CN-117036748-B - Migration correction method, system and storage medium for high-resolution remote sensing image annotation

CN117036748BCN 117036748 BCN117036748 BCN 117036748BCN-117036748-B

Abstract

The embodiment of the application provides a migration correction method, a migration correction system and a storage medium for high-resolution remote sensing image annotation, and belongs to the field of remote sensing image processing. The method comprises the steps of obtaining a sample image and a target image, wherein the sample image and the target image comprise at least one target object, the sample image is provided with labeling information of the target object, the target image is input into a pre-trained prediction model to obtain a target image prediction probability map, sample plaques are determined according to the labeling information, a matching area of the sample plaques is determined in the prediction probability map, similarity values of all pixel points in the at least one sample plaque and the matching area are calculated, all the sample plaques are matched in the matching area in an overall constraint mode based on the similarity values, a target matching position is obtained, and finally a target labeling result of the target object in the target image is obtained. The method and the device can improve the accuracy of the remote sensing image data annotation migration correction in the same area and different time phases.

Inventors

WANG JIE
ZHANG WEI
LIU QIANG

Assignees

鹏城实验室

Dates

Publication Date: 20260505
Application Date: 20230731

Claims (13)

1. The migration correction method for the high-resolution remote sensing image annotation is characterized by comprising the following steps of: Acquiring a sample image and a target image, wherein the sample image and the target image both comprise at least one target object, and the sample image is configured with labeling information of the target object; Inputting the target image into a pre-trained prediction model, obtaining pixel probability values of which each pixel point in the target image is the target object, and establishing a prediction probability map of the target image according to each pixel probability value; determining a sample plaque of the target object in the sample image according to the labeling information, determining a matching area of the sample plaque in the predictive probability map, and correcting and adjusting the position of the sample plaque in the corresponding matching area; calculating the similarity value of at least one sample patch and each pixel point in the matching area, and determining a weighted bipartite graph by taking the similarity of each candidate matching point as a weight according to the candidate matching point corresponding to each sample patch and the sample patch corresponding to each candidate matching point, wherein the weighted bipartite graph is used for representing the weight relation between each sample patch and each candidate matching point; Determining target matching positions based on each sample plaque and each alternative matching point from the weighted bipartite graph by taking total maximum weight matching as constraint conditions, so that each sample plaque after constraint matching integrally corresponds to the target object in the target image; And obtaining a target labeling result of the target object in the target image according to the target matching position.
2. The migration correction method for high-resolution remote sensing image annotation according to claim 1, wherein a target plaque is obtained according to a target annotation result of the target image, wherein the target plaque comprises a target plaque characteristic value and target plaque annotation information corresponding to the target plaque characteristic value; after obtaining the target labeling result of the target object in the target image according to the target matching position, the method further comprises the following steps: acquiring a weight parameter value for amplification; Performing difference operation on the weight parameter value and the unit weight to obtain a difference weight parameter value; Multiplying the weight parameter value with the sample plaque characteristic value to obtain a first sub-value, multiplying the difference weight parameter value with the target plaque characteristic value to obtain a second sub-value, and adding the first sub-value and the second sub-value to obtain first amplification information; Multiplying the weight parameter value with the sample plaque annotation information to obtain a third sub-value, multiplying the difference weight parameter value with the target plaque annotation information to obtain a fourth sub-value, and adding the third sub-value and the fourth sub-value to obtain second amplification information; Obtaining an amplified sample after amplification improvement according to the first amplification information and the second amplification information, wherein the amplified sample is used for carrying out iterative improvement on the target labeling result.
3. The method for migration correction of high-resolution remote sensing image annotation according to claim 1, wherein determining a sample plaque in the sample image where the target object is located according to the annotation information comprises: Acquiring a positioning reference point and a resolution parameter of the target image; according to the positioning reference point and the resolution parameter, positioning adjustment and resolution adjustment are carried out on the sample image so that the coordinate positions and the resolutions of the adjusted sample image and the target image are consistent, and a preprocessed sample image is obtained; performing feature conversion on the preprocessed sample image according to the image features of the target image to obtain a converted sample image, and obtaining label information after conversion of the sample image according to the converted sample image; and determining a sample plaque of the target object in the sample image according to the converted labeling information.
4. The method for mobility correction of high resolution remote sensing image annotation according to claim 3, wherein the performing feature conversion on the preprocessed sample image according to the image feature of the target image to obtain a converted sample image comprises: acquiring first spectrum information of the preprocessed sample image and second spectrum information of the target image; If the first spectrum information and the second spectrum information are in a preset spectrum range, performing relative radiation correction operation on the preprocessed sample image to obtain a converted sample image; If the first spectrum information and the second spectrum information are not in the preset spectrum range, performing linear quantization operation on the sample image after pretreatment, so that the first spectrum information and the second spectrum information after linear quantization are in the preset spectrum range, and obtaining the sample image after conversion; Or performing style migration operation on the preprocessed sample image according to a pre-trained network model to obtain a converted sample image.
5. The migration correction method of high resolution remote sensing image annotation according to claim 1, wherein the inputting the target image into a pre-trained prediction model to obtain pixel probability values of each pixel point in the target image as the target object, and establishing a prediction probability map of the target image according to each pixel probability value comprises: Obtaining a pre-trained prediction model; inputting the target image into the prediction model, and carrying out probability prediction on each pixel point in the target image according to the prediction model to obtain a prediction probability value of each pixel point; if the predicted probability value is in a preset probability range, determining the pixel point as the region label of the target image; Calculating a vegetation index of the target image, and determining a green land mask of the target image according to the vegetation index; Expanding a predicted area according to the area label and the green land mask, and according to the expanded predicted area; and obtaining a prediction probability map corresponding to the target image according to the prediction region.
6. The migration correction method of high-resolution remote sensing image annotation according to claim 1, wherein the pre-trained prediction model is obtained by training the following steps: Acquiring a preset prediction training set, wherein the prediction training set comprises a plurality of training sample images for training, and setting corresponding training labels for the training sample images; Selecting any training sample image from the prediction training set, and inputting the training sample image into the prediction model to obtain a training prediction probability value of each sample area of the training sample image; And calculating a prediction loss value of the prediction model according to the training prediction probability value and the training label, and adjusting parameters of the prediction model according to the prediction loss value to obtain the trained prediction model.
7. The method for migration correction of high-resolution remote sensing image annotation according to claim 1, wherein determining the matching region of the sample plaque in the predictive probability map comprises: Obtaining geographic coordinate information; outputting the sample plaque to the prediction probability map correspondingly according to the geographic coordinate information to obtain a constraint matching map; acquiring a preset searching radius, wherein the searching radius is set according to the predictive probability map; and determining a matching area of the sample plaque on the constraint matching graph according to the searching radius, wherein the matching area comprises four sub-matching areas.
8. The method for migration modification of high-resolution remote sensing image annotation according to claim 7, wherein after calculating the similarity value between at least one of the sample patches and each pixel point in the matching region, further comprising: acquiring a preset distance threshold; Determining a pixel point corresponding to the maximum similarity value of each sub-matching region as a sub-alternative matching point of each sub-matching region; selecting a sub-candidate matching point with the largest similarity value in the sub-candidate matching points as a reference matching point; if the distance between the rest sub-candidate matching points and the reference matching points is smaller than the distance threshold value, determining the sub-candidate matching points as error matching points; And determining the alternative matching points corresponding to the sample patches according to the sub alternative matching points and the error matching points.
9. The method for migration modification of high-resolution remote sensing image annotation according to claim 8, wherein after determining the candidate matching point corresponding to the sample patch according to the sub-candidate matching point and the error matching point, the method further comprises: if the distance between the first alternative matching point of the first matching area and the second alternative matching point of the second matching area is smaller than the preset distance threshold value, determining that the first alternative matching point and the second alternative matching point are the same alternative matching point; and determining that the sample patches corresponding to the same candidate matching point comprise a first sample patch corresponding to the first candidate matching point and a second sample patch corresponding to the second candidate matching point.
10. The method of claim 1, wherein determining target matching locations based on each sample patch and each candidate matching point from the weighted bipartite graph comprises: If the similarity value of the sample plaque at the target matching position is not lower than a preset similarity threshold value, obtaining a corresponding sample plaque; If the sample plaque does not have the target matching position, obtaining a deleted sample plaque; If the similarity value of the sample plaque is lower than the similarity threshold value, obtaining a deleted sample plaque or a newly added sample plaque; And obtaining the target labeling result according to the corresponding sample plaque, the deleted sample plaque and the newly added sample plaque, wherein the target labeling result comprises a target label and a certainty corresponding to the target label, and the certainty is used for representing a credibility measure between the target label and an actual labeling result.
11. A migration correction system for high resolution remote sensing image annotation, the system comprising: the device comprises an acquisition module, a display module and a display module, wherein the acquisition module is used for acquiring a sample image and a target image, wherein the sample image and the target image both comprise at least one target object, and the sample image is configured with labeling information of the target object; The probability prediction module is used for inputting the target image into a pre-trained prediction model to obtain pixel probability values of which each pixel point in the target image is the target object, and establishing a prediction probability map of the target image according to each pixel probability value; the matching module is used for determining a sample plaque of the target object in the sample image according to the marking information, determining a matching area of the sample plaque in the prediction probability map, and correcting and adjusting the position of the sample plaque in the corresponding matching area; the computing module is used for computing the similarity value of at least one sample patch and each pixel point in the matching area, and determining a weighted bipartite graph by taking the similarity of each candidate matching point as a weight according to the candidate matching point corresponding to each sample patch and the sample patch corresponding to each candidate matching point, wherein the weighted bipartite graph is used for representing the weight relation between each sample patch and each candidate matching point; Determining target matching positions based on each sample plaque and each alternative matching point from the weighted bipartite graph by taking total maximum weight matching as constraint conditions, so that each sample plaque after constraint matching integrally corresponds to the target object in the target image; And the result module is used for obtaining a target marking result of the target object in the target image according to the target matching position.
12. An electronic device comprising a memory and a processor, the memory storing a computer program, the processor implementing the migration correction method for high resolution remote sensing image annotation according to any of claims 1 to 10 when executing the computer program.
13. A computer readable storage medium storing a computer program, wherein the computer program when executed by a processor implements the migration correction method for high resolution remote sensing image annotation according to any one of claims 1 to 10.

Description

Migration correction method, system and storage medium for high-resolution remote sensing image annotation Technical Field The application relates to the field of remote sensing image processing, in particular to a migration correction method, a migration correction system and a storage medium for high-resolution remote sensing image labels. Background Remote sensing (remote sensing) refers to a non-contact, remote detection technique. In the development of the current society, the function of remote sensing technology is more and more important, and powerful technical support and data support are provided for the fields of geographic information systems and space planning, agricultural production and natural resource management, weather forecast and disaster monitoring, infrastructure construction, traffic management and the like. When the remote sensing image is subjected to data processing, the data marking is usually required, the purpose of the remote sensing image marking is to help an operator to better understand or further utilize the remote sensing image, and under the condition that the data marking of the same region and one time phase remote sensing image is finished, the analysis is performed on the same region and the other time phase remote sensing image, so that the earth surface information change condition of the region in different time periods can be obtained. In the related art, the data annotation of the remote sensing image is subjected to annotation migration, but the accuracy of the remote sensing image data annotation migration is reduced due to the difference of imaging angles and the difference between sensors used for shooting. Disclosure of Invention The embodiment of the application mainly aims to provide a migration correction method, a migration correction system and a storage medium for high-resolution remote sensing image annotation, which can improve the accuracy of migration correction of remote sensing image data annotation of the same area and different time phases. In order to achieve the above objective, a first aspect of the present application provides a migration correction method for labeling a high-resolution remote sensing image, which includes obtaining a sample image and a target image, wherein the sample image and the target image each include at least one target object, the sample image is configured with labeling information of the target object, inputting the target image into a pre-trained prediction model to obtain pixel probability values of each pixel point in the target image as the target object, establishing a prediction probability map of the target image according to each pixel probability value, determining a sample plaque where the target object is located in the sample image according to the labeling information, determining a matching region of the sample plaque in the prediction probability map, calculating a similarity value of each pixel point in at least one sample plaque and the matching region, and constraining and matching each sample plaque in the matching region based on the similarity value, so as to obtain a target matching position, so that each sample after constraint matching corresponds to the target object in the whole target image, and the target object is labeled in the target image according to the target position. In some embodiments, according to a target labeling result of the target image, a target plaque is obtained, the target plaque comprises a target plaque characteristic value and target plaque labeling information corresponding to the target plaque characteristic value, the sample plaque comprises a sample plaque characteristic value and sample plaque labeling information corresponding to the sample plaque characteristic value, after the target labeling result of the target object in the target image is obtained according to the target matching position, the method further comprises the steps of obtaining a weight parameter value for amplification, performing difference operation on the weight parameter value and unit weight to obtain a differential weight parameter value, multiplying the weight parameter value and the sample plaque characteristic value to obtain a first sub-value, multiplying the differential weight parameter value and the target plaque characteristic value to obtain a second sub-value, adding the first sub-value and the second sub-value to obtain first amplification information, multiplying the weight parameter value and the sample plaque labeling information to obtain a third sub-value, multiplying the differential weight parameter value and the target plaque labeling information to obtain a fourth sub-value, adding the differential weight parameter value and the fourth amplification value to obtain the improved sample labeling information, and performing iteration on the improved sample labeling result. In some embodiments, the determining the sample plaque of the target object in the sample