028] The feature fusion unit is used to input the standardized processed data into a preset artificial intelligence model, and through the feature extraction module based on the shift window attention mechanism, fuse spectral features and spatial features to obtain hierarchical fused features.

[0029] The classification and detection unit is used to process the hierarchical fusion features through the classification and detection module, and output the coastal area coverage classification result and target detection result;

[0030] An optimization and adjustment unit is used to introduce a tidal adaptation model to correct classification and detection biases, thereby improving the anti-interference capability and accuracy of the mapping results.

[0031] Thirdly, the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor being configured to execute the computer program to implement the above-described artificial intelligence-based coastal area coverage mapping optimization method.

[0032] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described artificial intelligence-based coastal area coverage mapping optimization method.

[0033] The technical solution of the present invention has the following advantages over the prior art:

[0034] This invention integrates multi-source remote sensing data and performs standardized preprocessing. It then leverages a feature extraction module based on a shifted window attention mechanism to deeply fuse spectral and spatial features. This approach reduces computational complexity and efficiently captures local features through the W-MSA mechanism, while simultaneously enabling cross-window information interaction and accurate extraction of global features through the SW-MSA mechanism. This effectively addresses the technical pain points of traditional mapping methods, such as the separation of spectral and spatial features, poor adaptability to multi-source data, and weak anti-interference capabilities. It achieves an overall classification accuracy of 92.4%, significantly improves the coverage type recognition rate (>91%) and ship target detection rate (>90%) in coastal areas, and is adaptable to complex coastal environments such as tidal dynamics. This provides high-precision and efficient mapping data support for coastal ecological protection, sustainable management, and climate change mitigation, demonstrating strong practicality and application value. Attached Figure Description

[0035] Other features, objects, and advantages of the present invention will become more apparent from the following detailed description of non-limiting embodiments with reference to the accompanying drawings:

[0036] Figure 1 This is a schematic diagram of the framework of an artificial intelligence-based coastal area coverage mapping optimization method according to an embodiment of the present invention;

[0037] Figure 2 This is a schematic diagram of the module framework of a coastal area coverage mapping optimization device based on artificial intelligence, according to an embodiment of the present invention.

[0038] Figure 3 This is a schematic diagram of the composition of a computing device according to an embodiment of the present invention. Detailed Implementation

[0039] The technical solutions of the present invention will be clearly and completely described below with reference to the accompanying drawings of the embodiments of the present invention. Obviously, the described embodiments are only some embodiments of the present invention, and not all embodiments. All other embodiments obtained by those skilled in the art based on the embodiments of the present invention without creative effort are within the scope of protection of the present invention.

[0040] The present invention will be further described in detail below with reference to the accompanying drawings.

[0041] Reference Figure 1In a first aspect, the present invention provides an artificial intelligence-based method for optimizing coastal area coverage mapping, comprising the following steps:

[0042] Acquire multi-source remote sensing data of coastal areas, and preprocess the multi-source remote sensing data to obtain standardized processed data;

[0043] The standardized data is input into a preset artificial intelligence model, and the feature extraction module of the preset artificial intelligence model fuses spectral features and spatial features to obtain hierarchical fused features. The feature extraction module is constructed based on the shift window attention mechanism.

[0044] The classification and detection module of the preset artificial intelligence model processes the hierarchical fusion features and outputs the coastal area coverage classification results and target detection results, thereby realizing the optimization of coastal area coverage mapping.

[0045] In practice, the process begins by acquiring multi-source remote sensing data of the coastal area. Preprocessing operations such as noise reduction, standardization, dimensionality reduction, and spatiotemporal registration are then performed to obtain standardized data with a unified format and reliable quality. This standardized data is then input into a pre-defined artificial intelligence model. The model's feature extraction module, built on a shift-window attention mechanism, can simultaneously capture local and global features, achieving hierarchical fusion of spectral and spatial features and outputting hierarchical fused features rich in discriminative information. Finally, the model's classification and detection module processes the hierarchical fused features, simultaneously outputting the coastal area coverage classification results and the detection results of targets such as ships, thus optimizing the coastal area coverage mapping. This invention innovatively introduces a shift-window attention mechanism into the feature extraction process, achieving deep fusion of spectral and spatial features from multi-source remote sensing data, and solving the technical bottlenecks of feature separation and poor multi-source data adaptability in traditional methods. Its overall classification accuracy reaches 92.4%, the coastal coverage type identification rate exceeds 91%, and the ship detection rate exceeds 90%. At the same time, it minimizes overfitting through cross-validation and hyperparameter optimization, providing high-precision and high-reliability surveying and mapping support for coastal ecological protection, sustainable management, and climate change mitigation. It has a wide range of applications and strong robustness.

[0046] In some embodiments, the multi-source remote sensing data includes at least two of hyperspectral data, optical airborne data, and multi-temporal satellite data, wherein the spectral coverage range of the hyperspectral data is 400-2500nm, and the image resolution of the optical airborne data is not less than 512×512 pixels.

[0047] It should be noted that multi-source remote sensing data includes at least two of the following: hyperspectral data, optical airborne data, and multi-temporal satellite data. Hyperspectral data has a spectral coverage range limited to 400-2500 nm, capable of capturing rich spectral information of materials. Optical airborne data has an image resolution of no less than 512×512 pixels, clearly presenting target information such as ships and coastal details. Multi-temporal satellite data can reflect dynamic characteristics of time dimensions such as tides and seasonal changes. By integrating remote sensing data of different types and parameters, a complementary information set is formed, providing comprehensive data support for subsequent preprocessing and feature extraction. This invention, by clearly defining the diversity of data sources, provides a rich information foundation for subsequent feature fusion and classification detection, avoiding the problem of insufficient mapping accuracy caused by the limitations of a single data source, and further improving the stability and applicability of the overall method.

[0048] In some embodiments, preprocessing the multi-source remote sensing data includes:

[0049] Denoising was performed on each type of remote sensing data to remove low signal-to-noise ratio bands and invalid interference bands.

[0050] Standardization is achieved through detector calibration to eliminate systematic errors introduced by data acquisition equipment.

[0051] Dimension reduction techniques are used to reduce redundancy in high-dimensional data while preserving key feature bands.

[0052] The processed data from different sources are formatted and spatiotemporally registered to obtain the standardized processed data.

[0053] Adaptive filtering and denoising are performed on each type of remote sensing data to remove low signal-to-noise ratio bands and invalid interference bands severely affected by noise. Standardization is achieved through detector calibration, converting data from different sources into a unified dimension and eliminating systematic errors caused by equipment differences. Principal component analysis or feature selection and other dimensional reduction techniques are used to remove redundant information from high-dimensional data while retaining key feature bands. Coordinate system unification and time node alignment are performed on all processed data to complete spatiotemporal registration, ultimately resulting in standardized processed data with uniform format and meeting quality standards. This invention effectively removes noise, systematic errors, and redundant information from data through step-by-step targeted processing operations, improving data quality and consistency. Noise reduction reduces the negative impacts of atmospheric scattering and equipment interference; standardization eliminates systematic biases from different acquisition devices; dimensional reduction balances data processing efficiency and feature integrity; and spatiotemporal alignment ensures the collaborative usability of multi-source data, laying a solid foundation for high-precision feature fusion and classification detection.

[0054] In some embodiments, the feature extraction module includes a block embedding unit, a SwinTransformer concatenation module, and a block merging unit connected in sequence. The SwinTransformer concatenation module contains two consecutive sub-modules, which respectively employ a window-based multi-head self-attention mechanism (W-MSA) and a shift-window-based multi-head self-attention mechanism (SW-MSA).

[0055] It should be noted that the feature extraction module consists of a block embedding unit, a Swing Transformer concatenation module, and a block merging unit connected in sequence. The Swing Transformer concatenation module contains two consecutive sub-modules. The first sub-module employs a window-based multi-head self-attention mechanism (W-MSA), limiting the computational scope to a single window, thus efficiently extracting local features while reducing computational complexity. The second sub-module employs a shift-window-based multi-head self-attention mechanism (SW-MSA), achieving cross-window information interaction through a sliding window to accurately capture global features. The block embedding unit is responsible for converting the input data into feature vectors that adapt to the model, while the block merging unit enhances the hierarchical nature of features through downsampling. The three work together to complete the hierarchical extraction and fusion of features. Through the collaborative work of the block embedding unit, the Swing Transformer concatenation module, and the block merging unit, combined with the complementary advantages of W-MSA and SW-MSA, the high computational complexity of the traditional Transformer global self-attention mechanism is solved, and the deficiency of insufficient long-distance information interaction in convolutional neural networks is overcome. It can efficiently extract hierarchical fused features, balancing the efficiency and completeness of feature extraction, providing core technical support for subsequent high-precision classification and detection.

[0056] In some embodiments, the process of fusing spectral features and spatial features includes:

[0057] The standardized data is divided into non-overlapping blocks and converted into a vector of preset dimensions by using block embedding units.

[0058] Local and global features are extracted through the Swin Transformer concatenation module, computational complexity is reduced through the multi-head self-attention mechanism (W-MSA), and cross-window information interaction is achieved through the multi-head self-attention mechanism (SW-MSA).

[0059] By downsampling the feature map through block merging units, the feature hierarchy is enhanced, and deep fusion of spectral and spatial features is achieved.

[0060] In its implementation, the standardized data is first divided into equal-sized, non-overlapping blocks using a block embedding unit. A linear embedding operation is then performed on each block, converting it into a feature vector of a preset dimension, achieving uniform adaptation of the data dimensions. Subsequently, the embedded feature vectors are input into a Swing Transformer concatenation module. Through the synergistic effect of W-MSA and SW-MSA, local and global features are extracted and fused. Finally, a block merging unit downsamples the fused feature map, reducing the image's height and width to half of their original size while doubling the feature dimensions, enhancing the hierarchy and abstraction of the features, ultimately outputting a hierarchical fused feature. This invention achieves deep coupling of two core features through step-by-step modular operations. The block embedding operation ensures dimensional adaptation of the data, the Swing Transformer concatenation module achieves complementarity between local and global features, and the block merging operation enhances the hierarchy of features. The entire process ensures sufficient feature fusion while controlling computational costs, effectively improving the discriminative power of the features and providing a crucial guarantee for improving subsequent classification and detection accuracy.

[0061] In some embodiments, the method further includes:

[0062] To address the impact of tidal dynamics on coastal areas, the mangrove inundation index was introduced as a tidal characterization parameter, and a tidal adaptation model was constructed by combining multi-temporal remote sensing data.

[0063] The tidal adaptation model is embedded into the classification and detection module to correct spectral feature distortion caused by tidal changes and improve the classification accuracy of intertidal zone cover.

[0064] In practical implementation, the mangrove inundation index is first introduced as a tidal characterization parameter, which can quantify the impact of tidal inundation on the spectral characteristics of coastal vegetation. Then, combining multi-temporal remote sensing data, the spectral response patterns of coastal cover types under different tidal states (high tide, low tide, and slack tide) are analyzed, and a tidal adaptation model is constructed. This model is embedded into the classification and detection module, dynamically correcting the spectral characteristics affected by tides during the classification process to compensate for feature distortions caused by tidal changes, thereby improving the accuracy of cover classification in the intertidal zone and ensuring the reliability of the mapping results. Addressing the key interference factor of dynamic tidal changes in coastal areas, the tidal adaptation model is introduced for targeted correction, effectively solving the problems of spectral characteristic distortion and visibility changes of mangroves and other vegetation caused by tides. By combining the mangrove inundation index with multi-temporal remote sensing data, the accuracy of cover classification in the intertidal zone is improved, the impact of complex tidal environments on mapping results is reduced, and the method can maintain stable high-precision output even in dynamic coastal environments, further expanding the applicable scenarios of the method.

[0065] In some embodiments, the coverage classification results output by the classification and detection module include at least six types of coastal coverage, such as mangroves, salt marshes, seagrass, rivers, land, and ocean. The target detection results include the location, quantity, and category information of ship targets, and the ship target detection accuracy is not less than 90.8%.

[0066] It should be noted that the classification and detection module includes a classification submodule and a detection submodule. The classification submodule uses a fully connected layer combined with a softmax activation function to classify the hierarchical fusion features, outputting classification results for at least six coastal cover types, including mangroves, salt marshes, seagrass, rivers, land, and ocean. The detection submodule uses a sliding window detection algorithm to locate and identify ship targets in the hierarchical fusion features, outputting the target's location coordinates, quantity, and category information. By setting a performance threshold of no less than 90.8% accuracy in ship target detection, the reliability of the detection results is ensured, meeting the requirements of target recognition accuracy in practical applications. This invention ensures that the method can meet the needs of practical application scenarios such as coastal ecological monitoring and ship management by clearly defining the specific range of output results. At the same time, the quantified accuracy indicators provide a clear basis for the performance verification of the method, enhancing its practicality and operability.

[0067] Reference Figure 2 Secondly, the present invention provides an artificial intelligence-based coastal area coverage mapping optimization device, comprising:

[0068] The data preprocessing unit is used to acquire multi-source remote sensing data of coastal areas and perform noise reduction, standardization, dimensionality reduction and spatiotemporal registration processing on the multi-source remote sensing data to obtain standardized data.

[0069] The feature fusion unit is used to input the standardized processed data into a preset artificial intelligence model, and through the feature extraction module based on the shift window attention mechanism, fuse spectral features and spatial features to obtain hierarchical fused features.

[0070] The classification and detection unit is used to process the hierarchical fusion features through the classification and detection module, and output the coastal area coverage classification result and target detection result;

[0071] An optimization and adjustment unit is used to introduce a tidal adaptation model to correct classification and detection biases, thereby improving the anti-interference capability and accuracy of the mapping results.

[0072] It should be noted that the data preprocessing unit is responsible for acquiring multi-source remote sensing data of the coastal area, performing preprocessing operations such as noise reduction, standardization, dimensionality reduction, and spatiotemporal registration, and outputting standardized processed data. The feature fusion unit inputs the standardized processed data into a preset artificial intelligence model, and through a feature extraction module based on a shift window attention mechanism, completes the fusion of spectral features and spatial features, outputting hierarchical fused features. The classification and detection unit performs classification and target detection on the hierarchical fused features, outputting coverage classification results and target detection results. The optimization and adjustment unit introduces a tidal adaptation model to correct the deviation of the classification and detection results, improves the anti-interference ability and accuracy of the mapping results, and finally outputs optimized coastal area coverage mapping results. This invention, through modular design, breaks down the functions of data processing, feature fusion, classification and detection, and optimization and adjustment into different units, with a clear structure and well-defined responsibilities. The collaborative work of each unit can efficiently execute the entire process of the aforementioned method, ensuring the feasibility of the method in engineering implementation, facilitating subsequent maintenance, upgrades, and expansion, and enhancing the practical application value and implementation capability of the method.

[0073] Reference Figure 3 Thirdly, the present invention provides a computer device, the computer device including a memory and a processor, the memory storing a computer program, and the processor being configured to execute the computer program to implement the above-described artificial intelligence-based coastal area coverage mapping optimization method.

[0074] It should be noted that the computer device includes a memory and a processor. The memory stores the computer program that implements the aforementioned AI-based coastal area coverage mapping optimization method. This program includes the execution logic for each stage, such as data preprocessing, feature fusion, classification and detection, and optimization adjustment. The processor is configured to retrieve the computer program from the memory and execute it. By calling hardware resources, it completes a series of operations, including data reading, computation, and result output, transforming the steps of the aforementioned method into a computer-executable instruction sequence. Ultimately, it outputs the coastal area coverage classification results and target detection results, completing the mapping optimization process. The processor executes the computer program stored in the memory, transforming the logical flow of the method into actual computational operations. This provides hardware support for the engineering application of the method, enabling high-precision mapping methods to move beyond the theoretical level and be implemented in practical coastal monitoring and management scenarios.

[0075] In some embodiments, the AI-based coastal area coverage mapping optimization method described above can be implemented using a computer device, which includes at least one processor, a communication bus, a memory, and at least one communication interface.

[0076] A processor can be a general-purpose central processing unit (CPU) or an application-specific integrated circuit (ASIC).

[0077] A communication bus can be used to transmit information between the aforementioned components.

[0078] The memory can be read-only memory (ROM) or other types of static storage devices capable of storing static information and instructions, random access memory (RAM) or other types of dynamic storage devices capable of storing information and instructions, or electrically erasable programmable read-only memory (EEPROM), compact disc read-only memory (CD-ROM) or other optical disc storage, optical disc storage (including compressed optical discs, laser discs, optical discs, universal optical discs, Blu-ray discs, etc.), magnetic disks or other magnetic storage devices, or any other medium capable of carrying or storing desired program code in the form of instructions or data structures and accessible by a computer, but not limited to these. The memory can exist independently and be connected to the processor via a communication bus. The memory can also be integrated with the processor.

[0079] The memory stores program code for executing the present invention, and its execution is controlled by a processor. The processor executes the program code stored in the memory. The program code may include one or more software modules. In the above embodiments, the artificial intelligence-based coastal area coverage mapping optimization method can be implemented by a processor and one or more software modules in the program code in the memory.

[0080] A communication interface is a device that uses any transceiver or similar device to communicate with other devices or communication networks, such as Ethernet, radio access network (RAN), wireless local area networks (WLAN), etc.

[0081] In a specific implementation, as one example, a computer device may include multiple processors, each of which may be a single-core (single-CPU) processor or a multi-core (multi-CPU) processor. Here, a processor may refer to one or more devices, circuits, and / or processing cores used to process data (e.g., computer program instructions).

[0082] The aforementioned computer device can be a general-purpose computer device or a special-purpose computer device. In specific implementations, the computer device can be a desktop computer, a portable computer, a network server, a handheld digital assistant (PDA), a mobile phone, a tablet computer, a wireless terminal device, a communication device, or an embedded device. This embodiment of the invention does not limit the type of computer device.

[0083] Fourthly, the present invention provides a computer-readable storage medium storing a computer program that, when executed by a processor, implements the above-described artificial intelligence-based coastal area coverage mapping optimization method.

[0084] It should be noted that the computer-readable storage medium takes the form of a carrier capable of storing computer programs, including USB flash drives, external hard drives, ROM, RAM, magnetic disks, optical disks, etc. This storage medium stores a computer program that implements an artificial intelligence-based coastal area coverage mapping optimization method. When this computer program is read and executed by a processor, it drives the processor to sequentially complete data preprocessing, feature fusion, classification detection, and optimization adjustments according to the steps described above, ultimately outputting the coastal area coverage classification results and target detection results, thus realizing the mapping optimization function. The computer program stored in the storage medium can be called and executed by any compatible processor, breaking the limitations of hardware devices and enabling the method to be flexibly applied to different computer devices and systems. This lowers the threshold for the promotion and application of the method, further enhancing its practicality and widespread applicability.

[0085] To provide a clearer understanding of the invention, the invention is further described below:

[0086] Reference Figure 1 In a first aspect, the present invention provides an artificial intelligence-based method for optimizing coastal area coverage mapping, the method comprising the following steps:

[0087] Multi-source remote sensing data acquisition and preprocessing

[0088] Acquire multi-source remote sensing data for coastal areas. This multi-source remote sensing data includes at least two of the following: hyperspectral data (such as the KSC dataset captured by the AVIRIS imaging spectrometer), optical aerial data (such as the MASATI ship detection dataset), and multi-temporal satellite data (such as the Landsat series data). The hyperspectral data has a spectral coverage range of 400-2500 nm and contains 145-176 effective characteristic bands. The optical aerial data has an image resolution of at least 512×512 pixels and can cover multiple targets including ocean, land, and ships.

[0089] Preprocessing of the multi-source remote sensing data specifically includes:

[0090] Noise reduction processing: An adaptive filtering algorithm is used to suppress noise for each type of remote sensing data, eliminating low signal-to-noise ratio bands affected by atmospheric scattering and equipment noise, as well as invalid interference bands such as water absorption;

[0091] Standardization: Eliminate systematic errors caused by different acquisition devices through detector calibration (such as the calibration standards of UT Space Research Center) to ensure that the data are in a uniform dimension;

[0092] Dimensionality reduction: Principal component analysis (PCA) or feature selection algorithms are used to reduce redundancy in high-dimensional data, retain key feature bands, and balance data processing efficiency with feature integrity.

[0093] Spatiotemporal registration: Aligning different source data with coordinate system 1 and time nodes to ensure spatiotemporal consistency of the data and obtain standardized processed data.

[0094] Layered fusion feature extraction

[0095] The standardized processed data is input into a preset artificial intelligence model, which is an STU-Network model based on Swing Transformer. Its feature extraction module is constructed based on the shift window attention mechanism and specifically includes a block embedding unit, a Swing Transformer concatenation module, and a block merging unit.

[0096] The process of fusing spectral and spatial features is as follows:

[0097] The block embedding unit divides the standardized data into non-overlapping blocks of equal size, and converts each block into a feature vector of a preset dimension through linear embedding, thereby achieving uniform adaptation of data dimensions.

[0098] The Swin Transformer concatenation module consists of two consecutive sub-modules. The first sub-module uses a window-based multi-head self-attention mechanism (W-MSA) to limit the computational scope to a single window, thereby reducing computational complexity while extracting local features. The second sub-module uses a shift-window-based multi-head self-attention mechanism (SW-MSA) to achieve cross-window information interaction through a sliding window and capture global features. The two sub-modules work together to achieve complementary fusion of local and global features.

[0099] The block merging unit downsamples the feature map, reducing the image height and width to half of the original size, while expanding the feature dimension to twice the original size, enhancing the feature hierarchy, and finally outputting hierarchical fused features.

[0100] Classification, Detection and Mapping Optimization

[0101] The hierarchical fusion features are processed by the classification and detection module of the preset artificial intelligence model:

[0102] The classification submodule uses a fully connected layer combined with a softmax activation function to classify the hierarchical fusion features and output classification results for at least six types of coastal cover, including mangroves, salt marshes, seagrass, rivers, land, and ocean.

[0103] The detection submodule is based on the sliding window detection algorithm to locate and identify targets such as ships, and outputs the target's position coordinates, quantity and category information;

[0104] To address the impact of tidal dynamics on coastal areas, the mangrove inundation index was introduced as a tidal characterization parameter. A tidal adaptation model was constructed by combining multi-temporal remote sensing data and embedded into the classification and detection module to correct spectral distortions caused by tidal changes and improve the accuracy of intertidal cover classification.

[0105] The final output includes coastal area coverage classification results and target detection results, achieving optimization of coastal area coverage mapping. The overall classification accuracy of this method reaches 92.4%, the coastal area coverage type recognition rate is no less than 91%, and the ship target detection accuracy is no less than 90.8%.

[0106] Reference Figure 2 Secondly, the present invention provides an artificial intelligence-based coastal area coverage mapping optimization device, the device comprising:

[0107] Data preprocessing unit: This unit acquires multi-source remote sensing data from coastal areas and performs noise reduction, standardization, dimensionality reduction, and spatiotemporal registration to obtain standardized data. This unit is adaptable to different types of remote sensing data input and ensures consistent data quality through adaptive filtering algorithms and calibration standards.

[0108] Feature fusion unit: This unit inputs standardized data into a pre-defined artificial intelligence model and, through a feature extraction module based on a shift window attention mechanism (block embedding unit, Swing Transformer concatenation module, and block merging unit), fuses spectral and spatial features to obtain hierarchically fused features. This unit achieves a balance between efficiency and completeness in feature extraction through the synergistic effect of W-MSA and SW-MSA.

[0109] The classification and detection unit processes the hierarchical fusion features through the classification and detection module, outputting coastal area coverage classification results and target detection results. The classification submodule supports accurate division of multiple coverage types, while the detection submodule enables efficient identification of ship targets.

[0110] Optimization and adjustment unit: used to introduce tidal adaptation model to correct classification and detection bias. By combining mangrove inundation index with multi-temporal data, it improves the anti-interference ability of mapping results against tidal changes and ensures stable intertidal mapping accuracy.

[0111] To make the technical solution of this application clearer, the following detailed description is provided in conjunction with specific embodiments.

[0112] Example 1: Coastal Area Coverage Mapping Based on Multi-Source Remote Sensing Data

[0113] Data preparation: Acquire the KSC hyperspectral dataset (512×614 pixels, 18m spatial resolution, 400-2500nm spectral range), the Botswana hyperspectral dataset (30m spatial resolution, 242 raw bands), the MASATI ship detection dataset (512×512 pixels, 7 types of labeled targets), and the LongKou coastal cover dataset, covering typical targets such as mangroves, salt marshes, ships, and land.

[0114] Preprocessing:

[0115] Denoising was performed on the KSC dataset, and 176 effective bands were retained after removing low signal-to-noise ratio bands.

[0116] Detector calibration was performed on the Botswana dataset, uncalibrated water absorption bands were removed, and 145 characteristic bands (ranges such as 10-55 and 82-97) were retained.

[0117] The PCA algorithm was used to reduce the dimensionality of the two types of hyperspectral data, retaining the principal components with a cumulative contribution rate of 95%.

[0118] The MASATI dataset was formatted and its coordinates were registered with the hyperspectral data to obtain standardized data.

[0119] Model Training: The standardized data was divided into training and test sets in a 7:3 ratio and input into the STU-Network model for training. The block embedding dimension of the model was set to 128, the window size of W-MSA and SW-MSA was set to 7, the block merging layer used a fully connected layer to compress the feature dimension, the number of training iterations was set to 100, and the learning rate was 0.001.

[0120] Classification and Detection: After training, the test set is input into the model. The feature extraction module obtains hierarchical fused features, and the classification and detection module outputs classification results and ship detection results. The mangrove inundation index is introduced to correct for tidal effects. Finally, the overall classification accuracy (OA) of the test set reaches 92.4%, the mangrove classification accuracy is 91.5%, and the ship detection accuracy is 90.8%.

[0121] Results validation: The model was validated using metrics such as OA, AA, Kappa coefficient, and IOU. The Kappa coefficient of the model reached 0.908, which is significantly better than the comparative models such as GAT (0.8796) and 3D-CNN (0.7919), proving the effectiveness of the method.

[0122] Example 2: Optimization of Intertidal Coastal Cover Mapping

[0123] For the Pearl River Estuary coastal area with significant tidal variations, supplementary multi-temporal Landsat satellite data (covering three time points: high tide, low tide, and slack tide) was used to construct a tidal adaptation model:

[0124] Based on multi-temporal data, the tidal variation pattern was extracted, and the spectral response curves of mangroves under different tidal conditions were calculated.

[0125] The mangrove inundation index (SMRI) was introduced to quantify the impact of tidal inundation on spectral characteristics.

[0126] The tidal adaptation model is embedded into the classification and detection module to dynamically correct the classification results under different tidal conditions;

[0127] The measured results show that the standard deviation of intertidal cover classification decreased from 8.3% in the traditional method to 3.2%, and the accuracy and stability were significantly improved.

[0128] The technical solutions provided by the embodiments disclosed in this invention have the following beneficial effects:

[0129] This invention achieves deep fusion of spectral and spatial features of multi-source remote sensing data by constructing a feature extraction module based on a shift window attention mechanism. This solves the technical bottleneck of separating the two types of features in traditional methods, and the overall classification accuracy reaches 92.4%, which is significantly better than existing KNN (OA 67.84%-77.40%), SVM (OA 73.73%-78.05%) and conventional CNN models (OA 71.43%-77.38%).

[0130] By adopting a multi-source data preprocessing workflow (noise reduction, standardization, dimensionality reduction, and spatiotemporal registration), the adaptability of heterogeneous data is improved, and it is compatible with various data types such as hyperspectral, optical aerospace, and multi-temporal satellite, thus expanding the applicable scenarios of the method.

[0131] By introducing a tidal adaptation model and a mangrove inundation index, the problem of spectral distortion caused by tidal changes was specifically solved, improving the accuracy of intertidal cover classification to over 91% and enhancing the method's anti-interference ability.

[0132] It achieves synergistic optimization of coastal coverage classification and ship target detection, with a ship detection accuracy of 90.8%, avoiding the inefficiency caused by independent tasks in traditional methods, and providing technical support for integrated monitoring in coastal areas;

[0133] The pre-defined artificial intelligence model combines W-MSA and SW-MSA to reduce computational complexity while ensuring global information extraction capabilities. Compared to the traditional Transformer model, it improves training efficiency by more than 30% and is suitable for large-scale coastal surveying tasks.

[0134] The technical solutions provided by the embodiments of the present invention have been described in detail above. Specific examples have been used to illustrate the principles and implementation methods of the embodiments of the present invention. The descriptions of the embodiments above are only for helping to understand the principles of the embodiments of the present invention. At the same time, for those skilled in the art, there will be changes in the specific implementation methods and application scope based on the embodiments of the present invention. Therefore, the content of this specification should not be construed as a limitation of the present invention.

Claims

1. A method for optimizing coastal area coverage mapping based on artificial intelligence, characterized in that, Includes the following steps: Acquire multi-source remote sensing data of coastal areas, and preprocess the multi-source remote sensing data to obtain standardized processed data; The standardized data is input into a preset artificial intelligence model, and the feature extraction module of the preset artificial intelligence model fuses spectral features and spatial features to obtain hierarchical fused features. The feature extraction module is constructed based on the shift window attention mechanism. The classification and detection module of the preset artificial intelligence model processes the hierarchical fusion features and outputs the coastal area coverage classification results and target detection results, thereby realizing the optimization of coastal area coverage mapping.

2. The method for optimizing coastal area coverage mapping based on artificial intelligence according to claim 1, characterized in that, The multi-source remote sensing data includes at least two of the following: hyperspectral data, optical airborne data, and multi-temporal satellite data. The spectral coverage range of the hyperspectral data is 400-2500nm, and the image resolution of the optical airborne data is not less than 512×512 pixels.

3. The method for optimizing coastal area coverage mapping based on artificial intelligence according to claim 1, characterized in that, Preprocessing the multi-source remote sensing data includes: Denoising was performed on each type of remote sensing data to remove low signal-to-noise ratio bands and invalid interference bands. Standardization is achieved through detector calibration to eliminate systematic errors introduced by data acquisition equipment. Dimension reduction techniques are used to reduce redundancy in high-dimensional data while preserving key feature bands. The processed data from different sources are formatted and spatiotemporally registered to obtain the standardized processed data.

4. The method for optimizing coastal area coverage mapping based on artificial intelligence according to claim 1, characterized in that, The feature extraction module includes a block embedding unit, a Swing Transformer concatenation module, and a block merging unit connected in sequence. The Swing Transformer concatenation module contains two consecutive sub-modules, which respectively adopt a window-based multi-head self-attention mechanism and a shift-window-based multi-head self-attention mechanism.

5. The method for optimizing coastal area coverage mapping based on artificial intelligence according to claim 4, characterized in that, The process of fusing spectral and spatial features includes: The standardized data is divided into non-overlapping blocks and converted into a vector of preset dimensions by using block embedding units. Local and global features are extracted through the Swin Transformer concatenation module, computational complexity is reduced through a multi-head self-attention mechanism, and cross-window information interaction is achieved through a multi-head self-attention mechanism. By downsampling the feature map through block merging units, the feature hierarchy is enhanced, and deep fusion of spectral and spatial features is achieved.

6. The method for optimizing coastal area coverage mapping based on artificial intelligence according to claim 1, characterized in that, The method further includes: To address the impact of tidal dynamics on coastal areas, the mangrove inundation index was introduced as a tidal characterization parameter, and a tidal adaptation model was constructed by combining multi-temporal remote sensing data. The tidal adaptation model is embedded into the classification and detection module to correct spectral feature distortion caused by tidal changes and improve the classification accuracy of intertidal zone cover.

7. The method for optimizing coastal area coverage mapping based on artificial intelligence according to claim 1, characterized in that, The coverage classification results output by the classification and detection module include one or more types of coastal coverage such as mangroves, salt marshes, seagrass, rivers, land, and ocean. The target detection results include the location, quantity, and category information of ship targets. The accuracy of the ship target detection is not less than 90.8%.

8. A coastal area coverage mapping optimization device based on artificial intelligence, characterized in that, include: The data preprocessing unit is used to acquire multi-source remote sensing data of coastal areas and perform noise reduction, standardization, dimensionality reduction and spatiotemporal registration processing on the multi-source remote sensing data to obtain standardized data. The feature fusion unit is used to input the standardized processed data into a preset artificial intelligence model, and through the feature extraction module based on the shift window attention mechanism, fuse spectral features and spatial features to obtain hierarchical fused features. The classification and detection unit is used to process the hierarchical fusion features through the classification and detection module, and output the coastal area coverage classification result and target detection result; An optimization and adjustment unit is used to introduce a tidal adaptation model to correct classification and detection biases, thereby improving the anti-interference capability and accuracy of the mapping results.

9. A computer device, characterized in that, The system includes a memory and a processor, the memory storing a computer program and the processor being configured to execute the computer program to implement the artificial intelligence-based coastal area coverage mapping optimization method according to any one of claims 1-7.

10. A computer-readable storage medium, characterized in that, The computer-readable storage medium stores a computer program that, when executed by a processor, implements the artificial intelligence-based coastal area coverage mapping optimization method according to any one of claims 1-7.

Images