Geographic surveying and mapping target intelligent detection method based on unmanned aerial vehicle remote sensing image

Abstract

The invention relates to the technical field of image recognition, and particularly discloses an unmanned aerial vehicle remote sensing image-based geographic surveying and mapping target intelligent detection method, which comprises the following steps of: 1, determining the overlapping degree of shot images according to the route of an unmanned aerial vehicle, and sequentially carrying out preprocessing operation on the collected remote sensing images to obtain preprocessed remote sensing images; step 2, selecting an image at the central position of the shooting area as a reference, and respectively extracting features of the reference image and the image to be registered by using a feature extraction network to obtain corresponding feature maps; according to the method, the original information is reserved, the accuracy of feature extraction and matching is improved, accurate registration and high-quality reconstruction are realized, the problem of how to improve the image splicing precision and efficiency as much as possible while the original image information is reserved in the prior art can be solved, the whole process is high in automation degree, manual intervention is reduced, and the image splicing efficiency is improved. The image processing efficiency is greatly improved, and the application of the unmanned aerial vehicle remote sensing image in the geographic surveying and mapping field is promoted.

Description

Technical Field

[0001] The present invention belongs to the technical field of image recognition, and in particular relates to an intelligent detection method for geographic surveying and mapping targets based on unmanned aerial vehicle remote sensing images. Background Art

[0002] Drone remote sensing technology refers to the use of high-definition imaging equipment carried by drones to capture image information of target areas while the drone is flying. Compared with satellite remote sensing, drone remote sensing technology has faster deployment speed, lower equipment cost, and less control difficulty, making it very suitable for detailed geographic mapping of local areas. Because drone remote sensing images are local images of the target area, in order to fully obtain the target area, drone remote sensing images need to be spliced.

[0003] The UAV remote sensing system uses a small camera's photosensitive sensor frame, which is smaller than the sensor frame used for satellite remote sensing. As a result, there are more influencing materials in the UAV remote sensing images, and the flight stability of the UAV is easily affected by natural factors, which can easily lead to differences in image imaging effects. Therefore, in the process of stitching UAV remote sensing images, obvious stitching marks will appear, resulting in obvious distortion of the stitched images, and the entire target area cannot be accurately acquired. With the continuous development of image acquisition technology, the resolution of acquired images is getting higher and higher. How to maximize the accuracy and efficiency of image stitching while retaining the original image information is still an urgent problem to be solved in the current UAV remote sensing image stitching task.

[0004] Therefore, it is necessary to propose an intelligent detection method for geographic surveying and mapping targets based on UAV remote sensing images to solve the problem existing in the existing technology of how to improve the accuracy and efficiency of image stitching as much as possible while retaining the original image information.

[0005] The above information disclosed in this background technology is only used to increase the understanding of the background technology of the present invention and therefore, it may include information that does not constitute the prior art known to ordinary technicians in this field. Summary of the invention

[0006] The purpose of the present invention is to provide an intelligent detection method for geographic surveying and mapping targets based on UAV remote sensing images to solve the problems raised in the above background technology.

[0007] To achieve the above object, the present invention provides the following technical solutions:

[0008] A method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images, comprising:

[0009] Step 1: determine the overlap of the captured images according to the UAV route, and perform preprocessing operations on the collected remote sensing images in turn to obtain preprocessed remote sensing images;

[0010] Step 2: Select the image at the center of the shooting area as a reference, use the feature extraction network to extract the features of the reference image and the image to be registered, and obtain the corresponding feature map;

[0011] Step 3: Calculate the similarity between the local descriptors in the feature graphs of the reference image and the image to be registered by matrix multiplication to obtain the matching information of the features of the two images;

[0012] Step 4: Use feature matching information to estimate the transformation relationship between the two images, and use the spatial transformation network to achieve image registration and alignment to complete the preliminary image stitching;

[0013] Step 5: Use a multi-resolution image fusion algorithm to reconstruct the initially stitched images into a complete image to obtain the final fused image.

[0014] Preferably, the relative orientation between the captured images is determined according to the geographic correspondence between two adjacent images in the geodetic coordinate system according to the drone route, combined with the positional relationship between the two images judged by human vision; Gaussian filtering is used to eliminate noise information in the image, and histogram normalization based on the LAB color space is used to perform image enhancement operations to obtain a preprocessed remote sensing image.

[0015] Preferably, a deep convolutional neural network is used to construct a feature extraction network. When processing an image, the network layers of the feature extraction network use the same parameters and weights, and a channel attention module is introduced to automatically adjust the feature weights of different channels to obtain a feature map of the corresponding image.

[0016] The feature extraction network processing structure is as follows: first, there are two layers of convolutional layers with a convolution kernel size of 3×3 and 64 channels, a pooling layer with a pooling kernel size of 2×2, and then two layers of convolutional layers with a convolution kernel size of 3×3 and 128 channels. A residual structure is formed between the two convolutional layers and a channel attention module is introduced. A pooling layer with a pooling kernel size of 4×4 is added. Then, the result of the pooling of the 4th convolution is expanded to 256 channels using a 1×1 convolution. There are three layers of convolutional layers with a convolution kernel size of 3×3 and 256 channels. The input of the 5th convolution and the output of the 7th convolution are added in the channel dimension. There is a pooling layer with a pooling kernel size of 8×8. Finally, there are three layers of convolutional layers with a convolution kernel size of 3×3 and 512 channels, and a pooling layer with a pooling kernel size of 2×2.

[0017] In the channel attention module, the average pooling operation assigns a weight to each channel, which can capture the global information of the input features and enable the network to better focus on important features. The maximum pooling operation makes the network more stable to small perturbations of the input image. First, the output feature map F of the third layer of convolution XAfter parallel average pooling and maximum pooling, the outputs are Avgpool(F X ) and Maxpool(F X ); Then, the attention is inferred by the multi-layer perceptron (MLP) to obtain the channel attention M C , which can be expressed as:

[0018] M C (F X )=σ[MLP(Avgpool(F))+MLP(Maxpool(F X ))]

[0019] In the formula, σ is the Sigmod function. Finally, the obtained channel attention M C Multiply it with the input feature map to obtain the enhanced or weakened feature map as the input of the 4th convolution layer.

[0020] Preferably, the feature map f of the reference image obtained by feature extraction A And the feature map f of the image to be registered B , the matching degree of all features is calculated based on matrix multiplication to obtain the relevant feature map C AB , through the ReLU activation function and L2 normalization, the weight of the features with higher matching degree in the relevant feature map is increased to obtain the final relevant feature f AB .

[0021] Preferably, the RANSAC algorithm is used to extract the final relevant features f AB A subset is randomly selected from the set, a system of equations is constructed and the matrix is ​​solved to form a transformation matrix, the transformation matrix is ​​optimized based on the remaining feature points, and the least squares method is used to solve the optimal parameters of the transformation matrix.

[0022] Preferably, the spatial transformation network includes: stacking 3 convolutional layer blocks, each convolutional layer block contains convolution, batch normalization and linear correction units for dimensionality reduction and reducing computational complexity; a spatial pyramid pooling for converting the spatial dimension of the feature map into a one-dimensional vector; a fully connected layer with 6 neurons for regressing the one-dimensional vector into 6 parameters of the radial transformation matrix between the two images.

[0023] Preferably, the Laplace pyramid structure is used to decompose the overlapping area of ​​the two UAV remote sensing images to be fused into different frequency spaces; different fusion operators are used to perform hierarchical fusion operations in their respective frequency domains, and images in all frequency domains are reconstructed and summarized into a complete image.

[0024] Assuming that the overlapping areas of the two UAV farmland remote sensing images to be fused are image A and image B, and the fused image is image F, the specific implementation process of the multi-resolution fusion algorithm is as follows:

[0025] (1) The two images A and B to be fused are subjected to Gaussian filtering, and five-layer Gaussian pyramids are constructed respectively, where the bottom layer G A1 and G A2 is the original image size, and the remaining layers are obtained by downsampling and Gaussian filtering the lower layer image;

[0026] (2) Construct a five-layer Laplace pyramid L A and L B , where the top layer L A5 =G A5 and L B5 =G B5 ,The remaining 1 to 4 layers of the Laplacian pyramid are obtained by subtracting the corresponding layers of the Gaussian pyramid from the upper layer image after upsampling and Gaussian convolution;

[0027] (3) For the Laplace pyramid L A and L B Each layer of L is constructed by simple weighted fusion to construct a synthetic pyramid L s , for L s The pyramid is reconstructed and the bottom layer image is taken as the final fused image. The reconstruction process formula is as follows:

[0028]

[0029] Where Expand is a custom operator for upsampling and Gaussian convolution operations.

[0030] A system for implementing any of the above methods comprises an image acquisition module, an image analysis module and an image fusion module, wherein:

[0031] The image acquisition module is used to receive and store remote sensing image data taken by the drone;

[0032] The image analysis module includes feature extraction network, matrix multiplication calculation, transformation relationship estimation and spatial transformation network, which are used to perform remote sensing image feature extraction, feature map local similarity calculation, transformation relationship estimation between images and image registration and alignment;

[0033] The image fusion module is used to stitch remote sensing images using a multi-resolution image fusion algorithm based on the analysis results to obtain images of the target area.

[0034] Preferably, the system further comprises a user interaction interface for displaying analysis results, receiving user input to adjust analysis parameters or view detailed reports.

[0035] Preferably, the image analysis module further includes an image preprocessing unit for performing denoising and enhancement processing on the acquired remote sensing image.

[0036] Compared with the prior art, the present invention has the following beneficial effects:

[0037] The present invention ensures that there is enough overlapping area between adjacent images, pre-processes the collected remote sensing images, retains the details and information of the images, uses the image at the center of the shooting area as a reference, uses a feature extraction network to extract features to obtain a feature map, and can more accurately capture the key features in the image, and then uses a similarity calculation method based on matrix multiplication to obtain feature matching information, which can quickly process a large number of feature descriptors and reduce calculation complexity. Then, the feature matching information is used to accurately estimate the transformation relationship between images, and the spatial transformation network is combined to achieve image registration and alignment, ensuring that images of different perspectives can be accurately aligned, reducing splicing gaps and distortion, and finally using a multi-resolution image fusion algorithm to reconstruct the spliced ​​images into a complete image, obtain the target area image, retain high-frequency details, and be suitable for subsequent geographic surveying and analysis applications.

[0038] The present invention can solve the problem in the prior art of how to improve the accuracy and efficiency of image stitching as much as possible while retaining the original image information by retaining the original information, improving the accuracy of feature extraction and matching, and achieving precise alignment and high-quality reconstruction. The entire process has a high degree of automation, reduces manual intervention, greatly improves the efficiency of image processing, and promotes the application of UAV remote sensing images in the field of geographic surveying and mapping.

[0039] The above summary is for illustrative purposes only and is not intended to be limiting in any way. In addition to the illustrative aspects, embodiments and features described above, further aspects, embodiments and features of the present invention will be readily apparent by reference to the accompanying drawings and the following detailed description. BRIEF DESCRIPTION OF THE DRAWINGS

[0040] Figure 1 This is a flow chart of the intelligent detection method for geographic surveying and mapping targets of the present invention. DETAILED DESCRIPTION

[0041] The following will be combined with the drawings in the embodiments of the present invention to clearly and completely describe the technical solutions in the embodiments of the present invention. Obviously, the described embodiments are only part of the embodiments of the present invention, not all of the embodiments. Based on the embodiments of the present invention, all other embodiments obtained by ordinary technicians in this field without creative work are within the scope of protection of the present invention.

[0042] It should be further understood that the term “and / or” used in the specification of this application and the appended claims refers to any combination and all possible combinations of one or more of the associated listed items, and includes these combinations.

[0043] Embodiment 1:

[0044] See also Figure 1 As shown, a method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images includes:

[0045] Step 1: determine the overlap of the captured images according to the UAV route, and perform preprocessing operations on the collected remote sensing images in turn to obtain preprocessed remote sensing images;

[0046] Step 2: Select the image at the center of the shooting area as a reference, use the feature extraction network to extract the features of the reference image and the image to be registered, and obtain the corresponding feature map;

[0047] Step 3: Calculate the similarity between the local descriptors in the feature graphs of the reference image and the image to be registered by matrix multiplication to obtain the matching information of the features of the two images;

[0048] Step 4: Use feature matching information to estimate the transformation relationship between the two images, and use the spatial transformation network to achieve image registration and alignment to complete the preliminary image stitching;

[0049] Step 5: Use a multi-resolution image fusion algorithm to reconstruct the initially stitched images into a complete image to obtain the final fused image.

[0050] According to the geographical correspondence between two adjacent images in the geodetic coordinate system according to the drone route, combined with the positional relationship between the two images judged by human vision, the relative orientation between the captured images is determined;

[0051] Gaussian filtering is used to remove noise information in the image, and histogram normalization based on LAB color space is used to perform image enhancement operations to obtain the preprocessed remote sensing image.

[0052] A deep convolutional neural network is used to construct a feature extraction network. When processing images, the network layers of the feature extraction network use the same parameters and weights, and a channel attention module is introduced to automatically adjust the feature weights of different channels to obtain the feature map of the corresponding image.

[0053] The feature map f of the reference image obtained after feature extraction A And the feature map f of the image to be registered B , the matching degree of all features is calculated based on matrix multiplication to obtain the relevant feature map C AB, through the ReLU activation function and L2 normalization, the final correlation feature f is obtained AB .

[0054] Its mathematical expression is:

[0055] C AB (i,j,α)=F B (i,j) T F A (i α ,j α )

[0056] Where i∈{1,…,h}, h is the height of the image, j∈{1,…,w}, w is the width of the image, and α=h(j α -1)+i α Yes(i α ,j α ) index variable, C AB (i, j, α) contains the difference between the images and also integrates the salient features of each image. α ,j α ) represents the position of a single feature in the feature map. AB Each element in (i,j,α) contains a pair of descriptors F A ∈f A and F B ∈f B The scalar product of A All features and feature f B The matching degree of a single feature descriptor with coordinates (i, j) in the image.

[0057] Use the RANSAC algorithm to extract the final relevant features f AB A subset is randomly selected from the set, a system of equations is constructed and the matrix is ​​solved to form a transformation matrix, the transformation matrix is ​​optimized based on the remaining feature points, and the least squares method is used to solve the optimal parameters of the transformation matrix.

[0058] The spatial transformer network includes: stacking 3 convolutional layer blocks, each of which contains convolution, batch normalization and linear correction units for dimensionality reduction and computational reduction; a spatial pyramid pooling layer for converting the spatial dimension of the feature map into a one-dimensional vector; and a fully connected layer with 6 neurons for regressing the one-dimensional vector into 6 parameters of the radial transformation matrix between the two images.

[0059] Considering the relevant feature f ABThe information contained is large, and the amount of direct convolution calculation is large, which reduces the overall efficiency of image registration. First, a convolution with a convolution kernel size of 1×1 is used for dimensionality reduction, and group convolution is introduced in the last two layers of convolution to reduce the amount of calculation. Group convolution is to group the input data, and each group of non-intersecting data is convolved by an independent convolution kernel, and finally the results of each group are spliced ​​together.

[0060] Let k be the size of the convolution kernel, C in , C out is the number of input and output channels, F in 、F out is the size of the input features and output features. When the last two layers of the network do not introduce group convolution, the amount of calculation N is:

[0061] N=2(k 2 ·C in ·C out ·F in ·F out )

[0062] Assuming the group size is G, the total amount of computation M after introducing group convolution in the last two layers of the network is:

[0063]

[0064] The ratio P of the computational amount of the last two layers of the network without introducing grouped convolution to the computational amount after introducing grouped convolution is:

[0065]

[0066] Therefore, the amount of calculation can be reduced and the network regression efficiency can be improved by adjusting the number of groups.

[0067] The Laplace pyramid structure is used to decompose the overlapping area of ​​the two UAV remote sensing images to be fused into different frequency spaces;

[0068] Different fusion operators are used in their respective frequency domains to perform hierarchical fusion operations, and all frequency domain images are reconstructed and aggregated into a complete image.

[0069] Embodiment 2:

[0070] A system for implementing any of the above methods comprises an image acquisition module, a user interaction interface, an image analysis module and an image fusion module, wherein:

[0071] The image acquisition module is used to receive and store remote sensing image data taken by the drone;

[0072] The image analysis module includes a feature extraction network, matrix multiplication calculation, transformation relationship estimation and spatial transformation network, which are used to perform remote sensing image feature extraction, feature map local similarity calculation, transformation relationship estimation between images and image registration and alignment; the image analysis module also includes an image preprocessing unit, which is used to perform denoising and enhancement processing on the acquired remote sensing images;

[0073] The image fusion module is used to stitch the remote sensing images using a multi-resolution image fusion algorithm according to the analysis results to obtain the target area image;

[0074] The user interface is used to display analysis results, receive user input to adjust analysis parameters, or view detailed reports.

[0075] As can be seen from the above, the present invention ensures that there is enough overlapping area between adjacent images, pre-processes the collected remote sensing images, retains the details and information of the images, uses the image at the center of the shooting area as a reference, and uses a feature extraction network to extract features to obtain a feature map, which can more accurately capture the key features in the image, and then uses a similarity calculation method based on matrix multiplication to obtain feature matching information, which can quickly process a large number of feature descriptors and reduce computational complexity. The feature matching information is then used to accurately estimate the transformation relationship between images, and the spatial transformation network is combined to achieve image registration and alignment, ensuring that images of different perspectives can be accurately aligned, reducing splicing gaps and distortion, and finally using a multi-resolution image fusion algorithm to reconstruct the spliced ​​images into a complete image, obtain the target area image, retain high-frequency details, and be suitable for subsequent geographic surveying and analysis applications.

[0076] Embodiment 3:

[0077] The embodiment of the present invention also provides a computer-readable storage medium, on which a computer program is stored, and when the computer program is executed by a processor, each process of the above-mentioned embodiment of the intelligent detection method for geographic surveying and mapping targets is implemented, and the same technical effect can be achieved. To avoid repetition, it is not repeated here. Among them, the computer-readable storage medium, such as a read-only memory (Read-Only Memory, referred to as ROM), a random access memory (Random ACGessMemory, referred to as RAM), a disk or an optical disk, etc.

[0078] In the description of this specification, the description with reference to the terms "one embodiment", "some embodiments", "examples", "specific examples" or "some examples" etc. means that the specific features, structures, materials or characteristics described in conjunction with the embodiment or example are included in at least one embodiment or example of the present invention. In this specification, the schematic representations of the above terms do not necessarily refer to the same embodiment or example. Moreover, the specific features, structures, materials or characteristics described may be combined in any one or more embodiments or examples in a suitable manner. In addition, those skilled in the art may combine and combine different embodiments or examples described in this specification and the features of different embodiments or examples, unless they are contradictory.

[0079] The present invention is described with reference to flowcharts and / or block diagrams of methods, devices (systems) and computer program products according to embodiments of the present invention. It should be understood that each process and / or block in the flowchart and / or block diagram, as well as the combination of processes and / or blocks in the flowchart and / or block diagram, can be implemented by computer program instructions. These computer program instructions can be provided to a processor of a general-purpose computer, a special-purpose computer, an embedded processor or other programmable data processing device to generate a program, so that the instructions executed by the processor of the computer or other programmable data processing device generate instructions for implementing the processes in the flowchart and / or block diagram. Figure 1 A process or multiple processes and / or boxes Figure 1 A way to specify functions in one or more boxes.

[0080] These computer program instructions may also be stored in a computer-readable memory capable of directing a computer or other programmable data processing device to operate in a specific manner, so that the instructions stored in the computer-readable memory produce an article of manufacture comprising an instruction device, which implements the process Figure 1 A process or multiple processes and / or boxes Figure 1 A function specified in one or more boxes.

[0081] These computer program instructions can also be loaded onto a computer or other programmable data processing device so that a series of operating steps are executed on the computer or other programmable device to produce a computer-implemented process, thereby providing instructions for implementing the process in the computer or other programmable device. Figure 1 A process or multiple processes and / or boxes Figure 1 The steps for the functions specified in one or more boxes.

[0082] In the drawings of the embodiments disclosed in the present invention, only the structures involved in the embodiments disclosed in the present invention are involved, and other structures can refer to the general design. In the absence of conflict, the same embodiment and different embodiments of the present invention can be combined with each other.

[0083] The flowcharts shown in the accompanying drawings are only examples and do not necessarily include all the contents and operations / steps, nor must they be executed in the order described. For example, some operations / steps may also be decomposed, combined or partially merged, so the actual execution order may change according to actual conditions.

[0084] Although embodiments of the present invention have been shown and described, it will be appreciated by those skilled in the art that various changes, modifications, substitutions and variations may be made to the embodiments without departing from the principles and spirit of the present invention, and that the scope of the present invention is defined by the appended claims and their equivalents.

Claims

1. An intelligent detection method for geographic surveying and mapping targets based on UAV remote sensing images, characterized in that: The following steps are involved: Step 1: determine the overlap of the captured images according to the UAV route, and perform preprocessing operations on the collected remote sensing images in turn to obtain preprocessed remote sensing images; Step 2: Select the image at the center of the shooting area as a reference, use the feature extraction network to extract the features of the reference image and the image to be registered, and obtain the corresponding feature map; Step 3: Calculate the similarity between the local descriptors in the feature graphs of the reference image and the image to be registered by matrix multiplication to obtain the matching information of the features of the two images; Step 4: Use feature matching information to estimate the transformation relationship between the two images, and use the spatial transformation network to achieve image registration and alignment to complete the preliminary image stitching; Step 5: Use a multi-resolution image fusion algorithm to reconstruct the initially stitched images into a complete image to obtain the final fused image.

2. The method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images according to claim 1 is characterized in that: The method of determining the overlap of the captured images according to the UAV route and sequentially performing preprocessing operations on the collected remote sensing images includes: According to the geographical correspondence between two adjacent images in the geodetic coordinate system according to the drone route, combined with the positional relationship between the two images judged by human vision, the relative orientation between the captured images is determined; Gaussian filtering is used to remove noise information in the image, and histogram normalization based on LAB color space is used to perform image enhancement operations to obtain the preprocessed remote sensing image.

3. The method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images according to claim 2 is characterized in that: The method of using a feature extraction network to respectively extract features of the reference image and the image to be registered comprises: A deep convolutional neural network is used to construct a feature extraction network. When processing images, the network layers of the feature extraction network use the same parameters and weights, and a channel attention module is introduced to automatically adjust the feature weights of different channels to obtain the feature map of the corresponding image.

4. The method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images according to claim 3 is characterized by: The method of calculating the similarity between the local descriptors in the feature graphs of the reference image and the image to be registered based on matrix multiplication includes: The feature map f of the reference image obtained after feature extraction A And the feature map f of the image to be registered B , the matching degree of all features is calculated based on matrix multiplication to obtain the relevant feature map C AB , through the ReLU activation function and L2 normalization, the final correlation feature f is obtained AB .

5. The method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images according to claim 4 is characterized in that: The method of estimating the transformation relationship between the two images by using feature matching information includes: Use the RANSAC algorithm to extract the final relevant features f AB A subset is randomly selected from the set, a system of equations is constructed and the matrix is ​​solved to form a transformation matrix, the transformation matrix is ​​optimized based on the remaining feature points, and the least squares method is used to solve the optimal parameters of the transformation matrix.

6. The method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images according to claim 5 is characterized in that: The spatial transformation network includes: stacking 3 convolutional layer blocks, each convolutional layer block contains convolution, batch normalization and linear correction units for dimensionality reduction and reducing the amount of calculation; a spatial pyramid pooling for converting the spatial dimension of the feature map into a one-dimensional vector; a fully connected layer with 6 neurons for regressing the one-dimensional vector into 6 parameters of the radial transformation matrix between the two images.

7. The method for intelligent detection of geographic surveying and mapping targets based on UAV remote sensing images according to claim 6 is characterized by: The multi-resolution image fusion algorithm is used to reconstruct the preliminary spliced ​​images into a complete image, including: The Laplace pyramid structure is used to decompose the overlapping area of ​​the two UAV remote sensing images to be fused into different frequency spaces; Different fusion operators are used in their respective frequency domains to perform hierarchical fusion operations, and all frequency domain images are reconstructed and aggregated into a complete image.

8. A system for implementing the method according to any one of claims 1 to 7, comprising an image acquisition module, an image analysis module and an image fusion module, wherein: The image acquisition module is used to receive and store remote sensing image data taken by the drone; The image analysis module includes feature extraction network, matrix multiplication calculation, transformation relationship estimation and spatial transformation network, which are used to perform remote sensing image feature extraction, feature map local similarity calculation, transformation relationship estimation between images and image registration and alignment; The image fusion module is used to stitch remote sensing images using a multi-resolution image fusion algorithm based on the analysis results to obtain images of the target area.

9. The system according to claim 8, characterized in that It also includes a user interface for displaying analysis results, receiving user input to adjust analysis parameters, or viewing detailed reports.

10. The system according to claim 8, characterized in that The image analysis module also includes an image preprocessing unit for performing denoising and enhancement processing on the acquired remote sensing images.

Images