Heterogeneous time-series image wetland monitoring method and device based on improved cascade forest

Abstract

The present invention proposes a heterogeneous remote sensing image wetland monitoring method and device based on improved cascade forest, which relates to the application field of remote sensing technology, and obtains time-series SAR images of target wetland areas; pre-processes the time-series SAR images to obtain registered SAR images; performs time-series fusion of the registered SAR images based on a feature dimension parallel network to obtain fused SAR images; inputs the fused SAR images into a sliding window scanning model to obtain the spatiotemporal polarization feature set corresponding to the fused SAR images; constructs an initial cascade forest model based on multiple completely random forests and ordinary random forests, and inputs the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model; inputs the fused SAR images into the cascade forest model to obtain the wetland object classification result of the target wetland area. The present invention is helpful to achieve high-precision classification of wetland objects.

Description

Technical Field

[0001] The present invention relates to the field of remote sensing technology application technology, and in particular to a heterogeneous time-series image wetland monitoring method and device based on improved cascade forest. Background Art

[0002] Remote sensing technology has been widely used in many fields such as environmental monitoring, land use planning and disaster prediction. Especially in wetland monitoring, the high efficiency, wide range and rich data levels it provides make it an irreplaceable data source. Wetland is one of the important ecosystems in the world. It has important ecological functions such as conserving water resources, purifying water quality, maintaining biodiversity, storing floods and preventing droughts, regulating climate and fixing carbon. It plays a key role in global ecological balance and human life. Therefore, high-quality, long-term and continuous wetland classification and mapping are crucial to understanding habitat areas, landscape pattern evolution, and the impact of natural and human disturbances on wetland ecosystems.

[0003] The Chinese patent with the publication number CN112884029B discloses a collaborative classification method integrating full polarimetric SAR and hyperspectral remote sensing, including: polarimetric decomposition and texture feature calculation of synthetic aperture radar (SAR) in the same area, resampling hyperspectral and SAR images to the same resolution; converting sample vector files into raster data and extracting labels as training samples for classifiers; calling machine learning methods such as maximum likelihood (ML), Mahalanobis distance (MD), and support vector machine (SVM) to train classifier models on image training data sets; stretching the resampled images into different one-dimensional feature vectors, and then combining them into a prediction data set; calling the previously trained model to predict label values. However, the above scheme does not make sufficient use of temporal information and spatial information when classifying wetlands, and has limited processing capabilities for large and complex data, resulting in reduced accuracy in wetland object classification. Therefore, it is very necessary to provide a heterogeneous temporal image wetland monitoring method and device based on improved cascade forests to achieve high-precision classification of wetland objects. Summary of the invention

[0004] In view of this, the present invention proposes a heterogeneous time-series image wetland monitoring method and device based on an improved cascade forest. The acquired time-series SAR images are fused through preprocessing and feature dimension parallel network, which can effectively remove the noise in the time-series SAR images and complete the registration. At the same time, the sliding window scanning method is applied to obtain the spatiotemporal polarization feature set, and the cascade forest is used as a classifier, which can make full use of the rich spatiotemporal polarization information of the fused image, thereby achieving high-precision classification of wetland objects.

[0005] The present invention provides a method for improving heterogeneous time-series image wetland monitoring of cascade forests, the method comprising:

[0006] Acquire time-series SAR images of the target wetland area;

[0007] Preprocessing the time-series SAR images to obtain registered SAR images;

[0008] Based on a feature dimension parallel network, the registered SAR images are temporally fused to obtain a fused SAR image;

[0009] Inputting the fused SAR image into a sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image;

[0010] Constructing an initial cascade forest model and inputting the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model;

[0011] The fused SAR image is input into the cascade forest model to obtain the wetland feature classification result of the target wetland area.

[0012] On the basis of the above technical solution, preferably, the preprocessing of the time-series SAR images to obtain the registered SAR images specifically includes:

[0013] Performing multi-view processing on the time-series SAR image to convert single-channel complex data in the time-series SAR image into single-channel intensity data;

[0014] Based on the refined Lee filter, the single-channel intensity data is subjected to a speckle noise removal operation to obtain a corrected SAR image;

[0015] The corrected SAR image is registered based on a digital elevation model to obtain the registered SAR image.

[0016] On the basis of the above technical solution, preferably, the characteristic dimension parallel network specifically includes:

[0017]

[0018]

[0019] Among them, F N represents the fused SAR image, Indicates the parallel symbol, F tn represents the polarimetric SAR data feature set for the nth year, represents the registered SAR image, A represents the length of the registered SAR image, B represents the width of the registered SAR image, and N represents the amount of time series information obtained by the registered SAR image.

[0020] More preferably, the fused SAR image is input into a sliding window scanning model to obtain the spatiotemporal polarization feature set corresponding to the fused SAR image, specifically comprising:

[0021] λ=ω×ω

[0022]

[0023] Among them, int() represents the direct rounding function, ω represents the length of the sliding window, and λ represents the area of ​​the sliding window. represents the feature vector of the pixel in the fused SAR image, represents the spatial feature vector of the central pixel in the sliding window, i represents the horizontal coordinate of the pixel in the fused SAR image, j represents the vertical coordinate of the pixel in the fused SAR image, represents a series symbol, and N represents the number of time series information obtained by the registered SAR images.

[0024] Further preferably, constructing an initial cascade forest model based on multiple completely random forests and ordinary random forests, and inputting the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model, specifically includes:

[0025] Constructing the initial cascade forest model based on multiple completely random forests and common random forests, wherein each cascade layer in the initial cascade forest model includes two random forests and two completely random forests;

[0026] Extracting several features from the spatiotemporal polarization feature set as candidate features and inputting them into the initial cascade forest model;

[0027] Determine whether the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same;

[0028] If the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same, the training of the initial cascade forest model is completed and the cascade forest model is obtained.

[0029] More preferably, the initial cascade forest model is constructed based on multiple completely random forests and common random forests, specifically including:

[0030]

[0031] in, represents the predicted class vector of a completely random forest, represents the class vector of another fully random forest prediction, represents the predicted class vector of a normal random forest, represents the class vector of another ordinary random forest prediction, Indicates the concatenation symbol, {p k (x ij )|k=1,2...C} represents any completely random forest or ordinary random forest for x ij The probability of predicting the kth class, X N is the candidate feature initially input to the cascade layer, and J() represents the cascade operation.

[0032] More preferably, the initial cascade forest model specifically includes:

[0033]

[0034] X m = J(X m-1 )

[0035] predict=Max(Ave(X m ))

[0036] Wherein, X1 represents the vector obtained by concatenating the class vector output after the first layer of cascade operation and the candidate feature, X2 represents the vector obtained by concatenating the class vector output after the second layer of cascade operation and the candidate feature, and X n represents the vector of the class vector output after the n-th layer cascade operation and the candidate feature concatenation, X m Represents the final class vector for prediction, predict represents the output category label, Max() represents the maximum value function, and Ave() represents the average value function.

[0037] In a second aspect of the present application, a heterogeneous time-series remote sensing image wetland monitoring device based on an improved cascade forest is provided, wherein the heterogeneous time-series remote sensing image wetland monitoring device comprises a data acquisition module, a data processing module, a time-series fusion module, a feature scanning module and a training processing module, wherein:

[0038] The data acquisition module is used to acquire time-series SAR images of the target wetland area;

[0039] The data processing module is used to pre-process the time-series SAR images to obtain registered SAR images;

[0040] The temporal fusion module is used to perform temporal fusion on the registered SAR images based on a feature dimension parallel network to obtain a fused SAR image;

[0041] The feature scanning module is used to input the fused SAR image into a sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image;

[0042] The training processing module is used to construct an initial cascade forest model and input the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model, input the fused SAR image into the cascade forest model, and obtain the wetland feature classification result of the target wetland area.

[0043] In a third aspect of the present application, an electronic device is provided, comprising a processor, a memory, a user interface and a network interface, wherein the memory is used to store instructions, the user interface and the network interface are used to communicate with other devices, and the processor is used to execute the instructions stored in the memory.

[0044] In a fourth aspect of the present application, a computer-readable storage medium is provided, on which a computer program is stored, and the computer program is executed by a processor to implement the steps of an improved heterogeneous time-series image wetland monitoring method for cascade forests.

[0045] The improved cascade forest heterogeneous time-series image wetland monitoring method and device of the present invention have the following beneficial effects compared with the prior art:

[0046] (1) The fusion of the acquired time-series SAR images through preprocessing and feature dimension parallel network can effectively remove the noise in the time-series SAR images and complete the registration. At the same time, the sliding window scanning method is used to obtain the spatiotemporal polarization feature set, and a relatively complete spatiotemporal polarization multidimensional feature space is constructed. In addition, the cascade forest is used as a classifier, which can adaptively and complexly process the huge input data and optimize the internal rules, so as to make full use of the rich spatiotemporal polarization information of the fused image to achieve higher accuracy, thereby realizing high-precision classification of wetland objects;

[0047] (2) By constructing a cascade forest with two random forests and two completely random forests in each cascade layer to process the complex and large spatiotemporal polarization multidimensional feature set, the cascade forest can be applied to the wetland feature classification of single-polarization and dual-polarization SAR data, thus avoiding the tedious process of manual feature extraction and complex parameter adjustment in traditional methods, and constructing a relatively complete spatiotemporal polarization multidimensional feature space. At the same time, the cascade forest can adapt to complexity, fully mine and utilize the fused image information, and achieve high-precision classification of wetland features. It can still show excellent classification effect even when the data samples are limited, has significant robustness, and can be widely used in wetland feature classification scenarios under different situations, and has portability. BRIEF DESCRIPTION OF THE DRAWINGS

[0048] In order to more clearly illustrate the embodiments of the present invention or the technical solutions in the prior art, the drawings required for use in the embodiments or the description of the prior art will be briefly introduced below. Obviously, the drawings described below are only some embodiments of the present invention. For ordinary technicians in this field, other drawings can be obtained based on these drawings without paying creative work.

[0049] Figure 1 A schematic diagram of the steps of an improved cascade forest heterogeneous time-series image wetland monitoring method provided by the present invention;

[0050] Figure 2 A schematic diagram of a model framework of an improved cascade forest heterogeneous time-series image wetland monitoring method provided by an embodiment of the present invention;

[0051] Figure 3 A diagram of classification results under different time series information and classifiers provided by an embodiment of the present invention;

[0052] Figure 4 A schematic diagram of the structure of a heterogeneous time-series remote sensing image wetland monitoring device provided by an embodiment of the present invention;

[0053] Figure 5 A schematic diagram of the structure of an electronic device provided by an embodiment of the present invention.

[0054] Explanation of the accompanying symbols: 1. Heterogeneous time-series remote sensing image wetland monitoring device; 11. Data acquisition module; 12. Data processing module; 13. Time-series fusion module; 14. Feature scanning module; 15. Training processing module; 2. Electronic device; 21. Processor; 22. Communication bus; 23. User interface; 24. Network interface; 25. Memory. DETAILED DESCRIPTION

[0055] The following will be combined with the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0056] Before introducing the embodiments of the present invention, some terms involved in the embodiments of the present invention are first defined and explained.

[0057] Multi-look: SAR multi-look processing averages the SLC data in azimuth and / or range, and the result is multi-look intensity data.

[0058] Number of views: In a SAR processor, a set of signal samples that divides the entire synthetic aperture into several subapertures, each of which represents an independent observation of the same scene.

[0059] Multi-view processing: In essence, it averages the resolution of the image in the distance and azimuth directions, that is, synthesizing several (number of views) single views into one (multi-view).

[0060] The present application embodiment discloses an improved cascade forest heterogeneous time-series image wetland monitoring method, such as Figure 1 and Figure 2 As shown, the method comprises steps S1 to S6.

[0061] Step S1, obtaining time-series SAR images of the target wetland area.

[0062] In this step, sample selection and classification system can be completed according to the actual situation of the wetland study area. The classification system can be based on wetland classification standards such as the Convention on Wetlands and the National Wetland Resources Survey and Monitoring Technical Regulations. At the same time, training samples and test samples are selected based on the principles of randomness and uniformity.

[0063] Step S2, preprocessing the time-series SAR images to obtain registered SAR images.

[0064] In this step, step S2 also includes steps S21 to S23.

[0065] Step S21 , performing multi-view processing on the time-series SAR images to convert single-channel complex data in the time-series SAR images into single-channel intensity data.

[0066] In this step, the range resolution (pixelspacingslantrange), azimuth resolution (pixelspacingazimuth), and incidence angle (incidenceangle) data are used to calculate the number of views, or the number of views is calculated based on the ground range resolution (resolution of the range and azimuth of the slant range) and the incidence angle; the time series SAR image is Fourier transformed to obtain a complete frequency domain image (with a complete Doppler bandwidth); according to the multi-looked number, the acquired frequency domain image is divided into multi-looked parts to obtain a multi-looked molecular sub-spectrum frequency domain image; the acquired sub-spectrum frequency domain image is Fourier transformed to obtain a multi-looked molecular sub-spectrum time domain image; the sub-spectrum time domain images are merged to obtain a complete multi-looked SAR image. Among them, after multi-look processing, the azimuth direction of the data will become less, and intuitively, the number of data rows will become less. If the row and column numbers of the data are to be kept unchanged, the frequency domain zero padding method can be used.

[0067] Step S22: performing a speckle noise removal operation on the single-channel intensity data based on a refined Lee filter to obtain a corrected SAR image.

[0068] In this step, the single-channel intensity data uses a refined Lee filter with a window size of 7*7 to suppress speckle noise.

[0069] Step S23, registering the corrected SAR image based on the digital elevation model to obtain a registered SAR image.

[0070] In this step, the DEM is obtained and geocoded and radiometrically calibrated to obtain the backscatter coefficient, and the VV, VH, HH, and HV data of each phase are further obtained. The same reference image is selected, and the data of each year are processed using the intensity cross-correlation method. The time series SAR data are aligned using a coarse-to-fine matching method to achieve sub-pixel level alignment accuracy.

[0071] In one example, time-series SAR images can be refined by Lee filtering to remove speckle noise. The refined Lee filtering is simulated using a neural network model based on an autoencoder architecture and adds a channel attention mechanism. Morphological filtering is then performed to enhance geometric contour information. The preprocessing steps can be expressed as follows:

[0072]

[0073] Lee(X)=X+f conv (Cat(f conv (X), f CBAM (f ReLU (f conv (X)))))

[0074] in, and are the input and preprocessed time series SAR images, respectively. A and B are the sizes of time series SAR images. Lee() is the refined Lee filter simulated by the network. MF() is the morphological filter. f conv (), f ReLU () are convolution and activation operations respectively; Cat() represents the concatenation of feature channel dimensions; f CBAM () represents the channel attention mechanism.

[0075] Step S3, performing temporal fusion on the registered SAR images based on the feature dimension parallel network to obtain a fused SAR image.

[0076] In this step, the feature dimension parallel network specifically includes:

[0077]

[0078] Among them, F N represents the fused SAR image, Indicates the parallel symbol, F tn represents the polarimetric SAR data feature set for the nth year, represents the registered SAR image, A represents the length of the registered SAR image, B represents the width of the registered SAR image, that is, A×B represents the size of the registered SAR image, and N represents the amount of time series information obtained by the registered SAR image.

[0079] Step S4, inputting the fused SAR image into a sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image.

[0080] In this step, in order to ensure that each pixel is located at the center of the window, the adjacent value filling operation is adopted to process the image boundary. When the sliding window is scanned, the center pixel a in the sliding window is ij The pixels that can obtain spatial information include a (i-ω)(j-ω) ,...,a (i+ω)(j+ω) , we get λ pieces of spatial information, and record the spatial feature vector as λ is calculated as follows:

[0081] λ=ω×ω

[0082] The sliding window scanning method is used to obtain rich neighborhood information of pixels. In order to ensure that each pixel is located at the center of the window, the adjacent value filling operation is used to process the boundary of the image.

[0083] The final filled new image is:

[0084]

[0085] The feature vector of each pixel after updating is:

[0086]

[0087] Among them, int() represents the direct rounding function, ω represents the length of the sliding window, and λ represents the area of ​​the sliding window. Represents the feature vector of the pixel in the fused SAR image, represents the spatial feature vector of the central pixel in the sliding window, i represents the horizontal coordinate of the pixel in the fused SAR image, j represents the vertical coordinate of the pixel in the fused SAR image, represents the concatenation symbol, and N represents the number of time series information obtained by registering SAR images.

[0088] Step S5, constructing an initial cascade forest model and inputting the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model.

[0089] In this step, step S2 also includes steps S51 to S54.

[0090] Step S51, constructing an initial cascade forest model based on multiple completely random forests and ordinary random forests, wherein each cascade layer in the initial cascade forest model includes two random forests and two completely random forests.

[0091] In this step, the cascade forest model consists of a random forest and a completely random forest. The random forest randomly selects features as candidate features (d is the number of features), and then the best feature is selected by calculating the gini value for splitting. The completely random forest only randomly selects one feature from the input features, and the number of categories to be predicted is selected as C. Each cascade layer is set with two random forests and two completely random forests.

[0092] An initial cascade forest model is constructed based on multiple completely random forests and ordinary random forests, including:

[0093]

[0094] in, represents the predicted class vector of a completely random forest, represents the class vector of another fully random forest prediction, represents the predicted class vector of a normal random forest, represents the class vector of another ordinary random forest prediction, Indicates the concatenation symbol, {p k (x ij )|k=1,2...C} represents any completely random forest or ordinary random forest for x ij The probability of predicting the kth class, X N is the candidate feature initially input to the cascade layer, and J() represents the cascade operation.

[0095] The initial cascade forest model specifically includes:

[0096]

[0097] When the initial cascade forest model is output at the mth layer, we have:

[0098] X m = J(X m-1 )

[0099] predict=Max(Ave(X m ))

[0100] Among them, X1 represents the vector of the class vector output after the first layer of cascade operation and the candidate feature concatenation, X2 represents the vector of the class vector output after the second layer of cascade operation and the candidate feature concatenation, X n represents the vector of the class vector output after the n-th layer cascade operation and the concatenation of the candidate features, X m Represents the final class vector for prediction, predict represents the output category label, Max() represents the maximum value function, and Ave() represents the average value function.

[0101] Step S52: extracting several features from the spatiotemporal polarization feature set as candidate features and inputting them into the initial cascade forest model.

[0102] Step S53, determining whether the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same.

[0103] Step S54: If the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same, the training of the initial cascade forest model is completed, and the cascade forest model is obtained.

[0104] In this step, the last prediction predict will be one of the C class labels. After each level, the cascade layer will evaluate the prediction results to decide whether to expand to the next level. When the accuracy is no longer improved, the expansion will stop and the final class vector will be output.

[0105] Step S6, inputting the fused SAR image into the cascade forest model to obtain the wetland feature classification result of the target wetland area.

[0106] In this step, the cascade forest model is used to process the pixels in the fused SAR image one by one to obtain the classification results of wetland features in the study area, and the improved cascade forest wetland feature classification method of fused SAR images is completed. By using cascade forests to process complex and large spatiotemporal polarization multidimensional feature sets, the tedious process of manual feature extraction and complex parameter adjustment in traditional methods is avoided. At the same time, cascade forests can adapt to complexity, fully explore and utilize the fused image information, achieve high-precision classification of wetland features, and still show excellent classification results when the data samples are limited, with significant robustness.

[0107] The spatiotemporal polarization feature set is input into the subsequent cascade forest. Each subtree in each forest is trained in parallel through the class vector set. A predicted class probability vector is given to each sample through the constructed forest. The class probability vectors of all forests are concatenated with the original class vector of the sample as the input sample of the next level of forest. After each level, the level will evaluate the prediction results to decide whether to expand the next level. When the accuracy is no longer improved, the expansion is stopped and the final class vector is output. Finally, the pixels in the multi-time series SAR images are processed one by one using the trained cascade forest model to obtain the results of wetland feature classification in the study area. The improved cascade forest wetland feature classification method integrating multi-time series images is completed. The classification results obtained have high classification accuracy and good visual effects.

[0108] like Figure 3 As shown, Figure 3 (a) is the classification result of RF on polarization SAR data with different time series information. Figure 3 (b) is the classification result of XGB on polarization SAR data with different time series information. Figure 3 (c) is the classification result of SVM on polarization SAR data with different time series information. Figure 3 (d) is the classification result of the present invention on polarization SAR data with different time series information.

[0109] Please refer to Tables 1 to 3. The following tables are quantitative evaluations of the effect of the heterogeneous time-series image wetland monitoring method based on the improved cascade forest. Therefore, the overall accuracy (OA), Kappa coefficient, producer accuracy (PA), and user accuracy (UA) are selected as evaluation indicators. Figure 3 The experimental results in Tables 1 to 3 show that the wetland feature classification results obtained by this method are closer to the true value of the surface, with less noise, better detail preservation and visual effect, high overall accuracy and kappa, and greatly increased PA and UA values ​​for each type of feature. Therefore, this method can significantly improve the effect of wetland feature classification from multiple aspects. Among them, Table 1 shows the accuracy evaluation results obtained by RF, SVM, XGBoost, LightGBM and the classifier used in this scheme, Table 2 shows the producer accuracy of different input time information data sets and supervised classification methods, and Table 3 shows the user accuracy of different input time information data sets and supervised classification methods.

[0110] Table 1

[0111]

[0112] Table 2

[0113]

[0114]

[0115] Table 3

[0116]

[0117] Based on the above method, the embodiment of the present application discloses a heterogeneous time series remote sensing image wetland monitoring device based on improved cascade forest, referring to Figure 4 The heterogeneous time-series remote sensing image wetland monitoring device 1 includes a data acquisition module 11, a data processing module 12, a time-series fusion module 13, a feature scanning module 14 and a training processing module 15, wherein:

[0118] The data acquisition module 11 is used to acquire time-series SAR images of the target wetland area;

[0119] The data processing module 12 is used to pre-process the time-series SAR images to obtain the registered SAR images;

[0120] The time series fusion module 13 is used to perform time series fusion on the registered SAR images based on the feature dimension parallel network to obtain a fused SAR image;

[0121] The feature scanning module 14 is used to input the fused SAR image into the sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image;

[0122] The training processing module 15 is used to construct an initial cascade forest model and input the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model, input the fused SAR image into the cascade forest model, and obtain the wetland feature classification results of the target wetland area.

[0123] In one example, the data processing module 12 is used to perform multi-view processing on the time-series SAR images to convert the single-channel complex data in the time-series SAR images into single-channel intensity data; remove the coherent speckle noise operation on the single-channel intensity data based on the refined Lee filter to obtain a corrected SAR image; and align the corrected SAR images based on the digital elevation model to obtain a registered SAR image.

[0124] In one example, the feature dimension parallel network specifically includes:

[0125]

[0126] Among them, F N represents the fused SAR image, Indicates the parallel symbol, F tn represents the polarimetric SAR data feature set for the nth year, represents the registered SAR image, A represents the length of the registered SAR image, B represents the width of the registered SAR image, and N represents the amount of time series information obtained by the registered SAR image.

[0127] In one example, the fused SAR image is input into a sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image, specifically including:

[0128] λ=ω×ω

[0129]

[0130] Among them, int() represents the direct rounding function, ω represents the length of the sliding window, and λ represents the area of ​​the sliding window. Represents the feature vector of the pixel in the fused SAR image, represents the spatial feature vector of the central pixel in the sliding window, i represents the horizontal coordinate of the pixel in the fused SAR image, j represents the vertical coordinate of the pixel in the fused SAR image, represents the concatenation symbol, and N represents the number of time series information obtained by registering SAR images.

[0131] In one example, the training processing module 15 is used to construct an initial cascade forest model based on multiple completely random forests and ordinary random forests, wherein each cascade layer in the initial cascade forest model includes two random forests and two completely random forests; extract a number of features from the spatiotemporal polarization feature set as candidate features and input them into the initial cascade forest model; determine whether the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same; if the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same, the training of the initial cascade forest model is completed, and the cascade forest model is obtained.

[0132] In one example, an initial cascade forest model is constructed based on multiple completely random forests and ordinary random forests, including:

[0133]

[0134] in, represents the predicted class vector of a completely random forest, represents the class vector of another fully random forest prediction, represents the predicted class vector of a normal random forest, represents the class vector of another ordinary random forest prediction, Indicates the concatenation symbol, {p k (x ij )|k=1,2...C} represents any completely random forest or ordinary random forest for xij The probability of predicting the kth class, X N is the candidate feature initially input to the cascade layer, and J() represents the cascade operation.

[0135] In one example, the initial cascade forest model specifically includes:

[0136]

[0137]

[0138] X m = J(X m-1 )

[0139] predict=Max(Ave(X m ))

[0140] Among them, X1 represents the vector of the class vector output after the first layer of cascade operation and the candidate feature concatenation, X2 represents the vector of the class vector output after the second layer of cascade operation and the candidate feature concatenation, X n represents the vector of the class vector output after the n-th layer cascade operation and the concatenation of the candidate features, X m Represents the final class vector for prediction, predict represents the output category label, Max() represents the maximum value function, and Ave() represents the average value function.

[0141] See also Figure 5 , is a schematic diagram of the structure of an electronic device provided in an embodiment of the present application. Figure 5 As shown, the electronic device 2 may include: at least one processor 21 , at least one network interface 24 , a user interface 23 , a memory 25 , and at least one communication bus 22 .

[0142] The communication bus 22 is used to realize the connection and communication between these components.

[0143] The user interface 23 may include a display screen (Display) and a camera (Camera), and the optional user interface 23 may also include a standard wired interface and a wireless interface.

[0144] The network interface 24 may optionally include a standard wired interface or a wireless interface (such as a WI-FI interface).

[0145] Among them, the processor 21 may include one or more processing cores. The processor 21 uses various interfaces and lines to connect various parts of the entire server, and executes various functions of the server and processes data by running or executing instructions, programs, code sets or instruction sets stored in the memory 25, and calling data stored in the memory 25. Optionally, the processor 21 can be implemented in at least one hardware form of digital signal processing (Digital Signal Processing, DSP), field programmable gate array (Field ~ Programmable Gate Array, FPGA), and programmable logic array (Programmable Logic Array, PLA). The processor 21 can integrate one or a combination of a central processing unit (Central Processing Unit, CPU), a graphics processor (Graphics Processing Unit, GPU) and a modem. Among them, the CPU mainly processes the operating system, user interface and application programs; the GPU is responsible for rendering and drawing the content to be displayed on the display screen; the modem is used to process wireless communications. It can be understood that the above-mentioned modem may not be integrated into the processor 21, and it can be implemented separately through a chip.

[0146] Among them, the memory 25 may include a random access memory (Random Access Memory, RAM) and may also include a read-only memory (Read-Only Memory). Optionally, the memory 25 includes a non-transitory computer-readable storage medium (non-transitory computer-readable storage medium). The memory 25 can be used to store instructions, programs, codes, code sets or instruction sets. The memory 25 may include a program storage area and a data storage area, wherein the program storage area may store instructions for implementing an operating system, instructions for at least one function (such as a touch function, a sound playback function, an image playback function, etc.), instructions for implementing the above-mentioned method embodiments, etc.; the data storage area may store data involved in the above-mentioned method embodiments, etc. The memory 25 may also be optionally at least one storage device located away from the aforementioned processor 21. As Figure 5 As shown, the memory 25 as a computer storage medium may include an operating system, a network communication module, a user interface module, and an application program for a method for simulating real-time playback of GNSS post-observation files.

[0147] exist Figure 5In the electronic device 2 shown, the user interface 23 is mainly used to provide an input interface for the user and obtain data input by the user; and the processor 21 can be used to call the application of the simulated real-time playback method of the GNSS post-observation file stored in the memory 25. When executed by one or more processors, the electronic device executes one or more methods in the above-mentioned embodiments.

[0148] A computer-readable storage medium stores instructions, which, when executed by one or more processors, enable the computer to execute one or more methods in the above-mentioned embodiments.

[0149] It should be noted that, for the aforementioned method embodiments, for the sake of simplicity, they are all described as a series of action combinations, but those skilled in the art should be aware that the present application is not limited by the order of the actions described, because according to the present application, certain steps can be performed in other orders or simultaneously. Secondly, those skilled in the art should also be aware that the embodiments described in the specification are all preferred embodiments, and the actions and modules involved are not necessarily required for the present application.

[0150] In the above embodiments, the description of each embodiment has its own emphasis. For parts that are not described in detail in a certain embodiment, reference can be made to the relevant descriptions of other embodiments.

[0151] In the several embodiments provided in this application, it should be understood that the disclosed devices can be implemented in other ways. For example, the device embodiments described above are only schematic, such as the division of units, which is only a logical function division. There may be other division methods in actual implementation, such as multiple units or components can be combined or integrated into another system, or some features can be ignored or not executed. Another point is that the mutual coupling or direct coupling or communication connection shown or discussed can be through some service interfaces, and the indirect coupling or communication connection of devices or units can be electrical or other forms.

[0152] The units described as separate components may or may not be physically separated, and the components shown as units may or may not be physical units, that is, they may be located in one place or distributed on multiple network units. Some or all of the units may be selected according to actual needs to achieve the purpose of the solution of this embodiment.

[0153] In addition, each functional unit in each embodiment of the present application may be integrated into one processing unit, or each unit may exist physically separately, or two or more units may be integrated into one unit. The above-mentioned integrated unit may be implemented in the form of hardware or in the form of software functional units.

[0154] If the integrated unit is implemented in the form of a software functional unit and sold or used as an independent product, it can be stored in a computer-readable memory. Based on this understanding, the technical solution of the present application, or the part that contributes to the prior art, or all or part of the technical solution can be embodied in the form of a software product, which is stored in a memory and includes several instructions for a computer device (which can be a personal computer, server or network device, etc.) to perform all or part of the steps of the various embodiments of the present application. The aforementioned memory includes: various media that can store program codes, such as USB flash drives, mobile hard drives, magnetic disks or optical disks.

[0155] The above description is only a preferred embodiment of the present invention and is not intended to limit the present invention. Any modifications, equivalent substitutions, improvements, etc. made within the spirit and principles of the present invention should be included in the protection scope of the present invention.

Claims

1. A heterogeneous time-series image wetland monitoring method based on improved cascade forest, characterized in that: The method comprises: Acquire time-series SAR images of the target wetland area; Preprocessing the time-series SAR images to obtain registered SAR images; Based on a feature dimension parallel network, the registered SAR images are temporally fused to obtain a fused SAR image; The characteristic dimension parallel network specifically includes: Among them, F N represents the fused SAR image, Indicates the parallel symbol, F tn represents the polarimetric SAR data feature set for the nth year, represents the registered SAR image, A represents the length of the registered SAR image, B represents the width of the registered SAR image, and N represents the amount of time series information obtained by the registered SAR image; Inputting the fused SAR image into a sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image; Inputting the fused SAR image into a sliding window scanning model to obtain the spatiotemporal polarization feature set corresponding to the fused SAR image, specifically comprising: λ=ω×ω Among them, int() represents the direct rounding function, ω represents the length of the sliding window, and λ represents the area of ​​the sliding window. represents the feature vector of the pixel in the fused SAR image, represents the spatial feature vector of the central pixel in the sliding window, i represents the horizontal coordinate of the pixel in the fused SAR image, j represents the vertical coordinate of the pixel in the fused SAR image, represents a serial connection symbol, and N represents the number of time series information obtained by the registered SAR images; Constructing an initial cascade forest model and inputting the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model; The fused SAR image is input into the cascade forest model to obtain the wetland feature classification result of the target wetland area.

2. The method according to claim 1, characterized in that The preprocessing of the time-series SAR images to obtain the registered SAR images specifically includes: Performing multi-view processing on the time-series SAR image to convert single-channel complex data in the time-series SAR image into single-channel intensity data; Based on refined Lee filtering, a speckle noise removal operation is performed on the single-channel intensity data to obtain a corrected SAR image; The corrected SAR image is registered based on a digital elevation model to obtain the registered SAR image.

3. The method according to claim 1, characterized in that An initial cascade forest model is constructed based on multiple completely random forests and common random forests, and the spatiotemporal polarization feature set is input into each level of the initial cascade forest model for training to obtain a cascade forest model, specifically including: Constructing the initial cascade forest model based on multiple completely random forests and common random forests, wherein each cascade layer in the initial cascade forest model includes two random forests and two completely random forests; Extracting several features from the spatiotemporal polarization feature set as candidate features and inputting them into the initial cascade forest model; Determine whether the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same; If the predicted values ​​output by the nth cascade layer and the n-1th cascade layer in the initial cascade forest model are the same, the training of the initial cascade forest model is completed and the cascade forest model is obtained.

4. The method according to claim 3, characterized in that The constructing of the initial cascade forest model based on multiple completely random forests and common random forests specifically includes: in, represents the predicted class vector of a completely random forest, represents the class vector of another fully random forest prediction, represents the predicted class vector of a normal random forest, represents the class vector of another ordinary random forest prediction, Indicates the concatenation symbol. Represents any completely random forest or ordinary random forest for x ij The probability of predicting the kth class, S includes RF1, RF2, CF1 and CF2, RF1 represents a completely random forest, RF2 represents another completely random forest, CF1 represents an ordinary random forest, CF2 represents another ordinary random forest, C represents the number of categories to be predicted by the completely random forest, X N is the candidate feature initially input to the cascade layer, and J() represents the cascade operation.

5. The method according to claim 4, characterized in that The initial cascade forest model specifically includes: X m =J(X m-1 ) predict=Max(Ave(X m )) Wherein, X1 represents the vector obtained by concatenating the class vector output after the first layer of cascade operation and the candidate feature, X2 represents the vector obtained by concatenating the class vector output after the second layer of cascade operation and the candidate feature, and X n represents the vector of the class vector output after the n-th layer cascade operation and the candidate feature concatenation, X m Represents the final class vector for prediction, predict represents the output category label, Max() represents the maximum value function, and Ave() represents the average value function.

6. A heterogeneous time-series remote sensing image wetland monitoring device based on improved cascade forest, characterized in that: The heterogeneous time-series remote sensing image wetland monitoring device (1) comprises a data acquisition module (11), a data processing module (12), a time-series fusion module (13), a feature scanning module (14) and a training processing module (15), wherein: The data acquisition module (11) is used to acquire time-series SAR images of the target wetland area; The data processing module (12) is used to pre-process the time-series SAR images to obtain registered SAR images; The time series fusion module (13) is used to perform time series fusion on the registered SAR images based on a feature dimension parallel network to obtain a fused SAR image; The characteristic dimension parallel network specifically includes: Among them, F N represents the fused SAR image, Indicates the parallel symbol, F tn represents the polarimetric SAR data feature set for the nth year, represents the registered SAR image, A represents the length of the registered SAR image, B represents the width of the registered SAR image, and N represents the amount of time series information obtained by the registered SAR image; The feature scanning module (14) is used to input the fused SAR image into a sliding window scanning model to obtain a spatiotemporal polarization feature set corresponding to the fused SAR image; Inputting the fused SAR image into a sliding window scanning model to obtain the spatiotemporal polarization feature set corresponding to the fused SAR image, specifically comprising: λ=ω×ω Among them, int() represents the direct rounding function, ω represents the length of the sliding window, and λ represents the area of ​​the sliding window. represents the feature vector of the pixel in the fused SAR image, represents the spatial feature vector of the central pixel in the sliding window, i represents the horizontal coordinate of the pixel in the fused SAR image, j represents the vertical coordinate of the pixel in the fused SAR image, represents a serial connection symbol, and N represents the number of time series information obtained by the registered SAR images; The training processing module (15) is used to construct an initial cascade forest model and input the spatiotemporal polarization feature set into each level of the initial cascade forest model for training to obtain a cascade forest model, and input the fused SAR image into the cascade forest model to obtain a wetland feature classification result of the target wetland area.

7. An electronic device, characterized in that: The electronic device (2) comprises a processor (21), a memory (25), a user interface (23) and a network interface (24), wherein the memory (25) is used to store instructions, the user interface (23) and the network interface (24) are used to communicate with other devices, and the processor (21) is used to execute the instructions stored in the memory (25) so that the electronic device (2) executes the method according to any one of claims 1 to 5.

8. A computer-readable storage medium having a computer program stored thereon, characterized in that: When the computer program is executed by a processor, the steps of the method according to any one of claims 1 to 5 are implemented.

Images