A method and system for identifying spartina alterniflora based on multi-source remote sensing images

Abstract

The application relates to a mutual flower rice identification method and system based on multi-source remote sensing images, and relates to the technical field of plant identification. The method comprises the following steps: acquiring pretreated hyperspectral image data and radar image data of a monitoring area, respectively performing optical feature and radar feature extraction, generating a multi-dimensional feature data set, determining a target sample data set based on the hyperspectral image data, the multi-dimensional feature data set, a preset random forest algorithm and a preset water body masking algorithm, wherein the target sample data set comprises a training sample subset and a non-water body feature data set; training a pre-constructed multi-scale residual network based on the target sample data set to obtain a mutual flower rice identification model, so that the mutual flower rice identification model can identify mutual flower rice in the monitoring area. The application achieves the technical effect of effectively identifying mutual flower rice by using multi-source remote sensing images in combination with various algorithms and models, and improving identification precision and reliability.

Description

Technical Field

[0001] This invention relates to the field of plant identification technology, and in particular to a method and system for identifying Spartina alterniflora based on multi-source remote sensing imagery. Background Technology

[0002] Spartina alterniflora is a typical invasive alien plant with extremely strong reproductive capacity and ecological adaptability. It has spread widely in coastal wetlands of my country, seriously threatening the structure and function of local ecosystems. Therefore, effective identification and monitoring of Spartina alterniflora is crucial for protecting local ecosystems.

[0003] In the identification of Spartina alterniflora, remote sensing technology, with its advantages of wide coverage, high timeliness, and non-contact operation, is gradually being widely used in wetland vegetation identification. Hyperspectral imagery, in particular, possesses rich spectral information and can be used for vegetation identification; radar imagery, with its all-weather, all-time imaging capabilities and sensitivity to vegetation structure, can effectively supplement the limitations of optical imagery. Furthermore, traditional classification methods, which were commonly used in the early stages, have played a certain role in processing remote sensing imagery.

[0004] However, hyperspectral imagery is easily affected by weather and tides in complex wetland environments, leading to inaccurate identification results. Traditional classification methods have limited capabilities in handling high-dimensional features and nonlinear relationships, making it difficult to meet the requirements for precise identification of Spartina alterniflora. Furthermore, existing patents for Spartina alterniflora identification primarily focus on single-type remote sensing data and classification techniques, failing to fully consider the advantages of combining hyperspectral and radar data. This makes it difficult to effectively address the complex and ever-changing coastal wetland environment and achieve high-precision, automated identification of Spartina alterniflora. Summary of the Invention

[0005] The technical problem to be solved by the present invention is to provide a method and system for identifying Spartina alterniflora based on multi-source remote sensing images, aiming to solve at least one of the above-mentioned technical problems.

[0006] The technical solution of the present invention to solve the above-mentioned technical problems is as follows: Firstly, this application provides a method for identifying Spartina alterniflora based on multi-source remote sensing imagery, employing the following technical solution: A method for identifying Spartina alterniflora based on multi-source remote sensing imagery includes: Acquire preprocessed hyperspectral and radar image data of the monitoring area; Optical features are extracted from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and radar features are extracted from the radar image data to obtain the polarization features of the monitoring area. Based on the spectral characteristics, vegetation index, and polarization characteristics of the monitored area, a multidimensional feature dataset is generated, which contains multiple feature channels. Based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm and preset water body masking algorithm, a target sample dataset is determined. The target sample dataset includes a training sample subset and a non-water body feature dataset corresponding to the training sample subset. The pre-constructed multi-scale residual network is trained based on the target sample dataset to obtain the Spartina alterniflora identification model, so that the Spartina alterniflora identification model can identify Spartina alterniflora in the monitoring area.

[0007] The beneficial effects of this invention are as follows: By acquiring hyperspectral image data and radar image data, and combining the advantages of both, the spectral features and vegetation indices of the hyperspectral image data can distinguish the subtle spectral differences between Spartina alterniflora and native vegetation, while the polarization features of the radar image data can effectively supplement the deficiencies of optical images; a multi-dimensional feature dataset is generated to more comprehensively reflect the characteristics of the monitoring area; the random forest algorithm is used to determine the target sample dataset, and key features are selected to improve the recognition accuracy; the water body masking algorithm removes water body interference; a multi-scale residual network is trained based on the sample dataset to obtain the recognition model, which can handle image data in complex and variable coastal wetland environments and capture the characteristic changes of Spartina alterniflora under different habitat conditions; finally, a Spartina alterniflora classification map is obtained, realizing high-precision automatic recognition of Spartina alterniflora in complex backgrounds, avoiding the salt-and-pepper phenomenon, and achieving good spatial consistency of classification results, providing technical support for the precise control and dynamic monitoring of the regeneration of Spartina alterniflora.

[0008] Based on the above technical solution, the present invention can be further improved as follows.

[0009] Furthermore, the acquisition of preprocessed hyperspectral image data and radar image data of the monitoring area includes: Acquire raw hyperspectral image data and raw radar image data of the monitored area; The original hyperspectral image data were sequentially subjected to radiometric calibration, atmospheric correction, and orthorectification to obtain a surface reflectance image. The original radar image data is sequentially processed with orbit correction, radiometric calibration, speckle filtering, and terrain correction, and the backscattering intensity value is converted into decibels to obtain the radar backscattering coefficient image. Using the surface reflectance image as a spatial reference, the radar backscattering coefficient image is spatially resampled to make the pixel resolution and spatial position of the radar backscattering coefficient image consistent with those of the surface reflectance image, thus obtaining preprocessed hyperspectral image data and radar image data.

[0010] The beneficial effects of adopting the above-mentioned further scheme are as follows: By performing radiometric calibration, atmospheric correction, and orthorectification on hyperspectral image data to obtain surface reflectivity images, the quality and accuracy of image data can be ensured; by performing orbit correction, radiometric calibration, speckle filtering, topographic correction, and unit conversion on radar image data to obtain radar backscattering coefficient images, the radar images can be made more suitable for subsequent analysis; resampling the radar backscattering coefficient images based on surface reflectivity images can ensure a one-to-one correspondence between the two image pixels, providing an accurate data foundation for subsequent identification of Spartina alterniflora.

[0011] Furthermore, the process of extracting optical features from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and extracting radar features from the radar image data to obtain the polarization features of the monitoring area, includes: Based on the preprocessed hyperspectral image data and the preset principal component analysis method, a set number of principal components are extracted to obtain spectral features; Based on the preprocessed hyperspectral image data, reflectance values ​​of the red, green, blue and near-infrared bands are extracted, and multiple vegetation indices are calculated based on the reflectance values ​​of the red, green, blue and near-infrared bands. Based on the radar image data, the backscattering coefficients of the horizontal transmission and horizontal reception polarization modes and the horizontal transmission and vertical reception polarization modes of the radar image data are extracted, and the ratio between the backscattering coefficients of the horizontal transmission and horizontal reception polarization modes and the horizontal transmission and vertical reception polarization modes is calculated to obtain the polarization characteristics.

[0012] The beneficial effects of adopting the above-mentioned further scheme are as follows: Principal component analysis (PCA) is used to extract principal components from the preprocessed hyperspectral image data, which yields the spectral characteristics of the monitored area and helps to mine the spectral information of *Spartina alterniflora*. Extracting red, green, blue, and near-infrared reflectance values ​​from the preprocessed hyperspectral image data and calculating multiple vegetation indices better reflects the vegetation characteristics of *Spartina alterniflora*. Extracting the backscattering coefficients of two polarization modes from radar image data and calculating their ratio yields polarization features, which supplements the deficiencies of optical images. Combining these features provides a more comprehensive and accurate data foundation for subsequent generation of multidimensional feature datasets and *Spartina alterniflora* identification, improving the accuracy and efficiency of *Spartina alterniflora* identification.

[0013] Furthermore, the determination of the target sample dataset based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm, and preset water body masking algorithm includes: Based on the hyperspectral image data, a true-color image is synthesized, and regions of interest containing Spartina alterniflora and background features are delineated on the true-color image based on visual interpretation. Each region of interest is then labeled with a category to obtain an initial sample dataset. The initial sample dataset is divided into a training sample subset and a validation sample subset. Based on the multidimensional feature dataset and the training sample subset, the feature vector corresponding to the training sample subset is extracted. Based on the feature vectors, a random forest model is trained to obtain the importance of each feature channel in the multidimensional feature dataset. Based on the importance of each feature channel and a preset number of key feature channels, determine the optimal feature subset; A water mask is generated based on the water index NDWI corresponding to the hyperspectral image data. Based on the water body mask, the feature data corresponding to the optimal feature subset is masked to generate the non-water body feature dataset; Based on the training sample subset and the non-water body feature dataset, the target sample dataset is determined.

[0014] The beneficial effects of adopting the above-mentioned further scheme are as follows: by synthesizing true-color images and visually interpreting and delineating regions of interest and labeling categories, an initial sample dataset can be obtained, providing a foundation for subsequent model training; dividing the initial sample dataset and extracting the feature vectors corresponding to the training sample subset can prepare for training the random forest model; training the random forest model to obtain the importance of each feature channel and determine the optimal feature subset can filter out the more critical features for Spartina alterniflora identification, improving the identification accuracy; generating a water body mask and masking the feature data corresponding to the optimal feature subset can remove water body interference, generate a feature data space containing only non-water body land cover types, and combine the training sample subset to determine the target sample dataset, making the data used for training the model more accurate, thereby improving the performance and classification accuracy of the Spartina alterniflora identification model.

[0015] Furthermore, the step of training a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora recognition model includes: Using the feature data of each sample pixel in the training sample subset within the corresponding preset neighborhood window in the non-aquatic feature dataset as input samples, a pre-constructed multi-scale residual network is trained based on a two-stage learning rate reduction strategy. The model parameters with the highest classification accuracy on the validation set during the training process are used as the model parameters of the Spartina alterniflora recognition model, thus obtaining the Spartina alterniflora recognition model. The first stage uses a first learning rate to train up to a first preset number of iterations, and the second stage uses a second learning rate lower than the first learning rate to train up to a second preset number of iterations.

[0016] The beneficial effects of adopting the above-mentioned further scheme are as follows: using the feature data within the preset neighborhood window of the corresponding pixel in the non-aquatic feature dataset as input samples, and training the multi-scale residual network with a two-stage learning rate reduction strategy, the non-aquatic feature data can be fully utilized. At the same time, the different learning rate settings in the two stages can achieve rapid convergence in the early stage of training and finely adjust the model parameters in the later stage. The model parameters with the highest classification accuracy on the validation set can be used as the parameters of the Spartina alterniflora recognition model, which can improve the classification accuracy and precision of the Spartina alterniflora recognition model.

[0017] Furthermore, before training the pre-constructed multi-scale residual network, the process also includes: The input sample is augmented with data, and the data augmentation method includes at least one of horizontal flipping, vertical flipping, and diagonal flipping.

[0018] The beneficial effects of adopting the above-mentioned further scheme are: before training the pre-constructed multi-scale residual network, performing at least one of the following data augmentation methods on the input samples (horizontal flip, vertical flip, and diagonal flip) can increase the number of training samples, avoid model overfitting, and help improve the accuracy and generalization ability of the Spartina alterniflora identification model.

[0019] Furthermore, it also includes: Based on the Spartina alterniflora classification map and the validation sample subset, calculate the confusion matrix; Based on the confusion matrix, the overall classification accuracy and Kappa coefficient are calculated to evaluate the classification accuracy of the Spartina alterniflora identification model.

[0020] The beneficial effects of adopting the above-mentioned further scheme are: by calculating the confusion matrix based on the Spartina alterniflora classification map and the validation sample subset, and then calculating the overall classification accuracy and Kappa coefficient, the classification accuracy of the Spartina alterniflora identification model can be objectively and quantitatively evaluated, which helps to judge the model performance and provides a basis for the optimization and improvement of the model.

[0021] Secondly, this application provides a system for identifying Spartina alterniflora based on multi-source remote sensing imagery, employing the following technical solution: A system for identifying Spartina alterniflora based on multi-source remote sensing imagery, comprising: The acquisition module is used to acquire preprocessed hyperspectral image data and radar image data of the monitoring area; The feature extraction module is used to extract optical features from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and to extract radar features from the radar image data to obtain the polarization features of the monitoring area. A multidimensional feature generation module is used to generate a multidimensional feature dataset based on the spectral features, vegetation index, and polarization features of the monitored area. The multidimensional feature dataset contains multiple feature channels. The sample determination module is used to determine the target sample dataset based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm and preset water body masking algorithm. The target sample dataset includes a training sample subset and a non-water body feature dataset corresponding to the training sample subset. The training module is used to train a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora identification model, so that the Spartina alterniflora identification model can identify Spartina alterniflora in the monitoring area.

[0022] Thirdly, this application provides an electronic device that adopts the following technical solution: An electronic device includes a memory and a processor, wherein the memory stores a computer program that can be loaded by the processor and executed as described in any of the first aspects of the Spartina alterniflora identification method based on multi-source remote sensing imagery.

[0023] Fourthly, this application provides a computer-readable storage medium, which adopts the following technical solution: A computer-readable storage medium storing a computer program capable of being loaded by a processor and executing the method for identifying Spartina alterniflora based on multi-source remote sensing imagery as described in any of the first aspects.

[0024] Additional aspects and advantages of this application will be set forth in part in the description which follows, and will become apparent from the description or may be learned by practice of this application. Attached Figure Description

[0025] Figure 1 This is a flowchart illustrating a method for identifying Spartina alterniflora based on multi-source remote sensing imagery, provided in one embodiment of the present invention. Figure 2 This is a schematic diagram of the structure of a Spartina alterniflora identification system based on multi-source remote sensing imagery, provided in one embodiment of the present invention. Figure 3 This is a schematic diagram of the structure of an electronic device provided in one embodiment of the present invention. Detailed Implementation

[0026] To make the objectives, technical solutions, and advantages of the embodiments of this application clearer, the technical solutions of the embodiments of this application will be clearly and completely described below with reference to the accompanying drawings. Obviously, the described embodiments are only some embodiments of this application, not all embodiments. Based on the embodiments of this application, all other embodiments obtained by those skilled in the art without creative effort are within the scope of protection of this application.

[0027] Furthermore, the term "and / or" in this article is merely a description of the relationship between related objects, indicating that three relationships can exist. For example, A and / or B can represent: A existing alone, A and B existing simultaneously, or B existing alone. Additionally, the character " / " in this article, unless otherwise specified, generally indicates that the preceding and following related objects have an "or" relationship.

[0028] This application provides a method for identifying Spartina alterniflora based on multi-source remote sensing imagery. This method can be executed by an electronic device, which can be a server or a mobile terminal device. The server can be an independent physical server, a server cluster or distributed system composed of multiple physical servers, or a cloud server that provides cloud computing services. The mobile terminal device can be a laptop computer, a desktop computer, etc., but is not limited to these.

[0029] like Figure 1 As shown, a method for identifying Spartina alterniflora based on multi-source remote sensing imagery mainly includes: S1, acquire preprocessed hyperspectral image data and radar image data of the monitoring area; In this embodiment of the application, specifically, S1 includes: Acquire raw hyperspectral image data and raw radar image data of the monitored area; The original hyperspectral image data were sequentially subjected to radiometric calibration, atmospheric correction, and orthorectification to obtain a surface reflectance image. The original radar image data is sequentially processed with orbit correction, radiometric calibration, speckle filtering, and terrain correction, and the backscattering intensity value is converted into decibels to obtain the radar backscattering coefficient image. Using the surface reflectance image as a spatial reference, the radar backscattering coefficient image is spatially resampled to make the pixel resolution and spatial position of the radar backscattering coefficient image consistent with those of the surface reflectance image, thus obtaining preprocessed hyperspectral image data and radar image data.

[0030] In the above embodiments, the raw hyperspectral image data can be obtained from professional remote sensing satellite data providers, such as data collected by the Gaofen-5 satellite. The raw radar image data can be Sentinel-1 radar imagery from the same period.

[0031] For raw hyperspectral image data, radiometric calibration, atmospheric correction, and orthorectification are performed sequentially. Radiometric calibration converts the raw digital quantization values ​​recorded by the sensor into physically meaningful radiance values, using specialized radiometric calibration software; different sensors have corresponding calibration coefficients. Atmospheric correction eliminates the influence of the atmosphere on spectral information; commonly used atmospheric correction methods include FLAASH and 6S. Orthorectification uses a high-precision digital elevation model (DEM) to geometrically correct the image, ensuring that the image's location accurately corresponds to its actual geographic coordinates, thus obtaining a surface reflectance image.

[0032] For raw radar image data, the following processes are performed sequentially: orbit correction, radiometric calibration, speckle filtering, and terrain correction, converting backscattering intensity values ​​to decibels. Orbit correction ensures accurate orbital information for the image, using orbital parameter files provided by the satellite. Radiometric calibration converts the raw radar image data into physically meaningful backscattering coefficients. Speckle filtering reduces speckle noise in the radar image; common filtering methods include Lee filtering and Gamma filtering. Terrain correction eliminates the influence of terrain undulations on the radar backscattering coefficients, using DEM data for correction. Finally, the backscattering intensity values ​​are converted to decibels to obtain the radar backscattering coefficient image. Using the surface reflectance image as a spatial reference, the radar backscattering coefficient image is spatially resampled. A cubic convolution resampling method can be used to ensure that the pixel resolution of the radar backscattering coefficient image is consistent with its spatial location and the surface reflectance image, resulting in preprocessed hyperspectral image data and radar image data.

[0033] S2, perform optical feature extraction on the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and perform radar feature extraction on the radar image data to obtain the polarization features of the monitoring area. In this embodiment of the application, the steps of extracting optical features from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and extracting radar features from the radar image data to obtain the polarization features of the monitoring area, include: Based on the preprocessed hyperspectral image data and the preset principal component analysis method, a set number of principal components are extracted to obtain spectral features; Based on the preprocessed hyperspectral image data, reflectance values ​​of the red, green, blue and near-infrared bands are extracted, and multiple vegetation indices are calculated based on the reflectance values ​​of the red, green, blue and near-infrared bands. Based on the radar image data, the backscattering coefficients of the horizontal transmission and horizontal reception polarization modes and the horizontal transmission and vertical reception polarization modes of the radar image data are extracted, and the ratio between the backscattering coefficients of the horizontal transmission and horizontal reception polarization modes and the horizontal transmission and vertical reception polarization modes is calculated to obtain the polarization characteristics.

[0034] In the above implementation, for the preprocessed Sentinel-1 radar image, two different polarization data are obtained: horizontal-horizontal polarization (VV) and horizontal-vertical polarization (VH). VH / VV is calculated from VH and VV, and VV, VH, and VH / VV are used as feature variables.

[0035] For the preprocessed Gaofen-5 image, principal component analysis (PCA) was used to extract three principal components (PCA1, PCA2, PCA3). Principal component analysis can reduce the dimensionality of high-dimensional spectral data, reduce data redundancy, and extract the principal components that best represent the data characteristics.

[0036] Based on preprocessed hyperspectral image data, reflectance values ​​for the red, green, blue, and near-infrared bands are extracted. These bands can be determined according to the center wavelength of the hyperspectral image, such as red (B55), green (B35), blue (B20), and near-infrared (B90). Based on these reflectance values, various vegetation indices are calculated, such as NDVI, RVI, GNDVI, NDVIre, CI, MCARI, MTCI, NDWI, PVI, CI, and BSI, all vegetation-related indices.

[0037] S3. Based on the spectral characteristics, vegetation index, and polarization characteristics of the monitored area, a multidimensional feature dataset is generated, which contains multiple feature channels. In this embodiment, spectral features, vegetation indices, and polarization features are stacked to generate a multidimensional feature dataset.

[0038] S4. Based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm and preset water body masking algorithm, determine the target sample dataset, which includes a training sample subset and a non-water body feature dataset corresponding to the training sample subset. In this embodiment of the application, specifically, S4 includes the following sub-steps: S41. Based on the hyperspectral image data, a true-color image is synthesized, and based on visual interpretation, regions of interest containing Spartina alterniflora and background features are delineated on the true-color image, and each region of interest is labeled with a category to obtain an initial sample dataset. S42, the initial sample dataset is divided into a training sample subset and a validation sample subset, and the feature vector corresponding to the training sample subset is extracted based on the multidimensional feature dataset and the training sample subset; S43, Based on the feature vector, train a random forest model to obtain the importance of each feature channel in the multidimensional feature dataset; S44, determine the optimal feature subset based on the importance of each feature channel and a preset number of key feature channels; S45, Generate a water mask based on the water index NDWI corresponding to the hyperspectral image data; S46, Based on the water body mask, the feature data corresponding to the optimal feature subset is masked to generate the non-water body feature dataset; S47. Based on the training sample subset and the non-water body feature dataset, determine the target sample dataset.

[0039] In the above implementation, the red, green, and blue bands of the hyperspectral image are combined to generate a true-color image. Regions of interest containing Spartina alterniflora and background features are delineated on the true-color image based on visual interpretation, for example, using ArcGIS software, by manually identifying and delineating the areas containing Spartina alterniflora and other features. Each region of interest is then labeled with its category to obtain an initial sample dataset. This initial sample dataset is then divided into a training sample subset and a validation sample subset in a 7:3 ratio.

[0040] Based on the multidimensional feature dataset and a subset of training samples, feature vectors corresponding to the subset of training samples are extracted. Based on these feature vectors, a random forest model is trained. The random forest model is an ensemble learning model composed of multiple decision trees. By training the random forest model, the importance of each feature channel in the multidimensional feature dataset can be obtained. The feature importance value typically ranges from 0 to 1; a larger value indicates a greater contribution of the feature to the model.

[0041] Based on the importance of each feature channel and a preset number of key feature channels, determine the optimal feature subset, such as selecting the top 8 feature channels with the highest importance values, including: PCA1, B55, NDVI, RVI, GNDVI, MCARI, VH, and VV.

[0042] The Water Body Index (NDWI) is calculated based on RGB image data. Pixels with an NDWI greater than 0 are classified as water bodies. This NDWI is used as a mask to mask the feature data within the multidimensional feature space in step four, generating a feature data space that contains only non-water body land cover types. The NDWI formula is as follows: ; In the formula, g and nir represent the green band and near-infrared band of the hyperspectral image, respectively.

[0043] S5. Based on the target sample dataset, a pre-constructed multi-scale residual network is trained to obtain a Spartina alterniflora identification model, so that the Spartina alterniflora identification model can identify Spartina alterniflora in the monitoring area.

[0044] In this embodiment of the application, the step of training a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora recognition model includes: Using the feature data of each sample pixel in the training sample subset within the corresponding preset neighborhood window in the non-aquatic feature dataset as input samples, a pre-constructed multi-scale residual network is trained based on a two-stage learning rate reduction strategy. The model parameters with the highest classification accuracy on the validation set during the training process are used as the model parameters of the Spartina alterniflora recognition model, thus obtaining the Spartina alterniflora recognition model.

[0045] The first stage uses a first learning rate to train up to a first preset number of iterations, and the second stage uses a second learning rate lower than the first learning rate to train up to a second preset number of iterations.

[0046] In this embodiment of the application, Spartina alterniflora is classified using a model based on the sample labels and corresponding classification features (PCA1, B55, NDVI, RVI, GNDVI, MCARI, VH, VV) in the training sample subset.

[0047] In this embodiment, the preset neighborhood window can be set to a 5x5 window. A pre-constructed multi-scale residual network is trained based on a two-stage learning rate reduction strategy. The first stage uses a first learning rate and trains to a first preset number of iterations, for example, a learning rate of 10 and 170 iterations. The second stage uses a second learning rate lower than the first learning rate and trains to a second preset number of iterations, for example, a learning rate of 0.1 and 30 iterations.

[0048] The model parameters that achieve the highest classification accuracy on the data validation subset during training are used as the model parameters for the Spartina alterniflora recognition model, thus obtaining the Spartina alterniflora recognition model.

[0049] Before training the pre-built multi-scale residual network, the input samples can be augmented. Data augmentation methods include at least one of horizontal flipping, vertical flipping, and diagonal flipping. This can increase the diversity of training data and improve the generalization ability of the model.

[0050] After training a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora recognition model, the method further includes: Based on the Spartina alterniflora classification map and the validation sample subset, calculate the confusion matrix; Based on the confusion matrix, the overall classification accuracy and Kappa coefficient are calculated to evaluate the classification accuracy of the Spartina alterniflora identification model.

[0051] This method acquires hyperspectral and radar imagery data, combining their advantages. The spectral features and vegetation indices of the hyperspectral imagery data can distinguish subtle spectral differences between Spartina alterniflora and native vegetation, while the polarization features of the radar imagery data can effectively supplement the deficiencies of optical imagery. A multi-dimensional feature dataset is generated to more comprehensively reflect the characteristics of the monitored area. The random forest algorithm is used to determine the target sample dataset, and key features are selected to improve recognition accuracy. The water body masking algorithm removes water body interference. Based on the sample dataset, a multi-scale residual network is trained to obtain the recognition model, which can handle imagery data in complex and variable coastal wetland environments and capture the characteristic changes of Spartina alterniflora under different habitat conditions. Finally, a Spartina alterniflora classification map is obtained, achieving high-precision automatic recognition of Spartina alterniflora in complex backgrounds, avoiding the salt-and-pepper phenomenon, and showing good spatial consistency in the classification results. This provides technical support for the precise control and dynamic monitoring of the resurgence of Spartina alterniflora.

[0052] Figure 2 A schematic diagram of a Spartina alterniflora identification system based on multi-source remote sensing imagery is shown.

[0053] like Figure 2 As shown, a Spartina alterniflora identification system 200 based on multi-source remote sensing imagery mainly includes: The acquisition module 201 is used to acquire preprocessed hyperspectral image data and radar image data of the monitoring area; The feature extraction module 202 is used to extract optical features from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and to extract radar features from the radar image data to obtain the polarization features of the monitoring area. The multidimensional feature generation module 203 is used to generate a multidimensional feature dataset based on the spectral features, vegetation index and polarization features of the monitored area. The multidimensional feature dataset contains multiple feature channels. The sample determination module 204 is used to determine the target sample dataset based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm and preset water body masking algorithm. The target sample dataset includes a training sample subset and a non-water body feature dataset corresponding to the training sample subset. The training module 205 is used to train a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora identification model, so that the Spartina alterniflora identification model can identify Spartina alterniflora in the monitoring area.

[0054] In one example, the module in any of the above devices may be one or more integrated circuits configured to implement the above methods, such as one or more application-specific integrated circuits (ASICs), or one or more digital signal processors (DSPs), or one or more field-programmable gate arrays (FPGAs), or a combination of at least two of these integrated circuit forms.

[0055] For example, when modules in a device can be implemented via a processing element scheduler, the processing element can be a general-purpose processor, such as a central processing unit (CPU) or other processor capable of calling programs. Alternatively, these modules can be integrated together as a system-on-a-chip (SOC).

[0056] In this application, various objects such as messages / information / devices / network elements / systems / apparatus / actions / operations / processes / concepts may be named. It is understood that these specific names do not constitute a limitation on the relevant objects. The names may be changed depending on the scenario, context, or usage habits. The understanding of the technical meaning of the technical terms in this application should be mainly determined from their functions and technical effects embodied / performed in the technical solution.

[0057] Those skilled in the art will understand that, for the sake of convenience and brevity, the specific working processes of the systems, devices, and modules described above can be referred to the corresponding processes in the foregoing method embodiments, and will not be repeated here.

[0058] Those skilled in the art will recognize that the modules and algorithm steps of the various examples described in conjunction with the embodiments disclosed herein can be implemented in electronic hardware, or a combination of computer software and electronic hardware. Whether these functions are implemented in hardware or software depends on the specific application and design constraints of the technical solution. Those skilled in the art can use different methods to implement the described functions for each specific application, but such implementation should not be considered beyond the scope of this application.

[0059] Figure 3 This is a structural block diagram of an electronic device 300 according to an embodiment of this application.

[0060] like Figure 3 As shown, the electronic device 300 includes a processor 301 and a memory 302, and may further include one or more of an information input / output (I / O) interface 303, a communication component 304, and a communication bus 305.

[0061] The processor 301 controls the overall operation of the electronic device 300 to complete all or part of the steps in the above-described method for identifying Spartina alterniflora based on multi-source remote sensing imagery. The memory 302 stores various types of data to support the operation of the electronic device 300. This data may include, for example, instructions for any application or method operating on the electronic device 300, as well as application-related data. The memory 302 can be implemented by any type of volatile or non-volatile storage device or a combination thereof, such as one or more of Static Random Access Memory (SRAM), Electrically Erasable Programmable Read-Only Memory (EEPROM), Erasable Programmable Read-Only Memory (EPROM), Programmable Read-Only Memory (PROM), Read-Only Memory (ROM), magnetic storage, flash memory, magnetic disk, or optical disk.

[0062] I / O interface 303 provides an interface between processor 301 and other interface modules, such as keyboards, mice, and buttons. These buttons can be virtual or physical. Communication component 304 is used to test wired or wireless communication between electronic device 300 and other devices. Wireless communication includes Wi-Fi, Bluetooth, Near Field Communication (NFC), 2G, 3G, or 4G, or a combination thereof. Therefore, the corresponding communication component 304 may include a Wi-Fi component, a Bluetooth component, and an NFC component.

[0063] The communication bus 305 may include a path for transmitting information between the aforementioned components. The communication bus 305 may be a PCI (Peripheral Component Interconnect) bus or an EISA (Extended Industry Standard Architecture) bus, etc. The communication bus 305 may be divided into an address bus, a data bus, a control bus, etc.

[0064] The electronic device 300 can be implemented by one or more application-specific integrated circuits (ASICs), digital signal processors (DSPs), digital signal processing devices (DSPDs), programmable logic devices (PLDs), field-programmable gate arrays (FPGAs), controllers, microcontrollers, microprocessors, or other electronic components to execute the Spartina alterniflora identification method based on multi-source remote sensing imagery given in the above embodiments.

[0065] The following describes the computer-readable storage medium provided in the embodiments of this application. The computer-readable storage medium described below can be referred to in correspondence with the above-described method for identifying Spartina alterniflora based on multi-source remote sensing images.

[0066] This application also provides a computer-readable storage medium storing a computer program, which, when executed by a processor, implements the steps of the above-described method for identifying Spartina alterniflora based on multi-source remote sensing images.

[0067] The computer-readable storage medium may include various media capable of storing program code, such as USB flash drives, portable hard drives, read-only memory (ROM), random access memory (RAM), magnetic disks, or optical disks.

[0068] The terms “comprising,” “including,” or any other variations thereof are intended to cover non-exclusive inclusion, such that a process, method, article, or apparatus that comprises a list of elements includes not only those elements but also other elements not expressly listed, or elements inherent to such process, method, article, or apparatus.

[0069] The above description is merely a preferred embodiment of this application and an explanation of the technical principles employed. Those skilled in the art should understand that the scope of this application is not limited to technical solutions formed by specific combinations of the above-described technical features, but should also cover other technical solutions formed by arbitrary combinations of the above-described technical features or their equivalents without departing from the foregoing application concept. For example, technical solutions formed by substituting the above features with (but not limited to) technical features with similar functions claimed in this application.

Claims

1. A method for identifying Spartina alterniflora based on multi-source remote sensing imagery, characterized in that, include: Acquire preprocessed hyperspectral and radar image data of the monitoring area; Optical features are extracted from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and radar features are extracted from the radar image data to obtain the polarization features of the monitoring area. Based on the spectral characteristics, vegetation index, and polarization characteristics of the monitored area, a multidimensional feature dataset is generated, which contains multiple feature channels. Based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm and preset water body masking algorithm, a target sample dataset is determined. The target sample dataset includes a training sample subset and a non-water body feature dataset corresponding to the training sample subset. The pre-constructed multi-scale residual network is trained based on the target sample dataset to obtain the Spartina alterniflora identification model, so that the Spartina alterniflora identification model can identify Spartina alterniflora in the monitoring area. The method for determining the target sample dataset based on hyperspectral image data, multidimensional feature datasets, a pre-defined random forest algorithm, and a pre-defined water body masking algorithm includes: Based on the hyperspectral image data, a true-color image is synthesized, and regions of interest containing Spartina alterniflora and background features are delineated on the true-color image based on visual interpretation. Each region of interest is then labeled with a category to obtain an initial sample dataset. The initial sample dataset is divided into a training sample subset and a validation sample subset. Based on the multidimensional feature dataset and the training sample subset, the feature vector corresponding to the training sample subset is extracted. Based on the feature vectors, a random forest model is trained to obtain the importance of each feature channel in the multidimensional feature dataset. Based on the importance of each feature channel and a preset number of key feature channels, determine the optimal feature subset; A water mask is generated based on the water index NDWI corresponding to the hyperspectral image data. Based on the water body mask, the feature data corresponding to the optimal feature subset is masked to generate the non-water body feature dataset; Based on the training sample subset and the non-water body feature dataset, the target sample dataset is determined; The step of training a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora recognition model includes: Using the feature data of each sample pixel in the training sample subset within the corresponding preset neighborhood window in the non-aquatic feature dataset as input samples, a pre-constructed multi-scale residual network is trained based on a two-stage learning rate reduction strategy. The model parameters with the highest classification accuracy on the validation set during the training process are used as the model parameters of the Spartina alterniflora recognition model, thus obtaining the Spartina alterniflora recognition model. The first stage uses a first learning rate to train up to a first preset number of iterations, and the second stage uses a second learning rate lower than the first learning rate to train up to a second preset number of iterations.

2. The method for identifying Spartina alterniflora based on multi-source remote sensing imagery according to claim 1, characterized in that, The acquisition of preprocessed hyperspectral image data and radar image data of the monitoring area includes: Acquire raw hyperspectral image data and raw radar image data of the monitored area; The original hyperspectral image data were sequentially subjected to radiometric calibration, atmospheric correction, and orthorectification to obtain a surface reflectance image. The original radar image data is sequentially processed with orbit correction, radiometric calibration, speckle filtering, and terrain correction, and the backscattering intensity value is converted into decibels to obtain the radar backscattering coefficient image. Using the surface reflectance image as a spatial reference, the radar backscattering coefficient image is spatially resampled to make the pixel resolution and spatial position of the radar backscattering coefficient image consistent with those of the surface reflectance image, thus obtaining preprocessed hyperspectral image data and radar image data.

3. The method for identifying Spartina alterniflora based on multi-source remote sensing imagery according to claim 1, characterized in that, The process of extracting optical features from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and extracting radar features from the radar image data to obtain the polarization features of the monitoring area, includes: Based on the preprocessed hyperspectral image data and the preset principal component analysis method, a set number of principal components are extracted to obtain spectral features; Based on the preprocessed hyperspectral image data, reflectance values ​​of the red, green, blue and near-infrared bands are extracted, and multiple vegetation indices are calculated based on the reflectance values ​​of the red, green, blue and near-infrared bands. Based on the radar image data, the backscattering coefficients of the horizontal transmission and horizontal reception polarization modes and the horizontal transmission and vertical reception polarization modes of the radar image data are extracted, and the ratio between the backscattering coefficients of the horizontal transmission and horizontal reception polarization modes and the horizontal transmission and vertical reception polarization modes is calculated to obtain the polarization characteristics.

4. The method for identifying Spartina alterniflora based on multi-source remote sensing imagery according to claim 1, characterized in that, Before training the pre-built multi-scale residual network, the following is also included: The input sample is augmented with data, and the data augmentation method includes at least one of horizontal flipping, vertical flipping, and diagonal flipping.

5. The method for identifying Spartina alterniflora based on multi-source remote sensing imagery according to claim 1, characterized in that, Also includes: Based on the Spartina alterniflora classification map and the aforementioned subset of validation samples, a confusion matrix is ​​calculated. Based on the confusion matrix, the overall classification accuracy and Kappa coefficient are calculated to evaluate the classification accuracy of the Spartina alterniflora identification model.

6. A system for identifying Spartina alterniflora based on multi-source remote sensing imagery, characterized in that, include: The acquisition module is used to acquire preprocessed hyperspectral image data and radar image data of the monitoring area; The feature extraction module is used to extract optical features from the hyperspectral image data to obtain the spectral features and vegetation index of the monitoring area, and to extract radar features from the radar image data to obtain the polarization features of the monitoring area. A multidimensional feature generation module is used to generate a multidimensional feature dataset based on the spectral features, vegetation index, and polarization features of the monitored area. The multidimensional feature dataset contains multiple feature channels. The sample determination module is used to determine the target sample dataset based on hyperspectral image data, multidimensional feature dataset, preset random forest algorithm and preset water body masking algorithm. The target sample dataset includes a training sample subset and a non-water body feature dataset corresponding to the training sample subset. The training module is used to train a pre-constructed multi-scale residual network based on the target sample dataset to obtain a Spartina alterniflora identification model, so that the Spartina alterniflora identification model can identify Spartina alterniflora in the monitoring area. The sample determination module is specifically used for: Based on the hyperspectral image data, a true-color image is synthesized, and regions of interest containing Spartina alterniflora and background features are delineated on the true-color image based on visual interpretation. Each region of interest is then labeled with a category to obtain an initial sample dataset. The initial sample dataset is divided into a training sample subset and a validation sample subset. Based on the multidimensional feature dataset and the training sample subset, the feature vector corresponding to the training sample subset is extracted. Based on the feature vectors, a random forest model is trained to obtain the importance of each feature channel in the multidimensional feature dataset. Based on the importance of each feature channel and a preset number of key feature channels, determine the optimal feature subset; A water mask is generated based on the water index NDWI corresponding to the hyperspectral image data. Based on the water body mask, the feature data corresponding to the optimal feature subset is masked to generate the non-water body feature dataset; Based on the training sample subset and the non-water body feature dataset, the target sample dataset is determined; The training module is specifically used for: Using the feature data of each sample pixel in the training sample subset within the corresponding preset neighborhood window in the non-aquatic feature dataset as input samples, a pre-constructed multi-scale residual network is trained based on a two-stage learning rate reduction strategy. The model parameters with the highest classification accuracy on the validation set during the training process are used as the model parameters of the Spartina alterniflora recognition model, thus obtaining the Spartina alterniflora recognition model. The first stage uses a first learning rate to train up to a first preset number of iterations, and the second stage uses a second learning rate lower than the first learning rate to train up to a second preset number of iterations.

7. An electronic device, characterized in that, Includes a processor, which is coupled to a memory; The processor is configured to execute a computer program stored in the memory, causing the electronic device to perform the method as described in any one of claims 1 to 5.

8. A computer-readable storage medium, characterized in that, Includes a computer program or instructions that, when run on a computer, cause the computer to perform the method as described in any one of claims 1-5.

Images