Data processing system and method based on remote sensing data

Abstract

The invention relates to the technical field of image data processing, and discloses a data processing system and method based on remote sensing data, and the processing system is composed of an image collection module, an image preprocessing module, a feature extraction module, an image matching module, an image splicing module and a registration calculation module. Through the denoising technology, random noise in the images is reduced, the feature points are more stable under the influence of noise, the feature extraction accuracy is further improved, after radiation correction is carried out, the radiation difference between the images can be corrected, image changes caused by different imaging conditions are reduced, and the image extraction accuracy is improved. When feature matching is carried out through a random sampling consistency method, wrong matching is filtered out through an inner point set, the quality of matching pairs is improved, reliable input is provided for the image registration stage, and the beneficial effect of improving the image preprocessing precision is achieved.

Description

Technical Field

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

[0002] In modern society, unmanned aerial vehicles (UAVs) have been widely used in various different fields, including military reconnaissance, environmental monitoring, geographic information collection, agricultural disaster assessment, etc. In these applications, the use of thermal infrared images has important value. Thermal infrared images provide a way of observation regardless of lighting conditions, and can reveal the heat distribution and heat generation of objects, and this information is crucial for many fields. However, due to the limitations of the perspective and shooting distance of the UAV during flight, it is usually impossible to obtain complete regional information at one time. Therefore, it is necessary to splice multiple obtained thermal infrared images to form a larger-scale and continuous thermal infrared image information. The existing thermal infrared image splicing technology is mainly based on traditional image matching, including image preprocessing, feature extraction, feature matching, and image fusion, etc. When traditional image splicing technology processes thermal infrared images, since no targeted optimization is performed on thermal infrared images, the processes of feature extraction and matching rely on the quality and clarity of the images. The characteristics of thermal infrared images result in insufficient accuracy of these processes, thus affecting the final splicing accuracy.

[0003] After retrieval, it is found that a remote sensing data processing system and method with the publication number of CN 116797953 B in the prior art, the system includes: an image collection module, an image preprocessing module, a feature extraction module, an image matching module, an image splicing module, a registration calculation module, and a system management module; the method includes the following steps: obtaining thermal infrared images, performing preprocessing to obtain key points, matching to obtain corresponding points, obtaining the target thermal infrared image after splicing, calculating the target structural similarity index value, and judging and adjusting the target thermal infrared image. By performing targeted preprocessing on thermal infrared images and further adjusting the spliced images, the splicing accuracy of thermal infrared images is effectively improved.

[0004] The foregoing technical solution considers multiple modules and steps in the design to improve the splicing accuracy of thermal infrared images. However, the effect of image preprocessing directly affects the subsequent steps. If the preprocessing is insufficient or inaccurate, it will lead to poor results in feature extraction and matching. The effect of image splicing is affected by various factors, such as the lighting, temperature change, cloud cover, etc. between different images, which results in seams or inconsistencies in the spliced image. Summary of the Invention

[0005] Technical Problem to be Solved In view of the deficiencies of the prior art, the present invention provides a data processing system and method based on remote sensing data, which has the advantages of improving the accuracy of images during preprocessing and ensuring that there are no seams in the images after splicing, thereby solving the above technical problems.

[0006] Technical solution: To achieve the above object, the present invention provides the following technical solution: A data processing system based on remote sensing data, the processing system is composed of an image collection module, an image preprocessing module, a feature extraction module, an image matching module, an image splicing module, and a registration calculation module; The image collection module is used to automatically collect thermal infrared images, including image data at different times and locations; The image preprocessing module is used to perform denoising, filtering, and radiometric correction on the collected images; The feature extraction module is used to extract key feature points from the preprocessed images, and the scale-invariant feature transform algorithm is used for the feature extraction; The image matching module is based on the method of random sample consensus to enhance the accuracy of matching; The image splicing module is used to seamlessly splice the matched images, adopts a multi-level fusion technology, dynamically adjusts the weights of the splicing areas, and eliminates seams; The registration calculation module is used to calculate the registration transformation between images to ensure the geometric consistency of the images; The expression of the image splicing module is: Calculation of the weight of the overlapping area:

[0007] Where: is a dynamically adjusted weight function, is the distance to the edge of the overlapping area, is the set maximum distance; Fusion:

[0008] Where: is the pixel value of the spliced image at position, and the two matched images are respectively and ; In order to eliminate the seam effect, the lighting and color are adjusted in the splicing area to further optimize the fusion effect:

[0009] Where is a function for realizing color and lighting adjustment, which is used to ensure that the splicing effect is natural and smooth.

[0010] Preferably, the expression of the image collection module is:

[0011] Among them, represents the set of thermal infrared image data collected at time and location The collected thermal infrared image data set, represents images at different times and locations; Image collection process:

[0012] Among them, represents the set of the time and location of image collection, represents the th time of collection of time and location.

[0013] Preferably, the complete image set of the image collection module:

[0014] Among them, represents the total set of all collected thermal infrared images.

[0015] Preferably, the image denoising expression in the image preprocessing module is: ; Among them, represents the denoised image, represents the originally collected image, is the denoising function; The image filtering expression is: ; Among them, represents the filtered image, is the filtering function; The radiation correction expression is: ; Among them, represents the image after radiation correction processing, is the radiation correction function.

[0016] Preferably, the complete preprocessing process expression of the image preprocessing module is:

[0017] Among them, represents the finally preprocessed image.

[0018] Preferably, the image feature extraction expression in the feature extraction module is: ; Among them, represents the set of extracted key feature points, is the image after preprocessing, is the Scale-Invariant Feature Transform (SIFT) function; The description expression of the feature points is as follows: ; where represents the set of descriptors of the feature points, represents the th description of the feature points.

[0019] Preferably, the feature point verification expression in the feature extraction module is: ; where represents the function for verifying the feature points; The expression of the complete feature extraction process in the feature extraction module is: ; where: represents the set of feature points after final extraction and verification.

[0020] Preferably, the feature matching expression in the image matching module is: ; where represents the obtained feature matching pairs, and are respectively the sets of feature points extracted from two images; The expression of the random sample consensus process in the image matching module is: ; where represents the estimated transformation model, represents the set of inliers obtained after calculation by the method, is the allowable error threshold, is the number of feature points for random sampling; The expression of the set of inliers in the image matching module is: where represents the set of inliers, is the feature point in the matching pair.

[0021] Preferably, the expression for defining the transformation model in the registration module is:

[0022] where represents the transformation model, is the coordinate of the original image, is the transformation parameter.

[0023] A data processing system and method based on remote sensing data includes the following steps: Step 1. Image preprocessing: Collect the original image data, perform denoising on the original image to obtain a denoised image, perform image filtering to further smooth the image, perform radiometric correction, and record the finally processed image as the basis for subsequent analysis; Step 2. Feature Extraction: Extract key feature points from the preprocessed image to identify and describe important features in the image. For the extracted feature points, generate corresponding feature descriptors to provide more abundant information and verify the extracted feature points. Step 3. Feature Matching: Match the feature points in two images to find similar features. Use the Random Sample Consensus (RANSAC) method to estimate the transformation relationship between the feature points and identify the inlier set, and distinguish valid matching pairs for subsequent processing. Step 4. Image Registration: Define a transformation model between images to achieve precise registration of the images. According to the matching result and the transformation model, align one image with the other to ensure the alignment of the two images.

[0024] Compared with the prior art, the present invention provides a data processing system and method based on remote sensing data, having the following beneficial effects: 1. Through the denoising technology, the present invention reduces the random noise in the image, making the feature points more stable under the influence of noise, thereby improving the accuracy of feature extraction. After radiometric correction, the radiometric differences between images are corrected, reducing the image variations caused by different imaging conditions, ensuring that the features are easy to match, improving the registration accuracy. The accurate feature point extraction and verification steps can ensure that the descriptions generated by the key features in the image are accurate, reduce redundant data, and ensure that the subsequent image matching process is more efficient and accurate. When performing feature matching by the RANSAC method, the inlier set is used to filter out incorrect matches, improving the quality of the matching pairs and providing reliable input for the image registration stage, achieving the beneficial effect of improving the accuracy of the image during preprocessing.

[0025] 2. Through an efficient feature point extraction method, the present invention ensures that consistent and important feature points are identified in the stitched images, which provides an accurate basis for subsequent matching and reduces the seam problems caused by incorrect feature points. During the feature matching process, the RANSAC algorithm is used to remove incorrect matching points and retain the inlier set, significantly improving the reliability of the selected feature points, thereby effectively reducing the seams during stitching. During the stitching process, the gradual fusion technology is applied to gradually transition in the seam area rather than suddenly switch, reducing the visual discomfort. This smooth transition makes the seam almost invisible, achieving the beneficial effect that the image after stitching does not have seams and other advantages. BRIEF DESCRIPTION OF THE DRAWINGS

[0026] Figure 1 It is a schematic flowchart of the method of the present invention. DETAILED DESCRIPTION OF THE INVENTION

[0027] Next, the technical solutions in the embodiments of the present invention will be clearly and completely described in conjunction with the accompanying drawings in the embodiments of the present invention. Obviously, the described embodiments are only a part of the embodiments of the present invention, rather than all the embodiments. All other embodiments obtained by those of ordinary skill in the art based on the embodiments of the present invention without creative efforts shall fall within the protection scope of the present invention.

[0028] Please refer to Figure 1 , a data processing system based on remote sensing data, the processing system is composed of an image collection module, an image preprocessing module, a feature extraction module, an image matching module, an image stitching module and a registration calculation module; The image collection module is used to automatically collect thermal infrared images, including image data at different times and locations; The image preprocessing module is used to perform denoising, filtering, and radiometric correction on the collected images; The feature extraction module is used to extract key feature points from the preprocessed images, and the scale-invariant feature transform algorithm is used for the feature extraction; The image matching module is based on the method of random sample consensus to enhance the accuracy of matching; The image stitching module is used to seamlessly stitch the matched images, adopts a multi-level fusion technology, dynamically adjusts the weights of the stitching areas, and eliminates seams; The registration calculation module is used to calculate the registration transformation between images to ensure the geometric consistency of the images; The expression of the image stitching module is: Calculation of the weight of the overlapping area:

[0029] Where: is a dynamically adjusted weight function, is the distance to the edge of the overlapping area, is the set maximum distance; Fusion:

[0030] Where: is the pixel value of the stitched image at the position, and the two matched images are respectively and ; In order to eliminate the seam effect, the lighting and color are adjusted in the stitching area to further optimize the fusion effect:

[0031] Where is a function to achieve color and lighting adjustment, which is used to ensure a natural and smooth stitching effect.

[0032] The image collection module can automatically collect thermal infrared images at different times and locations, reducing manual intervention and improving the efficiency of data collection.

[0033] The image preprocessing module ensures the quality of the input images through processes such as denoising, filtering, and radiometric correction, reducing errors in subsequent processing.

[0034] The Scale-Invariant Feature Transform (SIFT) algorithm is used to extract key feature points, making the system highly robust to images of different scales and rotations and improving the accuracy of feature matching.

[0035] The image matching module is based on the Random Sample Consensus (RANSAC) method, which can effectively filter out incorrect matches, enhance the accuracy of matching, and ensure the reliability of the stitching effect.

[0036] The image stitching module uses multi-level fusion technology and dynamic weight adjustment to achieve seamless stitching, eliminate seams, and provide a more natural visual effect.

[0037] Through the calculation of the weights in the overlapping areas, the system can dynamically adjust the weights of the stitching areas according to the distance of the pixels to the edges of the overlapping areas, making the stitching effect smoother.

[0038] Adjust the lighting and color in the stitching areas to further optimize the fusion effect, ensuring that the stitched images are consistent in color and brightness and avoiding visual abruptness.

[0039] The registration calculation module ensures the geometric consistency between images, avoiding stitching problems caused by image deformation or misalignment and improving the accuracy of the overall image.

[0040] The system can process image data at different times and locations, adapting to various remote sensing application scenarios such as environmental monitoring and urban planning.

[0041] By generating high-quality stitched images, the system provides a reliable basis for subsequent data analysis and decision support, enhancing the application value of remote sensing data.

[0042] Through denoising technology, the random noise in the images is reduced, making the feature points more stable under the influence of noise, and thus improving the accuracy of feature extraction.

[0043] Apply filtering technology to smooth the images and remove small detail interferences, thereby ensuring the extraction of important feature points rather than irrelevant noise or interference signals and improving the overall quality of the images.

[0044] After radiometric correction, radiometric differences between images are corrected, reducing image variations due to different imaging conditions such as lighting, sensor properties, etc. This process ensures that features can be easily matched, improving registration accuracy.

[0045] Accurate feature point extraction and verification steps can ensure accurate description of key features in the image, reduce redundant data, and ensure that the subsequent image matching process is more efficient and accurate.

[0046] When performing feature matching through the random sampling consistency method, the inlier set is used to filter out false matches, improve the quality of the matching pairs, and provide a reliable input for the image registration stage.

[0047] In the process of image registration, we rely on more accurate transformation models to better map the relationship between images, thereby achieving higher-precision image overlap and alignment.

[0048] Through efficient feature point extraction methods, we ensure that consistent and important feature points are identified in the stitched images. This provides an accurate basis for subsequent matching and reduces seam problems caused by feature point errors.

[0049] In the feature matching process, the RANSAC algorithm is used to remove incorrect matching points and retain the internal point set. In this way, the reliability of the selected feature points can be significantly improved, thereby effectively reducing the seams during splicing.

[0050] Use appropriate transformation models (such as affine transformation, perspective transformation, etc.) to accurately describe the geometric relationship between images, achieve more precise image alignment, and reduce seams caused by transformation errors.

[0051] Perform radiation correction and color balancing to ensure that the splicing area is consistent in brightness and tone. This treatment effectively reduces the visual conflict of the splicing part and avoids the obvious existence of the seam.

[0052] In the splicing process, the gradient fusion technology is applied to gradually transition in the seam area instead of switching suddenly, reducing visual discomfort. This smooth transition makes the seam almost invisible.

[0053] Use global optimization technology to adjust the entire stitched image to ensure the spatial consistency and coordination of the stitched parts, thereby improving the overall stitching quality.

[0054] 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 alterations may be made to the embodiments without departing from the principles and spirit of the invention, the scope of which is defined by the appended claims and their equivalents.

Claims

1. A data processing system based on remote sensing data, characterized in that: The processing system is composed of an image collection module, an image preprocessing module, a feature extraction module, an image matching module, an image stitching module and a registration calculation module; The image collection module is used to automatically collect thermal infrared images, including image data at different times and locations; The image preprocessing module is used to perform denoising, filtering, and radiation correction on the collected images; The feature extraction module is used to extract key feature points from the preprocessed image, and the feature extraction uses a scale-invariant feature transformation algorithm; The image matching module is based on the random sampling consistency method to enhance the matching accuracy; The image stitching module is used to seamlessly stitch the matched images, using multi-level fusion technology to dynamically adjust the weight of the stitching area and eliminate seams; The registration calculation module is used to calculate the registration transformation between images to ensure the geometric consistency of the images; The image stitching module expression is: Overlap area weight calculation: in: is a dynamically adjusted weight function, is the distance to the edge of the overlap region, is the maximum distance set; Fusion: in: After stitching, the image The pixel values ​​of the two matched images are and ; In order to eliminate the seam effect, adjust the lighting and color in the stitching area to further optimize the fusion effect: in It is a function that implements color and lighting adjustments to ensure a natural and smooth stitching effect.

2. A data processing system based on remote sensing data according to claim 1, characterized in that: The image collection module expression is: in, Indicates at time and location The thermal infrared image data set collected, images representing different times and places; Image collection process: in, represents the collection of time and place where the image was collected, Indicates The time and date of the collection.

3. A data processing system based on remote sensing data according to claim 2, characterized in that: The complete image collection of the image collection module: in, Represents the total set of all collected thermal infrared images.

4. A data processing system based on remote sensing data according to claim 1, characterized in that: The image denoising expression in the image preprocessing module is: ;in, represents the denoised image, represents the original collected image, is the denoising function; The image filtering expression is: ;in, represents the filtered image, is the filter function; The radiation correction expression is: ;in, Represents the image after radiometric correction. is the radiation correction function.

5. A data processing system based on remote sensing data according to claim 4, characterized in that: The complete preprocessing process expression of the image preprocessing module is: in, Represents the final preprocessed image.

6. A data processing system based on remote sensing data according to claim 1, characterized in that: The image feature extraction expression in the feature extraction module is: ;in, represents the set of extracted key feature points, is the pre-processed image. It is the scale-invariant feature transform algorithm SIFT function; The feature point description expression is: ;in, A set of descriptors representing feature points, Indicates Description of feature points.

7. A data processing system based on remote sensing data according to claim 6, characterized in that: The feature point verification expression in the feature extraction module is: ;in, A function representing a verification feature point; The complete feature extraction process expression in the feature extraction module is: ;in: Represents the final set of feature points extracted and verified.

8. The data processing system based on remote sensing data according to claim 1, characterized in that: The feature matching expression in the image matching module is: ;in, represents the obtained feature matching pair, and They are the sets of feature points extracted from the two images; The expression of the random sampling consistency process in the image matching module is: ;in, represents the estimated transformation model, Indicates passing The set of interior points obtained by the method is: is the allowable error threshold, is the number of feature points used for random sampling; The expression of the interior point set in the image matching module is: in, represents the interior point set, are the feature points in the matching pair.

9. The data processing system based on remote sensing data according to claim 1, characterized in that: The transformation model definition expression in the registration module is: in, represents the transformation model, are the coordinates of the original image, is the transformation parameter.

10. A method for data processing based on remote sensing data according to claim 9, characterized in that: The following steps are involved: Step 1: Image preprocessing: collect the original image data, denoise the original image to obtain the denoised image, filter the image, further smooth the image, perform radiation correction, and record the final processed image as the basis for subsequent analysis; Step 2: Feature extraction: Extract key feature points from the preprocessed image to identify and describe important features in the image. Generate corresponding feature descriptors for the extracted feature points to provide richer information and verify the extracted feature points. Step 3: Feature matching: Match the feature points in the two images to find similar features, use the random sampling consistency method to estimate the transformation relationship between the feature points, identify the internal point set, and identify the valid matching pairs for subsequent processing; Step 4: Image registration: Define the transformation model between images to achieve accurate image registration; based on the matching results and the transformation model, align one image with the other to ensure the alignment of the two images.

Images