Coral reef mask automatic extraction method and system based on artificial intelligence

Abstract

The invention relates to the technical field of image recognition, and discloses a coral reef mask automatic extraction method and system based on artificial intelligence. The method comprises the following steps: collecting a multi-source coral reef remote sensing image, and generating a standardized data set through environment adaptive preprocessing; constructing a multi-scale feature pyramid encoder to extract five layers of feature information; an environment perception module is established to calculate an environment complexity vector and re-calibrate a feature weight to form adaptive feature representation; setting a multi-level fusion decoder to execute progressive up-sampling and cross-scale fusion, and generating a probability distribution diagram; and outputting a coral reef mask result by applying an adaptive threshold segmentation algorithm in combination with morphological post-processing. The coral reef mask extraction method and device solve the problems that an existing coral reef mask extraction technology is insufficient in multi-scale feature fusion and insufficient in environment adaptive capacity in a complex marine environment, and improve the coral reef mask extraction precision and environment adaptability.

Description

Technical Field

[0001] The present application relates to the field of image recognition technology, and in particular to an artificial intelligence-based method and system for automatically extracting coral reef masks. Background Art

[0002] As a vital component of marine ecosystems, monitoring the spatial distribution of coral reefs is crucial for marine biodiversity conservation and ecological assessment. Traditional coral reef monitoring relies primarily on diving surveys, satellite remote sensing, and drone aerial photography. With the development of artificial intelligence (AI), deep learning-based automatic coral reef identification technology is gaining popularity. Existing AI-based coral reef mask extraction techniques primarily employ convolutional neural networks for end-to-end image segmentation, using training data to automatically identify coral reef areas and generate masks.

[0003] However, existing technologies have significant shortcomings in complex marine environments: first, traditional CNN architectures lack the ability to effectively integrate coral reef features at different scales, and cannot fully capture multi-level feature information from global distribution to pixel-level details; second, existing methods lack environmental adaptation mechanisms, and when faced with complex underwater environments such as weak light, turbid water, and biological occlusion, they are unable to dynamically adjust feature extraction strategies, resulting in a significant decrease in segmentation accuracy; in addition, traditional decoder structures lack effective utilization of cross-scale information in the feature fusion process, making it difficult to accurately reconstruct coral reef boundary details.

[0004] Based on a step-by-step analysis of the aforementioned technical deficiencies, the fundamental problem with existing technologies lies in the lack of a complete set of multi-scale feature adaptive extraction and fusion mechanisms. Specifically, the image preprocessing stage lacks an adaptive processing strategy for marine imagery of varying quality, resulting in uneven input quality for subsequent feature extraction. The feature encoding stage lacks a multi-scale pyramid structure for deep feature representation, making it impossible to fully capture the multi-layered spatial characteristics of coral reefs. The feature weight allocation stage lacks an adaptive adjustment mechanism based on environmental complexity, making it impossible to dynamically optimize feature weights based on the specific marine environment. The decoding and fusion stage lacks a progressive cross-scale feature fusion strategy, making it difficult to accurately reconstruct coral reef mask boundaries. The mask generation stage lacks adaptive threshold selection and intelligent post-processing mechanisms, affecting the accuracy and completeness of the final mask. Summary of the Invention

[0005] This application provides an artificial intelligence-based automatic coral reef mask extraction method and system, which is used to solve the problems of insufficient multi-scale feature fusion and insufficient environmental adaptability of existing coral reef mask extraction technology in complex marine environments, and improve the accuracy and environmental adaptability of coral reef mask extraction.

[0006] In a first aspect, the present application provides an artificial intelligence-based automatic coral reef mask extraction method, the artificial intelligence-based automatic coral reef mask extraction method comprising: Step S101: Collect multi-source coral reef remote sensing images, analyze image quality using an environment-adaptive preprocessing algorithm, and perform layered preprocessing to generate a standardized coral reef image dataset; Step S102: construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid; Step S103: Establish an environmental perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector, and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation; Step S104: setting a multi-level fusion decoder to receive the adaptive coral reef feature representation, performing progressive upsampling and cross-scale feature fusion, and generating a coral reef probability distribution map; Step S105 : applying an adaptive threshold segmentation algorithm to process the coral reef probability distribution map, combining morphological post-processing to eliminate noise and smooth boundaries, and outputting a coral reef mask result.

[0007] In a second aspect, the present application provides an artificial intelligence-based automatic coral reef mask extraction system, the artificial intelligence-based automatic coral reef mask extraction system comprising: The acquisition module is used to collect multi-source coral reef remote sensing images, analyze image quality through an environment-adaptive preprocessing algorithm, and perform layered preprocessing to generate a standardized coral reef image dataset; An input module is used to construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid; An analysis module is used to establish an environmental perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation; a fusion module, configured to set a multi-level fusion decoder to receive the adaptive coral reef feature representation, perform progressive upsampling and cross-scale feature fusion, and generate a coral reef probability distribution map; The output module is used to apply an adaptive threshold segmentation algorithm to process the coral reef probability distribution map, combine morphological post-processing to eliminate noise and smooth boundaries, and output a coral reef mask result.

[0008] In a third aspect, an artificial intelligence-based automatic extraction device for coral reef masks is provided, comprising: a memory and at least one processor, wherein the memory stores instructions; the at least one processor calls the instructions in the memory to enable the artificial intelligence-based automatic extraction device for coral reef masks to execute the above-mentioned artificial intelligence-based automatic extraction method for coral reef masks.

[0009] In a fourth aspect, a computer-readable storage medium is provided, wherein instructions are stored in the computer-readable storage medium, which, when executed on a computer, enables the computer to execute the above-mentioned artificial intelligence-based automatic extraction method for coral reef masks.

[0010] The technical solution provided in this application addresses the inability of traditional preprocessing methods to cope with the varying quality of complex marine environments by constructing an environmentally adaptive preprocessing algorithm for multi-source coral reef remote sensing imagery, ensuring consistent data quality during subsequent processing. The design of a multi-scale feature pyramid encoder overcomes the limitations of traditional CNN architectures. Through a five-layer feature pyramid structure, it achieves complete feature coverage from global semantics to pixel-level details, significantly enhancing the expressiveness of coral reef features. The introduction of an environmental perception module and an environmental complexity vector enables the system to dynamically adjust feature weights based on environmental factors such as lighting, water quality, and occlusion. The resulting adaptive coral reef feature representation maintains stable recognition performance across a variety of complex marine environments. A multi-level fusion decoder effectively integrates feature information at different levels through progressive upsampling and cross-scale feature fusion mechanisms, generating coral reef probability distribution maps with higher spatial accuracy and boundary clarity. An adaptive threshold segmentation algorithm combined with an intelligent mask generation strategy based on morphological post-processing not only eliminates the limitations of traditional fixed threshold methods but also effectively removes noise and smoothes boundaries through morphological operations. The resulting coral reef mask output significantly outperforms existing technologies in terms of accuracy and completeness.

[0011] In the specific application field of coral reef monitoring and ecological protection, the core algorithm features of this application play a key role. The environmental adaptive preprocessing algorithm targets the special properties of marine remote sensing images. Through quality assessment and hierarchical processing strategies, it ensures that image data of different sources and qualities can meet unified processing standards, laying a solid foundation for subsequent intelligent analysis. The multi-scale feature pyramid encoder is particularly suitable for the identification of marine organisms with complex spatial structures such as coral reefs. Its hierarchical feature extraction mechanism can simultaneously capture the overall distribution pattern and local texture details of coral reefs, which is crucial for accurately distinguishing coral reefs from other seabed features. The environmental complexity vector calculation and adaptive weight recalibration mechanism are specifically designed for the complexity and variability of the underwater environment, enabling the system to automatically adjust the recognition strategy under different marine environmental conditions. This adaptive capability has important practical value for large-scale, long-term coral reef monitoring projects, ensuring the consistency and reliability of the monitoring results. BRIEF DESCRIPTION OF THE DRAWINGS

[0012] In order to more clearly illustrate the technical solutions of the embodiments of the present invention, the following briefly introduces the drawings required for use in the description of the embodiments. Obviously, the drawings described below are some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without creative work.

[0013] Figure 1 Schematic diagram of an embodiment of the method for automatic extraction of coral reef masks based on artificial intelligence in an embodiment of the present application; Figure 2 This is a schematic diagram of an embodiment of the processing flow of the method for automatically extracting coral reef masks based on artificial intelligence in an embodiment of the present application; Figure 3 This is a schematic diagram of an embodiment of an artificial intelligence-based automatic coral reef mask extraction system in an embodiment of the present application; Figure 4 It is a schematic block diagram of the structure of an automatic coral reef mask extraction device based on artificial intelligence in an embodiment of the present invention. DETAILED DESCRIPTION

[0014] The embodiments of the present application provide a method and system for automatic extraction of coral reef masks based on artificial intelligence. The terms "first", "second", "third", "fourth", etc. (if any) in the specification and claims of this application and the above-mentioned drawings are used to distinguish similar objects and are not necessarily used to describe a specific order or sequence. It should be understood that the numbers used in this way are interchangeable where appropriate so that the embodiments described herein can be implemented in an order other than that illustrated or described herein. In addition, the terms "including" or "having" and any variations thereof are intended to cover non-exclusive inclusions. For example, a process, method, system, product or device that includes a series of steps or units is not necessarily limited to those steps or units explicitly listed, but may include other steps or units that are not explicitly listed or inherent to these processes, methods, products or devices.

[0015] This application relates to the field of image recognition technology and discloses an artificial intelligence-based method and system for automatic coral reef mask extraction. Multi-source coral reef remote sensing images are collected and a standardized data set is generated through environmental adaptive preprocessing. A multi-scale feature pyramid encoder is constructed to extract five layers of feature information. An environmental perception module is established to calculate the environmental complexity vector and recalibrate the feature weights to form an adaptive feature representation. A multi-level fusion decoder is set up to perform progressive upsampling and cross-scale fusion to generate a probability distribution map. An adaptive threshold segmentation algorithm is applied in combination with morphological post-processing to output the coral reef mask result. This method solves the problems of insufficient multi-scale feature fusion and insufficient environmental adaptability of existing coral reef mask extraction technologies in complex marine environments, thereby improving the accuracy and environmental adaptability of coral reef mask extraction.

[0016] For ease of understanding, the specific process of the embodiment of the present application is described below. Figure 1 In the embodiments of the present application, an embodiment of the method for automatically extracting coral reef masks based on artificial intelligence includes: Step S101: Collect multi-source coral reef remote sensing images, analyze image quality using an environment-adaptive preprocessing algorithm, and perform layered preprocessing to generate a standardized coral reef image dataset; Step S102: construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid; Step S103: Establish an environmental perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector, and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation; Step S104: setting a multi-level fusion decoder to receive the adaptive coral reef feature representation, performing progressive upsampling and cross-scale feature fusion, and generating a coral reef probability distribution map; Step S105: Apply an adaptive threshold segmentation algorithm to process the coral reef probability distribution map, combine it with morphological post-processing to eliminate noise and smooth the boundaries, and output the coral reef mask result.

[0017] It is understandable that the execution subject of this application can be an artificial intelligence-based coral reef mask automatic extraction system, or a terminal or a server, which is not limited here. The embodiment of this application is described by taking the server as the execution subject as an example.

[0018] Specifically, the project begins with multi-source data collection, integrating various data sources, including satellite imagery, drone aerial photography, and underwater photography, to ensure comprehensive data coverage. To address the unique challenges of underwater environments, such as uneven illumination, water turbidity, and biological occlusion, an environmentally adaptive preprocessing algorithm is employed to assess the quality of input images and perform stratified processing. The quality assessment model analyzes image brightness distribution, color shift, and water turbidity, classifying the images into three levels: high, medium, and low. Appropriate preprocessing operations, including size normalization, color space conversion, contrast enhancement, and multi-scale illumination compensation, are then performed on each level to generate a standardized coral reef image dataset.

[0019] During the feature extraction phase, the system constructed a multi-scale feature pyramid encoder based on a modified ResNet50 architecture, extending the network depth by adding fine-grained feature layers. The encoder employs a bidirectional pyramid structure, with a bottom-up path extracting features at each stage and a top-down path transmitting high-level semantic information. Lateral connections are used to achieve feature fusion. The introduction of depthwise separable convolutions significantly reduces computational complexity while maintaining feature extraction accuracy. The encoder ultimately outputs a five-layer feature pyramid, corresponding to coral reef feature information at different scales, forming a complete feature representation system, from global semantic features to pixel-level details.

[0020] The environmental perception module consists of three parallel branch networks, responsible for assessing lighting conditions, water quality, and occlusion levels, respectively. Each branch network analyzes feature maps at different levels, calculating metrics such as illumination uniformity, water clarity, and occlusion complexity. These are then integrated into an environmental complexity vector via a fully connected layer. Based on this vector, the system employs a channel-wise attention mechanism to dynamically adjust the weights of features at each layer. This mechanism prioritizes high-resolution features in bright conditions, prioritizes low-resolution global features in low-light or turbid environments, and balances the weights of features at all scales in high-occlusion scenarios, forming an adaptive feature representation that adapts to environmental changes.

[0021] The multi-level fusion decoder employs a symmetrical upsampling structure and gradually restores spatial resolution through a progressive feature fusion strategy. Each decoding block incorporates a multi-scale receptive field module, employing convolution kernels of varying sizes to process feature information in parallel. This module, combined with a boundary enhancement module, optimizes edge detection. The decoding process organically combines high-level semantic information with low-level detail features, ultimately generating a coral reef probability distribution map that accurately reflects the likelihood that each pixel belongs to a coral reef area.

[0022] During the result generation phase, the system applies an improved adaptive threshold segmentation algorithm, adding connected region constraints to the traditional inter-class variance maximization to ensure the continuity of the segmentation results. Morphological post-processing eliminates noise points and fills holes through a sequence of opening and closing operations, while boundary smoothing effectively improves the quality of the mask edges. The final coral reef mask output is verified by a quality assessment module, providing reliable data support for coral reef ecological monitoring and protection. The entire processing flow is comprehensively optimized for the specific characteristics of underwater environments, effectively addressing the shortcomings of traditional methods in feature fusion and environmental adaptability, and improving the accuracy and reliability of coral reef monitoring.

[0023] In a specific embodiment, the process of executing step S101 may specifically include the following steps: The image quality assessment model is used to detect weak light areas, water turbidity, and color shift in multi-source coral reef remote sensing images, and an image quality assessment vector is obtained. A hierarchical preprocessing strategy is implemented on coral reef remote sensing images based on image quality assessment vectors. Size normalization and color space conversion are performed on high-quality images. Histogram equalization and contrast enhancement are added to medium-quality images. A multi-scale Retinex algorithm is used for illumination compensation and detail enhancement on low-quality images to obtain quality-optimized images. Perform random rotation, horizontal and vertical flipping, random cropping and elastic deformation on the quality optimized image to obtain augmented image data; Based on the amplified image data, the resolution of 512×512 pixels was adjusted and the pixel values ​​were normalized to the interval [0,1] to obtain a standardized coral reef image dataset.

[0024] Specifically, the acquisition and processing of remote sensing images of coral reefs face technical challenges posed by the complex and ever-changing marine environment. Multi-source data exhibit significant differences in lighting conditions, water transparency, and color fidelity. The image quality assessment model utilizes a three-channel parallel architecture to process the input image. The low-light detection branch analyzes the brightness component of the RGB channels and quantifies the degree of illumination unevenness by calculating the average pixel value and variance of the local area. Higher values ​​indicate worse lighting conditions. The water turbidity detection branch operates in the HSV color space, focusing on analyzing the histogram distribution of the saturation channel. Combined with high-frequency component energy calculations, turbid waters typically exhibit high-frequency energy attenuation. The color shift detection branch converts the image to the LAB color space and calculates the mean shift of the a and b channels. Color distortion is determined when the shift exceeds a threshold. The outputs of the three branches are normalized and concatenated into a three-dimensional quality assessment vector. The values ​​of each dimension of the vector range from 0 to 1, corresponding to the quantitative scores of illumination, turbidity, and color shift, respectively.

[0025] A dynamic processing path was established based on a hierarchical preprocessing strategy for quality assessment vectors. Images with scores below 0.3 in all dimensions of the vector were considered high-quality. The processing pipeline included bilinear interpolation to resize to the target resolution and conversion from RGB to LAB color space, maintaining the linear relationship between pixel values. For moderate-quality images with scores between 0.3 and 0.6, contrast-constrained adaptive histogram equalization was added to the basic processing, with a tile size of 16×16 pixels and a clipping factor of 0.03 to avoid excessive noise enhancement. Low-quality images (any score exceeding 0.6) were subjected to an improved multi-scale Retinex algorithm with three Gaussian kernel sizes (15, 80, and 200 pixels). Illumination estimation and compensation were performed in the luminance channel, with the compensation strength positively correlated with the quality score. Chroma channel compensation was enabled for turbidity scores above 0.7. The processed images were validated using the SSIM metric to ensure a minimum 15% improvement in structural similarity.

[0026] During the data augmentation phase, a mapping relationship between the original image and the derived samples was established. Random rotation was performed to uniformly sample the angles within the range of 0-360 degrees, and edge reflection filling was used to maintain the integrity of the coral reef area. The probability of horizontal and vertical flipping was set to 0.5 to generate mirrored samples to increase perspective diversity. Random cropping was performed to extract a 512×512 pixel area from the original image while ensuring the integrity of the main coral reef. The overlap rate was controlled within 30%. Elastic deformation applied a B-spline-based grid deformation algorithm with a grid spacing of 64 pixels and a maximum deformation displacement of no more than 20% of the grid spacing to avoid excessive geometric distortion. The number of samples after augmentation was expanded to eight times that of the original data, and the transformation parameters of each sample were recorded for subsequent analysis.

[0027] During the standardization phase, resolution uniformity and numerical normalization are performed. A bicubic interpolation algorithm resizes enhanced images of varying sizes to a standard 512×512 resolution. Interpolation weights are calculated based on the Lanczos kernel function, preserving high-frequency details while suppressing ringing artifacts. Pixel value normalization utilizes global statistics, linearly mapping each channel's values ​​to the [0, 1] interval based on the maximum and minimum pixel values ​​of the entire dataset. A metadata recording system is established during processing to store each image's quality assessment vector, processing parameters, and enhancement method, forming a complete preprocessing traceability chain. The standardized image datasets are categorized and stored by quality level, maintaining a balanced sample distribution and providing high-quality input for subsequent model training.

[0028] Step S102: construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid.

[0029] In a specific embodiment, the process of executing step S102 may specifically include the following steps: A multi-scale feature pyramid encoder is constructed using an improved ResNet backbone network. The network depth is expanded by adding fine-grained feature layers to obtain a five-layer feature pyramid structure. The standardized coral reef image dataset is input into a five-layer feature pyramid structure to extract global semantic features, regional boundary features, local texture features, detail edge features, and pixel-level features, respectively, to obtain a multi-scale feature map group. Based on the multi-scale feature map group, depthwise separable convolution is used to replace traditional convolution for feature processing. Through lateral connection and top-down path fusion, high-level semantic information and low-level features are added element-wise to obtain a fused feature representation. The channel dimension of the fused feature representation is expanded so that each feature map contains the same number of channels, resulting in a five-layer coral reef feature pyramid.

[0030] Specifically, the multi-scale feature pyramid encoder is based on the improved ResNet50 backbone network for architectural expansion. By introducing cross-level feature interaction mechanism and depth-wise separable convolution optimization, a feature extraction system with a five-layer pyramid structure is constructed.

[0031] The standardized coral reef image is fed into the encoder as a three-channel matrix with a resolution of 512×512. In the first processing stage, a fine-grained feature extraction branch is added after the original ResNet50 stem module. The standard convolutional layer uses a 7×7 kernel size and downsampling with a stride of 2 to obtain the initial feature map. Simultaneously, the parallel branch employs a series of 3×3 convolutions to preserve more detailed information. The outputs of the two paths are concatenated in the channel dimension to form a 256-dimensional low-level feature representation. This dual-path design maintains the receptive field while avoiding the loss of high-frequency information caused by early downsampling, which is particularly beneficial for preserving the outline of coral reef edges.

[0032] Furthermore, a fifth feature extraction stage is added to the four standard residual blocks of ResNet50, constructing a multi-scale receptive field by modifying the dilation rate. The first four stages maintain the original network's [3, 4, 6, 3] layer configuration, reducing the output feature map size to 1 / 4, 1 / 8, 1 / 16, and 1 / 32 of the input size, respectively. The newly added fifth stage replaces ordinary convolution with dilated convolution, maintaining 1 / 32 resolution while setting the dilation factor to [1, 2, 4], enabling a single neuron to capture structural features at different scales within the coral reef region. The outputs of each stage are processed using batch normalization and the ReLU activation function to form a feature pyramid with a clear semantic hierarchy.

[0033] The fusion feature architecture utilizes a bidirectional feature pyramid network. The bottom-up path directly extracts the output of each stage as the base feature map, while the top-down path transfers high-level semantic features to low-resolution layers through 2x nearest neighbor upsampling. Horizontal connections use 1×1 convolutions to align the channel dimensions, ensuring element-wise addition of features from different levels. The introduction of depthwise separable convolutions significantly reduces computational complexity. After 3×3 spatial convolutions independently process each input channel, 1×1 point-by-point convolutions achieve cross-channel information fusion.

[0034] The channel dimension expansion module uses grouped convolution to process the fused feature maps, dividing the 256-dimensional input channels into four 64-dimensional subspaces. Each subspace undergoes an independent 3×3 convolution before being merged. This structure ensures rich feature representation while avoiding the over-parameterization problem of fully connected convolution. The expanded five-layer feature maps are uniformly resized to 256 channels, corresponding to resolutions of 128×128, 64×64, 32×32, 16×16, and 8×8, respectively, forming a complete coral reef feature pyramid.

[0035] Step S103: Establish an environmental perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation.

[0036] In a specific embodiment, the process of executing step S103 may specifically include the following steps: The environmental perception module consists of three parallel branch networks: illumination assessment branch, water quality assessment branch, and occlusion assessment branch. The five-layer coral reef feature pyramid is input into each branch for environmental analysis, and the illumination uniformity index, water quality clarity index, and occlusion complexity index are obtained. Based on the illumination uniformity index, water quality clarity index and occlusion complexity index, a fusion calculation is performed through the fully connected layer to obtain the environment complexity vector; According to the environmental complexity vector, the channel attention mechanism is used to distribute weights on the five-layer coral reef feature pyramid. The weights of high-resolution features are enhanced in well-lit environments, the weights of low-resolution global features are enhanced in low-light or turbid environments, and the weights of features at all scales are balanced in highly occluded environments. The weight distribution results are obtained. The weight distribution result is element-wise multiplied with the five-layer coral reef feature pyramid to obtain the adaptive coral reef feature representation.

[0037] Specifically, in coral reef remote sensing image analysis, the design of the environmental perception module is optimized for complex underwater environmental conditions. Five layers of feature pyramids are used as input, with resolutions of 128×128, 64×64, 32×32, 16×16, and 8×8 pixels, respectively, and each layer contains 256 feature channels. The illumination assessment branch uses 128×128 high-resolution features as input, which retain sufficient spatial detail. The branch structure consists of a global average pooling layer and two fully connected layers. The 256-dimensional vector output by the pooling layer is compressed into 3D features through a fully connected network, corresponding to the image brightness mean, variance, and contrast metrics. These statistics are linearly combined to generate an illumination uniformity score in the range of 0–1, with the variance term assigned a weight of 0.6 to ensure sensitive detection of areas with uneven illumination.

[0038] The water quality assessment branch processes 64×64 mid-level features, which retain sufficient spatial detail while also allowing for a certain degree of semantic abstraction. This branch employs a channel-wise attention mechanism, calculating the importance weight of each channel through 1×1 convolutions to prioritize the contribution of high-frequency texture features. The weighted feature maps undergo spatial pyramid pooling, calculating energy distribution at different grid scales. Higher output values ​​indicate more severe turbidity caused by water scattering and suspended matter.

[0039] The occlusion assessment branch analyzes the anisotropic properties of the 32×32 feature map. Coral reefs exhibit a unique texture orientation distribution under biological occlusion. This branch calculates the feature map gradient using gradient operators in three directions (0°, 45°, and 90°) and calculates the distribution entropy of the gradient magnitude in each direction. High entropy values ​​indicate a chaotic texture orientation, suggesting occlusion such as algae cover or schooling fish.

[0040] Furthermore, the outputs of the three branches are interactively calculated at the fusion layer. Light and water quality indicators are coupled through a gating mechanism, designed to be multiplied to reflect their synergistic effect: when light is insufficient and water is turbid, the product exponentially amplifies the severity of the environment. The occlusion coefficient is retained as an independent dimension, as its formation is relatively independent of light and water conditions. A fully connected network maps these three parameters into a three-dimensional environmental vector. The vector modulus is normalized using a sigmoid function, and the directional angle is used to determine the dominant interference type. This vector provides a compact representation of the environmental state, and its mathematical properties ensure robustness under complex environmental conditions.

[0041] The channel attention mechanism generates a five-dimensional weight distribution based on the environment vector. The design uses the vector's modulus as input to a sigmoid function to generate a basic scaling factor. The directional angle is decomposed into three axial components using trigonometric functions, controlling the relative weights of high-, medium-, and low-resolution features. The implementation consists of three fully connected layers. The first layer expands the 3D environment vector into 64-dimensional hidden features. The second layer compresses it to 5 dimensions corresponding to the pyramid level. The final layer applies a softmax to ensure weight normalization. This structure assigns a weight of more than 0.35 to the 128×128 feature in bright light (vectors pointing to the first quadrant) and increases the weight of the 16×16 feature to 0.4 in turbid conditions (vectors pointing to the second quadrant), consistent with the discernible nature of coral reefs in diverse environments.

[0042] During the weight application phase, feature recalibration is achieved using a channel-by-channel multiplication method, where each of the five layers of feature maps is multiplied by the corresponding weight coefficients, preserving the original dimensionality. This fine-grained adjustment is more precise than traditional inter-layer weight assignment. For example, in the case of partial occlusion, only the channel response of the affected spatial region is reduced, rather than the entire layer of features. During this process, the feature dimension is gradually compressed from 256 channels to a single-channel probability output, while the spatial resolution remains unchanged at 512×512, ensuring that each input pixel has a unique probability prediction value.

[0043] Step S104: Setting a multi-level fusion decoder to receive the adaptive coral reef feature representation, performing progressive upsampling and cross-scale feature fusion, and generating a coral reef probability distribution map.

[0044] In a specific embodiment, the process of executing step S104 may specifically include the following steps: Constructing a multi-level fusion decoder consists of five decoding blocks, each corresponding to a feature layer of the adaptive coral reef feature representation. The decoding process is performed through a symmetrical upsampling structure to obtain the decoder framework. The adaptive coral reef feature representation is input into the decoder framework, and a progressive feature fusion strategy is used to upsample the highest-level features through transposed convolution and perform jump connections with the corresponding encoder features. The features are fused layer by layer until the original resolution is restored, resulting in progressive fused features. Based on the progressive fusion feature, a multi-scale receptive field module is introduced in each decoding block. Convolution kernels of different sizes are used to process the fusion features in parallel. The boundary enhancement module is combined with the Sobel operator to detect edge information. The edge features are weightedly fused with the semantic features to obtain the boundary enhancement features. Final decoding output processing is performed on the boundary-enhanced features to generate a coral reef probability distribution map.

[0045] Specifically, the designed multi-level fusion decoder realizes the conversion process from adaptive feature representation to high-precision probability distribution map through progressive upsampling and cross-scale feature fusion mechanism.

[0046] First, the decoder architecture adopts a five-layer pyramid structure symmetrical to the encoder, with each decoding block corresponding to a feature layer of a specific resolution. The input adaptive coral reef feature representation consists of five sets of feature maps with resolutions increasing from 8×8 to 128×128, each with 256 channels. The decoding process starts with the lowest resolution 8×8 features, which are upsampled by a factor of 2 via transposed convolution. The kernel size is set to 4×4 to ensure sufficient neighborhood information is involved in the calculation. The upsampled features are then channel-wise concatenated with the 16×16 features passed from the encoder path via skip connections. Before concatenation, 1×1 convolution is used to unify the channel dimensions to 128 to avoid feature dilution. This progressive fusion strategy gradually recovers spatial detail information while maintaining high-level semantic integrity.

[0047] Secondly, a multi-scale receptive field module is embedded at the core of each decoding block, utilizing a three-way parallel convolutional architecture to process and fuse features. The first path uses a 3×3 standard convolution to capture local neighborhood features, suitable for coral reef texture analysis. The second path uses a 5×5 dilated convolution (with a dilation ratio of 2) to expand the receptive field to 9×9, detecting the structure of medium-sized coral communities. The third path applies a 7×7 depthwise separable convolution, covering a larger area while reducing computational effort and identifying overall patterns in coral reef distribution. After concatenating the three outputs in the channel dimension, a 1×1 convolution is used to implement cross-scale feature interaction, generating a 256-dimensional multi-scale fused feature.

[0048] Furthermore, the boundary enhancement module improves edge localization accuracy by integrating spatial gradient information. The Sobel operator, implemented as a 3×3 convolution kernel, calculates the spatial gradients of the feature map in both the horizontal and vertical directions. The gradient magnitudes are then fused using the L2 norm to form an edge intensity map. This intensity map is gated and added to the multi-scale fused features. The gating coefficients are adaptively generated from the feature map to suppress edge responses in flat areas and enhance contour information in boundary regions.

[0049] Finally, the decoding stage uses a cascade of convolutional layers to process boundary-enhanced features. The first layer uses 3×3 convolutions combined with batch normalization and the LeakyReLU activation function to transform the features. The second layer uses 1×1 convolutions to compress the number of channels to one dimension, outputting a single-channel feature response map. All decoding block outputs are uniformly upsampled to a 512×512 resolution using bilinear interpolation. The sigmoid function maps the response values ​​to probabilities in the [0, 1] range, forming a two-dimensional distribution matrix of coral reef presence probabilities. The value of each pixel in this matrix represents the probability that the location belongs to a coral reef area. After thresholding, a binary mask can be directly generated.

[0050] In a specific embodiment, the step of performing final decoding and outputting processing on the boundary enhancement features to generate a coral reef probability distribution map may specifically include the following steps: The boundary enhancement features are transferred through the residual connection structure. Each decoding block contains two convolutional layers, a batch normalization layer and an activation function to perform deep feature processing to obtain deep decoding features. Based on the deep decoding features, a fully convolutional network structure is used to perform pixel-level classification calculations. The convolution kernel is used to calculate the classification probability value of each pixel position belonging to the coral reef area to obtain the pixel classification probability; The Sigmoid activation function is performed on the pixel classification probability to map the probability value to the range of zero to one to obtain the normalized probability value; The normalized probability values ​​are reconstructed according to the pixel positions to form a complete two-dimensional probability distribution matrix and generate a coral reef probability distribution map.

[0051] Specifically, boundary-enhanced features are fed into the final decoding stage as 512×512×256 tensors, and a residual connection structure establishes a multi-level feature transfer path. 128×128 resolution features are upsampled to the target size via bilinear interpolation, 64×64 features are transposed convolutionally restored to restore spatial details, and 32×32 features preserve global context. The three-level residual features are each resized to 64 channels using 1×1 convolutions and then weighted summed with the current layer features. The weight coefficients are dynamically generated by a three-layer fully connected network to ensure adaptive feature fusion. The first convolutional layer within the decoding block processes the 256-dimensional input using a 3×3 kernel. The output features are then distributed through a normalization layer. A moving average statistic records the global mean and variance during training, while the training set statistics are fixed during inference. The LeakyReLU activation function introduces a nonlinear transformation, with a slope of 0.01 in the negative interval to prevent vanishing gradients. The second convolutional layer compresses the channel to 128 dimensions through a 1×1 kernel size to reduce the complexity of subsequent calculations. The convolution weights are initialized to He normal distribution and the bias term is initialized to zero.

[0052] The fully convolutional classifier processes 128-dimensional deep decoding features. The classification layer uses a single 3×3 convolution kernel, with kernel parameters designed to be a circular distribution with positive weights at the center and negative weights at the edges, based on coral reef morphology. A dot product response is computed at each spatial location using a sliding window. Parallel computation on a 512×512 grid generates a complete response map. Responses are mapped to probabilities using a sigmoid function. A temperature coefficient of 0.8 controls the gradient slope, and the output range is strictly constrained to the 0-1 range. The probability distribution matrix reconstruction process maintains spatial coordinate correspondence, with row numbers corresponding to image latitude and column numbers to longitude. 32-bit floating-point storage ensures numerical accuracy. Probability values ​​in coral reef areas are typically above 0.7, while those in background areas are below 0.3. The transition zone exhibits a smooth gradient, which provides a reliable foundation for subsequent adaptive threshold segmentation. During processing, the feature dimension is gradually compressed from 256 channels to a single-channel probability output, while the spatial resolution remains constant at 512×512, ensuring that each input pixel has a unique probability prediction value.

[0053] Step S105: Apply an adaptive threshold segmentation algorithm to process the coral reef probability distribution map, combine it with morphological post-processing to eliminate noise and smooth the boundaries, and output the coral reef mask result.

[0054] In a specific embodiment, the process of executing step S105 may specifically include the following steps: The histogram distribution characteristics of the coral reef probability distribution map are analyzed, and the adaptive segmentation threshold is obtained by calculating the threshold point with the maximum inter-class variance and combining it with the connectivity constraint of the coral reef area. Based on the adaptive segmentation threshold, the coral reef probability distribution map is subjected to binary segmentation processing. Pixels with probability values ​​higher than the threshold are marked as coral reef areas, and pixels with probability values ​​lower than the threshold are marked as background areas, thus obtaining an initial binary mask. The initial binary mask is processed by a sequence of morphological opening and closing operations. The structuring element is first used to perform an erosion and dilation operation to remove noise points, and then an erosion and dilation operation is performed to fill holes and breaks to obtain a morphologically optimized mask. The morphologically optimized mask is smoothed by a Gaussian filter to eliminate jagged artifacts at the boundary and calculate the comprehensive score of the mask quality to obtain the coral reef mask result.

[0055] Specifically, the probability distribution map of coral reefs is input into the threshold segmentation stage as a single-channel floating-point matrix with a resolution of 512×512. The value range of each element in the matrix is ​​[0,1], which represents the probability that the corresponding pixel belongs to the coral reef area. The adaptive threshold calculation adopts the improved Otsu algorithm, which introduces the connected region constraint condition on the basis of the traditional inter-class variance maximization. The probability histogram divides the [0,1] interval into 100 bins with an interval of 0.01, and the pixel frequency of each bin is counted to form a distribution curve. The algorithm traverses all possible threshold candidate points and for each candidate threshold t Calculate the prospect (probability ≥ t ) and background (probability < t ) and calculate the area ratio of the largest connected region in the foreground pixels. The objective function is designed as a weighted product of the inter-class variance and the proportion of the connected region. The weight coefficients are set to 0.7 and 0.3 based on the morphological characteristics of the coral reef to ensure that the segmentation results maintain the inter-class distinction and conform to the continuous distribution characteristics of the coral reef. t Through exhaustive search, the calculation process is accelerated in parallel on the GPU, and the processing time is controlled within 8ms.

[0056] During the binarization phase, the probability matrix is ​​compared pixel by pixel with a threshold t. Pixels with a probability ≥ t are set to 1 (coral reef), and all others are set to 0 (background), generating an initial binary mask. This mask uses a four-connected region analysis to remove isolated noise points smaller than 50 pixels and fill internal holes smaller than 100 pixels. Morphological processing utilizes a sequence of circular structuring elements. During the opening phase, a circular kernel with a radius of 3 pixels is first eroded to eliminate edge artifacts, followed by a dilation operation to restore the main area. During the closing phase, a circular kernel with a radius of 5 pixels is first dilated to fill small fractures, followed by an erosion operation to refine the boundary contours. Morphological operations are implemented using FPGA hardware acceleration, with processing time for a single image exceeding 5ms.

[0057] In the boundary optimization stage, the binary mask after morphological processing is fused with the original probability map. σ =1.5, performs weighted smoothing on the area within 3 pixels of the binary boundary. The weight coefficient is taken from the corresponding position value of the probability map to achieve a soft transition from the hard segmentation boundary to the probability gradient. The quality assessment module calculates three indicators of the mask: boundary tortuosity (calculated by chain code analysis), regional compactness (4 π The three metrics are linearly combined to form a quality score on a scale of 0-100, with masks scoring below 70 triggering a manual review process. The final coral reef mask is saved in GeoTIFF format, with the threshold parameters and quality score recorded as metadata.

[0058] The above describes the method for automatically extracting coral reef masks based on artificial intelligence in the embodiment of the present application. The following describes the processing flow of the method for automatically extracting coral reef masks based on artificial intelligence in the embodiment of the present application. Figure 2 In the embodiment of the present application, an embodiment of the processing flow of the method for automatically extracting coral reef masks based on artificial intelligence includes: The AI-based automatic coral reef mask extraction method comprises four core processes, forming a complete end-to-end chain. First, the data normalization process intelligently classifies the input raw coral reef image into high / medium / low quality levels. Normalization, color conversion, histogram equalization, contrast enhancement, and multi-scale Reinex processing are performed for each quality level, resulting in a unified output of 512×512 resolution Lab color space data. Second, the feature extraction process constructs a five-layer feature pyramid using an improved ResNet architecture. Using techniques such as depthwise separable convolution, dilated convolution, and the SE attention mechanism, the five-layer feature pyramid is extracted within a multi-scale space ranging from 128×128 to 8×8. These features include global semantic features, local texture features, regional boundary features, detailed edge features, and pixel-level features. Furthermore, for environmental adaptation, the illumination, water quality, and occlusion assessment branches are run in parallel, generating an environmental complexity vector using a learnable dynamic weighting formula. Finally, in the decoding output process, a three-way parallel feature fusion strategy is adopted: the first way uses standard 3×3 convolution to extract local neighborhood features, the second way uses 5×5 void convolution to expand the receptive field, and the third way applies 7×7 depthwise separable convolution to reduce computational complexity. The three-way features are added element by element to achieve multi-scale information fusion, and bilinear interpolation is used to complete progressive upsampling, gradually reconstructing from 8×8 to 128×128 resolution. Combined with the hybrid threshold segmentation of the improved Otsu algorithm, the mask is output after morphological optimization of the 5×5 circular structure element.

[0059] The above describes the processing flow of the method for automatically extracting coral reef masks based on artificial intelligence in the embodiment of the present application. The following describes the automatic extraction system for coral reef masks based on artificial intelligence in the embodiment of the present application. Figure 3 In the embodiment of the present application, an embodiment of the coral reef mask automatic extraction system based on artificial intelligence includes: The acquisition module 301 is used to acquire multi-source coral reef remote sensing images, analyze the image quality through an environment-adaptive preprocessing algorithm, and perform layered preprocessing to generate a standardized coral reef image dataset; Input module 302 is used to construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid; An analysis module 303 is configured to establish an environment perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector, and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation; A fusion module 304 is configured to set a multi-level fusion decoder to receive the adaptive coral reef feature representation, perform progressive upsampling and cross-scale feature fusion, and generate a coral reef probability distribution map; The output module 305 is used to process the coral reef probability distribution map using an adaptive threshold segmentation algorithm, remove noise and smooth boundaries in combination with morphological post-processing, and output a coral reef mask result.

[0060] above Figure 3 The artificial intelligence-based automatic extraction system for coral reef masks in an embodiment of the present invention is described in detail from the perspective of modular functional entities. The artificial intelligence-based automatic extraction device for coral reef masks in an embodiment of the present invention is described in detail from the perspective of hardware processing.

[0061] Reference Figure 4 In an embodiment of the present invention, there is also provided an artificial intelligence-based coral reef mask automatic extraction device 400. The artificial intelligence-based coral reef mask automatic extraction device can be a server, and its internal structure can be as follows: Figure 4 As shown, the AI-based automatic coral reef mask extraction device includes a processor 402, memory 403, display screen 404, input device 405, network interface 406, and database 407 connected via a system bus 401. The computer-designed processor 402 provides computing and control capabilities. The memory 403 of the AI-based automatic coral reef mask extraction device includes a non-volatile storage medium 4031 and internal memory 4032. The non-volatile storage medium 4031 stores an operating system and a computer program. The internal memory 4032 provides an environment for the operating system and computer program in the non-volatile storage medium to run. The database 407 of the AI-based automatic coral reef mask extraction device stores the corresponding data in this embodiment. The network interface 406 of the AI-based automatic coral reef mask extraction device is used to communicate with an external terminal via a network connection. When executed by the processor, the computer program implements the above-described method.

[0062] Those skilled in the art will understand that Figure 4 The structure shown in the figure is merely a block diagram of a portion of the structure related to the solution of the present invention, and does not constitute a limitation on the artificial intelligence-based automatic extraction device for coral reef masks to which the solution of the present invention is applied.

[0063] The present invention also provides a computer-readable storage medium, which may be a non-volatile computer-readable storage medium or a volatile computer-readable storage medium. The computer-readable storage medium stores instructions, which, when executed on a computer, enable the computer to execute the steps of the artificial intelligence-based coral reef mask automatic extraction method.

[0064] Those skilled in the art will clearly understand that, for the convenience and brevity of description, the specific working processes of the above-described systems, systems and units can refer to the corresponding processes in the aforementioned method embodiments and will not be repeated here.

[0065] If the integrated unit is implemented as a software functional unit and sold or used as a standalone product, it can be stored in a computer-readable storage medium. Based on this understanding, the technical solution of the present invention, or the portion that contributes to the prior art, or all or part of the technical solution, can be embodied in the form of a software product. This computer software product, stored in a storage medium, includes instructions for enabling an artificial intelligence-based coral reef mask automatic extraction device (which can be a personal computer, server, or network device, etc.) to perform all or part of the steps of the method described in each embodiment of the present invention. The aforementioned storage medium includes various media capable of storing program code, such as a USB flash drive, a mobile hard drive, a read-only memory (ROM), a random access memory (RAM), a magnetic disk, or an optical disk.

[0066] The above embodiments are only used to illustrate the technical solutions of the present invention, rather than to limit the same. Although the present invention has been described in detail with reference to the aforementioned embodiments, those skilled in the art should understand that they can still modify the technical solutions described in the aforementioned embodiments, or make equivalent replacements for some of the technical features therein. However, these modifications or replacements do not deviate the essence of the corresponding technical solutions from the spirit and scope of the technical solutions of the various embodiments of the present invention.

Claims

1. An artificial intelligence-based automatic extraction method for coral reef masks, characterized in that: The method comprises: Step S101: Collect multi-source coral reef remote sensing images, analyze image quality using an environment-adaptive preprocessing algorithm, and perform layered preprocessing to generate a standardized coral reef image dataset; Step S102: construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid; Step S103: Establish an environmental perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector, and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation; Step S104: setting a multi-level fusion decoder to receive the adaptive coral reef feature representation, performing progressive upsampling and cross-scale feature fusion, and generating a coral reef probability distribution map; Step S105 : applying an adaptive threshold segmentation algorithm to process the coral reef probability distribution map, combining morphological post-processing to eliminate noise and smooth boundaries, and outputting a coral reef mask result.

2. The method for automatically extracting coral reef masks based on artificial intelligence according to claim 1, wherein: The step S101 further includes: The image quality assessment model is used to detect weak light areas, water turbidity, and color shift in multi-source coral reef remote sensing images, and an image quality assessment vector is obtained. Based on the image quality assessment vector, a layered preprocessing strategy is performed on the coral reef remote sensing image, size normalization and color space conversion are performed on high-quality images, histogram equalization and contrast enhancement are added to medium-quality images, and a multi-scale Retinex algorithm is used to perform illumination compensation and detail enhancement on low-quality images to obtain quality-optimized images; Performing random rotation, horizontal and vertical flipping, random cropping, and elastic deformation data enhancement operations on the quality optimized image to obtain augmented image data; Based on the amplified image data, a 512×512 pixel resolution adjustment process is performed and the pixel values ​​are normalized to the [0, 1] interval process to obtain a standardized coral reef image dataset.

3. The method for automatically extracting coral reef masks based on artificial intelligence according to claim 1, wherein: The step S102 further includes: A multi-scale feature pyramid encoder is constructed using an improved ResNet backbone network. The network depth is expanded by adding fine-grained feature layers to obtain a five-layer feature pyramid structure. Inputting the standardized coral reef image dataset into the five-layer feature pyramid structure, respectively extracting global semantic features, regional boundary features, local texture features, detail edge features and pixel-level features to obtain a multi-scale feature map group; Based on the multi-scale feature map group, depthwise separable convolution is used to replace traditional convolution for feature processing, and element-wise addition of high-level semantic information and low-level features is achieved through lateral connection and top-down path fusion to obtain a fused feature representation; The fused feature representation is subjected to channel dimension expansion processing so that each feature map contains the same number of channels, thereby obtaining a five-layer coral reef feature pyramid.

4. The method for automatically extracting coral reef masks based on artificial intelligence according to claim 1, wherein: The step S103 further includes: The environmental perception module is constructed, which includes three parallel branch networks: illumination assessment branch, water quality assessment branch, and occlusion assessment branch. The five-layer coral reef feature pyramid is input into each branch for environmental analysis to obtain the illumination uniformity index, water quality clarity index, and occlusion complexity index. Based on the illumination uniformity index, water quality clarity index and occlusion complexity index, a fusion calculation is performed through a fully connected layer to obtain an environment complexity vector; According to the environmental complexity vector, the five-layer coral reef feature pyramid is weighted using a channel attention mechanism, whereby the weights of high-resolution features are enhanced for well-lit environments, the weights of low-resolution global features are enhanced for low-light or turbid environments, and the weights of features at all scales are balanced for highly occluded environments, thereby obtaining a weighted distribution result. An element-by-element multiplication operation is performed on the weight distribution result and the five-layer coral reef feature pyramid to obtain an adaptive coral reef feature representation.

5. The method for automatically extracting coral reef masks based on artificial intelligence according to claim 1, characterized in that: The step S104 further includes: Constructing a multi-level fusion decoder includes five decoding blocks, each decoding block corresponds to a feature layer of the adaptive coral reef feature representation, and performing decoding processing through a symmetrical upsampling structure to obtain a decoder framework; The adaptive coral reef feature representation is input into the decoder framework, and a progressive feature fusion strategy is adopted to upsample the highest layer features through transposed convolution and perform jump connections with the corresponding encoder features, and fuse them layer by layer until the original resolution is restored to obtain progressive fused features; Based on the progressive fusion feature, a multi-scale receptive field module is introduced in each decoding block, and convolution kernels of different sizes are used to process the fusion features in parallel. The boundary enhancement module is combined with the Sobel operator to detect edge information, and the edge features are weightedly fused with the semantic features to obtain the boundary enhancement feature. Final decoding output processing is performed on the boundary enhancement features to generate a coral reef probability distribution map.

6. The method for automatically extracting coral reef masks based on artificial intelligence according to claim 5, characterized in that: The performing of a final decoding output process on the boundary enhancement feature to generate a coral reef probability distribution map includes: The boundary enhancement features are transferred through a residual connection structure. Each decoding block contains two convolutional layers, a batch normalization layer and an activation function to perform deep feature processing to obtain deep decoding features. Based on the deep decoding features, a fully convolutional network structure is used to perform pixel-level classification calculations, and the classification probability value of each pixel position belonging to the coral reef area is calculated through the convolution kernel to obtain the pixel classification probability; Performing Sigmoid activation function processing on the pixel classification probability, mapping the probability value to the range of zero to one, and obtaining a normalized probability value; A complete two-dimensional probability distribution matrix is ​​formed by reconstructing pixel positions according to the normalized probability values ​​to generate a coral reef probability distribution map.

7. The method for automatically extracting coral reef masks based on artificial intelligence according to claim 1, characterized in that: The step S105 further includes: Performing a histogram distribution feature analysis on the coral reef probability distribution map, and obtaining an adaptive segmentation threshold by calculating a threshold point with the maximum inter-class variance and combining it with the connectivity constraint of the coral reef area; performing a binary segmentation process on the coral reef probability distribution map based on the adaptive segmentation threshold, marking pixels with probability values ​​higher than the threshold as coral reef areas, and marking pixels with probability values ​​lower than the threshold as background areas, to obtain an initial binary mask; Performing a sequence of morphological opening and closing operations on the initial binary mask, first performing an erosion and dilation operation using a structural element to remove noise points, and then performing an erosion and dilation operation to fill holes and breaks, to obtain a morphologically optimized mask; The morphologically optimized mask is subjected to boundary smoothing processing using a Gaussian filter to eliminate boundary jagged artifacts and calculate a comprehensive score of mask quality to obtain a coral reef mask result.

8. An artificial intelligence-based automatic coral reef mask extraction system, characterized in that: For implementing the method for automatically extracting coral reef masks based on artificial intelligence according to any one of claims 1 to 7, the automatic extraction system for coral reef masks based on artificial intelligence comprises: The acquisition module is used to collect multi-source coral reef remote sensing images, analyze image quality through an environment-adaptive preprocessing algorithm, and perform layered preprocessing to generate a standardized coral reef image dataset; An input module is used to construct a multi-scale feature pyramid encoder, input the standardized coral reef image dataset into the encoder to extract feature information of different resolutions, and output a five-layer coral reef feature pyramid; An analysis module is used to establish an environmental perception module to analyze the five-layer coral reef feature pyramid, calculate the environmental complexity vector and recalibrate the feature weights of each layer to form an adaptive coral reef feature representation; a fusion module, configured to set a multi-level fusion decoder to receive the adaptive coral reef feature representation, perform progressive upsampling and cross-scale feature fusion, and generate a coral reef probability distribution map; The output module is used to apply an adaptive threshold segmentation algorithm to process the coral reef probability distribution map, combine morphological post-processing to eliminate noise and smooth boundaries, and output a coral reef mask result.

9. An artificial intelligence-based automatic coral reef mask extraction device, characterized in that: The method comprises a memory and a processor, wherein the memory stores a computer program that can be run on the processor, and when the processor executes the computer program, the method for automatically extracting coral reef masks based on artificial intelligence according to any one of claims 1 to 7 is implemented.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the processor is enabled to perform the method for automatic extraction of coral reef masks based on artificial intelligence according to any one of claims 1 to 7.

Images