CN116403011A - Establishment Method of Effective Data Sample Group Based on SAR-Optical Image Matching - Google Patents

Abstract

The invention discloses a method for establishing an effective data sample group based on SAR-Optical image matching, which comprises the steps of obtaining SAR images and Optical remote sensing image pairs; rectangular template region matching is carried out on the SAR image and the optical remote sensing image pair, manual selection of data to be detected by naked eyes is not needed in the matching, the method has the advantages that manual screening and classification errors are reduced, meanwhile, the data set manufacturing efficiency is improved, the time and labor cost are reduced, and the method is more suitable for the existing matching algorithm based on deep learning; and finally, calibrating and aligning all screened effective SAR images and Optical remote sensing image pairs to generate an effective data sample group based on SAR-Optical image matching, removing invalid data through a threshold method, retaining the effective data, improving the effectiveness of the data, solving the technical problem of how to automatically generate an effective data set and reducing the processing cost of manual data.

Description

Establishment Method of Effective Data Sample Group Based on SAR-Optical Image Matching

technical field

The invention belongs to the technical field of image processing, in particular to a method for establishing an effective data sample group based on SAR-Optical image matching.

Background technique

Image is the most important form of representation of natural scene information in computers, so image processing technology has always been the focus and hotspot of theoretical research and practical application. Usually, a single sensor can only obtain one aspect of the scene information, while multiple sensors can obtain various aspects of the scene information, for example, visible light images can obtain the spectral reflectance information of the visible light band of the scene, remote sensing images can obtain the characteristics of the ground features in the shooting area, SAR The image can obtain an all-weather radar image of the shooting area. Heterogeneous image matching and image fusion are the technical basis for comprehensive utilization of complementary image information collected by multiple sensors. Heterogeneous image matching refers to the process of aligning multiple images of the same scene collected by different sensors in pixel space, while heterogeneous image fusion refers to the process of synthesizing complementary information in heterogeneous images. Due to the basic status of heterogeneous image matching and fusion, it is widely used in various military and civilian fields, such as navigation guidance, remote sensing scene analysis, target recognition, etc.

In heterogeneous image matching tasks, the matching of SAR (Synthetic Aperture Radar) and visible light (Optical) images is the most representative and has important military application prospects. Due to the fundamental difference in the imaging principles of synthetic aperture radar and optical images, there are complex nonlinear radiation differences between their images. Therefore, the traditional matching methods based on feature points and templates often cannot achieve ideal results.

And there are following several technical problems in the prior art:

1) In the basic process of making SAR-Optical data sets, a lot of manual intervention is still required, especially in the part of effective area screening. The manual inspection method is not only inefficient but also very costly;

2) In addition, because the existing SAR-Optical dataset relies on manual screening of samples, people cannot fully understand the features and regions that the network needs to learn, so the manually screened data often cannot adapt to the existing neural network;

3) Considering that the matching algorithm in recent years has gradually changed from feature dependence to data dependence, the size of the data scale will sometimes determine the performance of the matching algorithm. The existing algorithms for cleaning data are slow, occupy a lot of resources, and can only process a single type of data. These problems greatly limit the development of data set scale.

Contents of the invention

The purpose of the present invention is to provide a method for establishing an effective data sample group based on SAR-Optical image matching, so as to solve the technical problem of how to automatically generate an effective data set and reduce the cost of manual data processing.

The present invention adopts following technical scheme:

Embodiment 1 of the present invention provides a method for establishing an effective data sample group based on SAR-Optical image matching, including:

Obtain SAR image and optical remote sensing image pair;

Perform template matching on SAR images and optical remote sensing image pairs to obtain effective SAR images and optical remote sensing image pairs;

All the selected effective SAR images and optical remote sensing image pairs are calibrated and aligned to generate an effective data sample group based on SAR-Optical image matching.

Optionally, performing template matching on the SAR image and the optical remote sensing image pair includes:

Randomly select the first rectangular template area and the second rectangular template area on the SAR image according to the template matching algorithm;

calculating a first length offset and a first width offset between the first rectangular template area and the second rectangular template area in the first preset offset direction;

An effective SAR image and an optical remote sensing image pair are obtained according to the first length offset and the first width offset.

Optionally, performing template matching on the SAR image and the optical remote sensing image pair also includes:

Randomly select the third rectangular template area and the fourth rectangular template area on the optical remote sensing image according to the template matching algorithm;

calculating a second length offset and a second width offset between the third rectangular template area and the fourth rectangular template area in a second preset offset direction;

The current effective SAR image and the optical remote sensing image pair are determined according to the second length offset and the second length offset.

Optionally, obtaining effective SAR images and optical remote sensing image pairs includes:

judging whether the first difference between the first length offset and the second length offset is greater than a first threshold;

judging whether the second difference between the second length offset and the second length offset is greater than a second threshold;

If the first difference is less than the first threshold and the second difference is less than the second threshold, then retaining the current SAR image and the optical remote sensing image pair as a valid SAR image and the optical remote sensing image pair;

Optionally, if the first difference is not smaller than the first threshold and the second difference is not smaller than the second threshold, then adjust the size of the template area or the size of different thresholds.

Optionally, performing calibration and alignment on all the effective SAR images and optical remote sensing image pairs screened out includes:

Separately extract the features of ground objects in effective SAR images and effective optical remote sensing images;

Calculating the alignment between features in valid SAR images and valid optical remote sensing images;

Several effective SAR images and optical remote sensing image pairs are screened out according to the alignment, and an effective data sample group based on SAR-Optical image matching is generated.

Optionally, the calculation method of the alignment includes:

表示I₁图像的像素方差，/>

表示图像I₂图像的像素方差，/>

表示I₂对I₁的像素期望方差，/>

表示I₁对I₂的像素期望方差，CI(I₁,I₂)表示有效SAR图像和有效光学遥感图像中地物特征之间的对齐度。Among them, I ₁ represents the feature of the effective SAR image, I ₂ represents the feature of the effective optical remote sensing image,

represents the pixel variance of the I ₁ image, />

represents the pixel variance of the image I ₂ image, />

represents the pixel expected variance of I ₂ to I ₁ , />

Indicates the pixel expected variance of I ₁ to I ₂ , and CI(I ₁ , I ₂ ) indicates the alignment between effective SAR image and effective optical remote sensing image.

Optionally, acquiring the SAR image and the optical remote sensing image pair includes:

Data format conversion for SAR images and optical remote sensing images;

Random sampling of SAR images and optical remote sensing images after data format conversion;

SAR images and optical remote sensing images collected by random sampling are deduplicated to generate SAR images and optical remote sensing image pairs.

Embodiment 2 of the present invention provides a device for establishing an effective data sample group based on SAR-Optical image matching, including a memory, a processor, and a computer program stored in the memory and operable on the processor, and the processor executes the computer program A method for establishing an effective data sample group based on SAR-Optical image matching as in any one of the above method embodiments is implemented at the same time.

The beneficial effects of the present invention are: 1) using the mature Google Earth Engine on the market to obtain SAR images and optical remote sensing image pairs, it is easy to use, easy to grasp, and has a wide social acceptance;

2) Template matching is performed on the SAR image and the optical remote sensing image pair. During the matching, it is not necessary to manually use the naked eye to select the test data, which reduces manual screening and classification errors, improves the efficiency of data set production, and reduces time and manpower. cost, and is more adaptable to current matching algorithms based on deep learning.

3) Calibrate and align all the selected valid SAR images and optical remote sensing image pairs, generate valid data sample groups based on SAR-Optical image matching, and remove invalid data by threshold method, retain valid data, and improve data validity.

Description of drawings

Fig. 1 is a flow chart of a method for establishing an effective data sample group based on SAR-Optical image matching provided by Embodiment 1 of the present invention;

FIG. 2 is a schematic flow chart of the steps of a method for establishing an effective data sample group based on SAR-Optical image matching provided by Embodiment 1 of the present invention;

Fig. 3 is a schematic diagram of setting a rectangular template area provided by Embodiment 1 of the present invention;

4 is a schematic diagram of the steps of an image calibration method based on a ground control point (GCP) library provided by Embodiment 1 of the present invention;

FIG. 5 is a schematic diagram of a network structure of a template matching algorithm provided by Embodiment 1 of the present invention;

FIG. 6 is a schematic transposition diagram of establishment of an effective data sample group based on SAR-Optical image matching provided by Embodiment 2 of the present invention.

Detailed ways

The present invention will be described in detail below in conjunction with the accompanying drawings and specific embodiments.

1. Embodiment 1 of the present invention provides a method for establishing an effective data sample group based on SAR-Optical image matching. In combination with FIGS. 1 and 2, the method includes:

Step 101, acquiring a SAR image and an optical remote sensing image pair;

Optionally, acquiring the SAR image and the optical remote sensing image pair includes:

Data format conversion for SAR images and optical remote sensing images;

Random sampling of SAR images and optical remote sensing images after data format conversion;

SAR images and optical remote sensing images collected by random sampling are deduplicated to generate SAR images and optical remote sensing image pairs.

The specific implementation of step 101 is: download the required SAR image and optical remote sensing image from Google Earth Engine, in order to expand the data source, or obtain the preset city or region from the public data set formed by the public radar satellite sampling data on the network SAR images and optical remote sensing images, and data format conversion, processing and converting them into easy-to-handle data.

Specifically, for downloaded public datasets, it is first necessary to adjust all the data into a unified format for easy processing.

Taking the data of Sentinel 1 and Sentinel 2 as an example, converting the data format refers to converting the data into GeoTiffs data format images, using the Export.image.toDrive function of GEE to export the downloaded SAR image data or optical remote sensing images to generate GeoTiff data; and the gray value of GeoTiff data is controlled within the range of ±2.5σ, and the pixel value is normalized to the interval [0, 1] to represent a relatively large range. If there are multiple bands, then The above corrective operation is realized on all bands. After completing the above operations, the data format conversion between SAR images and optical remote sensing images can be realized. After the data format conversion, random sampling and selection of the required ground object area data.

In order to remove overlapping random sampling data, use GEE's built-in ee.ImageCollection.mosaic() function and ee.Image.clip() function to set a function that can trim the size of the sampling data for each random sampling. In short, the ee.ImageCollection.mosaic() function is used to process those partially overlapping images, eliminate invalid SAR images and optical remote sensing images, and improve heterogeneous matching.

For the images obtained from Google Earth Engine, first sample, de-duplicate, unify the format with the public dataset, and integrate and filter out the available data.

After downloading the required image from Google Earth Engine, you need to set the area for random sampling, set the sampling seed, and generally select 100 points on the surface of the preset city or the required area for random sampling, and then select from the preset urban area 50 points. Shape details of different land or urban areas are provided by public domain Geodata services with a sampling scale of 1:50m.

If the positions of the two points obtained are quite close, it is necessary to remove one of the points through deduplication processing to ensure that there is no overlapping part during sampling, and keep the non-overlapping SAR image and optical remote sensing image pair. For the sampled image, the image deduplication work is performed, and the above-mentioned GEE built-in ee.ImageCollection.mosaic() function and ee.Image.clip() function are used for processing.

After the sampling is completed, the Google Earth Engine data will be unified in a format, and then the public dataset data and Google Earth Engine data will be integrated to screen unavailable data. For example, an effective data sample group based on SAR-Optical image matching requires that the cloud coverage of the Sentinel-2 image is less than 1% and all the VV-IW bands of the Sentinel-1 are available. Use the GEE tool to filter the image according to the above conditions, so as to Filter and process the sampling data, if there is data that does not meet the conditions, remove it.

Step 102, performing template matching on the SAR image and the optical remote sensing image pair to obtain an effective SAR image and optical remote sensing image pair;

In one embodiment, it should be noted that the cloud coverage data of Sentinel-2 particles is only a globally shared parameter, and it is not accurate to filter SAR images and optical remote sensing images according to parameters, and there are a bunch of particles in the whole particle For the image data covered by clouds, only a small part of the cloud covers the local image data, which are all invalid data, and only a very small part of the data without any occlusion is valid data. However, the template matching method is used to set a rectangular template area in the SAR image and the optical remote sensing image respectively. It should be noted that the template area contains the characteristics of the target object. SAR image and optical remote sensing image pair. Therefore, the template matching method is used to remove large invalid areas and images with severe cloud coverage in SAR images and optical remote sensing images, to achieve rough matching between SAR images and optical remote sensing images, and to preliminarily screen out effective data to achieve the effect of data cleaning.

Optionally, performing template matching on the SAR image and the optical remote sensing image pair includes:

Randomly select the first rectangular template area and the second rectangular template area on the SAR image according to the template matching algorithm;

The first rectangular template area and the second rectangular template area have a first length offset and a first width offset according to a first preset offset direction;

An effective SAR image and an optical remote sensing image pair are obtained according to the first length offset and the first width offset.

Optionally, performing template matching on the SAR image and the optical remote sensing image pair also includes:

Randomly select the third rectangular template area and the fourth rectangular template area on the optical remote sensing image according to the template matching algorithm;

The third rectangular template area and the fourth rectangular template area have a second length offset and a second width offset according to a second preset offset direction;

The current effective SAR image and the optical remote sensing image pair are determined according to the second length offset and the second length offset.

Optionally, obtaining effective SAR images and optical remote sensing image pairs includes:

judging whether the first difference between the first length offset and the second length offset is greater than a first threshold;

judging whether the second difference between the second length offset and the second length offset is greater than a second threshold;

If the first difference is less than the first threshold and the second difference is less than the second threshold, then retaining the current SAR image and the optical remote sensing image pair as a valid SAR image and the optical remote sensing image pair;

Optionally, if the first difference is not smaller than the first threshold and the second difference is not smaller than the second threshold, then adjust the size of the template area or the size of different thresholds.

In one embodiment, as shown in FIG. 3 , firstly, two rectangular template regions are randomly set in the SAR image to be detected, for example, region A is the first rectangular template region, and region B is the second rectangular template region, wherein the region There is a certain amount of offset between area B and area A according to the preset direction. For example, in the two-dimensional coordinate system, there is a first length offset Δx in the horizontal direction and a first width offset Δy in the vertical direction relative to the area B relative to the area A. The corresponding template is set in the optical remote sensing image in the same way, for example, the area A' is the third rectangular template area, and the area B' is the fourth rectangular template area. There is a certain amount of offset between the area B' and the area A' according to the preset direction. For example, in a two-dimensional coordinate system, there is a second length offset Δx' in the horizontal direction and a second width offset Δy' in the vertical direction between the region B' and the region A'.

To this end, further calculate the first difference between the first length offset and the second length offset between the SAR image and the optical remote sensing image, that is, whether (Δx-Δx′) is greater than the value set in the template matching algorithm The first threshold; simultaneously judge the second difference between the first width offset and the second width offset between the SAR image and the remote sensing optical image, that is, whether (Δy-Δy') is greater than in the template matching algorithm The second threshold is set, only when the first difference is smaller than the first threshold and the second difference is smaller than the second threshold, it indicates that the currently processed SAR image and optical remote sensing image are valid SAR image and optical remote sensing image pair.

If the first difference is not smaller than the first threshold and the second difference is not smaller than the second threshold, then adjust the size of the template area or the size of different thresholds.

It should be noted that, usually, if there is a high adaptability in this region, then Δx, Δy will be very close to Δx′, Δy′. If the fitness of this region is low, then Δx and Δx', Δy and Δy' will show large deviations. By setting a certain threshold for the deviation between Δx, Δy and Δx′, Δy′, most of the unsuitable regions can be screened out.

It should be noted that the size and position of the rectangular template area are not fixed throughout the cleaning process, and can be continuously changed to meet different invalid area screening requirements and image types. For example, when dealing with a small invalid area on the data, the existence of a large rectangular template area may not be detected, or the calculated offset is very small and does not meet the threshold. Therefore, an additional coarse-to-fine rectangular template area setting strategy can be added.

In the coarse-to-fine rectangular template area setting strategy, a pair of larger rectangular template areas are first set as sliding windows, which can be adjusted according to the size of the area where the feature to be detected is actually located, and can be used in the entire Slide on the reference image to include all the image features to be detected, so as to more accurately calculate the offset of the rectangular template area in the SAR image and optical remote sensing image.

All SAR images and optical remote sensing images are detected repeatedly through the template detection method to increase the accuracy of data screening.

And by setting various thresholds in this embodiment, an adjustment data set can be constructed according to the requirements of the network. In the entire data set construction process, data cleaning is used cyclically for the same batch of data to gradually remove all unsuitable images and retain suitable images, thereby effectively replacing manual intervention steps in the data set preparation process and realizing large-scale data set preparation Process automation.

Step 103, perform calibration and alignment on all the selected effective SAR images and optical remote sensing image pairs, and generate an effective data sample group based on SAR-Optical image matching.

In one embodiment, it should be noted that in the pair of SAR images and optical remote sensing images screened out through the above step 102, since the actually collected SAR image information is the image information formed by the backscattering of ground objects, the image presents For the surface image information of an area, in order to better align the SAR image and the optical remote sensing image, the SAR image and the optical remote sensing image are firstly calibrated using the image calibration method based on the ground control point (GCP) library, as shown in Figure 4 As shown, the image to be calibrated is used as the input of the calibrator, and the optical remote sensing image of the target area is obtained through the (GCP) library, and the center point position of the optical remote sensing image can be directly extracted; in addition, since the SAR image presents the surface of the target area Surface image information, for this purpose, it is necessary to obtain SAR orbital parameters and imaging parameters, expand the SAR image, and calculate the position information of the four corner points of the expanded SAR image, and calculate the SAR image according to the position information of the four corner points. The position of the center point, the center point position of the SAR image and the center point position of the optical remote sensing optical image in the (GCP) library are calibrated, and the initial matching area is extracted by using affine transformation and resampling to ensure that the same image can be found from the two images. Image slices or features of the same object in the target area; finally, cross-correlation precise matching is performed on the center point position of the SAR image and the initial matching area of the optical remote sensing optical image in the (GCP) library.

Optionally, calibrate and align all the effective SAR images and optical remote sensing image pairs screened out, and generate an effective data sample group based on SAR-Optical image matching including:

Separately extract the features of ground objects in effective SAR images and effective optical remote sensing images;

Calculating the alignment between features in valid SAR images and valid optical remote sensing images;

Several effective SAR images and optical remote sensing image pairs are screened out according to the alignment, and an effective data sample group based on SAR-Optical image matching is generated.

In one embodiment, as shown in Figure 5, based on the Siamese network architecture, the effective SAR images and optical remote sensing image pairs screened out are used as the input of the rectangular template area matching algorithm network, and the same CNN feature extractor is used to respectively Effective SAR image and effective optical remote sensing image feature feature, after determining the feature map of the SAR image and the feature map of the optical remote sensing image, calculate the distance between the SAR image and the effective optical remote sensing image feature feature according to the following method Alignment:

表示I₁图像的像素方差，/>

表示图像I₂图像的像素方差，/>

表示I₂对I₁的像素期望方差，

表示I₁对I₂的像素期望方差，CI(I₁,I₂)表示有效SAR图像和有效光学遥感图像中地物特征之间的对齐度。Among them, I ₁ represents the feature of the effective SAR image, I ₂ represents the feature of the effective optical remote sensing image,

represents the pixel variance of the I ₁ image, />

represents the pixel variance of the image I ₂ image, />

Denotes the pixel-wise expected variance of I ₂ to I ₁ ,

Indicates the pixel expected variance of I ₁ to I ₂ , and CI(I ₁ , I ₂ ) indicates the alignment between effective SAR image and effective optical remote sensing image.

In order to realize the alignment of the SAR image and the optical image, the alignment degree is calculated by using the interaction variance, which reflects the stability of the gray levels of the two images corresponding to each other. The main idea is that if two images are aligned, each gray level of one image corresponds to the gray level of the other image at the pixel position is the most stable, that is, the variance is the smallest. To this end, all valid SAR images and optical remote sensing image pairs that meet the preset alignment threshold requirements can be screened out according to the size of the variance or the size of the alignment, and an effective data sample group based on SAR-Optical image matching can be generated. Accurate matching of SAR images and optical images is realized.

It should be noted that, as shown in Figure 5, after extracting the features of the SAR image and the effective optical remote sensing image, Fourier convolution is used to perform correlation calculation on the extracted feature map. Since there is a certain downsampling operation in the feature extraction network, it is necessary to upsample the correlation map at the end to restore the original spatial resolution accuracy.

This solution trains a larger model on the server side and deploys a lightweight model on the mobile side. On the server side, this solution uses the classic ResNet network structure, and enhances the robustness and recognition of image features through the channel attention module. The lightweight model deployed on the actual mobile terminal mainly adopts a network structure based on mobilenetv2, which has a smaller amount of parameters and calculations.

When calculating the correlation between the feature map of the SAR image and the feature map of the optical remote sensing image, the present invention adopts a dense InforNCE loss function based on contrastive learning and uses a multi-granularity regularization scheme, which can effectively avoid training stage overfitting problem.

And during the whole training process, the correlation score is obtained by calculating the correct position of the feature map of the SAR image and the feature map of the optical remote sensing image:

Among them, _sp is the correlation score at the correct position p between the feature map of the SAR image and the feature map of the optical remote sensing image, and s _j is the feature map of the SAR image and the feature map of the optical remote sensing image The total correlation score during the matching process, τ is the temperature coefficient used to moderately non-linearly scale the correlation scores at different locations, and R represents all locations of the correlation score map. The loss function can effectively deal with the multi-mode distribution problem in the matching problem, thereby effectively supervising the training of the entire network, and enhancing the stability and generalization of network feature learning and extraction.

To speed up the matching process, Fourier (FFT) convolutions are used instead of ordinary convolutions. In the process of FFT convolution, FFT convolution is first performed on the SAR image as the convolution kernel and the reference optical remote sensing image. During convolution, padding is performed on the SAR image to make it the same size as the reference image. Compared with the operation of ordinary convolution, the total calculation amount is significantly reduced.

Through the rectangular template area matching algorithm provided by the present invention and combined with SAR images, the rough matching of two heterogeneous images is realized; further, the fine matching of the two heterogeneous images is realized by calibrating and aligning the features of effective optical remote sensing images, and the final improvement of the modified model is improved. matching efficiency and accuracy.

Embodiment 2 of the present invention provides a device 500 for establishing an effective data sample group based on SAR-Optical image matching, as shown in FIG. 6 , including a memory 510, a processor 520, and an A computer program 530, when the processor executes the computer program, implements a method for establishing an effective data sample group based on SAR-Optical image matching as in any one of the above method embodiments.

It should be noted that, the above-mentioned device 500 for establishing an effective data sample group based on SAR-Optical image matching can realize real-time establishment of an effective data sample group based on SAR-Optical image matching in accordance with the method steps of Embodiment 1, which is not mentioned here. Let me repeat.

Claims (9)

1. The method for establishing the valid data sample group based on SAR-Optical image matching is characterized by comprising the following steps of:

acquiring SAR images and optical remote sensing image pairs;

template matching is carried out on the SAR image and the optical remote sensing image pair, and an effective SAR image and the optical remote sensing image pair are obtained;

and performing calibration alignment on the effective SAR image and the Optical remote sensing image pair to generate an effective data sample group based on SAR-Optical image matching.

2. The method of claim 1, wherein the performing template matching on the SAR image and the Optical remote sensing image pair comprises:

randomly selecting a first rectangular template area and a second rectangular template area on the SAR image according to a template matching algorithm;

calculating a first length offset and a first width offset of the first rectangular template area and the second rectangular template area in a first preset offset direction;

and obtaining an effective SAR image and an optical remote sensing image pair according to the first length offset and the first width offset.

3. The method of claim 2, wherein the performing template matching on the SAR image and the Optical remote sensing image pair further comprises:

randomly selecting a third rectangular template area and a fourth rectangular template area on the optical remote sensing image according to a template matching algorithm;

calculating a second length offset and a second width offset of the third rectangular template area and the fourth rectangular template area in a second preset offset direction;

and determining a current effective SAR image and an optical remote sensing image pair according to the second length offset and the second length offset.

4. A method of establishing a population of valid data samples based on SAR-Optical image matching as claimed in any one of claims 2 or 3, wherein template matching the SAR image and the Optical remote sensing image pair comprises:

judging whether a first difference value between the first length offset and the second length offset is larger than a first threshold value;

judging whether a second difference value between the second length offset and the second length offset is larger than a second threshold value;

and if the first difference value is smaller than a first threshold value and the second difference value is smaller than a second threshold value, reserving the current SAR image and the optical remote sensing image pair as an effective SAR image and optical remote sensing image pair.

5. The method for creating a valid data sample group based on SAR-Optical image matching according to claim 4, wherein the size of the template region or the size of the different threshold is adjusted if the first difference is not smaller than a first threshold and the second difference is not smaller than a second threshold.

6. The method of claim 1, wherein the step of aligning the screened valid SAR image and the Optical remote sensing image pairs comprises:

respectively extracting ground feature characteristics in an effective SAR image and an effective optical remote sensing image;

calculating the alignment degree between the ground feature characteristics in the effective SAR image and the effective optical remote sensing image;

and screening a plurality of effective SAR images and the effective Optical remote sensing image pairs according to the alignment degree, and generating an effective data sample group based on SAR-Optical image matching.

7. The method for establishing the valid data sample group based on the SAR-Optical image matching as set forth in claim 6, wherein the alignment degree calculating method comprises the following steps:

wherein I is ₁ Representing the feature characteristics of the effective SAR image, I ₂ Representing the feature of the effective optical remote sensing image,

representation I ₁ Pixel variance of image,/>

Representing image I ₂ Pixel variance of image,/>

Representation I ₂ Pair I ₁ Is used to determine the desired variance of the pixels of (1),

representation I ₁ Pair I ₂ Is of the pixel expected variance, CI (I ₁ ,I ₂ ) And representing the alignment degree between the ground feature characteristics in the effective SAR image and the effective optical remote sensing image.

8. The method for creating the valid data sample group based on SAR-Optical image matching of claim 1, wherein obtaining the SAR image and the Optical remote sensing image pair comprises:

performing data format conversion on the SAR image and the optical remote sensing image;

randomly sampling the SAR image and the optical remote sensing image after the data format conversion;

and performing de-duplication processing on the SAR image and the optical remote sensing image which are acquired by random sampling, and generating an SAR image and optical remote sensing image pair.

9. The apparatus for creating a valid data sample group based on SAR-Optical image matching according to claim 1, comprising a memory, a processor, and a computer program stored in the memory and executable on the processor, wherein the processor implements a method for creating a valid data sample group based on SAR-Optical image matching according to any one of claims 1 to 8 when executing the computer program.

Images