Wetland carbon storage estimation method and system based on remote sensing technology

Abstract

The present application relates to the field of intelligent carbon stock estimation, and specifically discloses a wetland carbon stock estimation method and system based on remote sensing technology, which collects remote sensing images of wetland areas through remote sensing sensors, and uses image analysis and processing algorithms based on deep learning to perform local area segmentation and image semantic feature association on the remote sensing images of the wetland areas, so as to intelligently obtain the vegetation type and quantity of each area, and obtain the carbon stock estimation value of the wetland based on the vegetation type and quantity of each area. In this way, remote sensing technology can be used to cover a wide range of wetland areas. At the same time, the deep learning algorithm can extract precise features from remote sensing images, and the automated image processing process reduces manual intervention, improves the efficiency and processing speed of wetland carbon stock estimation, and also improves the accuracy of carbon stock estimation.

Description

Technical Field

[0001] The present application relates to the field of intelligent estimation, and more specifically, to a wetland carbon storage estimation method and system based on remote sensing technology. Background Art

[0002] Wetland ecosystems play a vital role in the global carbon cycle because they are able to accumulate and store large amounts of carbon for long periods of time under their unique anaerobic conditions. Wetland types such as peatlands, mangroves and swamps actively absorb carbon dioxide from the atmosphere through photosynthesis of vegetation. Over time, these organic materials accumulate in the soil, forming carbon-rich layers that form carbon sinks. Carbon stock estimation is a key part of studying the Earth's carbon cycle. By estimating and understanding the carbon storage capacity of different ecosystems, measures can be taken to increase their carbon sink potential, which can help promote sustainable development practices such as rational forestry management and wetland protection.

[0003] However, the traditional method of estimating wetland carbon storage is to conduct field surveys and collect soil and plant samples at different locations in the wetland for laboratory analysis, which may require a lot of manpower and time, resulting in low efficiency. Secondly, a vast wetland area may contain a variety of different ecosystems and terrains, and ground sampling may only cover a limited area, and may not fully represent the carbon storage of the entire wetland due to the limitations of the sampling points.

[0004] Therefore, an optimized method for estimating wetland carbon storage is desired. Summary of the invention

[0005] In order to solve the above-mentioned technical problems, the present application is proposed. The embodiment of the present application provides a wetland carbon stock estimation method and system based on remote sensing technology, which collects remote sensing images of wetland areas through remote sensing sensors, and uses image analysis and processing algorithms based on deep learning to perform local area segmentation and image semantic feature association on the remote sensing images of the wetland areas, so as to intelligently obtain the vegetation type and quantity of each area, and based on the vegetation type and quantity of each area, obtain the carbon stock estimation value of the wetland. In this way, remote sensing technology can be used to cover a wide range of wetland areas. At the same time, the deep learning algorithm can extract precise features from remote sensing images, and the automated image processing process reduces manual intervention, improves the efficiency and processing speed of wetland carbon stock estimation, and also improves the accuracy of carbon stock estimation.

[0006] According to one aspect of the present application, a method for estimating wetland carbon storage based on remote sensing technology is provided, which comprises:

[0007] Collect remote sensing images of wetland areas through remote sensing sensors;

[0008] After segmenting the remote sensing images of the wetland area to obtain a set of local wetland remote sensing images, extracting wetland remote sensing local area features from the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps;

[0009] Performing wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps;

[0010] The set of local wetland remote sensing image context semantic association feature maps and each corresponding group of local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps;

[0011] Performing feature enhancement on the set of multi-scale local wetland remote sensing image feature maps to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors;

[0012] Based on the set of enhanced multi-scale local wetland remote sensing image feature vectors, the vegetation type and quantity in each area are obtained;

[0013] Based on the vegetation type and quantity in each area, an estimated carbon storage value of the wetland is obtained.

[0014] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, after the remote sensing image of the wetland area is segmented to obtain a set of local wetland remote sensing images, the set of local wetland remote sensing images is subjected to wetland remote sensing local area feature extraction to obtain a set of local wetland remote sensing image feature maps, including: segmenting the remote sensing image of the wetland area to obtain the set of local wetland remote sensing images; and inputting the set of local wetland remote sensing images into a wetland remote sensing local area feature extraction module based on a fully convolutional neural network to obtain a set of local wetland remote sensing image feature maps.

[0015] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, the set of local wetland remote sensing image feature maps is subjected to wetland remote sensing local image semantic feature association to obtain a set of local wetland remote sensing image contextual semantic association feature maps, including: inputting the set of local wetland remote sensing image feature maps into a wetland remote sensing local image semantic feature association extractor based on Swin Transformer to obtain a set of local wetland remote sensing image contextual semantic association feature maps.

[0016] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, the set of local wetland remote sensing image context semantic association feature maps and each corresponding group of local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps, including: expanding the local wetland remote sensing image context semantic association feature map and the local wetland remote sensing image feature map to obtain a local wetland remote sensing image context semantic association feature vector and a local wetland remote sensing image feature vector; performing feature intrinsic alignment of the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector based on essential feature recognition to obtain a multi-scale local wetland remote sensing image feature vector; and performing dimensional reconstruction of the multi-scale local wetland remote sensing image feature vector to obtain the multi-scale local wetland remote sensing image feature map.

[0017] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector are subjected to feature intrinsic alignment based on essential feature recognition to obtain a multi-scale local wetland remote sensing image feature vector, including: performing normalization modulation based on vector norming on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a pre-aligned local wetland remote sensing image context semantic association feature vector and a pre-aligned local wetland remote sensing image feature vector; performing essential feature recognition on the pre-aligned local wetland remote sensing image context semantic association feature vector and the pre-aligned local wetland remote sensing image feature vector to obtain a first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and a second principal component pre-aligned local wetland remote sensing image feature vector; performing intrinsic fine-grained alignment and fusion on the first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and the second principal component pre-aligned local wetland remote sensing image feature vector to obtain the multi-scale local wetland remote sensing image feature vector.

[0018] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, the set of multi-scale local wetland remote sensing image feature maps is feature enhanced to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors, including: inputting the set of multi-scale local wetland remote sensing image feature maps into a bidirectional attention mechanism model to obtain the set of enhanced multi-scale local wetland remote sensing image feature vectors.

[0019] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, the vegetation type and quantity of each area are obtained based on the set of enhanced multi-scale local wetland remote sensing image feature vectors, including: inputting the set of enhanced multi-scale local wetland remote sensing image feature vectors into a vegetation type determiner based on a classifier to obtain the vegetation type of each area; inputting the set of enhanced multi-scale local wetland remote sensing image feature vectors into a vegetation quantity estimator based on a decoder to obtain the vegetation quantity of each area.

[0020] In the above-mentioned wetland carbon stock estimation method based on remote sensing technology, the carbon stock estimation value of the wetland is obtained based on the vegetation type and quantity of each area, including: inputting the vegetation type and vegetation quantity of each area into the carbon stock estimation model to obtain the carbon stock estimation value of the wetland.

[0021] According to another aspect of the present application, a wetland carbon storage estimation system based on remote sensing technology is provided, which includes:

[0022] A remote sensing image acquisition module for wetland areas, used to acquire remote sensing images of wetland areas through remote sensing sensors;

[0023] A wetland remote sensing image segmentation feature extraction module is used to segment the remote sensing image of the wetland area to obtain a set of local wetland remote sensing images, and then extract wetland remote sensing local area features from the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps;

[0024] A wetland remote sensing local area semantic association module is used to perform wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps;

[0025] A multi-scale local wetland remote sensing image feature fusion module is used to fuse each corresponding group of local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature map in the set of local wetland remote sensing image context semantic association feature maps to obtain a set of multi-scale local wetland remote sensing image feature maps;

[0026] An image feature enhancement module, used for performing feature enhancement on the set of multi-scale local wetland remote sensing image feature maps to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors;

[0027] A vegetation type and quantity generation module, used to obtain the vegetation type and quantity of each area based on the set of enhanced multi-scale local wetland remote sensing image feature vectors;

[0028] The carbon storage estimation module is used to obtain the estimated carbon storage value of the wetland based on the vegetation type and quantity of each area.

[0029] In the above-mentioned wetland carbon stock estimation system based on remote sensing technology, the wetland remote sensing image segmentation feature extraction module is used to: segment the remote sensing image of the wetland area to obtain a set of local wetland remote sensing images; and input the set of local wetland remote sensing images into the wetland remote sensing local area feature extraction module based on the fully convolutional neural network to obtain a set of local wetland remote sensing image feature maps.

[0030] Compared with the prior art, the present application provides a wetland carbon storage estimation method and system based on remote sensing technology, which collects remote sensing images of wetland areas through remote sensing sensors, and uses image analysis and processing algorithms based on deep learning to perform local area segmentation and image semantic feature association on the remote sensing images of the wetland areas, so as to intelligently obtain the vegetation type and quantity in each area, and based on the vegetation type and quantity in each area, obtain the carbon storage estimation value of the wetland. In this way, remote sensing technology can be used to cover a wide range of wetland areas. At the same time, the deep learning algorithm can extract precise features from remote sensing images, and the automated image processing process reduces manual intervention, improves the efficiency and processing speed of wetland carbon storage estimation, and also improves the accuracy of carbon storage estimation. BRIEF DESCRIPTION OF THE DRAWINGS

[0031] By describing the embodiments of the present application in more detail in conjunction with the accompanying drawings, the above and other purposes, features and advantages of the present application will become more apparent. The accompanying drawings are used to provide a further understanding of the embodiments of the present application and constitute a part of the specification. Together with the embodiments of the present application, they are used to explain the present application and do not constitute a limitation of the present application. In the accompanying drawings, the same reference numerals generally represent the same components or steps.

[0032] Figure 1 Flow chart of a wetland carbon storage estimation method based on remote sensing technology according to an embodiment of the present application.

[0033] Figure 2 The present invention is a flowchart of segmenting the remote sensing images of the wetland area to obtain a set of local wetland remote sensing images in the wetland carbon stock estimation method based on remote sensing technology according to an embodiment of the present application, and then performing wetland remote sensing local area feature extraction on the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps.

[0034] Figure 3 This is a flowchart for obtaining the vegetation type and quantity in each area based on the set of enhanced multi-scale local wetland remote sensing image feature vectors in the wetland carbon storage estimation method based on remote sensing technology according to an embodiment of the present application.

[0035] Figure 4 4 is a block diagram of a wetland carbon storage estimation system based on remote sensing technology according to an embodiment of the present application. DETAILED DESCRIPTION

[0036] The embodiments of the present disclosure will be described in more detail below with reference to the accompanying drawings. Although certain embodiments of the present disclosure are shown in the accompanying drawings, it should be understood that the present disclosure can be implemented in various forms and should not be construed as being limited to the embodiments described herein, but rather these embodiments are provided for a more thorough and complete understanding of the present disclosure. It should be understood that the drawings and embodiments of the present disclosure are only for exemplary purposes and are not intended to limit the scope of protection of the present disclosure.

[0037] It should be understood that the various steps described in the method embodiments of the present disclosure may be performed in different orders and / or in parallel. In addition, the method embodiments may include additional steps and / or omit the steps shown. The scope of the present disclosure is not limited in this respect.

[0038] In the description of the embodiments of the present disclosure, the term "including" and similar terms should be understood as open inclusion, that is, "including but not limited to". The term "based on" should be understood as "based at least in part on". The term "one embodiment" or "the embodiment" should be understood as "at least one embodiment". The terms "first", "second", etc. may refer to different or the same objects. Other explicit and implicit definitions may also be included below.

[0039] It should be noted that the modifications of "one" and "plurality" mentioned in the present disclosure are illustrative rather than restrictive, and those skilled in the art should understand that unless otherwise clearly indicated in the context, it should be understood as "one or more".

[0040] Various exemplary embodiments, features and aspects of the present disclosure will be described in detail below with reference to the accompanying drawings. The same reference numerals in the accompanying drawings represent elements with the same or similar functions. Although various aspects of the embodiments are shown in the accompanying drawings, the drawings are not necessarily drawn to scale unless otherwise specified.

[0041] Wetland ecosystems play a vital role in the global carbon cycle because they are able to accumulate and store large amounts of carbon for long periods of time under their unique anaerobic conditions. Wetland types such as peatlands, mangroves and swamps actively absorb carbon dioxide from the atmosphere through photosynthesis of vegetation. Over time, these organic materials accumulate in the soil, forming carbon-rich layers that form carbon sinks. Carbon stock estimation is a key part of studying the Earth's carbon cycle. By estimating and understanding the carbon storage capacity of different ecosystems, measures can be taken to increase their carbon sink potential, which can help promote sustainable development practices such as rational forestry management and wetland protection.

[0042] However, the traditional method of estimating wetland carbon storage is to conduct field surveys and collect soil and plant samples at different locations in the wetland for laboratory analysis, which may require a lot of manpower and time, resulting in low efficiency. Secondly, a vast wetland area may contain a variety of different ecosystems and terrains, and ground sampling may only cover a limited area, and may not fully represent the carbon storage of the entire wetland due to the limitations of the sampling points.

[0043] It should be understood that remote sensing technology refers to a technology that does not directly contact the target object, but obtains the target characteristics by recording the electromagnetic wave information emitted or reflected by the target object. Remote sensing technology is widely used in geographic information systems (GIS), environmental monitoring, weather forecasting, agricultural management, urban planning, geological exploration, military reconnaissance and other fields. With the continuous development of remote sensing technology, remote sensing technology has also played a huge role in wetland carbon storage estimation. That is, remote sensing technology can cover a large range of wetland areas and realize comprehensive monitoring of wetland ecosystems. Through remote sensing technology, scientists and environmental managers can more effectively monitor and evaluate the health status and carbon storage capacity of wetland ecosystems, providing important tools for wetland protection and global carbon cycle research.

[0044] Therefore, a wetland carbon storage estimation method based on remote sensing technology is desired, which collects remote sensing images of wetland areas through remote sensing sensors, and uses image analysis and processing algorithms based on deep learning to perform local area segmentation and image semantic feature association on the remote sensing images of the wetland areas, so as to intelligently obtain the vegetation type and quantity of each area, and obtain the carbon storage estimation value of the wetland based on the vegetation type and quantity of each area. In this way, remote sensing technology can be used to cover a wide range of wetland areas. At the same time, the deep learning algorithm can extract accurate features from remote sensing images, and the automated image processing process reduces manual intervention, improves the efficiency and processing speed of wetland carbon storage estimation, and also improves the accuracy of carbon storage estimation.

[0045] Figure 1 Flow chart of a wetland carbon storage estimation method based on remote sensing technology according to an embodiment of the present application. Figure 1As shown, according to the wetland carbon stock estimation method based on remote sensing technology according to the embodiment of the present application, the method comprises: S110, collecting remote sensing images of a wetland area by means of a remote sensing sensor; S120, segmenting the remote sensing images of the wetland area to obtain a set of local wetland remote sensing images, and then performing wetland remote sensing local area feature extraction on the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps; S130, performing wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps; S140, associating the local wetland remote sensing image context semantic association feature maps The set of semantic association feature maps and the set of local wetland remote sensing image feature maps, each corresponding group of local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps; S150, feature enhancement is performed on the set of multi-scale local wetland remote sensing image feature maps to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors; S160, based on the set of enhanced multi-scale local wetland remote sensing image feature vectors, the vegetation type and quantity of each area are obtained; S170, based on the vegetation type and quantity of each area, the carbon storage estimate of the wetland is obtained.

[0046] In step S110, remote sensing images of wetland areas are collected by remote sensing sensors. It should be understood that, considering that the remote sensing images of the wetland areas refer to visual data of the wetland areas obtained using remote sensing technology, these data may be optical images, radar images, infrared images or other types of electromagnetic wave images. Wetland remote sensing images are helpful in monitoring vegetation types, coverage and growth conditions, which are key factors in assessing wetland carbon reserves. Based on this, in the technical solution of the present application, remote sensing images of wetland areas are collected by remote sensing sensors, and the remote sensing images are analyzed and processed to provide vegetation characteristic information about the wetlands, so as to better understand the state of the wetland ecosystem and assess its carbon reserves.

[0047] In step S120, after segmenting the remote sensing images of the wetland area to obtain a set of local wetland remote sensing images, wetland remote sensing local area features are extracted from the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps. Figure 2 The present invention is a flow chart of the method for estimating wetland carbon storage based on remote sensing technology according to an embodiment of the present application, wherein after the remote sensing image of the wetland area is segmented to obtain a set of local wetland remote sensing images, the set of local wetland remote sensing images is subjected to wetland remote sensing local area feature extraction to obtain a set of local wetland remote sensing image feature maps. Specifically, in the embodiment of the present application, Figure 2As shown, after the remote sensing image of the wetland area is segmented to obtain a set of local wetland remote sensing images, wetland remote sensing local area features are extracted for the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps, including: S210, segmenting the remote sensing image of the wetland area to obtain the set of local wetland remote sensing images; S220, inputting the set of local wetland remote sensing images into a wetland remote sensing local area feature extraction module based on a fully convolutional neural network to obtain a set of local wetland remote sensing image feature maps.

[0048] Specifically, in step S210, the remote sensing image of the wetland area is segmented to obtain a set of local wetland remote sensing images. Accordingly, it is considered that different local areas in the remote sensing image of the wetland area have different vegetation characteristics, such as different ecological characteristic information such as different vegetation types. Therefore, in the technical solution of the present application, the remote sensing image of the wetland area is segmented to obtain a set of local wetland remote sensing images. It should be understood that segmentation allows for a more detailed analysis of local areas of the remote sensing image of the wetland area, which helps to identify and understand ecological characteristics and changes at a small scale, that is, smaller image blocks may make it easier to identify vegetation details in the wetland, such as small water bodies, vegetation patches or other important ecological features, so that wetland data can be processed and analyzed more effectively.

[0049] Specifically, in step S220, the set of local wetland remote sensing images is respectively input into the wetland remote sensing local area feature extraction module based on the full convolutional neural network to obtain the set of local wetland remote sensing image feature maps. Furthermore, in order to capture and mine the ecological feature information and structure contained in each local wetland remote sensing image in the set of local wetland remote sensing images, in the technical solution of the present application, the set of local wetland remote sensing images is respectively input into the wetland remote sensing local area feature extraction module based on the full convolutional neural network to obtain the set of local wetland remote sensing image feature maps. In particular, the wetland remote sensing local area feature extraction module based on the full convolutional neural network surrounds the full convolutional neural network model. It is worth mentioning that the full convolutional neural network is particularly good at extracting features from images, which is crucial for understanding the complex ecological and environmental information in wetland remote sensing images, and the full convolutional network can process image features of different scales, which helps to capture micro and macro changes in wetland ecosystems and provide richer feature representations.

[0050] In step S130, the set of local wetland remote sensing image feature maps is subjected to wetland remote sensing local image semantic feature association to obtain a set of local wetland remote sensing image contextual semantic association feature maps. Specifically, in an embodiment of the present application, the set of local wetland remote sensing image feature maps is subjected to wetland remote sensing local image semantic feature association to obtain a set of local wetland remote sensing image contextual semantic association feature maps, including: inputting the set of local wetland remote sensing image feature maps into a wetland remote sensing local image semantic feature association extractor based on Swin Transformer to obtain the set of local wetland remote sensing image contextual semantic association feature maps. Accordingly, it is considered that each local wetland remote sensing image feature map in the set of local wetland remote sensing image feature maps has a contextual semantic association influence relationship between local features. Based on this, in order to understand and extract the semantic dependencies between contexts from the local wetland remote sensing image feature map, in the technical solution of the present application, the set of local wetland remote sensing image feature maps is input into the wetland remote sensing local image semantic feature association extractor based on Swin Transformer to obtain a set of local wetland remote sensing image context semantic association feature maps. In particular, the essence of the wetland remote sensing local image semantic feature association extractor based on Swin Transformer is Swin Transformer. It is worth mentioning that Swin Transformer, as a model based on the attention mechanism, can effectively capture the contextual information in the image, which is crucial for understanding the complexity and diversity of wetland ecosystems, and can extract richer semantic information from local wetland remote sensing images.

[0051] In step S140, the set of local wetland remote sensing image context semantic association feature maps and each group of corresponding local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps. It should be understood that the local wetland remote sensing image context semantic association feature map and the local wetland remote sensing image feature map respectively express the semantic association between wetland feature information and context in the local image domain. Therefore, in order to combine the context semantic association features and the original feature map to improve the accuracy of wetland type classification and state recognition, in the technical solution of the present application, the set of local wetland remote sensing image context semantic association feature maps and each group of corresponding local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps, so that the feature information of local wetlands can be more comprehensively and richly represented.

[0052] In particular, in the technical solution of the present application, it is considered that although the directly fused local wetland remote sensing image context semantic association feature map and the local wetland remote sensing image feature map are both derived from the remote sensing image of the same area, the information they carry has different properties and levels. The local wetland remote sensing image feature map mainly captures spatial detail information, such as texture, shape, etc.; while the local wetland remote sensing image context semantic association feature map focuses on expressing the correlation between regions and high-level semantic information, such as the distribution pattern of different vegetation types. If these two types of feature maps are simply fused without considering the inherent differences between them, it may cause a mismatch between high-level semantic information and low-level spatial detail information, thereby causing information distortion or loss of important details. In order to solve this problem, a feature intrinsic alignment method based on essential feature recognition is adopted to effectively solve the alignment problem between high- and low-level features. This method ensures that in the fusion process, not only the original spatial detail information is retained, but also the high-level semantic information is accurately transmitted by analyzing and adjusting the intrinsic features in the feature map. The multi-scale local wetland remote sensing image feature vectors obtained in this way can more accurately reflect the actual conditions of the wetlands and provide more reliable data support for subsequent vegetation type judgment, vegetation quantity estimation and carbon storage estimation.

[0053] Specifically, in an embodiment of the present application, the set of the local wetland remote sensing image context semantic association feature maps and each group of corresponding local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of the local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps, including: expanding the local wetland remote sensing image context semantic association feature map and the local wetland remote sensing image feature map to obtain a local wetland remote sensing image context semantic association feature vector and a local wetland remote sensing image feature vector; performing feature intrinsic alignment based on essential feature recognition on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a multi-scale local wetland remote sensing image feature vector; and performing dimensional reconstruction on the multi-scale local wetland remote sensing image feature vector to obtain the multi-scale local wetland remote sensing image feature map.

[0054] More specifically, in an embodiment of the present application, the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector are subjected to intrinsic feature alignment based on essential feature recognition to obtain a multi-scale local wetland remote sensing image feature vector, including: performing normalization modulation based on vector norming on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a pre-aligned local wetland remote sensing image context semantic association feature vector and a pre-aligned local wetland remote sensing image feature vector; performing essential feature recognition on the pre-aligned local wetland remote sensing image context semantic association feature vector and the pre-aligned local wetland remote sensing image feature vector to obtain a first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and a second principal component pre-aligned local wetland remote sensing image feature vector; performing intrinsic fine-grained alignment and fusion on the first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and the second principal component pre-aligned local wetland remote sensing image feature vector to obtain the multi-scale local wetland remote sensing image feature vector.

[0055] In an embodiment of the present application, specifically, another implementation expression of performing feature intrinsic alignment based on essential feature recognition on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a multi-scale local wetland remote sensing image feature vector can be: processing the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector using the following formula to obtain a multi-scale local wetland remote sensing image feature vector; wherein the formula is:

[0056]

[0057]

[0058]

[0059]

[0060]

[0061] in, Represents the contextual semantic association feature vector of the local wetland remote sensing image, represents the local wetland remote sensing image feature vector, represents the square of the vector's two-norm, The first character representing the contextual semantic association feature vector of the local wetland remote sensing image The eigenvalues ​​at the positions, The first character vector representing the local wetland remote sensing image The eigenvalues ​​at the positions, represents the contextual semantic association feature vector of the pre-aligned local wetland remote sensing image, represents the pre-aligned local wetland remote sensing image feature vector, Indicates essential feature recognition, represents the sequence of first eigendecomposition vectors, represents the first diagonal matrix, and Represent the first position and the first diagonal matrix respectively. The eigenvalues ​​at the positions, represents the transpose of a vector, represents vector concatenation, , , The first, second and third sequences of the first principal component pre-aligned local wetland remote sensing image contextual semantic association feature vectors are represented by feature vectors, represents the contextual semantic association feature vector of the first principal component pre-aligned local wetland remote sensing image, represents the sequence of the second eigendecomposition vectors, represents the first diagonal matrix, and Represent the first and second positions of the second diagonal matrix respectively. The eigenvalues ​​at the positions, represents vector concatenation, , , The first, second and third sequences of the second principal component pre-aligned local wetland remote sensing image feature vectors are represented by feature vectors, represents the second principal component pre-aligned local wetland remote sensing image feature vector, It means subtracting by position. It means point multiplication by position. It means adding by position. represents point convolution, Represents the multi-scale local wetland remote sensing image feature vector.

[0062] That is, in the technical solution of the present application, the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector are subjected to intrinsic feature alignment based on essential feature recognition to obtain a multi-scale local wetland remote sensing image feature vector, and the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector are first subjected to normalization modulation based on vector norm assignment to obtain a pre-aligned local wetland remote sensing image context semantic association feature vector and a pre-aligned local wetland remote sensing image feature vector. This process is not just for simply adjusting the scale, but is an important technology in feature engineering, which aims to solve the problem of inconsistent scales between features in a data set. By calculating the modulus of the feature vector and dividing each vector by its own modulus, it can be ensured that all feature vectors have the same length, thereby eliminating the impact of scale differences between different feature vectors. Normalization processing not only contributes to numerical stability in subsequent processing, but also makes the comparison between feature vectors more fair and reasonable, laying the foundation for subsequent feature alignment.

[0063] Next, the pre-aligned local wetland remote sensing image context semantic association feature vector and the pre-aligned local wetland remote sensing image feature vector are subjected to essential feature recognition to obtain the first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and the second principal component pre-aligned local wetland remote sensing image feature vector. That is, after completing the normalization modulation, the pre-aligned local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector are subjected to essential feature recognition. As an unsupervised linear dimensionality reduction method, essential feature recognition is widely used in data compression and feature extraction. This process determines the principal components by finding the direction of the maximum variance of the data. These principal components are new coordinate axes in the original feature space and can capture the structure of the data with minimal information loss. It is worth mentioning that by selecting the eigenvectors corresponding to the first few maximum eigenvalues, the dimension of the feature space can be effectively reduced while maintaining the key information of the data. This dimensionality reduction can not only accelerate the model training process, but also help the model to generalize better and reduce the impact of noise, thereby improving the performance of the model.

[0064] Finally, the first principal component pre-aligned local wetland remote sensing image contextual semantic association feature vector and the second principal component pre-aligned local wetland remote sensing image feature vector are intrinsically fine-grained aligned and fused to obtain the multi-scale local wetland remote sensing image feature vector. The goal of this step is to find the best matching point or area between the two principal component feature vectors to achieve alignment between the two. Fine-grained alignment goes beyond the traditional rough fusion methods such as simple weighted averaging or splicing, and needs to consider the time series characteristics or spatial distribution characteristics of the feature vector. The fusion after alignment is not a simple numerical operation, but requires the design of a specific mechanism to determine how to combine the information of the two feature vectors. This content-based intelligent fusion method can more effectively utilize multi-source information and improve the representation ability and decision-making quality of the model.

[0065] In step S150, the set of multi-scale local wetland remote sensing image feature maps is feature enhanced to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors. Specifically, in an embodiment of the present application, the set of multi-scale local wetland remote sensing image feature maps is feature enhanced to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors, including: inputting the set of multi-scale local wetland remote sensing image feature maps into a two-way attention mechanism model to obtain the set of enhanced multi-scale local wetland remote sensing image feature vectors. Accordingly, considering that each multi-scale local wetland remote sensing image feature map in the set of multi-scale local wetland remote sensing image feature maps contains key feature information and redundant information about wetland remote sensing. Based on this, in the technical solution of the present application, the set of multi-scale local wetland remote sensing image feature maps is input into a two-way attention mechanism model to capture and extract important wetland feature information, thereby obtaining the set of enhanced multi-scale local wetland remote sensing image feature vectors. It should be understood that the bidirectional attention mechanism model can simultaneously capture the feature information in the horizontal and vertical directions of the multi-scale local wetland remote sensing image feature map, and can adaptively select the most effective features for the current task, ignore irrelevant or redundant features, and obtain an enhanced multi-scale local wetland remote sensing image feature map with representation capabilities. Then, in order to retain and reduce the calculation of the model, the enhanced multi-scale local wetland remote sensing image feature map is pooled to obtain an enhanced multi-scale local wetland remote sensing image feature vector.

[0066] In step S160, the vegetation type and quantity in each area are obtained based on the set of enhanced multi-scale local wetland remote sensing image feature vectors. Figure 3 The present invention is a flowchart of obtaining the vegetation type and quantity in each area based on the set of enhanced multi-scale local wetland remote sensing image feature vectors in the wetland carbon storage estimation method based on remote sensing technology according to the embodiment of the present application. Specifically, in the embodiment of the present application, Figure 3As shown, based on the set of enhanced multi-scale local wetland remote sensing image feature vectors, the vegetation type and quantity of each area are obtained, including: S310, inputting the set of enhanced multi-scale local wetland remote sensing image feature vectors into a vegetation type determiner based on a classifier to obtain the vegetation type of each area; S320, inputting the set of enhanced multi-scale local wetland remote sensing image feature vectors into a vegetation quantity estimator based on a decoder to obtain the vegetation quantity of each area. That is, the vegetation type and quantity of each area are intelligently obtained by performing classification processing and decoding processing on the set of enhanced multi-scale local wetland remote sensing image feature vectors. It should be understood that the vegetation type determiner based on the classifier can accurately identify and distinguish different vegetation types from complex remote sensing data. The vegetation quantity estimator based on the decoder is usually used to recover or estimate the specific values ​​of the original data from the feature map, such as vegetation coverage or biomass. In this way, carbon storage can be evaluated more accurately.

[0067] In step S170, based on the vegetation type and quantity of each region, the carbon storage estimate of the wetland is obtained. Specifically, in an embodiment of the present application, based on the vegetation type and quantity of each region, the carbon storage estimate of the wetland is obtained, including: inputting the vegetation type and vegetation quantity of each region into a carbon storage estimation model to obtain the carbon storage estimate of the wetland. It should be understood that different vegetation types have different biomass and growth rates. For example, the biomass of forests is usually higher than that of grasslands or wetland plants. Therefore, the difference in vegetation types directly affects the estimation of carbon storage. In addition, the amount of vegetation, usually measured by coverage, can also directly affect the carbon absorption capacity of the wetland. In particular, areas with high vegetation coverage usually have higher carbon fixation potential. Based on this, in the technical solution of the present application, the vegetation type and vegetation quantity of each region are input into a carbon storage estimation model to obtain the carbon storage estimate of the wetland. In particular, the carbon stock estimation model has been trained with a large amount of actual data, which means that it has been optimized based on extensive historical data and can more accurately reflect the relationship between different vegetation types and their quantities and actual carbon storage. This ensures that the model has high accuracy and reliability when dealing with diverse wetland ecosystems. In this way, not only the efficiency of carbon stock assessment work is improved, but also the accuracy of the results is enhanced.

[0068] In summary, a wetland carbon stock estimation method based on remote sensing technology based on the embodiment of the present application is explained, which collects remote sensing images of wetland areas through remote sensing sensors, and uses deep learning-based image analysis and processing algorithms to perform local area segmentation and image semantic feature association on the remote sensing images of the wetland areas, so as to intelligently obtain the vegetation type and quantity in each area, and based on the vegetation type and quantity in each area, obtain the carbon stock estimation value of the wetland. In this way, remote sensing technology can be used to cover a wide range of wetland areas. At the same time, the deep learning algorithm can extract precise features from remote sensing images, and the automated image processing process reduces manual intervention, improves the efficiency and processing speed of wetland carbon stock estimation, and also improves the accuracy of carbon stock estimation.

[0069] Figure 4 FIG. 1 is a block diagram of a wetland carbon storage estimation system based on remote sensing technology according to an embodiment of the present application. Figure 4 As shown, according to the wetland carbon storage estimation system 100 based on remote sensing technology of the embodiment of the present application, it includes: a wetland area remote sensing image acquisition module 110, which is used to collect remote sensing images of wetland areas through remote sensing sensors; a wetland remote sensing image segmentation feature extraction module 120, which is used to segment the remote sensing images of the wetland area to obtain a set of local wetland remote sensing images, and then perform wetland remote sensing local area feature extraction on the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps; a wetland remote sensing local area semantic association module 130, which is used to perform wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps; a multi-scale local wetland remote sensing image feature fusion module 14 0, used for fusing the set of local wetland remote sensing image context semantic association feature maps and each group of corresponding local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of local wetland remote sensing image feature maps to obtain a set of multi-scale local wetland remote sensing image feature maps; image feature enhancement module 150, used for performing feature enhancement on the set of multi-scale local wetland remote sensing image feature maps to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors; vegetation type and quantity generation module 160, used for obtaining the vegetation type and quantity of each area based on the set of enhanced multi-scale local wetland remote sensing image feature vectors; carbon stock estimation module 170, used for obtaining the carbon stock estimation value of the wetland based on the vegetation type and quantity of each area.

[0070] Here, those skilled in the art will appreciate that the specific operations of each step in the wetland carbon storage estimation system based on remote sensing technology have been described in the reference above. Figures 1 to 3It has been introduced in detail in the description of the remote sensing technology-based wetland carbon stock estimation method, and therefore, its repeated description will be omitted.

[0071] The above specific implementation methods further illustrate the objectives, technical solutions and beneficial effects of the present invention in detail. It should be understood that the above are only specific implementation methods of the present invention and are not intended to limit the scope of protection of the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the scope of protection of the present invention.

Claims

1. A wetland carbon storage estimation method based on remote sensing technology, characterized in that: include: Collect remote sensing images of wetland areas through remote sensing sensors; After segmenting the remote sensing images of the wetland area to obtain a set of local wetland remote sensing images, extracting wetland remote sensing local area features from the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps; Performing wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps; The set of local wetland remote sensing image context semantic association feature maps and each corresponding group of local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature maps in the set of local wetland remote sensing image feature maps are fused to obtain a set of multi-scale local wetland remote sensing image feature maps; Performing feature enhancement on the set of multi-scale local wetland remote sensing image feature maps to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors; Based on the set of enhanced multi-scale local wetland remote sensing image feature vectors, the vegetation type and quantity in each area are obtained; Based on the vegetation type and quantity in each area, an estimated carbon storage value of the wetland is obtained; The method comprises fusing the set of the local wetland remote sensing image context semantic association feature maps and each corresponding group of the local wetland remote sensing image context semantic association feature maps and the local wetland remote sensing image feature maps in the set of the local wetland remote sensing image feature maps to obtain a set of multi-scale local wetland remote sensing image feature maps, including: Expanding the local wetland remote sensing image context semantic association feature map and the local wetland remote sensing image feature map to obtain a local wetland remote sensing image context semantic association feature vector and a local wetland remote sensing image feature vector; Performing feature intrinsic alignment based on essential feature recognition on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a multi-scale local wetland remote sensing image feature vector; Reconstructing the dimension of the multi-scale local wetland remote sensing image feature vector to obtain the multi-scale local wetland remote sensing image feature map; The method of performing feature intrinsic alignment based on essential feature recognition on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a multi-scale local wetland remote sensing image feature vector includes: Performing normalization modulation based on vector norming on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a pre-aligned local wetland remote sensing image context semantic association feature vector and a pre-aligned local wetland remote sensing image feature vector; Performing essential feature recognition on the pre-aligned local wetland remote sensing image context semantic association feature vector and the pre-aligned local wetland remote sensing image feature vector to obtain a first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and a second principal component pre-aligned local wetland remote sensing image feature vector; The first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and the second principal component pre-aligned local wetland remote sensing image feature vector are intrinsically fine-grained aligned and fused to obtain the multi-scale local wetland remote sensing image feature vector.

2. The wetland carbon storage estimation method based on remote sensing technology according to claim 1 is characterized in that: After segmenting the remote sensing image of the wetland area to obtain a set of local wetland remote sensing images, extracting wetland remote sensing local area features from the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps, including: Segmenting the remote sensing image of the wetland area to obtain a set of local wetland remote sensing images; The set of local wetland remote sensing images is respectively input into a wetland remote sensing local area feature extraction module based on a fully convolutional neural network to obtain a set of local wetland remote sensing image feature maps.

3. The wetland carbon storage estimation method based on remote sensing technology according to claim 2 is characterized in that: The method of performing wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps includes: The set of local wetland remote sensing image feature maps is input into a wetland remote sensing local image semantic feature association extractor based on Swin Transformer to obtain a set of local wetland remote sensing image context semantic association feature maps.

4. The wetland carbon storage estimation method based on remote sensing technology according to claim 3 is characterized in that: The set of multi-scale local wetland remote sensing image feature maps is feature enhanced to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors, including: inputting the set of multi-scale local wetland remote sensing image feature maps into a bidirectional attention mechanism model to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors.

5. The wetland carbon storage estimation method based on remote sensing technology according to claim 4 is characterized in that: Based on the set of enhanced multi-scale local wetland remote sensing image feature vectors, the vegetation type and quantity of each area are obtained, including: Inputting the set of enhanced multi-scale local wetland remote sensing image feature vectors into a classifier-based vegetation type determiner to obtain the vegetation type of each area; The set of enhanced multi-scale local wetland remote sensing image feature vectors is input into a decoder-based vegetation quantity estimator to obtain the vegetation quantity in each area.

6. The wetland carbon storage estimation method based on remote sensing technology according to claim 5 is characterized in that: Based on the vegetation type and quantity in each area, the estimated carbon storage value of the wetland is obtained, including: The vegetation type and quantity of each area are input into the carbon storage estimation model to obtain the carbon storage estimation value of the wetland.

7. A wetland carbon storage estimation system based on remote sensing technology, characterized in that: include: A remote sensing image acquisition module for wetland areas, used to acquire remote sensing images of wetland areas through remote sensing sensors; A wetland remote sensing image segmentation feature extraction module is used to segment the remote sensing image of the wetland area to obtain a set of local wetland remote sensing images, and then extract wetland remote sensing local area features from the set of local wetland remote sensing images to obtain a set of local wetland remote sensing image feature maps; A wetland remote sensing local area semantic association module is used to perform wetland remote sensing local image semantic feature association on the set of local wetland remote sensing image feature maps to obtain a set of local wetland remote sensing image context semantic association feature maps; A multi-scale local wetland remote sensing image feature fusion module is used to fuse each corresponding group of local wetland remote sensing image context semantic association feature maps and local wetland remote sensing image feature map in the set of local wetland remote sensing image context semantic association feature maps to obtain a set of multi-scale local wetland remote sensing image feature maps; An image feature enhancement module, used for performing feature enhancement on the set of multi-scale local wetland remote sensing image feature maps to obtain a set of enhanced multi-scale local wetland remote sensing image feature vectors; A vegetation type and quantity generation module, used to obtain the vegetation type and quantity of each area based on the set of enhanced multi-scale local wetland remote sensing image feature vectors; A carbon storage estimation module, used to obtain an estimated carbon storage value of the wetland based on the vegetation type and quantity of each area; The method comprises fusing the set of the local wetland remote sensing image context semantic association feature maps and each corresponding group of the local wetland remote sensing image context semantic association feature maps and the local wetland remote sensing image feature maps in the set of the local wetland remote sensing image feature maps to obtain a set of multi-scale local wetland remote sensing image feature maps, including: Expanding the local wetland remote sensing image context semantic association feature map and the local wetland remote sensing image feature map to obtain a local wetland remote sensing image context semantic association feature vector and a local wetland remote sensing image feature vector; Performing feature intrinsic alignment based on essential feature recognition on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a multi-scale local wetland remote sensing image feature vector; Reconstructing the dimension of the multi-scale local wetland remote sensing image feature vector to obtain the multi-scale local wetland remote sensing image feature map; The method of performing feature intrinsic alignment based on essential feature recognition on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a multi-scale local wetland remote sensing image feature vector includes: Performing normalization modulation based on vector norming on the local wetland remote sensing image context semantic association feature vector and the local wetland remote sensing image feature vector to obtain a pre-aligned local wetland remote sensing image context semantic association feature vector and a pre-aligned local wetland remote sensing image feature vector; Performing essential feature recognition on the pre-aligned local wetland remote sensing image context semantic association feature vector and the pre-aligned local wetland remote sensing image feature vector to obtain a first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and a second principal component pre-aligned local wetland remote sensing image feature vector; The first principal component pre-aligned local wetland remote sensing image context semantic association feature vector and the second principal component pre-aligned local wetland remote sensing image feature vector are intrinsically fine-grained aligned and fused to obtain the multi-scale local wetland remote sensing image feature vector.

8. The wetland carbon storage estimation system based on remote sensing technology according to claim 7 is characterized in that: The wetland remote sensing image segmentation feature extraction module is used to: Segmenting the remote sensing image of the wetland area to obtain a set of local wetland remote sensing images; The set of local wetland remote sensing images is respectively input into a wetland remote sensing local area feature extraction module based on a fully convolutional neural network to obtain a set of local wetland remote sensing image feature maps.

Images