Labeling method and system based on grassland remote sensing data

Abstract

The invention provides an annotation method and system based on grassland remote sensing data. Comprising the steps of obtaining multi-source remote sensing data of a grassland; wherein the multi-source remote sensing data comprises spectral data, thermal infrared data and texture information data; performing data alignment and standardization processing on the multi-source remote sensing data; adopting a multi-branch convolutional network to perform feature extraction on the multi-source remote sensing data after data alignment and standardization processing, and performing weighted fusion on features through a cross-modal attention mechanism to obtain a joint feature map; initial labeling is carried out based on the joint feature map, an initial labeling result is obtained, dynamic optimization is carried out in combination with actually measured data of the unmanned aerial vehicle, and an optimized labeling result is obtained; and generating a vector annotation file according to the optimized annotation result. According to the tagging method and system based on the grassland remote sensing data, the tagging efficiency, the tagging precision and the like can be remarkably improved.

Description

Technical Field

[0001] The present invention relates to the technical field of grassland remote sensing data processing, and in particular to a labeling method and system based on grassland remote sensing data. Background Art

[0002] Currently, grassland remote sensing monitoring mainly uses manual visual interpretation or traditional semi-automatic labeling tools, which have significant bottlenecks in terms of efficiency, accuracy, and data utilization. Specifically, it is manifested as follows:

[0003] 1. The contradiction between efficiency and accuracy is prominent.

[0004] Manual labeling requires interpreting image features pixel by pixel, resulting in poor timeliness for large-scale grassland monitoring and difficulty in meeting the needs of dynamic monitoring. Traditional semi-automatic tools rely on a single data source (such as visible light images), have insufficient accuracy in identifying the boundaries of complex landforms such as areas with gradually changing vegetation coverage and desertification edges, and are easily affected by lighting conditions.

[0005] 2. The ability to collaborate with multi-source data is insufficient.

[0006] Existing technologies have not effectively integrated multi-dimensional data features such as multi-spectral (such as NDVI, EVI indices), radar (SAR backscattering coefficient), and high-resolution three-dimensional terrain. For example, visible light data is difficult to penetrate clouds, resulting in missing labels during the rainy season; radar data is sensitive to surface humidity but has not been cross-validated with spectral vegetation indices; the spatio-temporal correlation features of multi-temporal data have not been deeply mined, etc.

[0007] 3. Adaptive defects of intelligent models

[0008] Deep learning models have two major limitations: the training data is limited by specific seasons or meteorological conditions (such as snow cover, dry periods), resulting in misjudgment of features when the model makes cross-period predictions. In addition, the labeling results lack a real-time verification feedback mechanism, and the model parameters cannot be updated through online learning, and errors are prone to accumulate with environmental changes. Summary of the Invention

[0009] The present invention aims to solve at least one of the technical problems existing in the prior art, and provides a labeling method and system based on grassland remote sensing data.

[0010] In a first aspect, the present invention provides a labeling method based on grassland remote sensing data, including:

[0011] Obtaining multi-source remote sensing data of the grassland; wherein, the multi-source remote sensing data includes spectral data, thermal infrared data, and texture information data;

[0012] Performing data alignment and standardization processing on the multi-source remote sensing data;

[0013] Adopt a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and standardization processing, and weighted-fuse the features through a cross-modal attention mechanism to obtain a joint feature map;

[0014] Based on the joint feature map, perform initial annotation to obtain an initial annotation result, and combine it with the actual measured data of the unmanned aerial vehicle for dynamic optimization to obtain an optimized annotation result;

[0015] Generate a vector annotation file from the optimized annotation result.

[0016] In some possible embodiments, the data alignment and standardization processing of the multi-source remote sensing data includes:

[0017] The data alignment processing includes the following steps:

[0018] Geometric correction: Use ground control points for image registration, and select affine transformation, polynomial transformation or DNN registration method to achieve image alignment;

[0019] Temporal registration: Complement the missing images by nearest-time interpolation to ensure the same time for different data sources;

[0020] Spatial resampling: Resample all data to the same resolution using bilinear interpolation, nearest-neighbor interpolation or pyramid downsampling;

[0021] The data standardization processing includes the following steps:

[0022] Radiometric correction: Use the 6S atmospheric correction model to remove the atmospheric influence and convert the DN value to surface reflectance;

[0023] Histogram matching: Unify the image brightness value distributions of different sensors to ensure data consistency;

[0024] Normalization: Use Min-Max normalization or Z-score standardization to process spectral data so that the features of different bands are on the same scale.

[0025] In some possible embodiments, the multi-branch convolutional network includes a spectral feature extraction network, a thermal infrared feature extraction network and a texture feature extraction network; among them,

[0026] The spectral feature extraction network uses a combination of 3×3 and 5×5 convolutional kernels to extract spectral information of different scales; and uses depthwise separable convolution to reduce the computational amount; and uses BN+ReLU to accelerate convergence and prevent gradient disappearance;

[0027] The thermal infrared feature extraction network uses larger convolutional kernels to enhance the receptive field of thermal infrared information and obtain regional thermal anomaly features;

[0028] The texture feature extraction network calculates contrast, entropy, and correlation features using the gray-level co-occurrence matrix and reduces the dimension through 1×1 convolution; and extracts multi-scale and multi-directional texture features using Gabor filtering.

[0029] In some possible embodiments, the multi-branch convolutional network is used to extract features from the multi-source remote sensing data after data alignment and normalization processing, and the features are weighted and fused through a cross-modal attention mechanism to obtain a joint feature map, including:

[0030] The spectral feature extraction network is used to extract vegetation indices and spectral curve features;

[0031] The thermal infrared feature extraction network is used to extract temperature anomaly features;

[0032] The texture feature extraction network is used to extract desertification and degradation pattern features;

[0033] Calculate the weights of spectral-texture, spectral-thermal infrared, and thermal infrared-texture;

[0034] The feature contribution degree is adjusted through the Query-Key-Value mechanism for weighted fusion to obtain the joint feature map.

[0035] In a second aspect, an embodiment of the present invention provides a labeling system based on grassland remote sensing data, including:

[0036] An acquisition module for acquiring multi-source remote sensing data of grasslands; wherein, the multi-source remote sensing data includes spectral data, thermal infrared data, and texture information data;

[0037] A processing module for performing data alignment and normalization processing on the multi-source remote sensing data;

[0038] An extraction module for using a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and normalization processing, and weighting and fusing the features through a cross-modal attention mechanism to obtain a joint feature map;

[0039] An optimization module for performing initial labeling based on the joint feature map to obtain an initial labeling result, and dynamically optimizing it in combination with the actual measured data of the unmanned aerial vehicle to obtain an optimized labeling result;

[0040] A generation module for generating a vector labeling file from the optimized labeling result.

[0041] In some possible embodiments, the processing module is specifically further used for:

[0042] The data alignment processing includes the following steps:

[0043] Geometric correction: Image registration is performed using ground control points, and affine transformation, polynomial transformation, or DNN registration method is selected to achieve image alignment;

[0044] Temporal registration: Missing images are completed by nearest-time interpolation to ensure consistent time for different data sources;

[0045] Spatial resampling: All data is resampled to the same resolution using bilinear interpolation, nearest-neighbor interpolation, or pyramid downsampling;

[0046] The data normalization process includes the following steps:

[0047] Radiometric correction: The 6S atmospheric correction model is used to remove the atmospheric influence and convert the DN value to surface reflectance;

[0048] Histogram matching: The brightness value distributions of images from different sensors are unified to ensure data consistency;

[0049] Normalization: Min-Max normalization or Z-score standardization is used to process spectral data so that the features of different bands are on the same scale.

[0050] In some possible embodiments, the spectral feature extraction network uses a combination of 3×3 and 5×5 convolutional kernels to extract spectral information at different scales; and, depthwise separable convolutions are used to reduce the computational amount; and BN+ReLU is used to accelerate convergence and prevent gradient vanishing;

[0051] The thermal infrared feature extraction network uses larger convolutional kernels to enhance the receptive field of thermal infrared information and obtain regional thermal anomaly features;

[0052] The texture feature extraction network calculates contrast, entropy, and correlation features using the gray-level co-occurrence matrix and reduces the dimension through 1×1 convolution; and Gabor filtering is used to extract multi-scale and multi-directional texture features.

[0053] In some possible embodiments, the extraction module is specifically further configured to:

[0054] Use the spectral feature extraction network to extract vegetation indices and spectral curve features;

[0055] Use the thermal infrared feature extraction network to extract temperature anomaly features;

[0056] Use the texture feature extraction network to extract desertification and degradation pattern features;

[0057] Calculate the weights of spectral-texture, spectral-thermal infrared, and thermal infrared-texture;

[0058] Adjust the feature contribution degree through the Query-Key-Value mechanism for weighted fusion to obtain the combined feature map.

[0059] In a third aspect, the present invention provides an electronic device, comprising:

[0060] One or more processors;

[0061] A storage unit for storing one or more programs, which, when executed by the one or more processors, enable the one or more processors to implement the method described above.

[0062] In a fourth aspect, the present invention provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, it can implement the method described above.

[0063] The annotation method and system based on grassland remote sensing data in the embodiments of the present invention have the following beneficial effects:

[0064] 1. Efficiency improvement: The automation annotation speed is increased by more than 80%, supporting large-scale grassland monitoring.

[0065] 2. Precision enhancement: The classification precision of vegetation coverage reaches more than 95% through multi-source data fusion.

[0066] 3. Dynamic adaptability: Through the real-time verification mechanism, the robustness of the model to seasonal changes is increased by 40%.

[0067] 4. Multidimensional analysis: Support the output of multi-dimensional annotation results such as desertification, degradation, and vegetation type.

[0068] 5. Scalability: The modular design is compatible with new remote sensing data sources (such as hyperspectral satellites). BRIEF DESCRIPTION OF THE DRAWINGS

[0069] Figure 1 It is a schematic structural diagram of an exemplary electronic device for the annotation method based on grassland remote sensing data according to an embodiment of the present invention;

[0070] Figure 2 It is a flowchart of the annotation method based on grassland remote sensing data according to another embodiment of the present invention;

[0071] Figure 3 It is a schematic structural diagram of the annotation system based on grassland remote sensing data according to another embodiment of the present invention. DETAILED DESCRIPTION OF THE EMBODIMENTS

[0072] The technical solution of the present invention will be clearly and completely described below with reference to the accompanying drawings. Apparently, the described embodiments are some, but not all, of the embodiments of the present invention. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0073] Figure 1 FIG. is a schematic structural diagram of an exemplary electronic device for implementing a method for annotating grassland remote sensing data according to an embodiment of the present invention. As Figure 1 shown, the electronic device 100 includes one or more processors 110, one or more storage devices 120, one or more input devices 130, one or more output devices 140, etc. These components are interconnected through a bus system 150 and / or other forms of connection mechanisms. It should be noted that Figure 1 the components and structures of the electronic device shown are exemplary and non-limiting. According to needs, the electronic device may also have other components and structures.

[0074] The processor 110 may be a central processing unit (CPU), or may be composed of multiple processing cores, or may be other forms of processing units with data processing capabilities and / or instruction execution capabilities, and may control other components in the electronic device 100 to execute desired functions.

[0075] The storage device 120 may include one or more computer program products, and the computer program products may include various forms of computer-readable storage media, such as volatile memory and / or non-volatile memory. The volatile memory may include, for example, random access memory (RAM) and / or cache memory, etc. The non-volatile memory may include, for example, read-only memory (ROM), hard disk, flash memory, etc. One or more computer program instructions may be stored on the computer-readable storage medium, and the processor may run the program instructions to implement the client functions (implemented by the processor) in the embodiments of the present disclosure described below and / or other desired functions. Various application programs and various data may also be stored in the computer-readable storage medium. For example, various data used and / or generated by the application programs, etc.

[0076] The input device 130 may be a device used by a user to input instructions, and may include one or more of a keyboard, a mouse, a microphone, and a touch screen, etc.

[0077] The output device 140 may output various information (such as images or sounds) to the outside (such as users), and may include one or more of a display, a speaker, etc.

[0078] Figure 2Flowchart of the annotation method based on grassland remote sensing data according to another embodiment of the present invention. As Figure 2 shown, an annotation method S200 based on grassland remote sensing data includes the following steps S201 to S205:

[0079] Step S201, obtain multi-source remote sensing data of the grassland; wherein, the multi-source remote sensing data includes spectral data, thermal infrared data, and texture information data.

[0080] Step S202, perform data alignment and standardization processing on the multi-source remote sensing data.

[0081] Specifically, in this step, the data alignment processing includes the following steps:

[0082] Geometric correction: Use ground control points for image registration, and select affine transformation, polynomial transformation, or DNN registration method to achieve image alignment;

[0083] Temporal registration: Complement missing images through nearest-time interpolation to ensure consistent time for different data sources;

[0084] Spatial resampling: Resample all data to the same resolution using bilinear interpolation, nearest-neighbor interpolation, or pyramid downsampling.

[0085] The data standardization processing includes the following steps:

[0086] Radiative correction: Use the 6S atmospheric correction model to remove the atmospheric influence and convert the DN value to the surface reflectance;

[0087] Histogram matching: Unify the image brightness value distributions of different sensors to ensure data consistency;

[0088] Normalization: Use Min-Max normalization or Z-score standardization to process the spectral data so that the features of different bands are on the same scale.

[0089] Step S203, adopt a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and standardization processing, and weighted-fuse the features through a cross-modal attention mechanism to obtain a joint feature map.

[0090] Specifically, in this step, the multi-branch convolutional network includes a spectral feature extraction network, a thermal infrared feature extraction network, and a texture feature extraction network to efficiently extract features from different data sources. The spectral feature extraction network uses a combination of 3×3 and 5×5 convolutional kernels to extract spectral information at different scales; in addition, depthwise separable convolution is used to reduce the computational amount; and BN+ReLU is used to accelerate convergence and prevent gradient disappearance. The thermal infrared feature extraction network uses larger convolutional kernels to enhance the receptive field of thermal infrared information and obtain regional thermal anomaly features. The texture feature extraction network calculates contrast, entropy, and correlation features using the gray-level co-occurrence matrix and reduces the dimension through 1×1 convolution; and uses Gabor filtering to extract multi-scale and multi-directional texture features.

[0091] In some embodiments, the multi-branch convolutional network is used to extract features from the multi-source remote sensing data after data alignment and normalization processing, and the features are weighted and fused through a cross-modal attention mechanism to obtain a joint feature map, including: using the spectral feature extraction network to extract vegetation index and spectral curve features; using the thermal infrared feature extraction network to extract temperature anomaly features; using the texture feature extraction network to extract desertification and degradation mode features; calculating the weights of spectrum-texture, spectrum-thermal infrared, and thermal infrared-texture; adjusting the feature contribution degree through the Query-Key-Value mechanism for weighted fusion to obtain the joint feature map.

[0092] Specifically, spectral features reflect information such as vegetation health status and soil moisture content. Common spectral features include vegetation indices, band ratios, etc. In this step, the spectral feature extraction step includes the following contents:

[0093] 1. Calculate vegetation indices (NDVI, EVI, SAVI, etc.)

[0094] Normalized Difference Vegetation Index (NDVI):

[0095]

[0096] Enhanced Vegetation Index (EVI):

[0097]

[0098] 2. Calculate the water index (NDWI):

[0099]

[0100] Reflects soil moisture and vegetation water content.

[0101] 3. Calculate the Soil Adjusted Vegetation Index (SAVI):

[0102]

[0103] Applicable to areas with sparse vegetation to reduce the influence of soil background.

[0104] Texture features reflect the spatial variation pattern of the surface of ground objects and can be extracted by methods such as the gray-level co-occurrence matrix (GLCM) or Gabor filtering. Specifically, in this step, the extraction of texture features includes the following steps:

[0105] Calculate GLCM texture features (gray-level co-occurrence matrix):

[0106] 1. Calculate the following statistical features:

[0107] 1). Contrast: Degree of local variation, used to distinguish different vegetation types.

[0108] 2). Homogeneity: Measure the similarity of adjacent pixels, used to judge the degree of desertification / degradation.

[0109] 3). Entropy: Measure the complexity of the image. A high entropy value indicates a surface with strong heterogeneity (such as bare soil).

[0110] Extract multi-scale texture information by Gabor filtering:

[0111] 1. Convolve with Gabor kernels of different directions and different frequencies to extract the texture of ground objects at different scales.

[0112] Formula:

[0113] · γ controls the directionality of the filter,

[0114] · σ controls the scale,

[0115] Where:

[0116] · f represents the filter frequency.

[0117] · Applicable to extracting the structural features of vegetation leaves and identifying the patterns of grassland degradation.

[0118] There may be problems of information redundancy or uneven weights in different feature data. Therefore, in this step, weighted fusion is performed through the attention mechanism, which specifically includes the following steps:

[0119] Channel attention mechanism (SE-Net)

[0120] 1. Calculate the global average pooling (GAP) of the spectral and texture feature maps:

[0121]

[0122] 2. Map to weights through a fully connected layer:

[0123] α c = σ(W2·ReLU(W1·s[[ID=,5]] c ))

[0124] 3. Weighted spectral and texture features:

[0125] X′ c = α c ·X c

[0126] Cross-Attention

[0127] 1. Construct query (Q), key (K), and value (V) matrices:

[0128] Q = W Q X 光谱 , K = W K X 纹理 , V = W V [[ID=3,6]]X 纹理

[0129] Calculate attention scores:

[0130]

[0131] 2. Weighted fusion features:

[0132] X 融合 = AV + X[[ID=,52]] 光谱 .

[0133] The specific steps to generate the joint feature map are as follows:

[0134] 1. Feature concatenation: Concatenate the fused spectral-texture feature maps along the channel dimension:

[0135] X 联合 = Concat(X 融合 , X 高阶特征 )

[0136] 2. Further feature extraction through CNN to generate the final joint feature map for classification or regression:

[0137] X final = ReLU(BN(Conv(X 联合 )))

[0138] Step S204: Based on the joint feature map, perform initial annotation to obtain the initial annotation result, and combine it with the actual measured data of the UAV for dynamic optimization to obtain the optimized annotation result.

[0139] Step S205: Generate a vector annotation file from the optimized annotation results.

[0140] The annotation method based on grassland remote sensing data in the embodiments of the present invention has the following beneficial effects:

[0141] 1. Efficiency improvement: The automated annotation speed is increased by more than 80%, supporting large-scale grassland monitoring.

[0142] 2. Precision enhancement: The classification precision of vegetation coverage reaches more than 95% through multi-source data fusion.

[0143] 3. Dynamic adaptability: Through a real-time verification mechanism, the robustness of the model to seasonal changes is increased by 40%.

[0144] 4. Multidimensional analysis: Support the output of multi-dimensional annotation results such as desertification, degradation, and vegetation types.

[0145] 5. Scalability: The modular design is compatible with new remote sensing data sources (such as hyperspectral satellites).

[0146] Figure 3 It is a schematic structural diagram of an annotation system based on grassland remote sensing data according to another embodiment of the present invention. The embodiments of the present invention provide an annotation system based on grassland remote sensing data. This annotation system can be applied to the annotation method described above. For specific details, please refer to the relevant descriptions above and will not be elaborated here.

[0147] As Figure 3 shown, the system includes: an acquisition module 301, a processing module 302, an extraction module 303, an optimization module 304, and a generation module 305. The acquisition module 301 is used to acquire multi-source remote sensing data of the grassland; among them, the multi-source remote sensing data includes spectral data, thermal infrared data, and texture information data; the processing module 302 is used to perform data alignment and standardization processing on the multi-source remote sensing data; the extraction module 303 is used to adopt a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and standardization processing, and weighted fusion of features through a cross-modal attention mechanism to obtain a joint feature map; the optimization module 304 is used to perform initial annotation based on the joint feature map to obtain an initial annotation result, and perform dynamic optimization in combination with the actual measured data of the unmanned aerial vehicle to obtain an optimized annotation result; the generation module 305 is used to generate a vector annotation file from the optimized annotation result.

[0148] In some embodiments, as Figure 3 shown, the processing module 303 is specifically further used for:

[0149] The data alignment processing includes the following steps:

[0150] Geometric correction: Image registration is performed using ground control points, and affine transformation, polynomial transformation, or DNN registration method is selected to achieve image alignment;

[0151] Temporal registration: The missing images are completed by nearest-time interpolation to ensure the same time for different data sources;

[0152] Spatial resampling: All data is resampled to the same resolution using bilinear interpolation, nearest-neighbor interpolation, or pyramid downsampling;

[0153] The data normalization process includes the following steps:

[0154] Radiometric correction: The 6S atmospheric correction model is used to remove the atmospheric influence and convert the DN value to surface reflectance;

[0155] Histogram matching: The brightness value distributions of images from different sensors are unified to ensure data consistency;

[0156] Normalization: Min-Max normalization or Z-score standardization is used to process spectral data so that the features of different bands are on the same scale.

[0157] In some embodiments, the spectral feature extraction network uses a combination of 3×3 and 5×5 convolutional kernels to extract spectral information at different scales; and depthwise separable convolution is used to reduce the computational amount; and BN+ReLU is used to accelerate convergence and prevent gradient disappearance; the thermal infrared feature extraction network uses larger convolutional kernels to enhance the receptive field of thermal infrared information and obtain regional thermal anomaly features; the texture feature extraction network uses the gray-level co-occurrence matrix to calculate contrast, entropy, and correlation features, and reduces the dimension through 1×1 convolution; and Gabor filtering is used to extract multi-scale and multi-directional texture features.

[0158] In some embodiments, as Figure 3 shown, the extraction module 303 is specifically further configured to: use the spectral feature extraction network to extract vegetation indices and spectral curve features; use the thermal infrared feature extraction network to extract temperature anomaly features; use the texture feature extraction network to extract desertification and degradation pattern features; calculate the weights of spectral-texture, spectral-thermal infrared, and thermal infrared-texture; adjust the feature contribution degree through the Query-Key-Value mechanism for weighted fusion to obtain the joint feature map.

[0159] The annotation system based on grassland remote sensing data in the embodiments of the present invention has the following beneficial effects:

[0160] 1. Efficiency improvement: The automated annotation speed is increased by more than z%, supporting large-scale grassland monitoring.

[0161] 2. Precision Enhancement: The classification accuracy of vegetation coverage reaches over 95% through multi-source data fusion.

[0162] 3. Dynamic Adaptability: The robustness of the model to seasonal changes is improved by 40% through a real-time verification mechanism.

[0163] 4. Multidimensional Analysis: Support the output of multi-dimensional annotation results such as desertification, degradation, and vegetation types.

[0164] 5. Scalability: The modular design is compatible with new remote sensing data sources (such as hyperspectral satellites).

[0165] Another aspect of the embodiments of the present invention provides a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it can implement the method described above.

[0166] Among them, the computer-readable medium can be included in the device, equipment, and system of the present disclosure, or can exist alone.

[0167] Among them, the computer-readable storage medium can be any tangible medium that contains or stores a program. It can be an electrical, magnetic, optical, electromagnetic, infrared, or semiconductor system, device, or equipment. More specific examples include, but are not limited to: an electrical connection with one or more wires, a portable computer disk, a hard disk, an optical fiber, a random access memory (RAM), a read-only memory (ROM), an erasable programmable read-only memory (EPROM or flash memory), a portable compact disk read-only memory (CD-ROM), an optical storage device, a magnetic storage device, or any suitable combination thereof.

[0168] Among them, the computer-readable storage medium can also include data signals propagated in a baseband or as part of a carrier wave, which carry computer-readable program codes. Specific examples include, but are not limited to, electromagnetic signals, optical signals, or any suitable combination thereof.

[0169] Finally, it should be noted that the above embodiments are only used to illustrate the technical solutions of the present invention, and are not intended to limit them. Although the present invention has been described in detail with reference to the foregoing embodiments, those of ordinary skill in the art should understand that they can still modify the technical solutions described in the foregoing embodiments, or perform equivalent replacements for some or all of the technical features. And these modifications or replacements do not make the essence of the corresponding technical solutions deviate from the scope of the technical solutions of the embodiments of the present invention.

Claims

1. A labeling method based on grassland remote sensing data, characterized in that, Including: Obtain multi-source remote sensing data of grasslands; wherein, the multi-source remote sensing data includes spectral data, thermal infrared data, and texture information data; Perform data alignment and standardization processing on the multi-source remote sensing data; Adopt a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and standardization processing, and perform weighted fusion of features through a cross-modal attention mechanism to obtain a joint feature map; Based on the joint feature map, perform initial annotation to obtain an initial annotation result, and perform dynamic optimization in combination with the actual measured data of the unmanned aerial vehicle to obtain an optimized annotation result; Generate a vector annotation file from the optimized annotation result.

2. The annotation method based on grassland remote sensing data according to claim 1, wherein The performing data alignment and standardization processing on the multi-source remote sensing data includes: The data alignment processing includes the following steps: Geometric correction: Use ground control points for image registration, and select affine transformation, polynomial transformation, or DNN registration method to achieve image alignment; Temporal registration: Complement missing images through nearest-time interpolation to ensure consistent time for different data sources; Spatial resampling: Resample all data to the same resolution using bilinear interpolation, nearest-neighbor interpolation, or pyramid downsampling; The data standardization processing includes the following steps: Radiative correction: Use the 6S atmospheric correction model to remove the atmospheric influence and convert the DN value to the surface reflectance; Histogram matching: Unify the brightness value distribution of images from different sensors to ensure data consistency; Normalization: Use Min-Max normalization or Z-score standardization to process spectral data so that features of different bands are on the same scale.

3. The annotation method based on grassland remote sensing data according to claim 1, characterized in that, The multi-branch convolutional network includes a spectral feature extraction network, a thermal infrared feature extraction network, and a texture feature extraction network; wherein, The spectral feature extraction network uses a combination of 3×3 and 5×5 convolutional kernels to extract spectral information of different scales; and, uses depthwise separable convolution to reduce the computational amount; and adopts BN+ReLU to accelerate convergence and prevent gradient disappearance; The thermal infrared feature extraction network uses larger convolutional kernels to enhance the receptive field of thermal infrared information and obtain regional thermal anomaly features; The texture feature extraction network uses the gray-level co-occurrence matrix to calculate contrast, entropy, and correlation features, and reduces the dimension through 1×1 convolution; and uses Gabor filtering to extract multi-scale and multi-directional texture features.

4. The annotation method based on grassland remote sensing data according to claim 3, wherein The adopting a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and standardization processing, and performing weighted fusion of features through a cross-modal attention mechanism to obtain a joint feature map includes: Use the spectral feature extraction network to extract vegetation index and spectral curve features; Use the thermal infrared feature extraction network to extract temperature anomaly features; Use the texture feature extraction network to extract desertification and degradation pattern features; Calculate the weights of spectrum-texture, spectrum-thermal infrared, and thermal infrared-texture; Adjust the feature contribution degree through the Query-Key-Value mechanism for weighted fusion to obtain the joint feature map.

5. A labeling system based on grassland remote sensing data, characterized in that, Including: An acquisition module for acquiring multi-source remote sensing data of grasslands; wherein, the multi-source remote sensing data includes spectral data, thermal infrared data, and texture information data; A processing module for performing data alignment and normalization processing on the multi-source remote sensing data; An extraction module for using a multi-branch convolutional network to extract features from the multi-source remote sensing data after data alignment and normalization processing, and weighted fusing the features through a cross-modal attention mechanism to obtain a joint feature map; An optimization module for performing initial annotation based on the joint feature map to obtain an initial annotation result, and dynamically optimizing it in combination with the actual measured data of the unmanned aerial vehicle to obtain an optimized annotation result; A generation module for generating a vector annotation file from the optimized annotation result.

6. The annotation system based on grassland remote sensing data according to claim 5, characterized in that Specifically, the processing module is further configured to: The data alignment processing includes the following steps: Geometric correction: Image registration is performed using ground control points, and affine transformation, polynomial transformation or DNN registration method is selected to achieve image alignment; Temporal registration: Missing images are complemented by nearest-time interpolation to ensure that the time of different data sources is consistent; Spatial resampling: All data is resampled to the same resolution using bilinear interpolation, nearest-neighbor interpolation or pyramid downsampling; The data normalization processing includes the following steps: Radiometric correction: The 6S atmospheric correction model is used to remove the atmospheric influence and convert the DN value to the surface reflectance; Histogram matching: The brightness value distributions of images of different sensors are unified to ensure data consistency; Normalization: Min-Max normalization or Z-score standardization is used to process spectral data so that the features of different bands are on the same scale.

7. The annotation system based on grassland remote sensing data according to claim 5, wherein The spectral feature extraction network uses a combination of 3×3 and 5×5 convolutional kernels to extract spectral information of different scales; and, depthwise separable convolution is used to reduce the computational amount; and BN+ReLU is used to accelerate convergence and prevent gradient disappearance; The thermal infrared feature extraction network uses larger convolutional kernels to enhance the receptive field of thermal infrared information and obtain regional thermal anomaly features; The texture feature extraction network calculates contrast, entropy, and correlation features using the gray-level co-occurrence matrix, and reduces the dimension through 1×1 convolution; and Gabor filtering is used to extract multi-scale and multi-directional texture features.

8. The annotation system based on grassland remote sensing data according to claim 7, characterized in that, Specifically, the extraction module is further configured to: Use the spectral feature extraction network to extract vegetation index and spectral curve features; Use the thermal infrared feature extraction network to extract temperature anomaly features; Use the texture feature extraction network to extract desertification and degradation pattern features; Calculate the weights of spectrum-texture, spectrum-thermal infrared, and thermal infrared-texture; Adjust the feature contribution degree through the Query-Key-Value mechanism for weighted fusion to obtain the joint feature map.

9. An electronic device, characterized in that, Including: One or more processors; A storage unit for storing one or more programs, which when executed by the one or more processors, can enable the one or more processors to implement the annotation method based on grassland remote sensing data according to any one of claims 1 to 4.

10. A computer-readable storage medium having a computer program stored thereon, characterized in that, When the computer program is executed by a processor, it can implement the annotation method based on grassland remote sensing data according to any one of claims 1 to 4.

Images