Satellite image recognition method and device based on multi-modal remote sensing data fusion

Abstract

The invention discloses a satellite image recognition method and device based on multi-mode remote sensing data fusion, and relates to the technical field of remote sensing and image processing, and the method comprises the steps: carrying out the preprocessing of an optical remote sensing image, an infrared remote sensing image and a radar remote sensing image, and obtaining a standardized optical feature map, an infrared feature map and a radar feature map, and performing deep feature extraction on the optical feature map, the infrared feature map and the radar feature map based on a cross-modal semantic embedding model to generate unified semantic features, inputting the unified semantic features into an image fusion model, generating a target fusion result based on a multi-source information fusion algorithm, and outputting the target fusion result. And generating a satellite image recognition result according to the target fusion result and the scene complexity evaluation index. Through the mode, the technical problem of low calculation efficiency in multi-modal remote sensing data fusion recognition is solved.

Description

Technical Field

[0001] This application relates to the field of remote sensing and image processing technology, and in particular to a satellite image recognition method and apparatus based on multimodal remote sensing data fusion. Background Technology

[0002] Currently, while combining the characteristics of different modalities in satellite image recognition using multimodal remote sensing data such as visible light, infrared, and synthetic aperture radar (SAR) can improve the accuracy of target detection and recognition, existing fusion methods still have limitations in feature extraction and fusion strategies. For example, in complex scenes, single-modal data is easily affected by environmental factors, leading to insufficient robustness of the recognition results. Furthermore, limited model generalization ability and low computational efficiency also constrain practical applications. These factors pose challenges to multimodal remote sensing data fusion technology in addressing diverse scenario requirements, especially in tasks with high real-time and accuracy demands.

[0003] The above content is only used to help understand the technical solution of the present invention and does not represent an admission that the above content is prior art. Summary of the Invention

[0004] The main objective of this application is to provide a satellite image recognition method and apparatus based on multimodal remote sensing data fusion, aiming to solve the technical problem of low computational efficiency in the existing multimodal remote sensing data fusion recognition technology.

[0005] To achieve the above objectives, this application provides a satellite image recognition method based on multimodal remote sensing data fusion, the method comprising: Optical remote sensing images, infrared remote sensing images, and radar remote sensing images are preprocessed to obtain standardized optical feature maps, infrared feature maps, and radar feature maps, respectively. Based on a cross-modal semantic embedding model, deep feature extraction is performed on the optical feature map, the infrared feature map, and the radar feature map to generate unified semantic features; The unified semantic features are input into the image fusion model, and the target fusion result is generated based on the multi-source information fusion algorithm; Based on the target fusion results and scene complexity evaluation indicators, satellite image recognition results are generated.

[0006] In one embodiment, the step of performing deep feature extraction on the optical feature map, the infrared feature map, and the radar feature map based on a cross-modal semantic embedding model to generate unified semantic features includes: A cross-modal semantic embedding model is constructed, which includes a feature extraction unit and a semantic mapping unit; In the feature extraction unit, local features are extracted from the optical feature map, the infrared feature map, and the radar feature map based on a convolutional neural network to generate corresponding local feature vectors. In the semantic mapping unit, the local feature vectors are weighted and fused through an adaptive weight allocation mechanism to generate unified semantic features.

[0007] In one embodiment, the step of weighted fusion of the local feature vectors through an adaptive weight allocation mechanism in the semantic mapping unit to generate unified semantic features includes: Determine the correlation coefficients between the local feature vectors of the optical feature map, the infrared feature map, and the radar feature map; Calculate the initial weight value of each local feature vector based on the correlation coefficient; The initial weight values ​​are adjusted using an activation function to generate the target weight allocation values; The local feature vectors are weighted and summed based on the target weight allocation values ​​to generate unified semantic features.

[0008] In one embodiment, the step of inputting the unified semantic features into the image fusion model and generating the target fusion result based on the multi-source information fusion algorithm includes: Extract the unified semantic features and input the unified semantic features into the image fusion model; In the image fusion model, semantic features are processed hierarchically based on a multi-scale feature pyramid structure to generate multi-scale feature maps; The multi-scale feature maps are fused layer by layer according to a multi-source information fusion algorithm to generate target candidate regions; Boundary box regression and classification operations are performed on the target candidate regions to generate target fusion results.

[0009] In one embodiment, the step of fusing the multi-scale feature maps layer by layer according to a multi-source information fusion algorithm to generate a target candidate region includes: Spatial attention is calculated for each layer of the feature map in the multi-scale feature map to generate a spatial attention weight matrix; The spatial attention weight matrix is ​​normalized to obtain normalized spatial attention weights; The normalized spatial attention weights are multiplied by the feature map of the corresponding layer to generate an enhanced feature map. The enhanced feature maps from all layers are superimposed to generate target candidate regions.

[0010] In one embodiment, the step of generating satellite image recognition results based on the target fusion result and the scene complexity evaluation index includes: Extract the target category, target location, and target confidence information from the target fusion result; Based on the target category and the target confidence information, a comprehensive score for target detection is calculated; By combining the target location information and scene complexity evaluation metrics, a priority ranking for image fusion is generated; Satellite image recognition results are generated based on the target fusion results, the comprehensive score, and the priority ranking.

[0011] In one embodiment, before the step of preprocessing the optical remote sensing image, infrared remote sensing image, and radar remote sensing image respectively, the method further includes: Determine the geographical scope and time window of the area to be identified; Based on the geographical range and the time window, determine the observation tasks of the optical sensor, infrared sensor, and radar sensor; The observation task is sent to the satellite control system, and the observation data returned by the satellite control system is received. The received observation data is format-converted and preliminarily verified to generate optical remote sensing images, infrared remote sensing images, and radar remote sensing images.

[0012] In one embodiment, the step of preprocessing the optical remote sensing image, infrared remote sensing image, and radar remote sensing image respectively to obtain standardized optical feature maps, infrared feature maps, and radar feature maps includes: Radiometric and geometric corrections are performed on optical remote sensing images to generate standardized optical feature maps; Temperature calibration and noise filtering are performed on infrared remote sensing images to generate standardized infrared feature maps; Polarization decomposition and scattering characteristic analysis are performed on radar remote sensing images to generate standardized radar feature maps.

[0013] In one embodiment, after the step of generating satellite image recognition results based on the target fusion result and the scene complexity evaluation index, the method further includes: Anomaly detection is performed on the satellite image recognition results. When it is determined that there is an abnormal target in the satellite image recognition results, the semantic features of the abnormal target are extracted. The semantic features of the abnormal target are input into the anomaly analysis unit to generate the anomaly information of the abnormal target; An abnormal alarm signal is generated based on the abnormal information.

[0014] Furthermore, to achieve the above objectives, this application also proposes a satellite image recognition device based on multimodal remote sensing data fusion. The satellite image recognition device based on multimodal remote sensing data fusion includes: The image processing module is used to preprocess optical remote sensing images, infrared remote sensing images, and radar remote sensing images respectively to obtain standardized optical feature maps, infrared feature maps, and radar feature maps. The feature extraction module is used to perform deep feature extraction on the optical feature map, the infrared feature map and the radar feature map based on the cross-modal semantic embedding model to generate unified semantic features; The target detection module is used to input the unified semantic features into the image fusion model and generate target fusion results based on the multi-source information fusion algorithm; The result output module is used to generate satellite image recognition results based on the target fusion results and scene complexity evaluation indicators.

[0015] Furthermore, to achieve the above objectives, this application also proposes a satellite image recognition device for multimodal remote sensing data fusion. The satellite image recognition device for multimodal remote sensing data fusion includes: a memory, a processor, and a computer program stored in the memory and executable on the processor. The computer program is configured to implement the steps of the satellite image recognition method for multimodal remote sensing data fusion as described above.

[0016] In addition, to achieve the above objectives, the present invention also proposes a storage medium, which is a computer-readable storage medium, on which a computer program is stored. When the computer program is executed by a processor, it implements the steps of the satellite image recognition method for multimodal remote sensing data fusion as described above.

[0017] In addition, to achieve the above objectives, this application also provides a computer program product, which includes a computer program that, when executed by a processor, implements the steps of the satellite image recognition method for multimodal remote sensing data fusion as described above.

[0018] This application provides a satellite image recognition method based on multimodal remote sensing data fusion. It preprocesses optical, infrared, and radar remote sensing images separately to obtain standardized optical, infrared, and radar feature maps. Deep feature extraction is then performed on these maps using a cross-modal semantic embedding model to generate unified semantic features. These unified semantic features are input into an image fusion model, and a target fusion result is generated based on a multi-source information fusion algorithm. Finally, a satellite image recognition result is generated based on the target fusion result and a scene complexity evaluation index. This method addresses the technical problem of low computational efficiency in multimodal remote sensing data fusion recognition. Attached Figure Description

[0019] The accompanying drawings, which are incorporated in and form part of this specification, illustrate embodiments consistent with this application and, together with the description, serve to explain the principles of this application.

[0020] To more clearly illustrate the technical solutions in the embodiments of this application or the prior art, the drawings used in the description of the embodiments or the prior art will be briefly introduced below. Obviously, for those skilled in the art, other drawings can be obtained based on these drawings without creative effort.

[0021] Figure 1 This is a flowchart illustrating an embodiment of the satellite image recognition method based on multimodal remote sensing data fusion according to this application. Figure 2 This is a schematic diagram of the model architecture of an embodiment of the satellite image recognition method based on multimodal remote sensing data fusion according to this application; Figure 3 This is a schematic diagram of the module structure of the satellite image recognition device for multimodal remote sensing data fusion according to an embodiment of this application; Figure 4 This is a schematic diagram of the equipment structure of the hardware operating environment involved in the satellite image recognition method for multimodal remote sensing data fusion in the embodiments of this application.

[0022] The realization of the purpose, functional features and advantages of this application will be further explained in conjunction with the embodiments and with reference to the accompanying drawings. Detailed Implementation

[0023] It should be understood that the specific embodiments described herein are merely illustrative of the technical solutions of this application and are not intended to limit this application.

[0024] To better understand the technical solution of this application, a detailed description will be provided below in conjunction with the accompanying drawings and specific implementation methods.

[0025] The main solution of this application embodiment is as follows: preprocessing optical remote sensing images, infrared remote sensing images, and radar remote sensing images respectively to obtain standardized optical feature maps, infrared feature maps, and radar feature maps; performing deep feature extraction on the optical feature maps, infrared feature maps, and radar feature maps based on a cross-modal semantic embedding model to generate unified semantic features; inputting the unified semantic features into an image fusion model to generate target fusion results based on a multi-source information fusion algorithm; and generating satellite image recognition results based on the target fusion results and scene complexity evaluation indicators.

[0026] Currently, while combining the characteristics of different modalities in satellite image recognition using multimodal remote sensing data such as visible light, infrared, and synthetic aperture radar (SAR) can improve the accuracy of target detection and recognition, existing fusion methods still have limitations in feature extraction and fusion strategies. For example, in complex scenes, single-modal data is easily affected by environmental factors, leading to insufficient robustness of the recognition results. Furthermore, limited model generalization ability and low computational efficiency also constrain practical applications. These factors pose challenges to multimodal remote sensing data fusion technology in addressing diverse scenario requirements, especially in tasks with high real-time and accuracy demands.

[0027] This application provides a solution that preprocesses optical, infrared, and radar remote sensing images to obtain standardized optical, infrared, and radar feature maps. Based on a cross-modal semantic embedding model, deep feature extraction is performed on these optical, infrared, and radar feature maps to generate unified semantic features. These unified semantic features are then input into an image fusion model, and a target fusion result is generated based on a multi-source information fusion algorithm. Finally, based on the target fusion result and a scene complexity evaluation index, a satellite image recognition result is generated. This approach addresses the technical problem of low computational efficiency in multimodal remote sensing data fusion and recognition.

[0028] It should be noted that the executing entity in this embodiment can be a computing service device with data processing, network communication, and program execution functions, such as a tablet computer, personal computer, or mobile phone, or an electronic device capable of performing the above functions, a satellite image recognition device for multimodal remote sensing data fusion, etc. This embodiment does not specifically limit it. The following uses a satellite image recognition device for multimodal remote sensing data fusion as an example to describe this embodiment and the following embodiments.

[0029] All actions involving the acquisition of signals, information, or data in this application are carried out in accordance with the relevant data protection laws and policies of the country where the application is located, and with the authorization of the owner of the relevant device.

[0030] This application provides a satellite image recognition method based on multimodal remote sensing data fusion, referring to... Figure 1 , Figure 1 This is a flowchart illustrating the first embodiment of the satellite image recognition method based on multimodal remote sensing data fusion according to this application.

[0031] In this embodiment, the satellite image recognition method based on multimodal remote sensing data fusion includes steps S10 to S40: Step S10: Preprocess the optical remote sensing image, infrared remote sensing image and radar remote sensing image respectively to obtain standardized optical feature map, infrared feature map and radar feature map; It should be noted that optical remote sensing images utilize the electromagnetic wave reflection characteristics of the visible and near-infrared bands (typically 0.4μm~1.0μm) to capture surface images through satellite / airborne optical sensors. Infrared remote sensing images are generated by sensors receiving the temperature distribution shadows produced by the mid-to-long-wave infrared radiation (typically 3μm~14μm) emitted by objects on the ground. Radar remote sensing images utilize synthetic aperture radar (SAR) to emit microwaves (common bands: X / C / L) and receive backscattered signals from the ground to generate active imaging data.

[0032] Additionally, it should be noted that optical feature maps are preprocessed, standardized optical remote sensing data that represent the geometric structure and reflectivity characteristics of surface objects in the form of a pixel matrix. Infrared feature maps refer to standardized infrared radiation data, characterizing the thermal properties of targets using temperature (K) or radiation flux density (W / m²). Radar feature maps are multidimensional feature matrices generated from polarimetric SAR data after scattering characteristic analysis, reflecting the dielectric properties and structural information of the Earth's surface.

[0033] It is understood that the steps of preprocessing optical remote sensing images, infrared remote sensing images, and radar remote sensing images respectively to obtain standardized optical feature maps, infrared feature maps, and radar feature maps include: Radiometric and geometric corrections are performed on optical remote sensing images to generate standardized optical feature maps; Temperature calibration and noise filtering are performed on infrared remote sensing images to generate standardized infrared feature maps; Polarization decomposition and scattering characteristic analysis are performed on radar remote sensing images to generate standardized radar feature maps.

[0034] In practical implementation, to eliminate the influence of atmospheric heat dissipation and illumination and obtain the true surface reflectance, optical remote sensing images can be preprocessed, including radiometric and geometric corrections. For radiometric correction, when calculating surface reflectance... When, the calculation formula is:

[0035] in, For the sensor to receive radiance, Atmospheric path radiation, This is the Earth-Sun distance correction factor. Solar irradiance outside the atmosphere. The zenith angle of the sun. For gas transmittance. In the parameter constraints, the camera radiometric resolution is 12-bit quantization, and the static MTF ≥ 0.15.

[0036] To eliminate distortions caused by terrain undulations and satellite attitude, geometric correction can be performed. Geometric correction is based on the orthorectification formula, which is:

[0037] in, For rotation matrix, For attitude angle, For camera focal length, This is the satellite position vector. The positioning error is ≤ 1 pixel.

[0038] When preprocessing infrared remote sensing images, the DN value can be converted into temperature based on Planck's radiation law. The conversion formula is as follows:

[0039]

[0040] in, For band Radiance This is the surface emissivity, typically taken as 0.95 to 0.99. It is Planck's constant. Boltzmann coefficient, At the speed of light, Absolute temperature These are calibration coefficients. The temperature value can be determined by solving the equation using Newton's iterative method.

[0041] When performing noise filtering, wavelet transform filtering can be used to suppress thermal noise. The filtering process can be expressed as:

[0042] in, The images are the image after wavelet transform filtering and the image before filtering. For wavelet transform / inverse transform, For threshold function, , The threshold is determined by the noise variance. estimate.

[0043] Polarization decomposition is performed on radar remote sensing image preprocessing to convert the fully polarimetric scattering matrix. .

[0044]

[0045] The covariance matrix was calculated using Cloude decomposition. Specifically;

[0046] Then feature decomposition is performed. , The characteristic matrix vectors satisfy the unitary matrix property. diagonal matrix , These are the scattering energies of the dominant scattering mechanism (such as specular reflection), the secondary scattering mechanism (such as dihedral scattering), and the random scattering energy (such as vegetation volume scattering).

[0047] Extract parameters, entropy average scattering angle .

[0048] Then, scattering characteristics analysis is performed to determine the scattering power and anisotropy.

[0049] Scattering power: .

[0050] Anisotropy: .

[0051] Scattering power and anisotropy determine the intensity and directional features of each pixel in a radar image. When generating a standardized radar feature map, scattering power provides a unified intensity benchmark for the feature map through normalization, while anisotropy adds directional features to the feature map, enabling the standardized radar feature map to more comprehensively reflect the physical characteristics of the target.

[0052] In one feasible implementation, before the step of preprocessing the optical remote sensing image, infrared remote sensing image, and radar remote sensing image respectively, the method further includes: Determine the geographical scope and time window of the area to be identified; Based on the geographical range and the time window, determine the observation tasks of the optical sensor, infrared sensor, and radar sensor; The observation task is sent to the satellite control system, and the observation data returned by the satellite control system is received. The received observation data is format-converted and preliminarily verified to generate optical remote sensing images, infrared remote sensing images, and radar remote sensing images.

[0053] It is important to note that the geographical extent and time window of the area to be identified are key parameters for mission planning. The geographical extent defines the specific boundaries of the area to be monitored, typically determined by geographic coordinates (such as latitude and longitude), e.g., urban areas, mountain ranges, or ocean areas. The time window specifies the monitoring period or frequency, such as a specific date range, a time of day, or a periodic monitoring cycle. These two parameters together determine the scope and timing of the observation mission, ensuring that satellite sensors collect data within the designated area and time frame.

[0054] In practice, the geographical scope and time window of the area to be identified must first be determined. Based on this, the observation tasks for optical, infrared, and radar sensors are determined, and these tasks are sent to the satellite control system to receive the observation data. Subsequently, the received data undergoes format conversion and preliminary verification to generate standardized optical, infrared, and radar remote sensing images. This process ensures the accuracy and effectiveness of data acquisition, laying the foundation for subsequent feature extraction and fusion processing.

[0055] Step S20: Based on the cross-modal semantic embedding model, perform deep feature extraction on the optical feature map, the infrared feature map, and the radar feature map to generate unified semantic features; It should be noted that cross-modal semantic embedding models are a type of deep learning model that can map data from different modalities into a shared semantic space. Unified semantic features refer to the feature representations obtained by mapping the local feature vectors of optical, infrared, and radar remote sensing images into the same semantic space after processing by a cross-modal semantic embedding model.

[0056] It is understood that the step of performing deep feature extraction on the optical feature map, the infrared feature map, and the radar feature map based on the cross-modal semantic embedding model to generate unified semantic features includes: A cross-modal semantic embedding model is constructed, which includes a feature extraction unit and a semantic mapping unit; In the feature extraction unit, local features are extracted from the optical feature map, the infrared feature map, and the radar feature map based on a convolutional neural network to generate corresponding local feature vectors. In the semantic mapping unit, the local feature vectors are weighted and fused through an adaptive weight allocation mechanism to generate unified semantic features.

[0057] In the specific implementation, refer to Figure 2 , Figure 2 This is a schematic diagram of the model architecture. Optical feature maps, infrared feature maps, and radar feature maps are input to the feature extraction unit. The feature extraction unit uses a multi-branch CNN, with each branch adapting to a different modality. Optical branch: Extracting spatial texture through 3×3 convolution; Infrared branch: Temperature channels are compressed via 1×1 convolution; Radar branch: The scattering matrix is ​​processed through polarimetric convolution.

[0058] The feature extraction unit outputs dimension-aligned local feature vectors.

[0059] After obtaining the local feature vectors, these vectors are input into the semantic mapping unit. A convolutional neural network then extracts local features from the feature maps of different modalities, generating corresponding local feature vectors. Finally, an adaptive weight allocation mechanism is used to weight and fuse these local feature vectors, ultimately generating a unified semantic feature.

[0060] In one feasible implementation, the step of weighted fusing the local feature vectors through an adaptive weight allocation mechanism in the semantic mapping unit to generate unified semantic features includes: Determine the correlation coefficients between the local feature vectors of the optical feature map, the infrared feature map, and the radar feature map; Calculate the initial weight value of each local feature vector based on the correlation coefficient; The initial weight values ​​are adjusted using an activation function to generate the target weight allocation values; The local feature vectors are weighted and summed based on the target weight allocation values ​​to generate unified semantic features.

[0061] In the specific implementation, the correlation coefficient between the local eigenvectors of the optical feature map, infrared feature map, and radar feature map is calculated. The formula for calculating the correlation coefficient is:

[0062]

[0063] in, For local feature vectors, These are optical feature maps, infrared feature maps, and radar feature maps, respectively.

[0064] Calculate the initial weight values ​​of each local feature vector based on the correlation coefficient. The formula for calculating the initial weight value is:

[0065] The initial weight values ​​are adjusted using an activation function to generate the target weight allocation values. When this happens, it can be expressed as: the initial weight value was not used.

[0066] ,

[0067] in, It is the Sigmoid function. It is the scene complexity factor. and These are learnable parameters.

[0068] The local feature vectors are weighted and summed according to the target weight allocation value to generate a unified semantic feature. By weighting and summing, the local feature vectors of different modalities are fused into a unified semantic feature vector, which can more comprehensively represent the semantic information of the input data.

[0069] Step S30: Input the unified semantic features into the image fusion model, and generate the target fusion result based on the multi-source information fusion algorithm; It should be noted that image fusion models are a type of deep learning model that integrates feature information from different modalities. This model typically contains multiple processing layers, enabling multi-scale processing and layer-by-layer fusion of the input features.

[0070] It is understood that the step of inputting the unified semantic features into the image fusion model and generating the target fusion result based on the multi-source information fusion algorithm includes: Extract the unified semantic features and input the unified semantic features into the image fusion model; In the image fusion model, semantic features are processed hierarchically based on a multi-scale feature pyramid structure to generate multi-scale feature maps; The multi-scale feature maps are fused layer by layer according to a multi-source information fusion algorithm to generate target candidate regions; Boundary box regression and classification operations are performed on the target candidate regions to generate target fusion results.

[0071] In its implementation, the satellite image recognition method based on multimodal remote sensing data fusion first inputs unified semantic features generated through a cross-modal semantic embedding model into the image fusion model. Within the model, the unified semantic features are processed hierarchically based on a multi-scale feature pyramid structure. Downsampling and upsampling operations generate feature maps at different scales, thereby capturing target information at different scales. Next, a multi-source information fusion algorithm is used to fuse these multi-scale feature maps layer by layer. Specifically, this involves calculating the spatial attention weight matrix of each layer's feature map, normalizing the weight matrix, and multiplying the normalized weights with the corresponding layer's feature map to generate an enhanced feature map. Finally, the enhanced feature maps from all layers are superimposed to generate the target candidate region.

[0072] Then, bounding box regression and classification operations are performed on the generated target candidate regions. Bounding box regression adjusts the position and size of the bounding boxes of the candidate regions by calculating offsets, thereby locating the targets more accurately; the classification operation uses the Softmax function to predict the probability distribution of the target category and determine the specific category of the target. Finally, the results of bounding box regression and classification are integrated to generate a target fusion result containing target location, category, and confidence information, providing comprehensive and accurate target information for satellite image recognition.

[0073] In one feasible implementation, the step of fusing the multi-scale feature maps layer by layer according to a multi-source information fusion algorithm to generate target candidate regions includes: Spatial attention is calculated for each layer of the feature map in the multi-scale feature map to generate a spatial attention weight matrix; The spatial attention weight matrix is ​​normalized to obtain normalized spatial attention weights; The normalized spatial attention weights are multiplied by the feature map of the corresponding layer to generate an enhanced feature map. The enhanced feature maps from all layers are superimposed to generate target candidate regions.

[0074] In the specific implementation, during the multi-scale feature map processing of remote sensing images, spatial attention calculation is first performed on each layer of the feature map to generate a corresponding spatial attention weight matrix, which identifies the importance of each pixel position. Then, the spatial attention weight matrix is ​​normalized to obtain normalized spatial attention weights, ensuring that the weight values ​​are within a uniform range and enhancing the model's stability and performance. Next, the normalized spatial attention weights are multiplied by the original feature map to generate an enhanced feature map, highlighting the features of key regions. Finally, the enhanced feature maps from each layer are superimposed to generate the target candidate region.

[0075] Step S40: Generate satellite image recognition results based on the target fusion results and scene complexity evaluation indicators.

[0076] It should be noted that the target fusion result is an intermediate result generated after processing by an image fusion model. It includes target category, location, and confidence information, and integrates feature information from multimodal data to provide data support for satellite image recognition. Scene complexity evaluation metrics are used to measure the complexity of a scene in a satellite image, including the number of targets, background clutter intensity, and target-background contrast, etc., to help assess the difficulty of the recognition task.

[0077] It is understood that the step of generating satellite image recognition results based on the target fusion result and the scene complexity evaluation index includes: Extract the target category, target location, and target confidence information from the target fusion result; Based on the target category and the target confidence information, a comprehensive score for target detection is calculated; By combining the target location information and scene complexity evaluation metrics, a priority ranking for image fusion is generated; Satellite image recognition results are generated based on the target fusion results, the comprehensive score, and the priority ranking.

[0078] In the specific implementation, the target category, location, and confidence information are first extracted from the target fusion result. Here, target category refers to the type of target, target location refers to the target's coordinates in the image, and target confidence information reflects the reliability of target recognition. Next, based on the target category and confidence information, a comprehensive score for target detection is calculated, using the following formula: .

[0079] Subsequently, by combining target location information and scene complexity evaluation metrics, a priority ranking for image fusion is generated to measure the processing priority of each region. Finally, by combining the target fusion results, the overall score, and the priority ranking, an accurate satellite image recognition result is generated.

[0080] After the step of generating satellite image recognition results based on the target fusion result and the scene complexity evaluation index, the method further includes: Anomaly detection is performed on the satellite image recognition results. When it is determined that there is an abnormal target in the satellite image recognition results, the semantic features of the abnormal target are extracted. The semantic features of the abnormal target are input into the anomaly analysis unit to generate the anomaly information of the abnormal target; An abnormal alarm signal is generated based on the abnormal information.

[0081] In its implementation, the system first employs anomaly detection algorithms to comprehensively screen the recognition results, determining whether any abnormal targets deviate from the expected patterns. Once an abnormal target is detected, its semantic features are immediately extracted; these features accurately describe the characteristics of the abnormal target. Then, the extracted semantic features are input into an anomaly analysis unit. This unit performs deep analysis of the semantic features to generate anomaly information containing the anomaly type, severity, and potential impact range. Finally, based on the anomaly information, an anomaly alarm signal is triggered and generated, promptly alerting relevant personnel to pay attention and take action, thereby effectively improving the security and practicality of the satellite image recognition system.

[0082] This embodiment provides a satellite image recognition method based on multimodal remote sensing data fusion. It preprocesses optical, infrared, and radar remote sensing images separately to obtain standardized optical, infrared, and radar feature maps. Deep feature extraction is then performed on these maps using a cross-modal semantic embedding model to generate unified semantic features. These unified semantic features are input into an image fusion model, and a target fusion result is generated based on a multi-source information fusion algorithm. Finally, a satellite image recognition result is generated based on the target fusion result and a scene complexity evaluation index. This method addresses the technical problem of low computational efficiency in multimodal remote sensing data fusion and recognition.

[0083] It should be noted that the above examples are only for understanding this application and do not constitute a limitation on the satellite image recognition method of multimodal remote sensing data fusion in this application. Any simple modifications based on this technical concept are within the protection scope of this application.

[0084] This application also provides a satellite image recognition device based on multimodal remote sensing data fusion; please refer to [reference needed]. Figure 3 The satellite image recognition device for multimodal remote sensing data fusion includes: Image processing module 10 is used to preprocess optical remote sensing images, infrared remote sensing images and radar remote sensing images respectively to obtain standardized optical feature maps, infrared feature maps and radar feature maps; The feature extraction module 20 is used to perform deep feature extraction on the optical feature map, the infrared feature map and the radar feature map based on the cross-modal semantic embedding model to generate unified semantic features; The target detection module 30 is used to input the unified semantic features into the image fusion model and generate the target fusion result based on the multi-source information fusion algorithm; The result output module 40 is used to generate satellite image recognition results based on the target fusion results and scene complexity evaluation indicators.

[0085] In one feasible implementation, the feature extraction module 20 is further configured to construct a cross-modal semantic embedding model, the cross-modal semantic embedding model including a feature extraction unit and a semantic mapping unit; In the feature extraction unit, local features are extracted from the optical feature map, the infrared feature map, and the radar feature map based on a convolutional neural network to generate corresponding local feature vectors. In the semantic mapping unit, the local feature vectors are weighted and fused through an adaptive weight allocation mechanism to generate unified semantic features.

[0086] In one feasible implementation, the feature extraction module 20 is further configured to determine the correlation coefficient between the local feature vectors of the optical feature map, the infrared feature map, and the radar feature map; Calculate the initial weight value of each local feature vector based on the correlation coefficient; The initial weight values ​​are adjusted using an activation function to generate the target weight allocation values; The local feature vectors are weighted and summed based on the target weight allocation values ​​to generate unified semantic features.

[0087] In one feasible implementation, the target detection module 30 is further configured to extract the unified semantic features and input the unified semantic features into the image fusion model; In the image fusion model, semantic features are processed hierarchically based on a multi-scale feature pyramid structure to generate multi-scale feature maps; The multi-scale feature maps are fused layer by layer according to a multi-source information fusion algorithm to generate target candidate regions; Boundary box regression and classification operations are performed on the target candidate regions to generate target fusion results.

[0088] In one feasible implementation, the target detection module 30 is further configured to perform spatial attention calculation on each layer of the feature map in the multi-scale feature map to generate a spatial attention weight matrix. The spatial attention weight matrix is ​​normalized to obtain normalized spatial attention weights; The normalized spatial attention weights are multiplied by the feature map of the corresponding layer to generate an enhanced feature map. The enhanced feature maps from all layers are superimposed to generate target candidate regions.

[0089] In one feasible implementation, the result output module 40 is further used to extract target category, target location and target confidence information from the target fusion result; Based on the target category and the target confidence information, a comprehensive score for target detection is calculated; By combining the target location information and scene complexity evaluation metrics, a priority ranking for image fusion is generated; Satellite image recognition results are generated based on the target fusion results, the comprehensive score, and the priority ranking.

[0090] In one feasible implementation, the image processing module 10 is further configured to determine the geographical extent and time window of the area to be identified; Based on the geographical range and the time window, determine the observation tasks of the optical sensor, infrared sensor, and radar sensor; The observation task is sent to the satellite control system, and the observation data returned by the satellite control system is received. The received observation data is format-converted and preliminarily verified to generate optical remote sensing images, infrared remote sensing images, and radar remote sensing images.

[0091] In one feasible implementation, the image processing module 10 is further configured to perform radiometric and geometric correction on the optical remote sensing image to generate a standardized optical feature map. Temperature calibration and noise filtering are performed on infrared remote sensing images to generate standardized infrared feature maps; Polarization decomposition and scattering characteristic analysis are performed on radar remote sensing images to generate standardized radar feature maps.

[0092] In one feasible implementation, the result output module 40 is further configured to perform anomaly detection on the satellite image recognition result, and when it is determined that there is an abnormal target in the satellite image recognition result, extract the semantic features of the abnormal target; The semantic features of the abnormal target are input into the anomaly analysis unit to generate the anomaly information of the abnormal target; An abnormal alarm signal is generated based on the abnormal information.

[0093] The satellite image recognition device for multimodal remote sensing data fusion provided in this application, employing the satellite image recognition method for multimodal remote sensing data fusion in the above embodiments, can solve the technical problem of low computational efficiency in multimodal remote sensing data fusion recognition. Compared with the prior art, the beneficial effects of the satellite image recognition device for multimodal remote sensing data fusion provided in this application are the same as those of the satellite image recognition method for multimodal remote sensing data fusion provided in the above embodiments, and other technical features in the satellite image recognition device for multimodal remote sensing data fusion are the same as those disclosed in the methods of the above embodiments, and will not be repeated here.

[0094] This application provides a satellite image recognition device for multimodal remote sensing data fusion. The satellite image recognition device for multimodal remote sensing data fusion includes: at least one processor; and a memory communicatively connected to the at least one processor; wherein the memory stores instructions executable by the at least one processor, and the instructions are executed by the at least one processor to enable the at least one processor to perform the satellite image recognition method for multimodal remote sensing data fusion in the above embodiment 1.

[0095] The following is for reference. Figure 4This document illustrates a structural schematic diagram of a satellite image recognition device suitable for implementing multimodal remote sensing data fusion in the embodiments of this application. The satellite image recognition device for multimodal remote sensing data fusion in the embodiments of this application may include, but is not limited to, mobile terminals such as mobile phones, laptops, digital broadcast receivers, PDAs (Personal Digital Assistants), PADs (Portable Application Description), PMPs (Portable Media Players), and in-vehicle terminals (e.g., in-vehicle navigation terminals), as well as fixed terminals such as digital TVs and desktop computers. Figure 4 The satellite image recognition device for multimodal remote sensing data fusion shown is merely an example and should not impose any limitations on the functionality and scope of use of the embodiments of this application.

[0096] like Figure 4 As shown, a satellite image recognition device for multimodal remote sensing data fusion may include a processing unit 1001 (e.g., a central processing unit, a graphics processing unit, etc.), which can perform various appropriate actions and processes according to a program stored in ROM (Read Only Memory) 1002 or a program loaded from storage device 1003 into RAM (Random Access Memory) 1004. RAM 1004 also stores various programs and data required for the operation of the satellite image recognition device for multimodal remote sensing data fusion. The processing unit 1001, ROM 1002, and RAM 1004 are interconnected via bus 1005. Input / output (I / O) interface 1006 is also connected to the bus. Typically, the following systems can be connected to I / O interface 1006: input devices 1007 including, for example, touchscreens, touchpads, keyboards, mice, image sensors, microphones, accelerometers, gyroscopes, etc.; output devices 1008 including, for example, LCDs (Liquid Crystal Displays), speakers, vibrators, etc.; storage devices 1003 including, for example, magnetic tapes, hard disks, etc.; and communication devices 1009. Communication device 1009 allows the satellite image recognition equipment for multimodal remote sensing data fusion to communicate wirelessly or wiredly with other devices to exchange data. Although the figure shows a satellite image recognition equipment for multimodal remote sensing data fusion with various systems, it should be understood that it is not required to implement or possess all the systems shown. More or fewer systems can be implemented alternatively.

[0097] Specifically, according to the embodiments disclosed in this application, the processes described above with reference to the flowcharts can be implemented as computer software programs. For example, embodiments disclosed in this application include a computer program product comprising a computer program carried on a computer-readable medium, the computer program containing program code for performing the methods shown in the flowcharts. In such embodiments, the computer program can be downloaded and installed from a network via a communication device, or installed from storage device 1003, or installed from ROM 1002. When the computer program is executed by processing device 1001, it performs the functions defined in the methods of the embodiments disclosed in this application.

[0098] The satellite image recognition device based on multimodal remote sensing data fusion provided in this application, employing the satellite image recognition method based on multimodal remote sensing data fusion described in the above embodiments, can solve the technical problems of satellite image recognition based on multimodal remote sensing data fusion. Compared with the prior art, the beneficial effects of the satellite image recognition device based on multimodal remote sensing data fusion provided in this application are the same as those of the satellite image recognition method based on multimodal remote sensing data fusion described in the above embodiments, and other technical features in this satellite image recognition device based on multimodal remote sensing data fusion are the same as those disclosed in the method of the previous embodiment, and will not be repeated here.

[0099] It should be understood that the various parts disclosed in this application can be implemented using hardware, software, firmware, or a combination thereof. In the description of the above embodiments, specific features, structures, materials, or characteristics can be combined in any suitable manner in one or more embodiments or examples.

[0100] The above are merely specific embodiments of this application, but the scope of protection of this application is not limited thereto. Any variations or substitutions that can be easily conceived by those skilled in the art within the scope of the technology disclosed in this application should be included within the scope of protection of this application. Therefore, the scope of protection of this application should be determined by the scope of the claims.

[0101] This application provides a computer-readable storage medium having computer-readable program instructions (i.e., a computer program) stored thereon, which are used to execute the satellite image recognition method for multimodal remote sensing data fusion in the above embodiments.

[0102] The computer-readable storage medium provided in this application may be, for example, a USB flash drive, but is not limited to, electrical, magnetic, optical, electromagnetic, infrared, or semiconductor systems, devices, or any combination thereof. More specific examples of computer-readable storage media may include, but are not limited to: electrical connections having one or more wires, portable computer disks, hard disks, RAM (Random Access Memory), ROM (Read Only Memory), EPROM (Erasable Programmable Read Only Memory or Flash Memory), optical fibers, CD-ROM (CD-Read Only Memory), optical storage devices, magnetic storage devices, or any suitable combination thereof. In this embodiment, the computer-readable storage medium may be any tangible medium containing or storing a program that can be used by or in conjunction with an instruction execution system, system, or device. The program code contained on the computer-readable storage medium may be transmitted using any suitable medium, including but not limited to: wires, optical cables, RF (Radio Frequency), etc., or any suitable combination thereof.

[0103] The aforementioned computer-readable storage medium may be included in a satellite image recognition device for multimodal remote sensing data fusion; or it may exist independently and not be assembled into a satellite image recognition device for multimodal remote sensing data fusion.

[0104] The aforementioned computer-readable storage medium carries one or more programs that, when executed by a satellite image recognition device that fuses multimodal remote sensing data, cause the satellite image recognition device to preprocess optical remote sensing images, infrared remote sensing images, and radar remote sensing images respectively to obtain standardized optical feature maps, infrared feature maps, and radar feature maps. Based on a cross-modal semantic embedding model, deep feature extraction is performed on the optical feature map, the infrared feature map, and the radar feature map to generate unified semantic features; The unified semantic features are input into the image fusion model, and the target fusion result is generated based on the multi-source information fusion algorithm; Based on the target fusion results and scene complexity evaluation indicators, satellite image recognition results are generated.

[0105] Computer program code for performing the operations of this application can be written in one or more programming languages ​​or a combination thereof, including object-oriented programming languages ​​such as Java, Smalltalk, and C++, as well as conventional procedural programming languages ​​such as the "C" language or similar programming languages. The program code can be executed entirely on the user's computer, partially on the user's computer, as a standalone software package, partially on the user's computer and partially on a remote computer, or entirely on a remote computer or server. In cases involving remote computers, the remote computer can be connected to the user's computer via any type of network—including LAN (Local Area Network) or WAN (Wide Area Network)—or can be connected to an external computer (e.g., via the Internet using an Internet service provider).

[0106] The flowcharts and block diagrams in the accompanying drawings illustrate the architecture, functionality, and operation of possible implementations of systems, methods, and computer program products according to various embodiments of this application. In this regard, each block in a flowchart or block diagram may represent a module, segment, or portion of code containing one or more executable instructions for implementing a specified logical function. It should also be noted that in some alternative implementations, the functions indicated in the blocks may occur in a different order than those indicated in the drawings. For example, two consecutively indicated blocks may actually be executed substantially in parallel, and they may sometimes be executed in reverse order, depending on the functions involved. It should also be noted that each block in the block diagrams and / or flowcharts, and combinations of blocks in the block diagrams and / or flowcharts, can be implemented using a dedicated hardware-based system that performs the specified function or operation, or using a combination of dedicated hardware and computer instructions.

[0107] The modules described in the embodiments of this application can be implemented in software or hardware. The names of the modules do not necessarily limit the functionality of the unit itself.

[0108] The readable storage medium provided in this application is a computer-readable storage medium that stores computer-readable program instructions (i.e., a computer program) for executing the above-described satellite image recognition method based on multimodal remote sensing data fusion, and is capable of solving the technical problem of satellite image recognition based on multimodal remote sensing data fusion. Compared with the prior art, the beneficial effects of the computer-readable storage medium provided in this application are the same as the beneficial effects of the satellite image recognition method based on multimodal remote sensing data fusion provided in the above embodiments, and will not be repeated here.

[0109] This application also provides a computer program product, including a computer program that, when executed by a processor, implements the steps of the satellite image recognition method for multimodal remote sensing data fusion as described above.

[0110] The computer program product provided in this application can solve the technical problem of satellite image recognition based on multimodal remote sensing data fusion. Compared with the prior art, the beneficial effects of the computer program product provided in this application are the same as those of the satellite image recognition method based on multimodal remote sensing data fusion provided in the above embodiments, and will not be repeated here.

[0111] The above are only some embodiments of this application and do not limit the patent scope of this application. All equivalent structural transformations made under the technical concept of this application and using the contents of the specification and drawings of this application, or direct / indirect applications in other related technical fields, are included in the patent protection scope of this application.

Claims

1. A satellite image recognition method based on multimodal remote sensing data fusion, characterized in that, The satellite image recognition method based on multimodal remote sensing data fusion includes: Optical remote sensing images, infrared remote sensing images, and radar remote sensing images are preprocessed to obtain standardized optical feature maps, infrared feature maps, and radar feature maps, respectively. Based on a cross-modal semantic embedding model, deep feature extraction is performed on the optical feature map, the infrared feature map, and the radar feature map to generate unified semantic features; The unified semantic features are input into the image fusion model, and the target fusion result is generated based on the multi-source information fusion algorithm; Based on the target fusion results and scene complexity evaluation indicators, satellite image recognition results are generated.

2. The method as described in claim 1, characterized in that, The steps of performing deep feature extraction on the optical feature map, the infrared feature map, and the radar feature map based on a cross-modal semantic embedding model to generate unified semantic features include: A cross-modal semantic embedding model is constructed, which includes a feature extraction unit and a semantic mapping unit; In the feature extraction unit, local features are extracted from the optical feature map, the infrared feature map, and the radar feature map based on a convolutional neural network to generate corresponding local feature vectors. In the semantic mapping unit, the local feature vectors are weighted and fused through an adaptive weight allocation mechanism to generate unified semantic features.

3. The method as described in claim 2, characterized in that, The step of weighted fusion of the local feature vectors through an adaptive weight allocation mechanism in the semantic mapping unit to generate unified semantic features includes: Determine the correlation coefficients between the local feature vectors of the optical feature map, the infrared feature map, and the radar feature map; Calculate the initial weight value of each local feature vector based on the correlation coefficient; The initial weight values ​​are adjusted using an activation function to generate the target weight allocation values; The local feature vectors are weighted and summed based on the target weight allocation values ​​to generate unified semantic features.

4. The method as described in claim 1, characterized in that, The step of inputting the unified semantic features into the image fusion model and generating the target fusion result based on the multi-source information fusion algorithm includes: Extract the unified semantic features and input the unified semantic features into the image fusion model; In the image fusion model, semantic features are processed hierarchically based on a multi-scale feature pyramid structure to generate multi-scale feature maps; The multi-scale feature maps are fused layer by layer according to a multi-source information fusion algorithm to generate target candidate regions; Boundary box regression and classification operations are performed on the target candidate regions to generate target fusion results.

5. The method as described in claim 4, characterized in that, The step of fusing the multi-scale feature maps layer by layer according to the multi-source information fusion algorithm to generate target candidate regions includes: Spatial attention is calculated for each layer of the feature map in the multi-scale feature map to generate a spatial attention weight matrix; The spatial attention weight matrix is ​​normalized to obtain normalized spatial attention weights; The normalized spatial attention weights are multiplied by the feature map of the corresponding layer to generate an enhanced feature map. The enhanced feature maps from all layers are superimposed to generate target candidate regions.

6. The method as described in claim 1, characterized in that, The step of generating satellite image recognition results based on the target fusion result and the scene complexity evaluation index includes: Extract the target category, target location, and target confidence information from the target fusion result; Based on the target category and the target confidence information, a comprehensive score for target detection is calculated; By combining the target location information and scene complexity evaluation metrics, a priority ranking for image fusion is generated; Satellite image recognition results are generated based on the target fusion results, the comprehensive score, and the priority ranking.

7. The method as described in claim 1, characterized in that, Before the step of preprocessing the optical remote sensing image, infrared remote sensing image, and radar remote sensing image respectively, the method further includes: Determine the geographical scope and time window of the area to be identified; Based on the geographical range and the time window, determine the observation tasks of the optical sensor, infrared sensor, and radar sensor; The observation task is sent to the satellite control system, and the observation data returned by the satellite control system is received. The received observation data is format-converted and preliminarily verified to generate optical remote sensing images, infrared remote sensing images, and radar remote sensing images.

8. The method as described in claim 1, characterized in that, The steps of preprocessing optical remote sensing images, infrared remote sensing images, and radar remote sensing images to obtain standardized optical feature maps, infrared feature maps, and radar feature maps include: Radiometric and geometric corrections are performed on optical remote sensing images to generate standardized optical feature maps; Temperature calibration and noise filtering are performed on infrared remote sensing images to generate standardized infrared feature maps; Polarization decomposition and scattering characteristic analysis are performed on radar remote sensing images to generate standardized radar feature maps.

9. The method according to any one of claims 1 to 8, characterized in that, After the step of generating satellite image recognition results based on the target fusion result and the scene complexity evaluation index, the method further includes: Anomaly detection is performed on the satellite image recognition results. When it is determined that there is an abnormal target in the satellite image recognition results, the semantic features of the abnormal target are extracted. The semantic features of the abnormal target are input into the anomaly analysis unit to generate the anomaly information of the abnormal target; An abnormal alarm signal is generated based on the abnormal information.

10. A satellite image recognition device based on multimodal remote sensing data fusion, characterized in that, The satellite image recognition device based on the multimodal remote sensing data fusion includes: The image processing module is used to preprocess optical remote sensing images, infrared remote sensing images, and radar remote sensing images respectively to obtain standardized optical feature maps, infrared feature maps, and radar feature maps. The feature extraction module is used to perform deep feature extraction on the optical feature map, the infrared feature map and the radar feature map based on the cross-modal semantic embedding model to generate unified semantic features; The target detection module is used to input the unified semantic features into the image fusion model and generate target fusion results based on the multi-source information fusion algorithm; The result output module is used to generate satellite image recognition results based on the target fusion results and scene complexity evaluation indicators.

Images