SuperMap's new generation remote sensing software technology system is built on the basis of Geospatial AI Foundations (AIF), including remote sensing data management, remote sensing data processing, remote sensing interpretation and analysis, remote sensing data visualization and other technical aspects.
Remote sensing data management is based on the mosaic dataset model, which supports the management of products at different levels such as remote sensing raw data, remote sensing primary products, orthorectified products, image fusion products, uniform color mosaic products, standard framing products, etc. Remote sensing data processing technology provides core capabilities including geometric correction, orthophoto production, stereo image production, image quality inspection, and radiation correction.
Remote sensing interpretation and analysis technology is based on the underlying rich remote sensing interpretation algorithms and can provide pre-trained models for different scenarios. Remote sensing data visualization includes raw image data visualization, image cache visualization, image service visualization, image uniform color preview, and image enhancement visualization.
Remote sensing data management
Remote sensing image data management is mainly achieved through mosaic datasets, which provide image data browsing, query, and update capabilities. Before querying and browsing, you need to create an image pyramid and an overview. In “image data query”, you can intuitively query the image of the outline sub-dataset of the mosaic dataset.
In image data update, when the image storage location changes, you can update the file path of the image in the mosaic dataset through "Respecify path"; when adding or deleting images in the mosaic dataset, it provides the ability to batch update image data.
Based on the above capabilities, SuperMap supports the management of multiple types of remote sensing data products, including remote sensing raw data, remote sensing primary products, ortho-rectified products, image fusion products, uniform mosaic products, standard segmented products, etc.
Among them, remote sensing raw data is relatively special. In addition to image files, it also includes metadata files and rational function model parameter files. For remote sensing raw data, mosaic datasets can read the corresponding metadata files based on different satellite types and add image data to the mosaic dataset for management to obtain rich information such as sensors, cloud coverage, data production time, etc. In addition, for the management of uniform mosaic products, it also provides the ability to display uniform color effects and mosaic effects in real-time.
Remote sensing data processing
The remote sensing data processing technology integrates the world's leading photogrammetry algorithms and is oriented to optical satellite remote sensing image data. It has the ability to automatically and quickly produce high-precision DOM, DSM/DEM data, and supports the processing of commonly used satellite remote sensing data such as Gaofen series, Ziyuan series, Beijing, Gaojing, Jilin, Sentinel, Landsat, IKONOS, WorldView, SPOT, etc.
Geometric correction
Geometric correction refers to the work of projection transformation, target space plane position correction and geometric matching correction between different sensor images to eliminate the geometric distortion of the image. This module mainly includes technologies such as image matching, regional network adjustment and image registration. Its main feature is that it realizes intelligent geometric processing by combining traditional photogrammetry technology with artificial intelligence technology.
On the one hand, this technology uses the built-in AI heterogeneous image matching model, introduces heterogeneous image feature detection and robust feature descr1ption methods, improves the number and accuracy of image matching, and ensures that good geometric correction effects can be obtained in weak texture areas; on the other hand, based on AI semantic information, it assists in eliminating non-ground points distributed on clouds, buildings, etc. during image matching, thereby improving the accuracy of geometric processing.
AI semantic-assisted geometric processing (right) technology effect
Orthophoto production
The orthophoto production module mainly includes: generating orthophotos, image fusion, image enhancement (true color output), image color grading, constructing mosaic lines, image de-clouding, image defogging and other technologies.
Image color grading adopts a method of fine-grained spatial segmentation and color consistency processing with the reference images of the corresponding areas. It can not only make the color-graded image consistent with the template image in the corresponding spatial position, but also better reflect the detail differences in color in space, and the tonal transition between objects is more natural and realistic.
Image de-clouding includes two steps: cloud detection and cloud replacement. Based on the optimized cloud detection AI model, cloud areas can be detected more accurately. After the user provides high-quality replacement images, it can support image de-clouding, automatically replace and color-grade the cloudy areas in the image, and obtain high-quality cloud-free images.
Image cloud removal example
Stereo-scopic image production
Dense matching and DSM filtering are two core processing processes in stereo image production. By introducing AI intelligent terrain data production technology, on the one hand, the built-in AI dense matching model can effectively reduce mismatching during dense matching, which can not only reduce the generation of valueless areas, but also make the DSM results free of holes, more complete, more natural in visual effects, and clearer in image contours; on the other hand, the built-in AI filtering model can be used to improve the adaptability to complex terrain areas during the filtering process, thereby improving the filtering processing effect in mountainous and urban areas.
Comparison of stereoscopic image production effects
Image quality inspection
The image quality inspection module mainly includes: plane accuracy inspection, edge accuracy inspection, regional accuracy inspection, pattern deformation detection and other technologies.
The inspection results of plane and edge accuracy can be visualized through thematic maps, which is convenient for users to intuitively evaluate the accuracy of the results; regional accuracy inspection can be used to conduct detailed block inspection of the entire image, which is convenient for users to understand the local accuracy exceeding the limit; pattern deformation detection can automatically extract the pattern deformation areas of mountains, roads and houses, reducing manual workload and improving efficiency.
Examples of thematic maps for plane accuracy check and regional accuracy check
To be continued...
