Wetland classification method, device and equipment, storage medium and product
By combining the preprocessing and feature fusion classification of hyperspectral images and lidar images, the problem of low accuracy of hyperspectral remote sensing images in wetland classification is solved, and more efficient wetland classification is achieved.
Patent Information
- Application Number
- CN202410263387.3
- Authority / Receiving Office
- CN · China
- Patent Type
- Applications(China)
- Current Assignee / Owner
- Filing Date
- 2024-03-07
- Publication Date
- 2025-09-09
AI Technical Summary
The accuracy of existing hyperspectral remote sensing images in classifying riverside/lakeside wetlands is low, and there are problems of spectral confusion and information missing.
Combining hyperspectral images and lidar images, through preprocessing, feature extraction, regulation of information attention and fusion classification, spectral enhancement features and elevation enhancement features are obtained, and finally wetland classification is performed.
It improves the accuracy and efficiency of wetland classification, enables a more comprehensive understanding of the three-dimensional structure and surface characteristics of wetlands, and avoids information loss and redundant interference.
Smart Images
Figure CN120612503A_ABST
Abstract
Description
Technical Field
[0001] The present invention relates to the field of data processing technology, and in particular to a wetland classification method, device, equipment, storage medium and computer program product. Background Art
[0002] With the rapid advancement of remote sensing technology, hyperspectral remote sensing images are becoming increasingly important in earth observation and resource management. Remote sensing technology captures the spectral information of the earth's surface in a highly detailed and accurate manner, providing key data support for fields such as environmental monitoring, land use planning, and natural resource management. However, despite the remarkable achievements of hyperspectral remote sensing in many aspects, single hyperspectral data still faces some challenges when processing complex environments. Problems such as spectral confusion and information loss limit its accuracy and applicability in specific application scenarios. For example, in the hyperspectral image classification task of riverside / lakeside wetlands, errors are prone to occur at the boundaries of the objects. Therefore, how to efficiently and accurately classify riverside / lakeside wetlands has become a technical problem that needs to be solved urgently. Summary of the Invention
[0003] The main purpose of the present invention is to provide a wetland classification method, device, equipment, storage medium and computer program product, aiming to solve the technical problem of low accuracy of wetland image classification using hyperspectral remote sensing images in the existing technology.
[0004] To achieve the above object, the present invention provides a wetland classification method, which comprises the following steps:
[0005] Preprocess the hyperspectral image and lidar image of the wetland to obtain the hyperspectral feature map and point cloud data feature map;
[0006] Extracting features from the hyperspectral feature map and the point cloud data feature map to obtain spectral spatial features and elevation features;
[0007] Regulating the information attention of the spectral spatial feature and the elevation feature to obtain a spectral enhancement feature and an elevation enhancement feature;
[0008] The spectral enhancement features and the elevation enhancement features are fused and classified to obtain a wetland classification result.
[0009] Optionally, the step of fusing and classifying the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result includes:
[0010] The spectral enhancement feature and the elevation enhancement feature are fused and enhanced to obtain a spectral cross-enhancement feature and an elevation cross-enhancement feature;
[0011] Selecting a target decision feature from the spectral cross-enhancement feature and the elevation cross-enhancement feature;
[0012] Determining a complementary fusion feature according to the spectral enhancement feature and the elevation enhancement feature;
[0013] A wetland classification result is determined based on the target decision feature and the complementary fusion feature.
[0014] Optionally, the step of extracting features from the hyperspectral feature map and the point cloud data feature map respectively to obtain spectral spatial features and elevation features includes:
[0015] Performing multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result;
[0016] Perform feature fusion according to the multi-scale feature extraction result to obtain a feature fusion result;
[0017] Downsampling the feature fusion result to obtain spectral spatial features;
[0018] Shallow feature extraction is performed on the point cloud data feature map to obtain elevation features.
[0019] Optionally, the step of performing multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result includes:
[0020] Dividing the hyperspectral feature map according to a preset number of branches to obtain a division result;
[0021] A convolution operation is performed on each branch according to the division result to obtain a multi-scale feature extraction result.
[0022] Optionally, the step of adjusting the information attention of the spectral spatial feature and the elevation feature to obtain the spectral enhancement feature and the elevation enhancement feature includes:
[0023] Performing dimension compression on the spectral spatial feature and the elevation feature respectively to obtain a spectral feature vector and an elevation feature vector;
[0024] Performing vector compression on the spectral feature vector and the elevation feature vector respectively to obtain a first vector compression result and a second vector compression result;
[0025] determining a spectrum enhancement feature according to the first vector compression result and the spectrum feature vector;
[0026] An elevation enhancement feature is determined according to the second vector compression result and the elevation feature vector.
[0027] Optionally, the step of preprocessing the hyperspectral image and the lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map includes:
[0028] The principal component analysis method is used to reduce the feature dimension of the hyperspectral image and lidar image of the wetland to obtain the first image and the second image;
[0029] Normalizing the first image and the second image respectively to obtain a first normalized image and a second normalized image;
[0030] Cutting the first normalized image to obtain a hyperspectral feature map;
[0031] The second normalized image is cut to obtain a point cloud data feature map.
[0032] In addition, to achieve the above-mentioned purpose, the present invention also provides a wetland classification device, which includes:
[0033] A preprocessing module is used to preprocess the hyperspectral image and lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map;
[0034] A feature extraction module is used to extract features from the hyperspectral feature map and the point cloud data feature map to obtain spectral spatial features and elevation features;
[0035] A control module, configured to control the information attention of the spectral spatial feature and the elevation feature to obtain a spectral enhancement feature and an elevation enhancement feature;
[0036] The fusion classification module is used to fuse and classify the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result.
[0037] In addition, to achieve the above-mentioned purpose, the present invention also proposes a wetland classification device, which includes: a memory, a processor, and a wetland classification program stored on the memory and runnable on the processor, and the wetland classification program is configured to implement the steps of the wetland classification method described above.
[0038] In addition, to achieve the above-mentioned purpose, the present invention also proposes a storage medium, on which a wetland classification program is stored. When the wetland classification program is executed by a processor, the steps of the wetland classification method described above are implemented.
[0039] In addition, to achieve the above-mentioned purpose, the present invention also provides a computer program product, which includes a wetland classification program, and when the wetland classification program is executed by a processor, it implements the steps of the wetland classification method described above.
[0040] The present invention preprocesses hyperspectral images and lidar images of wetlands to obtain hyperspectral feature maps and point cloud data feature maps; performs feature extraction on the hyperspectral feature maps and point cloud data feature maps respectively to obtain spectral spatial features and elevation features; regulates the information attention of the spectral spatial features and the elevation features to obtain spectral enhancement features and elevation enhancement features; and fuses and classifies the spectral enhancement features and the elevation enhancement features to obtain wetland classification results. Since the present invention uses elevation information provided by lidar images and spectral spatial information provided by hyperspectral images to identify and classify wetlands, compared to existing methods of wetland image classification using hyperspectral remote sensing images, the above-mentioned method of the present invention can more comprehensively understand the three-dimensional structure and surface features of wetlands, thereby improving wetland classification efficiency. BRIEF DESCRIPTION OF THE DRAWINGS
[0041] Figure 1 It is a structural diagram of a wetland classification device in a hardware operating environment involved in an embodiment of the present invention;
[0042] Figure 2 This is a flow chart of the first embodiment of the wetland classification method of the present invention;
[0043] Figure 3 This is a schematic structural diagram of a first embodiment of a wetland classification method according to the present invention;
[0044] Figure 4 is a schematic diagram of a second embodiment of the wetland classification method of the present invention;
[0045] Figure 5 This is a structural block diagram of the first embodiment of the wetland classification device of the present invention.
[0046] The purpose, features and advantages of the present invention will be further described with reference to the accompanying drawings and in conjunction with the embodiments. DETAILED DESCRIPTION
[0047] It should be understood that the specific embodiments described herein are only used to explain the present invention and are not intended to limit the present invention.
[0048] Reference Figure 1 , Figure 1 This is a schematic diagram of the wetland classification device structure in the hardware operating environment involved in the embodiment of the present invention.
[0049] like Figure 1As shown, the wetland classification device may include: a processor 1001, such as a central processing unit (CPU), a communication bus 1002, a user interface 1003, a network interface 1004, and a memory 1005. The communication bus 1002 is used to implement communication between these components. The user interface 1003 may include a display screen and an input unit such as a keyboard. Optionally, the user interface 1003 may also include a standard wired interface or a wireless interface. The network interface 1004 may optionally include a standard wired interface or a wireless interface (such as a wireless fidelity (WI-FI) interface). The memory 1005 may be a high-speed random access memory (RAM) or a stable non-volatile memory (NVM), such as a disk storage device. The memory 1005 may also be a storage device independent of the aforementioned processor 1001.
[0050] Those skilled in the art will understand that Figure 1 The structure shown in does not constitute a limitation on the wetland classification equipment, and may include more or fewer components than shown in the figure, or combine certain components, or arrange the components differently.
[0051] like Figure 1 As shown, the memory 1005 as a storage medium may include an operating system, a network communication module, a user interface module and a wetland classification program.
[0052] exist Figure 1 In the wetland classification device shown, the network interface 1004 is mainly used for data communication with the network server; the user interface 1003 is mainly used for data interaction with the user; the processor 1001 and the memory 1005 in the wetland classification device of the present invention can be set in the wetland classification device, and the wetland classification device calls the wetland classification program stored in the memory 1005 through the processor 1001, and executes the wetland classification method provided by the embodiment of the present invention.
[0053] Based on the above wetland classification device, the embodiment of the present invention provides a wetland classification method, referring to Figure 2 , Figure 2 Schematic diagram of the flow chart of the first embodiment of the wetland classification method of the present invention.
[0054] In this embodiment, the wetland classification method includes the following steps:
[0055] Step S10: Preprocessing the hyperspectral image and lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map.
[0056] It should be noted that the execution subject of this embodiment can be a computing service device with data processing, network communication, and program execution capabilities, such as a mobile phone, tablet computer, or personal computer, or an electronic device or wetland classification device capable of performing the aforementioned functions. This embodiment and the following embodiments will be described below using the wetland classification device as an example.
[0057] It should be noted that the preprocessing of the hyperspectral image and the lidar image of the wetland to obtain the hyperspectral feature map and the point cloud data feature map can be performed by performing feature dimensionality reduction, normalization, image cutting and other processing on the hyperspectral image and the lidar image of the wetland to obtain the hyperspectral feature map and the point cloud data feature map. Specifically, the hyperspectral image of the wetland is processed by performing feature dimensionality reduction, normalization, image cutting and other processing to obtain the hyperspectral feature map F HSI ; Perform feature dimension reduction, normalization, image cutting and other processing on the wetland lidar image to obtain the point cloud data feature map F Lidar The laser radar image is Lidar data, which is hereinafter referred to as X Lidar Characterization, the hyperspectral image is X HSI representation.
[0058] Furthermore, in order to improve the wetland classification accuracy, the step S10 may include: performing feature dimensionality reduction on the hyperspectral image and the lidar image of the wetland using a principal component analysis method to obtain a first image and a second image;
[0059] Normalizing the first image and the second image respectively to obtain a first normalized image and a second normalized image;
[0060] Cutting the first normalized image to obtain a hyperspectral feature map;
[0061] The second normalized image is cut to obtain a point cloud data feature map.
[0062] It should be noted that the principal component analysis (PCA) dimensionality reduction method can reduce data dimensions and extract key features. The principal component analysis method is used to perform feature dimensionality reduction on the hyperspectral image and lidar image of the wetland to obtain the first image and the second image. The method specifically includes: performing feature dimensionality reduction on the hyperspectral image of the wetland to obtain the first image, and performing feature dimensionality reduction on the lidar image of the wetland to obtain the second image. Normalizing the first image and the second image respectively to obtain a first normalized image and a second normalized image specifically includes: normalizing the first image to obtain a first normalized image, and normalizing the second image to obtain a second normalized image. Cutting the first normalized image to obtain a hyperspectral feature map can be cutting the first normalized image into patch blocks of size P*P to obtain a hyperspectral feature map F. HSI The second normalized image is cut into patches of size P*P to obtain a point cloud data feature map F. Lidar Where P is the preset image size, which can be customized according to the required classification accuracy.
[0063] Step S20: performing feature extraction on the hyperspectral feature map and the point cloud data feature map respectively to obtain spectral spatial features and elevation features.
[0064] It should be noted that the feature extraction of the hyperspectral feature map and the point cloud data feature map is respectively performed to obtain the spectral space feature and the elevation feature, which can be respectively extracted from the hyperspectral feature map F HSI And point cloud data feature map F Lidar The spectral spatial information and elevation information of the hyperspectral image are extracted based on deep CNN to extract multi-level spatial information and multi-scale features, mining the deep spectral spatial information of the hyperspectral image, and using two-layer CNN to extract elevation information from Lidar data to obtain spectral spatial features f and f respectively. hsi and elevation feature f Lidar .
[0065] Step S30: regulating the information attention of the spectral spatial feature and the elevation feature to obtain a spectral enhancement feature and an elevation enhancement feature.
[0066] It should be noted that the information attention of the spectral spatial feature and the elevation feature is adjusted to obtain the spectral enhancement feature and the elevation enhancement feature, which can be obtained by adjusting the spectral spatial feature f of the extracted hyperspectral image. hsi and the elevation feature f of Lidar data LidarThrough the self-attention mechanism, the attention of each information is regulated to enhance the features of different perception levels and obtain the height enhancement features. and spectral enhancement features
[0067] Furthermore, the step S30 may include: performing dimension compression on the spectral spatial feature and the elevation feature respectively to obtain a spectral feature vector and an elevation feature vector;
[0068] Performing vector compression on the spectral feature vector and the elevation feature vector respectively to obtain a first vector compression result and a second vector compression result;
[0069] determining a spectrum enhancement feature according to the first vector compression result and the spectrum feature vector;
[0070] An elevation enhancement feature is determined according to the second vector compression result and the elevation feature vector.
[0071] It should be noted that the spectral spatial feature and the elevation feature are dimensional compressed to obtain the spectral feature vector and the elevation feature vector respectively. hsi and elevation feature f Lidar The spectral spatial features of each patch block in the image are compressed into a scale from becomes The one-dimensional vector of is used to obtain a spectral feature vector and an elevation feature vector. The spectral feature vector and the elevation feature vector are respectively compressed to obtain a first vector compression result and a second vector compression result. The feature vectors (i.e., the spectral feature vector and the elevation feature vector) are respectively compressed into a matrix of size (B, num_class) through a linear layer to obtain a first vector compression result. and the second vector compression result Where num_class is the number of wetland categories, obtained during the model training phase.
[0072] The determining of the spectrum enhancement feature according to the first vector compression result and the spectrum feature vector may be a first vector compression result. Apply the Sigmoid function to convert the matrix into the probability of each category of the compressed fusion features in the batch, and obtain the probability W Hsi , multiply the probability by the compressed result of the first vector element by element to obtain the self-attention feature and the spectral enhancement feature Refer to the following formula:
[0073]
[0074] Among them, the function mul() is used to perform multiplication calculations.
[0075] The determining of the elevation enhancement feature according to the second vector compression result and the elevation feature vector may be a step of compressing the second vector result. Apply the Sigmoid function to convert the matrix into the probability of each category of the compressed fusion features in the batch, and obtain the probability W Lidar , multiply the probability by the compressed second vector compression result element by element to obtain the self-attention feature and the elevation enhancement feature Refer to the following formula:
[0076]
[0077] Step S40: fusing and classifying the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result.
[0078] It should be noted that the spectral enhancement features and the elevation enhancement features are fused and classified to obtain the wetland classification result, which can be obtained by fusing the spectral enhancement features extracted separately. and elevation enhancement features By fusing and enhancing the interaction modules, on the one hand, the spectral spatial information of each is cross-enhanced to obtain the cross-enhanced feature O Lidar and O Hsi On the other hand, the complementary fusion feature O is obtained by adding the respective fusion features and balancing the pixel values. i , the three output results (O Lidar , O Hsi and O1) fusion, taking the maximum value as the standard, screening the cross-enhancement feature O Lidar and O Hsi , and then fuse the screening results with the complementary fusion feature O1 to output the final wetland classification result O.
[0079] Furthermore, in order to improve wetland classification efficiency, the step S40 may include: fusing and enhancing the spectral enhancement feature and the elevation enhancement feature to obtain a spectral cross-enhancement feature and an elevation cross-enhancement feature;
[0080] Selecting a target decision feature from the spectral cross-enhancement feature and the elevation cross-enhancement feature;
[0081] Determining a complementary fusion feature according to the spectral enhancement feature and the elevation enhancement feature;
[0082] A wetland classification result is determined based on the target decision feature and the complementary fusion feature.
[0083] It should be noted that the spectral enhancement feature and the elevation enhancement feature are fused and enhanced to obtain the spectral cross enhancement feature and the elevation cross enhancement feature. On the one hand, the classification probability W of each category is calculated based on the compressed fusion feature in the hyperspectral batch. Hsi As a weight, the spectral spatial information of the Lidar data (ie, the laser radar image) is compensated to obtain the elevation cross-enhancement feature O Lidar On the other hand, the classification probability W of each category is calculated based on the compressed fusion features in the Lidar data batch. Lidar As a weight, the elevation information of the hyperspectral image is compensated to obtain the spectral cross-enhancement feature O Hsi , refer to the following formula:
[0084]
[0085] The selecting of the target decision feature from the spectral cross-enhancement feature and the elevation cross-enhancement feature may be performed by taking a maximum value as a criterion to select the target decision feature from the spectral cross-enhancement feature and the elevation cross-enhancement feature.
[0086] Determining the complementary fusion feature according to the spectral enhancement feature and the elevation enhancement feature may be to balance the contributions of the spectral enhancement feature and the elevation enhancement feature to obtain the complementary fusion feature O1, which may be specifically referred to the following formula:
[0087]
[0088] Determining the wetland classification result according to the target decision feature and the complementary fusion feature may be adding the target decision feature and the complementary fusion feature to obtain the wetland classification result O, which may be specifically referred to the following formula:
[0089] O=maximum(O Lidar ,O Hsi )+O1
[0090] In the specific implementation, please refer to Figure 3 , Figure 3This is a structural diagram of the first embodiment of the wetland classification method of the present invention; this embodiment includes a spectral spatial enhancement module, a self-attention module and a fusion enhancement interaction module. This embodiment uses the spectral spatial information of the hyperspectral image and the elevation information of the Lidar data to classify wetlands. The spectral spatial enhancement module obtains the deep features of the hyperspectral image, avoiding problems such as information loss, redundant information interference and noise interference, thereby providing more reliable input data for subsequent wetland classification tasks; the self-attention module is applied to multimodal data respectively to obtain their respective features and probabilities, highlighting the importance of each modal data, and avoiding the problem of key information loss; the fusion enhancement interaction module balances the contribution of multimodal data, highlights key information, ensures that each modality receives appropriate attention and fusion, and prevents a certain modality from dominating the entire fusion result, thereby improving the balance of fusion and completing the classification of riverside / lakeside wetlands.
[0091] This embodiment preprocesses the hyperspectral image and the lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map; performs feature extraction on the hyperspectral feature map and the point cloud data feature map respectively to obtain spectral spatial features and elevation features; regulates the information attention of the spectral spatial features and the elevation features to obtain spectral enhancement features and elevation enhancement features; and fuses and classifies the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result. Since this embodiment uses the elevation information provided by the lidar image and the spectral spatial information provided by the hyperspectral image to identify and classify wetlands, compared to the existing method of classifying wetland images through hyperspectral remote sensing images, the above method of this embodiment can more comprehensively understand the three-dimensional structure and surface features of the wetland, thereby improving the efficiency of wetland classification.
[0092] refer to Figure 4 , Figure 4 Schematic diagram of the flow chart of the second embodiment of the wetland classification method of the present invention.
[0093] Based on the above first embodiment, in this embodiment, step S20 includes:
[0094] Step S201: performing multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result.
[0095] It should be noted that the multi-scale feature extraction is performed on the hyperspectral feature map to obtain the multi-scale feature extraction result. The multi-scale feature extraction result can be obtained by performing convolution operations on the hyperspectral feature map using multiple convolution layers with convolution kernel sizes of 1*1, 3*3 and 5*5, respectively, to obtain multi-scale feature extraction results under multiple branches.
[0096] Furthermore, in order to improve wetland classification efficiency, the step S201 may include: dividing the hyperspectral feature map according to a preset number of branches to obtain a division result;
[0097] A convolution operation is performed on each branch according to the division result to obtain a multi-scale feature extraction result.
[0098] It should be noted that the preset number of branches can be 3 branches. Corresponding to the convolution layers with convolution kernel sizes of 1*1, 3*3 and 5*5 respectively. The division of the hyperspectral feature map according to the preset number of branches can be performed by convolution operations on the hyperspectral feature map through convolution layers with convolution kernel sizes of 1*1, 3*3 and 5*5 respectively, and the hyperspectral feature map F Hsi It is divided into three branches. Among them, F Hsi ∈R B*C*H*W , B represents the batch size, R represents the set of real numbers, C represents the number of channels, H and W represent the length and width of the patch, and reducing the number of channels from C to C / 3, the three branches in the partitioning result are: and
[0099] It should be noted that the convolution operation is performed on each branch according to the division result to obtain the multi-scale feature extraction result, which can be the feature map obtained by the 1*1 convolution layer. Apply 3*3 and 5*5 convolution operations respectively, and we get and
[0100] The other two branches and Then after 1*1 convolution operation, we get and Finally, four feature branches are generated and The number of channels in each branch is halved to C / 6 to complete multi-scale feature extraction, where and
[0101] Step S202: performing feature fusion according to the multi-scale feature extraction result to obtain a feature fusion result.
[0102] It should be noted that the feature fusion is performed according to the multi-scale feature extraction result, and the feature fusion result can be obtained by adding the features of the first and third branches and the features of the second and fourth branches element by element, respectively. and in, For details, refer to the following formula to calculate the feature fusion results. and
[0103]
[0104] Step S203: down-sampling the feature fusion result to obtain spectral spatial features.
[0105] It should be noted that the downsampling of the feature fusion result to obtain the spectral spatial feature can be the feature obtained according to the first and third branches. Perform downsampling operation to obtain Then the downsampled features are and the features obtained from the second and fourth branches Perform element-by-element addition to obtain the final output f with spectral spatial characteristics hsi , That is, the spectral spatial characteristics, which can be referred to the following formula:
[0106]
[0107] Step S204: performing shallow feature extraction on the point cloud data feature map to obtain elevation features.
[0108] It should be noted that the shallow feature extraction of the point cloud data feature map to obtain the elevation feature can be performed by extracting the shallow features of the point cloud data feature map through two layers of CNN, each layer of which is conv2d, BN and ReLU with a stride of 3*3 to obtain the elevation feature f Lidar ,
[0109] This embodiment performs multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result; performs feature fusion based on the multi-scale feature extraction result to obtain a feature fusion result; downsamples the feature fusion result to obtain spectral spatial features; and performs shallow feature extraction on the point cloud data feature map to obtain elevation features. By acquiring deep features from the hyperspectral image, this embodiment effectively avoids problems such as information loss, redundant information interference, and noise interference. This improves the quality of the extracted data and provides more reliable and accurate feature information for subsequent wetland classification tasks.
[0110] Reference Figure 5 , Figure 5 This is a structural block diagram of the first embodiment of the wetland classification device of the present invention.
[0111] like Figure 5 As shown, the wetland classification device proposed in the embodiment of the present invention includes:
[0112] A preprocessing module 10 is used to preprocess the hyperspectral image and the lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map;
[0113] A feature extraction module 20 is used to extract features from the hyperspectral feature map and the point cloud data feature map to obtain spectral spatial features and elevation features;
[0114] A control module 30 is used to control the information attention of the spectral spatial feature and the elevation feature to obtain a spectral enhancement feature and an elevation enhancement feature;
[0115] The fusion classification module 40 is used to fuse and classify the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result.
[0116] This embodiment preprocesses the hyperspectral image and the lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map; performs feature extraction on the hyperspectral feature map and the point cloud data feature map respectively to obtain spectral spatial features and elevation features; regulates the information attention of the spectral spatial features and the elevation features to obtain spectral enhancement features and elevation enhancement features; and fuses and classifies the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result. Since this embodiment uses the elevation information provided by the lidar image and the spectral spatial information provided by the hyperspectral image to identify and classify wetlands, compared to the existing method of classifying wetland images through hyperspectral remote sensing images, the above method of this embodiment can more comprehensively understand the three-dimensional structure and surface features of the wetland, thereby improving the efficiency of wetland classification.
[0117] It should be noted that the workflow described above is merely illustrative and does not limit the scope of protection of the present invention. In practical applications, technicians in this field can select part or all of it according to actual needs to achieve the purpose of the embodiment scheme, and no limitation is made here.
[0118] In addition, for technical details not fully described in this embodiment, reference can be made to the wetland classification method provided in any embodiment of the present invention, and will not be repeated here.
[0119] Based on the above-mentioned first embodiment of the wetland classification device of the present invention, a second embodiment of the wetland classification device of the present invention is proposed.
[0120] In this embodiment, the fusion classification module 40 is further configured to fuse and enhance the spectral enhancement feature and the elevation enhancement feature to obtain a spectral cross-enhancement feature and an elevation cross-enhancement feature.
[0121] Selecting a target decision feature from the spectral cross-enhancement feature and the elevation cross-enhancement feature;
[0122] Determining a complementary fusion feature according to the spectral enhancement feature and the elevation enhancement feature;
[0123] A wetland classification result is determined based on the target decision feature and the complementary fusion feature.
[0124] Furthermore, the feature extraction module 20 is further configured to perform multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result;
[0125] Perform feature fusion according to the multi-scale feature extraction result to obtain a feature fusion result;
[0126] Downsampling the feature fusion result to obtain spectral spatial features;
[0127] Shallow feature extraction is performed on the point cloud data feature map to obtain elevation features.
[0128] Furthermore, the feature extraction module 20 is further configured to divide the hyperspectral feature map according to a preset number of branches to obtain a division result;
[0129] A convolution operation is performed on each branch according to the division result to obtain a multi-scale feature extraction result.
[0130] Furthermore, the control module 30 is further configured to perform dimension compression on the spectral spatial feature and the elevation feature respectively to obtain a spectral feature vector and an elevation feature vector;
[0131] Performing vector compression on the spectral feature vector and the elevation feature vector respectively to obtain a first vector compression result and a second vector compression result;
[0132] determining a spectrum enhancement feature according to the first vector compression result and the spectrum feature vector;
[0133] An elevation enhancement feature is determined according to the second vector compression result and the elevation feature vector.
[0134] Furthermore, the pre-processing module 10 is further configured to perform feature dimensionality reduction on the hyperspectral image and the lidar image of the wetland using a principal component analysis method to obtain a first image and a second image;
[0135] Normalizing the first image and the second image respectively to obtain a first normalized image and a second normalized image;
[0136] Cutting the first normalized image to obtain a hyperspectral feature map;
[0137] The second normalized image is cut to obtain a point cloud data feature map.
[0138] Other embodiments or specific implementations of the wetland classification device of the present invention can refer to the above-mentioned method embodiments and will not be repeated here.
[0139] In addition, an embodiment of the present invention further provides a storage medium, on which a wetland classification program is stored. When the wetland classification program is executed by a processor, the steps of the wetland classification method described above are implemented.
[0140] In addition, an embodiment of the present invention further provides a computer program product, including a wetland classification program, which implements the steps of the wetland classification method described above when executed by a processor.
[0141] The specific implementation of the computer program product of the present invention is basically the same as the embodiments of the above-mentioned wetland classification method, and will not be repeated here.
[0142] It should be noted that, in this document, the terms "comprises," "includes," or any other variations thereof are intended to encompass non-exclusive inclusion, such that a process, method, article, or system comprising a series of elements includes not only those elements but also other elements not explicitly listed, or elements inherent to such process, method, article, or system. In the absence of further limitations, an element defined by the phrase "comprising a ..." does not exclude the presence of other identical elements in the process, method, article, or system comprising the element.
[0143] The serial numbers of the above embodiments of the present invention are for description only and do not represent the advantages or disadvantages of the embodiments.
[0144] Through the description of the above embodiments, those skilled in the art can clearly understand that the above-mentioned embodiment methods can be implemented by means of software plus the necessary general hardware platform, and of course can also be implemented by hardware, but in many cases the former is a better embodiment. Based on this understanding, the technical solution of the present invention, or the part that contributes to the prior art, can be embodied in the form of a software product, which is stored in a storage medium (such as read-only memory / random access memory, magnetic disk, optical disk), and includes a number of instructions for enabling a terminal device (which can be a mobile phone, computer, server, air conditioner, or network device, etc.) to execute the methods described in each embodiment of the present invention.
[0145] The above are only preferred embodiments of the present invention and are not intended to limit the patent scope of the present invention. Any equivalent structure or equivalent process transformation made using the contents of the present invention description and drawings, or directly or indirectly applied in other related technical fields, are also included in the patent protection scope of the present invention.
Claims
1. A wetland classification method, characterized in that: The wetland classification method comprises the following steps: Preprocess the hyperspectral image and lidar image of the wetland to obtain the hyperspectral feature map and point cloud data feature map; Extracting features from the hyperspectral feature map and the point cloud data feature map to obtain spectral spatial features and elevation features; Regulating the information attention of the spectral spatial feature and the elevation feature to obtain a spectral enhancement feature and an elevation enhancement feature; The spectral enhancement features and the elevation enhancement features are fused and classified to obtain a wetland classification result.
2. The wetland classification method according to claim 1, wherein: The step of fusing and classifying the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result includes: The spectral enhancement feature and the elevation enhancement feature are fused and enhanced to obtain a spectral cross-enhancement feature and an elevation cross-enhancement feature; Selecting a target decision feature from the spectral cross-enhancement feature and the elevation cross-enhancement feature; Determining a complementary fusion feature according to the spectral enhancement feature and the elevation enhancement feature; A wetland classification result is determined based on the target decision feature and the complementary fusion feature.
3. The wetland classification method according to claim 1, wherein: The step of extracting features from the hyperspectral feature map and the point cloud data feature map to obtain spectral spatial features and elevation features comprises: Performing multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result; Perform feature fusion according to the multi-scale feature extraction result to obtain a feature fusion result; Downsampling the feature fusion result to obtain spectral spatial features; Shallow feature extraction is performed on the point cloud data feature map to obtain elevation features.
4. The wetland classification method according to claim 3, wherein: The step of performing multi-scale feature extraction on the hyperspectral feature map to obtain a multi-scale feature extraction result includes: Dividing the hyperspectral feature map according to a preset number of branches to obtain a division result; A convolution operation is performed on each branch according to the division result to obtain a multi-scale feature extraction result.
5. The wetland classification method according to any one of claims 1 to 4, characterized in that: The step of regulating the information attention of the spectral spatial feature and the elevation feature to obtain the spectral enhancement feature and the elevation enhancement feature includes: Performing dimension compression on the spectral spatial feature and the elevation feature respectively to obtain a spectral feature vector and an elevation feature vector; Performing vector compression on the spectral feature vector and the elevation feature vector respectively to obtain a first vector compression result and a second vector compression result; determining a spectrum enhancement feature according to the first vector compression result and the spectrum feature vector; An elevation enhancement feature is determined according to the second vector compression result and the elevation feature vector.
6. The wetland classification method according to any one of claims 1 to 4, characterized in that: The step of preprocessing the hyperspectral image and the lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map includes: The principal component analysis method is used to reduce the feature dimension of the hyperspectral image and lidar image of the wetland to obtain the first image and the second image; Normalizing the first image and the second image respectively to obtain a first normalized image and a second normalized image; Cutting the first normalized image to obtain a hyperspectral feature map; The second normalized image is cut to obtain a point cloud data feature map.
7. A wetland classification device, characterized in that: The wetland classification device comprises: A preprocessing module is used to preprocess the hyperspectral image and lidar image of the wetland to obtain a hyperspectral feature map and a point cloud data feature map; A feature extraction module is used to extract features from the hyperspectral feature map and the point cloud data feature map to obtain spectral spatial features and elevation features; A control module, configured to control the information attention of the spectral spatial feature and the elevation feature to obtain a spectral enhancement feature and an elevation enhancement feature; The fusion classification module is used to fuse and classify the spectral enhancement features and the elevation enhancement features to obtain a wetland classification result.
8. A wetland classification device, characterized in that: The device includes: a memory, a processor, and a wetland classification program stored in the memory and executable on the processor, wherein the wetland classification program is configured to implement the steps of the wetland classification method according to any one of claims 1 to 6.
9. A storage medium, characterized in that: The storage medium stores a wetland classification program, which, when executed by a processor, implements the steps of the wetland classification method according to any one of claims 1 to 6.
10. A computer program product, characterized in that The computer program product comprises a wetland classification program, which implements the steps of the wetland classification method according to any one of claims 1 to 6 when executed by a processor.